[PATCH v1 33/50] mips: octeon: Add misc remaining header files

Stefan Roese sr at denx.de
Fri Dec 11 17:05:55 CET 2020


From: Aaron Williams <awilliams at marvell.com>

Import misc remaining header files from 2013 U-Boot. These will be used
by the later added drivers to support PCIe and networking on the MIPS
Octeon II / III platforms.

Signed-off-by: Aaron Williams <awilliams at marvell.com>
Signed-off-by: Stefan Roese <sr at denx.de>
---

 .../mach-octeon/include/mach/cvmx-address.h   |  209 ++
 .../mach-octeon/include/mach/cvmx-cmd-queue.h |  441 +++
 .../mach-octeon/include/mach/cvmx-csr-enums.h |   87 +
 arch/mips/mach-octeon/include/mach/cvmx-csr.h |   78 +
 .../mach-octeon/include/mach/cvmx-error.h     |  456 +++
 arch/mips/mach-octeon/include/mach/cvmx-fpa.h |  217 ++
 .../mips/mach-octeon/include/mach/cvmx-fpa1.h |  196 ++
 .../mips/mach-octeon/include/mach/cvmx-fpa3.h |  566 ++++
 .../include/mach/cvmx-global-resources.h      |  213 ++
 arch/mips/mach-octeon/include/mach/cvmx-gmx.h |   16 +
 .../mach-octeon/include/mach/cvmx-hwfau.h     |  606 ++++
 .../mach-octeon/include/mach/cvmx-hwpko.h     |  570 ++++
 arch/mips/mach-octeon/include/mach/cvmx-ilk.h |  154 +
 arch/mips/mach-octeon/include/mach/cvmx-ipd.h |  233 ++
 .../mach-octeon/include/mach/cvmx-packet.h    |   40 +
 .../mips/mach-octeon/include/mach/cvmx-pcie.h |  279 ++
 arch/mips/mach-octeon/include/mach/cvmx-pip.h | 1080 ++++++
 .../include/mach/cvmx-pki-resources.h         |  157 +
 arch/mips/mach-octeon/include/mach/cvmx-pki.h |  970 ++++++
 .../mach/cvmx-pko-internal-ports-range.h      |   43 +
 .../include/mach/cvmx-pko3-queue.h            |  175 +
 arch/mips/mach-octeon/include/mach/cvmx-pow.h | 2991 +++++++++++++++++
 arch/mips/mach-octeon/include/mach/cvmx-qlm.h |  304 ++
 .../mach-octeon/include/mach/cvmx-scratch.h   |  113 +
 arch/mips/mach-octeon/include/mach/cvmx-wqe.h | 1462 ++++++++
 25 files changed, 11656 insertions(+)
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-address.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-cmd-queue.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-csr-enums.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-csr.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-error.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-fpa.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-fpa1.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-fpa3.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-global-resources.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-gmx.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-hwfau.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-hwpko.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-ilk.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-ipd.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-packet.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-pcie.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-pip.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-pki-resources.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-pki.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-pko-internal-ports-range.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-pko3-queue.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-pow.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-qlm.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-scratch.h
 create mode 100644 arch/mips/mach-octeon/include/mach/cvmx-wqe.h

diff --git a/arch/mips/mach-octeon/include/mach/cvmx-address.h b/arch/mips/mach-octeon/include/mach/cvmx-address.h
new file mode 100644
index 0000000000..984f574a75
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-address.h
@@ -0,0 +1,209 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ *
+ * Typedefs and defines for working with Octeon physical addresses.
+ */
+
+#ifndef __CVMX_ADDRESS_H__
+#define __CVMX_ADDRESS_H__
+
+typedef enum {
+	CVMX_MIPS_SPACE_XKSEG = 3LL,
+	CVMX_MIPS_SPACE_XKPHYS = 2LL,
+	CVMX_MIPS_SPACE_XSSEG = 1LL,
+	CVMX_MIPS_SPACE_XUSEG = 0LL
+} cvmx_mips_space_t;
+
+typedef enum {
+	CVMX_MIPS_XKSEG_SPACE_KSEG0 = 0LL,
+	CVMX_MIPS_XKSEG_SPACE_KSEG1 = 1LL,
+	CVMX_MIPS_XKSEG_SPACE_SSEG = 2LL,
+	CVMX_MIPS_XKSEG_SPACE_KSEG3 = 3LL
+} cvmx_mips_xkseg_space_t;
+
+/* decodes <14:13> of a kseg3 window address */
+typedef enum {
+	CVMX_ADD_WIN_SCR = 0L,
+	CVMX_ADD_WIN_DMA = 1L,
+	CVMX_ADD_WIN_UNUSED = 2L,
+	CVMX_ADD_WIN_UNUSED2 = 3L
+} cvmx_add_win_dec_t;
+
+/* decode within DMA space */
+typedef enum {
+	CVMX_ADD_WIN_DMA_ADD = 0L,
+	CVMX_ADD_WIN_DMA_SENDMEM = 1L,
+	/* store data must be normal DRAM memory space address in this case */
+	CVMX_ADD_WIN_DMA_SENDDMA = 2L,
+	/* see CVMX_ADD_WIN_DMA_SEND_DEC for data contents */
+	CVMX_ADD_WIN_DMA_SENDIO = 3L,
+	/* store data must be normal IO space address in this case */
+	CVMX_ADD_WIN_DMA_SENDSINGLE = 4L,
+	/* no write buffer data needed/used */
+} cvmx_add_win_dma_dec_t;
+
+/**
+ *   Physical Address Decode
+ *
+ * Octeon-I HW never interprets this X (<39:36> reserved
+ * for future expansion), software should set to 0.
+ *
+ *  - 0x0 XXX0 0000 0000 to      DRAM         Cached
+ *  - 0x0 XXX0 0FFF FFFF
+ *
+ *  - 0x0 XXX0 1000 0000 to      Boot Bus     Uncached  (Converted to 0x1 00X0 1000 0000
+ *  - 0x0 XXX0 1FFF FFFF         + EJTAG                           to 0x1 00X0 1FFF FFFF)
+ *
+ *  - 0x0 XXX0 2000 0000 to      DRAM         Cached
+ *  - 0x0 XXXF FFFF FFFF
+ *
+ *  - 0x1 00X0 0000 0000 to      Boot Bus     Uncached
+ *  - 0x1 00XF FFFF FFFF
+ *
+ *  - 0x1 01X0 0000 0000 to      Other NCB    Uncached
+ *  - 0x1 FFXF FFFF FFFF         devices
+ *
+ * Decode of all Octeon addresses
+ */
+typedef union {
+	u64 u64;
+	struct {
+		cvmx_mips_space_t R : 2;
+		u64 offset : 62;
+	} sva;
+
+	struct {
+		u64 zeroes : 33;
+		u64 offset : 31;
+	} suseg;
+
+	struct {
+		u64 ones : 33;
+		cvmx_mips_xkseg_space_t sp : 2;
+		u64 offset : 29;
+	} sxkseg;
+
+	struct {
+		cvmx_mips_space_t R : 2;
+		u64 cca : 3;
+		u64 mbz : 10;
+		u64 pa : 49;
+	} sxkphys;
+
+	struct {
+		u64 mbz : 15;
+		u64 is_io : 1;
+		u64 did : 8;
+		u64 unaddr : 4;
+		u64 offset : 36;
+	} sphys;
+
+	struct {
+		u64 zeroes : 24;
+		u64 unaddr : 4;
+		u64 offset : 36;
+	} smem;
+
+	struct {
+		u64 mem_region : 2;
+		u64 mbz : 13;
+		u64 is_io : 1;
+		u64 did : 8;
+		u64 unaddr : 4;
+		u64 offset : 36;
+	} sio;
+
+	struct {
+		u64 ones : 49;
+		cvmx_add_win_dec_t csrdec : 2;
+		u64 addr : 13;
+	} sscr;
+
+	/* there should only be stores to IOBDMA space, no loads */
+	struct {
+		u64 ones : 49;
+		cvmx_add_win_dec_t csrdec : 2;
+		u64 unused2 : 3;
+		cvmx_add_win_dma_dec_t type : 3;
+		u64 addr : 7;
+	} sdma;
+
+	struct {
+		u64 didspace : 24;
+		u64 unused : 40;
+	} sfilldidspace;
+} cvmx_addr_t;
+
+/* These macros for used by 32 bit applications */
+
+#define CVMX_MIPS32_SPACE_KSEG0	     1l
+#define CVMX_ADD_SEG32(segment, add) (((s32)segment << 31) | (s32)(add))
+
+/*
+ * Currently all IOs are performed using XKPHYS addressing. Linux uses the
+ * CvmMemCtl register to enable XKPHYS addressing to IO space from user mode.
+ * Future OSes may need to change the upper bits of IO addresses. The
+ * following define controls the upper two bits for all IO addresses generated
+ * by the simple executive library
+ */
+#define CVMX_IO_SEG CVMX_MIPS_SPACE_XKPHYS
+
+/* These macros simplify the process of creating common IO addresses */
+#define CVMX_ADD_SEG(segment, add) ((((u64)segment) << 62) | (add))
+
+#define CVMX_ADD_IO_SEG(add) (add)
+
+#define CVMX_ADDR_DIDSPACE(did)	   (((CVMX_IO_SEG) << 22) | ((1ULL) << 8) | (did))
+#define CVMX_ADDR_DID(did)	   (CVMX_ADDR_DIDSPACE(did) << 40)
+#define CVMX_FULL_DID(did, subdid) (((did) << 3) | (subdid))
+
+/* from include/ncb_rsl_id.v */
+#define CVMX_OCT_DID_MIS  0ULL /* misc stuff */
+#define CVMX_OCT_DID_GMX0 1ULL
+#define CVMX_OCT_DID_GMX1 2ULL
+#define CVMX_OCT_DID_PCI  3ULL
+#define CVMX_OCT_DID_KEY  4ULL
+#define CVMX_OCT_DID_FPA  5ULL
+#define CVMX_OCT_DID_DFA  6ULL
+#define CVMX_OCT_DID_ZIP  7ULL
+#define CVMX_OCT_DID_RNG  8ULL
+#define CVMX_OCT_DID_IPD  9ULL
+#define CVMX_OCT_DID_PKT  10ULL
+#define CVMX_OCT_DID_TIM  11ULL
+#define CVMX_OCT_DID_TAG  12ULL
+/* the rest are not on the IO bus */
+#define CVMX_OCT_DID_L2C  16ULL
+#define CVMX_OCT_DID_LMC  17ULL
+#define CVMX_OCT_DID_SPX0 18ULL
+#define CVMX_OCT_DID_SPX1 19ULL
+#define CVMX_OCT_DID_PIP  20ULL
+#define CVMX_OCT_DID_ASX0 22ULL
+#define CVMX_OCT_DID_ASX1 23ULL
+#define CVMX_OCT_DID_IOB  30ULL
+
+#define CVMX_OCT_DID_PKT_SEND	 CVMX_FULL_DID(CVMX_OCT_DID_PKT, 2ULL)
+#define CVMX_OCT_DID_TAG_SWTAG	 CVMX_FULL_DID(CVMX_OCT_DID_TAG, 0ULL)
+#define CVMX_OCT_DID_TAG_TAG1	 CVMX_FULL_DID(CVMX_OCT_DID_TAG, 1ULL)
+#define CVMX_OCT_DID_TAG_TAG2	 CVMX_FULL_DID(CVMX_OCT_DID_TAG, 2ULL)
+#define CVMX_OCT_DID_TAG_TAG3	 CVMX_FULL_DID(CVMX_OCT_DID_TAG, 3ULL)
+#define CVMX_OCT_DID_TAG_NULL_RD CVMX_FULL_DID(CVMX_OCT_DID_TAG, 4ULL)
+#define CVMX_OCT_DID_TAG_TAG5	 CVMX_FULL_DID(CVMX_OCT_DID_TAG, 5ULL)
+#define CVMX_OCT_DID_TAG_CSR	 CVMX_FULL_DID(CVMX_OCT_DID_TAG, 7ULL)
+#define CVMX_OCT_DID_FAU_FAI	 CVMX_FULL_DID(CVMX_OCT_DID_IOB, 0ULL)
+#define CVMX_OCT_DID_TIM_CSR	 CVMX_FULL_DID(CVMX_OCT_DID_TIM, 0ULL)
+#define CVMX_OCT_DID_KEY_RW	 CVMX_FULL_DID(CVMX_OCT_DID_KEY, 0ULL)
+#define CVMX_OCT_DID_PCI_6	 CVMX_FULL_DID(CVMX_OCT_DID_PCI, 6ULL)
+#define CVMX_OCT_DID_MIS_BOO	 CVMX_FULL_DID(CVMX_OCT_DID_MIS, 0ULL)
+#define CVMX_OCT_DID_PCI_RML	 CVMX_FULL_DID(CVMX_OCT_DID_PCI, 0ULL)
+#define CVMX_OCT_DID_IPD_CSR	 CVMX_FULL_DID(CVMX_OCT_DID_IPD, 7ULL)
+#define CVMX_OCT_DID_DFA_CSR	 CVMX_FULL_DID(CVMX_OCT_DID_DFA, 7ULL)
+#define CVMX_OCT_DID_MIS_CSR	 CVMX_FULL_DID(CVMX_OCT_DID_MIS, 7ULL)
+#define CVMX_OCT_DID_ZIP_CSR	 CVMX_FULL_DID(CVMX_OCT_DID_ZIP, 0ULL)
+
+/* Cast to unsigned long long, mainly for use in printfs. */
+#define CAST_ULL(v) ((unsigned long long)(v))
+
+#define UNMAPPED_PTR(x) ((1ULL << 63) | (x))
+
+#endif /* __CVMX_ADDRESS_H__ */
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-cmd-queue.h b/arch/mips/mach-octeon/include/mach/cvmx-cmd-queue.h
new file mode 100644
index 0000000000..ddc294348c
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-cmd-queue.h
@@ -0,0 +1,441 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ *
+ * Support functions for managing command queues used for
+ * various hardware blocks.
+ *
+ * The common command queue infrastructure abstracts out the
+ * software necessary for adding to Octeon's chained queue
+ * structures. These structures are used for commands to the
+ * PKO, ZIP, DFA, RAID, HNA, and DMA engine blocks. Although each
+ * hardware unit takes commands and CSRs of different types,
+ * they all use basic linked command buffers to store the
+ * pending request. In general, users of the CVMX API don't
+ * call cvmx-cmd-queue functions directly. Instead the hardware
+ * unit specific wrapper should be used. The wrappers perform
+ * unit specific validation and CSR writes to submit the
+ * commands.
+ *
+ * Even though most software will never directly interact with
+ * cvmx-cmd-queue, knowledge of its internal workings can help
+ * in diagnosing performance problems and help with debugging.
+ *
+ * Command queue pointers are stored in a global named block
+ * called "cvmx_cmd_queues". Except for the PKO queues, each
+ * hardware queue is stored in its own cache line to reduce SMP
+ * contention on spin locks. The PKO queues are stored such that
+ * every 16th queue is next to each other in memory. This scheme
+ * allows for queues being in separate cache lines when there
+ * are low number of queues per port. With 16 queues per port,
+ * the first queue for each port is in the same cache area. The
+ * second queues for each port are in another area, etc. This
+ * allows software to implement very efficient lockless PKO with
+ * 16 queues per port using a minimum of cache lines per core.
+ * All queues for a given core will be isolated in the same
+ * cache area.
+ *
+ * In addition to the memory pointer layout, cvmx-cmd-queue
+ * provides an optimized fair ll/sc locking mechanism for the
+ * queues. The lock uses a "ticket / now serving" model to
+ * maintain fair order on contended locks. In addition, it uses
+ * predicted locking time to limit cache contention. When a core
+ * know it must wait in line for a lock, it spins on the
+ * internal cycle counter to completely eliminate any causes of
+ * bus traffic.
+ */
+
+#ifndef __CVMX_CMD_QUEUE_H__
+#define __CVMX_CMD_QUEUE_H__
+
+/**
+ * By default we disable the max depth support. Most programs
+ * don't use it and it slows down the command queue processing
+ * significantly.
+ */
+#ifndef CVMX_CMD_QUEUE_ENABLE_MAX_DEPTH
+#define CVMX_CMD_QUEUE_ENABLE_MAX_DEPTH 0
+#endif
+
+/**
+ * Enumeration representing all hardware blocks that use command
+ * queues. Each hardware block has up to 65536 sub identifiers for
+ * multiple command queues. Not all chips support all hardware
+ * units.
+ */
+typedef enum {
+	CVMX_CMD_QUEUE_PKO_BASE = 0x00000,
+#define CVMX_CMD_QUEUE_PKO(queue)                                                                  \
+	((cvmx_cmd_queue_id_t)(CVMX_CMD_QUEUE_PKO_BASE + (0xffff & (queue))))
+	CVMX_CMD_QUEUE_ZIP = 0x10000,
+#define CVMX_CMD_QUEUE_ZIP_QUE(queue)                                                              \
+	((cvmx_cmd_queue_id_t)(CVMX_CMD_QUEUE_ZIP + (0xffff & (queue))))
+	CVMX_CMD_QUEUE_DFA = 0x20000,
+	CVMX_CMD_QUEUE_RAID = 0x30000,
+	CVMX_CMD_QUEUE_DMA_BASE = 0x40000,
+#define CVMX_CMD_QUEUE_DMA(queue)                                                                  \
+	((cvmx_cmd_queue_id_t)(CVMX_CMD_QUEUE_DMA_BASE + (0xffff & (queue))))
+	CVMX_CMD_QUEUE_BCH = 0x50000,
+#define CVMX_CMD_QUEUE_BCH(queue) ((cvmx_cmd_queue_id_t)(CVMX_CMD_QUEUE_BCH + (0xffff & (queue))))
+	CVMX_CMD_QUEUE_HNA = 0x60000,
+	CVMX_CMD_QUEUE_END = 0x70000,
+} cvmx_cmd_queue_id_t;
+
+#define CVMX_CMD_QUEUE_ZIP3_QUE(node, queue)                                                       \
+	((cvmx_cmd_queue_id_t)((node) << 24 | CVMX_CMD_QUEUE_ZIP | (0xffff & (queue))))
+
+/**
+ * Command write operations can fail if the command queue needs
+ * a new buffer and the associated FPA pool is empty. It can also
+ * fail if the number of queued command words reaches the maximum
+ * set at initialization.
+ */
+typedef enum {
+	CVMX_CMD_QUEUE_SUCCESS = 0,
+	CVMX_CMD_QUEUE_NO_MEMORY = -1,
+	CVMX_CMD_QUEUE_FULL = -2,
+	CVMX_CMD_QUEUE_INVALID_PARAM = -3,
+	CVMX_CMD_QUEUE_ALREADY_SETUP = -4,
+} cvmx_cmd_queue_result_t;
+
+typedef struct {
+	/* First 64-bit word: */
+	u64 fpa_pool : 16;
+	u64 base_paddr : 48;
+	s32 index;
+	u16 max_depth;
+	u16 pool_size_m1;
+} __cvmx_cmd_queue_state_t;
+
+/**
+ * command-queue locking uses a fair ticket spinlock algo,
+ * with 64-bit tickets for endianness-neutrality and
+ * counter overflow protection.
+ * Lock is free when both counters are of equal value.
+ */
+typedef struct {
+	u64 ticket;
+	u64 now_serving;
+} __cvmx_cmd_queue_lock_t;
+
+/**
+ * @INTERNAL
+ * This structure contains the global state of all command queues.
+ * It is stored in a bootmem named block and shared by all
+ * applications running on Octeon. Tickets are stored in a different
+ * cache line that queue information to reduce the contention on the
+ * ll/sc used to get a ticket. If this is not the case, the update
+ * of queue state causes the ll/sc to fail quite often.
+ */
+typedef struct {
+	__cvmx_cmd_queue_lock_t lock[(CVMX_CMD_QUEUE_END >> 16) * 256];
+	__cvmx_cmd_queue_state_t state[(CVMX_CMD_QUEUE_END >> 16) * 256];
+} __cvmx_cmd_queue_all_state_t;
+
+extern __cvmx_cmd_queue_all_state_t *__cvmx_cmd_queue_state_ptrs[CVMX_MAX_NODES];
+
+/**
+ * @INTERNAL
+ * Internal function to handle the corner cases
+ * of adding command words to a queue when the current
+ * block is getting full.
+ */
+cvmx_cmd_queue_result_t __cvmx_cmd_queue_write_raw(cvmx_cmd_queue_id_t queue_id,
+						   __cvmx_cmd_queue_state_t *qptr, int cmd_count,
+						   const u64 *cmds);
+
+/**
+ * Initialize a command queue for use. The initial FPA buffer is
+ * allocated and the hardware unit is configured to point to the
+ * new command queue.
+ *
+ * @param queue_id  Hardware command queue to initialize.
+ * @param max_depth Maximum outstanding commands that can be queued.
+ * @param fpa_pool  FPA pool the command queues should come from.
+ * @param pool_size Size of each buffer in the FPA pool (bytes)
+ *
+ * @return CVMX_CMD_QUEUE_SUCCESS or a failure code
+ */
+cvmx_cmd_queue_result_t cvmx_cmd_queue_initialize(cvmx_cmd_queue_id_t queue_id, int max_depth,
+						  int fpa_pool, int pool_size);
+
+/**
+ * Shutdown a queue a free it's command buffers to the FPA. The
+ * hardware connected to the queue must be stopped before this
+ * function is called.
+ *
+ * @param queue_id Queue to shutdown
+ *
+ * @return CVMX_CMD_QUEUE_SUCCESS or a failure code
+ */
+cvmx_cmd_queue_result_t cvmx_cmd_queue_shutdown(cvmx_cmd_queue_id_t queue_id);
+
+/**
+ * Return the number of command words pending in the queue. This
+ * function may be relatively slow for some hardware units.
+ *
+ * @param queue_id Hardware command queue to query
+ *
+ * @return Number of outstanding commands
+ */
+int cvmx_cmd_queue_length(cvmx_cmd_queue_id_t queue_id);
+
+/**
+ * Return the command buffer to be written to. The purpose of this
+ * function is to allow CVMX routine access to the low level buffer
+ * for initial hardware setup. User applications should not call this
+ * function directly.
+ *
+ * @param queue_id Command queue to query
+ *
+ * @return Command buffer or NULL on failure
+ */
+void *cvmx_cmd_queue_buffer(cvmx_cmd_queue_id_t queue_id);
+
+/**
+ * @INTERNAL
+ * Retrieve or allocate command queue state named block
+ */
+cvmx_cmd_queue_result_t __cvmx_cmd_queue_init_state_ptr(unsigned int node);
+
+/**
+ * @INTERNAL
+ * Get the index into the state arrays for the supplied queue id.
+ *
+ * @param queue_id Queue ID to get an index for
+ *
+ * @return Index into the state arrays
+ */
+static inline unsigned int __cvmx_cmd_queue_get_index(cvmx_cmd_queue_id_t queue_id)
+{
+	/* Warning: This code currently only works with devices that have 256
+	 * queues or less.  Devices with more than 16 queues are laid out in
+	 * memory to allow cores quick access to every 16th queue. This reduces
+	 * cache thrashing when you are running 16 queues per port to support
+	 * lockless operation
+	 */
+	unsigned int unit = (queue_id >> 16) & 0xff;
+	unsigned int q = (queue_id >> 4) & 0xf;
+	unsigned int core = queue_id & 0xf;
+
+	return (unit << 8) | (core << 4) | q;
+}
+
+static inline int __cvmx_cmd_queue_get_node(cvmx_cmd_queue_id_t queue_id)
+{
+	unsigned int node = queue_id >> 24;
+	return node;
+}
+
+/**
+ * @INTERNAL
+ * Lock the supplied queue so nobody else is updating it at the same
+ * time as us.
+ *
+ * @param queue_id Queue ID to lock
+ *
+ */
+static inline void __cvmx_cmd_queue_lock(cvmx_cmd_queue_id_t queue_id)
+{
+}
+
+/**
+ * @INTERNAL
+ * Unlock the queue, flushing all writes.
+ *
+ * @param queue_id Queue ID to lock
+ *
+ */
+static inline void __cvmx_cmd_queue_unlock(cvmx_cmd_queue_id_t queue_id)
+{
+	CVMX_SYNCWS; /* nudge out the unlock. */
+}
+
+/**
+ * @INTERNAL
+ * Initialize a command-queue lock to "unlocked" state.
+ */
+static inline void __cvmx_cmd_queue_lock_init(cvmx_cmd_queue_id_t queue_id)
+{
+	unsigned int index = __cvmx_cmd_queue_get_index(queue_id);
+	unsigned int node = __cvmx_cmd_queue_get_node(queue_id);
+
+	__cvmx_cmd_queue_state_ptrs[node]->lock[index] = (__cvmx_cmd_queue_lock_t){ 0, 0 };
+	CVMX_SYNCWS;
+}
+
+/**
+ * @INTERNAL
+ * Get the queue state structure for the given queue id
+ *
+ * @param queue_id Queue id to get
+ *
+ * @return Queue structure or NULL on failure
+ */
+static inline __cvmx_cmd_queue_state_t *__cvmx_cmd_queue_get_state(cvmx_cmd_queue_id_t queue_id)
+{
+	unsigned int index;
+	unsigned int node;
+	__cvmx_cmd_queue_state_t *qptr;
+
+	node = __cvmx_cmd_queue_get_node(queue_id);
+	index = __cvmx_cmd_queue_get_index(queue_id);
+
+	if (cvmx_unlikely(!__cvmx_cmd_queue_state_ptrs[node]))
+		__cvmx_cmd_queue_init_state_ptr(node);
+
+	qptr = &__cvmx_cmd_queue_state_ptrs[node]->state[index];
+	return qptr;
+}
+
+/**
+ * Write an arbitrary number of command words to a command queue.
+ * This is a generic function; the fixed number of command word
+ * functions yield higher performance.
+ *
+ * @param queue_id  Hardware command queue to write to
+ * @param use_locking
+ *                  Use internal locking to ensure exclusive access for queue
+ *                  updates. If you don't use this locking you must ensure
+ *                  exclusivity some other way. Locking is strongly recommended.
+ * @param cmd_count Number of command words to write
+ * @param cmds      Array of commands to write
+ *
+ * @return CVMX_CMD_QUEUE_SUCCESS or a failure code
+ */
+static inline cvmx_cmd_queue_result_t
+cvmx_cmd_queue_write(cvmx_cmd_queue_id_t queue_id, bool use_locking, int cmd_count, const u64 *cmds)
+{
+	cvmx_cmd_queue_result_t ret = CVMX_CMD_QUEUE_SUCCESS;
+	u64 *cmd_ptr;
+
+	__cvmx_cmd_queue_state_t *qptr = __cvmx_cmd_queue_get_state(queue_id);
+
+	/* Make sure nobody else is updating the same queue */
+	if (cvmx_likely(use_locking))
+		__cvmx_cmd_queue_lock(queue_id);
+
+	/* Most of the time there is lots of free words in current block */
+	if (cvmx_unlikely((qptr->index + cmd_count) >= qptr->pool_size_m1)) {
+		/* The rare case when nearing end of block */
+		ret = __cvmx_cmd_queue_write_raw(queue_id, qptr, cmd_count, cmds);
+	} else {
+		cmd_ptr = (u64 *)cvmx_phys_to_ptr((u64)qptr->base_paddr);
+		/* Loop easy for compiler to unroll for the likely case */
+		while (cmd_count > 0) {
+			cmd_ptr[qptr->index++] = *cmds++;
+			cmd_count--;
+		}
+	}
+
+	/* All updates are complete. Release the lock and return */
+	if (cvmx_likely(use_locking))
+		__cvmx_cmd_queue_unlock(queue_id);
+	else
+		CVMX_SYNCWS;
+
+	return ret;
+}
+
+/**
+ * Simple function to write two command words to a command queue.
+ *
+ * @param queue_id Hardware command queue to write to
+ * @param use_locking
+ *                 Use internal locking to ensure exclusive access for queue
+ *                 updates. If you don't use this locking you must ensure
+ *                 exclusivity some other way. Locking is strongly recommended.
+ * @param cmd1     Command
+ * @param cmd2     Command
+ *
+ * @return CVMX_CMD_QUEUE_SUCCESS or a failure code
+ */
+static inline cvmx_cmd_queue_result_t cvmx_cmd_queue_write2(cvmx_cmd_queue_id_t queue_id,
+							    bool use_locking, u64 cmd1, u64 cmd2)
+{
+	cvmx_cmd_queue_result_t ret = CVMX_CMD_QUEUE_SUCCESS;
+	u64 *cmd_ptr;
+
+	__cvmx_cmd_queue_state_t *qptr = __cvmx_cmd_queue_get_state(queue_id);
+
+	/* Make sure nobody else is updating the same queue */
+	if (cvmx_likely(use_locking))
+		__cvmx_cmd_queue_lock(queue_id);
+
+	if (cvmx_unlikely((qptr->index + 2) >= qptr->pool_size_m1)) {
+		/* The rare case when nearing end of block */
+		u64 cmds[2];
+
+		cmds[0] = cmd1;
+		cmds[1] = cmd2;
+		ret = __cvmx_cmd_queue_write_raw(queue_id, qptr, 2, cmds);
+	} else {
+		/* Likely case to work fast */
+		cmd_ptr = (u64 *)cvmx_phys_to_ptr((u64)qptr->base_paddr);
+		cmd_ptr += qptr->index;
+		qptr->index += 2;
+		cmd_ptr[0] = cmd1;
+		cmd_ptr[1] = cmd2;
+	}
+
+	/* All updates are complete. Release the lock and return */
+	if (cvmx_likely(use_locking))
+		__cvmx_cmd_queue_unlock(queue_id);
+	else
+		CVMX_SYNCWS;
+
+	return ret;
+}
+
+/**
+ * Simple function to write three command words to a command queue.
+ *
+ * @param queue_id Hardware command queue to write to
+ * @param use_locking
+ *                 Use internal locking to ensure exclusive access for queue
+ *                 updates. If you don't use this locking you must ensure
+ *                 exclusivity some other way. Locking is strongly recommended.
+ * @param cmd1     Command
+ * @param cmd2     Command
+ * @param cmd3     Command
+ *
+ * @return CVMX_CMD_QUEUE_SUCCESS or a failure code
+ */
+static inline cvmx_cmd_queue_result_t
+cvmx_cmd_queue_write3(cvmx_cmd_queue_id_t queue_id, bool use_locking, u64 cmd1, u64 cmd2, u64 cmd3)
+{
+	cvmx_cmd_queue_result_t ret = CVMX_CMD_QUEUE_SUCCESS;
+	__cvmx_cmd_queue_state_t *qptr = __cvmx_cmd_queue_get_state(queue_id);
+	u64 *cmd_ptr;
+
+	/* Make sure nobody else is updating the same queue */
+	if (cvmx_likely(use_locking))
+		__cvmx_cmd_queue_lock(queue_id);
+
+	if (cvmx_unlikely((qptr->index + 3) >= qptr->pool_size_m1)) {
+		/* Most of the time there is lots of free words in current block */
+		u64 cmds[3];
+
+		cmds[0] = cmd1;
+		cmds[1] = cmd2;
+		cmds[2] = cmd3;
+		ret = __cvmx_cmd_queue_write_raw(queue_id, qptr, 3, cmds);
+	} else {
+		cmd_ptr = (u64 *)cvmx_phys_to_ptr((u64)qptr->base_paddr);
+		cmd_ptr += qptr->index;
+		qptr->index += 3;
+		cmd_ptr[0] = cmd1;
+		cmd_ptr[1] = cmd2;
+		cmd_ptr[2] = cmd3;
+	}
+
+	/* All updates are complete. Release the lock and return */
+	if (cvmx_likely(use_locking))
+		__cvmx_cmd_queue_unlock(queue_id);
+	else
+		CVMX_SYNCWS;
+
+	return ret;
+}
+
+#endif /* __CVMX_CMD_QUEUE_H__ */
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-csr-enums.h b/arch/mips/mach-octeon/include/mach/cvmx-csr-enums.h
new file mode 100644
index 0000000000..a8625b4228
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-csr-enums.h
@@ -0,0 +1,87 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ *
+ * Definitions for enumerations used with Octeon CSRs.
+ */
+
+#ifndef __CVMX_CSR_ENUMS_H__
+#define __CVMX_CSR_ENUMS_H__
+
+typedef enum {
+	CVMX_IPD_OPC_MODE_STT = 0LL,
+	CVMX_IPD_OPC_MODE_STF = 1LL,
+	CVMX_IPD_OPC_MODE_STF1_STT = 2LL,
+	CVMX_IPD_OPC_MODE_STF2_STT = 3LL
+} cvmx_ipd_mode_t;
+
+/**
+ * Enumeration representing the amount of packet processing
+ * and validation performed by the input hardware.
+ */
+typedef enum {
+	CVMX_PIP_PORT_CFG_MODE_NONE = 0ull,
+	CVMX_PIP_PORT_CFG_MODE_SKIPL2 = 1ull,
+	CVMX_PIP_PORT_CFG_MODE_SKIPIP = 2ull
+} cvmx_pip_port_parse_mode_t;
+
+/**
+ * This enumeration controls how a QoS watcher matches a packet.
+ *
+ * @deprecated  This enumeration was used with cvmx_pip_config_watcher which has
+ *              been deprecated.
+ */
+typedef enum {
+	CVMX_PIP_QOS_WATCH_DISABLE = 0ull,
+	CVMX_PIP_QOS_WATCH_PROTNH = 1ull,
+	CVMX_PIP_QOS_WATCH_TCP = 2ull,
+	CVMX_PIP_QOS_WATCH_UDP = 3ull
+} cvmx_pip_qos_watch_types;
+
+/**
+ * This enumeration is used in PIP tag config to control how
+ * POW tags are generated by the hardware.
+ */
+typedef enum {
+	CVMX_PIP_TAG_MODE_TUPLE = 0ull,
+	CVMX_PIP_TAG_MODE_MASK = 1ull,
+	CVMX_PIP_TAG_MODE_IP_OR_MASK = 2ull,
+	CVMX_PIP_TAG_MODE_TUPLE_XOR_MASK = 3ull
+} cvmx_pip_tag_mode_t;
+
+/**
+ * Tag type definitions
+ */
+typedef enum {
+	CVMX_POW_TAG_TYPE_ORDERED = 0L,
+	CVMX_POW_TAG_TYPE_ATOMIC = 1L,
+	CVMX_POW_TAG_TYPE_NULL = 2L,
+	CVMX_POW_TAG_TYPE_NULL_NULL = 3L
+} cvmx_pow_tag_type_t;
+
+/**
+ * LCR bits 0 and 1 control the number of bits per character. See the following table for encodings:
+ *
+ * - 00 = 5 bits (bits 0-4 sent)
+ * - 01 = 6 bits (bits 0-5 sent)
+ * - 10 = 7 bits (bits 0-6 sent)
+ * - 11 = 8 bits (all bits sent)
+ */
+typedef enum {
+	CVMX_UART_BITS5 = 0,
+	CVMX_UART_BITS6 = 1,
+	CVMX_UART_BITS7 = 2,
+	CVMX_UART_BITS8 = 3
+} cvmx_uart_bits_t;
+
+typedef enum {
+	CVMX_UART_IID_NONE = 1,
+	CVMX_UART_IID_RX_ERROR = 6,
+	CVMX_UART_IID_RX_DATA = 4,
+	CVMX_UART_IID_RX_TIMEOUT = 12,
+	CVMX_UART_IID_TX_EMPTY = 2,
+	CVMX_UART_IID_MODEM = 0,
+	CVMX_UART_IID_BUSY = 7
+} cvmx_uart_iid_t;
+
+#endif /* __CVMX_CSR_ENUMS_H__ */
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-csr.h b/arch/mips/mach-octeon/include/mach/cvmx-csr.h
new file mode 100644
index 0000000000..730d54bb92
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-csr.h
@@ -0,0 +1,78 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ *
+ * Configuration and status register (CSR) address and type definitions for
+ * Octoen.
+ */
+
+#ifndef __CVMX_CSR_H__
+#define __CVMX_CSR_H__
+
+#include "cvmx-csr-enums.h"
+#include "cvmx-pip-defs.h"
+
+typedef cvmx_pip_prt_cfgx_t cvmx_pip_port_cfg_t;
+
+/* The CSRs for bootbus region zero used to be independent of the
+    other 1-7. As of SDK 1.7.0 these were combined. These macros
+    are for backwards compactability */
+#define CVMX_MIO_BOOT_REG_CFG0 CVMX_MIO_BOOT_REG_CFGX(0)
+#define CVMX_MIO_BOOT_REG_TIM0 CVMX_MIO_BOOT_REG_TIMX(0)
+
+/* The CN3XXX and CN58XX chips used to not have a LMC number
+    passed to the address macros. These are here to supply backwards
+    compatibility with old code. Code should really use the new addresses
+    with bus arguments for support on other chips */
+#define CVMX_LMC_BIST_CTL	  CVMX_LMCX_BIST_CTL(0)
+#define CVMX_LMC_BIST_RESULT	  CVMX_LMCX_BIST_RESULT(0)
+#define CVMX_LMC_COMP_CTL	  CVMX_LMCX_COMP_CTL(0)
+#define CVMX_LMC_CTL		  CVMX_LMCX_CTL(0)
+#define CVMX_LMC_CTL1		  CVMX_LMCX_CTL1(0)
+#define CVMX_LMC_DCLK_CNT_HI	  CVMX_LMCX_DCLK_CNT_HI(0)
+#define CVMX_LMC_DCLK_CNT_LO	  CVMX_LMCX_DCLK_CNT_LO(0)
+#define CVMX_LMC_DCLK_CTL	  CVMX_LMCX_DCLK_CTL(0)
+#define CVMX_LMC_DDR2_CTL	  CVMX_LMCX_DDR2_CTL(0)
+#define CVMX_LMC_DELAY_CFG	  CVMX_LMCX_DELAY_CFG(0)
+#define CVMX_LMC_DLL_CTL	  CVMX_LMCX_DLL_CTL(0)
+#define CVMX_LMC_DUAL_MEMCFG	  CVMX_LMCX_DUAL_MEMCFG(0)
+#define CVMX_LMC_ECC_SYND	  CVMX_LMCX_ECC_SYND(0)
+#define CVMX_LMC_FADR		  CVMX_LMCX_FADR(0)
+#define CVMX_LMC_IFB_CNT_HI	  CVMX_LMCX_IFB_CNT_HI(0)
+#define CVMX_LMC_IFB_CNT_LO	  CVMX_LMCX_IFB_CNT_LO(0)
+#define CVMX_LMC_MEM_CFG0	  CVMX_LMCX_MEM_CFG0(0)
+#define CVMX_LMC_MEM_CFG1	  CVMX_LMCX_MEM_CFG1(0)
+#define CVMX_LMC_OPS_CNT_HI	  CVMX_LMCX_OPS_CNT_HI(0)
+#define CVMX_LMC_OPS_CNT_LO	  CVMX_LMCX_OPS_CNT_LO(0)
+#define CVMX_LMC_PLL_BWCTL	  CVMX_LMCX_PLL_BWCTL(0)
+#define CVMX_LMC_PLL_CTL	  CVMX_LMCX_PLL_CTL(0)
+#define CVMX_LMC_PLL_STATUS	  CVMX_LMCX_PLL_STATUS(0)
+#define CVMX_LMC_READ_LEVEL_CTL	  CVMX_LMCX_READ_LEVEL_CTL(0)
+#define CVMX_LMC_READ_LEVEL_DBG	  CVMX_LMCX_READ_LEVEL_DBG(0)
+#define CVMX_LMC_READ_LEVEL_RANKX CVMX_LMCX_READ_LEVEL_RANKX(0)
+#define CVMX_LMC_RODT_COMP_CTL	  CVMX_LMCX_RODT_COMP_CTL(0)
+#define CVMX_LMC_RODT_CTL	  CVMX_LMCX_RODT_CTL(0)
+#define CVMX_LMC_WODT_CTL	  CVMX_LMCX_WODT_CTL0(0)
+#define CVMX_LMC_WODT_CTL0	  CVMX_LMCX_WODT_CTL0(0)
+#define CVMX_LMC_WODT_CTL1	  CVMX_LMCX_WODT_CTL1(0)
+
+/* The CN3XXX and CN58XX chips used to not have a TWSI bus number
+    passed to the address macros. These are here to supply backwards
+    compatibility with old code. Code should really use the new addresses
+    with bus arguments for support on other chips */
+#define CVMX_MIO_TWS_INT	 CVMX_MIO_TWSX_INT(0)
+#define CVMX_MIO_TWS_SW_TWSI	 CVMX_MIO_TWSX_SW_TWSI(0)
+#define CVMX_MIO_TWS_SW_TWSI_EXT CVMX_MIO_TWSX_SW_TWSI_EXT(0)
+#define CVMX_MIO_TWS_TWSI_SW	 CVMX_MIO_TWSX_TWSI_SW(0)
+
+/* The CN3XXX and CN58XX chips used to not have a SMI/MDIO bus number
+    passed to the address macros. These are here to supply backwards
+    compatibility with old code. Code should really use the new addresses
+    with bus arguments for support on other chips */
+#define CVMX_SMI_CLK	CVMX_SMIX_CLK(0)
+#define CVMX_SMI_CMD	CVMX_SMIX_CMD(0)
+#define CVMX_SMI_EN	CVMX_SMIX_EN(0)
+#define CVMX_SMI_RD_DAT CVMX_SMIX_RD_DAT(0)
+#define CVMX_SMI_WR_DAT CVMX_SMIX_WR_DAT(0)
+
+#endif /* __CVMX_CSR_H__ */
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-error.h b/arch/mips/mach-octeon/include/mach/cvmx-error.h
new file mode 100644
index 0000000000..9a13ed4224
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-error.h
@@ -0,0 +1,456 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ *
+ * Interface to the Octeon extended error status.
+ */
+
+#ifndef __CVMX_ERROR_H__
+#define __CVMX_ERROR_H__
+
+/**
+ * There are generally many error status bits associated with a
+ * single logical group. The enumeration below is used to
+ * communicate high level groups to the error infastructure so
+ * error status bits can be enable or disabled in large groups.
+ */
+typedef enum {
+	CVMX_ERROR_GROUP_INTERNAL,
+	CVMX_ERROR_GROUP_L2C,
+	CVMX_ERROR_GROUP_ETHERNET,
+	CVMX_ERROR_GROUP_MGMT_PORT,
+	CVMX_ERROR_GROUP_PCI,
+	CVMX_ERROR_GROUP_SRIO,
+	CVMX_ERROR_GROUP_USB,
+	CVMX_ERROR_GROUP_LMC,
+	CVMX_ERROR_GROUP_ILK,
+	CVMX_ERROR_GROUP_DFM,
+	CVMX_ERROR_GROUP_ILA,
+} cvmx_error_group_t;
+
+/**
+ * Flags representing special handling for some error registers.
+ * These flags are passed to cvmx_error_initialize() to control
+ * the handling of bits where the same flags were passed to the
+ * added cvmx_error_info_t.
+ */
+typedef enum {
+	CVMX_ERROR_TYPE_NONE = 0,
+	CVMX_ERROR_TYPE_SBE = 1 << 0,
+	CVMX_ERROR_TYPE_DBE = 1 << 1,
+} cvmx_error_type_t;
+
+/**
+ * When registering for interest in an error status register, the
+ * type of the register needs to be known by cvmx-error. Most
+ * registers are either IO64 or IO32, but some blocks contain
+ * registers that can't be directly accessed. A good example of
+ * would be PCIe extended error state stored in config space.
+ */
+typedef enum {
+	__CVMX_ERROR_REGISTER_NONE,
+	CVMX_ERROR_REGISTER_IO64,
+	CVMX_ERROR_REGISTER_IO32,
+	CVMX_ERROR_REGISTER_PCICONFIG,
+	CVMX_ERROR_REGISTER_SRIOMAINT,
+} cvmx_error_register_t;
+
+struct cvmx_error_info;
+/**
+ * Error handling functions must have the following prototype.
+ */
+typedef int (*cvmx_error_func_t)(const struct cvmx_error_info *info);
+
+/**
+ * This structure is passed to all error handling functions.
+ */
+typedef struct cvmx_error_info {
+	cvmx_error_register_t reg_type;
+	u64 status_addr;
+	u64 status_mask;
+	u64 enable_addr;
+	u64 enable_mask;
+	cvmx_error_type_t flags;
+	cvmx_error_group_t group;
+	int group_index;
+	cvmx_error_func_t func;
+	u64 user_info;
+	struct {
+		cvmx_error_register_t reg_type;
+		u64 status_addr;
+		u64 status_mask;
+	} parent;
+} cvmx_error_info_t;
+
+/**
+ * Initialize the error status system. This should be called once
+ * before any other functions are called. This function adds default
+ * handlers for most all error events but does not enable them. Later
+ * calls to cvmx_error_enable() are needed.
+ *
+ * @param flags  Optional flags.
+ *
+ * @return Zero on success, negative on failure.
+ */
+int cvmx_error_initialize(void);
+
+/**
+ * Poll the error status registers and call the appropriate error
+ * handlers. This should be called in the RSL interrupt handler
+ * for your application or operating system.
+ *
+ * @return Number of error handlers called. Zero means this call
+ *         found no errors and was spurious.
+ */
+int cvmx_error_poll(void);
+
+/**
+ * Register to be called when an error status bit is set. Most users
+ * will not need to call this function as cvmx_error_initialize()
+ * registers default handlers for most error conditions. This function
+ * is normally used to add more handlers without changing the existing
+ * handlers.
+ *
+ * @param new_info Information about the handler for a error register. The
+ *                 structure passed is copied and can be destroyed after the
+ *                 call. All members of the structure must be populated, even the
+ *                 parent information.
+ *
+ * @return Zero on success, negative on failure.
+ */
+int cvmx_error_add(const cvmx_error_info_t *new_info);
+
+/**
+ * Remove all handlers for a status register and mask. Normally
+ * this function should not be called. Instead a new handler should be
+ * installed to replace the existing handler. In the even that all
+ * reporting of a error bit should be removed, then use this
+ * function.
+ *
+ * @param reg_type Type of the status register to remove
+ * @param status_addr
+ *                 Status register to remove.
+ * @param status_mask
+ *                 All handlers for this status register with this mask will be
+ *                 removed.
+ * @param old_info If not NULL, this is filled with information about the handler
+ *                 that was removed.
+ *
+ * @return Zero on success, negative on failure (not found).
+ */
+int cvmx_error_remove(cvmx_error_register_t reg_type, u64 status_addr, u64 status_mask,
+		      cvmx_error_info_t *old_info);
+
+/**
+ * Change the function and user_info for an existing error status
+ * register. This function should be used to replace the default
+ * handler with an application specific version as needed.
+ *
+ * @param reg_type Type of the status register to change
+ * @param status_addr
+ *                 Status register to change.
+ * @param status_mask
+ *                 All handlers for this status register with this mask will be
+ *                 changed.
+ * @param new_func New function to use to handle the error status
+ * @param new_user_info
+ *                 New user info parameter for the function
+ * @param old_func If not NULL, the old function is returned. Useful for restoring
+ *                 the old handler.
+ * @param old_user_info
+ *                 If not NULL, the old user info parameter.
+ *
+ * @return Zero on success, negative on failure
+ */
+int cvmx_error_change_handler(cvmx_error_register_t reg_type, u64 status_addr, u64 status_mask,
+			      cvmx_error_func_t new_func, u64 new_user_info,
+			      cvmx_error_func_t *old_func, u64 *old_user_info);
+
+/**
+ * Enable all error registers for a logical group. This should be
+ * called whenever a logical group is brought online.
+ *
+ * @param group  Logical group to enable
+ * @param group_index
+ *               Index for the group as defined in the cvmx_error_group_t
+ *               comments.
+ *
+ * @return Zero on success, negative on failure.
+ */
+/*
+ * Rather than conditionalize the calls throughout the executive to not enable
+ * interrupts in Uboot, simply make the enable function do nothing
+ */
+static inline int cvmx_error_enable_group(cvmx_error_group_t group, int group_index)
+{
+	return 0;
+}
+
+/**
+ * Disable all error registers for a logical group. This should be
+ * called whenever a logical group is brought offline. Many blocks
+ * will report spurious errors when offline unless this function
+ * is called.
+ *
+ * @param group  Logical group to disable
+ * @param group_index
+ *               Index for the group as defined in the cvmx_error_group_t
+ *               comments.
+ *
+ * @return Zero on success, negative on failure.
+ */
+/*
+ * Rather than conditionalize the calls throughout the executive to not disable
+ * interrupts in Uboot, simply make the enable function do nothing
+ */
+static inline int cvmx_error_disable_group(cvmx_error_group_t group, int group_index)
+{
+	return 0;
+}
+
+/**
+ * Enable all handlers for a specific status register mask.
+ *
+ * @param reg_type Type of the status register
+ * @param status_addr
+ *                 Status register address
+ * @param status_mask
+ *                 All handlers for this status register with this mask will be
+ *                 enabled.
+ *
+ * @return Zero on success, negative on failure.
+ */
+int cvmx_error_enable(cvmx_error_register_t reg_type, u64 status_addr, u64 status_mask);
+
+/**
+ * Disable all handlers for a specific status register and mask.
+ *
+ * @param reg_type Type of the status register
+ * @param status_addr
+ *                 Status register address
+ * @param status_mask
+ *                 All handlers for this status register with this mask will be
+ *                 disabled.
+ *
+ * @return Zero on success, negative on failure.
+ */
+int cvmx_error_disable(cvmx_error_register_t reg_type, u64 status_addr, u64 status_mask);
+
+/**
+ * @INTERNAL
+ * Function for processing non leaf error status registers. This function
+ * calls all handlers for this passed register and all children linked
+ * to it.
+ *
+ * @param info   Error register to check
+ *
+ * @return Number of error status bits found or zero if no bits were set.
+ */
+int __cvmx_error_decode(const cvmx_error_info_t *info);
+
+/**
+ * @INTERNAL
+ * This error bit handler simply prints a message and clears the status bit
+ *
+ * @param info   Error register to check
+ *
+ * @return
+ */
+int __cvmx_error_display(const cvmx_error_info_t *info);
+
+/**
+ * Find the handler for a specific status register and mask
+ *
+ * @param status_addr
+ *                Status register address
+ *
+ * @return  Return the handler on success or null on failure.
+ */
+cvmx_error_info_t *cvmx_error_get_index(u64 status_addr);
+
+void __cvmx_install_gmx_error_handler_for_xaui(void);
+
+/**
+ * 78xx related
+ */
+/**
+ * Compare two INTSN values.
+ *
+ * @param key INTSN value to search for
+ * @param data current entry from the searched array
+ *
+ * @return Negative, 0 or positive when respectively key is less than,
+ *		equal or greater than data.
+ */
+int cvmx_error_intsn_cmp(const void *key, const void *data);
+
+/**
+ * @INTERNAL
+ *
+ * @param intsn   Interrupt source number to display
+ *
+ * @param node Node number
+ *
+ * @return Zero on success, -1 on error
+ */
+int cvmx_error_intsn_display_v3(int node, u32 intsn);
+
+/**
+ * Initialize the error status system for cn78xx. This should be called once
+ * before any other functions are called. This function enables the interrupts
+ * described in the array.
+ *
+ * @param node Node number
+ *
+ * @return Zero on success, negative on failure.
+ */
+int cvmx_error_initialize_cn78xx(int node);
+
+/**
+ * Enable interrupt for a specific INTSN.
+ *
+ * @param node Node number
+ * @param intsn Interrupt source number
+ *
+ * @return Zero on success, negative on failure.
+ */
+int cvmx_error_intsn_enable_v3(int node, u32 intsn);
+
+/**
+ * Disable interrupt for a specific INTSN.
+ *
+ * @param node Node number
+ * @param intsn Interrupt source number
+ *
+ * @return Zero on success, negative on failure.
+ */
+int cvmx_error_intsn_disable_v3(int node, u32 intsn);
+
+/**
+ * Clear interrupt for a specific INTSN.
+ *
+ * @param intsn Interrupt source number
+ *
+ * @return Zero on success, negative on failure.
+ */
+int cvmx_error_intsn_clear_v3(int node, u32 intsn);
+
+/**
+ * Enable interrupts for a specific CSR(all the bits/intsn in the csr).
+ *
+ * @param node Node number
+ * @param csr_address CSR address
+ *
+ * @return Zero on success, negative on failure.
+ */
+int cvmx_error_csr_enable_v3(int node, u64 csr_address);
+
+/**
+ * Disable interrupts for a specific CSR (all the bits/intsn in the csr).
+ *
+ * @param node Node number
+ * @param csr_address CSR address
+ *
+ * @return Zero
+ */
+int cvmx_error_csr_disable_v3(int node, u64 csr_address);
+
+/**
+ * Enable all error registers for a logical group. This should be
+ * called whenever a logical group is brought online.
+ *
+ * @param group  Logical group to enable
+ * @param xipd_port  The IPD port value
+ *
+ * @return Zero.
+ */
+int cvmx_error_enable_group_v3(cvmx_error_group_t group, int xipd_port);
+
+/**
+ * Disable all error registers for a logical group.
+ *
+ * @param group  Logical group to enable
+ * @param xipd_port  The IPD port value
+ *
+ * @return Zero.
+ */
+int cvmx_error_disable_group_v3(cvmx_error_group_t group, int xipd_port);
+
+/**
+ * Enable all error registers for a specific category in a logical group.
+ * This should be called whenever a logical group is brought online.
+ *
+ * @param group  Logical group to enable
+ * @param type   Category in a logical group to enable
+ * @param xipd_port  The IPD port value
+ *
+ * @return Zero.
+ */
+int cvmx_error_enable_group_type_v3(cvmx_error_group_t group, cvmx_error_type_t type,
+				    int xipd_port);
+
+/**
+ * Disable all error registers for a specific category in a logical group.
+ * This should be called whenever a logical group is brought online.
+ *
+ * @param group  Logical group to disable
+ * @param type   Category in a logical group to disable
+ * @param xipd_port  The IPD port value
+ *
+ * @return Zero.
+ */
+int cvmx_error_disable_group_type_v3(cvmx_error_group_t group, cvmx_error_type_t type,
+				     int xipd_port);
+
+/**
+ * Clear all error registers for a logical group.
+ *
+ * @param group  Logical group to disable
+ * @param xipd_port  The IPD port value
+ *
+ * @return Zero.
+ */
+int cvmx_error_clear_group_v3(cvmx_error_group_t group, int xipd_port);
+
+/**
+ * Enable all error registers for a particular category.
+ *
+ * @param node  CCPI node
+ * @param type  category to enable
+ *
+ *@return Zero.
+ */
+int cvmx_error_enable_type_v3(int node, cvmx_error_type_t type);
+
+/**
+ * Disable all error registers for a particular category.
+ *
+ * @param node  CCPI node
+ * @param type  category to disable
+ *
+ *@return Zero.
+ */
+int cvmx_error_disable_type_v3(int node, cvmx_error_type_t type);
+
+void cvmx_octeon_hang(void) __attribute__((__noreturn__));
+
+/**
+ * @INTERNAL
+ *
+ * Process L2C single and multi-bit ECC errors
+ *
+ */
+int __cvmx_cn7xxx_l2c_l2d_ecc_error_display(int node, int intsn);
+
+/**
+ * Handle L2 cache TAG ECC errors and noway errors
+ *
+ * @param	CCPI node
+ * @param	intsn	intsn from error array.
+ * @param	remote	true for remote node (cn78xx only)
+ *
+ * @return	1 if handled, 0 if not handled
+ */
+int __cvmx_cn7xxx_l2c_tag_error_display(int node, int intsn, bool remote);
+
+#endif
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-fpa.h b/arch/mips/mach-octeon/include/mach/cvmx-fpa.h
new file mode 100644
index 0000000000..297fb3f4a2
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-fpa.h
@@ -0,0 +1,217 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ *
+ * Interface to the hardware Free Pool Allocator.
+ */
+
+#ifndef __CVMX_FPA_H__
+#define __CVMX_FPA_H__
+
+#include "cvmx-scratch.h"
+#include "cvmx-fpa-defs.h"
+#include "cvmx-fpa1.h"
+#include "cvmx-fpa3.h"
+
+#define CVMX_FPA_MIN_BLOCK_SIZE 128
+#define CVMX_FPA_ALIGNMENT	128
+#define CVMX_FPA_POOL_NAME_LEN	16
+
+/* On CN78XX in backward-compatible mode, pool is mapped to AURA */
+#define CVMX_FPA_NUM_POOLS                                                                         \
+	(octeon_has_feature(OCTEON_FEATURE_FPA3) ? cvmx_fpa3_num_auras() : CVMX_FPA1_NUM_POOLS)
+
+/**
+ * Structure to store FPA pool configuration parameters.
+ */
+struct cvmx_fpa_pool_config {
+	s64 pool_num;
+	u64 buffer_size;
+	u64 buffer_count;
+};
+
+typedef struct cvmx_fpa_pool_config cvmx_fpa_pool_config_t;
+
+/**
+ * Return the name of the pool
+ *
+ * @param pool_num   Pool to get the name of
+ * @return The name
+ */
+const char *cvmx_fpa_get_name(int pool_num);
+
+/**
+ * Initialize FPA per node
+ */
+int cvmx_fpa_global_init_node(int node);
+
+/**
+ * Enable the FPA
+ */
+static inline void cvmx_fpa_enable(void)
+{
+	if (!octeon_has_feature(OCTEON_FEATURE_FPA3))
+		cvmx_fpa1_enable();
+	else
+		cvmx_fpa_global_init_node(cvmx_get_node_num());
+}
+
+/**
+ * Disable the FPA
+ */
+static inline void cvmx_fpa_disable(void)
+{
+	if (!octeon_has_feature(OCTEON_FEATURE_FPA3))
+		cvmx_fpa1_disable();
+	/* FPA3 does not have a disable function */
+}
+
+/**
+ * @INTERNAL
+ * @deprecated OBSOLETE
+ *
+ * Kept for transition assistance only
+ */
+static inline void cvmx_fpa_global_initialize(void)
+{
+	cvmx_fpa_global_init_node(cvmx_get_node_num());
+}
+
+/**
+ * @INTERNAL
+ *
+ * Convert FPA1 style POOL into FPA3 AURA in
+ * backward compatibility mode.
+ */
+static inline cvmx_fpa3_gaura_t cvmx_fpa1_pool_to_fpa3_aura(cvmx_fpa1_pool_t pool)
+{
+	if ((octeon_has_feature(OCTEON_FEATURE_FPA3))) {
+		unsigned int node = cvmx_get_node_num();
+		cvmx_fpa3_gaura_t aura = __cvmx_fpa3_gaura(node, pool);
+		return aura;
+	}
+	return CVMX_FPA3_INVALID_GAURA;
+}
+
+/**
+ * Get a new block from the FPA
+ *
+ * @param pool   Pool to get the block from
+ * @return Pointer to the block or NULL on failure
+ */
+static inline void *cvmx_fpa_alloc(u64 pool)
+{
+	/* FPA3 is handled differently */
+	if ((octeon_has_feature(OCTEON_FEATURE_FPA3))) {
+		return cvmx_fpa3_alloc(cvmx_fpa1_pool_to_fpa3_aura(pool));
+	} else
+		return cvmx_fpa1_alloc(pool);
+}
+
+/**
+ * Asynchronously get a new block from the FPA
+ *
+ * The result of cvmx_fpa_async_alloc() may be retrieved using
+ * cvmx_fpa_async_alloc_finish().
+ *
+ * @param scr_addr Local scratch address to put response in.  This is a byte
+ *		   address but must be 8 byte aligned.
+ * @param pool      Pool to get the block from
+ */
+static inline void cvmx_fpa_async_alloc(u64 scr_addr, u64 pool)
+{
+	if ((octeon_has_feature(OCTEON_FEATURE_FPA3))) {
+		return cvmx_fpa3_async_alloc(scr_addr, cvmx_fpa1_pool_to_fpa3_aura(pool));
+	} else
+		return cvmx_fpa1_async_alloc(scr_addr, pool);
+}
+
+/**
+ * Retrieve the result of cvmx_fpa_async_alloc
+ *
+ * @param scr_addr The Local scratch address.  Must be the same value
+ * passed to cvmx_fpa_async_alloc().
+ *
+ * @param pool Pool the block came from.  Must be the same value
+ * passed to cvmx_fpa_async_alloc.
+ *
+ * @return Pointer to the block or NULL on failure
+ */
+static inline void *cvmx_fpa_async_alloc_finish(u64 scr_addr, u64 pool)
+{
+	if ((octeon_has_feature(OCTEON_FEATURE_FPA3)))
+		return cvmx_fpa3_async_alloc_finish(scr_addr, cvmx_fpa1_pool_to_fpa3_aura(pool));
+	else
+		return cvmx_fpa1_async_alloc_finish(scr_addr, pool);
+}
+
+/**
+ * Free a block allocated with a FPA pool.
+ * Does NOT provide memory ordering in cases where the memory block was
+ * modified by the core.
+ *
+ * @param ptr    Block to free
+ * @param pool   Pool to put it in
+ * @param num_cache_lines
+ *               Cache lines to invalidate
+ */
+static inline void cvmx_fpa_free_nosync(void *ptr, u64 pool, u64 num_cache_lines)
+{
+	/* FPA3 is handled differently */
+	if ((octeon_has_feature(OCTEON_FEATURE_FPA3)))
+		cvmx_fpa3_free_nosync(ptr, cvmx_fpa1_pool_to_fpa3_aura(pool), num_cache_lines);
+	else
+		cvmx_fpa1_free_nosync(ptr, pool, num_cache_lines);
+}
+
+/**
+ * Free a block allocated with a FPA pool.  Provides required memory
+ * ordering in cases where memory block was modified by core.
+ *
+ * @param ptr    Block to free
+ * @param pool   Pool to put it in
+ * @param num_cache_lines
+ *               Cache lines to invalidate
+ */
+static inline void cvmx_fpa_free(void *ptr, u64 pool, u64 num_cache_lines)
+{
+	if ((octeon_has_feature(OCTEON_FEATURE_FPA3)))
+		cvmx_fpa3_free(ptr, cvmx_fpa1_pool_to_fpa3_aura(pool), num_cache_lines);
+	else
+		cvmx_fpa1_free(ptr, pool, num_cache_lines);
+}
+
+/**
+ * Setup a FPA pool to control a new block of memory.
+ * This can only be called once per pool. Make sure proper
+ * locking enforces this.
+ *
+ * @param pool       Pool to initialize
+ * @param name       Constant character string to name this pool.
+ *                   String is not copied.
+ * @param buffer     Pointer to the block of memory to use. This must be
+ *                   accessible by all processors and external hardware.
+ * @param block_size Size for each block controlled by the FPA
+ * @param num_blocks Number of blocks
+ *
+ * @return the pool number on Success,
+ *         -1 on failure
+ */
+int cvmx_fpa_setup_pool(int pool, const char *name, void *buffer, u64 block_size, u64 num_blocks);
+
+int cvmx_fpa_shutdown_pool(int pool);
+
+/**
+ * Gets the block size of buffer in specified pool
+ * @param pool	 Pool to get the block size from
+ * @return       Size of buffer in specified pool
+ */
+unsigned int cvmx_fpa_get_block_size(int pool);
+
+int cvmx_fpa_is_pool_available(int pool_num);
+u64 cvmx_fpa_get_pool_owner(int pool_num);
+int cvmx_fpa_get_max_pools(void);
+int cvmx_fpa_get_current_count(int pool_num);
+int cvmx_fpa_validate_pool(int pool);
+
+#endif /*  __CVM_FPA_H__ */
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-fpa1.h b/arch/mips/mach-octeon/include/mach/cvmx-fpa1.h
new file mode 100644
index 0000000000..6985083a5d
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-fpa1.h
@@ -0,0 +1,196 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ *
+ * Interface to the hardware Free Pool Allocator on Octeon chips.
+ * These are the legacy models, i.e. prior to CN78XX/CN76XX.
+ */
+
+#ifndef __CVMX_FPA1_HW_H__
+#define __CVMX_FPA1_HW_H__
+
+#include "cvmx-scratch.h"
+#include "cvmx-fpa-defs.h"
+#include "cvmx-fpa3.h"
+
+/* Legacy pool range is 0..7 and 8 on CN68XX */
+typedef int cvmx_fpa1_pool_t;
+
+#define CVMX_FPA1_NUM_POOLS    8
+#define CVMX_FPA1_INVALID_POOL ((cvmx_fpa1_pool_t)-1)
+#define CVMX_FPA1_NAME_SIZE    16
+
+/**
+ * Structure describing the data format used for stores to the FPA.
+ */
+typedef union {
+	u64 u64;
+	struct {
+		u64 scraddr : 8;
+		u64 len : 8;
+		u64 did : 8;
+		u64 addr : 40;
+	} s;
+} cvmx_fpa1_iobdma_data_t;
+
+/*
+ * Allocate or reserve the specified fpa pool.
+ *
+ * @param pool	  FPA pool to allocate/reserve. If -1 it
+ *                finds an empty pool to allocate.
+ * @return        Alloctaed pool number or CVMX_FPA1_POOL_INVALID
+ *                if fails to allocate the pool
+ */
+cvmx_fpa1_pool_t cvmx_fpa1_reserve_pool(cvmx_fpa1_pool_t pool);
+
+/**
+ * Free the specified fpa pool.
+ * @param pool	   Pool to free
+ * @return         0 for success -1 failure
+ */
+int cvmx_fpa1_release_pool(cvmx_fpa1_pool_t pool);
+
+static inline void cvmx_fpa1_free(void *ptr, cvmx_fpa1_pool_t pool, u64 num_cache_lines)
+{
+	cvmx_addr_t newptr;
+
+	newptr.u64 = cvmx_ptr_to_phys(ptr);
+	newptr.sfilldidspace.didspace = CVMX_ADDR_DIDSPACE(CVMX_FULL_DID(CVMX_OCT_DID_FPA, pool));
+	/* Make sure that any previous writes to memory go out before we free
+	 * this buffer.  This also serves as a barrier to prevent GCC from
+	 * reordering operations to after the free.
+	 */
+	CVMX_SYNCWS;
+	/* value written is number of cache lines not written back */
+	cvmx_write_io(newptr.u64, num_cache_lines);
+}
+
+static inline void cvmx_fpa1_free_nosync(void *ptr, cvmx_fpa1_pool_t pool,
+					 unsigned int num_cache_lines)
+{
+	cvmx_addr_t newptr;
+
+	newptr.u64 = cvmx_ptr_to_phys(ptr);
+	newptr.sfilldidspace.didspace = CVMX_ADDR_DIDSPACE(CVMX_FULL_DID(CVMX_OCT_DID_FPA, pool));
+	/* Prevent GCC from reordering around free */
+	asm volatile("" : : : "memory");
+	/* value written is number of cache lines not written back */
+	cvmx_write_io(newptr.u64, num_cache_lines);
+}
+
+/**
+ * Enable the FPA for use. Must be performed after any CSR
+ * configuration but before any other FPA functions.
+ */
+static inline void cvmx_fpa1_enable(void)
+{
+	cvmx_fpa_ctl_status_t status;
+
+	status.u64 = csr_rd(CVMX_FPA_CTL_STATUS);
+	if (status.s.enb) {
+		/*
+		 * CN68XXP1 should not reset the FPA (doing so may break
+		 * the SSO, so we may end up enabling it more than once.
+		 * Just return and don't spew messages.
+		 */
+		return;
+	}
+
+	status.u64 = 0;
+	status.s.enb = 1;
+	csr_wr(CVMX_FPA_CTL_STATUS, status.u64);
+}
+
+/**
+ * Reset FPA to disable. Make sure buffers from all FPA pools are freed
+ * before disabling FPA.
+ */
+static inline void cvmx_fpa1_disable(void)
+{
+	cvmx_fpa_ctl_status_t status;
+
+	if (OCTEON_IS_MODEL(OCTEON_CN68XX_PASS1))
+		return;
+
+	status.u64 = csr_rd(CVMX_FPA_CTL_STATUS);
+	status.s.reset = 1;
+	csr_wr(CVMX_FPA_CTL_STATUS, status.u64);
+}
+
+static inline void *cvmx_fpa1_alloc(cvmx_fpa1_pool_t pool)
+{
+	u64 address;
+
+	for (;;) {
+		address = csr_rd(CVMX_ADDR_DID(CVMX_FULL_DID(CVMX_OCT_DID_FPA, pool)));
+		if (cvmx_likely(address)) {
+			return cvmx_phys_to_ptr(address);
+		} else {
+			if (csr_rd(CVMX_FPA_QUEX_AVAILABLE(pool)) > 0)
+				udelay(50);
+			else
+				return NULL;
+		}
+	}
+}
+
+/**
+ * Asynchronously get a new block from the FPA
+ * @INTERNAL
+ *
+ * The result of cvmx_fpa_async_alloc() may be retrieved using
+ * cvmx_fpa_async_alloc_finish().
+ *
+ * @param scr_addr Local scratch address to put response in.  This is a byte
+ *		   address but must be 8 byte aligned.
+ * @param pool      Pool to get the block from
+ */
+static inline void cvmx_fpa1_async_alloc(u64 scr_addr, cvmx_fpa1_pool_t pool)
+{
+	cvmx_fpa1_iobdma_data_t data;
+
+	/* Hardware only uses 64 bit aligned locations, so convert from byte
+	 * address to 64-bit index
+	 */
+	data.u64 = 0ull;
+	data.s.scraddr = scr_addr >> 3;
+	data.s.len = 1;
+	data.s.did = CVMX_FULL_DID(CVMX_OCT_DID_FPA, pool);
+	data.s.addr = 0;
+
+	cvmx_scratch_write64(scr_addr, 0ull);
+	CVMX_SYNCW;
+	cvmx_send_single(data.u64);
+}
+
+/**
+ * Retrieve the result of cvmx_fpa_async_alloc
+ * @INTERNAL
+ *
+ * @param scr_addr The Local scratch address.  Must be the same value
+ * passed to cvmx_fpa_async_alloc().
+ *
+ * @param pool Pool the block came from.  Must be the same value
+ * passed to cvmx_fpa_async_alloc.
+ *
+ * @return Pointer to the block or NULL on failure
+ */
+static inline void *cvmx_fpa1_async_alloc_finish(u64 scr_addr, cvmx_fpa1_pool_t pool)
+{
+	u64 address;
+
+	CVMX_SYNCIOBDMA;
+
+	address = cvmx_scratch_read64(scr_addr);
+	if (cvmx_likely(address))
+		return cvmx_phys_to_ptr(address);
+	else
+		return cvmx_fpa1_alloc(pool);
+}
+
+static inline u64 cvmx_fpa1_get_available(cvmx_fpa1_pool_t pool)
+{
+	return csr_rd(CVMX_FPA_QUEX_AVAILABLE(pool));
+}
+
+#endif /* __CVMX_FPA1_HW_H__ */
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-fpa3.h b/arch/mips/mach-octeon/include/mach/cvmx-fpa3.h
new file mode 100644
index 0000000000..229982b831
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-fpa3.h
@@ -0,0 +1,566 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ *
+ * Interface to the CN78XX Free Pool Allocator, a.k.a. FPA3
+ */
+
+#include "cvmx-address.h"
+#include "cvmx-fpa-defs.h"
+#include "cvmx-scratch.h"
+
+#ifndef __CVMX_FPA3_H__
+#define __CVMX_FPA3_H__
+
+typedef struct {
+	unsigned res0 : 6;
+	unsigned node : 2;
+	unsigned res1 : 2;
+	unsigned lpool : 6;
+	unsigned valid_magic : 16;
+} cvmx_fpa3_pool_t;
+
+typedef struct {
+	unsigned res0 : 6;
+	unsigned node : 2;
+	unsigned res1 : 6;
+	unsigned laura : 10;
+	unsigned valid_magic : 16;
+} cvmx_fpa3_gaura_t;
+
+#define CVMX_FPA3_VALID_MAGIC	0xf9a3
+#define CVMX_FPA3_INVALID_GAURA ((cvmx_fpa3_gaura_t){ 0, 0, 0, 0, 0 })
+#define CVMX_FPA3_INVALID_POOL	((cvmx_fpa3_pool_t){ 0, 0, 0, 0, 0 })
+
+static inline bool __cvmx_fpa3_aura_valid(cvmx_fpa3_gaura_t aura)
+{
+	if (aura.valid_magic != CVMX_FPA3_VALID_MAGIC)
+		return false;
+	return true;
+}
+
+static inline bool __cvmx_fpa3_pool_valid(cvmx_fpa3_pool_t pool)
+{
+	if (pool.valid_magic != CVMX_FPA3_VALID_MAGIC)
+		return false;
+	return true;
+}
+
+static inline cvmx_fpa3_gaura_t __cvmx_fpa3_gaura(int node, int laura)
+{
+	cvmx_fpa3_gaura_t aura;
+
+	if (node < 0)
+		node = cvmx_get_node_num();
+	if (laura < 0)
+		return CVMX_FPA3_INVALID_GAURA;
+
+	aura.node = node;
+	aura.laura = laura;
+	aura.valid_magic = CVMX_FPA3_VALID_MAGIC;
+	return aura;
+}
+
+static inline cvmx_fpa3_pool_t __cvmx_fpa3_pool(int node, int lpool)
+{
+	cvmx_fpa3_pool_t pool;
+
+	if (node < 0)
+		node = cvmx_get_node_num();
+	if (lpool < 0)
+		return CVMX_FPA3_INVALID_POOL;
+
+	pool.node = node;
+	pool.lpool = lpool;
+	pool.valid_magic = CVMX_FPA3_VALID_MAGIC;
+	return pool;
+}
+
+#undef CVMX_FPA3_VALID_MAGIC
+
+/**
+ * Structure describing the data format used for stores to the FPA.
+ */
+typedef union {
+	u64 u64;
+	struct {
+		u64 scraddr : 8;
+		u64 len : 8;
+		u64 did : 8;
+		u64 addr : 40;
+	} s;
+	struct {
+		u64 scraddr : 8;
+		u64 len : 8;
+		u64 did : 8;
+		u64 node : 4;
+		u64 red : 1;
+		u64 reserved2 : 9;
+		u64 aura : 10;
+		u64 reserved3 : 16;
+	} cn78xx;
+} cvmx_fpa3_iobdma_data_t;
+
+/**
+ * Struct describing load allocate operation addresses for FPA pool.
+ */
+union cvmx_fpa3_load_data {
+	u64 u64;
+	struct {
+		u64 seg : 2;
+		u64 reserved1 : 13;
+		u64 io : 1;
+		u64 did : 8;
+		u64 node : 4;
+		u64 red : 1;
+		u64 reserved2 : 9;
+		u64 aura : 10;
+		u64 reserved3 : 16;
+	};
+};
+
+typedef union cvmx_fpa3_load_data cvmx_fpa3_load_data_t;
+
+/**
+ * Struct describing store free operation addresses from FPA pool.
+ */
+union cvmx_fpa3_store_addr {
+	u64 u64;
+	struct {
+		u64 seg : 2;
+		u64 reserved1 : 13;
+		u64 io : 1;
+		u64 did : 8;
+		u64 node : 4;
+		u64 reserved2 : 10;
+		u64 aura : 10;
+		u64 fabs : 1;
+		u64 reserved3 : 3;
+		u64 dwb_count : 9;
+		u64 reserved4 : 3;
+	};
+};
+
+typedef union cvmx_fpa3_store_addr cvmx_fpa3_store_addr_t;
+
+enum cvmx_fpa3_pool_alignment_e {
+	FPA_NATURAL_ALIGNMENT,
+	FPA_OFFSET_ALIGNMENT,
+	FPA_OPAQUE_ALIGNMENT
+};
+
+#define CVMX_FPA3_AURAX_LIMIT_MAX ((1ull << 40) - 1)
+
+/**
+ * @INTERNAL
+ * Accessor functions to return number of POOLS in an FPA3
+ * depending on SoC model.
+ * The number is per-node for models supporting multi-node configurations.
+ */
+static inline int cvmx_fpa3_num_pools(void)
+{
+	if (OCTEON_IS_MODEL(OCTEON_CN78XX))
+		return 64;
+	if (OCTEON_IS_MODEL(OCTEON_CNF75XX))
+		return 32;
+	if (OCTEON_IS_MODEL(OCTEON_CN73XX))
+		return 32;
+	printf("ERROR: %s: Unknowm model\n", __func__);
+	return -1;
+}
+
+/**
+ * @INTERNAL
+ * Accessor functions to return number of AURAS in an FPA3
+ * depending on SoC model.
+ * The number is per-node for models supporting multi-node configurations.
+ */
+static inline int cvmx_fpa3_num_auras(void)
+{
+	if (OCTEON_IS_MODEL(OCTEON_CN78XX))
+		return 1024;
+	if (OCTEON_IS_MODEL(OCTEON_CNF75XX))
+		return 512;
+	if (OCTEON_IS_MODEL(OCTEON_CN73XX))
+		return 512;
+	printf("ERROR: %s: Unknowm model\n", __func__);
+	return -1;
+}
+
+/**
+ * Get the FPA3 POOL underneath FPA3 AURA, containing all its buffers
+ *
+ */
+static inline cvmx_fpa3_pool_t cvmx_fpa3_aura_to_pool(cvmx_fpa3_gaura_t aura)
+{
+	cvmx_fpa3_pool_t pool;
+	cvmx_fpa_aurax_pool_t aurax_pool;
+
+	aurax_pool.u64 = cvmx_read_csr_node(aura.node, CVMX_FPA_AURAX_POOL(aura.laura));
+
+	pool = __cvmx_fpa3_pool(aura.node, aurax_pool.s.pool);
+	return pool;
+}
+
+/**
+ * Get a new block from the FPA pool
+ *
+ * @param aura  - aura number
+ * @return pointer to the block or NULL on failure
+ */
+static inline void *cvmx_fpa3_alloc(cvmx_fpa3_gaura_t aura)
+{
+	u64 address;
+	cvmx_fpa3_load_data_t load_addr;
+
+	load_addr.u64 = 0;
+	load_addr.seg = CVMX_MIPS_SPACE_XKPHYS;
+	load_addr.io = 1;
+	load_addr.did = 0x29; /* Device ID. Indicates FPA. */
+	load_addr.node = aura.node;
+	load_addr.red = 0; /* Perform RED on allocation.
+				  * FIXME to use config option
+				  */
+	load_addr.aura = aura.laura;
+
+	address = cvmx_read64_uint64(load_addr.u64);
+	if (!address)
+		return NULL;
+	return cvmx_phys_to_ptr(address);
+}
+
+/**
+ * Asynchronously get a new block from the FPA
+ *
+ * The result of cvmx_fpa_async_alloc() may be retrieved using
+ * cvmx_fpa_async_alloc_finish().
+ *
+ * @param scr_addr Local scratch address to put response in.  This is a byte
+ *		   address but must be 8 byte aligned.
+ * @param aura     Global aura to get the block from
+ */
+static inline void cvmx_fpa3_async_alloc(u64 scr_addr, cvmx_fpa3_gaura_t aura)
+{
+	cvmx_fpa3_iobdma_data_t data;
+
+	/* Hardware only uses 64 bit aligned locations, so convert from byte
+	 * address to 64-bit index
+	 */
+	data.u64 = 0ull;
+	data.cn78xx.scraddr = scr_addr >> 3;
+	data.cn78xx.len = 1;
+	data.cn78xx.did = 0x29;
+	data.cn78xx.node = aura.node;
+	data.cn78xx.aura = aura.laura;
+	cvmx_scratch_write64(scr_addr, 0ull);
+
+	CVMX_SYNCW;
+	cvmx_send_single(data.u64);
+}
+
+/**
+ * Retrieve the result of cvmx_fpa3_async_alloc
+ *
+ * @param scr_addr The Local scratch address.  Must be the same value
+ * passed to cvmx_fpa_async_alloc().
+ *
+ * @param aura Global aura the block came from.  Must be the same value
+ * passed to cvmx_fpa_async_alloc.
+ *
+ * @return Pointer to the block or NULL on failure
+ */
+static inline void *cvmx_fpa3_async_alloc_finish(u64 scr_addr, cvmx_fpa3_gaura_t aura)
+{
+	u64 address;
+
+	CVMX_SYNCIOBDMA;
+
+	address = cvmx_scratch_read64(scr_addr);
+	if (cvmx_likely(address))
+		return cvmx_phys_to_ptr(address);
+	else
+		/* Try regular alloc if async failed */
+		return cvmx_fpa3_alloc(aura);
+}
+
+/**
+ * Free a pointer back to the pool.
+ *
+ * @param aura   global aura number
+ * @param ptr    physical address of block to free.
+ * @param num_cache_lines Cache lines to invalidate
+ */
+static inline void cvmx_fpa3_free(void *ptr, cvmx_fpa3_gaura_t aura, unsigned int num_cache_lines)
+{
+	cvmx_fpa3_store_addr_t newptr;
+	cvmx_addr_t newdata;
+
+	newdata.u64 = cvmx_ptr_to_phys(ptr);
+
+	/* Make sure that any previous writes to memory go out before we free
+	   this buffer. This also serves as a barrier to prevent GCC from
+	   reordering operations to after the free. */
+	CVMX_SYNCWS;
+
+	newptr.u64 = 0;
+	newptr.seg = CVMX_MIPS_SPACE_XKPHYS;
+	newptr.io = 1;
+	newptr.did = 0x29; /* Device id, indicates FPA */
+	newptr.node = aura.node;
+	newptr.aura = aura.laura;
+	newptr.fabs = 0; /* Free absolute. FIXME to use config option */
+	newptr.dwb_count = num_cache_lines;
+
+	cvmx_write_io(newptr.u64, newdata.u64);
+}
+
+/**
+ * Free a pointer back to the pool without flushing the write buffer.
+ *
+ * @param aura   global aura number
+ * @param ptr    physical address of block to free.
+ * @param num_cache_lines Cache lines to invalidate
+ */
+static inline void cvmx_fpa3_free_nosync(void *ptr, cvmx_fpa3_gaura_t aura,
+					 unsigned int num_cache_lines)
+{
+	cvmx_fpa3_store_addr_t newptr;
+	cvmx_addr_t newdata;
+
+	newdata.u64 = cvmx_ptr_to_phys(ptr);
+
+	/* Prevent GCC from reordering writes to (*ptr) */
+	asm volatile("" : : : "memory");
+
+	newptr.u64 = 0;
+	newptr.seg = CVMX_MIPS_SPACE_XKPHYS;
+	newptr.io = 1;
+	newptr.did = 0x29; /* Device id, indicates FPA */
+	newptr.node = aura.node;
+	newptr.aura = aura.laura;
+	newptr.fabs = 0; /* Free absolute. FIXME to use config option */
+	newptr.dwb_count = num_cache_lines;
+
+	cvmx_write_io(newptr.u64, newdata.u64);
+}
+
+static inline int cvmx_fpa3_pool_is_enabled(cvmx_fpa3_pool_t pool)
+{
+	cvmx_fpa_poolx_cfg_t pool_cfg;
+
+	if (!__cvmx_fpa3_pool_valid(pool))
+		return -1;
+
+	pool_cfg.u64 = cvmx_read_csr_node(pool.node, CVMX_FPA_POOLX_CFG(pool.lpool));
+	return pool_cfg.cn78xx.ena;
+}
+
+static inline int cvmx_fpa3_config_red_params(unsigned int node, int qos_avg_en, int red_lvl_dly,
+					      int avg_dly)
+{
+	cvmx_fpa_gen_cfg_t fpa_cfg;
+	cvmx_fpa_red_delay_t red_delay;
+
+	fpa_cfg.u64 = cvmx_read_csr_node(node, CVMX_FPA_GEN_CFG);
+	fpa_cfg.s.avg_en = qos_avg_en;
+	fpa_cfg.s.lvl_dly = red_lvl_dly;
+	cvmx_write_csr_node(node, CVMX_FPA_GEN_CFG, fpa_cfg.u64);
+
+	red_delay.u64 = cvmx_read_csr_node(node, CVMX_FPA_RED_DELAY);
+	red_delay.s.avg_dly = avg_dly;
+	cvmx_write_csr_node(node, CVMX_FPA_RED_DELAY, red_delay.u64);
+	return 0;
+}
+
+/**
+ * Gets the buffer size of the specified pool,
+ *
+ * @param aura Global aura number
+ * @return Returns size of the buffers in the specified pool.
+ */
+static inline int cvmx_fpa3_get_aura_buf_size(cvmx_fpa3_gaura_t aura)
+{
+	cvmx_fpa3_pool_t pool;
+	cvmx_fpa_poolx_cfg_t pool_cfg;
+	int block_size;
+
+	pool = cvmx_fpa3_aura_to_pool(aura);
+
+	pool_cfg.u64 = cvmx_read_csr_node(pool.node, CVMX_FPA_POOLX_CFG(pool.lpool));
+	block_size = pool_cfg.cn78xx.buf_size << 7;
+	return block_size;
+}
+
+/**
+ * Return the number of available buffers in an AURA
+ *
+ * @param aura to receive count for
+ * @return available buffer count
+ */
+static inline long long cvmx_fpa3_get_available(cvmx_fpa3_gaura_t aura)
+{
+	cvmx_fpa3_pool_t pool;
+	cvmx_fpa_poolx_available_t avail_reg;
+	cvmx_fpa_aurax_cnt_t cnt_reg;
+	cvmx_fpa_aurax_cnt_limit_t limit_reg;
+	long long ret;
+
+	pool = cvmx_fpa3_aura_to_pool(aura);
+
+	/* Get POOL available buffer count */
+	avail_reg.u64 = cvmx_read_csr_node(pool.node, CVMX_FPA_POOLX_AVAILABLE(pool.lpool));
+
+	/* Get AURA current available count */
+	cnt_reg.u64 = cvmx_read_csr_node(aura.node, CVMX_FPA_AURAX_CNT(aura.laura));
+	limit_reg.u64 = cvmx_read_csr_node(aura.node, CVMX_FPA_AURAX_CNT_LIMIT(aura.laura));
+
+	if (limit_reg.cn78xx.limit < cnt_reg.cn78xx.cnt)
+		return 0;
+
+	/* Calculate AURA-based buffer allowance */
+	ret = limit_reg.cn78xx.limit - cnt_reg.cn78xx.cnt;
+
+	/* Use POOL real buffer availability when less then allowance */
+	if (ret > (long long)avail_reg.cn78xx.count)
+		ret = avail_reg.cn78xx.count;
+
+	return ret;
+}
+
+/**
+ * Configure the QoS parameters of an FPA3 AURA
+ *
+ * @param aura is the FPA3 AURA handle
+ * @param ena_bp enables backpressure when outstanding count exceeds 'bp_thresh'
+ * @param ena_red enables random early discard when outstanding count exceeds 'pass_thresh'
+ * @param pass_thresh is the maximum count to invoke flow control
+ * @param drop_thresh is the count threshold to begin dropping packets
+ * @param bp_thresh is the back-pressure threshold
+ *
+ */
+static inline void cvmx_fpa3_setup_aura_qos(cvmx_fpa3_gaura_t aura, bool ena_red, u64 pass_thresh,
+					    u64 drop_thresh, bool ena_bp, u64 bp_thresh)
+{
+	unsigned int shift = 0;
+	u64 shift_thresh;
+	cvmx_fpa_aurax_cnt_limit_t limit_reg;
+	cvmx_fpa_aurax_cnt_levels_t aura_level;
+
+	if (!__cvmx_fpa3_aura_valid(aura))
+		return;
+
+	/* Get AURAX count limit for validation */
+	limit_reg.u64 = cvmx_read_csr_node(aura.node, CVMX_FPA_AURAX_CNT_LIMIT(aura.laura));
+
+	if (pass_thresh < 256)
+		pass_thresh = 255;
+
+	if (drop_thresh <= pass_thresh || drop_thresh > limit_reg.cn78xx.limit)
+		drop_thresh = limit_reg.cn78xx.limit;
+
+	if (bp_thresh < 256 || bp_thresh > limit_reg.cn78xx.limit)
+		bp_thresh = limit_reg.cn78xx.limit >> 1;
+
+	shift_thresh = (bp_thresh > drop_thresh) ? bp_thresh : drop_thresh;
+
+	/* Calculate shift so that the largest threshold fits in 8 bits */
+	for (shift = 0; shift < (1 << 6); shift++) {
+		if (0 == ((shift_thresh >> shift) & ~0xffull))
+			break;
+	};
+
+	aura_level.u64 = cvmx_read_csr_node(aura.node, CVMX_FPA_AURAX_CNT_LEVELS(aura.laura));
+	aura_level.s.pass = pass_thresh >> shift;
+	aura_level.s.drop = drop_thresh >> shift;
+	aura_level.s.bp = bp_thresh >> shift;
+	aura_level.s.shift = shift;
+	aura_level.s.red_ena = ena_red;
+	aura_level.s.bp_ena = ena_bp;
+	cvmx_write_csr_node(aura.node, CVMX_FPA_AURAX_CNT_LEVELS(aura.laura), aura_level.u64);
+}
+
+cvmx_fpa3_gaura_t cvmx_fpa3_reserve_aura(int node, int desired_aura_num);
+int cvmx_fpa3_release_aura(cvmx_fpa3_gaura_t aura);
+cvmx_fpa3_pool_t cvmx_fpa3_reserve_pool(int node, int desired_pool_num);
+int cvmx_fpa3_release_pool(cvmx_fpa3_pool_t pool);
+int cvmx_fpa3_is_aura_available(int node, int aura_num);
+int cvmx_fpa3_is_pool_available(int node, int pool_num);
+
+cvmx_fpa3_pool_t cvmx_fpa3_setup_fill_pool(int node, int desired_pool, const char *name,
+					   unsigned int block_size, unsigned int num_blocks,
+					   void *buffer);
+
+/**
+ * Function to attach an aura to an existing pool
+ *
+ * @param node - configure fpa on this node
+ * @param pool - configured pool to attach aura to
+ * @param desired_aura - pointer to aura to use, set to -1 to allocate
+ * @param name - name to register
+ * @param block_size - size of buffers to use
+ * @param num_blocks - number of blocks to allocate
+ *
+ * @return configured gaura on success, CVMX_FPA3_INVALID_GAURA on failure
+ */
+cvmx_fpa3_gaura_t cvmx_fpa3_set_aura_for_pool(cvmx_fpa3_pool_t pool, int desired_aura,
+					      const char *name, unsigned int block_size,
+					      unsigned int num_blocks);
+
+/**
+ * Function to setup and initialize a pool.
+ *
+ * @param node - configure fpa on this node
+ * @param desired_aura - aura to use, -1 for dynamic allocation
+ * @param name - name to register
+ * @param block_size - size of buffers in pool
+ * @param num_blocks - max number of buffers allowed
+ */
+cvmx_fpa3_gaura_t cvmx_fpa3_setup_aura_and_pool(int node, int desired_aura, const char *name,
+						void *buffer, unsigned int block_size,
+						unsigned int num_blocks);
+
+int cvmx_fpa3_shutdown_aura_and_pool(cvmx_fpa3_gaura_t aura);
+int cvmx_fpa3_shutdown_aura(cvmx_fpa3_gaura_t aura);
+int cvmx_fpa3_shutdown_pool(cvmx_fpa3_pool_t pool);
+const char *cvmx_fpa3_get_pool_name(cvmx_fpa3_pool_t pool);
+int cvmx_fpa3_get_pool_buf_size(cvmx_fpa3_pool_t pool);
+const char *cvmx_fpa3_get_aura_name(cvmx_fpa3_gaura_t aura);
+
+/* FIXME: Need a different macro for stage2 of u-boot */
+
+static inline void cvmx_fpa3_stage2_init(int aura, int pool, u64 stack_paddr, int stacklen,
+					 int buffer_sz, int buf_cnt)
+{
+	cvmx_fpa_poolx_cfg_t pool_cfg;
+
+	/* Configure pool stack */
+	cvmx_write_csr_node(0, CVMX_FPA_POOLX_STACK_BASE(pool), stack_paddr);
+	cvmx_write_csr_node(0, CVMX_FPA_POOLX_STACK_ADDR(pool), stack_paddr);
+	cvmx_write_csr_node(0, CVMX_FPA_POOLX_STACK_END(pool), stack_paddr + stacklen);
+
+	/* Configure pool with buffer size */
+	pool_cfg.u64 = 0;
+	pool_cfg.cn78xx.nat_align = 1;
+	pool_cfg.cn78xx.buf_size = buffer_sz >> 7;
+	pool_cfg.cn78xx.l_type = 0x2;
+	pool_cfg.cn78xx.ena = 0;
+	cvmx_write_csr_node(0, CVMX_FPA_POOLX_CFG(pool), pool_cfg.u64);
+	/* Reset pool before starting */
+	pool_cfg.cn78xx.ena = 1;
+	cvmx_write_csr_node(0, CVMX_FPA_POOLX_CFG(pool), pool_cfg.u64);
+
+	cvmx_write_csr_node(0, CVMX_FPA_AURAX_CFG(aura), 0);
+	cvmx_write_csr_node(0, CVMX_FPA_AURAX_CNT_ADD(aura), buf_cnt);
+	cvmx_write_csr_node(0, CVMX_FPA_AURAX_POOL(aura), (u64)pool);
+}
+
+static inline void cvmx_fpa3_stage2_disable(int aura, int pool)
+{
+	cvmx_write_csr_node(0, CVMX_FPA_AURAX_POOL(aura), 0);
+	cvmx_write_csr_node(0, CVMX_FPA_POOLX_CFG(pool), 0);
+	cvmx_write_csr_node(0, CVMX_FPA_POOLX_STACK_BASE(pool), 0);
+	cvmx_write_csr_node(0, CVMX_FPA_POOLX_STACK_ADDR(pool), 0);
+	cvmx_write_csr_node(0, CVMX_FPA_POOLX_STACK_END(pool), 0);
+}
+
+#endif /* __CVMX_FPA3_H__ */
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-global-resources.h b/arch/mips/mach-octeon/include/mach/cvmx-global-resources.h
new file mode 100644
index 0000000000..28c32ddbe1
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-global-resources.h
@@ -0,0 +1,213 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ */
+
+#ifndef _CVMX_GLOBAL_RESOURCES_T_
+#define _CVMX_GLOBAL_RESOURCES_T_
+
+#define CVMX_GLOBAL_RESOURCES_DATA_NAME "cvmx-global-resources"
+
+/*In macros below abbreviation GR stands for global resources. */
+#define CVMX_GR_TAG_INVALID                                                                        \
+	cvmx_get_gr_tag('i', 'n', 'v', 'a', 'l', 'i', 'd', '.', '.', '.', '.', '.', '.', '.', '.', \
+			'.')
+/*Tag for pko que table range. */
+#define CVMX_GR_TAG_PKO_QUEUES                                                                     \
+	cvmx_get_gr_tag('c', 'v', 'm', '_', 'p', 'k', 'o', '_', 'q', 'u', 'e', 'u', 's', '.', '.', \
+			'.')
+/*Tag for a pko internal ports range */
+#define CVMX_GR_TAG_PKO_IPORTS                                                                     \
+	cvmx_get_gr_tag('c', 'v', 'm', '_', 'p', 'k', 'o', '_', 'i', 'p', 'o', 'r', 't', '.', '.', \
+			'.')
+#define CVMX_GR_TAG_FPA                                                                            \
+	cvmx_get_gr_tag('c', 'v', 'm', '_', 'f', 'p', 'a', '.', '.', '.', '.', '.', '.', '.', '.', \
+			'.')
+#define CVMX_GR_TAG_FAU                                                                            \
+	cvmx_get_gr_tag('c', 'v', 'm', '_', 'f', 'a', 'u', '.', '.', '.', '.', '.', '.', '.', '.', \
+			'.')
+#define CVMX_GR_TAG_SSO_GRP(n)                                                                     \
+	cvmx_get_gr_tag('c', 'v', 'm', '_', 's', 's', 'o', '_', '0', (n) + '0', '.', '.', '.',     \
+			'.', '.', '.');
+#define CVMX_GR_TAG_TIM(n)                                                                         \
+	cvmx_get_gr_tag('c', 'v', 'm', '_', 't', 'i', 'm', '_', (n) + '0', '.', '.', '.', '.',     \
+			'.', '.', '.')
+#define CVMX_GR_TAG_CLUSTERS(x)                                                                    \
+	cvmx_get_gr_tag('c', 'v', 'm', '_', 'c', 'l', 'u', 's', 't', 'e', 'r', '_', (x + '0'),     \
+			'.', '.', '.')
+#define CVMX_GR_TAG_CLUSTER_GRP(x)                                                                 \
+	cvmx_get_gr_tag('c', 'v', 'm', '_', 'c', 'l', 'g', 'r', 'p', '_', (x + '0'), '.', '.',     \
+			'.', '.', '.')
+#define CVMX_GR_TAG_STYLE(x)                                                                       \
+	cvmx_get_gr_tag('c', 'v', 'm', '_', 's', 't', 'y', 'l', 'e', '_', (x + '0'), '.', '.',     \
+			'.', '.', '.')
+#define CVMX_GR_TAG_QPG_ENTRY(x)                                                                   \
+	cvmx_get_gr_tag('c', 'v', 'm', '_', 'q', 'p', 'g', 'e', 't', '_', (x + '0'), '.', '.',     \
+			'.', '.', '.')
+#define CVMX_GR_TAG_BPID(x)                                                                        \
+	cvmx_get_gr_tag('c', 'v', 'm', '_', 'b', 'p', 'i', 'd', 's', '_', (x + '0'), '.', '.',     \
+			'.', '.', '.')
+#define CVMX_GR_TAG_MTAG_IDX(x)                                                                    \
+	cvmx_get_gr_tag('c', 'v', 'm', '_', 'm', 't', 'a', 'g', 'x', '_', (x + '0'), '.', '.',     \
+			'.', '.', '.')
+#define CVMX_GR_TAG_PCAM(x, y, z)                                                                  \
+	cvmx_get_gr_tag('c', 'v', 'm', '_', 'p', 'c', 'a', 'm', '_', (x + '0'), (y + '0'),         \
+			(z + '0'), '.', '.', '.', '.')
+
+#define CVMX_GR_TAG_CIU3_IDT(_n)                                                                   \
+	cvmx_get_gr_tag('c', 'v', 'm', '_', 'c', 'i', 'u', '3', '_', ((_n) + '0'), '_', 'i', 'd',  \
+			't', '.', '.')
+
+/* Allocation of the 512 SW INTSTs (in the  12 bit SW INTSN space) */
+#define CVMX_GR_TAG_CIU3_SWINTSN(_n)                                                               \
+	cvmx_get_gr_tag('c', 'v', 'm', '_', 'c', 'i', 'u', '3', '_', ((_n) + '0'), '_', 's', 'w',  \
+			'i', 's', 'n')
+
+#define TAG_INIT_PART(A, B, C, D, E, F, G, H)                                                      \
+	((((u64)(A) & 0xff) << 56) | (((u64)(B) & 0xff) << 48) | (((u64)(C) & 0xff) << 40) |             \
+	 (((u64)(D) & 0xff) << 32) | (((u64)(E) & 0xff) << 24) | (((u64)(F) & 0xff) << 16) |             \
+	 (((u64)(G) & 0xff) << 8) | (((u64)(H) & 0xff)))
+
+struct global_resource_tag {
+	u64 lo;
+	u64 hi;
+};
+
+enum cvmx_resource_err { CVMX_RESOURCE_ALLOC_FAILED = -1, CVMX_RESOURCE_ALREADY_RESERVED = -2 };
+
+/*
+ * @INTERNAL
+ * Creates a tag from the specified characters.
+ */
+static inline struct global_resource_tag cvmx_get_gr_tag(char a, char b, char c, char d, char e,
+							 char f, char g, char h, char i, char j,
+							 char k, char l, char m, char n, char o,
+							 char p)
+{
+	struct global_resource_tag tag;
+
+	tag.lo = TAG_INIT_PART(a, b, c, d, e, f, g, h);
+	tag.hi = TAG_INIT_PART(i, j, k, l, m, n, o, p);
+	return tag;
+}
+
+static inline int cvmx_gr_same_tag(struct global_resource_tag gr1, struct global_resource_tag gr2)
+{
+	return (gr1.hi == gr2.hi) && (gr1.lo == gr2.lo);
+}
+
+/*
+ * @INTERNAL
+ * Creates a global resource range that can hold the specified number of
+ * elements
+ * @param tag is the tag of the range. The taga is created using the method
+ * cvmx_get_gr_tag()
+ * @param nelements is the number of elements to be held in the resource range.
+ */
+int cvmx_create_global_resource_range(struct global_resource_tag tag, int nelements);
+
+/*
+ * @INTERNAL
+ * Allocate nelements in the global resource range with the specified tag. It
+ * is assumed that prior
+ * to calling this the global resource range has already been created using
+ * cvmx_create_global_resource_range().
+ * @param tag is the tag of the global resource range.
+ * @param nelements is the number of elements to be allocated.
+ * @param owner is a 64 bit number that identifes the owner of this range.
+ * @aligment specifes the required alignment of the returned base number.
+ * @return returns the base of the allocated range. -1 return value indicates
+ * failure.
+ */
+int cvmx_allocate_global_resource_range(struct global_resource_tag tag, u64 owner, int nelements,
+					int alignment);
+
+/*
+ * @INTERNAL
+ * Allocate nelements in the global resource range with the specified tag.
+ * The elements allocated need not be contiguous. It is assumed that prior to
+ * calling this the global resource range has already
+ * been created using cvmx_create_global_resource_range().
+ * @param tag is the tag of the global resource range.
+ * @param nelements is the number of elements to be allocated.
+ * @param owner is a 64 bit number that identifes the owner of the allocated
+ * elements.
+ * @param allocated_elements returns indexs of the allocated entries.
+ * @return returns 0 on success and -1 on failure.
+ */
+int cvmx_resource_alloc_many(struct global_resource_tag tag, u64 owner, int nelements,
+			     int allocated_elements[]);
+int cvmx_resource_alloc_reverse(struct global_resource_tag, u64 owner);
+/*
+ * @INTERNAL
+ * Reserve nelements starting from base in the global resource range with the
+ * specified tag.
+ * It is assumed that prior to calling this the global resource range has
+ * already been created using cvmx_create_global_resource_range().
+ * @param tag is the tag of the global resource range.
+ * @param nelements is the number of elements to be allocated.
+ * @param owner is a 64 bit number that identifes the owner of this range.
+ * @base specifies the base start of nelements.
+ * @return returns the base of the allocated range. -1 return value indicates
+ * failure.
+ */
+int cvmx_reserve_global_resource_range(struct global_resource_tag tag, u64 owner, int base,
+				       int nelements);
+/*
+ * @INTERNAL
+ * Free nelements starting at base in the global resource range with the
+ * specified tag.
+ * @param tag is the tag of the global resource range.
+ * @param base is the base number
+ * @param nelements is the number of elements that are to be freed.
+ * @return returns 0 if successful and -1 on failure.
+ */
+int cvmx_free_global_resource_range_with_base(struct global_resource_tag tag, int base,
+					      int nelements);
+
+/*
+ * @INTERNAL
+ * Free nelements with the bases specified in bases[] with the
+ * specified tag.
+ * @param tag is the tag of the global resource range.
+ * @param bases is an array containing the bases to be freed.
+ * @param nelements is the number of elements that are to be freed.
+ * @return returns 0 if successful and -1 on failure.
+ */
+int cvmx_free_global_resource_range_multiple(struct global_resource_tag tag, int bases[],
+					     int nelements);
+/*
+ * @INTERNAL
+ * Free elements from the specified owner in the global resource range with the
+ * specified tag.
+ * @param tag is the tag of the global resource range.
+ * @param owner is the owner of resources that are to be freed.
+ * @return returns 0 if successful and -1 on failure.
+ */
+int cvmx_free_global_resource_range_with_owner(struct global_resource_tag tag, int owner);
+
+/*
+ * @INTERNAL
+ * Frees all the global resources that have been created.
+ * For use only from the bootloader, when it shutdown and boots up the
+ * application or kernel.
+ */
+int free_global_resources(void);
+
+u64 cvmx_get_global_resource_owner(struct global_resource_tag tag, int base);
+/*
+ * @INTERNAL
+ * Shows the global resource range with the specified tag. Use mainly for debug.
+ */
+void cvmx_show_global_resource_range(struct global_resource_tag tag);
+
+/*
+ * @INTERNAL
+ * Shows all the global resources. Used mainly for debug.
+ */
+void cvmx_global_resources_show(void);
+
+u64 cvmx_allocate_app_id(void);
+u64 cvmx_get_app_id(void);
+
+#endif
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-gmx.h b/arch/mips/mach-octeon/include/mach/cvmx-gmx.h
new file mode 100644
index 0000000000..2df7da102a
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-gmx.h
@@ -0,0 +1,16 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ *
+ * Interface to the GMX hardware.
+ */
+
+#ifndef __CVMX_GMX_H__
+#define __CVMX_GMX_H__
+
+/* CSR typedefs have been moved to cvmx-gmx-defs.h */
+
+int cvmx_gmx_set_backpressure_override(u32 interface, u32 port_mask);
+int cvmx_agl_set_backpressure_override(u32 interface, u32 port_mask);
+
+#endif
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-hwfau.h b/arch/mips/mach-octeon/include/mach/cvmx-hwfau.h
new file mode 100644
index 0000000000..59772190aa
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-hwfau.h
@@ -0,0 +1,606 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ *
+ * Interface to the hardware Fetch and Add Unit.
+ */
+
+/**
+ * @file
+ *
+ * Interface to the hardware Fetch and Add Unit.
+ *
+ */
+
+#ifndef __CVMX_HWFAU_H__
+#define __CVMX_HWFAU_H__
+
+typedef int cvmx_fau_reg64_t;
+typedef int cvmx_fau_reg32_t;
+typedef int cvmx_fau_reg16_t;
+typedef int cvmx_fau_reg8_t;
+
+#define CVMX_FAU_REG_ANY -1
+
+/*
+ * Octeon Fetch and Add Unit (FAU)
+ */
+
+#define CVMX_FAU_LOAD_IO_ADDRESS cvmx_build_io_address(0x1e, 0)
+#define CVMX_FAU_BITS_SCRADDR	 63, 56
+#define CVMX_FAU_BITS_LEN	 55, 48
+#define CVMX_FAU_BITS_INEVAL	 35, 14
+#define CVMX_FAU_BITS_TAGWAIT	 13, 13
+#define CVMX_FAU_BITS_NOADD	 13, 13
+#define CVMX_FAU_BITS_SIZE	 12, 11
+#define CVMX_FAU_BITS_REGISTER	 10, 0
+
+#define CVMX_FAU_MAX_REGISTERS_8 (2048)
+
+typedef enum {
+	CVMX_FAU_OP_SIZE_8 = 0,
+	CVMX_FAU_OP_SIZE_16 = 1,
+	CVMX_FAU_OP_SIZE_32 = 2,
+	CVMX_FAU_OP_SIZE_64 = 3
+} cvmx_fau_op_size_t;
+
+/**
+ * Tagwait return definition. If a timeout occurs, the error
+ * bit will be set. Otherwise the value of the register before
+ * the update will be returned.
+ */
+typedef struct {
+	u64 error : 1;
+	s64 value : 63;
+} cvmx_fau_tagwait64_t;
+
+/**
+ * Tagwait return definition. If a timeout occurs, the error
+ * bit will be set. Otherwise the value of the register before
+ * the update will be returned.
+ */
+typedef struct {
+	u64 error : 1;
+	s32 value : 31;
+} cvmx_fau_tagwait32_t;
+
+/**
+ * Tagwait return definition. If a timeout occurs, the error
+ * bit will be set. Otherwise the value of the register before
+ * the update will be returned.
+ */
+typedef struct {
+	u64 error : 1;
+	s16 value : 15;
+} cvmx_fau_tagwait16_t;
+
+/**
+ * Tagwait return definition. If a timeout occurs, the error
+ * bit will be set. Otherwise the value of the register before
+ * the update will be returned.
+ */
+typedef struct {
+	u64 error : 1;
+	int8_t value : 7;
+} cvmx_fau_tagwait8_t;
+
+/**
+ * Asynchronous tagwait return definition. If a timeout occurs,
+ * the error bit will be set. Otherwise the value of the
+ * register before the update will be returned.
+ */
+typedef union {
+	u64 u64;
+	struct {
+		u64 invalid : 1;
+		u64 data : 63; /* unpredictable if invalid is set */
+	} s;
+} cvmx_fau_async_tagwait_result_t;
+
+#define SWIZZLE_8  0
+#define SWIZZLE_16 0
+#define SWIZZLE_32 0
+
+/**
+ * @INTERNAL
+ * Builds a store I/O address for writing to the FAU
+ *
+ * @param noadd  0 = Store value is atomically added to the current value
+ *               1 = Store value is atomically written over the current value
+ * @param reg    FAU atomic register to access. 0 <= reg < 2048.
+ *               - Step by 2 for 16 bit access.
+ *               - Step by 4 for 32 bit access.
+ *               - Step by 8 for 64 bit access.
+ * @return Address to store for atomic update
+ */
+static inline u64 __cvmx_hwfau_store_address(u64 noadd, u64 reg)
+{
+	return (CVMX_ADD_IO_SEG(CVMX_FAU_LOAD_IO_ADDRESS) |
+		cvmx_build_bits(CVMX_FAU_BITS_NOADD, noadd) |
+		cvmx_build_bits(CVMX_FAU_BITS_REGISTER, reg));
+}
+
+/**
+ * @INTERNAL
+ * Builds a I/O address for accessing the FAU
+ *
+ * @param tagwait Should the atomic add wait for the current tag switch
+ *                operation to complete.
+ *                - 0 = Don't wait
+ *                - 1 = Wait for tag switch to complete
+ * @param reg     FAU atomic register to access. 0 <= reg < 2048.
+ *                - Step by 2 for 16 bit access.
+ *                - Step by 4 for 32 bit access.
+ *                - Step by 8 for 64 bit access.
+ * @param value   Signed value to add.
+ *                Note: When performing 32 and 64 bit access, only the low
+ *                22 bits are available.
+ * @return Address to read from for atomic update
+ */
+static inline u64 __cvmx_hwfau_atomic_address(u64 tagwait, u64 reg, s64 value)
+{
+	return (CVMX_ADD_IO_SEG(CVMX_FAU_LOAD_IO_ADDRESS) |
+		cvmx_build_bits(CVMX_FAU_BITS_INEVAL, value) |
+		cvmx_build_bits(CVMX_FAU_BITS_TAGWAIT, tagwait) |
+		cvmx_build_bits(CVMX_FAU_BITS_REGISTER, reg));
+}
+
+/**
+ * Perform an atomic 64 bit add
+ *
+ * @param reg     FAU atomic register to access. 0 <= reg < 2048.
+ *                - Step by 8 for 64 bit access.
+ * @param value   Signed value to add.
+ *                Note: Only the low 22 bits are available.
+ * @return Value of the register before the update
+ */
+static inline s64 cvmx_hwfau_fetch_and_add64(cvmx_fau_reg64_t reg, s64 value)
+{
+	return cvmx_read64_int64(__cvmx_hwfau_atomic_address(0, reg, value));
+}
+
+/**
+ * Perform an atomic 32 bit add
+ *
+ * @param reg     FAU atomic register to access. 0 <= reg < 2048.
+ *                - Step by 4 for 32 bit access.
+ * @param value   Signed value to add.
+ *                Note: Only the low 22 bits are available.
+ * @return Value of the register before the update
+ */
+static inline s32 cvmx_hwfau_fetch_and_add32(cvmx_fau_reg32_t reg, s32 value)
+{
+	reg ^= SWIZZLE_32;
+	return cvmx_read64_int32(__cvmx_hwfau_atomic_address(0, reg, value));
+}
+
+/**
+ * Perform an atomic 16 bit add
+ *
+ * @param reg     FAU atomic register to access. 0 <= reg < 2048.
+ *                - Step by 2 for 16 bit access.
+ * @param value   Signed value to add.
+ * @return Value of the register before the update
+ */
+static inline s16 cvmx_hwfau_fetch_and_add16(cvmx_fau_reg16_t reg, s16 value)
+{
+	reg ^= SWIZZLE_16;
+	return cvmx_read64_int16(__cvmx_hwfau_atomic_address(0, reg, value));
+}
+
+/**
+ * Perform an atomic 8 bit add
+ *
+ * @param reg     FAU atomic register to access. 0 <= reg < 2048.
+ * @param value   Signed value to add.
+ * @return Value of the register before the update
+ */
+static inline int8_t cvmx_hwfau_fetch_and_add8(cvmx_fau_reg8_t reg, int8_t value)
+{
+	reg ^= SWIZZLE_8;
+	return cvmx_read64_int8(__cvmx_hwfau_atomic_address(0, reg, value));
+}
+
+/**
+ * Perform an atomic 64 bit add after the current tag switch
+ * completes
+ *
+ * @param reg    FAU atomic register to access. 0 <= reg < 2048.
+ *               - Step by 8 for 64 bit access.
+ * @param value  Signed value to add.
+ *               Note: Only the low 22 bits are available.
+ * @return If a timeout occurs, the error bit will be set. Otherwise
+ *         the value of the register before the update will be
+ *         returned
+ */
+static inline cvmx_fau_tagwait64_t cvmx_hwfau_tagwait_fetch_and_add64(cvmx_fau_reg64_t reg,
+								      s64 value)
+{
+	union {
+		u64 i64;
+		cvmx_fau_tagwait64_t t;
+	} result;
+	result.i64 = cvmx_read64_int64(__cvmx_hwfau_atomic_address(1, reg, value));
+	return result.t;
+}
+
+/**
+ * Perform an atomic 32 bit add after the current tag switch
+ * completes
+ *
+ * @param reg    FAU atomic register to access. 0 <= reg < 2048.
+ *               - Step by 4 for 32 bit access.
+ * @param value  Signed value to add.
+ *               Note: Only the low 22 bits are available.
+ * @return If a timeout occurs, the error bit will be set. Otherwise
+ *         the value of the register before the update will be
+ *         returned
+ */
+static inline cvmx_fau_tagwait32_t cvmx_hwfau_tagwait_fetch_and_add32(cvmx_fau_reg32_t reg,
+								      s32 value)
+{
+	union {
+		u64 i32;
+		cvmx_fau_tagwait32_t t;
+	} result;
+	reg ^= SWIZZLE_32;
+	result.i32 = cvmx_read64_int32(__cvmx_hwfau_atomic_address(1, reg, value));
+	return result.t;
+}
+
+/**
+ * Perform an atomic 16 bit add after the current tag switch
+ * completes
+ *
+ * @param reg    FAU atomic register to access. 0 <= reg < 2048.
+ *               - Step by 2 for 16 bit access.
+ * @param value  Signed value to add.
+ * @return If a timeout occurs, the error bit will be set. Otherwise
+ *         the value of the register before the update will be
+ *         returned
+ */
+static inline cvmx_fau_tagwait16_t cvmx_hwfau_tagwait_fetch_and_add16(cvmx_fau_reg16_t reg,
+								      s16 value)
+{
+	union {
+		u64 i16;
+		cvmx_fau_tagwait16_t t;
+	} result;
+	reg ^= SWIZZLE_16;
+	result.i16 = cvmx_read64_int16(__cvmx_hwfau_atomic_address(1, reg, value));
+	return result.t;
+}
+
+/**
+ * Perform an atomic 8 bit add after the current tag switch
+ * completes
+ *
+ * @param reg    FAU atomic register to access. 0 <= reg < 2048.
+ * @param value  Signed value to add.
+ * @return If a timeout occurs, the error bit will be set. Otherwise
+ *         the value of the register before the update will be
+ *         returned
+ */
+static inline cvmx_fau_tagwait8_t cvmx_hwfau_tagwait_fetch_and_add8(cvmx_fau_reg8_t reg,
+								    int8_t value)
+{
+	union {
+		u64 i8;
+		cvmx_fau_tagwait8_t t;
+	} result;
+	reg ^= SWIZZLE_8;
+	result.i8 = cvmx_read64_int8(__cvmx_hwfau_atomic_address(1, reg, value));
+	return result.t;
+}
+
+/**
+ * @INTERNAL
+ * Builds I/O data for async operations
+ *
+ * @param scraddr Scratch pad byte address to write to.  Must be 8 byte aligned
+ * @param value   Signed value to add.
+ *                Note: When performing 32 and 64 bit access, only the low
+ *                22 bits are available.
+ * @param tagwait Should the atomic add wait for the current tag switch
+ *                operation to complete.
+ *                - 0 = Don't wait
+ *                - 1 = Wait for tag switch to complete
+ * @param size    The size of the operation:
+ *                - CVMX_FAU_OP_SIZE_8  (0) = 8 bits
+ *                - CVMX_FAU_OP_SIZE_16 (1) = 16 bits
+ *                - CVMX_FAU_OP_SIZE_32 (2) = 32 bits
+ *                - CVMX_FAU_OP_SIZE_64 (3) = 64 bits
+ * @param reg     FAU atomic register to access. 0 <= reg < 2048.
+ *                - Step by 2 for 16 bit access.
+ *                - Step by 4 for 32 bit access.
+ *                - Step by 8 for 64 bit access.
+ * @return Data to write using cvmx_send_single
+ */
+static inline u64 __cvmx_fau_iobdma_data(u64 scraddr, s64 value, u64 tagwait,
+					 cvmx_fau_op_size_t size, u64 reg)
+{
+	return (CVMX_FAU_LOAD_IO_ADDRESS | cvmx_build_bits(CVMX_FAU_BITS_SCRADDR, scraddr >> 3) |
+		cvmx_build_bits(CVMX_FAU_BITS_LEN, 1) |
+		cvmx_build_bits(CVMX_FAU_BITS_INEVAL, value) |
+		cvmx_build_bits(CVMX_FAU_BITS_TAGWAIT, tagwait) |
+		cvmx_build_bits(CVMX_FAU_BITS_SIZE, size) |
+		cvmx_build_bits(CVMX_FAU_BITS_REGISTER, reg));
+}
+
+/**
+ * Perform an async atomic 64 bit add. The old value is
+ * placed in the scratch memory at byte address scraddr.
+ *
+ * @param scraddr Scratch memory byte address to put response in.
+ *                Must be 8 byte aligned.
+ * @param reg     FAU atomic register to access. 0 <= reg < 2048.
+ *                - Step by 8 for 64 bit access.
+ * @param value   Signed value to add.
+ *                Note: Only the low 22 bits are available.
+ * @return Placed in the scratch pad register
+ */
+static inline void cvmx_hwfau_async_fetch_and_add64(u64 scraddr, cvmx_fau_reg64_t reg, s64 value)
+{
+	cvmx_send_single(__cvmx_fau_iobdma_data(scraddr, value, 0, CVMX_FAU_OP_SIZE_64, reg));
+}
+
+/**
+ * Perform an async atomic 32 bit add. The old value is
+ * placed in the scratch memory at byte address scraddr.
+ *
+ * @param scraddr Scratch memory byte address to put response in.
+ *                Must be 8 byte aligned.
+ * @param reg     FAU atomic register to access. 0 <= reg < 2048.
+ *                - Step by 4 for 32 bit access.
+ * @param value   Signed value to add.
+ *                Note: Only the low 22 bits are available.
+ * @return Placed in the scratch pad register
+ */
+static inline void cvmx_hwfau_async_fetch_and_add32(u64 scraddr, cvmx_fau_reg32_t reg, s32 value)
+{
+	cvmx_send_single(__cvmx_fau_iobdma_data(scraddr, value, 0, CVMX_FAU_OP_SIZE_32, reg));
+}
+
+/**
+ * Perform an async atomic 16 bit add. The old value is
+ * placed in the scratch memory at byte address scraddr.
+ *
+ * @param scraddr Scratch memory byte address to put response in.
+ *                Must be 8 byte aligned.
+ * @param reg     FAU atomic register to access. 0 <= reg < 2048.
+ *                - Step by 2 for 16 bit access.
+ * @param value   Signed value to add.
+ * @return Placed in the scratch pad register
+ */
+static inline void cvmx_hwfau_async_fetch_and_add16(u64 scraddr, cvmx_fau_reg16_t reg, s16 value)
+{
+	cvmx_send_single(__cvmx_fau_iobdma_data(scraddr, value, 0, CVMX_FAU_OP_SIZE_16, reg));
+}
+
+/**
+ * Perform an async atomic 8 bit add. The old value is
+ * placed in the scratch memory at byte address scraddr.
+ *
+ * @param scraddr Scratch memory byte address to put response in.
+ *                Must be 8 byte aligned.
+ * @param reg     FAU atomic register to access. 0 <= reg < 2048.
+ * @param value   Signed value to add.
+ * @return Placed in the scratch pad register
+ */
+static inline void cvmx_hwfau_async_fetch_and_add8(u64 scraddr, cvmx_fau_reg8_t reg, int8_t value)
+{
+	cvmx_send_single(__cvmx_fau_iobdma_data(scraddr, value, 0, CVMX_FAU_OP_SIZE_8, reg));
+}
+
+/**
+ * Perform an async atomic 64 bit add after the current tag
+ * switch completes.
+ *
+ * @param scraddr Scratch memory byte address to put response in.
+ *                Must be 8 byte aligned.
+ *                If a timeout occurs, the error bit (63) will be set. Otherwise
+ *                the value of the register before the update will be
+ *                returned
+ * @param reg     FAU atomic register to access. 0 <= reg < 2048.
+ *                - Step by 8 for 64 bit access.
+ * @param value   Signed value to add.
+ *                Note: Only the low 22 bits are available.
+ * @return Placed in the scratch pad register
+ */
+static inline void cvmx_hwfau_async_tagwait_fetch_and_add64(u64 scraddr, cvmx_fau_reg64_t reg,
+							    s64 value)
+{
+	cvmx_send_single(__cvmx_fau_iobdma_data(scraddr, value, 1, CVMX_FAU_OP_SIZE_64, reg));
+}
+
+/**
+ * Perform an async atomic 32 bit add after the current tag
+ * switch completes.
+ *
+ * @param scraddr Scratch memory byte address to put response in.
+ *                Must be 8 byte aligned.
+ *                If a timeout occurs, the error bit (63) will be set. Otherwise
+ *                the value of the register before the update will be
+ *                returned
+ * @param reg     FAU atomic register to access. 0 <= reg < 2048.
+ *                - Step by 4 for 32 bit access.
+ * @param value   Signed value to add.
+ *                Note: Only the low 22 bits are available.
+ * @return Placed in the scratch pad register
+ */
+static inline void cvmx_hwfau_async_tagwait_fetch_and_add32(u64 scraddr, cvmx_fau_reg32_t reg,
+							    s32 value)
+{
+	cvmx_send_single(__cvmx_fau_iobdma_data(scraddr, value, 1, CVMX_FAU_OP_SIZE_32, reg));
+}
+
+/**
+ * Perform an async atomic 16 bit add after the current tag
+ * switch completes.
+ *
+ * @param scraddr Scratch memory byte address to put response in.
+ *                Must be 8 byte aligned.
+ *                If a timeout occurs, the error bit (63) will be set. Otherwise
+ *                the value of the register before the update will be
+ *                returned
+ * @param reg     FAU atomic register to access. 0 <= reg < 2048.
+ *                - Step by 2 for 16 bit access.
+ * @param value   Signed value to add.
+ * @return Placed in the scratch pad register
+ */
+static inline void cvmx_hwfau_async_tagwait_fetch_and_add16(u64 scraddr, cvmx_fau_reg16_t reg,
+							    s16 value)
+{
+	cvmx_send_single(__cvmx_fau_iobdma_data(scraddr, value, 1, CVMX_FAU_OP_SIZE_16, reg));
+}
+
+/**
+ * Perform an async atomic 8 bit add after the current tag
+ * switch completes.
+ *
+ * @param scraddr Scratch memory byte address to put response in.
+ *                Must be 8 byte aligned.
+ *                If a timeout occurs, the error bit (63) will be set. Otherwise
+ *                the value of the register before the update will be
+ *                returned
+ * @param reg     FAU atomic register to access. 0 <= reg < 2048.
+ * @param value   Signed value to add.
+ * @return Placed in the scratch pad register
+ */
+static inline void cvmx_hwfau_async_tagwait_fetch_and_add8(u64 scraddr, cvmx_fau_reg8_t reg,
+							   int8_t value)
+{
+	cvmx_send_single(__cvmx_fau_iobdma_data(scraddr, value, 1, CVMX_FAU_OP_SIZE_8, reg));
+}
+
+/**
+ * Perform an atomic 64 bit add
+ *
+ * @param reg     FAU atomic register to access. 0 <= reg < 2048.
+ *                - Step by 8 for 64 bit access.
+ * @param value   Signed value to add.
+ */
+static inline void cvmx_hwfau_atomic_add64(cvmx_fau_reg64_t reg, s64 value)
+{
+	cvmx_write64_int64(__cvmx_hwfau_store_address(0, reg), value);
+}
+
+/**
+ * Perform an atomic 32 bit add
+ *
+ * @param reg     FAU atomic register to access. 0 <= reg < 2048.
+ *                - Step by 4 for 32 bit access.
+ * @param value   Signed value to add.
+ */
+static inline void cvmx_hwfau_atomic_add32(cvmx_fau_reg32_t reg, s32 value)
+{
+	reg ^= SWIZZLE_32;
+	cvmx_write64_int32(__cvmx_hwfau_store_address(0, reg), value);
+}
+
+/**
+ * Perform an atomic 16 bit add
+ *
+ * @param reg     FAU atomic register to access. 0 <= reg < 2048.
+ *                - Step by 2 for 16 bit access.
+ * @param value   Signed value to add.
+ */
+static inline void cvmx_hwfau_atomic_add16(cvmx_fau_reg16_t reg, s16 value)
+{
+	reg ^= SWIZZLE_16;
+	cvmx_write64_int16(__cvmx_hwfau_store_address(0, reg), value);
+}
+
+/**
+ * Perform an atomic 8 bit add
+ *
+ * @param reg     FAU atomic register to access. 0 <= reg < 2048.
+ * @param value   Signed value to add.
+ */
+static inline void cvmx_hwfau_atomic_add8(cvmx_fau_reg8_t reg, int8_t value)
+{
+	reg ^= SWIZZLE_8;
+	cvmx_write64_int8(__cvmx_hwfau_store_address(0, reg), value);
+}
+
+/**
+ * Perform an atomic 64 bit write
+ *
+ * @param reg     FAU atomic register to access. 0 <= reg < 2048.
+ *                - Step by 8 for 64 bit access.
+ * @param value   Signed value to write.
+ */
+static inline void cvmx_hwfau_atomic_write64(cvmx_fau_reg64_t reg, s64 value)
+{
+	cvmx_write64_int64(__cvmx_hwfau_store_address(1, reg), value);
+}
+
+/**
+ * Perform an atomic 32 bit write
+ *
+ * @param reg     FAU atomic register to access. 0 <= reg < 2048.
+ *                - Step by 4 for 32 bit access.
+ * @param value   Signed value to write.
+ */
+static inline void cvmx_hwfau_atomic_write32(cvmx_fau_reg32_t reg, s32 value)
+{
+	reg ^= SWIZZLE_32;
+	cvmx_write64_int32(__cvmx_hwfau_store_address(1, reg), value);
+}
+
+/**
+ * Perform an atomic 16 bit write
+ *
+ * @param reg     FAU atomic register to access. 0 <= reg < 2048.
+ *                - Step by 2 for 16 bit access.
+ * @param value   Signed value to write.
+ */
+static inline void cvmx_hwfau_atomic_write16(cvmx_fau_reg16_t reg, s16 value)
+{
+	reg ^= SWIZZLE_16;
+	cvmx_write64_int16(__cvmx_hwfau_store_address(1, reg), value);
+}
+
+/**
+ * Perform an atomic 8 bit write
+ *
+ * @param reg     FAU atomic register to access. 0 <= reg < 2048.
+ * @param value   Signed value to write.
+ */
+static inline void cvmx_hwfau_atomic_write8(cvmx_fau_reg8_t reg, int8_t value)
+{
+	reg ^= SWIZZLE_8;
+	cvmx_write64_int8(__cvmx_hwfau_store_address(1, reg), value);
+}
+
+/** Allocates 64bit FAU register.
+ *  @return value is the base address of allocated FAU register
+ */
+int cvmx_fau64_alloc(int reserve);
+
+/** Allocates 32bit FAU register.
+ *  @return value is the base address of allocated FAU register
+ */
+int cvmx_fau32_alloc(int reserve);
+
+/** Allocates 16bit FAU register.
+ *  @return value is the base address of allocated FAU register
+ */
+int cvmx_fau16_alloc(int reserve);
+
+/** Allocates 8bit FAU register.
+ *  @return value is the base address of allocated FAU register
+ */
+int cvmx_fau8_alloc(int reserve);
+
+/** Frees the specified FAU register.
+ *  @param address Base address of register to release.
+ *  @return 0 on success; -1 on failure
+ */
+int cvmx_fau_free(int address);
+
+/** Display the fau registers array
+ */
+void cvmx_fau_show(void);
+
+#endif /* __CVMX_HWFAU_H__ */
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-hwpko.h b/arch/mips/mach-octeon/include/mach/cvmx-hwpko.h
new file mode 100644
index 0000000000..459c19bbc0
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-hwpko.h
@@ -0,0 +1,570 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ *
+ * Interface to the hardware Packet Output unit.
+ *
+ * Starting with SDK 1.7.0, the PKO output functions now support
+ * two types of locking. CVMX_PKO_LOCK_ATOMIC_TAG continues to
+ * function similarly to previous SDKs by using POW atomic tags
+ * to preserve ordering and exclusivity. As a new option, you
+ * can now pass CVMX_PKO_LOCK_CMD_QUEUE which uses a ll/sc
+ * memory based locking instead. This locking has the advantage
+ * of not affecting the tag state but doesn't preserve packet
+ * ordering. CVMX_PKO_LOCK_CMD_QUEUE is appropriate in most
+ * generic code while CVMX_PKO_LOCK_CMD_QUEUE should be used
+ * with hand tuned fast path code.
+ *
+ * Some of other SDK differences visible to the command command
+ * queuing:
+ * - PKO indexes are no longer stored in the FAU. A large
+ *   percentage of the FAU register block used to be tied up
+ *   maintaining PKO queue pointers. These are now stored in a
+ *   global named block.
+ * - The PKO <b>use_locking</b> parameter can now have a global
+ *   effect. Since all application use the same named block,
+ *   queue locking correctly applies across all operating
+ *   systems when using CVMX_PKO_LOCK_CMD_QUEUE.
+ * - PKO 3 word commands are now supported. Use
+ *   cvmx_pko_send_packet_finish3().
+ */
+
+#ifndef __CVMX_HWPKO_H__
+#define __CVMX_HWPKO_H__
+
+#include "cvmx-hwfau.h"
+#include "cvmx-fpa.h"
+#include "cvmx-pow.h"
+#include "cvmx-cmd-queue.h"
+#include "cvmx-helper.h"
+#include "cvmx-helper-util.h"
+#include "cvmx-helper-cfg.h"
+
+/* Adjust the command buffer size by 1 word so that in the case of using only
+** two word PKO commands no command words stradle buffers.  The useful values
+** for this are 0 and 1. */
+#define CVMX_PKO_COMMAND_BUFFER_SIZE_ADJUST (1)
+
+#define CVMX_PKO_MAX_OUTPUT_QUEUES_STATIC 256
+#define CVMX_PKO_MAX_OUTPUT_QUEUES                                                                 \
+	((OCTEON_IS_OCTEON2() || OCTEON_IS_MODEL(OCTEON_CN70XX)) ? 256 : 128)
+#define CVMX_PKO_NUM_OUTPUT_PORTS                                                                  \
+	((OCTEON_IS_MODEL(OCTEON_CN63XX)) ? 44 : (OCTEON_IS_MODEL(OCTEON_CN66XX) ? 48 : 40))
+#define CVMX_PKO_MEM_QUEUE_PTRS_ILLEGAL_PID 63
+#define CVMX_PKO_QUEUE_STATIC_PRIORITY	    9
+#define CVMX_PKO_ILLEGAL_QUEUE		    0xFFFF
+#define CVMX_PKO_MAX_QUEUE_DEPTH	    0
+
+typedef enum {
+	CVMX_PKO_SUCCESS,
+	CVMX_PKO_INVALID_PORT,
+	CVMX_PKO_INVALID_QUEUE,
+	CVMX_PKO_INVALID_PRIORITY,
+	CVMX_PKO_NO_MEMORY,
+	CVMX_PKO_PORT_ALREADY_SETUP,
+	CVMX_PKO_CMD_QUEUE_INIT_ERROR
+} cvmx_pko_return_value_t;
+
+/**
+ * This enumeration represents the differnet locking modes supported by PKO.
+ */
+typedef enum {
+	CVMX_PKO_LOCK_NONE = 0,
+	CVMX_PKO_LOCK_ATOMIC_TAG = 1,
+	CVMX_PKO_LOCK_CMD_QUEUE = 2,
+} cvmx_pko_lock_t;
+
+typedef struct cvmx_pko_port_status {
+	u32 packets;
+	u64 octets;
+	u64 doorbell;
+} cvmx_pko_port_status_t;
+
+/**
+ * This structure defines the address to use on a packet enqueue
+ */
+typedef union {
+	u64 u64;
+	struct {
+		cvmx_mips_space_t mem_space : 2;
+		u64 reserved : 13;
+		u64 is_io : 1;
+		u64 did : 8;
+		u64 reserved2 : 4;
+		u64 reserved3 : 15;
+		u64 port : 9;
+		u64 queue : 9;
+		u64 reserved4 : 3;
+	} s;
+} cvmx_pko_doorbell_address_t;
+
+/**
+ * Structure of the first packet output command word.
+ */
+typedef union {
+	u64 u64;
+	struct {
+		cvmx_fau_op_size_t size1 : 2;
+		cvmx_fau_op_size_t size0 : 2;
+		u64 subone1 : 1;
+		u64 reg1 : 11;
+		u64 subone0 : 1;
+		u64 reg0 : 11;
+		u64 le : 1;
+		u64 n2 : 1;
+		u64 wqp : 1;
+		u64 rsp : 1;
+		u64 gather : 1;
+		u64 ipoffp1 : 7;
+		u64 ignore_i : 1;
+		u64 dontfree : 1;
+		u64 segs : 6;
+		u64 total_bytes : 16;
+	} s;
+} cvmx_pko_command_word0_t;
+
+/**
+ * Call before any other calls to initialize the packet
+ * output system.
+ */
+
+void cvmx_pko_hw_init(u8 pool, unsigned int bufsize);
+
+/**
+ * Enables the packet output hardware. It must already be
+ * configured.
+ */
+void cvmx_pko_enable(void);
+
+/**
+ * Disables the packet output. Does not affect any configuration.
+ */
+void cvmx_pko_disable(void);
+
+/**
+ * Shutdown and free resources required by packet output.
+ */
+
+void cvmx_pko_shutdown(void);
+
+/**
+ * Configure a output port and the associated queues for use.
+ *
+ * @param port       Port to configure.
+ * @param base_queue First queue number to associate with this port.
+ * @param num_queues Number of queues t oassociate with this port
+ * @param priority   Array of priority levels for each queue. Values are
+ *                   allowed to be 1-8. A value of 8 get 8 times the traffic
+ *                   of a value of 1. There must be num_queues elements in the
+ *                   array.
+ */
+cvmx_pko_return_value_t cvmx_pko_config_port(int port, int base_queue, int num_queues,
+					     const u8 priority[]);
+
+/**
+ * Ring the packet output doorbell. This tells the packet
+ * output hardware that "len" command words have been added
+ * to its pending list.  This command includes the required
+ * CVMX_SYNCWS before the doorbell ring.
+ *
+ * WARNING: This function may have to look up the proper PKO port in
+ * the IPD port to PKO port map, and is thus slower than calling
+ * cvmx_pko_doorbell_pkoid() directly if the PKO port identifier is
+ * known.
+ *
+ * @param ipd_port   The IPD port corresponding the to pko port the packet is for
+ * @param queue  Queue the packet is for
+ * @param len    Length of the command in 64 bit words
+ */
+static inline void cvmx_pko_doorbell(u64 ipd_port, u64 queue, u64 len)
+{
+	cvmx_pko_doorbell_address_t ptr;
+	u64 pko_port;
+
+	pko_port = ipd_port;
+	if (octeon_has_feature(OCTEON_FEATURE_PKND))
+		pko_port = cvmx_helper_cfg_ipd2pko_port_base(ipd_port);
+
+	ptr.u64 = 0;
+	ptr.s.mem_space = CVMX_IO_SEG;
+	ptr.s.did = CVMX_OCT_DID_PKT_SEND;
+	ptr.s.is_io = 1;
+	ptr.s.port = pko_port;
+	ptr.s.queue = queue;
+	/* Need to make sure output queue data is in DRAM before doorbell write */
+	CVMX_SYNCWS;
+	cvmx_write_io(ptr.u64, len);
+}
+
+/**
+ * Prepare to send a packet.  This may initiate a tag switch to
+ * get exclusive access to the output queue structure, and
+ * performs other prep work for the packet send operation.
+ *
+ * cvmx_pko_send_packet_finish() MUST be called after this function is called,
+ * and must be called with the same port/queue/use_locking arguments.
+ *
+ * The use_locking parameter allows the caller to use three
+ * possible locking modes.
+ * - CVMX_PKO_LOCK_NONE
+ *      - PKO doesn't do any locking. It is the responsibility
+ *          of the application to make sure that no other core
+ *          is accessing the same queue at the same time.
+ * - CVMX_PKO_LOCK_ATOMIC_TAG
+ *      - PKO performs an atomic tagswitch to insure exclusive
+ *          access to the output queue. This will maintain
+ *          packet ordering on output.
+ * - CVMX_PKO_LOCK_CMD_QUEUE
+ *      - PKO uses the common command queue locks to insure
+ *          exclusive access to the output queue. This is a
+ *          memory based ll/sc. This is the most portable
+ *          locking mechanism.
+ *
+ * NOTE: If atomic locking is used, the POW entry CANNOT be
+ * descheduled, as it does not contain a valid WQE pointer.
+ *
+ * @param port   Port to send it on, this can be either IPD port or PKO
+ *		 port.
+ * @param queue  Queue to use
+ * @param use_locking
+ *               CVMX_PKO_LOCK_NONE, CVMX_PKO_LOCK_ATOMIC_TAG, or CVMX_PKO_LOCK_CMD_QUEUE
+ */
+static inline void cvmx_pko_send_packet_prepare(u64 port __attribute__((unused)), u64 queue,
+						cvmx_pko_lock_t use_locking)
+{
+	if (use_locking == CVMX_PKO_LOCK_ATOMIC_TAG) {
+		/*
+		 * Must do a full switch here to handle all cases.  We use a
+		 * fake WQE pointer, as the POW does not access this memory.
+		 * The WQE pointer and group are only used if this work is
+		 * descheduled, which is not supported by the
+		 * cvmx_pko_send_packet_prepare/cvmx_pko_send_packet_finish
+		 * combination. Note that this is a special case in which these
+		 * fake values can be used - this is not a general technique.
+		 */
+		u32 tag = CVMX_TAG_SW_BITS_INTERNAL << CVMX_TAG_SW_SHIFT |
+			  CVMX_TAG_SUBGROUP_PKO << CVMX_TAG_SUBGROUP_SHIFT |
+			  (CVMX_TAG_SUBGROUP_MASK & queue);
+		cvmx_pow_tag_sw_full((cvmx_wqe_t *)cvmx_phys_to_ptr(0x80), tag,
+				     CVMX_POW_TAG_TYPE_ATOMIC, 0);
+	}
+}
+
+#define cvmx_pko_send_packet_prepare_pkoid cvmx_pko_send_packet_prepare
+
+/**
+ * Complete packet output. cvmx_pko_send_packet_prepare() must be called exactly once before this,
+ * and the same parameters must be passed to both cvmx_pko_send_packet_prepare() and
+ * cvmx_pko_send_packet_finish().
+ *
+ * WARNING: This function may have to look up the proper PKO port in
+ * the IPD port to PKO port map, and is thus slower than calling
+ * cvmx_pko_send_packet_finish_pkoid() directly if the PKO port
+ * identifier is known.
+ *
+ * @param ipd_port   The IPD port corresponding the to pko port the packet is for
+ * @param queue  Queue to use
+ * @param pko_command
+ *               PKO HW command word
+ * @param packet Packet to send
+ * @param use_locking
+ *               CVMX_PKO_LOCK_NONE, CVMX_PKO_LOCK_ATOMIC_TAG, or CVMX_PKO_LOCK_CMD_QUEUE
+ *
+ * @return returns CVMX_PKO_SUCCESS on success, or error code on failure of output
+ */
+static inline cvmx_pko_return_value_t
+cvmx_hwpko_send_packet_finish(u64 ipd_port, u64 queue, cvmx_pko_command_word0_t pko_command,
+			      cvmx_buf_ptr_t packet, cvmx_pko_lock_t use_locking)
+{
+	cvmx_cmd_queue_result_t result;
+
+	if (use_locking == CVMX_PKO_LOCK_ATOMIC_TAG)
+		cvmx_pow_tag_sw_wait();
+
+	result = cvmx_cmd_queue_write2(CVMX_CMD_QUEUE_PKO(queue),
+				       (use_locking == CVMX_PKO_LOCK_CMD_QUEUE), pko_command.u64,
+				       packet.u64);
+	if (cvmx_likely(result == CVMX_CMD_QUEUE_SUCCESS)) {
+		cvmx_pko_doorbell(ipd_port, queue, 2);
+		return CVMX_PKO_SUCCESS;
+	} else if ((result == CVMX_CMD_QUEUE_NO_MEMORY) || (result == CVMX_CMD_QUEUE_FULL)) {
+		return CVMX_PKO_NO_MEMORY;
+	} else {
+		return CVMX_PKO_INVALID_QUEUE;
+	}
+}
+
+/**
+ * Complete packet output. cvmx_pko_send_packet_prepare() must be called exactly once before this,
+ * and the same parameters must be passed to both cvmx_pko_send_packet_prepare() and
+ * cvmx_pko_send_packet_finish().
+ *
+ * WARNING: This function may have to look up the proper PKO port in
+ * the IPD port to PKO port map, and is thus slower than calling
+ * cvmx_pko_send_packet_finish3_pkoid() directly if the PKO port
+ * identifier is known.
+ *
+ * @param ipd_port   The IPD port corresponding the to pko port the packet is for
+ * @param queue  Queue to use
+ * @param pko_command
+ *               PKO HW command word
+ * @param packet Packet to send
+ * @param addr   Plysical address of a work queue entry or physical address to zero on complete.
+ * @param use_locking
+ *               CVMX_PKO_LOCK_NONE, CVMX_PKO_LOCK_ATOMIC_TAG, or CVMX_PKO_LOCK_CMD_QUEUE
+ *
+ * @return returns CVMX_PKO_SUCCESS on success, or error code on failure of output
+ */
+static inline cvmx_pko_return_value_t
+cvmx_hwpko_send_packet_finish3(u64 ipd_port, u64 queue, cvmx_pko_command_word0_t pko_command,
+			       cvmx_buf_ptr_t packet, u64 addr, cvmx_pko_lock_t use_locking)
+{
+	cvmx_cmd_queue_result_t result;
+
+	if (use_locking == CVMX_PKO_LOCK_ATOMIC_TAG)
+		cvmx_pow_tag_sw_wait();
+
+	result = cvmx_cmd_queue_write3(CVMX_CMD_QUEUE_PKO(queue),
+				       (use_locking == CVMX_PKO_LOCK_CMD_QUEUE), pko_command.u64,
+				       packet.u64, addr);
+	if (cvmx_likely(result == CVMX_CMD_QUEUE_SUCCESS)) {
+		cvmx_pko_doorbell(ipd_port, queue, 3);
+		return CVMX_PKO_SUCCESS;
+	} else if ((result == CVMX_CMD_QUEUE_NO_MEMORY) || (result == CVMX_CMD_QUEUE_FULL)) {
+		return CVMX_PKO_NO_MEMORY;
+	} else {
+		return CVMX_PKO_INVALID_QUEUE;
+	}
+}
+
+/**
+ * Get the first pko_port for the (interface, index)
+ *
+ * @param interface
+ * @param index
+ */
+int cvmx_pko_get_base_pko_port(int interface, int index);
+
+/**
+ * Get the number of pko_ports for the (interface, index)
+ *
+ * @param interface
+ * @param index
+ */
+int cvmx_pko_get_num_pko_ports(int interface, int index);
+
+/**
+ * For a given port number, return the base pko output queue
+ * for the port.
+ *
+ * @param port   IPD port number
+ * @return Base output queue
+ */
+int cvmx_pko_get_base_queue(int port);
+
+/**
+ * For a given port number, return the number of pko output queues.
+ *
+ * @param port   IPD port number
+ * @return Number of output queues
+ */
+int cvmx_pko_get_num_queues(int port);
+
+/**
+ * Sets the internal FPA pool data structure for PKO comamnd queue.
+ * @param pool	fpa pool number yo use
+ * @param buffer_size	buffer size of pool
+ * @param buffer_count	number of buufers to allocate to pool
+ *
+ * @note the caller is responsable for setting up the pool with
+ * an appropriate buffer size and sufficient buffer count.
+ */
+void cvmx_pko_set_cmd_que_pool_config(s64 pool, u64 buffer_size, u64 buffer_count);
+
+/**
+ * Get the status counters for a port.
+ *
+ * @param ipd_port Port number (ipd_port) to get statistics for.
+ * @param clear    Set to 1 to clear the counters after they are read
+ * @param status   Where to put the results.
+ *
+ * Note:
+ *     - Only the doorbell for the base queue of the ipd_port is
+ *       collected.
+ *     - Retrieving the stats involves writing the index through
+ *       CVMX_PKO_REG_READ_IDX and reading the stat CSRs, in that
+ *       order. It is not MP-safe and caller should guarantee
+ *       atomicity.
+ */
+void cvmx_pko_get_port_status(u64 ipd_port, u64 clear, cvmx_pko_port_status_t *status);
+
+/**
+ * Rate limit a PKO port to a max packets/sec. This function is only
+ * supported on CN57XX, CN56XX, CN55XX, and CN54XX.
+ *
+ * @param port      Port to rate limit
+ * @param packets_s Maximum packet/sec
+ * @param burst     Maximum number of packets to burst in a row before rate
+ *                  limiting cuts in.
+ *
+ * @return Zero on success, negative on failure
+ */
+int cvmx_pko_rate_limit_packets(int port, int packets_s, int burst);
+
+/**
+ * Rate limit a PKO port to a max bits/sec. This function is only
+ * supported on CN57XX, CN56XX, CN55XX, and CN54XX.
+ *
+ * @param port   Port to rate limit
+ * @param bits_s PKO rate limit in bits/sec
+ * @param burst  Maximum number of bits to burst before rate
+ *               limiting cuts in.
+ *
+ * @return Zero on success, negative on failure
+ */
+int cvmx_pko_rate_limit_bits(int port, u64 bits_s, int burst);
+
+/**
+ * @INTERNAL
+ *
+ * Retrieve the PKO pipe number for a port
+ *
+ * @param interface
+ * @param index
+ *
+ * @return negative on error.
+ *
+ * This applies only to the non-loopback interfaces.
+ *
+ */
+int __cvmx_pko_get_pipe(int interface, int index);
+
+/**
+ * For a given PKO port number, return the base output queue
+ * for the port.
+ *
+ * @param pko_port   PKO port number
+ * @return           Base output queue
+ */
+int cvmx_pko_get_base_queue_pkoid(int pko_port);
+
+/**
+ * For a given PKO port number, return the number of output queues
+ * for the port.
+ *
+ * @param pko_port	PKO port number
+ * @return		the number of output queues
+ */
+int cvmx_pko_get_num_queues_pkoid(int pko_port);
+
+/**
+ * Ring the packet output doorbell. This tells the packet
+ * output hardware that "len" command words have been added
+ * to its pending list.  This command includes the required
+ * CVMX_SYNCWS before the doorbell ring.
+ *
+ * @param pko_port   Port the packet is for
+ * @param queue  Queue the packet is for
+ * @param len    Length of the command in 64 bit words
+ */
+static inline void cvmx_pko_doorbell_pkoid(u64 pko_port, u64 queue, u64 len)
+{
+	cvmx_pko_doorbell_address_t ptr;
+
+	ptr.u64 = 0;
+	ptr.s.mem_space = CVMX_IO_SEG;
+	ptr.s.did = CVMX_OCT_DID_PKT_SEND;
+	ptr.s.is_io = 1;
+	ptr.s.port = pko_port;
+	ptr.s.queue = queue;
+	/* Need to make sure output queue data is in DRAM before doorbell write */
+	CVMX_SYNCWS;
+	cvmx_write_io(ptr.u64, len);
+}
+
+/**
+ * Complete packet output. cvmx_pko_send_packet_prepare() must be called exactly once before this,
+ * and the same parameters must be passed to both cvmx_pko_send_packet_prepare() and
+ * cvmx_pko_send_packet_finish_pkoid().
+ *
+ * @param pko_port   Port to send it on
+ * @param queue  Queue to use
+ * @param pko_command
+ *               PKO HW command word
+ * @param packet Packet to send
+ * @param use_locking
+ *               CVMX_PKO_LOCK_NONE, CVMX_PKO_LOCK_ATOMIC_TAG, or CVMX_PKO_LOCK_CMD_QUEUE
+ *
+ * @return returns CVMX_PKO_SUCCESS on success, or error code on failure of output
+ */
+static inline cvmx_pko_return_value_t
+cvmx_hwpko_send_packet_finish_pkoid(int pko_port, u64 queue, cvmx_pko_command_word0_t pko_command,
+				    cvmx_buf_ptr_t packet, cvmx_pko_lock_t use_locking)
+{
+	cvmx_cmd_queue_result_t result;
+
+	if (use_locking == CVMX_PKO_LOCK_ATOMIC_TAG)
+		cvmx_pow_tag_sw_wait();
+
+	result = cvmx_cmd_queue_write2(CVMX_CMD_QUEUE_PKO(queue),
+				       (use_locking == CVMX_PKO_LOCK_CMD_QUEUE), pko_command.u64,
+				       packet.u64);
+	if (cvmx_likely(result == CVMX_CMD_QUEUE_SUCCESS)) {
+		cvmx_pko_doorbell_pkoid(pko_port, queue, 2);
+		return CVMX_PKO_SUCCESS;
+	} else if ((result == CVMX_CMD_QUEUE_NO_MEMORY) || (result == CVMX_CMD_QUEUE_FULL)) {
+		return CVMX_PKO_NO_MEMORY;
+	} else {
+		return CVMX_PKO_INVALID_QUEUE;
+	}
+}
+
+/**
+ * Complete packet output. cvmx_pko_send_packet_prepare() must be called exactly once before this,
+ * and the same parameters must be passed to both cvmx_pko_send_packet_prepare() and
+ * cvmx_pko_send_packet_finish_pkoid().
+ *
+ * @param pko_port   The PKO port the packet is for
+ * @param queue  Queue to use
+ * @param pko_command
+ *               PKO HW command word
+ * @param packet Packet to send
+ * @param addr   Plysical address of a work queue entry or physical address to zero on complete.
+ * @param use_locking
+ *               CVMX_PKO_LOCK_NONE, CVMX_PKO_LOCK_ATOMIC_TAG, or CVMX_PKO_LOCK_CMD_QUEUE
+ *
+ * @return returns CVMX_PKO_SUCCESS on success, or error code on failure of output
+ */
+static inline cvmx_pko_return_value_t
+cvmx_hwpko_send_packet_finish3_pkoid(u64 pko_port, u64 queue, cvmx_pko_command_word0_t pko_command,
+				     cvmx_buf_ptr_t packet, u64 addr, cvmx_pko_lock_t use_locking)
+{
+	cvmx_cmd_queue_result_t result;
+
+	if (use_locking == CVMX_PKO_LOCK_ATOMIC_TAG)
+		cvmx_pow_tag_sw_wait();
+
+	result = cvmx_cmd_queue_write3(CVMX_CMD_QUEUE_PKO(queue),
+				       (use_locking == CVMX_PKO_LOCK_CMD_QUEUE), pko_command.u64,
+				       packet.u64, addr);
+	if (cvmx_likely(result == CVMX_CMD_QUEUE_SUCCESS)) {
+		cvmx_pko_doorbell_pkoid(pko_port, queue, 3);
+		return CVMX_PKO_SUCCESS;
+	} else if ((result == CVMX_CMD_QUEUE_NO_MEMORY) || (result == CVMX_CMD_QUEUE_FULL)) {
+		return CVMX_PKO_NO_MEMORY;
+	} else {
+		return CVMX_PKO_INVALID_QUEUE;
+	}
+}
+
+/*
+ * Obtain the number of PKO commands pending in a queue
+ *
+ * @param queue is the queue identifier to be queried
+ * @return the number of commands pending transmission or -1 on error
+ */
+int cvmx_pko_queue_pend_count(cvmx_cmd_queue_id_t queue);
+
+void cvmx_pko_set_cmd_queue_pool_buffer_count(u64 buffer_count);
+
+#endif /* __CVMX_HWPKO_H__ */
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-ilk.h b/arch/mips/mach-octeon/include/mach/cvmx-ilk.h
new file mode 100644
index 0000000000..727298352c
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-ilk.h
@@ -0,0 +1,154 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ *
+ * This file contains defines for the ILK interface
+ */
+
+#ifndef __CVMX_ILK_H__
+#define __CVMX_ILK_H__
+
+/* CSR typedefs have been moved to cvmx-ilk-defs.h */
+
+/*
+ * Note: this macro must match the first ilk port in the ipd_port_map_68xx[]
+ * and ipd_port_map_78xx[] arrays.
+ */
+static inline int CVMX_ILK_GBL_BASE(void)
+{
+	if (OCTEON_IS_MODEL(OCTEON_CN68XX))
+		return 5;
+	if (OCTEON_IS_MODEL(OCTEON_CN78XX))
+		return 6;
+	return -1;
+}
+
+static inline int CVMX_ILK_QLM_BASE(void)
+{
+	if (OCTEON_IS_MODEL(OCTEON_CN68XX))
+		return 1;
+	if (OCTEON_IS_MODEL(OCTEON_CN78XX))
+		return 4;
+	return -1;
+}
+
+typedef struct {
+	int intf_en : 1;
+	int la_mode : 1;
+	int reserved : 14; /* unused */
+	int lane_speed : 16;
+	/* add more here */
+} cvmx_ilk_intf_t;
+
+#define CVMX_NUM_ILK_INTF 2
+static inline int CVMX_ILK_MAX_LANES(void)
+{
+	if (OCTEON_IS_MODEL(OCTEON_CN68XX))
+		return 8;
+	if (OCTEON_IS_MODEL(OCTEON_CN78XX))
+		return 16;
+	return -1;
+}
+
+extern unsigned short cvmx_ilk_lane_mask[CVMX_MAX_NODES][CVMX_NUM_ILK_INTF];
+
+typedef struct {
+	unsigned int pipe;
+	unsigned int chan;
+} cvmx_ilk_pipe_chan_t;
+
+#define CVMX_ILK_MAX_PIPES 45
+/* Max number of channels allowed */
+#define CVMX_ILK_MAX_CHANS 256
+
+extern int cvmx_ilk_chans[CVMX_MAX_NODES][CVMX_NUM_ILK_INTF];
+
+typedef struct {
+	unsigned int chan;
+	unsigned int pknd;
+} cvmx_ilk_chan_pknd_t;
+
+#define CVMX_ILK_MAX_PKNDS 16 /* must be <45 */
+
+typedef struct {
+	int *chan_list; /* for discrete channels. or, must be null */
+	unsigned int num_chans;
+
+	unsigned int chan_start; /* for continuous channels */
+	unsigned int chan_end;
+	unsigned int chan_step;
+
+	unsigned int clr_on_rd;
+} cvmx_ilk_stats_ctrl_t;
+
+#define CVMX_ILK_MAX_CAL      288
+#define CVMX_ILK_MAX_CAL_IDX  (CVMX_ILK_MAX_CAL / 8)
+#define CVMX_ILK_TX_MIN_CAL   1
+#define CVMX_ILK_RX_MIN_CAL   1
+#define CVMX_ILK_CAL_GRP_SZ   8
+#define CVMX_ILK_PIPE_BPID_SZ 7
+#define CVMX_ILK_ENT_CTRL_SZ  2
+#define CVMX_ILK_RX_FIFO_WM   0x200
+
+typedef enum { PIPE_BPID = 0, LINK, XOFF, XON } cvmx_ilk_cal_ent_ctrl_t;
+
+typedef struct {
+	unsigned char pipe_bpid;
+	cvmx_ilk_cal_ent_ctrl_t ent_ctrl;
+} cvmx_ilk_cal_entry_t;
+
+typedef enum { CVMX_ILK_LPBK_DISA = 0, CVMX_ILK_LPBK_ENA } cvmx_ilk_lpbk_ena_t;
+
+typedef enum { CVMX_ILK_LPBK_INT = 0, CVMX_ILK_LPBK_EXT } cvmx_ilk_lpbk_mode_t;
+
+/**
+ * This header is placed in front of all received ILK look-aside mode packets
+ */
+typedef union {
+	u64 u64;
+
+	struct {
+		u32 reserved_63_57 : 7;	  /* bits 63...57 */
+		u32 nsp_cmd : 5;	  /* bits 56...52 */
+		u32 nsp_flags : 4;	  /* bits 51...48 */
+		u32 nsp_grp_id_upper : 6; /* bits 47...42 */
+		u32 reserved_41_40 : 2;	  /* bits 41...40 */
+		/* Protocol type, 1 for LA mode packet */
+		u32 la_mode : 1;	  /* bit  39      */
+		u32 nsp_grp_id_lower : 2; /* bits 38...37 */
+		u32 nsp_xid_upper : 4;	  /* bits 36...33 */
+		/* ILK channel number, 0 or 1 */
+		u32 ilk_channel : 1;   /* bit  32      */
+		u32 nsp_xid_lower : 8; /* bits 31...24 */
+		/* Unpredictable, may be any value */
+		u32 reserved_23_0 : 24; /* bits 23...0  */
+	} s;
+} cvmx_ilk_la_nsp_compact_hdr_t;
+
+typedef struct cvmx_ilk_LA_mode_struct {
+	int ilk_LA_mode;
+	int ilk_LA_mode_cal_ena;
+} cvmx_ilk_LA_mode_t;
+
+extern cvmx_ilk_LA_mode_t cvmx_ilk_LA_mode[CVMX_NUM_ILK_INTF];
+
+int cvmx_ilk_use_la_mode(int interface, int channel);
+int cvmx_ilk_start_interface(int interface, unsigned short num_lanes);
+int cvmx_ilk_start_interface_la(int interface, unsigned char num_lanes);
+int cvmx_ilk_set_pipe(int interface, int pipe_base, unsigned int pipe_len);
+int cvmx_ilk_tx_set_channel(int interface, cvmx_ilk_pipe_chan_t *pch, unsigned int num_chs);
+int cvmx_ilk_rx_set_pknd(int interface, cvmx_ilk_chan_pknd_t *chpknd, unsigned int num_pknd);
+int cvmx_ilk_enable(int interface);
+int cvmx_ilk_disable(int interface);
+int cvmx_ilk_get_intf_ena(int interface);
+int cvmx_ilk_get_chan_info(int interface, unsigned char **chans, unsigned char *num_chan);
+cvmx_ilk_la_nsp_compact_hdr_t cvmx_ilk_enable_la_header(int ipd_port, int mode);
+void cvmx_ilk_show_stats(int interface, cvmx_ilk_stats_ctrl_t *pstats);
+int cvmx_ilk_cal_setup_rx(int interface, int cal_depth, cvmx_ilk_cal_entry_t *pent, int hi_wm,
+			  unsigned char cal_ena);
+int cvmx_ilk_cal_setup_tx(int interface, int cal_depth, cvmx_ilk_cal_entry_t *pent,
+			  unsigned char cal_ena);
+int cvmx_ilk_lpbk(int interface, cvmx_ilk_lpbk_ena_t enable, cvmx_ilk_lpbk_mode_t mode);
+int cvmx_ilk_la_mode_enable_rx_calendar(int interface);
+
+#endif /* __CVMX_ILK_H__ */
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-ipd.h b/arch/mips/mach-octeon/include/mach/cvmx-ipd.h
new file mode 100644
index 0000000000..cdff36fffb
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-ipd.h
@@ -0,0 +1,233 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ *
+ * Interface to the hardware Input Packet Data unit.
+ */
+
+#ifndef __CVMX_IPD_H__
+#define __CVMX_IPD_H__
+
+#include "cvmx-pki.h"
+
+/* CSR typedefs have been moved to cvmx-ipd-defs.h */
+
+typedef cvmx_ipd_1st_mbuff_skip_t cvmx_ipd_mbuff_not_first_skip_t;
+typedef cvmx_ipd_1st_next_ptr_back_t cvmx_ipd_second_next_ptr_back_t;
+
+typedef struct cvmx_ipd_tag_fields {
+	u64 ipv6_src_ip : 1;
+	u64 ipv6_dst_ip : 1;
+	u64 ipv6_src_port : 1;
+	u64 ipv6_dst_port : 1;
+	u64 ipv6_next_header : 1;
+	u64 ipv4_src_ip : 1;
+	u64 ipv4_dst_ip : 1;
+	u64 ipv4_src_port : 1;
+	u64 ipv4_dst_port : 1;
+	u64 ipv4_protocol : 1;
+	u64 input_port : 1;
+} cvmx_ipd_tag_fields_t;
+
+typedef struct cvmx_pip_port_config {
+	u64 parse_mode;
+	u64 tag_type;
+	u64 tag_mode;
+	cvmx_ipd_tag_fields_t tag_fields;
+} cvmx_pip_port_config_t;
+
+typedef struct cvmx_ipd_config_struct {
+	u64 first_mbuf_skip;
+	u64 not_first_mbuf_skip;
+	u64 ipd_enable;
+	u64 enable_len_M8_fix;
+	u64 cache_mode;
+	cvmx_fpa_pool_config_t packet_pool;
+	cvmx_fpa_pool_config_t wqe_pool;
+	cvmx_pip_port_config_t port_config;
+} cvmx_ipd_config_t;
+
+extern cvmx_ipd_config_t cvmx_ipd_cfg;
+
+/**
+ * Gets the fpa pool number of packet pool
+ */
+static inline s64 cvmx_fpa_get_packet_pool(void)
+{
+	return (cvmx_ipd_cfg.packet_pool.pool_num);
+}
+
+/**
+ * Gets the buffer size of packet pool buffer
+ */
+static inline u64 cvmx_fpa_get_packet_pool_block_size(void)
+{
+	return (cvmx_ipd_cfg.packet_pool.buffer_size);
+}
+
+/**
+ * Gets the buffer count of packet pool
+ */
+static inline u64 cvmx_fpa_get_packet_pool_buffer_count(void)
+{
+	return (cvmx_ipd_cfg.packet_pool.buffer_count);
+}
+
+/**
+ * Gets the fpa pool number of wqe pool
+ */
+static inline s64 cvmx_fpa_get_wqe_pool(void)
+{
+	return (cvmx_ipd_cfg.wqe_pool.pool_num);
+}
+
+/**
+ * Gets the buffer size of wqe pool buffer
+ */
+static inline u64 cvmx_fpa_get_wqe_pool_block_size(void)
+{
+	return (cvmx_ipd_cfg.wqe_pool.buffer_size);
+}
+
+/**
+ * Gets the buffer count of wqe pool
+ */
+static inline u64 cvmx_fpa_get_wqe_pool_buffer_count(void)
+{
+	return (cvmx_ipd_cfg.wqe_pool.buffer_count);
+}
+
+/**
+ * Sets the ipd related configuration in internal structure which is then used
+ * for seting IPD hardware block
+ */
+int cvmx_ipd_set_config(cvmx_ipd_config_t ipd_config);
+
+/**
+ * Gets the ipd related configuration from internal structure.
+ */
+void cvmx_ipd_get_config(cvmx_ipd_config_t *ipd_config);
+
+/**
+ * Sets the internal FPA pool data structure for packet buffer pool.
+ * @param pool	fpa pool number yo use
+ * @param buffer_size	buffer size of pool
+ * @param buffer_count	number of buufers to allocate to pool
+ */
+void cvmx_ipd_set_packet_pool_config(s64 pool, u64 buffer_size, u64 buffer_count);
+
+/**
+ * Sets the internal FPA pool data structure for wqe pool.
+ * @param pool	fpa pool number yo use
+ * @param buffer_size	buffer size of pool
+ * @param buffer_count	number of buufers to allocate to pool
+ */
+void cvmx_ipd_set_wqe_pool_config(s64 pool, u64 buffer_size, u64 buffer_count);
+
+/**
+ * Gets the FPA packet buffer pool parameters.
+ */
+static inline void cvmx_fpa_get_packet_pool_config(s64 *pool, u64 *buffer_size, u64 *buffer_count)
+{
+	if (pool)
+		*pool = cvmx_ipd_cfg.packet_pool.pool_num;
+	if (buffer_size)
+		*buffer_size = cvmx_ipd_cfg.packet_pool.buffer_size;
+	if (buffer_count)
+		*buffer_count = cvmx_ipd_cfg.packet_pool.buffer_count;
+}
+
+/**
+ * Sets the FPA packet buffer pool parameters.
+ */
+static inline void cvmx_fpa_set_packet_pool_config(s64 pool, u64 buffer_size, u64 buffer_count)
+{
+	cvmx_ipd_set_packet_pool_config(pool, buffer_size, buffer_count);
+}
+
+/**
+ * Gets the FPA WQE pool parameters.
+ */
+static inline void cvmx_fpa_get_wqe_pool_config(s64 *pool, u64 *buffer_size, u64 *buffer_count)
+{
+	if (pool)
+		*pool = cvmx_ipd_cfg.wqe_pool.pool_num;
+	if (buffer_size)
+		*buffer_size = cvmx_ipd_cfg.wqe_pool.buffer_size;
+	if (buffer_count)
+		*buffer_count = cvmx_ipd_cfg.wqe_pool.buffer_count;
+}
+
+/**
+ * Sets the FPA WQE pool parameters.
+ */
+static inline void cvmx_fpa_set_wqe_pool_config(s64 pool, u64 buffer_size, u64 buffer_count)
+{
+	cvmx_ipd_set_wqe_pool_config(pool, buffer_size, buffer_count);
+}
+
+/**
+ * Configure IPD
+ *
+ * @param mbuff_size Packets buffer size in 8 byte words
+ * @param first_mbuff_skip
+ *                   Number of 8 byte words to skip in the first buffer
+ * @param not_first_mbuff_skip
+ *                   Number of 8 byte words to skip in each following buffer
+ * @param first_back Must be same as first_mbuff_skip / 128
+ * @param second_back
+ *                   Must be same as not_first_mbuff_skip / 128
+ * @param wqe_fpa_pool
+ *                   FPA pool to get work entries from
+ * @param cache_mode
+ * @param back_pres_enable_flag
+ *                   Enable or disable port back pressure at a global level.
+ *                   This should always be 1 as more accurate control can be
+ *                   found in IPD_PORTX_BP_PAGE_CNT[BP_ENB].
+ */
+void cvmx_ipd_config(u64 mbuff_size, u64 first_mbuff_skip, u64 not_first_mbuff_skip, u64 first_back,
+		     u64 second_back, u64 wqe_fpa_pool, cvmx_ipd_mode_t cache_mode,
+		     u64 back_pres_enable_flag);
+/**
+ * Enable IPD
+ */
+void cvmx_ipd_enable(void);
+
+/**
+ * Disable IPD
+ */
+void cvmx_ipd_disable(void);
+
+void __cvmx_ipd_free_ptr(void);
+
+void cvmx_ipd_set_packet_pool_buffer_count(u64 buffer_count);
+void cvmx_ipd_set_wqe_pool_buffer_count(u64 buffer_count);
+
+/**
+ * Setup Random Early Drop on a specific input queue
+ *
+ * @param queue  Input queue to setup RED on (0-7)
+ * @param pass_thresh
+ *               Packets will begin slowly dropping when there are less than
+ *               this many packet buffers free in FPA 0.
+ * @param drop_thresh
+ *               All incoming packets will be dropped when there are less
+ *               than this many free packet buffers in FPA 0.
+ * @return Zero on success. Negative on failure
+ */
+int cvmx_ipd_setup_red_queue(int queue, int pass_thresh, int drop_thresh);
+
+/**
+ * Setup Random Early Drop to automatically begin dropping packets.
+ *
+ * @param pass_thresh
+ *               Packets will begin slowly dropping when there are less than
+ *               this many packet buffers free in FPA 0.
+ * @param drop_thresh
+ *               All incoming packets will be dropped when there are less
+ *               than this many free packet buffers in FPA 0.
+ * @return Zero on success. Negative on failure
+ */
+int cvmx_ipd_setup_red(int pass_thresh, int drop_thresh);
+
+#endif /*  __CVMX_IPD_H__ */
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-packet.h b/arch/mips/mach-octeon/include/mach/cvmx-packet.h
new file mode 100644
index 0000000000..f3cfe9c64f
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-packet.h
@@ -0,0 +1,40 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ *
+ * Packet buffer defines.
+ */
+
+#ifndef __CVMX_PACKET_H__
+#define __CVMX_PACKET_H__
+
+union cvmx_buf_ptr_pki {
+	u64 u64;
+	struct {
+		u64 size : 16;
+		u64 packet_outside_wqe : 1;
+		u64 rsvd0 : 5;
+		u64 addr : 42;
+	};
+};
+
+typedef union cvmx_buf_ptr_pki cvmx_buf_ptr_pki_t;
+
+/**
+ * This structure defines a buffer pointer on Octeon
+ */
+union cvmx_buf_ptr {
+	void *ptr;
+	u64 u64;
+	struct {
+		u64 i : 1;
+		u64 back : 4;
+		u64 pool : 3;
+		u64 size : 16;
+		u64 addr : 40;
+	} s;
+};
+
+typedef union cvmx_buf_ptr cvmx_buf_ptr_t;
+
+#endif /*  __CVMX_PACKET_H__ */
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-pcie.h b/arch/mips/mach-octeon/include/mach/cvmx-pcie.h
new file mode 100644
index 0000000000..a819196c02
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-pcie.h
@@ -0,0 +1,279 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ */
+
+#ifndef __CVMX_PCIE_H__
+#define __CVMX_PCIE_H__
+
+#define CVMX_PCIE_MAX_PORTS 4
+#define CVMX_PCIE_PORTS                                                                            \
+	((OCTEON_IS_MODEL(OCTEON_CN78XX) || OCTEON_IS_MODEL(OCTEON_CN73XX)) ?                      \
+		       CVMX_PCIE_MAX_PORTS :                                                             \
+		       (OCTEON_IS_MODEL(OCTEON_CN70XX) ? 3 : 2))
+
+/*
+ * The physical memory base mapped by BAR1.  256MB at the end of the
+ * first 4GB.
+ */
+#define CVMX_PCIE_BAR1_PHYS_BASE ((1ull << 32) - (1ull << 28))
+#define CVMX_PCIE_BAR1_PHYS_SIZE BIT_ULL(28)
+
+/*
+ * The RC base of BAR1.  gen1 has a 39-bit BAR2, gen2 has 41-bit BAR2,
+ * place BAR1 so it is the same for both.
+ */
+#define CVMX_PCIE_BAR1_RC_BASE BIT_ULL(41)
+
+typedef union {
+	u64 u64;
+	struct {
+		u64 upper : 2;		 /* Normally 2 for XKPHYS */
+		u64 reserved_49_61 : 13; /* Must be zero */
+		u64 io : 1;		 /* 1 for IO space access */
+		u64 did : 5;		 /* PCIe DID = 3 */
+		u64 subdid : 3;		 /* PCIe SubDID = 1 */
+		u64 reserved_38_39 : 2;	 /* Must be zero */
+		u64 node : 2;		 /* Numa node number */
+		u64 es : 2;		 /* Endian swap = 1 */
+		u64 port : 2;		 /* PCIe port 0,1 */
+		u64 reserved_29_31 : 3;	 /* Must be zero */
+		u64 ty : 1;
+		u64 bus : 8;
+		u64 dev : 5;
+		u64 func : 3;
+		u64 reg : 12;
+	} config;
+	struct {
+		u64 upper : 2;		 /* Normally 2 for XKPHYS */
+		u64 reserved_49_61 : 13; /* Must be zero */
+		u64 io : 1;		 /* 1 for IO space access */
+		u64 did : 5;		 /* PCIe DID = 3 */
+		u64 subdid : 3;		 /* PCIe SubDID = 2 */
+		u64 reserved_38_39 : 2;	 /* Must be zero */
+		u64 node : 2;		 /* Numa node number */
+		u64 es : 2;		 /* Endian swap = 1 */
+		u64 port : 2;		 /* PCIe port 0,1 */
+		u64 address : 32;	 /* PCIe IO address */
+	} io;
+	struct {
+		u64 upper : 2;		 /* Normally 2 for XKPHYS */
+		u64 reserved_49_61 : 13; /* Must be zero */
+		u64 io : 1;		 /* 1 for IO space access */
+		u64 did : 5;		 /* PCIe DID = 3 */
+		u64 subdid : 3;		 /* PCIe SubDID = 3-6 */
+		u64 reserved_38_39 : 2;	 /* Must be zero */
+		u64 node : 2;		 /* Numa node number */
+		u64 address : 36;	 /* PCIe Mem address */
+	} mem;
+} cvmx_pcie_address_t;
+
+/**
+ * Return the Core virtual base address for PCIe IO access. IOs are
+ * read/written as an offset from this address.
+ *
+ * @param pcie_port PCIe port the IO is for
+ *
+ * @return 64bit Octeon IO base address for read/write
+ */
+u64 cvmx_pcie_get_io_base_address(int pcie_port);
+
+/**
+ * Size of the IO address region returned at address
+ * cvmx_pcie_get_io_base_address()
+ *
+ * @param pcie_port PCIe port the IO is for
+ *
+ * @return Size of the IO window
+ */
+u64 cvmx_pcie_get_io_size(int pcie_port);
+
+/**
+ * Return the Core virtual base address for PCIe MEM access. Memory is
+ * read/written as an offset from this address.
+ *
+ * @param pcie_port PCIe port the IO is for
+ *
+ * @return 64bit Octeon IO base address for read/write
+ */
+u64 cvmx_pcie_get_mem_base_address(int pcie_port);
+
+/**
+ * Size of the Mem address region returned at address
+ * cvmx_pcie_get_mem_base_address()
+ *
+ * @param pcie_port PCIe port the IO is for
+ *
+ * @return Size of the Mem window
+ */
+u64 cvmx_pcie_get_mem_size(int pcie_port);
+
+/**
+ * Initialize a PCIe port for use in host(RC) mode. It doesn't enumerate the bus.
+ *
+ * @param pcie_port PCIe port to initialize
+ *
+ * @return Zero on success
+ */
+int cvmx_pcie_rc_initialize(int pcie_port);
+
+/**
+ * Shutdown a PCIe port and put it in reset
+ *
+ * @param pcie_port PCIe port to shutdown
+ *
+ * @return Zero on success
+ */
+int cvmx_pcie_rc_shutdown(int pcie_port);
+
+/**
+ * Read 8bits from a Device's config space
+ *
+ * @param pcie_port PCIe port the device is on
+ * @param bus       Sub bus
+ * @param dev       Device ID
+ * @param fn        Device sub function
+ * @param reg       Register to access
+ *
+ * @return Result of the read
+ */
+u8 cvmx_pcie_config_read8(int pcie_port, int bus, int dev, int fn, int reg);
+
+/**
+ * Read 16bits from a Device's config space
+ *
+ * @param pcie_port PCIe port the device is on
+ * @param bus       Sub bus
+ * @param dev       Device ID
+ * @param fn        Device sub function
+ * @param reg       Register to access
+ *
+ * @return Result of the read
+ */
+u16 cvmx_pcie_config_read16(int pcie_port, int bus, int dev, int fn, int reg);
+
+/**
+ * Read 32bits from a Device's config space
+ *
+ * @param pcie_port PCIe port the device is on
+ * @param bus       Sub bus
+ * @param dev       Device ID
+ * @param fn        Device sub function
+ * @param reg       Register to access
+ *
+ * @return Result of the read
+ */
+u32 cvmx_pcie_config_read32(int pcie_port, int bus, int dev, int fn, int reg);
+
+/**
+ * Write 8bits to a Device's config space
+ *
+ * @param pcie_port PCIe port the device is on
+ * @param bus       Sub bus
+ * @param dev       Device ID
+ * @param fn        Device sub function
+ * @param reg       Register to access
+ * @param val       Value to write
+ */
+void cvmx_pcie_config_write8(int pcie_port, int bus, int dev, int fn, int reg, u8 val);
+
+/**
+ * Write 16bits to a Device's config space
+ *
+ * @param pcie_port PCIe port the device is on
+ * @param bus       Sub bus
+ * @param dev       Device ID
+ * @param fn        Device sub function
+ * @param reg       Register to access
+ * @param val       Value to write
+ */
+void cvmx_pcie_config_write16(int pcie_port, int bus, int dev, int fn, int reg, u16 val);
+
+/**
+ * Write 32bits to a Device's config space
+ *
+ * @param pcie_port PCIe port the device is on
+ * @param bus       Sub bus
+ * @param dev       Device ID
+ * @param fn        Device sub function
+ * @param reg       Register to access
+ * @param val       Value to write
+ */
+void cvmx_pcie_config_write32(int pcie_port, int bus, int dev, int fn, int reg, u32 val);
+
+/**
+ * Read a PCIe config space register indirectly. This is used for
+ * registers of the form PCIEEP_CFG??? and PCIERC?_CFG???.
+ *
+ * @param pcie_port  PCIe port to read from
+ * @param cfg_offset Address to read
+ *
+ * @return Value read
+ */
+u32 cvmx_pcie_cfgx_read(int pcie_port, u32 cfg_offset);
+u32 cvmx_pcie_cfgx_read_node(int node, int pcie_port, u32 cfg_offset);
+
+/**
+ * Write a PCIe config space register indirectly. This is used for
+ * registers of the form PCIEEP_CFG??? and PCIERC?_CFG???.
+ *
+ * @param pcie_port  PCIe port to write to
+ * @param cfg_offset Address to write
+ * @param val        Value to write
+ */
+void cvmx_pcie_cfgx_write(int pcie_port, u32 cfg_offset, u32 val);
+void cvmx_pcie_cfgx_write_node(int node, int pcie_port, u32 cfg_offset, u32 val);
+
+/**
+ * Write a 32bit value to the Octeon NPEI register space
+ *
+ * @param address Address to write to
+ * @param val     Value to write
+ */
+static inline void cvmx_pcie_npei_write32(u64 address, u32 val)
+{
+	cvmx_write64_uint32(address ^ 4, val);
+	cvmx_read64_uint32(address ^ 4);
+}
+
+/**
+ * Read a 32bit value from the Octeon NPEI register space
+ *
+ * @param address Address to read
+ * @return The result
+ */
+static inline u32 cvmx_pcie_npei_read32(u64 address)
+{
+	return cvmx_read64_uint32(address ^ 4);
+}
+
+/**
+ * Initialize a PCIe port for use in target(EP) mode.
+ *
+ * @param pcie_port PCIe port to initialize
+ *
+ * @return Zero on success
+ */
+int cvmx_pcie_ep_initialize(int pcie_port);
+
+/**
+ * Wait for posted PCIe read/writes to reach the other side of
+ * the internal PCIe switch. This will insure that core
+ * read/writes are posted before anything after this function
+ * is called. This may be necessary when writing to memory that
+ * will later be read using the DMA/PKT engines.
+ *
+ * @param pcie_port PCIe port to wait for
+ */
+void cvmx_pcie_wait_for_pending(int pcie_port);
+
+/**
+ * Returns if a PCIe port is in host or target mode.
+ *
+ * @param pcie_port PCIe port number (PEM number)
+ *
+ * @return 0 if PCIe port is in target mode, !0 if in host mode.
+ */
+int cvmx_pcie_is_host_mode(int pcie_port);
+
+#endif
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-pip.h b/arch/mips/mach-octeon/include/mach/cvmx-pip.h
new file mode 100644
index 0000000000..013f533fb7
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-pip.h
@@ -0,0 +1,1080 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ *
+ * Interface to the hardware Packet Input Processing unit.
+ */
+
+#ifndef __CVMX_PIP_H__
+#define __CVMX_PIP_H__
+
+#include "cvmx-wqe.h"
+#include "cvmx-pki.h"
+#include "cvmx-helper-pki.h"
+
+#include "cvmx-helper.h"
+#include "cvmx-helper-util.h"
+#include "cvmx-pki-resources.h"
+
+#define CVMX_PIP_NUM_INPUT_PORTS 46
+#define CVMX_PIP_NUM_WATCHERS	 8
+
+/*
+ * Encodes the different error and exception codes
+ */
+typedef enum {
+	CVMX_PIP_L4_NO_ERR = 0ull,
+	/*        1  = TCP (UDP) packet not long enough to cover TCP (UDP) header */
+	CVMX_PIP_L4_MAL_ERR = 1ull,
+	/*        2  = TCP/UDP checksum failure */
+	CVMX_PIP_CHK_ERR = 2ull,
+	/*        3  = TCP/UDP length check (TCP/UDP length does not match IP length) */
+	CVMX_PIP_L4_LENGTH_ERR = 3ull,
+	/*        4  = illegal TCP/UDP port (either source or dest port is zero) */
+	CVMX_PIP_BAD_PRT_ERR = 4ull,
+	/*        8  = TCP flags = FIN only */
+	CVMX_PIP_TCP_FLG8_ERR = 8ull,
+	/*        9  = TCP flags = 0 */
+	CVMX_PIP_TCP_FLG9_ERR = 9ull,
+	/*        10 = TCP flags = FIN+RST+* */
+	CVMX_PIP_TCP_FLG10_ERR = 10ull,
+	/*        11 = TCP flags = SYN+URG+* */
+	CVMX_PIP_TCP_FLG11_ERR = 11ull,
+	/*        12 = TCP flags = SYN+RST+* */
+	CVMX_PIP_TCP_FLG12_ERR = 12ull,
+	/*        13 = TCP flags = SYN+FIN+* */
+	CVMX_PIP_TCP_FLG13_ERR = 13ull
+} cvmx_pip_l4_err_t;
+
+typedef enum {
+	CVMX_PIP_IP_NO_ERR = 0ull,
+	/*        1 = not IPv4 or IPv6 */
+	CVMX_PIP_NOT_IP = 1ull,
+	/*        2 = IPv4 header checksum violation */
+	CVMX_PIP_IPV4_HDR_CHK = 2ull,
+	/*        3 = malformed (packet not long enough to cover IP hdr) */
+	CVMX_PIP_IP_MAL_HDR = 3ull,
+	/*        4 = malformed (packet not long enough to cover len in IP hdr) */
+	CVMX_PIP_IP_MAL_PKT = 4ull,
+	/*        5 = TTL / hop count equal zero */
+	CVMX_PIP_TTL_HOP = 5ull,
+	/*        6 = IPv4 options / IPv6 early extension headers */
+	CVMX_PIP_OPTS = 6ull
+} cvmx_pip_ip_exc_t;
+
+/**
+ * NOTES
+ *       late collision (data received before collision)
+ *            late collisions cannot be detected by the receiver
+ *            they would appear as JAM bits which would appear as bad FCS
+ *            or carrier extend error which is CVMX_PIP_EXTEND_ERR
+ */
+typedef enum {
+	/**
+	 * No error
+	 */
+	CVMX_PIP_RX_NO_ERR = 0ull,
+
+	CVMX_PIP_PARTIAL_ERR =
+		1ull, /* RGM+SPI            1 = partially received packet (buffering/bandwidth not adequate) */
+	CVMX_PIP_JABBER_ERR =
+		2ull, /* RGM+SPI            2 = receive packet too large and truncated */
+	CVMX_PIP_OVER_FCS_ERR =
+		3ull, /* RGM                3 = max frame error (pkt len > max frame len) (with FCS error) */
+	CVMX_PIP_OVER_ERR =
+		4ull, /* RGM+SPI            4 = max frame error (pkt len > max frame len) */
+	CVMX_PIP_ALIGN_ERR =
+		5ull, /* RGM                5 = nibble error (data not byte multiple - 100M and 10M only) */
+	CVMX_PIP_UNDER_FCS_ERR =
+		6ull, /* RGM                6 = min frame error (pkt len < min frame len) (with FCS error) */
+	CVMX_PIP_GMX_FCS_ERR = 7ull, /* RGM                7 = FCS error */
+	CVMX_PIP_UNDER_ERR =
+		8ull, /* RGM+SPI            8 = min frame error (pkt len < min frame len) */
+	CVMX_PIP_EXTEND_ERR = 9ull, /* RGM                9 = Frame carrier extend error */
+	CVMX_PIP_TERMINATE_ERR =
+		9ull, /* XAUI               9 = Packet was terminated with an idle cycle */
+	CVMX_PIP_LENGTH_ERR =
+		10ull, /* RGM               10 = length mismatch (len did not match len in L2 length/type) */
+	CVMX_PIP_DAT_ERR =
+		11ull, /* RGM               11 = Frame error (some or all data bits marked err) */
+	CVMX_PIP_DIP_ERR = 11ull, /*     SPI           11 = DIP4 error */
+	CVMX_PIP_SKIP_ERR =
+		12ull, /* RGM               12 = packet was not large enough to pass the skipper - no inspection could occur */
+	CVMX_PIP_NIBBLE_ERR =
+		13ull, /* RGM               13 = studder error (data not repeated - 100M and 10M only) */
+	CVMX_PIP_PIP_FCS = 16L, /* RGM+SPI           16 = FCS error */
+	CVMX_PIP_PIP_SKIP_ERR =
+		17L, /* RGM+SPI+PCI       17 = packet was not large enough to pass the skipper - no inspection could occur */
+	CVMX_PIP_PIP_L2_MAL_HDR =
+		18L, /* RGM+SPI+PCI       18 = malformed l2 (packet not long enough to cover L2 hdr) */
+	CVMX_PIP_PUNY_ERR =
+		47L /* SGMII             47 = PUNY error (packet was 4B or less when FCS stripping is enabled) */
+	/* NOTES
+	 *       xx = late collision (data received before collision)
+	 *            late collisions cannot be detected by the receiver
+	 *            they would appear as JAM bits which would appear as bad FCS
+	 *            or carrier extend error which is CVMX_PIP_EXTEND_ERR
+	 */
+} cvmx_pip_rcv_err_t;
+
+/**
+ * This defines the err_code field errors in the work Q entry
+ */
+typedef union {
+	cvmx_pip_l4_err_t l4_err;
+	cvmx_pip_ip_exc_t ip_exc;
+	cvmx_pip_rcv_err_t rcv_err;
+} cvmx_pip_err_t;
+
+/**
+ * Status statistics for a port
+ */
+typedef struct {
+	u64 dropped_octets;
+	u64 dropped_packets;
+	u64 pci_raw_packets;
+	u64 octets;
+	u64 packets;
+	u64 multicast_packets;
+	u64 broadcast_packets;
+	u64 len_64_packets;
+	u64 len_65_127_packets;
+	u64 len_128_255_packets;
+	u64 len_256_511_packets;
+	u64 len_512_1023_packets;
+	u64 len_1024_1518_packets;
+	u64 len_1519_max_packets;
+	u64 fcs_align_err_packets;
+	u64 runt_packets;
+	u64 runt_crc_packets;
+	u64 oversize_packets;
+	u64 oversize_crc_packets;
+	u64 inb_packets;
+	u64 inb_octets;
+	u64 inb_errors;
+	u64 mcast_l2_red_packets;
+	u64 bcast_l2_red_packets;
+	u64 mcast_l3_red_packets;
+	u64 bcast_l3_red_packets;
+} cvmx_pip_port_status_t;
+
+/**
+ * Definition of the PIP custom header that can be prepended
+ * to a packet by external hardware.
+ */
+typedef union {
+	u64 u64;
+	struct {
+		u64 rawfull : 1;
+		u64 reserved0 : 5;
+		cvmx_pip_port_parse_mode_t parse_mode : 2;
+		u64 reserved1 : 1;
+		u64 skip_len : 7;
+		u64 grpext : 2;
+		u64 nqos : 1;
+		u64 ngrp : 1;
+		u64 ntt : 1;
+		u64 ntag : 1;
+		u64 qos : 3;
+		u64 grp : 4;
+		u64 rs : 1;
+		cvmx_pow_tag_type_t tag_type : 2;
+		u64 tag : 32;
+	} s;
+} cvmx_pip_pkt_inst_hdr_t;
+
+enum cvmx_pki_pcam_match {
+	CVMX_PKI_PCAM_MATCH_IP,
+	CVMX_PKI_PCAM_MATCH_IPV4,
+	CVMX_PKI_PCAM_MATCH_IPV6,
+	CVMX_PKI_PCAM_MATCH_TCP
+};
+
+/* CSR typedefs have been moved to cvmx-pip-defs.h */
+static inline int cvmx_pip_config_watcher(int index, int type, u16 match, u16 mask, int grp,
+					  int qos)
+{
+	if (index >= CVMX_PIP_NUM_WATCHERS) {
+		debug("ERROR: pip watcher %d is > than supported\n", index);
+		return -1;
+	}
+	if (octeon_has_feature(OCTEON_FEATURE_PKI)) {
+		/* store in software for now, only when the watcher is enabled program the entry*/
+		if (type == CVMX_PIP_QOS_WATCH_PROTNH) {
+			qos_watcher[index].field = CVMX_PKI_PCAM_TERM_L3_FLAGS;
+			qos_watcher[index].data = (u32)(match << 16);
+			qos_watcher[index].data_mask = (u32)(mask << 16);
+			qos_watcher[index].advance = 0;
+		} else if (type == CVMX_PIP_QOS_WATCH_TCP) {
+			qos_watcher[index].field = CVMX_PKI_PCAM_TERM_L4_PORT;
+			qos_watcher[index].data = 0x060000;
+			qos_watcher[index].data |= (u32)match;
+			qos_watcher[index].data_mask = (u32)(mask);
+			qos_watcher[index].advance = 0;
+		} else if (type == CVMX_PIP_QOS_WATCH_UDP) {
+			qos_watcher[index].field = CVMX_PKI_PCAM_TERM_L4_PORT;
+			qos_watcher[index].data = 0x110000;
+			qos_watcher[index].data |= (u32)match;
+			qos_watcher[index].data_mask = (u32)(mask);
+			qos_watcher[index].advance = 0;
+		} else if (type == 0x4 /*CVMX_PIP_QOS_WATCH_ETHERTYPE*/) {
+			qos_watcher[index].field = CVMX_PKI_PCAM_TERM_ETHTYPE0;
+			if (match == 0x8100) {
+				debug("ERROR: default vlan entry already exist, cant set watcher\n");
+				return -1;
+			}
+			qos_watcher[index].data = (u32)(match << 16);
+			qos_watcher[index].data_mask = (u32)(mask << 16);
+			qos_watcher[index].advance = 4;
+		} else {
+			debug("ERROR: Unsupported watcher type %d\n", type);
+			return -1;
+		}
+		if (grp >= 32) {
+			debug("ERROR: grp %d out of range for backward compat 78xx\n", grp);
+			return -1;
+		}
+		qos_watcher[index].sso_grp = (u8)(grp << 3 | qos);
+		qos_watcher[index].configured = 1;
+	} else {
+		/* Implement it later */
+	}
+	return 0;
+}
+
+static inline int __cvmx_pip_set_tag_type(int node, int style, int tag_type, int field)
+{
+	struct cvmx_pki_style_config style_cfg;
+	int style_num;
+	int pcam_offset;
+	int bank;
+	struct cvmx_pki_pcam_input pcam_input;
+	struct cvmx_pki_pcam_action pcam_action;
+
+	/* All other style parameters remain same except tag type */
+	cvmx_pki_read_style_config(node, style, CVMX_PKI_CLUSTER_ALL, &style_cfg);
+	style_cfg.parm_cfg.tag_type = (enum cvmx_sso_tag_type)tag_type;
+	style_num = cvmx_pki_style_alloc(node, -1);
+	if (style_num < 0) {
+		debug("ERROR: style not available to set tag type\n");
+		return -1;
+	}
+	cvmx_pki_write_style_config(node, style_num, CVMX_PKI_CLUSTER_ALL, &style_cfg);
+	memset(&pcam_input, 0, sizeof(pcam_input));
+	memset(&pcam_action, 0, sizeof(pcam_action));
+	pcam_input.style = style;
+	pcam_input.style_mask = 0xff;
+	if (field == CVMX_PKI_PCAM_MATCH_IP) {
+		pcam_input.field = CVMX_PKI_PCAM_TERM_ETHTYPE0;
+		pcam_input.field_mask = 0xff;
+		pcam_input.data = 0x08000000;
+		pcam_input.data_mask = 0xffff0000;
+		pcam_action.pointer_advance = 4;
+		/* legacy will write to all clusters*/
+		bank = 0;
+		pcam_offset = cvmx_pki_pcam_entry_alloc(node, CVMX_PKI_FIND_AVAL_ENTRY, bank,
+							CVMX_PKI_CLUSTER_ALL);
+		if (pcam_offset < 0) {
+			debug("ERROR: pcam entry not available to enable qos watcher\n");
+			cvmx_pki_style_free(node, style_num);
+			return -1;
+		}
+		pcam_action.parse_mode_chg = CVMX_PKI_PARSE_NO_CHG;
+		pcam_action.layer_type_set = CVMX_PKI_LTYPE_E_NONE;
+		pcam_action.style_add = (u8)(style_num - style);
+		cvmx_pki_pcam_write_entry(node, pcam_offset, CVMX_PKI_CLUSTER_ALL, pcam_input,
+					  pcam_action);
+		field = CVMX_PKI_PCAM_MATCH_IPV6;
+	}
+	if (field == CVMX_PKI_PCAM_MATCH_IPV4) {
+		pcam_input.field = CVMX_PKI_PCAM_TERM_ETHTYPE0;
+		pcam_input.field_mask = 0xff;
+		pcam_input.data = 0x08000000;
+		pcam_input.data_mask = 0xffff0000;
+		pcam_action.pointer_advance = 4;
+	} else if (field == CVMX_PKI_PCAM_MATCH_IPV6) {
+		pcam_input.field = CVMX_PKI_PCAM_TERM_ETHTYPE0;
+		pcam_input.field_mask = 0xff;
+		pcam_input.data = 0x86dd00000;
+		pcam_input.data_mask = 0xffff0000;
+		pcam_action.pointer_advance = 4;
+	} else if (field == CVMX_PKI_PCAM_MATCH_TCP) {
+		pcam_input.field = CVMX_PKI_PCAM_TERM_L4_PORT;
+		pcam_input.field_mask = 0xff;
+		pcam_input.data = 0x60000;
+		pcam_input.data_mask = 0xff0000;
+		pcam_action.pointer_advance = 0;
+	}
+	pcam_action.parse_mode_chg = CVMX_PKI_PARSE_NO_CHG;
+	pcam_action.layer_type_set = CVMX_PKI_LTYPE_E_NONE;
+	pcam_action.style_add = (u8)(style_num - style);
+	bank = pcam_input.field & 0x01;
+	pcam_offset = cvmx_pki_pcam_entry_alloc(node, CVMX_PKI_FIND_AVAL_ENTRY, bank,
+						CVMX_PKI_CLUSTER_ALL);
+	if (pcam_offset < 0) {
+		debug("ERROR: pcam entry not available to enable qos watcher\n");
+		cvmx_pki_style_free(node, style_num);
+		return -1;
+	}
+	cvmx_pki_pcam_write_entry(node, pcam_offset, CVMX_PKI_CLUSTER_ALL, pcam_input, pcam_action);
+	return style_num;
+}
+
+/* Only for legacy internal use */
+static inline int __cvmx_pip_enable_watcher_78xx(int node, int index, int style)
+{
+	struct cvmx_pki_style_config style_cfg;
+	struct cvmx_pki_qpg_config qpg_cfg;
+	struct cvmx_pki_pcam_input pcam_input;
+	struct cvmx_pki_pcam_action pcam_action;
+	int style_num;
+	int qpg_offset;
+	int pcam_offset;
+	int bank;
+
+	if (!qos_watcher[index].configured) {
+		debug("ERROR: qos watcher %d should be configured before enable\n", index);
+		return -1;
+	}
+	/* All other style parameters remain same except grp and qos and qps base */
+	cvmx_pki_read_style_config(node, style, CVMX_PKI_CLUSTER_ALL, &style_cfg);
+	cvmx_pki_read_qpg_entry(node, style_cfg.parm_cfg.qpg_base, &qpg_cfg);
+	qpg_cfg.qpg_base = CVMX_PKI_FIND_AVAL_ENTRY;
+	qpg_cfg.grp_ok = qos_watcher[index].sso_grp;
+	qpg_cfg.grp_bad = qos_watcher[index].sso_grp;
+	qpg_offset = cvmx_helper_pki_set_qpg_entry(node, &qpg_cfg);
+	if (qpg_offset == -1) {
+		debug("Warning: no new qpg entry available to enable watcher\n");
+		return -1;
+	}
+	/* try to reserve the style, if it is not configured already, reserve
+	   and configure it */
+	style_cfg.parm_cfg.qpg_base = qpg_offset;
+	style_num = cvmx_pki_style_alloc(node, -1);
+	if (style_num < 0) {
+		debug("ERROR: style not available to enable qos watcher\n");
+		cvmx_pki_qpg_entry_free(node, qpg_offset, 1);
+		return -1;
+	}
+	cvmx_pki_write_style_config(node, style_num, CVMX_PKI_CLUSTER_ALL, &style_cfg);
+	/* legacy will write to all clusters*/
+	bank = qos_watcher[index].field & 0x01;
+	pcam_offset = cvmx_pki_pcam_entry_alloc(node, CVMX_PKI_FIND_AVAL_ENTRY, bank,
+						CVMX_PKI_CLUSTER_ALL);
+	if (pcam_offset < 0) {
+		debug("ERROR: pcam entry not available to enable qos watcher\n");
+		cvmx_pki_style_free(node, style_num);
+		cvmx_pki_qpg_entry_free(node, qpg_offset, 1);
+		return -1;
+	}
+	memset(&pcam_input, 0, sizeof(pcam_input));
+	memset(&pcam_action, 0, sizeof(pcam_action));
+	pcam_input.style = style;
+	pcam_input.style_mask = 0xff;
+	pcam_input.field = qos_watcher[index].field;
+	pcam_input.field_mask = 0xff;
+	pcam_input.data = qos_watcher[index].data;
+	pcam_input.data_mask = qos_watcher[index].data_mask;
+	pcam_action.parse_mode_chg = CVMX_PKI_PARSE_NO_CHG;
+	pcam_action.layer_type_set = CVMX_PKI_LTYPE_E_NONE;
+	pcam_action.style_add = (u8)(style_num - style);
+	pcam_action.pointer_advance = qos_watcher[index].advance;
+	cvmx_pki_pcam_write_entry(node, pcam_offset, CVMX_PKI_CLUSTER_ALL, pcam_input, pcam_action);
+	return 0;
+}
+
+/**
+ * Configure an ethernet input port
+ *
+ * @param ipd_port Port number to configure
+ * @param port_cfg Port hardware configuration
+ * @param port_tag_cfg Port POW tagging configuration
+ */
+static inline void cvmx_pip_config_port(u64 ipd_port, cvmx_pip_prt_cfgx_t port_cfg,
+					cvmx_pip_prt_tagx_t port_tag_cfg)
+{
+	struct cvmx_pki_qpg_config qpg_cfg;
+	int qpg_offset;
+	u8 tcp_tag = 0xff;
+	u8 ip_tag = 0xaa;
+	int style, nstyle, n4style, n6style;
+
+	if (octeon_has_feature(OCTEON_FEATURE_PKI)) {
+		struct cvmx_pki_port_config pki_prt_cfg;
+		struct cvmx_xport xp = cvmx_helper_ipd_port_to_xport(ipd_port);
+
+		cvmx_pki_get_port_config(ipd_port, &pki_prt_cfg);
+		style = pki_prt_cfg.pkind_cfg.initial_style;
+		if (port_cfg.s.ih_pri || port_cfg.s.vlan_len || port_cfg.s.pad_len)
+			debug("Warning: 78xx: use different config for this option\n");
+		pki_prt_cfg.style_cfg.parm_cfg.minmax_sel = port_cfg.s.len_chk_sel;
+		pki_prt_cfg.style_cfg.parm_cfg.lenerr_en = port_cfg.s.lenerr_en;
+		pki_prt_cfg.style_cfg.parm_cfg.maxerr_en = port_cfg.s.maxerr_en;
+		pki_prt_cfg.style_cfg.parm_cfg.minerr_en = port_cfg.s.minerr_en;
+		pki_prt_cfg.style_cfg.parm_cfg.fcs_chk = port_cfg.s.crc_en;
+		if (port_cfg.s.grp_wat || port_cfg.s.qos_wat || port_cfg.s.grp_wat_47 ||
+		    port_cfg.s.qos_wat_47) {
+			u8 group_mask = (u8)(port_cfg.s.grp_wat | (u8)(port_cfg.s.grp_wat_47 << 4));
+			u8 qos_mask = (u8)(port_cfg.s.qos_wat | (u8)(port_cfg.s.qos_wat_47 << 4));
+			int i;
+
+			for (i = 0; i < CVMX_PIP_NUM_WATCHERS; i++) {
+				if ((group_mask & (1 << i)) || (qos_mask & (1 << i)))
+					__cvmx_pip_enable_watcher_78xx(xp.node, i, style);
+			}
+		}
+		if (port_tag_cfg.s.tag_mode) {
+			if (OCTEON_IS_MODEL(OCTEON_CN78XX_PASS1_X))
+				cvmx_printf("Warning: mask tag is not supported in 78xx pass1\n");
+			else {
+			}
+			/* need to implement for 78xx*/
+		}
+		if (port_cfg.s.tag_inc)
+			debug("Warning: 78xx uses differnet method for tag generation\n");
+		pki_prt_cfg.style_cfg.parm_cfg.rawdrp = port_cfg.s.rawdrp;
+		pki_prt_cfg.pkind_cfg.parse_en.inst_hdr = port_cfg.s.inst_hdr;
+		if (port_cfg.s.hg_qos)
+			pki_prt_cfg.style_cfg.parm_cfg.qpg_qos = CVMX_PKI_QPG_QOS_HIGIG;
+		else if (port_cfg.s.qos_vlan)
+			pki_prt_cfg.style_cfg.parm_cfg.qpg_qos = CVMX_PKI_QPG_QOS_VLAN;
+		else if (port_cfg.s.qos_diff)
+			pki_prt_cfg.style_cfg.parm_cfg.qpg_qos = CVMX_PKI_QPG_QOS_DIFFSERV;
+		if (port_cfg.s.qos_vod)
+			debug("Warning: 78xx needs pcam entries installed to achieve qos_vod\n");
+		if (port_cfg.s.qos) {
+			cvmx_pki_read_qpg_entry(xp.node, pki_prt_cfg.style_cfg.parm_cfg.qpg_base,
+						&qpg_cfg);
+			qpg_cfg.qpg_base = CVMX_PKI_FIND_AVAL_ENTRY;
+			qpg_cfg.grp_ok |= port_cfg.s.qos;
+			qpg_cfg.grp_bad |= port_cfg.s.qos;
+			qpg_offset = cvmx_helper_pki_set_qpg_entry(xp.node, &qpg_cfg);
+			if (qpg_offset == -1)
+				debug("Warning: no new qpg entry available, will not modify qos\n");
+			else
+				pki_prt_cfg.style_cfg.parm_cfg.qpg_base = qpg_offset;
+		}
+		if (port_tag_cfg.s.grp != pki_dflt_sso_grp[xp.node].group) {
+			cvmx_pki_read_qpg_entry(xp.node, pki_prt_cfg.style_cfg.parm_cfg.qpg_base,
+						&qpg_cfg);
+			qpg_cfg.qpg_base = CVMX_PKI_FIND_AVAL_ENTRY;
+			qpg_cfg.grp_ok |= (u8)(port_tag_cfg.s.grp << 3);
+			qpg_cfg.grp_bad |= (u8)(port_tag_cfg.s.grp << 3);
+			qpg_offset = cvmx_helper_pki_set_qpg_entry(xp.node, &qpg_cfg);
+			if (qpg_offset == -1)
+				debug("Warning: no new qpg entry available, will not modify group\n");
+			else
+				pki_prt_cfg.style_cfg.parm_cfg.qpg_base = qpg_offset;
+		}
+		pki_prt_cfg.pkind_cfg.parse_en.dsa_en = port_cfg.s.dsa_en;
+		pki_prt_cfg.pkind_cfg.parse_en.hg_en = port_cfg.s.higig_en;
+		pki_prt_cfg.style_cfg.tag_cfg.tag_fields.layer_c_src =
+			port_tag_cfg.s.ip6_src_flag | port_tag_cfg.s.ip4_src_flag;
+		pki_prt_cfg.style_cfg.tag_cfg.tag_fields.layer_c_dst =
+			port_tag_cfg.s.ip6_dst_flag | port_tag_cfg.s.ip4_dst_flag;
+		pki_prt_cfg.style_cfg.tag_cfg.tag_fields.ip_prot_nexthdr =
+			port_tag_cfg.s.ip6_nxth_flag | port_tag_cfg.s.ip4_pctl_flag;
+		pki_prt_cfg.style_cfg.tag_cfg.tag_fields.layer_d_src =
+			port_tag_cfg.s.ip6_sprt_flag | port_tag_cfg.s.ip4_sprt_flag;
+		pki_prt_cfg.style_cfg.tag_cfg.tag_fields.layer_d_dst =
+			port_tag_cfg.s.ip6_dprt_flag | port_tag_cfg.s.ip4_dprt_flag;
+		pki_prt_cfg.style_cfg.tag_cfg.tag_fields.input_port = port_tag_cfg.s.inc_prt_flag;
+		pki_prt_cfg.style_cfg.tag_cfg.tag_fields.first_vlan = port_tag_cfg.s.inc_vlan;
+		pki_prt_cfg.style_cfg.tag_cfg.tag_fields.second_vlan = port_tag_cfg.s.inc_vs;
+
+		if (port_tag_cfg.s.tcp6_tag_type == port_tag_cfg.s.tcp4_tag_type)
+			tcp_tag = port_tag_cfg.s.tcp6_tag_type;
+		if (port_tag_cfg.s.ip6_tag_type == port_tag_cfg.s.ip4_tag_type)
+			ip_tag = port_tag_cfg.s.ip6_tag_type;
+		pki_prt_cfg.style_cfg.parm_cfg.tag_type =
+			(enum cvmx_sso_tag_type)port_tag_cfg.s.non_tag_type;
+		if (tcp_tag == ip_tag && tcp_tag == port_tag_cfg.s.non_tag_type)
+			pki_prt_cfg.style_cfg.parm_cfg.tag_type = (enum cvmx_sso_tag_type)tcp_tag;
+		else if (tcp_tag == ip_tag) {
+			/* allocate and copy style */
+			/* modify tag type */
+			/*pcam entry for ip6 && ip4 match*/
+			/* default is non tag type */
+			__cvmx_pip_set_tag_type(xp.node, style, ip_tag, CVMX_PKI_PCAM_MATCH_IP);
+		} else if (ip_tag == port_tag_cfg.s.non_tag_type) {
+			/* allocate and copy style */
+			/* modify tag type */
+			/*pcam entry for tcp6 & tcp4 match*/
+			/* default is non tag type */
+			__cvmx_pip_set_tag_type(xp.node, style, tcp_tag, CVMX_PKI_PCAM_MATCH_TCP);
+		} else {
+			if (ip_tag != 0xaa) {
+				nstyle = __cvmx_pip_set_tag_type(xp.node, style, ip_tag,
+								 CVMX_PKI_PCAM_MATCH_IP);
+				if (tcp_tag != 0xff)
+					__cvmx_pip_set_tag_type(xp.node, nstyle, tcp_tag,
+								CVMX_PKI_PCAM_MATCH_TCP);
+				else {
+					n4style = __cvmx_pip_set_tag_type(xp.node, nstyle, ip_tag,
+									  CVMX_PKI_PCAM_MATCH_IPV4);
+					__cvmx_pip_set_tag_type(xp.node, n4style,
+								port_tag_cfg.s.tcp4_tag_type,
+								CVMX_PKI_PCAM_MATCH_TCP);
+					n6style = __cvmx_pip_set_tag_type(xp.node, nstyle, ip_tag,
+									  CVMX_PKI_PCAM_MATCH_IPV6);
+					__cvmx_pip_set_tag_type(xp.node, n6style,
+								port_tag_cfg.s.tcp6_tag_type,
+								CVMX_PKI_PCAM_MATCH_TCP);
+				}
+			} else {
+				n4style = __cvmx_pip_set_tag_type(xp.node, style,
+								  port_tag_cfg.s.ip4_tag_type,
+								  CVMX_PKI_PCAM_MATCH_IPV4);
+				n6style = __cvmx_pip_set_tag_type(xp.node, style,
+								  port_tag_cfg.s.ip6_tag_type,
+								  CVMX_PKI_PCAM_MATCH_IPV6);
+				if (tcp_tag != 0xff) {
+					__cvmx_pip_set_tag_type(xp.node, n4style, tcp_tag,
+								CVMX_PKI_PCAM_MATCH_TCP);
+					__cvmx_pip_set_tag_type(xp.node, n6style, tcp_tag,
+								CVMX_PKI_PCAM_MATCH_TCP);
+				} else {
+					__cvmx_pip_set_tag_type(xp.node, n4style,
+								port_tag_cfg.s.tcp4_tag_type,
+								CVMX_PKI_PCAM_MATCH_TCP);
+					__cvmx_pip_set_tag_type(xp.node, n6style,
+								port_tag_cfg.s.tcp6_tag_type,
+								CVMX_PKI_PCAM_MATCH_TCP);
+				}
+			}
+		}
+		pki_prt_cfg.style_cfg.parm_cfg.qpg_dis_padd = !port_tag_cfg.s.portadd_en;
+
+		if (port_cfg.s.mode == 0x1)
+			pki_prt_cfg.pkind_cfg.initial_parse_mode = CVMX_PKI_PARSE_LA_TO_LG;
+		else if (port_cfg.s.mode == 0x2)
+			pki_prt_cfg.pkind_cfg.initial_parse_mode = CVMX_PKI_PARSE_LC_TO_LG;
+		else
+			pki_prt_cfg.pkind_cfg.initial_parse_mode = CVMX_PKI_PARSE_NOTHING;
+		/* This is only for backward compatibility, not all the parameters are supported in 78xx */
+		cvmx_pki_set_port_config(ipd_port, &pki_prt_cfg);
+	} else {
+		if (octeon_has_feature(OCTEON_FEATURE_PKND)) {
+			int interface, index, pknd;
+
+			interface = cvmx_helper_get_interface_num(ipd_port);
+			index = cvmx_helper_get_interface_index_num(ipd_port);
+			pknd = cvmx_helper_get_pknd(interface, index);
+
+			ipd_port = pknd; /* overload port_num with pknd */
+		}
+		csr_wr(CVMX_PIP_PRT_CFGX(ipd_port), port_cfg.u64);
+		csr_wr(CVMX_PIP_PRT_TAGX(ipd_port), port_tag_cfg.u64);
+	}
+}
+
+/**
+ * Configure the VLAN priority to QoS queue mapping.
+ *
+ * @param vlan_priority
+ *               VLAN priority (0-7)
+ * @param qos    QoS queue for packets matching this watcher
+ */
+static inline void cvmx_pip_config_vlan_qos(u64 vlan_priority, u64 qos)
+{
+	if (!octeon_has_feature(OCTEON_FEATURE_PKND)) {
+		cvmx_pip_qos_vlanx_t pip_qos_vlanx;
+
+		pip_qos_vlanx.u64 = 0;
+		pip_qos_vlanx.s.qos = qos;
+		csr_wr(CVMX_PIP_QOS_VLANX(vlan_priority), pip_qos_vlanx.u64);
+	}
+}
+
+/**
+ * Configure the Diffserv to QoS queue mapping.
+ *
+ * @param diffserv Diffserv field value (0-63)
+ * @param qos      QoS queue for packets matching this watcher
+ */
+static inline void cvmx_pip_config_diffserv_qos(u64 diffserv, u64 qos)
+{
+	if (!octeon_has_feature(OCTEON_FEATURE_PKND)) {
+		cvmx_pip_qos_diffx_t pip_qos_diffx;
+
+		pip_qos_diffx.u64 = 0;
+		pip_qos_diffx.s.qos = qos;
+		csr_wr(CVMX_PIP_QOS_DIFFX(diffserv), pip_qos_diffx.u64);
+	}
+}
+
+/**
+ * Get the status counters for a port for older non PKI chips.
+ *
+ * @param port_num Port number (ipd_port) to get statistics for.
+ * @param clear    Set to 1 to clear the counters after they are read
+ * @param status   Where to put the results.
+ */
+static inline void cvmx_pip_get_port_stats(u64 port_num, u64 clear, cvmx_pip_port_status_t *status)
+{
+	cvmx_pip_stat_ctl_t pip_stat_ctl;
+	cvmx_pip_stat0_prtx_t stat0;
+	cvmx_pip_stat1_prtx_t stat1;
+	cvmx_pip_stat2_prtx_t stat2;
+	cvmx_pip_stat3_prtx_t stat3;
+	cvmx_pip_stat4_prtx_t stat4;
+	cvmx_pip_stat5_prtx_t stat5;
+	cvmx_pip_stat6_prtx_t stat6;
+	cvmx_pip_stat7_prtx_t stat7;
+	cvmx_pip_stat8_prtx_t stat8;
+	cvmx_pip_stat9_prtx_t stat9;
+	cvmx_pip_stat10_x_t stat10;
+	cvmx_pip_stat11_x_t stat11;
+	cvmx_pip_stat_inb_pktsx_t pip_stat_inb_pktsx;
+	cvmx_pip_stat_inb_octsx_t pip_stat_inb_octsx;
+	cvmx_pip_stat_inb_errsx_t pip_stat_inb_errsx;
+	int interface = cvmx_helper_get_interface_num(port_num);
+	int index = cvmx_helper_get_interface_index_num(port_num);
+
+	pip_stat_ctl.u64 = 0;
+	pip_stat_ctl.s.rdclr = clear;
+	csr_wr(CVMX_PIP_STAT_CTL, pip_stat_ctl.u64);
+
+	if (octeon_has_feature(OCTEON_FEATURE_PKND)) {
+		int pknd = cvmx_helper_get_pknd(interface, index);
+		/*
+		 * PIP_STAT_CTL[MODE] 0 means pkind.
+		 */
+		stat0.u64 = csr_rd(CVMX_PIP_STAT0_X(pknd));
+		stat1.u64 = csr_rd(CVMX_PIP_STAT1_X(pknd));
+		stat2.u64 = csr_rd(CVMX_PIP_STAT2_X(pknd));
+		stat3.u64 = csr_rd(CVMX_PIP_STAT3_X(pknd));
+		stat4.u64 = csr_rd(CVMX_PIP_STAT4_X(pknd));
+		stat5.u64 = csr_rd(CVMX_PIP_STAT5_X(pknd));
+		stat6.u64 = csr_rd(CVMX_PIP_STAT6_X(pknd));
+		stat7.u64 = csr_rd(CVMX_PIP_STAT7_X(pknd));
+		stat8.u64 = csr_rd(CVMX_PIP_STAT8_X(pknd));
+		stat9.u64 = csr_rd(CVMX_PIP_STAT9_X(pknd));
+		stat10.u64 = csr_rd(CVMX_PIP_STAT10_X(pknd));
+		stat11.u64 = csr_rd(CVMX_PIP_STAT11_X(pknd));
+	} else {
+		if (port_num >= 40) {
+			stat0.u64 = csr_rd(CVMX_PIP_XSTAT0_PRTX(port_num));
+			stat1.u64 = csr_rd(CVMX_PIP_XSTAT1_PRTX(port_num));
+			stat2.u64 = csr_rd(CVMX_PIP_XSTAT2_PRTX(port_num));
+			stat3.u64 = csr_rd(CVMX_PIP_XSTAT3_PRTX(port_num));
+			stat4.u64 = csr_rd(CVMX_PIP_XSTAT4_PRTX(port_num));
+			stat5.u64 = csr_rd(CVMX_PIP_XSTAT5_PRTX(port_num));
+			stat6.u64 = csr_rd(CVMX_PIP_XSTAT6_PRTX(port_num));
+			stat7.u64 = csr_rd(CVMX_PIP_XSTAT7_PRTX(port_num));
+			stat8.u64 = csr_rd(CVMX_PIP_XSTAT8_PRTX(port_num));
+			stat9.u64 = csr_rd(CVMX_PIP_XSTAT9_PRTX(port_num));
+			if (OCTEON_IS_MODEL(OCTEON_CN6XXX)) {
+				stat10.u64 = csr_rd(CVMX_PIP_XSTAT10_PRTX(port_num));
+				stat11.u64 = csr_rd(CVMX_PIP_XSTAT11_PRTX(port_num));
+			}
+		} else {
+			stat0.u64 = csr_rd(CVMX_PIP_STAT0_PRTX(port_num));
+			stat1.u64 = csr_rd(CVMX_PIP_STAT1_PRTX(port_num));
+			stat2.u64 = csr_rd(CVMX_PIP_STAT2_PRTX(port_num));
+			stat3.u64 = csr_rd(CVMX_PIP_STAT3_PRTX(port_num));
+			stat4.u64 = csr_rd(CVMX_PIP_STAT4_PRTX(port_num));
+			stat5.u64 = csr_rd(CVMX_PIP_STAT5_PRTX(port_num));
+			stat6.u64 = csr_rd(CVMX_PIP_STAT6_PRTX(port_num));
+			stat7.u64 = csr_rd(CVMX_PIP_STAT7_PRTX(port_num));
+			stat8.u64 = csr_rd(CVMX_PIP_STAT8_PRTX(port_num));
+			stat9.u64 = csr_rd(CVMX_PIP_STAT9_PRTX(port_num));
+			if (OCTEON_IS_OCTEON2() || OCTEON_IS_MODEL(OCTEON_CN70XX)) {
+				stat10.u64 = csr_rd(CVMX_PIP_STAT10_PRTX(port_num));
+				stat11.u64 = csr_rd(CVMX_PIP_STAT11_PRTX(port_num));
+			}
+		}
+	}
+	if (octeon_has_feature(OCTEON_FEATURE_PKND)) {
+		int pknd = cvmx_helper_get_pknd(interface, index);
+
+		pip_stat_inb_pktsx.u64 = csr_rd(CVMX_PIP_STAT_INB_PKTS_PKNDX(pknd));
+		pip_stat_inb_octsx.u64 = csr_rd(CVMX_PIP_STAT_INB_OCTS_PKNDX(pknd));
+		pip_stat_inb_errsx.u64 = csr_rd(CVMX_PIP_STAT_INB_ERRS_PKNDX(pknd));
+	} else {
+		pip_stat_inb_pktsx.u64 = csr_rd(CVMX_PIP_STAT_INB_PKTSX(port_num));
+		pip_stat_inb_octsx.u64 = csr_rd(CVMX_PIP_STAT_INB_OCTSX(port_num));
+		pip_stat_inb_errsx.u64 = csr_rd(CVMX_PIP_STAT_INB_ERRSX(port_num));
+	}
+
+	status->dropped_octets = stat0.s.drp_octs;
+	status->dropped_packets = stat0.s.drp_pkts;
+	status->octets = stat1.s.octs;
+	status->pci_raw_packets = stat2.s.raw;
+	status->packets = stat2.s.pkts;
+	status->multicast_packets = stat3.s.mcst;
+	status->broadcast_packets = stat3.s.bcst;
+	status->len_64_packets = stat4.s.h64;
+	status->len_65_127_packets = stat4.s.h65to127;
+	status->len_128_255_packets = stat5.s.h128to255;
+	status->len_256_511_packets = stat5.s.h256to511;
+	status->len_512_1023_packets = stat6.s.h512to1023;
+	status->len_1024_1518_packets = stat6.s.h1024to1518;
+	status->len_1519_max_packets = stat7.s.h1519;
+	status->fcs_align_err_packets = stat7.s.fcs;
+	status->runt_packets = stat8.s.undersz;
+	status->runt_crc_packets = stat8.s.frag;
+	status->oversize_packets = stat9.s.oversz;
+	status->oversize_crc_packets = stat9.s.jabber;
+	if (OCTEON_IS_OCTEON2() || OCTEON_IS_MODEL(OCTEON_CN70XX)) {
+		status->mcast_l2_red_packets = stat10.s.mcast;
+		status->bcast_l2_red_packets = stat10.s.bcast;
+		status->mcast_l3_red_packets = stat11.s.mcast;
+		status->bcast_l3_red_packets = stat11.s.bcast;
+	}
+	status->inb_packets = pip_stat_inb_pktsx.s.pkts;
+	status->inb_octets = pip_stat_inb_octsx.s.octs;
+	status->inb_errors = pip_stat_inb_errsx.s.errs;
+}
+
+/**
+ * Get the status counters for a port.
+ *
+ * @param port_num Port number (ipd_port) to get statistics for.
+ * @param clear    Set to 1 to clear the counters after they are read
+ * @param status   Where to put the results.
+ */
+static inline void cvmx_pip_get_port_status(u64 port_num, u64 clear, cvmx_pip_port_status_t *status)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_PKI)) {
+		unsigned int node = cvmx_get_node_num();
+
+		cvmx_pki_get_port_stats(node, port_num, (struct cvmx_pki_port_stats *)status);
+	} else {
+		cvmx_pip_get_port_stats(port_num, clear, status);
+	}
+}
+
+/**
+ * Configure the hardware CRC engine
+ *
+ * @param interface Interface to configure (0 or 1)
+ * @param invert_result
+ *                 Invert the result of the CRC
+ * @param reflect  Reflect
+ * @param initialization_vector
+ *                 CRC initialization vector
+ */
+static inline void cvmx_pip_config_crc(u64 interface, u64 invert_result, u64 reflect,
+				       u32 initialization_vector)
+{
+	/* Only CN38XX & CN58XX */
+}
+
+/**
+ * Clear all bits in a tag mask. This should be called on
+ * startup before any calls to cvmx_pip_tag_mask_set. Each bit
+ * set in the final mask represent a byte used in the packet for
+ * tag generation.
+ *
+ * @param mask_index Which tag mask to clear (0..3)
+ */
+static inline void cvmx_pip_tag_mask_clear(u64 mask_index)
+{
+	u64 index;
+	cvmx_pip_tag_incx_t pip_tag_incx;
+
+	pip_tag_incx.u64 = 0;
+	pip_tag_incx.s.en = 0;
+	for (index = mask_index * 16; index < (mask_index + 1) * 16; index++)
+		csr_wr(CVMX_PIP_TAG_INCX(index), pip_tag_incx.u64);
+}
+
+/**
+ * Sets a range of bits in the tag mask. The tag mask is used
+ * when the cvmx_pip_port_tag_cfg_t tag_mode is non zero.
+ * There are four separate masks that can be configured.
+ *
+ * @param mask_index Which tag mask to modify (0..3)
+ * @param offset     Offset into the bitmask to set bits at. Use the GCC macro
+ *                   offsetof() to determine the offsets into packet headers.
+ *                   For example, offsetof(ethhdr, protocol) returns the offset
+ *                   of the ethernet protocol field.  The bitmask selects which bytes
+ *                   to include the the tag, with bit offset X selecting byte at offset X
+ *                   from the beginning of the packet data.
+ * @param len        Number of bytes to include. Usually this is the sizeof()
+ *                   the field.
+ */
+static inline void cvmx_pip_tag_mask_set(u64 mask_index, u64 offset, u64 len)
+{
+	while (len--) {
+		cvmx_pip_tag_incx_t pip_tag_incx;
+		u64 index = mask_index * 16 + offset / 8;
+
+		pip_tag_incx.u64 = csr_rd(CVMX_PIP_TAG_INCX(index));
+		pip_tag_incx.s.en |= 0x80 >> (offset & 0x7);
+		csr_wr(CVMX_PIP_TAG_INCX(index), pip_tag_incx.u64);
+		offset++;
+	}
+}
+
+/**
+ * Set byte count for Max-Sized and Min Sized frame check.
+ *
+ * @param interface   Which interface to set the limit
+ * @param max_size    Byte count for Max-Size frame check
+ */
+static inline void cvmx_pip_set_frame_check(int interface, u32 max_size)
+{
+	cvmx_pip_frm_len_chkx_t frm_len;
+
+	/* max_size and min_size are passed as 0, reset to default values. */
+	if (max_size < 1536)
+		max_size = 1536;
+
+	/* On CN68XX frame check is enabled for a pkind n and
+	   PIP_PRT_CFG[len_chk_sel] selects which set of
+	   MAXLEN/MINLEN to use. */
+	if (octeon_has_feature(OCTEON_FEATURE_PKND)) {
+		int port;
+		int num_ports = cvmx_helper_ports_on_interface(interface);
+
+		for (port = 0; port < num_ports; port++) {
+			if (octeon_has_feature(OCTEON_FEATURE_PKI)) {
+				int ipd_port;
+
+				ipd_port = cvmx_helper_get_ipd_port(interface, port);
+				cvmx_pki_set_max_frm_len(ipd_port, max_size);
+			} else {
+				int pknd;
+				int sel;
+				cvmx_pip_prt_cfgx_t config;
+
+				pknd = cvmx_helper_get_pknd(interface, port);
+				config.u64 = csr_rd(CVMX_PIP_PRT_CFGX(pknd));
+				sel = config.s.len_chk_sel;
+				frm_len.u64 = csr_rd(CVMX_PIP_FRM_LEN_CHKX(sel));
+				frm_len.s.maxlen = max_size;
+				csr_wr(CVMX_PIP_FRM_LEN_CHKX(sel), frm_len.u64);
+			}
+		}
+	}
+	/* on cn6xxx and cn7xxx models, PIP_FRM_LEN_CHK0 applies to
+	 *     all incoming traffic */
+	else if (OCTEON_IS_OCTEON2() || OCTEON_IS_MODEL(OCTEON_CN70XX)) {
+		frm_len.u64 = csr_rd(CVMX_PIP_FRM_LEN_CHKX(0));
+		frm_len.s.maxlen = max_size;
+		csr_wr(CVMX_PIP_FRM_LEN_CHKX(0), frm_len.u64);
+	}
+}
+
+/**
+ * Initialize Bit Select Extractor config. Their are 8 bit positions and valids
+ * to be used when using the corresponding extractor.
+ *
+ * @param bit     Bit Select Extractor to use
+ * @param pos     Which position to update
+ * @param val     The value to update the position with
+ */
+static inline void cvmx_pip_set_bsel_pos(int bit, int pos, int val)
+{
+	cvmx_pip_bsel_ext_posx_t bsel_pos;
+
+	/* The bit select extractor is available in CN61XX and CN68XX pass2.0 onwards. */
+	if (!octeon_has_feature(OCTEON_FEATURE_BIT_EXTRACTOR))
+		return;
+
+	if (bit < 0 || bit > 3) {
+		debug("ERROR: cvmx_pip_set_bsel_pos: Invalid Bit-Select Extractor (%d) passed\n",
+		      bit);
+		return;
+	}
+
+	bsel_pos.u64 = csr_rd(CVMX_PIP_BSEL_EXT_POSX(bit));
+	switch (pos) {
+	case 0:
+		bsel_pos.s.pos0_val = 1;
+		bsel_pos.s.pos0 = val & 0x7f;
+		break;
+	case 1:
+		bsel_pos.s.pos1_val = 1;
+		bsel_pos.s.pos1 = val & 0x7f;
+		break;
+	case 2:
+		bsel_pos.s.pos2_val = 1;
+		bsel_pos.s.pos2 = val & 0x7f;
+		break;
+	case 3:
+		bsel_pos.s.pos3_val = 1;
+		bsel_pos.s.pos3 = val & 0x7f;
+		break;
+	case 4:
+		bsel_pos.s.pos4_val = 1;
+		bsel_pos.s.pos4 = val & 0x7f;
+		break;
+	case 5:
+		bsel_pos.s.pos5_val = 1;
+		bsel_pos.s.pos5 = val & 0x7f;
+		break;
+	case 6:
+		bsel_pos.s.pos6_val = 1;
+		bsel_pos.s.pos6 = val & 0x7f;
+		break;
+	case 7:
+		bsel_pos.s.pos7_val = 1;
+		bsel_pos.s.pos7 = val & 0x7f;
+		break;
+	default:
+		debug("Warning: cvmx_pip_set_bsel_pos: Invalid pos(%d)\n", pos);
+		break;
+	}
+	csr_wr(CVMX_PIP_BSEL_EXT_POSX(bit), bsel_pos.u64);
+}
+
+/**
+ * Initialize offset and skip values to use by bit select extractor.
+
+ * @param bit	Bit Select Extractor to use
+ * @param offset	Offset to add to extractor mem addr to get final address
+ *			to lookup table.
+ * @param skip		Number of bytes to skip from start of packet 0-64
+ */
+static inline void cvmx_pip_bsel_config(int bit, int offset, int skip)
+{
+	cvmx_pip_bsel_ext_cfgx_t bsel_cfg;
+
+	/* The bit select extractor is available in CN61XX and CN68XX pass2.0 onwards. */
+	if (!octeon_has_feature(OCTEON_FEATURE_BIT_EXTRACTOR))
+		return;
+
+	bsel_cfg.u64 = csr_rd(CVMX_PIP_BSEL_EXT_CFGX(bit));
+	bsel_cfg.s.offset = offset;
+	bsel_cfg.s.skip = skip;
+	csr_wr(CVMX_PIP_BSEL_EXT_CFGX(bit), bsel_cfg.u64);
+}
+
+/**
+ * Get the entry for the Bit Select Extractor Table.
+ * @param work   pointer to work queue entry
+ * @return       Index of the Bit Select Extractor Table
+ */
+static inline int cvmx_pip_get_bsel_table_index(cvmx_wqe_t *work)
+{
+	int bit = cvmx_wqe_get_port(work) & 0x3;
+	/* Get the Bit select table index. */
+	int index;
+	int y;
+	cvmx_pip_bsel_ext_cfgx_t bsel_cfg;
+	cvmx_pip_bsel_ext_posx_t bsel_pos;
+
+	/* The bit select extractor is available in CN61XX and CN68XX pass2.0 onwards. */
+	if (!octeon_has_feature(OCTEON_FEATURE_BIT_EXTRACTOR))
+		return -1;
+
+	bsel_cfg.u64 = csr_rd(CVMX_PIP_BSEL_EXT_CFGX(bit));
+	bsel_pos.u64 = csr_rd(CVMX_PIP_BSEL_EXT_POSX(bit));
+
+	for (y = 0; y < 8; y++) {
+		char *ptr = (char *)cvmx_phys_to_ptr(work->packet_ptr.s.addr);
+		int bit_loc = 0;
+		int bit;
+
+		ptr += bsel_cfg.s.skip;
+		switch (y) {
+		case 0:
+			ptr += (bsel_pos.s.pos0 >> 3);
+			bit_loc = 7 - (bsel_pos.s.pos0 & 0x3);
+			break;
+		case 1:
+			ptr += (bsel_pos.s.pos1 >> 3);
+			bit_loc = 7 - (bsel_pos.s.pos1 & 0x3);
+			break;
+		case 2:
+			ptr += (bsel_pos.s.pos2 >> 3);
+			bit_loc = 7 - (bsel_pos.s.pos2 & 0x3);
+			break;
+		case 3:
+			ptr += (bsel_pos.s.pos3 >> 3);
+			bit_loc = 7 - (bsel_pos.s.pos3 & 0x3);
+			break;
+		case 4:
+			ptr += (bsel_pos.s.pos4 >> 3);
+			bit_loc = 7 - (bsel_pos.s.pos4 & 0x3);
+			break;
+		case 5:
+			ptr += (bsel_pos.s.pos5 >> 3);
+			bit_loc = 7 - (bsel_pos.s.pos5 & 0x3);
+			break;
+		case 6:
+			ptr += (bsel_pos.s.pos6 >> 3);
+			bit_loc = 7 - (bsel_pos.s.pos6 & 0x3);
+			break;
+		case 7:
+			ptr += (bsel_pos.s.pos7 >> 3);
+			bit_loc = 7 - (bsel_pos.s.pos7 & 0x3);
+			break;
+		}
+		bit = (*ptr >> bit_loc) & 1;
+		index |= bit << y;
+	}
+	index += bsel_cfg.s.offset;
+	index &= 0x1ff;
+	return index;
+}
+
+static inline int cvmx_pip_get_bsel_qos(cvmx_wqe_t *work)
+{
+	int index = cvmx_pip_get_bsel_table_index(work);
+	cvmx_pip_bsel_tbl_entx_t bsel_tbl;
+
+	/* The bit select extractor is available in CN61XX and CN68XX pass2.0 onwards. */
+	if (!octeon_has_feature(OCTEON_FEATURE_BIT_EXTRACTOR))
+		return -1;
+
+	bsel_tbl.u64 = csr_rd(CVMX_PIP_BSEL_TBL_ENTX(index));
+
+	return bsel_tbl.s.qos;
+}
+
+static inline int cvmx_pip_get_bsel_grp(cvmx_wqe_t *work)
+{
+	int index = cvmx_pip_get_bsel_table_index(work);
+	cvmx_pip_bsel_tbl_entx_t bsel_tbl;
+
+	/* The bit select extractor is available in CN61XX and CN68XX pass2.0 onwards. */
+	if (!octeon_has_feature(OCTEON_FEATURE_BIT_EXTRACTOR))
+		return -1;
+
+	bsel_tbl.u64 = csr_rd(CVMX_PIP_BSEL_TBL_ENTX(index));
+
+	return bsel_tbl.s.grp;
+}
+
+static inline int cvmx_pip_get_bsel_tt(cvmx_wqe_t *work)
+{
+	int index = cvmx_pip_get_bsel_table_index(work);
+	cvmx_pip_bsel_tbl_entx_t bsel_tbl;
+
+	/* The bit select extractor is available in CN61XX and CN68XX pass2.0 onwards. */
+	if (!octeon_has_feature(OCTEON_FEATURE_BIT_EXTRACTOR))
+		return -1;
+
+	bsel_tbl.u64 = csr_rd(CVMX_PIP_BSEL_TBL_ENTX(index));
+
+	return bsel_tbl.s.tt;
+}
+
+static inline int cvmx_pip_get_bsel_tag(cvmx_wqe_t *work)
+{
+	int index = cvmx_pip_get_bsel_table_index(work);
+	int port = cvmx_wqe_get_port(work);
+	int bit = port & 0x3;
+	int upper_tag = 0;
+	cvmx_pip_bsel_tbl_entx_t bsel_tbl;
+	cvmx_pip_bsel_ext_cfgx_t bsel_cfg;
+	cvmx_pip_prt_tagx_t prt_tag;
+
+	/* The bit select extractor is available in CN61XX and CN68XX pass2.0 onwards. */
+	if (!octeon_has_feature(OCTEON_FEATURE_BIT_EXTRACTOR))
+		return -1;
+
+	bsel_tbl.u64 = csr_rd(CVMX_PIP_BSEL_TBL_ENTX(index));
+	bsel_cfg.u64 = csr_rd(CVMX_PIP_BSEL_EXT_CFGX(bit));
+
+	prt_tag.u64 = csr_rd(CVMX_PIP_PRT_TAGX(port));
+	if (prt_tag.s.inc_prt_flag == 0)
+		upper_tag = bsel_cfg.s.upper_tag;
+	return bsel_tbl.s.tag | ((bsel_cfg.s.tag << 8) & 0xff00) | ((upper_tag << 16) & 0xffff0000);
+}
+
+#endif /*  __CVMX_PIP_H__ */
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-pki-resources.h b/arch/mips/mach-octeon/include/mach/cvmx-pki-resources.h
new file mode 100644
index 0000000000..79b99b0bd7
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-pki-resources.h
@@ -0,0 +1,157 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ *
+ * Resource management for PKI resources.
+ */
+
+#ifndef __CVMX_PKI_RESOURCES_H__
+#define __CVMX_PKI_RESOURCES_H__
+
+/**
+ * This function allocates/reserves a style from pool of global styles per node.
+ * @param node	 node to allocate style from.
+ * @param style	 style to allocate, if -1 it will be allocated
+		 first available style from style resource. If index is positive
+		 number and in range, it will try to allocate specified style.
+ * @return	 style number on success, -1 on failure.
+ */
+int cvmx_pki_style_alloc(int node, int style);
+
+/**
+ * This function allocates/reserves a cluster group from per node
+   cluster group resources.
+ * @param node		node to allocate cluster group from.
+   @param cl_grp	cluster group to allocate/reserve, if -1 ,
+			allocate any available cluster group.
+ * @return		cluster group number or -1 on failure
+ */
+int cvmx_pki_cluster_grp_alloc(int node, int cl_grp);
+
+/**
+ * This function allocates/reserves a cluster from per node
+   cluster resources.
+ * @param node		node to allocate cluster group from.
+   @param cluster_mask	mask of clusters  to allocate/reserve, if -1 ,
+			allocate any available clusters.
+ * @param num_clusters	number of clusters that will be allocated
+ */
+int cvmx_pki_cluster_alloc(int node, int num_clusters, u64 *cluster_mask);
+
+/**
+ * This function allocates/reserves a pcam entry from node
+ * @param node		node to allocate pcam entry from.
+   @param index	index of pacm entry (0-191), if -1 ,
+			allocate any available pcam entry.
+ * @param bank		pcam bank where to allocate/reserve pcan entry from
+ * @param cluster_mask  mask of clusters from which pcam entry is needed.
+ * @return		pcam entry of -1 on failure
+ */
+int cvmx_pki_pcam_entry_alloc(int node, int index, int bank, u64 cluster_mask);
+
+/**
+ * This function allocates/reserves QPG table entries per node.
+ * @param node		node number.
+ * @param base_offset	base_offset in qpg table. If -1, first available
+			qpg base_offset will be allocated. If base_offset is positive
+			number and in range, it will try to allocate specified base_offset.
+   @param count		number of consecutive qpg entries to allocate. They will be consecutive
+			from base offset.
+ * @return		qpg table base offset number on success, -1 on failure.
+ */
+int cvmx_pki_qpg_entry_alloc(int node, int base_offset, int count);
+
+/**
+ * This function frees a style from pool of global styles per node.
+ * @param node	 node to free style from.
+ * @param style	 style to free
+ * @return	 0 on success, -1 on failure.
+ */
+int cvmx_pki_style_free(int node, int style);
+
+/**
+ * This function frees a cluster group from per node
+   cluster group resources.
+ * @param node		node to free cluster group from.
+   @param cl_grp	cluster group to free
+ * @return		0 on success or -1 on failure
+ */
+int cvmx_pki_cluster_grp_free(int node, int cl_grp);
+
+/**
+ * This function frees QPG table entries per node.
+ * @param node		node number.
+ * @param base_offset	base_offset in qpg table. If -1, first available
+ *			qpg base_offset will be allocated. If base_offset is positive
+ *			number and in range, it will try to allocate specified base_offset.
+ * @param count		number of consecutive qpg entries to allocate. They will be consecutive
+ *			from base offset.
+ * @return		qpg table base offset number on success, -1 on failure.
+ */
+int cvmx_pki_qpg_entry_free(int node, int base_offset, int count);
+
+/**
+ * This function frees  clusters  from per node
+   clusters resources.
+ * @param node		node to free clusters from.
+ * @param cluster_mask  mask of clusters need freeing
+ * @return		0 on success or -1 on failure
+ */
+int cvmx_pki_cluster_free(int node, u64 cluster_mask);
+
+/**
+ * This function frees a pcam entry from node
+ * @param node		node to allocate pcam entry from.
+   @param index	index of pacm entry (0-191) needs to be freed.
+ * @param bank		pcam bank where to free pcam entry from
+ * @param cluster_mask  mask of clusters from which pcam entry is freed.
+ * @return		0 on success OR -1 on failure
+ */
+int cvmx_pki_pcam_entry_free(int node, int index, int bank, u64 cluster_mask);
+
+/**
+ * This function allocates/reserves a bpid from pool of global bpid per node.
+ * @param node	node to allocate bpid from.
+ * @param bpid	bpid  to allocate, if -1 it will be allocated
+ *		first available boid from bpid resource. If index is positive
+ *		number and in range, it will try to allocate specified bpid.
+ * @return	bpid number on success,
+ *		-1 on alloc failure.
+ *		-2 on resource already reserved.
+ */
+int cvmx_pki_bpid_alloc(int node, int bpid);
+
+/**
+ * This function frees a bpid from pool of global bpid per node.
+ * @param node	 node to free bpid from.
+ * @param bpid	 bpid to free
+ * @return	 0 on success, -1 on failure or
+ */
+int cvmx_pki_bpid_free(int node, int bpid);
+
+/**
+ * This function frees all the PKI software resources
+ * (clusters, styles, qpg_entry, pcam_entry etc) for the specified node
+ */
+
+/**
+ * This function allocates/reserves an index from pool of global MTAG-IDX per node.
+ * @param node	node to allocate index from.
+ * @param idx	index  to allocate, if -1 it will be allocated
+ * @return	MTAG index number on success,
+ *		-1 on alloc failure.
+ *		-2 on resource already reserved.
+ */
+int cvmx_pki_mtag_idx_alloc(int node, int idx);
+
+/**
+ * This function frees an index from pool of global MTAG-IDX per node.
+ * @param node	 node to free bpid from.
+ * @param bpid	 bpid to free
+ * @return	 0 on success, -1 on failure or
+ */
+int cvmx_pki_mtag_idx_free(int node, int idx);
+
+void __cvmx_pki_global_rsrc_free(int node);
+
+#endif /*  __CVM_PKI_RESOURCES_H__ */
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-pki.h b/arch/mips/mach-octeon/include/mach/cvmx-pki.h
new file mode 100644
index 0000000000..c1feb55a1f
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-pki.h
@@ -0,0 +1,970 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ *
+ * Interface to the hardware Packet Input Data unit.
+ */
+
+#ifndef __CVMX_PKI_H__
+#define __CVMX_PKI_H__
+
+#include "cvmx-fpa3.h"
+#include "cvmx-helper-util.h"
+#include "cvmx-helper-cfg.h"
+#include "cvmx-error.h"
+
+/* PKI AURA and BPID count are equal to FPA AURA count */
+#define CVMX_PKI_NUM_AURA	       (cvmx_fpa3_num_auras())
+#define CVMX_PKI_NUM_BPID	       (cvmx_fpa3_num_auras())
+#define CVMX_PKI_NUM_SSO_GROUP	       (cvmx_sso_num_xgrp())
+#define CVMX_PKI_NUM_CLUSTER_GROUP_MAX 1
+#define CVMX_PKI_NUM_CLUSTER_GROUP     (cvmx_pki_num_cl_grp())
+#define CVMX_PKI_NUM_CLUSTER	       (cvmx_pki_num_clusters())
+
+/* FIXME: Reduce some of these values, convert to routines XXX */
+#define CVMX_PKI_NUM_CHANNEL	    4096
+#define CVMX_PKI_NUM_PKIND	    64
+#define CVMX_PKI_NUM_INTERNAL_STYLE 256
+#define CVMX_PKI_NUM_FINAL_STYLE    64
+#define CVMX_PKI_NUM_QPG_ENTRY	    2048
+#define CVMX_PKI_NUM_MTAG_IDX	    (32 / 4) /* 32 registers grouped by 4*/
+#define CVMX_PKI_NUM_LTYPE	    32
+#define CVMX_PKI_NUM_PCAM_BANK	    2
+#define CVMX_PKI_NUM_PCAM_ENTRY	    192
+#define CVMX_PKI_NUM_FRAME_CHECK    2
+#define CVMX_PKI_NUM_BELTYPE	    32
+#define CVMX_PKI_MAX_FRAME_SIZE	    65535
+#define CVMX_PKI_FIND_AVAL_ENTRY    (-1)
+#define CVMX_PKI_CLUSTER_ALL	    0xf
+
+#ifdef CVMX_SUPPORT_SEPARATE_CLUSTER_CONFIG
+#define CVMX_PKI_TOTAL_PCAM_ENTRY                                                                  \
+	((CVMX_PKI_NUM_CLUSTER) * (CVMX_PKI_NUM_PCAM_BANK) * (CVMX_PKI_NUM_PCAM_ENTRY))
+#else
+#define CVMX_PKI_TOTAL_PCAM_ENTRY (CVMX_PKI_NUM_PCAM_BANK * CVMX_PKI_NUM_PCAM_ENTRY)
+#endif
+
+static inline unsigned int cvmx_pki_num_clusters(void)
+{
+	if (OCTEON_IS_MODEL(OCTEON_CN73XX) || OCTEON_IS_MODEL(OCTEON_CNF75XX))
+		return 2;
+	return 4;
+}
+
+static inline unsigned int cvmx_pki_num_cl_grp(void)
+{
+	if (OCTEON_IS_MODEL(OCTEON_CN73XX) || OCTEON_IS_MODEL(OCTEON_CNF75XX) ||
+	    OCTEON_IS_MODEL(OCTEON_CN78XX))
+		return 1;
+	return 0;
+}
+
+enum cvmx_pki_pkind_parse_mode {
+	CVMX_PKI_PARSE_LA_TO_LG = 0,  /* Parse LA(L2) to LG */
+	CVMX_PKI_PARSE_LB_TO_LG = 1,  /* Parse LB(custom) to LG */
+	CVMX_PKI_PARSE_LC_TO_LG = 3,  /* Parse LC(L3) to LG */
+	CVMX_PKI_PARSE_LG = 0x3f,     /* Parse LG */
+	CVMX_PKI_PARSE_NOTHING = 0x7f /* Parse nothing */
+};
+
+enum cvmx_pki_parse_mode_chg {
+	CVMX_PKI_PARSE_NO_CHG = 0x0,
+	CVMX_PKI_PARSE_SKIP_TO_LB = 0x1,
+	CVMX_PKI_PARSE_SKIP_TO_LC = 0x3,
+	CVMX_PKI_PARSE_SKIP_TO_LD = 0x7,
+	CVMX_PKI_PARSE_SKIP_TO_LG = 0x3f,
+	CVMX_PKI_PARSE_SKIP_ALL = 0x7f,
+};
+
+enum cvmx_pki_l2_len_mode { PKI_L2_LENCHK_EQUAL_GREATER = 0, PKI_L2_LENCHK_EQUAL_ONLY };
+
+enum cvmx_pki_cache_mode {
+	CVMX_PKI_OPC_MODE_STT = 0LL,	  /* All blocks write through DRAM,*/
+	CVMX_PKI_OPC_MODE_STF = 1LL,	  /* All blocks into L2 */
+	CVMX_PKI_OPC_MODE_STF1_STT = 2LL, /* 1st block L2, rest DRAM */
+	CVMX_PKI_OPC_MODE_STF2_STT = 3LL  /* 1st, 2nd blocks L2, rest DRAM */
+};
+
+/**
+ * Tag type definitions
+ */
+enum cvmx_sso_tag_type {
+	CVMX_SSO_TAG_TYPE_ORDERED = 0L,
+	CVMX_SSO_TAG_TYPE_ATOMIC = 1L,
+	CVMX_SSO_TAG_TYPE_UNTAGGED = 2L,
+	CVMX_SSO_TAG_TYPE_EMPTY = 3L
+};
+
+enum cvmx_pki_qpg_qos {
+	CVMX_PKI_QPG_QOS_NONE = 0,
+	CVMX_PKI_QPG_QOS_VLAN,
+	CVMX_PKI_QPG_QOS_MPLS,
+	CVMX_PKI_QPG_QOS_DSA_SRC,
+	CVMX_PKI_QPG_QOS_DIFFSERV,
+	CVMX_PKI_QPG_QOS_HIGIG,
+};
+
+enum cvmx_pki_wqe_vlan { CVMX_PKI_USE_FIRST_VLAN = 0, CVMX_PKI_USE_SECOND_VLAN };
+
+/**
+ * Controls how the PKI statistics counters are handled
+ * The PKI_STAT*_X registers can be indexed either by port kind (pkind), or
+ * final style. (Does not apply to the PKI_STAT_INB* registers.)
+ *    0 = X represents the packet’s pkind
+ *    1 = X represents the low 6-bits of packet’s final style
+ */
+enum cvmx_pki_stats_mode { CVMX_PKI_STAT_MODE_PKIND, CVMX_PKI_STAT_MODE_STYLE };
+
+enum cvmx_pki_fpa_wait { CVMX_PKI_DROP_PKT, CVMX_PKI_WAIT_PKT };
+
+#define PKI_BELTYPE_E__NONE_M 0x0
+#define PKI_BELTYPE_E__MISC_M 0x1
+#define PKI_BELTYPE_E__IP4_M  0x2
+#define PKI_BELTYPE_E__IP6_M  0x3
+#define PKI_BELTYPE_E__TCP_M  0x4
+#define PKI_BELTYPE_E__UDP_M  0x5
+#define PKI_BELTYPE_E__SCTP_M 0x6
+#define PKI_BELTYPE_E__SNAP_M 0x7
+
+/* PKI_BELTYPE_E_t */
+enum cvmx_pki_beltype {
+	CVMX_PKI_BELTYPE_NONE = PKI_BELTYPE_E__NONE_M,
+	CVMX_PKI_BELTYPE_MISC = PKI_BELTYPE_E__MISC_M,
+	CVMX_PKI_BELTYPE_IP4 = PKI_BELTYPE_E__IP4_M,
+	CVMX_PKI_BELTYPE_IP6 = PKI_BELTYPE_E__IP6_M,
+	CVMX_PKI_BELTYPE_TCP = PKI_BELTYPE_E__TCP_M,
+	CVMX_PKI_BELTYPE_UDP = PKI_BELTYPE_E__UDP_M,
+	CVMX_PKI_BELTYPE_SCTP = PKI_BELTYPE_E__SCTP_M,
+	CVMX_PKI_BELTYPE_SNAP = PKI_BELTYPE_E__SNAP_M,
+	CVMX_PKI_BELTYPE_MAX = CVMX_PKI_BELTYPE_SNAP
+};
+
+struct cvmx_pki_frame_len {
+	u16 maxlen;
+	u16 minlen;
+};
+
+struct cvmx_pki_tag_fields {
+	u64 layer_g_src : 1;
+	u64 layer_f_src : 1;
+	u64 layer_e_src : 1;
+	u64 layer_d_src : 1;
+	u64 layer_c_src : 1;
+	u64 layer_b_src : 1;
+	u64 layer_g_dst : 1;
+	u64 layer_f_dst : 1;
+	u64 layer_e_dst : 1;
+	u64 layer_d_dst : 1;
+	u64 layer_c_dst : 1;
+	u64 layer_b_dst : 1;
+	u64 input_port : 1;
+	u64 mpls_label : 1;
+	u64 first_vlan : 1;
+	u64 second_vlan : 1;
+	u64 ip_prot_nexthdr : 1;
+	u64 tag_sync : 1;
+	u64 tag_spi : 1;
+	u64 tag_gtp : 1;
+	u64 tag_vni : 1;
+};
+
+struct cvmx_pki_pkind_parse {
+	u64 mpls_en : 1;
+	u64 inst_hdr : 1;
+	u64 lg_custom : 1;
+	u64 fulc_en : 1;
+	u64 dsa_en : 1;
+	u64 hg2_en : 1;
+	u64 hg_en : 1;
+};
+
+struct cvmx_pki_pool_config {
+	int pool_num;
+	cvmx_fpa3_pool_t pool;
+	u64 buffer_size;
+	u64 buffer_count;
+};
+
+struct cvmx_pki_qpg_config {
+	int qpg_base;
+	int port_add;
+	int aura_num;
+	int grp_ok;
+	int grp_bad;
+	int grptag_ok;
+	int grptag_bad;
+};
+
+struct cvmx_pki_aura_config {
+	int aura_num;
+	int pool_num;
+	cvmx_fpa3_pool_t pool;
+	cvmx_fpa3_gaura_t aura;
+	int buffer_count;
+};
+
+struct cvmx_pki_cluster_grp_config {
+	int grp_num;
+	u64 cluster_mask; /* Bit mask of cluster assigned to this cluster group */
+};
+
+struct cvmx_pki_sso_grp_config {
+	int group;
+	int priority;
+	int weight;
+	int affinity;
+	u64 core_mask;
+	u8 core_mask_set;
+};
+
+/* This is per style structure for configuring port parameters,
+ * it is kind of of profile which can be assigned to any port.
+ * If multiple ports are assigned same style be aware that modifying
+ * that style will modify the respective parameters for all the ports
+ * which are using this style
+ */
+struct cvmx_pki_style_parm {
+	bool ip6_udp_opt;
+	bool lenerr_en;
+	bool maxerr_en;
+	bool minerr_en;
+	u8 lenerr_eqpad;
+	u8 minmax_sel;
+	bool qpg_dis_grptag;
+	bool fcs_strip;
+	bool fcs_chk;
+	bool rawdrp;
+	bool force_drop;
+	bool nodrop;
+	bool qpg_dis_padd;
+	bool qpg_dis_grp;
+	bool qpg_dis_aura;
+	u16 qpg_base;
+	enum cvmx_pki_qpg_qos qpg_qos;
+	u8 qpg_port_sh;
+	u8 qpg_port_msb;
+	u8 apad_nip;
+	u8 wqe_vs;
+	enum cvmx_sso_tag_type tag_type;
+	bool pkt_lend;
+	u8 wqe_hsz;
+	u16 wqe_skip;
+	u16 first_skip;
+	u16 later_skip;
+	enum cvmx_pki_cache_mode cache_mode;
+	u8 dis_wq_dat;
+	u64 mbuff_size;
+	bool len_lg;
+	bool len_lf;
+	bool len_le;
+	bool len_ld;
+	bool len_lc;
+	bool len_lb;
+	bool csum_lg;
+	bool csum_lf;
+	bool csum_le;
+	bool csum_ld;
+	bool csum_lc;
+	bool csum_lb;
+};
+
+/* This is per style structure for configuring port's tag configuration,
+ * it is kind of of profile which can be assigned to any port.
+ * If multiple ports are assigned same style be aware that modiying that style
+ * will modify the respective parameters for all the ports which are
+ * using this style */
+enum cvmx_pki_mtag_ptrsel {
+	CVMX_PKI_MTAG_PTRSEL_SOP = 0,
+	CVMX_PKI_MTAG_PTRSEL_LA = 8,
+	CVMX_PKI_MTAG_PTRSEL_LB = 9,
+	CVMX_PKI_MTAG_PTRSEL_LC = 10,
+	CVMX_PKI_MTAG_PTRSEL_LD = 11,
+	CVMX_PKI_MTAG_PTRSEL_LE = 12,
+	CVMX_PKI_MTAG_PTRSEL_LF = 13,
+	CVMX_PKI_MTAG_PTRSEL_LG = 14,
+	CVMX_PKI_MTAG_PTRSEL_VL = 15,
+};
+
+struct cvmx_pki_mask_tag {
+	bool enable;
+	int base;   /* CVMX_PKI_MTAG_PTRSEL_XXX */
+	int offset; /* Offset from base. */
+	u64 val;    /* Bitmask:
+		1 = enable, 0 = disabled for each byte in the 64-byte array.*/
+};
+
+struct cvmx_pki_style_tag_cfg {
+	struct cvmx_pki_tag_fields tag_fields;
+	struct cvmx_pki_mask_tag mask_tag[4];
+};
+
+struct cvmx_pki_style_config {
+	struct cvmx_pki_style_parm parm_cfg;
+	struct cvmx_pki_style_tag_cfg tag_cfg;
+};
+
+struct cvmx_pki_pkind_config {
+	u8 cluster_grp;
+	bool fcs_pres;
+	struct cvmx_pki_pkind_parse parse_en;
+	enum cvmx_pki_pkind_parse_mode initial_parse_mode;
+	u8 fcs_skip;
+	u8 inst_skip;
+	int initial_style;
+	bool custom_l2_hdr;
+	u8 l2_scan_offset;
+	u64 lg_scan_offset;
+};
+
+struct cvmx_pki_port_config {
+	struct cvmx_pki_pkind_config pkind_cfg;
+	struct cvmx_pki_style_config style_cfg;
+};
+
+struct cvmx_pki_global_parse {
+	u64 virt_pen : 1;
+	u64 clg_pen : 1;
+	u64 cl2_pen : 1;
+	u64 l4_pen : 1;
+	u64 il3_pen : 1;
+	u64 l3_pen : 1;
+	u64 mpls_pen : 1;
+	u64 fulc_pen : 1;
+	u64 dsa_pen : 1;
+	u64 hg_pen : 1;
+};
+
+struct cvmx_pki_tag_sec {
+	u16 dst6;
+	u16 src6;
+	u16 dst;
+	u16 src;
+};
+
+struct cvmx_pki_global_config {
+	u64 cluster_mask[CVMX_PKI_NUM_CLUSTER_GROUP_MAX];
+	enum cvmx_pki_stats_mode stat_mode;
+	enum cvmx_pki_fpa_wait fpa_wait;
+	struct cvmx_pki_global_parse gbl_pen;
+	struct cvmx_pki_tag_sec tag_secret;
+	struct cvmx_pki_frame_len frm_len[CVMX_PKI_NUM_FRAME_CHECK];
+	enum cvmx_pki_beltype ltype_map[CVMX_PKI_NUM_BELTYPE];
+	int pki_enable;
+};
+
+#define CVMX_PKI_PCAM_TERM_E_NONE_M	 0x0
+#define CVMX_PKI_PCAM_TERM_E_L2_CUSTOM_M 0x2
+#define CVMX_PKI_PCAM_TERM_E_HIGIGD_M	 0x4
+#define CVMX_PKI_PCAM_TERM_E_HIGIG_M	 0x5
+#define CVMX_PKI_PCAM_TERM_E_SMACH_M	 0x8
+#define CVMX_PKI_PCAM_TERM_E_SMACL_M	 0x9
+#define CVMX_PKI_PCAM_TERM_E_DMACH_M	 0xA
+#define CVMX_PKI_PCAM_TERM_E_DMACL_M	 0xB
+#define CVMX_PKI_PCAM_TERM_E_GLORT_M	 0x12
+#define CVMX_PKI_PCAM_TERM_E_DSA_M	 0x13
+#define CVMX_PKI_PCAM_TERM_E_ETHTYPE0_M	 0x18
+#define CVMX_PKI_PCAM_TERM_E_ETHTYPE1_M	 0x19
+#define CVMX_PKI_PCAM_TERM_E_ETHTYPE2_M	 0x1A
+#define CVMX_PKI_PCAM_TERM_E_ETHTYPE3_M	 0x1B
+#define CVMX_PKI_PCAM_TERM_E_MPLS0_M	 0x1E
+#define CVMX_PKI_PCAM_TERM_E_L3_SIPHH_M	 0x1F
+#define CVMX_PKI_PCAM_TERM_E_L3_SIPMH_M	 0x20
+#define CVMX_PKI_PCAM_TERM_E_L3_SIPML_M	 0x21
+#define CVMX_PKI_PCAM_TERM_E_L3_SIPLL_M	 0x22
+#define CVMX_PKI_PCAM_TERM_E_L3_FLAGS_M	 0x23
+#define CVMX_PKI_PCAM_TERM_E_L3_DIPHH_M	 0x24
+#define CVMX_PKI_PCAM_TERM_E_L3_DIPMH_M	 0x25
+#define CVMX_PKI_PCAM_TERM_E_L3_DIPML_M	 0x26
+#define CVMX_PKI_PCAM_TERM_E_L3_DIPLL_M	 0x27
+#define CVMX_PKI_PCAM_TERM_E_LD_VNI_M	 0x28
+#define CVMX_PKI_PCAM_TERM_E_IL3_FLAGS_M 0x2B
+#define CVMX_PKI_PCAM_TERM_E_LF_SPI_M	 0x2E
+#define CVMX_PKI_PCAM_TERM_E_L4_SPORT_M	 0x2f
+#define CVMX_PKI_PCAM_TERM_E_L4_PORT_M	 0x30
+#define CVMX_PKI_PCAM_TERM_E_LG_CUSTOM_M 0x39
+
+enum cvmx_pki_term {
+	CVMX_PKI_PCAM_TERM_NONE = CVMX_PKI_PCAM_TERM_E_NONE_M,
+	CVMX_PKI_PCAM_TERM_L2_CUSTOM = CVMX_PKI_PCAM_TERM_E_L2_CUSTOM_M,
+	CVMX_PKI_PCAM_TERM_HIGIGD = CVMX_PKI_PCAM_TERM_E_HIGIGD_M,
+	CVMX_PKI_PCAM_TERM_HIGIG = CVMX_PKI_PCAM_TERM_E_HIGIG_M,
+	CVMX_PKI_PCAM_TERM_SMACH = CVMX_PKI_PCAM_TERM_E_SMACH_M,
+	CVMX_PKI_PCAM_TERM_SMACL = CVMX_PKI_PCAM_TERM_E_SMACL_M,
+	CVMX_PKI_PCAM_TERM_DMACH = CVMX_PKI_PCAM_TERM_E_DMACH_M,
+	CVMX_PKI_PCAM_TERM_DMACL = CVMX_PKI_PCAM_TERM_E_DMACL_M,
+	CVMX_PKI_PCAM_TERM_GLORT = CVMX_PKI_PCAM_TERM_E_GLORT_M,
+	CVMX_PKI_PCAM_TERM_DSA = CVMX_PKI_PCAM_TERM_E_DSA_M,
+	CVMX_PKI_PCAM_TERM_ETHTYPE0 = CVMX_PKI_PCAM_TERM_E_ETHTYPE0_M,
+	CVMX_PKI_PCAM_TERM_ETHTYPE1 = CVMX_PKI_PCAM_TERM_E_ETHTYPE1_M,
+	CVMX_PKI_PCAM_TERM_ETHTYPE2 = CVMX_PKI_PCAM_TERM_E_ETHTYPE2_M,
+	CVMX_PKI_PCAM_TERM_ETHTYPE3 = CVMX_PKI_PCAM_TERM_E_ETHTYPE3_M,
+	CVMX_PKI_PCAM_TERM_MPLS0 = CVMX_PKI_PCAM_TERM_E_MPLS0_M,
+	CVMX_PKI_PCAM_TERM_L3_SIPHH = CVMX_PKI_PCAM_TERM_E_L3_SIPHH_M,
+	CVMX_PKI_PCAM_TERM_L3_SIPMH = CVMX_PKI_PCAM_TERM_E_L3_SIPMH_M,
+	CVMX_PKI_PCAM_TERM_L3_SIPML = CVMX_PKI_PCAM_TERM_E_L3_SIPML_M,
+	CVMX_PKI_PCAM_TERM_L3_SIPLL = CVMX_PKI_PCAM_TERM_E_L3_SIPLL_M,
+	CVMX_PKI_PCAM_TERM_L3_FLAGS = CVMX_PKI_PCAM_TERM_E_L3_FLAGS_M,
+	CVMX_PKI_PCAM_TERM_L3_DIPHH = CVMX_PKI_PCAM_TERM_E_L3_DIPHH_M,
+	CVMX_PKI_PCAM_TERM_L3_DIPMH = CVMX_PKI_PCAM_TERM_E_L3_DIPMH_M,
+	CVMX_PKI_PCAM_TERM_L3_DIPML = CVMX_PKI_PCAM_TERM_E_L3_DIPML_M,
+	CVMX_PKI_PCAM_TERM_L3_DIPLL = CVMX_PKI_PCAM_TERM_E_L3_DIPLL_M,
+	CVMX_PKI_PCAM_TERM_LD_VNI = CVMX_PKI_PCAM_TERM_E_LD_VNI_M,
+	CVMX_PKI_PCAM_TERM_IL3_FLAGS = CVMX_PKI_PCAM_TERM_E_IL3_FLAGS_M,
+	CVMX_PKI_PCAM_TERM_LF_SPI = CVMX_PKI_PCAM_TERM_E_LF_SPI_M,
+	CVMX_PKI_PCAM_TERM_L4_PORT = CVMX_PKI_PCAM_TERM_E_L4_PORT_M,
+	CVMX_PKI_PCAM_TERM_L4_SPORT = CVMX_PKI_PCAM_TERM_E_L4_SPORT_M,
+	CVMX_PKI_PCAM_TERM_LG_CUSTOM = CVMX_PKI_PCAM_TERM_E_LG_CUSTOM_M
+};
+
+#define CVMX_PKI_DMACH_SHIFT	  32
+#define CVMX_PKI_DMACH_MASK	  cvmx_build_mask(16)
+#define CVMX_PKI_DMACL_MASK	  CVMX_PKI_DATA_MASK_32
+#define CVMX_PKI_DATA_MASK_32	  cvmx_build_mask(32)
+#define CVMX_PKI_DATA_MASK_16	  cvmx_build_mask(16)
+#define CVMX_PKI_DMAC_MATCH_EXACT cvmx_build_mask(48)
+
+struct cvmx_pki_pcam_input {
+	u64 style;
+	u64 style_mask; /* bits: 1-match, 0-dont care */
+	enum cvmx_pki_term field;
+	u32 field_mask; /* bits: 1-match, 0-dont care */
+	u64 data;
+	u64 data_mask; /* bits: 1-match, 0-dont care */
+};
+
+struct cvmx_pki_pcam_action {
+	enum cvmx_pki_parse_mode_chg parse_mode_chg;
+	enum cvmx_pki_layer_type layer_type_set;
+	int style_add;
+	int parse_flag_set;
+	int pointer_advance;
+};
+
+struct cvmx_pki_pcam_config {
+	int in_use;
+	int entry_num;
+	u64 cluster_mask;
+	struct cvmx_pki_pcam_input pcam_input;
+	struct cvmx_pki_pcam_action pcam_action;
+};
+
+/**
+ * Status statistics for a port
+ */
+struct cvmx_pki_port_stats {
+	u64 dropped_octets;
+	u64 dropped_packets;
+	u64 pci_raw_packets;
+	u64 octets;
+	u64 packets;
+	u64 multicast_packets;
+	u64 broadcast_packets;
+	u64 len_64_packets;
+	u64 len_65_127_packets;
+	u64 len_128_255_packets;
+	u64 len_256_511_packets;
+	u64 len_512_1023_packets;
+	u64 len_1024_1518_packets;
+	u64 len_1519_max_packets;
+	u64 fcs_align_err_packets;
+	u64 runt_packets;
+	u64 runt_crc_packets;
+	u64 oversize_packets;
+	u64 oversize_crc_packets;
+	u64 inb_packets;
+	u64 inb_octets;
+	u64 inb_errors;
+	u64 mcast_l2_red_packets;
+	u64 bcast_l2_red_packets;
+	u64 mcast_l3_red_packets;
+	u64 bcast_l3_red_packets;
+};
+
+/**
+ * PKI Packet Instruction Header Structure (PKI_INST_HDR_S)
+ */
+typedef union {
+	u64 u64;
+	struct {
+		u64 w : 1;    /* INST_HDR size: 0 = 2 bytes, 1 = 4 or 8 bytes */
+		u64 raw : 1;  /* RAW packet indicator in WQE[RAW]: 1 = enable */
+		u64 utag : 1; /* Use INST_HDR[TAG] to compute WQE[TAG]: 1 = enable */
+		u64 uqpg : 1; /* Use INST_HDR[QPG] to compute QPG: 1 = enable */
+		u64 rsvd1 : 1;
+		u64 pm : 3; /* Packet parsing mode. Legal values = 0x0..0x7 */
+		u64 sl : 8; /* Number of bytes in INST_HDR. */
+		/* The following fields are not present, if INST_HDR[W] = 0: */
+		u64 utt : 1; /* Use INST_HDR[TT] to compute WQE[TT]: 1 = enable */
+		u64 tt : 2;  /* INST_HDR[TT] => WQE[TT], if INST_HDR[UTT] = 1 */
+		u64 rsvd2 : 2;
+		u64 qpg : 11; /* INST_HDR[QPG] => QPG, if INST_HDR[UQPG] = 1 */
+		u64 tag : 32; /* INST_HDR[TAG] => WQE[TAG], if INST_HDR[UTAG] = 1 */
+	} s;
+} cvmx_pki_inst_hdr_t;
+
+/**
+ * This function assignes the clusters to a group, later pkind can be
+ * configured to use that group depending on number of clusters pkind
+ * would use. A given cluster can only be enabled in a single cluster group.
+ * Number of clusters assign to that group determines how many engine can work
+ * in parallel to process the packet. Eack cluster can process x MPPS.
+ *
+ * @param node	Node
+ * @param cluster_group Group to attach clusters to.
+ * @param cluster_mask The mask of clusters which needs to be assigned to the group.
+ */
+static inline int cvmx_pki_attach_cluster_to_group(int node, u64 cluster_group, u64 cluster_mask)
+{
+	cvmx_pki_icgx_cfg_t pki_cl_grp;
+
+	if (cluster_group >= CVMX_PKI_NUM_CLUSTER_GROUP) {
+		debug("ERROR: config cluster group %d", (int)cluster_group);
+		return -1;
+	}
+	pki_cl_grp.u64 = cvmx_read_csr_node(node, CVMX_PKI_ICGX_CFG(cluster_group));
+	pki_cl_grp.s.clusters = cluster_mask;
+	cvmx_write_csr_node(node, CVMX_PKI_ICGX_CFG(cluster_group), pki_cl_grp.u64);
+	return 0;
+}
+
+static inline void cvmx_pki_write_global_parse(int node, struct cvmx_pki_global_parse gbl_pen)
+{
+	cvmx_pki_gbl_pen_t gbl_pen_reg;
+
+	gbl_pen_reg.u64 = cvmx_read_csr_node(node, CVMX_PKI_GBL_PEN);
+	gbl_pen_reg.s.virt_pen = gbl_pen.virt_pen;
+	gbl_pen_reg.s.clg_pen = gbl_pen.clg_pen;
+	gbl_pen_reg.s.cl2_pen = gbl_pen.cl2_pen;
+	gbl_pen_reg.s.l4_pen = gbl_pen.l4_pen;
+	gbl_pen_reg.s.il3_pen = gbl_pen.il3_pen;
+	gbl_pen_reg.s.l3_pen = gbl_pen.l3_pen;
+	gbl_pen_reg.s.mpls_pen = gbl_pen.mpls_pen;
+	gbl_pen_reg.s.fulc_pen = gbl_pen.fulc_pen;
+	gbl_pen_reg.s.dsa_pen = gbl_pen.dsa_pen;
+	gbl_pen_reg.s.hg_pen = gbl_pen.hg_pen;
+	cvmx_write_csr_node(node, CVMX_PKI_GBL_PEN, gbl_pen_reg.u64);
+}
+
+static inline void cvmx_pki_write_tag_secret(int node, struct cvmx_pki_tag_sec tag_secret)
+{
+	cvmx_pki_tag_secret_t tag_secret_reg;
+
+	tag_secret_reg.u64 = cvmx_read_csr_node(node, CVMX_PKI_TAG_SECRET);
+	tag_secret_reg.s.dst6 = tag_secret.dst6;
+	tag_secret_reg.s.src6 = tag_secret.src6;
+	tag_secret_reg.s.dst = tag_secret.dst;
+	tag_secret_reg.s.src = tag_secret.src;
+	cvmx_write_csr_node(node, CVMX_PKI_TAG_SECRET, tag_secret_reg.u64);
+}
+
+static inline void cvmx_pki_write_ltype_map(int node, enum cvmx_pki_layer_type layer,
+					    enum cvmx_pki_beltype backend)
+{
+	cvmx_pki_ltypex_map_t ltype_map;
+
+	if (layer > CVMX_PKI_LTYPE_E_MAX || backend > CVMX_PKI_BELTYPE_MAX) {
+		debug("ERROR: invalid ltype beltype mapping\n");
+		return;
+	}
+	ltype_map.u64 = cvmx_read_csr_node(node, CVMX_PKI_LTYPEX_MAP(layer));
+	ltype_map.s.beltype = backend;
+	cvmx_write_csr_node(node, CVMX_PKI_LTYPEX_MAP(layer), ltype_map.u64);
+}
+
+/**
+ * This function enables the cluster group to start parsing.
+ *
+ * @param node    Node number.
+ * @param cl_grp  Cluster group to enable parsing.
+ */
+static inline int cvmx_pki_parse_enable(int node, unsigned int cl_grp)
+{
+	cvmx_pki_icgx_cfg_t pki_cl_grp;
+
+	if (cl_grp >= CVMX_PKI_NUM_CLUSTER_GROUP) {
+		debug("ERROR: pki parse en group %d", (int)cl_grp);
+		return -1;
+	}
+	pki_cl_grp.u64 = cvmx_read_csr_node(node, CVMX_PKI_ICGX_CFG(cl_grp));
+	pki_cl_grp.s.pena = 1;
+	cvmx_write_csr_node(node, CVMX_PKI_ICGX_CFG(cl_grp), pki_cl_grp.u64);
+	return 0;
+}
+
+/**
+ * This function enables the PKI to send bpid level backpressure to CN78XX inputs.
+ *
+ * @param node Node number.
+ */
+static inline void cvmx_pki_enable_backpressure(int node)
+{
+	cvmx_pki_buf_ctl_t pki_buf_ctl;
+
+	pki_buf_ctl.u64 = cvmx_read_csr_node(node, CVMX_PKI_BUF_CTL);
+	pki_buf_ctl.s.pbp_en = 1;
+	cvmx_write_csr_node(node, CVMX_PKI_BUF_CTL, pki_buf_ctl.u64);
+}
+
+/**
+ * Clear the statistics counters for a port.
+ *
+ * @param node Node number.
+ * @param port Port number (ipd_port) to get statistics for.
+ *    Make sure PKI_STATS_CTL:mode is set to 0 for collecting per port/pkind stats.
+ */
+void cvmx_pki_clear_port_stats(int node, u64 port);
+
+/**
+ * Get the status counters for index from PKI.
+ *
+ * @param node	  Node number.
+ * @param index   PKIND number, if PKI_STATS_CTL:mode = 0 or
+ *     style(flow) number, if PKI_STATS_CTL:mode = 1
+ * @param status  Where to put the results.
+ */
+void cvmx_pki_get_stats(int node, int index, struct cvmx_pki_port_stats *status);
+
+/**
+ * Get the statistics counters for a port.
+ *
+ * @param node	 Node number
+ * @param port   Port number (ipd_port) to get statistics for.
+ *    Make sure PKI_STATS_CTL:mode is set to 0 for collecting per port/pkind stats.
+ * @param status Where to put the results.
+ */
+static inline void cvmx_pki_get_port_stats(int node, u64 port, struct cvmx_pki_port_stats *status)
+{
+	int xipd = cvmx_helper_node_to_ipd_port(node, port);
+	int xiface = cvmx_helper_get_interface_num(xipd);
+	int index = cvmx_helper_get_interface_index_num(port);
+	int pknd = cvmx_helper_get_pknd(xiface, index);
+
+	cvmx_pki_get_stats(node, pknd, status);
+}
+
+/**
+ * Get the statistics counters for a flow represented by style in PKI.
+ *
+ * @param node Node number.
+ * @param style_num Style number to get statistics for.
+ *    Make sure PKI_STATS_CTL:mode is set to 1 for collecting per style/flow stats.
+ * @param status Where to put the results.
+ */
+static inline void cvmx_pki_get_flow_stats(int node, u64 style_num,
+					   struct cvmx_pki_port_stats *status)
+{
+	cvmx_pki_get_stats(node, style_num, status);
+}
+
+/**
+ * Show integrated PKI configuration.
+ *
+ * @param node	   node number
+ */
+int cvmx_pki_config_dump(unsigned int node);
+
+/**
+ * Show integrated PKI statistics.
+ *
+ * @param node	   node number
+ */
+int cvmx_pki_stats_dump(unsigned int node);
+
+/**
+ * Clear PKI statistics.
+ *
+ * @param node	   node number
+ */
+void cvmx_pki_stats_clear(unsigned int node);
+
+/**
+ * This function enables PKI.
+ *
+ * @param node	 node to enable pki in.
+ */
+void cvmx_pki_enable(int node);
+
+/**
+ * This function disables PKI.
+ *
+ * @param node	node to disable pki in.
+ */
+void cvmx_pki_disable(int node);
+
+/**
+ * This function soft resets PKI.
+ *
+ * @param node	node to enable pki in.
+ */
+void cvmx_pki_reset(int node);
+
+/**
+ * This function sets the clusters in PKI.
+ *
+ * @param node	node to set clusters in.
+ */
+int cvmx_pki_setup_clusters(int node);
+
+/**
+ * This function reads global configuration of PKI block.
+ *
+ * @param node    Node number.
+ * @param gbl_cfg Pointer to struct to read global configuration
+ */
+void cvmx_pki_read_global_config(int node, struct cvmx_pki_global_config *gbl_cfg);
+
+/**
+ * This function writes global configuration of PKI into hw.
+ *
+ * @param node    Node number.
+ * @param gbl_cfg Pointer to struct to global configuration
+ */
+void cvmx_pki_write_global_config(int node, struct cvmx_pki_global_config *gbl_cfg);
+
+/**
+ * This function reads per pkind parameters in hardware which defines how
+ * the incoming packet is processed.
+ *
+ * @param node   Node number.
+ * @param pkind  PKI supports a large number of incoming interfaces and packets
+ *     arriving on different interfaces or channels may want to be processed
+ *     differently. PKI uses the pkind to determine how the incoming packet
+ *     is processed.
+ * @param pkind_cfg	Pointer to struct conatining pkind configuration read
+ *     from hardware.
+ */
+int cvmx_pki_read_pkind_config(int node, int pkind, struct cvmx_pki_pkind_config *pkind_cfg);
+
+/**
+ * This function writes per pkind parameters in hardware which defines how
+ * the incoming packet is processed.
+ *
+ * @param node   Node number.
+ * @param pkind  PKI supports a large number of incoming interfaces and packets
+ *     arriving on different interfaces or channels may want to be processed
+ *     differently. PKI uses the pkind to determine how the incoming packet
+ *     is processed.
+ * @param pkind_cfg	Pointer to struct conatining pkind configuration need
+ *     to be written in hardware.
+ */
+int cvmx_pki_write_pkind_config(int node, int pkind, struct cvmx_pki_pkind_config *pkind_cfg);
+
+/**
+ * This function reads parameters associated with tag configuration in hardware.
+ *
+ * @param node	 Node number.
+ * @param style  Style to configure tag for.
+ * @param cluster_mask  Mask of clusters to configure the style for.
+ * @param tag_cfg  Pointer to tag configuration struct.
+ */
+void cvmx_pki_read_tag_config(int node, int style, u64 cluster_mask,
+			      struct cvmx_pki_style_tag_cfg *tag_cfg);
+
+/**
+ * This function writes/configures parameters associated with tag
+ * configuration in hardware.
+ *
+ * @param node  Node number.
+ * @param style  Style to configure tag for.
+ * @param cluster_mask  Mask of clusters to configure the style for.
+ * @param tag_cfg  Pointer to taf configuration struct.
+ */
+void cvmx_pki_write_tag_config(int node, int style, u64 cluster_mask,
+			       struct cvmx_pki_style_tag_cfg *tag_cfg);
+
+/**
+ * This function reads parameters associated with style in hardware.
+ *
+ * @param node	Node number.
+ * @param style  Style to read from.
+ * @param cluster_mask  Mask of clusters style belongs to.
+ * @param style_cfg  Pointer to style config struct.
+ */
+void cvmx_pki_read_style_config(int node, int style, u64 cluster_mask,
+				struct cvmx_pki_style_config *style_cfg);
+
+/**
+ * This function writes/configures parameters associated with style in hardware.
+ *
+ * @param node  Node number.
+ * @param style  Style to configure.
+ * @param cluster_mask  Mask of clusters to configure the style for.
+ * @param style_cfg  Pointer to style config struct.
+ */
+void cvmx_pki_write_style_config(int node, u64 style, u64 cluster_mask,
+				 struct cvmx_pki_style_config *style_cfg);
+/**
+ * This function reads qpg entry at specified offset from qpg table
+ *
+ * @param node  Node number.
+ * @param offset  Offset in qpg table to read from.
+ * @param qpg_cfg  Pointer to structure containing qpg values
+ */
+int cvmx_pki_read_qpg_entry(int node, int offset, struct cvmx_pki_qpg_config *qpg_cfg);
+
+/**
+ * This function writes qpg entry at specified offset in qpg table
+ *
+ * @param node  Node number.
+ * @param offset  Offset in qpg table to write to.
+ * @param qpg_cfg  Pointer to stricture containing qpg values.
+ */
+void cvmx_pki_write_qpg_entry(int node, int offset, struct cvmx_pki_qpg_config *qpg_cfg);
+
+/**
+ * This function writes pcam entry at given offset in pcam table in hardware
+ *
+ * @param node  Node number.
+ * @param index	 Offset in pcam table.
+ * @param cluster_mask  Mask of clusters in which to write pcam entry.
+ * @param input  Input keys to pcam match passed as struct.
+ * @param action  PCAM match action passed as struct
+ */
+int cvmx_pki_pcam_write_entry(int node, int index, u64 cluster_mask,
+			      struct cvmx_pki_pcam_input input, struct cvmx_pki_pcam_action action);
+/**
+ * Configures the channel which will receive backpressure from the specified bpid.
+ * Each channel listens for backpressure on a specific bpid.
+ * Each bpid can backpressure multiple channels.
+ * @param node  Node number.
+ * @param bpid  BPID from which channel will receive backpressure.
+ * @param channel  Channel number to receive backpressue.
+ */
+int cvmx_pki_write_channel_bpid(int node, int channel, int bpid);
+
+/**
+ * Configures the bpid on which, specified channel will
+ * assert backpressure.
+ * Each bpid receives backpressure from auras.
+ * Multiple auras can backpressure single bpid.
+ * @param node  Node number.
+ * @param aura  Number which will assert backpressure on that bpid.
+ * @param bpid  To assert backpressure on.
+ */
+int cvmx_pki_write_aura_bpid(int node, int aura, int bpid);
+
+/**
+ * Enables/Disabled QoS (RED Drop, Tail Drop & backpressure) for the* PKI aura.
+ *
+ * @param node  Node number
+ * @param aura  To enable/disable QoS on.
+ * @param ena_red  Enable/Disable RED drop between pass and drop level
+ *    1-enable 0-disable
+ * @param ena_drop  Enable/disable tail drop when max drop level exceeds
+ *    1-enable 0-disable
+ * @param ena_bp  Enable/Disable asserting backpressure on bpid when
+ *    max DROP level exceeds.
+ *    1-enable 0-disable
+ */
+int cvmx_pki_enable_aura_qos(int node, int aura, bool ena_red, bool ena_drop, bool ena_bp);
+
+/**
+ * This function gives the initial style used by that pkind.
+ *
+ * @param node  Node number.
+ * @param pkind  PKIND number.
+ */
+int cvmx_pki_get_pkind_style(int node, int pkind);
+
+/**
+ * This function sets the wqe buffer mode. First packet data buffer can reside
+ * either in same buffer as wqe OR it can go in separate buffer. If used the later mode,
+ * make sure software allocate enough buffers to now have wqe separate from packet data.
+ *
+ * @param node  Node number.
+ * @param style  Style to configure.
+ * @param pkt_outside_wqe
+ *    0 = The packet link pointer will be at word [FIRST_SKIP] immediately
+ *    followed by packet data, in the same buffer as the work queue entry.
+ *    1 = The packet link pointer will be at word [FIRST_SKIP] in a new
+ *    buffer separate from the work queue entry. Words following the
+ *    WQE in the same cache line will be zeroed, other lines in the
+ *    buffer will not be modified and will retain stale data (from the
+ *    buffer’s previous use). This setting may decrease the peak PKI
+ *    performance by up to half on small packets.
+ */
+void cvmx_pki_set_wqe_mode(int node, u64 style, bool pkt_outside_wqe);
+
+/**
+ * This function sets the Packet mode of all ports and styles to little-endian.
+ * It Changes write operations of packet data to L2C to
+ * be in little-endian. Does not change the WQE header format, which is
+ * properly endian neutral.
+ *
+ * @param node  Node number.
+ * @param style  Style to configure.
+ */
+void cvmx_pki_set_little_endian(int node, u64 style);
+
+/**
+ * Enables/Disables L2 length error check and max & min frame length checks.
+ *
+ * @param node  Node number.
+ * @param pknd  PKIND to disable error for.
+ * @param l2len_err	 L2 length error check enable.
+ * @param maxframe_err	Max frame error check enable.
+ * @param minframe_err	Min frame error check enable.
+ *    1 -- Enabel err checks
+ *    0 -- Disable error checks
+ */
+void cvmx_pki_endis_l2_errs(int node, int pknd, bool l2len_err, bool maxframe_err,
+			    bool minframe_err);
+
+/**
+ * Enables/Disables fcs check and fcs stripping on the pkind.
+ *
+ * @param node  Node number.
+ * @param pknd  PKIND to apply settings on.
+ * @param fcs_chk  Enable/disable fcs check.
+ *    1 -- enable fcs error check.
+ *    0 -- disable fcs error check.
+ * @param fcs_strip	 Strip L2 FCS bytes from packet, decrease WQE[LEN] by 4 bytes
+ *    1 -- strip L2 FCS.
+ *    0 -- Do not strip L2 FCS.
+ */
+void cvmx_pki_endis_fcs_check(int node, int pknd, bool fcs_chk, bool fcs_strip);
+
+/**
+ * This function shows the qpg table entries, read directly from hardware.
+ *
+ * @param node  Node number.
+ * @param num_entry  Number of entries to print.
+ */
+void cvmx_pki_show_qpg_entries(int node, u16 num_entry);
+
+/**
+ * This function shows the pcam table in raw format read directly from hardware.
+ *
+ * @param node  Node number.
+ */
+void cvmx_pki_show_pcam_entries(int node);
+
+/**
+ * This function shows the valid entries in readable format,
+ * read directly from hardware.
+ *
+ * @param node  Node number.
+ */
+void cvmx_pki_show_valid_pcam_entries(int node);
+
+/**
+ * This function shows the pkind attributes in readable format,
+ * read directly from hardware.
+ * @param node  Node number.
+ * @param pkind  PKIND number to print.
+ */
+void cvmx_pki_show_pkind_attributes(int node, int pkind);
+
+/**
+ * @INTERNAL
+ * This function is called by cvmx_helper_shutdown() to extract all FPA buffers
+ * out of the PKI. After this function completes, all FPA buffers that were
+ * prefetched by PKI will be in the appropriate FPA pool.
+ * This functions does not reset the PKI.
+ * WARNING: It is very important that PKI be reset soon after a call to this function.
+ *
+ * @param node  Node number.
+ */
+void __cvmx_pki_free_ptr(int node);
+
+#endif
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-pko-internal-ports-range.h b/arch/mips/mach-octeon/include/mach/cvmx-pko-internal-ports-range.h
new file mode 100644
index 0000000000..1fb49b3fb6
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-pko-internal-ports-range.h
@@ -0,0 +1,43 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ */
+
+#ifndef __CVMX_INTERNAL_PORTS_RANGE__
+#define __CVMX_INTERNAL_PORTS_RANGE__
+
+/*
+ * Allocated a block of internal ports for the specified interface/port
+ *
+ * @param  interface  the interface for which the internal ports are requested
+ * @param  port       the index of the port within in the interface for which the internal ports
+ *                    are requested.
+ * @param  count      the number of internal ports requested
+ *
+ * @return  0 on success
+ *         -1 on failure
+ */
+int cvmx_pko_internal_ports_alloc(int interface, int port, u64 count);
+
+/*
+ * Free the internal ports associated with the specified interface/port
+ *
+ * @param  interface  the interface for which the internal ports are requested
+ * @param  port       the index of the port within in the interface for which the internal ports
+ *                    are requested.
+ *
+ * @return  0 on success
+ *         -1 on failure
+ */
+int cvmx_pko_internal_ports_free(int interface, int port);
+
+/*
+ * Frees up all the allocated internal ports.
+ */
+void cvmx_pko_internal_ports_range_free_all(void);
+
+void cvmx_pko_internal_ports_range_show(void);
+
+int __cvmx_pko_internal_ports_range_init(void);
+
+#endif
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-pko3-queue.h b/arch/mips/mach-octeon/include/mach/cvmx-pko3-queue.h
new file mode 100644
index 0000000000..5f83989049
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-pko3-queue.h
@@ -0,0 +1,175 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ */
+
+#ifndef __CVMX_PKO3_QUEUE_H__
+#define __CVMX_PKO3_QUEUE_H__
+
+/**
+ * @INTERNAL
+ *
+ * Find or allocate global port/dq map table
+ * which is a named table, contains entries for
+ * all possible OCI nodes.
+ *
+ * The table global pointer is stored in core-local variable
+ * so that every core will call this function once, on first use.
+ */
+int __cvmx_pko3_dq_table_setup(void);
+
+/*
+ * Get the base Descriptor Queue number for an IPD port on the local node
+ */
+int cvmx_pko3_get_queue_base(int ipd_port);
+
+/*
+ * Get the number of Descriptor Queues assigned for an IPD port
+ */
+int cvmx_pko3_get_queue_num(int ipd_port);
+
+/**
+ * Get L1/Port Queue number assigned to interface port.
+ *
+ * @param xiface is interface number.
+ * @param index is port index.
+ */
+int cvmx_pko3_get_port_queue(int xiface, int index);
+
+/*
+ * Configure L3 through L5 Scheduler Queues and Descriptor Queues
+ *
+ * The Scheduler Queues in Levels 3 to 5 and Descriptor Queues are
+ * configured one-to-one or many-to-one to a single parent Scheduler
+ * Queues. The level of the parent SQ is specified in an argument,
+ * as well as the number of children to attach to the specific parent.
+ * The children can have fair round-robin or priority-based scheduling
+ * when multiple children are assigned a single parent.
+ *
+ * @param node is the OCI node location for the queues to be configured
+ * @param parent_level is the level of the parent queue, 2 to 5.
+ * @param parent_queue is the number of the parent Scheduler Queue
+ * @param child_base is the number of the first child SQ or DQ to assign to
+ * @param parent
+ * @param child_count is the number of consecutive children to assign
+ * @param stat_prio_count is the priority setting for the children L2 SQs
+ *
+ * If <stat_prio_count> is -1, the Ln children will have equal Round-Robin
+ * relationship with eachother. If <stat_prio_count> is 0, all Ln children
+ * will be arranged in Weighted-Round-Robin, with the first having the most
+ * precedence. If <stat_prio_count> is between 1 and 8, it indicates how
+ * many children will have static priority settings (with the first having
+ * the most precedence), with the remaining Ln children having WRR scheduling.
+ *
+ * @returns 0 on success, -1 on failure.
+ *
+ * Note: this function supports the configuration of node-local unit.
+ */
+int cvmx_pko3_sq_config_children(unsigned int node, unsigned int parent_level,
+				 unsigned int parent_queue, unsigned int child_base,
+				 unsigned int child_count, int stat_prio_count);
+
+/*
+ * @INTERNAL
+ * Register a range of Descriptor Queues wth an interface port
+ *
+ * This function poulates the DQ-to-IPD translation table
+ * used by the application to retrieve the DQ range (typically ordered
+ * by priority) for a given IPD-port, which is either a physical port,
+ * or a channel on a channelized interface (i.e. ILK).
+ *
+ * @param xiface is the physical interface number
+ * @param index is either a physical port on an interface
+ * @param or a channel of an ILK interface
+ * @param dq_base is the first Descriptor Queue number in a consecutive range
+ * @param dq_count is the number of consecutive Descriptor Queues leading
+ * @param the same channel or port.
+ *
+ * Only a consecurive range of Descriptor Queues can be associated with any
+ * given channel/port, and usually they are ordered from most to least
+ * in terms of scheduling priority.
+ *
+ * Note: thus function only populates the node-local translation table.
+ *
+ * @returns 0 on success, -1 on failure.
+ */
+int __cvmx_pko3_ipd_dq_register(int xiface, int index, unsigned int dq_base, unsigned int dq_count);
+
+/**
+ * @INTERNAL
+ *
+ * Unregister DQs associated with CHAN_E (IPD port)
+ */
+int __cvmx_pko3_ipd_dq_unregister(int xiface, int index);
+
+/*
+ * Map channel number in PKO
+ *
+ * @param node is to specify the node to which this configuration is applied.
+ * @param pq_num specifies the Port Queue (i.e. L1) queue number.
+ * @param l2_l3_q_num  specifies L2/L3 queue number.
+ * @param channel specifies the channel number to map to the queue.
+ *
+ * The channel assignment applies to L2 or L3 Shaper Queues depending
+ * on the setting of channel credit level.
+ *
+ * @return returns none.
+ */
+void cvmx_pko3_map_channel(unsigned int node, unsigned int pq_num, unsigned int l2_l3_q_num,
+			   u16 channel);
+
+int cvmx_pko3_pq_config(unsigned int node, unsigned int mac_num, unsigned int pq_num);
+
+int cvmx_pko3_port_cir_set(unsigned int node, unsigned int pq_num, unsigned long rate_kbips,
+			   unsigned int burst_bytes, int adj_bytes);
+int cvmx_pko3_dq_cir_set(unsigned int node, unsigned int pq_num, unsigned long rate_kbips,
+			 unsigned int burst_bytes);
+int cvmx_pko3_dq_pir_set(unsigned int node, unsigned int pq_num, unsigned long rate_kbips,
+			 unsigned int burst_bytes);
+typedef enum {
+	CVMX_PKO3_SHAPE_RED_STALL,
+	CVMX_PKO3_SHAPE_RED_DISCARD,
+	CVMX_PKO3_SHAPE_RED_PASS
+} red_action_t;
+
+void cvmx_pko3_dq_red(unsigned int node, unsigned int dq_num, red_action_t red_act,
+		      int8_t len_adjust);
+
+/**
+ * Macros to deal with short floating point numbers,
+ * where unsigned exponent, and an unsigned normalized
+ * mantissa are represented each with a defined field width.
+ *
+ */
+#define CVMX_SHOFT_MANT_BITS 8
+#define CVMX_SHOFT_EXP_BITS  4
+
+/**
+ * Convert short-float to an unsigned integer
+ * Note that it will lose precision.
+ */
+#define CVMX_SHOFT_TO_U64(m, e)                                                                    \
+	((((1ull << CVMX_SHOFT_MANT_BITS) | (m)) << (e)) >> CVMX_SHOFT_MANT_BITS)
+
+/**
+ * Convert to short-float from an unsigned integer
+ */
+#define CVMX_SHOFT_FROM_U64(ui, m, e)                                                              \
+	do {                                                                                       \
+		unsigned long long u;                                                              \
+		unsigned int k;                                                                    \
+		k = (1ull << (CVMX_SHOFT_MANT_BITS + 1)) - 1;                                      \
+		(e) = 0;                                                                           \
+		u = (ui) << CVMX_SHOFT_MANT_BITS;                                                  \
+		while ((u) > k) {                                                                  \
+			u >>= 1;                                                                   \
+			(e)++;                                                                     \
+		}                                                                                  \
+		(m) = u & (k >> 1);                                                                \
+	} while (0);
+
+#define CVMX_SHOFT_MAX()                                                                           \
+	CVMX_SHOFT_TO_U64((1 << CVMX_SHOFT_MANT_BITS) - 1, (1 << CVMX_SHOFT_EXP_BITS) - 1)
+#define CVMX_SHOFT_MIN() CVMX_SHOFT_TO_U64(0, 0)
+
+#endif /* __CVMX_PKO3_QUEUE_H__ */
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-pow.h b/arch/mips/mach-octeon/include/mach/cvmx-pow.h
new file mode 100644
index 0000000000..0680ca258f
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-pow.h
@@ -0,0 +1,2991 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ *
+ * Interface to the hardware Scheduling unit.
+ *
+ * New, starting with SDK 1.7.0, cvmx-pow supports a number of
+ * extended consistency checks. The define
+ * CVMX_ENABLE_POW_CHECKS controls the runtime insertion of POW
+ * internal state checks to find common programming errors. If
+ * CVMX_ENABLE_POW_CHECKS is not defined, checks are by default
+ * enabled. For example, cvmx-pow will check for the following
+ * program errors or POW state inconsistency.
+ * - Requesting a POW operation with an active tag switch in
+ *   progress.
+ * - Waiting for a tag switch to complete for an excessively
+ *   long period. This is normally a sign of an error in locking
+ *   causing deadlock.
+ * - Illegal tag switches from NULL_NULL.
+ * - Illegal tag switches from NULL.
+ * - Illegal deschedule request.
+ * - WQE pointer not matching the one attached to the core by
+ *   the POW.
+ */
+
+#ifndef __CVMX_POW_H__
+#define __CVMX_POW_H__
+
+#include "cvmx-wqe.h"
+#include "cvmx-pow-defs.h"
+#include "cvmx-sso-defs.h"
+#include "cvmx-address.h"
+#include "cvmx-coremask.h"
+
+/* Default to having all POW constancy checks turned on */
+#ifndef CVMX_ENABLE_POW_CHECKS
+#define CVMX_ENABLE_POW_CHECKS 1
+#endif
+
+/*
+ * Special type for CN78XX style SSO groups (0..255),
+ * for distinction from legacy-style groups (0..15)
+ */
+typedef union {
+	u8 xgrp;
+	/* Fields that map XGRP for backwards compatibility */
+	struct __attribute__((__packed__)) {
+		u8 group : 5;
+		u8 qus : 3;
+	};
+} cvmx_xgrp_t;
+
+/*
+ * Softwsare-only structure to convey a return value
+ * containing multiple information fields about an work queue entry
+ */
+typedef struct {
+	u32 tag;
+	u16 index;
+	u8 grp; /* Legacy group # (0..15) */
+	u8 tag_type;
+} cvmx_pow_tag_info_t;
+
+/**
+ * Wait flag values for pow functions.
+ */
+typedef enum {
+	CVMX_POW_WAIT = 1,
+	CVMX_POW_NO_WAIT = 0,
+} cvmx_pow_wait_t;
+
+/**
+ *  POW tag operations.  These are used in the data stored to the POW.
+ */
+typedef enum {
+	CVMX_POW_TAG_OP_SWTAG = 0L,
+	CVMX_POW_TAG_OP_SWTAG_FULL = 1L,
+	CVMX_POW_TAG_OP_SWTAG_DESCH = 2L,
+	CVMX_POW_TAG_OP_DESCH = 3L,
+	CVMX_POW_TAG_OP_ADDWQ = 4L,
+	CVMX_POW_TAG_OP_UPDATE_WQP_GRP = 5L,
+	CVMX_POW_TAG_OP_SET_NSCHED = 6L,
+	CVMX_POW_TAG_OP_CLR_NSCHED = 7L,
+	CVMX_POW_TAG_OP_NOP = 15L
+} cvmx_pow_tag_op_t;
+
+/**
+ * This structure defines the store data on a store to POW
+ */
+typedef union {
+	u64 u64;
+	struct {
+		u64 no_sched : 1;
+		u64 unused : 2;
+		u64 index : 13;
+		cvmx_pow_tag_op_t op : 4;
+		u64 unused2 : 2;
+		u64 qos : 3;
+		u64 grp : 4;
+		cvmx_pow_tag_type_t type : 3;
+		u64 tag : 32;
+	} s_cn38xx;
+	struct {
+		u64 no_sched : 1;
+		cvmx_pow_tag_op_t op : 4;
+		u64 unused1 : 4;
+		u64 index : 11;
+		u64 unused2 : 1;
+		u64 grp : 6;
+		u64 unused3 : 3;
+		cvmx_pow_tag_type_t type : 2;
+		u64 tag : 32;
+	} s_cn68xx_clr;
+	struct {
+		u64 no_sched : 1;
+		cvmx_pow_tag_op_t op : 4;
+		u64 unused1 : 12;
+		u64 qos : 3;
+		u64 unused2 : 1;
+		u64 grp : 6;
+		u64 unused3 : 3;
+		cvmx_pow_tag_type_t type : 2;
+		u64 tag : 32;
+	} s_cn68xx_add;
+	struct {
+		u64 no_sched : 1;
+		cvmx_pow_tag_op_t op : 4;
+		u64 unused1 : 16;
+		u64 grp : 6;
+		u64 unused3 : 3;
+		cvmx_pow_tag_type_t type : 2;
+		u64 tag : 32;
+	} s_cn68xx_other;
+	struct {
+		u64 rsvd_62_63 : 2;
+		u64 grp : 10;
+		cvmx_pow_tag_type_t type : 2;
+		u64 no_sched : 1;
+		u64 rsvd_48 : 1;
+		cvmx_pow_tag_op_t op : 4;
+		u64 rsvd_42_43 : 2;
+		u64 wqp : 42;
+	} s_cn78xx_other;
+
+} cvmx_pow_tag_req_t;
+
+union cvmx_pow_tag_req_addr {
+	u64 u64;
+	struct {
+		u64 mem_region : 2;
+		u64 reserved_49_61 : 13;
+		u64 is_io : 1;
+		u64 did : 8;
+		u64 addr : 40;
+	} s;
+	struct {
+		u64 mem_region : 2;
+		u64 reserved_49_61 : 13;
+		u64 is_io : 1;
+		u64 did : 8;
+		u64 node : 4;
+		u64 tag : 32;
+		u64 reserved_0_3 : 4;
+	} s_cn78xx;
+};
+
+/**
+ * This structure describes the address to load stuff from POW
+ */
+typedef union {
+	u64 u64;
+	/**
+	 * Address for new work request loads (did<2:0> == 0)
+	 */
+	struct {
+		u64 mem_region : 2;
+		u64 reserved_49_61 : 13;
+		u64 is_io : 1;
+		u64 did : 8;
+		u64 reserved_4_39 : 36;
+		u64 wait : 1;
+		u64 reserved_0_2 : 3;
+	} swork;
+	struct {
+		u64 mem_region : 2;
+		u64 reserved_49_61 : 13;
+		u64 is_io : 1;
+		u64 did : 8;
+		u64 node : 4;
+		u64 reserved_32_35 : 4;
+		u64 indexed : 1;
+		u64 grouped : 1;
+		u64 rtngrp : 1;
+		u64 reserved_16_28 : 13;
+		u64 index : 12;
+		u64 wait : 1;
+		u64 reserved_0_2 : 3;
+	} swork_78xx;
+	/**
+	 * Address for loads to get POW internal status
+	 */
+	struct {
+		u64 mem_region : 2;
+		u64 reserved_49_61 : 13;
+		u64 is_io : 1;
+		u64 did : 8;
+		u64 reserved_10_39 : 30;
+		u64 coreid : 4;
+		u64 get_rev : 1;
+		u64 get_cur : 1;
+		u64 get_wqp : 1;
+		u64 reserved_0_2 : 3;
+	} sstatus;
+	/**
+	 * Address for loads to get 68XX SS0 internal status
+	 */
+	struct {
+		u64 mem_region : 2;
+		u64 reserved_49_61 : 13;
+		u64 is_io : 1;
+		u64 did : 8;
+		u64 reserved_14_39 : 26;
+		u64 coreid : 5;
+		u64 reserved_6_8 : 3;
+		u64 opcode : 3;
+		u64 reserved_0_2 : 3;
+	} sstatus_cn68xx;
+	/**
+	 * Address for memory loads to get POW internal state
+	 */
+	struct {
+		u64 mem_region : 2;
+		u64 reserved_49_61 : 13;
+		u64 is_io : 1;
+		u64 did : 8;
+		u64 reserved_16_39 : 24;
+		u64 index : 11;
+		u64 get_des : 1;
+		u64 get_wqp : 1;
+		u64 reserved_0_2 : 3;
+	} smemload;
+	/**
+	 * Address for memory loads to get SSO internal state
+	 */
+	struct {
+		u64 mem_region : 2;
+		u64 reserved_49_61 : 13;
+		u64 is_io : 1;
+		u64 did : 8;
+		u64 reserved_20_39 : 20;
+		u64 index : 11;
+		u64 reserved_6_8 : 3;
+		u64 opcode : 3;
+		u64 reserved_0_2 : 3;
+	} smemload_cn68xx;
+	/**
+	 * Address for index/pointer loads
+	 */
+	struct {
+		u64 mem_region : 2;
+		u64 reserved_49_61 : 13;
+		u64 is_io : 1;
+		u64 did : 8;
+		u64 reserved_9_39 : 31;
+		u64 qosgrp : 4;
+		u64 get_des_get_tail : 1;
+		u64 get_rmt : 1;
+		u64 reserved_0_2 : 3;
+	} sindexload;
+	/**
+	 * Address for a Index/Pointer loads to get SSO internal state
+	 */
+	struct {
+		u64 mem_region : 2;
+		u64 reserved_49_61 : 13;
+		u64 is_io : 1;
+		u64 did : 8;
+		u64 reserved_15_39 : 25;
+		u64 qos_grp : 6;
+		u64 reserved_6_8 : 3;
+		u64 opcode : 3;
+		u64 reserved_0_2 : 3;
+	} sindexload_cn68xx;
+	/**
+	 * Address for NULL_RD request (did<2:0> == 4)
+	 * when this is read, HW attempts to change the state to NULL if it is NULL_NULL
+	 * (the hardware cannot switch from NULL_NULL to NULL if a POW entry is not available -
+	 * software may need to recover by finishing another piece of work before a POW
+	 * entry can ever become available.)
+	 */
+	struct {
+		u64 mem_region : 2;
+		u64 reserved_49_61 : 13;
+		u64 is_io : 1;
+		u64 did : 8;
+		u64 reserved_0_39 : 40;
+	} snull_rd;
+} cvmx_pow_load_addr_t;
+
+/**
+ * This structure defines the response to a load/SENDSINGLE to POW (except CSR reads)
+ */
+typedef union {
+	u64 u64;
+	/**
+	 * Response to new work request loads
+	 */
+	struct {
+		u64 no_work : 1;
+		u64 pend_switch : 1;
+		u64 tt : 2;
+		u64 reserved_58_59 : 2;
+		u64 grp : 10;
+		u64 reserved_42_47 : 6;
+		u64 addr : 42;
+	} s_work;
+
+	/**
+	 * Result for a POW Status Load (when get_cur==0 and get_wqp==0)
+	 */
+	struct {
+		u64 reserved_62_63 : 2;
+		u64 pend_switch : 1;
+		u64 pend_switch_full : 1;
+		u64 pend_switch_null : 1;
+		u64 pend_desched : 1;
+		u64 pend_desched_switch : 1;
+		u64 pend_nosched : 1;
+		u64 pend_new_work : 1;
+		u64 pend_new_work_wait : 1;
+		u64 pend_null_rd : 1;
+		u64 pend_nosched_clr : 1;
+		u64 reserved_51 : 1;
+		u64 pend_index : 11;
+		u64 pend_grp : 4;
+		u64 reserved_34_35 : 2;
+		u64 pend_type : 2;
+		u64 pend_tag : 32;
+	} s_sstatus0;
+	/**
+	 * Result for a SSO Status Load (when opcode is SL_PENDTAG)
+	 */
+	struct {
+		u64 pend_switch : 1;
+		u64 pend_get_work : 1;
+		u64 pend_get_work_wait : 1;
+		u64 pend_nosched : 1;
+		u64 pend_nosched_clr : 1;
+		u64 pend_desched : 1;
+		u64 pend_alloc_we : 1;
+		u64 reserved_48_56 : 9;
+		u64 pend_index : 11;
+		u64 reserved_34_36 : 3;
+		u64 pend_type : 2;
+		u64 pend_tag : 32;
+	} s_sstatus0_cn68xx;
+	/**
+	 * Result for a POW Status Load (when get_cur==0 and get_wqp==1)
+	 */
+	struct {
+		u64 reserved_62_63 : 2;
+		u64 pend_switch : 1;
+		u64 pend_switch_full : 1;
+		u64 pend_switch_null : 1;
+		u64 pend_desched : 1;
+		u64 pend_desched_switch : 1;
+		u64 pend_nosched : 1;
+		u64 pend_new_work : 1;
+		u64 pend_new_work_wait : 1;
+		u64 pend_null_rd : 1;
+		u64 pend_nosched_clr : 1;
+		u64 reserved_51 : 1;
+		u64 pend_index : 11;
+		u64 pend_grp : 4;
+		u64 pend_wqp : 36;
+	} s_sstatus1;
+	/**
+	 * Result for a SSO Status Load (when opcode is SL_PENDWQP)
+	 */
+	struct {
+		u64 pend_switch : 1;
+		u64 pend_get_work : 1;
+		u64 pend_get_work_wait : 1;
+		u64 pend_nosched : 1;
+		u64 pend_nosched_clr : 1;
+		u64 pend_desched : 1;
+		u64 pend_alloc_we : 1;
+		u64 reserved_51_56 : 6;
+		u64 pend_index : 11;
+		u64 reserved_38_39 : 2;
+		u64 pend_wqp : 38;
+	} s_sstatus1_cn68xx;
+
+	struct {
+		u64 pend_switch : 1;
+		u64 pend_get_work : 1;
+		u64 pend_get_work_wait : 1;
+		u64 pend_nosched : 1;
+		u64 pend_nosched_clr : 1;
+		u64 pend_desched : 1;
+		u64 pend_alloc_we : 1;
+		u64 reserved_56 : 1;
+		u64 prep_index : 12;
+		u64 reserved_42_43 : 2;
+		u64 pend_tag : 42;
+	} s_sso_ppx_pendwqp_cn78xx;
+	/**
+	 * Result for a POW Status Load (when get_cur==1, get_wqp==0, and get_rev==0)
+	 */
+	struct {
+		u64 reserved_62_63 : 2;
+		u64 link_index : 11;
+		u64 index : 11;
+		u64 grp : 4;
+		u64 head : 1;
+		u64 tail : 1;
+		u64 tag_type : 2;
+		u64 tag : 32;
+	} s_sstatus2;
+	/**
+	 * Result for a SSO Status Load (when opcode is SL_TAG)
+	 */
+	struct {
+		u64 reserved_57_63 : 7;
+		u64 index : 11;
+		u64 reserved_45 : 1;
+		u64 grp : 6;
+		u64 head : 1;
+		u64 tail : 1;
+		u64 reserved_34_36 : 3;
+		u64 tag_type : 2;
+		u64 tag : 32;
+	} s_sstatus2_cn68xx;
+
+	struct {
+		u64 tailc : 1;
+		u64 reserved_60_62 : 3;
+		u64 index : 12;
+		u64 reserved_46_47 : 2;
+		u64 grp : 10;
+		u64 head : 1;
+		u64 tail : 1;
+		u64 tt : 2;
+		u64 tag : 32;
+	} s_sso_ppx_tag_cn78xx;
+	/**
+	 * Result for a POW Status Load (when get_cur==1, get_wqp==0, and get_rev==1)
+	 */
+	struct {
+		u64 reserved_62_63 : 2;
+		u64 revlink_index : 11;
+		u64 index : 11;
+		u64 grp : 4;
+		u64 head : 1;
+		u64 tail : 1;
+		u64 tag_type : 2;
+		u64 tag : 32;
+	} s_sstatus3;
+	/**
+	 * Result for a SSO Status Load (when opcode is SL_WQP)
+	 */
+	struct {
+		u64 reserved_58_63 : 6;
+		u64 index : 11;
+		u64 reserved_46 : 1;
+		u64 grp : 6;
+		u64 reserved_38_39 : 2;
+		u64 wqp : 38;
+	} s_sstatus3_cn68xx;
+
+	struct {
+		u64 reserved_58_63 : 6;
+		u64 grp : 10;
+		u64 reserved_42_47 : 6;
+		u64 tag : 42;
+	} s_sso_ppx_wqp_cn78xx;
+	/**
+	 * Result for a POW Status Load (when get_cur==1, get_wqp==1, and get_rev==0)
+	 */
+	struct {
+		u64 reserved_62_63 : 2;
+		u64 link_index : 11;
+		u64 index : 11;
+		u64 grp : 4;
+		u64 wqp : 36;
+	} s_sstatus4;
+	/**
+	 * Result for a SSO Status Load (when opcode is SL_LINKS)
+	 */
+	struct {
+		u64 reserved_46_63 : 18;
+		u64 index : 11;
+		u64 reserved_34 : 1;
+		u64 grp : 6;
+		u64 head : 1;
+		u64 tail : 1;
+		u64 reserved_24_25 : 2;
+		u64 revlink_index : 11;
+		u64 reserved_11_12 : 2;
+		u64 link_index : 11;
+	} s_sstatus4_cn68xx;
+
+	struct {
+		u64 tailc : 1;
+		u64 reserved_60_62 : 3;
+		u64 index : 12;
+		u64 reserved_38_47 : 10;
+		u64 grp : 10;
+		u64 head : 1;
+		u64 tail : 1;
+		u64 reserved_25 : 1;
+		u64 revlink_index : 12;
+		u64 link_index_vld : 1;
+		u64 link_index : 12;
+	} s_sso_ppx_links_cn78xx;
+	/**
+	 * Result for a POW Status Load (when get_cur==1, get_wqp==1, and get_rev==1)
+	 */
+	struct {
+		u64 reserved_62_63 : 2;
+		u64 revlink_index : 11;
+		u64 index : 11;
+		u64 grp : 4;
+		u64 wqp : 36;
+	} s_sstatus5;
+	/**
+	 * Result For POW Memory Load (get_des == 0 and get_wqp == 0)
+	 */
+	struct {
+		u64 reserved_51_63 : 13;
+		u64 next_index : 11;
+		u64 grp : 4;
+		u64 reserved_35 : 1;
+		u64 tail : 1;
+		u64 tag_type : 2;
+		u64 tag : 32;
+	} s_smemload0;
+	/**
+	 * Result For SSO Memory Load (opcode is ML_TAG)
+	 */
+	struct {
+		u64 reserved_38_63 : 26;
+		u64 tail : 1;
+		u64 reserved_34_36 : 3;
+		u64 tag_type : 2;
+		u64 tag : 32;
+	} s_smemload0_cn68xx;
+
+	struct {
+		u64 reserved_39_63 : 25;
+		u64 tail : 1;
+		u64 reserved_34_36 : 3;
+		u64 tag_type : 2;
+		u64 tag : 32;
+	} s_sso_iaq_ppx_tag_cn78xx;
+	/**
+	 * Result For POW Memory Load (get_des == 0 and get_wqp == 1)
+	 */
+	struct {
+		u64 reserved_51_63 : 13;
+		u64 next_index : 11;
+		u64 grp : 4;
+		u64 wqp : 36;
+	} s_smemload1;
+	/**
+	 * Result For SSO Memory Load (opcode is ML_WQPGRP)
+	 */
+	struct {
+		u64 reserved_48_63 : 16;
+		u64 nosched : 1;
+		u64 reserved_46 : 1;
+		u64 grp : 6;
+		u64 reserved_38_39 : 2;
+		u64 wqp : 38;
+	} s_smemload1_cn68xx;
+
+	/**
+	 * Entry structures for the CN7XXX chips.
+	 */
+	struct {
+		u64 reserved_39_63 : 25;
+		u64 tailc : 1;
+		u64 tail : 1;
+		u64 reserved_34_36 : 3;
+		u64 tt : 2;
+		u64 tag : 32;
+	} s_sso_ientx_tag_cn78xx;
+
+	struct {
+		u64 reserved_62_63 : 2;
+		u64 head : 1;
+		u64 nosched : 1;
+		u64 reserved_56_59 : 4;
+		u64 grp : 8;
+		u64 reserved_42_47 : 6;
+		u64 wqp : 42;
+	} s_sso_ientx_wqpgrp_cn73xx;
+
+	struct {
+		u64 reserved_62_63 : 2;
+		u64 head : 1;
+		u64 nosched : 1;
+		u64 reserved_58_59 : 2;
+		u64 grp : 10;
+		u64 reserved_42_47 : 6;
+		u64 wqp : 42;
+	} s_sso_ientx_wqpgrp_cn78xx;
+
+	struct {
+		u64 reserved_38_63 : 26;
+		u64 pend_switch : 1;
+		u64 reserved_34_36 : 3;
+		u64 pend_tt : 2;
+		u64 pend_tag : 32;
+	} s_sso_ientx_pendtag_cn78xx;
+
+	struct {
+		u64 reserved_26_63 : 38;
+		u64 prev_index : 10;
+		u64 reserved_11_15 : 5;
+		u64 next_index_vld : 1;
+		u64 next_index : 10;
+	} s_sso_ientx_links_cn73xx;
+
+	struct {
+		u64 reserved_28_63 : 36;
+		u64 prev_index : 12;
+		u64 reserved_13_15 : 3;
+		u64 next_index_vld : 1;
+		u64 next_index : 12;
+	} s_sso_ientx_links_cn78xx;
+
+	/**
+	 * Result For POW Memory Load (get_des == 1)
+	 */
+	struct {
+		u64 reserved_51_63 : 13;
+		u64 fwd_index : 11;
+		u64 grp : 4;
+		u64 nosched : 1;
+		u64 pend_switch : 1;
+		u64 pend_type : 2;
+		u64 pend_tag : 32;
+	} s_smemload2;
+	/**
+	 * Result For SSO Memory Load (opcode is ML_PENTAG)
+	 */
+	struct {
+		u64 reserved_38_63 : 26;
+		u64 pend_switch : 1;
+		u64 reserved_34_36 : 3;
+		u64 pend_type : 2;
+		u64 pend_tag : 32;
+	} s_smemload2_cn68xx;
+
+	struct {
+		u64 pend_switch : 1;
+		u64 pend_get_work : 1;
+		u64 pend_get_work_wait : 1;
+		u64 pend_nosched : 1;
+		u64 pend_nosched_clr : 1;
+		u64 pend_desched : 1;
+		u64 pend_alloc_we : 1;
+		u64 reserved_34_56 : 23;
+		u64 pend_tt : 2;
+		u64 pend_tag : 32;
+	} s_sso_ppx_pendtag_cn78xx;
+	/**
+	 * Result For SSO Memory Load (opcode is ML_LINKS)
+	 */
+	struct {
+		u64 reserved_24_63 : 40;
+		u64 fwd_index : 11;
+		u64 reserved_11_12 : 2;
+		u64 next_index : 11;
+	} s_smemload3_cn68xx;
+
+	/**
+	 * Result For POW Index/Pointer Load (get_rmt == 0/get_des_get_tail == 0)
+	 */
+	struct {
+		u64 reserved_52_63 : 12;
+		u64 free_val : 1;
+		u64 free_one : 1;
+		u64 reserved_49 : 1;
+		u64 free_head : 11;
+		u64 reserved_37 : 1;
+		u64 free_tail : 11;
+		u64 loc_val : 1;
+		u64 loc_one : 1;
+		u64 reserved_23 : 1;
+		u64 loc_head : 11;
+		u64 reserved_11 : 1;
+		u64 loc_tail : 11;
+	} sindexload0;
+	/**
+	 * Result for SSO Index/Pointer Load(opcode ==
+	 * IPL_IQ/IPL_DESCHED/IPL_NOSCHED)
+	 */
+	struct {
+		u64 reserved_28_63 : 36;
+		u64 queue_val : 1;
+		u64 queue_one : 1;
+		u64 reserved_24_25 : 2;
+		u64 queue_head : 11;
+		u64 reserved_11_12 : 2;
+		u64 queue_tail : 11;
+	} sindexload0_cn68xx;
+	/**
+	 * Result For POW Index/Pointer Load (get_rmt == 0/get_des_get_tail == 1)
+	 */
+	struct {
+		u64 reserved_52_63 : 12;
+		u64 nosched_val : 1;
+		u64 nosched_one : 1;
+		u64 reserved_49 : 1;
+		u64 nosched_head : 11;
+		u64 reserved_37 : 1;
+		u64 nosched_tail : 11;
+		u64 des_val : 1;
+		u64 des_one : 1;
+		u64 reserved_23 : 1;
+		u64 des_head : 11;
+		u64 reserved_11 : 1;
+		u64 des_tail : 11;
+	} sindexload1;
+	/**
+	 * Result for SSO Index/Pointer Load(opcode == IPL_FREE0/IPL_FREE1/IPL_FREE2)
+	 */
+	struct {
+		u64 reserved_60_63 : 4;
+		u64 qnum_head : 2;
+		u64 qnum_tail : 2;
+		u64 reserved_28_55 : 28;
+		u64 queue_val : 1;
+		u64 queue_one : 1;
+		u64 reserved_24_25 : 2;
+		u64 queue_head : 11;
+		u64 reserved_11_12 : 2;
+		u64 queue_tail : 11;
+	} sindexload1_cn68xx;
+	/**
+	 * Result For POW Index/Pointer Load (get_rmt == 1/get_des_get_tail == 0)
+	 */
+	struct {
+		u64 reserved_39_63 : 25;
+		u64 rmt_is_head : 1;
+		u64 rmt_val : 1;
+		u64 rmt_one : 1;
+		u64 rmt_head : 36;
+	} sindexload2;
+	/**
+	 * Result For POW Index/Pointer Load (get_rmt == 1/get_des_get_tail == 1)
+	 */
+	struct {
+		u64 reserved_39_63 : 25;
+		u64 rmt_is_head : 1;
+		u64 rmt_val : 1;
+		u64 rmt_one : 1;
+		u64 rmt_tail : 36;
+	} sindexload3;
+	/**
+	 * Response to NULL_RD request loads
+	 */
+	struct {
+		u64 unused : 62;
+		u64 state : 2;
+	} s_null_rd;
+
+} cvmx_pow_tag_load_resp_t;
+
+typedef union {
+	u64 u64;
+	struct {
+		u64 reserved_57_63 : 7;
+		u64 index : 11;
+		u64 reserved_45 : 1;
+		u64 grp : 6;
+		u64 head : 1;
+		u64 tail : 1;
+		u64 reserved_34_36 : 3;
+		u64 tag_type : 2;
+		u64 tag : 32;
+	} s;
+} cvmx_pow_sl_tag_resp_t;
+
+/**
+ * This structure describes the address used for stores to the POW.
+ *  The store address is meaningful on stores to the POW.  The hardware assumes that an aligned
+ *  64-bit store was used for all these stores.
+ *  Note the assumption that the work queue entry is aligned on an 8-byte
+ *  boundary (since the low-order 3 address bits must be zero).
+ *  Note that not all fields are used by all operations.
+ *
+ *  NOTE: The following is the behavior of the pending switch bit at the PP
+ *       for POW stores (i.e. when did<7:3> == 0xc)
+ *     - did<2:0> == 0      => pending switch bit is set
+ *     - did<2:0> == 1      => no affect on the pending switch bit
+ *     - did<2:0> == 3      => pending switch bit is cleared
+ *     - did<2:0> == 7      => no affect on the pending switch bit
+ *     - did<2:0> == others => must not be used
+ *     - No other loads/stores have an affect on the pending switch bit
+ *     - The switch bus from POW can clear the pending switch bit
+ *
+ *  NOTE: did<2:0> == 2 is used by the HW for a special single-cycle ADDWQ command
+ *  that only contains the pointer). SW must never use did<2:0> == 2.
+ */
+typedef union {
+	u64 u64;
+	struct {
+		u64 mem_reg : 2;
+		u64 reserved_49_61 : 13;
+		u64 is_io : 1;
+		u64 did : 8;
+		u64 addr : 40;
+	} stag;
+} cvmx_pow_tag_store_addr_t; /* FIXME- this type is unused */
+
+/**
+ * Decode of the store data when an IOBDMA SENDSINGLE is sent to POW
+ */
+typedef union {
+	u64 u64;
+	struct {
+		u64 scraddr : 8;
+		u64 len : 8;
+		u64 did : 8;
+		u64 unused : 36;
+		u64 wait : 1;
+		u64 unused2 : 3;
+	} s;
+	struct {
+		u64 scraddr : 8;
+		u64 len : 8;
+		u64 did : 8;
+		u64 node : 4;
+		u64 unused1 : 4;
+		u64 indexed : 1;
+		u64 grouped : 1;
+		u64 rtngrp : 1;
+		u64 unused2 : 13;
+		u64 index_grp_mask : 12;
+		u64 wait : 1;
+		u64 unused3 : 3;
+	} s_cn78xx;
+} cvmx_pow_iobdma_store_t;
+
+/* CSR typedefs have been moved to cvmx-pow-defs.h */
+
+/*enum for group priority parameters which needs modification*/
+enum cvmx_sso_group_modify_mask {
+	CVMX_SSO_MODIFY_GROUP_PRIORITY = 0x01,
+	CVMX_SSO_MODIFY_GROUP_WEIGHT = 0x02,
+	CVMX_SSO_MODIFY_GROUP_AFFINITY = 0x04
+};
+
+/**
+ * @INTERNAL
+ * Return the number of SSO groups for a given SoC model
+ */
+static inline unsigned int cvmx_sso_num_xgrp(void)
+{
+	if (OCTEON_IS_MODEL(OCTEON_CN78XX))
+		return 256;
+	if (OCTEON_IS_MODEL(OCTEON_CNF75XX))
+		return 64;
+	if (OCTEON_IS_MODEL(OCTEON_CN73XX))
+		return 64;
+	printf("ERROR: %s: Unknown model\n", __func__);
+	return 0;
+}
+
+/**
+ * @INTERNAL
+ * Return the number of POW groups on current model.
+ * In case of CN78XX/CN73XX this is the number of equivalent
+ * "legacy groups" on the chip when it is used in backward
+ * compatible mode.
+ */
+static inline unsigned int cvmx_pow_num_groups(void)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE))
+		return cvmx_sso_num_xgrp() >> 3;
+	else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE))
+		return 64;
+	else
+		return 16;
+}
+
+/**
+ * @INTERNAL
+ * Return the number of mask-set registers.
+ */
+static inline unsigned int cvmx_sso_num_maskset(void)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE))
+		return 2;
+	else
+		return 1;
+}
+
+/**
+ * Get the POW tag for this core. This returns the current
+ * tag type, tag, group, and POW entry index associated with
+ * this core. Index is only valid if the tag type isn't NULL_NULL.
+ * If a tag switch is pending this routine returns the tag before
+ * the tag switch, not after.
+ *
+ * @return Current tag
+ */
+static inline cvmx_pow_tag_info_t cvmx_pow_get_current_tag(void)
+{
+	cvmx_pow_load_addr_t load_addr;
+	cvmx_pow_tag_info_t result;
+
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_sso_sl_ppx_tag_t sl_ppx_tag;
+		cvmx_xgrp_t xgrp;
+		int node, core;
+
+		CVMX_SYNCS;
+		node = cvmx_get_node_num();
+		core = cvmx_get_local_core_num();
+		sl_ppx_tag.u64 = csr_rd_node(node, CVMX_SSO_SL_PPX_TAG(core));
+		result.index = sl_ppx_tag.s.index;
+		result.tag_type = sl_ppx_tag.s.tt;
+		result.tag = sl_ppx_tag.s.tag;
+
+		/* Get native XGRP value */
+		xgrp.xgrp = sl_ppx_tag.s.grp;
+
+		/* Return legacy style group 0..15 */
+		result.grp = xgrp.group;
+	} else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE)) {
+		cvmx_pow_sl_tag_resp_t load_resp;
+
+		load_addr.u64 = 0;
+		load_addr.sstatus_cn68xx.mem_region = CVMX_IO_SEG;
+		load_addr.sstatus_cn68xx.is_io = 1;
+		load_addr.sstatus_cn68xx.did = CVMX_OCT_DID_TAG_TAG5;
+		load_addr.sstatus_cn68xx.coreid = cvmx_get_core_num();
+		load_addr.sstatus_cn68xx.opcode = 3;
+		load_resp.u64 = csr_rd(load_addr.u64);
+		result.grp = load_resp.s.grp;
+		result.index = load_resp.s.index;
+		result.tag_type = load_resp.s.tag_type;
+		result.tag = load_resp.s.tag;
+	} else {
+		cvmx_pow_tag_load_resp_t load_resp;
+
+		load_addr.u64 = 0;
+		load_addr.sstatus.mem_region = CVMX_IO_SEG;
+		load_addr.sstatus.is_io = 1;
+		load_addr.sstatus.did = CVMX_OCT_DID_TAG_TAG1;
+		load_addr.sstatus.coreid = cvmx_get_core_num();
+		load_addr.sstatus.get_cur = 1;
+		load_resp.u64 = csr_rd(load_addr.u64);
+		result.grp = load_resp.s_sstatus2.grp;
+		result.index = load_resp.s_sstatus2.index;
+		result.tag_type = load_resp.s_sstatus2.tag_type;
+		result.tag = load_resp.s_sstatus2.tag;
+	}
+	return result;
+}
+
+/**
+ * Get the POW WQE for this core. This returns the work queue
+ * entry currently associated with this core.
+ *
+ * @return WQE pointer
+ */
+static inline cvmx_wqe_t *cvmx_pow_get_current_wqp(void)
+{
+	cvmx_pow_load_addr_t load_addr;
+	cvmx_pow_tag_load_resp_t load_resp;
+
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_sso_sl_ppx_wqp_t sso_wqp;
+		int node = cvmx_get_node_num();
+		int core = cvmx_get_local_core_num();
+
+		sso_wqp.u64 = csr_rd_node(node, CVMX_SSO_SL_PPX_WQP(core));
+		if (sso_wqp.s.wqp)
+			return (cvmx_wqe_t *)cvmx_phys_to_ptr(sso_wqp.s.wqp);
+		return (cvmx_wqe_t *)0;
+	}
+	if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE)) {
+		load_addr.u64 = 0;
+		load_addr.sstatus_cn68xx.mem_region = CVMX_IO_SEG;
+		load_addr.sstatus_cn68xx.is_io = 1;
+		load_addr.sstatus_cn68xx.did = CVMX_OCT_DID_TAG_TAG5;
+		load_addr.sstatus_cn68xx.coreid = cvmx_get_core_num();
+		load_addr.sstatus_cn68xx.opcode = 4;
+		load_resp.u64 = csr_rd(load_addr.u64);
+		if (load_resp.s_sstatus3_cn68xx.wqp)
+			return (cvmx_wqe_t *)cvmx_phys_to_ptr(load_resp.s_sstatus3_cn68xx.wqp);
+		else
+			return (cvmx_wqe_t *)0;
+	} else {
+		load_addr.u64 = 0;
+		load_addr.sstatus.mem_region = CVMX_IO_SEG;
+		load_addr.sstatus.is_io = 1;
+		load_addr.sstatus.did = CVMX_OCT_DID_TAG_TAG1;
+		load_addr.sstatus.coreid = cvmx_get_core_num();
+		load_addr.sstatus.get_cur = 1;
+		load_addr.sstatus.get_wqp = 1;
+		load_resp.u64 = csr_rd(load_addr.u64);
+		return (cvmx_wqe_t *)cvmx_phys_to_ptr(load_resp.s_sstatus4.wqp);
+	}
+}
+
+/**
+ * @INTERNAL
+ * Print a warning if a tag switch is pending for this core
+ *
+ * @param function Function name checking for a pending tag switch
+ */
+static inline void __cvmx_pow_warn_if_pending_switch(const char *function)
+{
+	u64 switch_complete;
+
+	CVMX_MF_CHORD(switch_complete);
+	cvmx_warn_if(!switch_complete, "%s called with tag switch in progress\n", function);
+}
+
+/**
+ * Waits for a tag switch to complete by polling the completion bit.
+ * Note that switches to NULL complete immediately and do not need
+ * to be waited for.
+ */
+static inline void cvmx_pow_tag_sw_wait(void)
+{
+	const u64 TIMEOUT_MS = 10; /* 10ms timeout */
+	u64 switch_complete;
+	u64 start_cycle;
+
+	if (CVMX_ENABLE_POW_CHECKS)
+		start_cycle = get_timer(0);
+
+	while (1) {
+		CVMX_MF_CHORD(switch_complete);
+		if (cvmx_likely(switch_complete))
+			break;
+
+		if (CVMX_ENABLE_POW_CHECKS) {
+			if (cvmx_unlikely(get_timer(start_cycle) > TIMEOUT_MS)) {
+				debug("WARNING: %s: Tag switch is taking a long time, possible deadlock\n",
+				      __func__);
+			}
+		}
+	}
+}
+
+/**
+ * Synchronous work request.  Requests work from the POW.
+ * This function does NOT wait for previous tag switches to complete,
+ * so the caller must ensure that there is not a pending tag switch.
+ *
+ * @param wait   When set, call stalls until work becomes available, or
+ *               times out. If not set, returns immediately.
+ *
+ * @return Returns the WQE pointer from POW. Returns NULL if no work was
+ * available.
+ */
+static inline cvmx_wqe_t *cvmx_pow_work_request_sync_nocheck(cvmx_pow_wait_t wait)
+{
+	cvmx_pow_load_addr_t ptr;
+	cvmx_pow_tag_load_resp_t result;
+
+	if (CVMX_ENABLE_POW_CHECKS)
+		__cvmx_pow_warn_if_pending_switch(__func__);
+
+	ptr.u64 = 0;
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		ptr.swork_78xx.node = cvmx_get_node_num();
+		ptr.swork_78xx.mem_region = CVMX_IO_SEG;
+		ptr.swork_78xx.is_io = 1;
+		ptr.swork_78xx.did = CVMX_OCT_DID_TAG_SWTAG;
+		ptr.swork_78xx.wait = wait;
+	} else {
+		ptr.swork.mem_region = CVMX_IO_SEG;
+		ptr.swork.is_io = 1;
+		ptr.swork.did = CVMX_OCT_DID_TAG_SWTAG;
+		ptr.swork.wait = wait;
+	}
+
+	result.u64 = csr_rd(ptr.u64);
+	if (result.s_work.no_work)
+		return NULL;
+	else
+		return (cvmx_wqe_t *)cvmx_phys_to_ptr(result.s_work.addr);
+}
+
+/**
+ * Synchronous work request.  Requests work from the POW.
+ * This function waits for any previous tag switch to complete before
+ * requesting the new work.
+ *
+ * @param wait   When set, call stalls until work becomes available, or
+ *               times out. If not set, returns immediately.
+ *
+ * @return Returns the WQE pointer from POW. Returns NULL if no work was
+ * available.
+ */
+static inline cvmx_wqe_t *cvmx_pow_work_request_sync(cvmx_pow_wait_t wait)
+{
+	/* Must not have a switch pending when requesting work */
+	cvmx_pow_tag_sw_wait();
+	return (cvmx_pow_work_request_sync_nocheck(wait));
+}
+
+/**
+ * Synchronous null_rd request.  Requests a switch out of NULL_NULL POW state.
+ * This function waits for any previous tag switch to complete before
+ * requesting the null_rd.
+ *
+ * @return Returns the POW state of type cvmx_pow_tag_type_t.
+ */
+static inline cvmx_pow_tag_type_t cvmx_pow_work_request_null_rd(void)
+{
+	cvmx_pow_load_addr_t ptr;
+	cvmx_pow_tag_load_resp_t result;
+
+	/* Must not have a switch pending when requesting work */
+	cvmx_pow_tag_sw_wait();
+
+	ptr.u64 = 0;
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		ptr.swork_78xx.mem_region = CVMX_IO_SEG;
+		ptr.swork_78xx.is_io = 1;
+		ptr.swork_78xx.did = CVMX_OCT_DID_TAG_NULL_RD;
+		ptr.swork_78xx.node = cvmx_get_node_num();
+	} else {
+		ptr.snull_rd.mem_region = CVMX_IO_SEG;
+		ptr.snull_rd.is_io = 1;
+		ptr.snull_rd.did = CVMX_OCT_DID_TAG_NULL_RD;
+	}
+	result.u64 = csr_rd(ptr.u64);
+	return (cvmx_pow_tag_type_t)result.s_null_rd.state;
+}
+
+/**
+ * Asynchronous work request.
+ * Work is requested from the POW unit, and should later be checked with
+ * function cvmx_pow_work_response_async.
+ * This function does NOT wait for previous tag switches to complete,
+ * so the caller must ensure that there is not a pending tag switch.
+ *
+ * @param scr_addr Scratch memory address that response will be returned to,
+ *     which is either a valid WQE, or a response with the invalid bit set.
+ *     Byte address, must be 8 byte aligned.
+ * @param wait 1 to cause response to wait for work to become available
+ *               (or timeout)
+ *             0 to cause response to return immediately
+ */
+static inline void cvmx_pow_work_request_async_nocheck(int scr_addr, cvmx_pow_wait_t wait)
+{
+	cvmx_pow_iobdma_store_t data;
+
+	if (CVMX_ENABLE_POW_CHECKS)
+		__cvmx_pow_warn_if_pending_switch(__func__);
+
+	/* scr_addr must be 8 byte aligned */
+	data.u64 = 0;
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		data.s_cn78xx.node = cvmx_get_node_num();
+		data.s_cn78xx.scraddr = scr_addr >> 3;
+		data.s_cn78xx.len = 1;
+		data.s_cn78xx.did = CVMX_OCT_DID_TAG_SWTAG;
+		data.s_cn78xx.wait = wait;
+	} else {
+		data.s.scraddr = scr_addr >> 3;
+		data.s.len = 1;
+		data.s.did = CVMX_OCT_DID_TAG_SWTAG;
+		data.s.wait = wait;
+	}
+	cvmx_send_single(data.u64);
+}
+
+/**
+ * Asynchronous work request.
+ * Work is requested from the POW unit, and should later be checked with
+ * function cvmx_pow_work_response_async.
+ * This function waits for any previous tag switch to complete before
+ * requesting the new work.
+ *
+ * @param scr_addr Scratch memory address that response will be returned to,
+ *     which is either a valid WQE, or a response with the invalid bit set.
+ *     Byte address, must be 8 byte aligned.
+ * @param wait 1 to cause response to wait for work to become available
+ *               (or timeout)
+ *             0 to cause response to return immediately
+ */
+static inline void cvmx_pow_work_request_async(int scr_addr, cvmx_pow_wait_t wait)
+{
+	/* Must not have a switch pending when requesting work */
+	cvmx_pow_tag_sw_wait();
+	cvmx_pow_work_request_async_nocheck(scr_addr, wait);
+}
+
+/**
+ * Gets result of asynchronous work request.  Performs a IOBDMA sync
+ * to wait for the response.
+ *
+ * @param scr_addr Scratch memory address to get result from
+ *                  Byte address, must be 8 byte aligned.
+ * @return Returns the WQE from the scratch register, or NULL if no work was
+ *         available.
+ */
+static inline cvmx_wqe_t *cvmx_pow_work_response_async(int scr_addr)
+{
+	cvmx_pow_tag_load_resp_t result;
+
+	CVMX_SYNCIOBDMA;
+	result.u64 = cvmx_scratch_read64(scr_addr);
+	if (result.s_work.no_work)
+		return NULL;
+	else
+		return (cvmx_wqe_t *)cvmx_phys_to_ptr(result.s_work.addr);
+}
+
+/**
+ * Checks if a work queue entry pointer returned by a work
+ * request is valid.  It may be invalid due to no work
+ * being available or due to a timeout.
+ *
+ * @param wqe_ptr pointer to a work queue entry returned by the POW
+ *
+ * @return 0 if pointer is valid
+ *         1 if invalid (no work was returned)
+ */
+static inline u64 cvmx_pow_work_invalid(cvmx_wqe_t *wqe_ptr)
+{
+	return (!wqe_ptr); /* FIXME: improve */
+}
+
+/**
+ * Starts a tag switch to the provided tag value and tag type.  Completion for
+ * the tag switch must be checked for separately.
+ * This function does NOT update the
+ * work queue entry in dram to match tag value and type, so the application must
+ * keep track of these if they are important to the application.
+ * This tag switch command must not be used for switches to NULL, as the tag
+ * switch pending bit will be set by the switch request, but never cleared by
+ * the hardware.
+ *
+ * NOTE: This should not be used when switching from a NULL tag.  Use
+ * cvmx_pow_tag_sw_full() instead.
+ *
+ * This function does no checks, so the caller must ensure that any previous tag
+ * switch has completed.
+ *
+ * @param tag      new tag value
+ * @param tag_type new tag type (ordered or atomic)
+ */
+static inline void cvmx_pow_tag_sw_nocheck(u32 tag, cvmx_pow_tag_type_t tag_type)
+{
+	union cvmx_pow_tag_req_addr ptr;
+	cvmx_pow_tag_req_t tag_req;
+
+	if (CVMX_ENABLE_POW_CHECKS) {
+		cvmx_pow_tag_info_t current_tag;
+
+		__cvmx_pow_warn_if_pending_switch(__func__);
+		current_tag = cvmx_pow_get_current_tag();
+		cvmx_warn_if(current_tag.tag_type == CVMX_POW_TAG_TYPE_NULL_NULL,
+			     "%s called with NULL_NULL tag\n", __func__);
+		cvmx_warn_if(current_tag.tag_type == CVMX_POW_TAG_TYPE_NULL,
+			     "%s called with NULL tag\n", __func__);
+		cvmx_warn_if((current_tag.tag_type == tag_type) && (current_tag.tag == tag),
+			     "%s called to perform a tag switch to the same tag\n", __func__);
+		cvmx_warn_if(
+			tag_type == CVMX_POW_TAG_TYPE_NULL,
+			"%s called to perform a tag switch to NULL. Use cvmx_pow_tag_sw_null() instead\n",
+			__func__);
+	}
+
+	/*
+	 * Note that WQE in DRAM is not updated here, as the POW does not read
+	 * from DRAM once the WQE is in flight.  See hardware manual for
+	 * complete details.
+	 * It is the application's responsibility to keep track of the
+	 * current tag value if that is important.
+	 */
+	tag_req.u64 = 0;
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		tag_req.s_cn78xx_other.op = CVMX_POW_TAG_OP_SWTAG;
+		tag_req.s_cn78xx_other.type = tag_type;
+	} else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE)) {
+		tag_req.s_cn68xx_other.op = CVMX_POW_TAG_OP_SWTAG;
+		tag_req.s_cn68xx_other.tag = tag;
+		tag_req.s_cn68xx_other.type = tag_type;
+	} else {
+		tag_req.s_cn38xx.op = CVMX_POW_TAG_OP_SWTAG;
+		tag_req.s_cn38xx.tag = tag;
+		tag_req.s_cn38xx.type = tag_type;
+	}
+	ptr.u64 = 0;
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		ptr.s_cn78xx.mem_region = CVMX_IO_SEG;
+		ptr.s_cn78xx.is_io = 1;
+		ptr.s_cn78xx.did = CVMX_OCT_DID_TAG_SWTAG;
+		ptr.s_cn78xx.node = cvmx_get_node_num();
+		ptr.s_cn78xx.tag = tag;
+	} else {
+		ptr.s.mem_region = CVMX_IO_SEG;
+		ptr.s.is_io = 1;
+		ptr.s.did = CVMX_OCT_DID_TAG_SWTAG;
+	}
+	/* Once this store arrives at POW, it will attempt the switch
+	   software must wait for the switch to complete separately */
+	cvmx_write_io(ptr.u64, tag_req.u64);
+}
+
+/**
+ * Starts a tag switch to the provided tag value and tag type.  Completion for
+ * the tag switch must be checked for separately.
+ * This function does NOT update the
+ * work queue entry in dram to match tag value and type, so the application must
+ * keep track of these if they are important to the application.
+ * This tag switch command must not be used for switches to NULL, as the tag
+ * switch pending bit will be set by the switch request, but never cleared by
+ * the hardware.
+ *
+ * NOTE: This should not be used when switching from a NULL tag.  Use
+ * cvmx_pow_tag_sw_full() instead.
+ *
+ * This function waits for any previous tag switch to complete, and also
+ * displays an error on tag switches to NULL.
+ *
+ * @param tag      new tag value
+ * @param tag_type new tag type (ordered or atomic)
+ */
+static inline void cvmx_pow_tag_sw(u32 tag, cvmx_pow_tag_type_t tag_type)
+{
+	/*
+	 * Note that WQE in DRAM is not updated here, as the POW does not read
+	 * from DRAM once the WQE is in flight.  See hardware manual for
+	 * complete details. It is the application's responsibility to keep
+	 * track of the current tag value if that is important.
+	 */
+
+	/*
+	 * Ensure that there is not a pending tag switch, as a tag switch
+	 * cannot be started if a previous switch is still pending.
+	 */
+	cvmx_pow_tag_sw_wait();
+	cvmx_pow_tag_sw_nocheck(tag, tag_type);
+}
+
+/**
+ * Starts a tag switch to the provided tag value and tag type.  Completion for
+ * the tag switch must be checked for separately.
+ * This function does NOT update the
+ * work queue entry in dram to match tag value and type, so the application must
+ * keep track of these if they are important to the application.
+ * This tag switch command must not be used for switches to NULL, as the tag
+ * switch pending bit will be set by the switch request, but never cleared by
+ * the hardware.
+ *
+ * This function must be used for tag switches from NULL.
+ *
+ * This function does no checks, so the caller must ensure that any previous tag
+ * switch has completed.
+ *
+ * @param wqp      pointer to work queue entry to submit.  This entry is
+ *                 updated to match the other parameters
+ * @param tag      tag value to be assigned to work queue entry
+ * @param tag_type type of tag
+ * @param group    group value for the work queue entry.
+ */
+static inline void cvmx_pow_tag_sw_full_nocheck(cvmx_wqe_t *wqp, u32 tag,
+						cvmx_pow_tag_type_t tag_type, u64 group)
+{
+	union cvmx_pow_tag_req_addr ptr;
+	cvmx_pow_tag_req_t tag_req;
+	unsigned int node = cvmx_get_node_num();
+	u64 wqp_phys = cvmx_ptr_to_phys(wqp);
+
+	if (CVMX_ENABLE_POW_CHECKS) {
+		cvmx_pow_tag_info_t current_tag;
+
+		__cvmx_pow_warn_if_pending_switch(__func__);
+		current_tag = cvmx_pow_get_current_tag();
+		cvmx_warn_if(current_tag.tag_type == CVMX_POW_TAG_TYPE_NULL_NULL,
+			     "%s called with NULL_NULL tag\n", __func__);
+		cvmx_warn_if((current_tag.tag_type == tag_type) && (current_tag.tag == tag),
+			     "%s called to perform a tag switch to the same tag\n", __func__);
+		cvmx_warn_if(
+			tag_type == CVMX_POW_TAG_TYPE_NULL,
+			"%s called to perform a tag switch to NULL. Use cvmx_pow_tag_sw_null() instead\n",
+			__func__);
+		if ((wqp != cvmx_phys_to_ptr(0x80)) && cvmx_pow_get_current_wqp())
+			cvmx_warn_if(wqp != cvmx_pow_get_current_wqp(),
+				     "%s passed WQE(%p) doesn't match the address in the POW(%p)\n",
+				     __func__, wqp, cvmx_pow_get_current_wqp());
+	}
+
+	/*
+	 * Note that WQE in DRAM is not updated here, as the POW does not
+	 * read from DRAM once the WQE is in flight.  See hardware manual
+	 * for complete details. It is the application's responsibility to
+	 * keep track of the current tag value if that is important.
+	 */
+	tag_req.u64 = 0;
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		unsigned int xgrp;
+
+		if (wqp_phys != 0x80) {
+			/* If WQE is valid, use its XGRP:
+			 * WQE GRP is 10 bits, and is mapped
+			 * to legacy GRP + QoS, includes node number.
+			 */
+			xgrp = wqp->word1.cn78xx.grp;
+			/* Use XGRP[node] too */
+			node = xgrp >> 8;
+			/* Modify XGRP with legacy group # from arg */
+			xgrp &= ~0xf8;
+			xgrp |= 0xf8 & (group << 3);
+
+		} else {
+			/* If no WQE, build XGRP with QoS=0 and current node */
+			xgrp = group << 3;
+			xgrp |= node << 8;
+		}
+		tag_req.s_cn78xx_other.op = CVMX_POW_TAG_OP_SWTAG_FULL;
+		tag_req.s_cn78xx_other.type = tag_type;
+		tag_req.s_cn78xx_other.grp = xgrp;
+		tag_req.s_cn78xx_other.wqp = wqp_phys;
+	} else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE)) {
+		tag_req.s_cn68xx_other.op = CVMX_POW_TAG_OP_SWTAG_FULL;
+		tag_req.s_cn68xx_other.tag = tag;
+		tag_req.s_cn68xx_other.type = tag_type;
+		tag_req.s_cn68xx_other.grp = group;
+	} else {
+		tag_req.s_cn38xx.op = CVMX_POW_TAG_OP_SWTAG_FULL;
+		tag_req.s_cn38xx.tag = tag;
+		tag_req.s_cn38xx.type = tag_type;
+		tag_req.s_cn38xx.grp = group;
+	}
+	ptr.u64 = 0;
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		ptr.s_cn78xx.mem_region = CVMX_IO_SEG;
+		ptr.s_cn78xx.is_io = 1;
+		ptr.s_cn78xx.did = CVMX_OCT_DID_TAG_SWTAG;
+		ptr.s_cn78xx.node = node;
+		ptr.s_cn78xx.tag = tag;
+	} else {
+		ptr.s.mem_region = CVMX_IO_SEG;
+		ptr.s.is_io = 1;
+		ptr.s.did = CVMX_OCT_DID_TAG_SWTAG;
+		ptr.s.addr = wqp_phys;
+	}
+	/* Once this store arrives at POW, it will attempt the switch
+	   software must wait for the switch to complete separately */
+	cvmx_write_io(ptr.u64, tag_req.u64);
+}
+
+/**
+ * Starts a tag switch to the provided tag value and tag type.
+ * Completion for the tag switch must be checked for separately.
+ * This function does NOT update the work queue entry in dram to match tag value
+ * and type, so the application must keep track of these if they are important
+ * to the application. This tag switch command must not be used for switches
+ * to NULL, as the tag switch pending bit will be set by the switch request,
+ * but never cleared by the hardware.
+ *
+ * This function must be used for tag switches from NULL.
+ *
+ * This function waits for any pending tag switches to complete
+ * before requesting the tag switch.
+ *
+ * @param wqp      Pointer to work queue entry to submit.
+ *     This entry is updated to match the other parameters
+ * @param tag      Tag value to be assigned to work queue entry
+ * @param tag_type Type of tag
+ * @param group    Group value for the work queue entry.
+ */
+static inline void cvmx_pow_tag_sw_full(cvmx_wqe_t *wqp, u32 tag, cvmx_pow_tag_type_t tag_type,
+					u64 group)
+{
+	/*
+	 * Ensure that there is not a pending tag switch, as a tag switch cannot
+	 * be started if a previous switch is still pending.
+	 */
+	cvmx_pow_tag_sw_wait();
+	cvmx_pow_tag_sw_full_nocheck(wqp, tag, tag_type, group);
+}
+
+/**
+ * Switch to a NULL tag, which ends any ordering or
+ * synchronization provided by the POW for the current
+ * work queue entry.  This operation completes immediately,
+ * so completion should not be waited for.
+ * This function does NOT wait for previous tag switches to complete,
+ * so the caller must ensure that any previous tag switches have completed.
+ */
+static inline void cvmx_pow_tag_sw_null_nocheck(void)
+{
+	union cvmx_pow_tag_req_addr ptr;
+	cvmx_pow_tag_req_t tag_req;
+
+	if (CVMX_ENABLE_POW_CHECKS) {
+		cvmx_pow_tag_info_t current_tag;
+
+		__cvmx_pow_warn_if_pending_switch(__func__);
+		current_tag = cvmx_pow_get_current_tag();
+		cvmx_warn_if(current_tag.tag_type == CVMX_POW_TAG_TYPE_NULL_NULL,
+			     "%s called with NULL_NULL tag\n", __func__);
+		cvmx_warn_if(current_tag.tag_type == CVMX_POW_TAG_TYPE_NULL,
+			     "%s called when we already have a NULL tag\n", __func__);
+	}
+	tag_req.u64 = 0;
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		tag_req.s_cn78xx_other.op = CVMX_POW_TAG_OP_SWTAG;
+		tag_req.s_cn78xx_other.type = CVMX_POW_TAG_TYPE_NULL;
+	} else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE)) {
+		tag_req.s_cn68xx_other.op = CVMX_POW_TAG_OP_SWTAG;
+		tag_req.s_cn68xx_other.type = CVMX_POW_TAG_TYPE_NULL;
+	} else {
+		tag_req.s_cn38xx.op = CVMX_POW_TAG_OP_SWTAG;
+		tag_req.s_cn38xx.type = CVMX_POW_TAG_TYPE_NULL;
+	}
+	ptr.u64 = 0;
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		ptr.s_cn78xx.mem_region = CVMX_IO_SEG;
+		ptr.s_cn78xx.is_io = 1;
+		ptr.s_cn78xx.did = CVMX_OCT_DID_TAG_TAG1;
+		ptr.s_cn78xx.node = cvmx_get_node_num();
+	} else {
+		ptr.s.mem_region = CVMX_IO_SEG;
+		ptr.s.is_io = 1;
+		ptr.s.did = CVMX_OCT_DID_TAG_TAG1;
+	}
+	cvmx_write_io(ptr.u64, tag_req.u64);
+}
+
+/**
+ * Switch to a NULL tag, which ends any ordering or
+ * synchronization provided by the POW for the current
+ * work queue entry.  This operation completes immediately,
+ * so completion should not be waited for.
+ * This function waits for any pending tag switches to complete
+ * before requesting the switch to NULL.
+ */
+static inline void cvmx_pow_tag_sw_null(void)
+{
+	/*
+	 * Ensure that there is not a pending tag switch, as a tag switch cannot
+	 * be started if a previous switch is still pending.
+	 */
+	cvmx_pow_tag_sw_wait();
+	cvmx_pow_tag_sw_null_nocheck();
+}
+
+/**
+ * Submits work to an input queue.
+ * This function updates the work queue entry in DRAM to match the arguments given.
+ * Note that the tag provided is for the work queue entry submitted, and
+ * is unrelated to the tag that the core currently holds.
+ *
+ * @param wqp      pointer to work queue entry to submit.
+ *                 This entry is updated to match the other parameters
+ * @param tag      tag value to be assigned to work queue entry
+ * @param tag_type type of tag
+ * @param qos      Input queue to add to.
+ * @param grp      group value for the work queue entry.
+ */
+static inline void cvmx_pow_work_submit(cvmx_wqe_t *wqp, u32 tag, cvmx_pow_tag_type_t tag_type,
+					u64 qos, u64 grp)
+{
+	union cvmx_pow_tag_req_addr ptr;
+	cvmx_pow_tag_req_t tag_req;
+
+	tag_req.u64 = 0;
+	ptr.u64 = 0;
+
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		unsigned int node = cvmx_get_node_num();
+		unsigned int xgrp;
+
+		xgrp = (grp & 0x1f) << 3;
+		xgrp |= (qos & 7);
+		xgrp |= 0x300 & (node << 8);
+
+		wqp->word1.cn78xx.rsvd_0 = 0;
+		wqp->word1.cn78xx.rsvd_1 = 0;
+		wqp->word1.cn78xx.tag = tag;
+		wqp->word1.cn78xx.tag_type = tag_type;
+		wqp->word1.cn78xx.grp = xgrp;
+		CVMX_SYNCWS;
+
+		tag_req.s_cn78xx_other.op = CVMX_POW_TAG_OP_ADDWQ;
+		tag_req.s_cn78xx_other.type = tag_type;
+		tag_req.s_cn78xx_other.wqp = cvmx_ptr_to_phys(wqp);
+		tag_req.s_cn78xx_other.grp = xgrp;
+
+		ptr.s_cn78xx.did = 0x66; // CVMX_OCT_DID_TAG_TAG6;
+		ptr.s_cn78xx.mem_region = CVMX_IO_SEG;
+		ptr.s_cn78xx.is_io = 1;
+		ptr.s_cn78xx.node = node;
+		ptr.s_cn78xx.tag = tag;
+	} else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE)) {
+		/* Reset all reserved bits */
+		wqp->word1.cn68xx.zero_0 = 0;
+		wqp->word1.cn68xx.zero_1 = 0;
+		wqp->word1.cn68xx.zero_2 = 0;
+		wqp->word1.cn68xx.qos = qos;
+		wqp->word1.cn68xx.grp = grp;
+
+		wqp->word1.tag = tag;
+		wqp->word1.tag_type = tag_type;
+
+		tag_req.s_cn68xx_add.op = CVMX_POW_TAG_OP_ADDWQ;
+		tag_req.s_cn68xx_add.type = tag_type;
+		tag_req.s_cn68xx_add.tag = tag;
+		tag_req.s_cn68xx_add.qos = qos;
+		tag_req.s_cn68xx_add.grp = grp;
+
+		ptr.s.mem_region = CVMX_IO_SEG;
+		ptr.s.is_io = 1;
+		ptr.s.did = CVMX_OCT_DID_TAG_TAG1;
+		ptr.s.addr = cvmx_ptr_to_phys(wqp);
+	} else {
+		/* Reset all reserved bits */
+		wqp->word1.cn38xx.zero_2 = 0;
+		wqp->word1.cn38xx.qos = qos;
+		wqp->word1.cn38xx.grp = grp;
+
+		wqp->word1.tag = tag;
+		wqp->word1.tag_type = tag_type;
+
+		tag_req.s_cn38xx.op = CVMX_POW_TAG_OP_ADDWQ;
+		tag_req.s_cn38xx.type = tag_type;
+		tag_req.s_cn38xx.tag = tag;
+		tag_req.s_cn38xx.qos = qos;
+		tag_req.s_cn38xx.grp = grp;
+
+		ptr.s.mem_region = CVMX_IO_SEG;
+		ptr.s.is_io = 1;
+		ptr.s.did = CVMX_OCT_DID_TAG_TAG1;
+		ptr.s.addr = cvmx_ptr_to_phys(wqp);
+	}
+	/* SYNC write to memory before the work submit.
+	 * This is necessary as POW may read values from DRAM at this time */
+	CVMX_SYNCWS;
+	cvmx_write_io(ptr.u64, tag_req.u64);
+}
+
+/**
+ * This function sets the group mask for a core.  The group mask
+ * indicates which groups each core will accept work from. There are
+ * 16 groups.
+ *
+ * @param core_num   core to apply mask to
+ * @param mask   Group mask, one bit for up to 64 groups.
+ *               Each 1 bit in the mask enables the core to accept work from
+ *               the corresponding group.
+ *               The CN68XX supports 64 groups, earlier models only support
+ *               16 groups.
+ *
+ * The CN78XX in backwards compatibility mode allows up to 32 groups,
+ * so the 'mask' argument has one bit for every of the legacy
+ * groups, and a '1' in the mask causes a total of 8 groups
+ * which share the legacy group numbher and 8 qos levels,
+ * to be enabled for the calling processor core.
+ * A '0' in the mask will disable the current core
+ * from receiving work from the associated group.
+ */
+static inline void cvmx_pow_set_group_mask(u64 core_num, u64 mask)
+{
+	u64 valid_mask;
+	int num_groups = cvmx_pow_num_groups();
+
+	if (num_groups >= 64)
+		valid_mask = ~0ull;
+	else
+		valid_mask = (1ull << num_groups) - 1;
+
+	if ((mask & valid_mask) == 0) {
+		printf("ERROR: %s empty group mask disables work on core# %llu, ignored.\n",
+		       __func__, (unsigned long long)core_num);
+		return;
+	}
+	cvmx_warn_if(mask & (~valid_mask), "%s group number range exceeded: %#llx\n", __func__,
+		     (unsigned long long)mask);
+
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		unsigned int mask_set;
+		cvmx_sso_ppx_sx_grpmskx_t grp_msk;
+		unsigned int core, node;
+		unsigned int rix;  /* Register index */
+		unsigned int grp;  /* Legacy group # */
+		unsigned int bit;  /* bit index */
+		unsigned int xgrp; /* native group # */
+
+		node = cvmx_coremask_core_to_node(core_num);
+		core = cvmx_coremask_core_on_node(core_num);
+
+		/* 78xx: 256 groups divided into 4 X 64 bit registers */
+		/* 73xx: 64 groups are in one register */
+		for (rix = 0; rix < (cvmx_sso_num_xgrp() >> 6); rix++) {
+			grp_msk.u64 = 0;
+			for (bit = 0; bit < 64; bit++) {
+				/* 8-bit native XGRP number */
+				xgrp = (rix << 6) | bit;
+				/* Legacy 5-bit group number */
+				grp = (xgrp >> 3) & 0x1f;
+				/* Inspect legacy mask by legacy group */
+				if (mask & (1ull << grp))
+					grp_msk.s.grp_msk |= 1ull << bit;
+				/* Pre-set to all 0's */
+			}
+			for (mask_set = 0; mask_set < cvmx_sso_num_maskset(); mask_set++) {
+				csr_wr_node(node, CVMX_SSO_PPX_SX_GRPMSKX(core, mask_set, rix),
+					    grp_msk.u64);
+			}
+		}
+	} else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE)) {
+		cvmx_sso_ppx_grp_msk_t grp_msk;
+
+		grp_msk.s.grp_msk = mask;
+		csr_wr(CVMX_SSO_PPX_GRP_MSK(core_num), grp_msk.u64);
+	} else {
+		cvmx_pow_pp_grp_mskx_t grp_msk;
+
+		grp_msk.u64 = csr_rd(CVMX_POW_PP_GRP_MSKX(core_num));
+		grp_msk.s.grp_msk = mask & 0xffff;
+		csr_wr(CVMX_POW_PP_GRP_MSKX(core_num), grp_msk.u64);
+	}
+}
+
+/**
+ * This function gets the group mask for a core.  The group mask
+ * indicates which groups each core will accept work from.
+ *
+ * @param core_num   core to apply mask to
+ * @return	Group mask, one bit for up to 64 groups.
+ *               Each 1 bit in the mask enables the core to accept work from
+ *               the corresponding group.
+ *               The CN68XX supports 64 groups, earlier models only support
+ *               16 groups.
+ *
+ * The CN78XX in backwards compatibility mode allows up to 32 groups,
+ * so the 'mask' argument has one bit for every of the legacy
+ * groups, and a '1' in the mask causes a total of 8 groups
+ * which share the legacy group numbher and 8 qos levels,
+ * to be enabled for the calling processor core.
+ * A '0' in the mask will disable the current core
+ * from receiving work from the associated group.
+ */
+static inline u64 cvmx_pow_get_group_mask(u64 core_num)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_sso_ppx_sx_grpmskx_t grp_msk;
+		unsigned int core, node, i;
+		int rix; /* Register index */
+		u64 mask = 0;
+
+		node = cvmx_coremask_core_to_node(core_num);
+		core = cvmx_coremask_core_on_node(core_num);
+
+		/* 78xx: 256 groups divided into 4 X 64 bit registers */
+		/* 73xx: 64 groups are in one register */
+		for (rix = (cvmx_sso_num_xgrp() >> 6) - 1; rix >= 0; rix--) {
+			/* read only mask_set=0 (both 'set' was written same) */
+			grp_msk.u64 = csr_rd_node(node, CVMX_SSO_PPX_SX_GRPMSKX(core, 0, rix));
+			/* ASSUME: (this is how mask bits got written) */
+			/* grp_mask[7:0]: all bits 0..7 are same */
+			/* grp_mask[15:8]: all bits 8..15 are same, etc */
+			/* DO: mask[7:0] = grp_mask.u64[56,48,40,32,24,16,8,0] */
+			for (i = 0; i < 8; i++)
+				mask |= (grp_msk.u64 & ((u64)1 << (i * 8))) >> (7 * i);
+			/* we collected 8 MSBs in mask[7:0], <<=8 and continue */
+			if (cvmx_likely(rix != 0))
+				mask <<= 8;
+		}
+		return mask & 0xFFFFFFFF;
+	} else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE)) {
+		cvmx_sso_ppx_grp_msk_t grp_msk;
+
+		grp_msk.u64 = csr_rd(CVMX_SSO_PPX_GRP_MSK(core_num));
+		return grp_msk.u64;
+	} else {
+		cvmx_pow_pp_grp_mskx_t grp_msk;
+
+		grp_msk.u64 = csr_rd(CVMX_POW_PP_GRP_MSKX(core_num));
+		return grp_msk.u64 & 0xffff;
+	}
+}
+
+/*
+ * Returns 0 if 78xx(73xx,75xx) is not programmed in legacy compatible mode
+ * Returns 1 if 78xx(73xx,75xx) is programmed in legacy compatible mode
+ * Returns 1 if octeon model is not 78xx(73xx,75xx)
+ */
+static inline u64 cvmx_pow_is_legacy78mode(u64 core_num)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_sso_ppx_sx_grpmskx_t grp_msk0, grp_msk1;
+		unsigned int core, node, i;
+		int rix; /* Register index */
+		u64 mask = 0;
+
+		node = cvmx_coremask_core_to_node(core_num);
+		core = cvmx_coremask_core_on_node(core_num);
+
+		/* 78xx: 256 groups divided into 4 X 64 bit registers */
+		/* 73xx: 64 groups are in one register */
+		/* 1) in order for the 78_SSO to be in legacy compatible mode
+		 * the both mask_sets should be programmed the same */
+		for (rix = (cvmx_sso_num_xgrp() >> 6) - 1; rix >= 0; rix--) {
+			/* read mask_set=0 (both 'set' was written same) */
+			grp_msk0.u64 = csr_rd_node(node, CVMX_SSO_PPX_SX_GRPMSKX(core, 0, rix));
+			grp_msk1.u64 = csr_rd_node(node, CVMX_SSO_PPX_SX_GRPMSKX(core, 1, rix));
+			if (grp_msk0.u64 != grp_msk1.u64) {
+				return 0;
+			}
+			/* (this is how mask bits should be written) */
+			/* grp_mask[7:0]: all bits 0..7 are same */
+			/* grp_mask[15:8]: all bits 8..15 are same, etc */
+			/* 2) in order for the 78_SSO to be in legacy compatible
+			 * mode above should be true (test only mask_set=0 */
+			for (i = 0; i < 8; i++) {
+				mask = (grp_msk0.u64 >> (i << 3)) & 0xFF;
+				if (!(mask == 0 || mask == 0xFF)) {
+					return 0;
+				}
+			}
+		}
+		/* if we come here, the 78_SSO is in legacy compatible mode */
+	}
+	return 1; /* the SSO/POW is in legacy (or compatible) mode */
+}
+
+/**
+ * This function sets POW static priorities for a core. Each input queue has
+ * an associated priority value.
+ *
+ * @param core_num   core to apply priorities to
+ * @param priority   Vector of 8 priorities, one per POW Input Queue (0-7).
+ *                   Highest priority is 0 and lowest is 7. A priority value
+ *                   of 0xF instructs POW to skip the Input Queue when
+ *                   scheduling to this specific core.
+ *                   NOTE: priorities should not have gaps in values, meaning
+ *                         {0,1,1,1,1,1,1,1} is a valid configuration while
+ *                         {0,2,2,2,2,2,2,2} is not.
+ */
+static inline void cvmx_pow_set_priority(u64 core_num, const u8 priority[])
+{
+	/* Detect gaps between priorities and flag error */
+	if (!octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		int i;
+		u32 prio_mask = 0;
+
+		for (i = 0; i < 8; i++)
+			if (priority[i] != 0xF)
+				prio_mask |= 1 << priority[i];
+
+		if (prio_mask ^ ((1 << cvmx_pop(prio_mask)) - 1)) {
+			debug("ERROR: POW static priorities should be contiguous (0x%llx)\n",
+			      (unsigned long long)prio_mask);
+			return;
+		}
+	}
+
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		unsigned int group;
+		unsigned int node = cvmx_get_node_num();
+		cvmx_sso_grpx_pri_t grp_pri;
+
+		/*grp_pri.s.weight = 0x3f; these will be anyway overwritten */
+		/*grp_pri.s.affinity = 0xf; by the next csr_rd_node(..), */
+
+		for (group = 0; group < cvmx_sso_num_xgrp(); group++) {
+			grp_pri.u64 = csr_rd_node(node, CVMX_SSO_GRPX_PRI(group));
+			grp_pri.s.pri = priority[group & 0x7];
+			csr_wr_node(node, CVMX_SSO_GRPX_PRI(group), grp_pri.u64);
+		}
+
+	} else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE)) {
+		cvmx_sso_ppx_qos_pri_t qos_pri;
+
+		qos_pri.u64 = csr_rd(CVMX_SSO_PPX_QOS_PRI(core_num));
+		qos_pri.s.qos0_pri = priority[0];
+		qos_pri.s.qos1_pri = priority[1];
+		qos_pri.s.qos2_pri = priority[2];
+		qos_pri.s.qos3_pri = priority[3];
+		qos_pri.s.qos4_pri = priority[4];
+		qos_pri.s.qos5_pri = priority[5];
+		qos_pri.s.qos6_pri = priority[6];
+		qos_pri.s.qos7_pri = priority[7];
+		csr_wr(CVMX_SSO_PPX_QOS_PRI(core_num), qos_pri.u64);
+	} else {
+		/* POW priorities on CN5xxx .. CN66XX */
+		cvmx_pow_pp_grp_mskx_t grp_msk;
+
+		grp_msk.u64 = csr_rd(CVMX_POW_PP_GRP_MSKX(core_num));
+		grp_msk.s.qos0_pri = priority[0];
+		grp_msk.s.qos1_pri = priority[1];
+		grp_msk.s.qos2_pri = priority[2];
+		grp_msk.s.qos3_pri = priority[3];
+		grp_msk.s.qos4_pri = priority[4];
+		grp_msk.s.qos5_pri = priority[5];
+		grp_msk.s.qos6_pri = priority[6];
+		grp_msk.s.qos7_pri = priority[7];
+
+		csr_wr(CVMX_POW_PP_GRP_MSKX(core_num), grp_msk.u64);
+	}
+}
+
+/**
+ * This function gets POW static priorities for a core. Each input queue has
+ * an associated priority value.
+ *
+ * @param[in]  core_num core to get priorities for
+ * @param[out] priority Pointer to u8[] where to return priorities
+ *			Vector of 8 priorities, one per POW Input Queue (0-7).
+ *			Highest priority is 0 and lowest is 7. A priority value
+ *			of 0xF instructs POW to skip the Input Queue when
+ *			scheduling to this specific core.
+ *                   NOTE: priorities should not have gaps in values, meaning
+ *                         {0,1,1,1,1,1,1,1} is a valid configuration while
+ *                         {0,2,2,2,2,2,2,2} is not.
+ */
+static inline void cvmx_pow_get_priority(u64 core_num, u8 priority[])
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		unsigned int group;
+		unsigned int node = cvmx_get_node_num();
+		cvmx_sso_grpx_pri_t grp_pri;
+
+		/* read priority only from the first 8 groups */
+		/* the next groups are programmed the same (periodicaly) */
+		for (group = 0; group < 8 /*cvmx_sso_num_xgrp() */; group++) {
+			grp_pri.u64 = csr_rd_node(node, CVMX_SSO_GRPX_PRI(group));
+			priority[group /* & 0x7 */] = grp_pri.s.pri;
+		}
+
+	} else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE)) {
+		cvmx_sso_ppx_qos_pri_t qos_pri;
+
+		qos_pri.u64 = csr_rd(CVMX_SSO_PPX_QOS_PRI(core_num));
+		priority[0] = qos_pri.s.qos0_pri;
+		priority[1] = qos_pri.s.qos1_pri;
+		priority[2] = qos_pri.s.qos2_pri;
+		priority[3] = qos_pri.s.qos3_pri;
+		priority[4] = qos_pri.s.qos4_pri;
+		priority[5] = qos_pri.s.qos5_pri;
+		priority[6] = qos_pri.s.qos6_pri;
+		priority[7] = qos_pri.s.qos7_pri;
+	} else {
+		/* POW priorities on CN5xxx .. CN66XX */
+		cvmx_pow_pp_grp_mskx_t grp_msk;
+
+		grp_msk.u64 = csr_rd(CVMX_POW_PP_GRP_MSKX(core_num));
+		priority[0] = grp_msk.s.qos0_pri;
+		priority[1] = grp_msk.s.qos1_pri;
+		priority[2] = grp_msk.s.qos2_pri;
+		priority[3] = grp_msk.s.qos3_pri;
+		priority[4] = grp_msk.s.qos4_pri;
+		priority[5] = grp_msk.s.qos5_pri;
+		priority[6] = grp_msk.s.qos6_pri;
+		priority[7] = grp_msk.s.qos7_pri;
+	}
+
+	/* Detect gaps between priorities and flag error - (optional) */
+	if (!octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		int i;
+		u32 prio_mask = 0;
+
+		for (i = 0; i < 8; i++)
+			if (priority[i] != 0xF)
+				prio_mask |= 1 << priority[i];
+
+		if (prio_mask ^ ((1 << cvmx_pop(prio_mask)) - 1)) {
+			debug("ERROR:%s: POW static priorities should be contiguous (0x%llx)\n",
+			      __func__, (unsigned long long)prio_mask);
+			return;
+		}
+	}
+}
+
+static inline void cvmx_sso_get_group_priority(int node, cvmx_xgrp_t xgrp, int *priority,
+					       int *weight, int *affinity)
+{
+	cvmx_sso_grpx_pri_t grp_pri;
+
+	if (!octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		debug("ERROR: %s is not supported on this chip)\n", __func__);
+		return;
+	}
+
+	grp_pri.u64 = csr_rd_node(node, CVMX_SSO_GRPX_PRI(xgrp.xgrp));
+	*affinity = grp_pri.s.affinity;
+	*priority = grp_pri.s.pri;
+	*weight = grp_pri.s.weight;
+}
+
+/**
+ * Performs a tag switch and then an immediate deschedule. This completes
+ * immediately, so completion must not be waited for.  This function does NOT
+ * update the wqe in DRAM to match arguments.
+ *
+ * This function does NOT wait for any prior tag switches to complete, so the
+ * calling code must do this.
+ *
+ * Note the following CAVEAT of the Octeon HW behavior when
+ * re-scheduling DE-SCHEDULEd items whose (next) state is
+ * ORDERED:
+ *   - If there are no switches pending at the time that the
+ *     HW executes the de-schedule, the HW will only re-schedule
+ *     the head of the FIFO associated with the given tag. This
+ *     means that in many respects, the HW treats this ORDERED
+ *     tag as an ATOMIC tag. Note that in the SWTAG_DESCH
+ *     case (to an ORDERED tag), the HW will do the switch
+ *     before the deschedule whenever it is possible to do
+ *     the switch immediately, so it may often look like
+ *     this case.
+ *   - If there is a pending switch to ORDERED at the time
+ *     the HW executes the de-schedule, the HW will perform
+ *     the switch at the time it re-schedules, and will be
+ *     able to reschedule any/all of the entries with the
+ *     same tag.
+ * Due to this behavior, the RECOMMENDATION to software is
+ * that they have a (next) state of ATOMIC when they
+ * DE-SCHEDULE. If an ORDERED tag is what was really desired,
+ * SW can choose to immediately switch to an ORDERED tag
+ * after the work (that has an ATOMIC tag) is re-scheduled.
+ * Note that since there are never any tag switches pending
+ * when the HW re-schedules, this switch can be IMMEDIATE upon
+ * the reception of the pointer during the re-schedule.
+ *
+ * @param tag      New tag value
+ * @param tag_type New tag type
+ * @param group    New group value
+ * @param no_sched Control whether this work queue entry will be rescheduled.
+ *                 - 1 : don't schedule this work
+ *                 - 0 : allow this work to be scheduled.
+ */
+static inline void cvmx_pow_tag_sw_desched_nocheck(u32 tag, cvmx_pow_tag_type_t tag_type, u64 group,
+						   u64 no_sched)
+{
+	union cvmx_pow_tag_req_addr ptr;
+	cvmx_pow_tag_req_t tag_req;
+
+	if (CVMX_ENABLE_POW_CHECKS) {
+		cvmx_pow_tag_info_t current_tag;
+
+		__cvmx_pow_warn_if_pending_switch(__func__);
+		current_tag = cvmx_pow_get_current_tag();
+		cvmx_warn_if(current_tag.tag_type == CVMX_POW_TAG_TYPE_NULL_NULL,
+			     "%s called with NULL_NULL tag\n", __func__);
+		cvmx_warn_if(current_tag.tag_type == CVMX_POW_TAG_TYPE_NULL,
+			     "%s called with NULL tag. Deschedule not allowed from NULL state\n",
+			     __func__);
+		cvmx_warn_if((current_tag.tag_type != CVMX_POW_TAG_TYPE_ATOMIC) &&
+			     (tag_type != CVMX_POW_TAG_TYPE_ATOMIC),
+			     "%s called where neither the before or after tag is ATOMIC\n",
+			     __func__);
+	}
+	tag_req.u64 = 0;
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_t *wqp = cvmx_pow_get_current_wqp();
+
+		if (!wqp) {
+			debug("ERROR: Failed to get WQE, %s\n", __func__);
+			return;
+		}
+		group &= 0x1f;
+		wqp->word1.cn78xx.tag = tag;
+		wqp->word1.cn78xx.tag_type = tag_type;
+		wqp->word1.cn78xx.grp = group << 3;
+		CVMX_SYNCWS;
+		tag_req.s_cn78xx_other.op = CVMX_POW_TAG_OP_SWTAG_DESCH;
+		tag_req.s_cn78xx_other.type = tag_type;
+		tag_req.s_cn78xx_other.grp = group << 3;
+		tag_req.s_cn78xx_other.no_sched = no_sched;
+	} else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE)) {
+		group &= 0x3f;
+		tag_req.s_cn68xx_other.op = CVMX_POW_TAG_OP_SWTAG_DESCH;
+		tag_req.s_cn68xx_other.tag = tag;
+		tag_req.s_cn68xx_other.type = tag_type;
+		tag_req.s_cn68xx_other.grp = group;
+		tag_req.s_cn68xx_other.no_sched = no_sched;
+	} else {
+		group &= 0x0f;
+		tag_req.s_cn38xx.op = CVMX_POW_TAG_OP_SWTAG_DESCH;
+		tag_req.s_cn38xx.tag = tag;
+		tag_req.s_cn38xx.type = tag_type;
+		tag_req.s_cn38xx.grp = group;
+		tag_req.s_cn38xx.no_sched = no_sched;
+	}
+	ptr.u64 = 0;
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		ptr.s.mem_region = CVMX_IO_SEG;
+		ptr.s.is_io = 1;
+		ptr.s.did = CVMX_OCT_DID_TAG_TAG3;
+		ptr.s_cn78xx.node = cvmx_get_node_num();
+		ptr.s_cn78xx.tag = tag;
+	} else {
+		ptr.s.mem_region = CVMX_IO_SEG;
+		ptr.s.is_io = 1;
+		ptr.s.did = CVMX_OCT_DID_TAG_TAG3;
+	}
+	cvmx_write_io(ptr.u64, tag_req.u64);
+}
+
+/**
+ * Performs a tag switch and then an immediate deschedule. This completes
+ * immediately, so completion must not be waited for.  This function does NOT
+ * update the wqe in DRAM to match arguments.
+ *
+ * This function waits for any prior tag switches to complete, so the
+ * calling code may call this function with a pending tag switch.
+ *
+ * Note the following CAVEAT of the Octeon HW behavior when
+ * re-scheduling DE-SCHEDULEd items whose (next) state is
+ * ORDERED:
+ *   - If there are no switches pending at the time that the
+ *     HW executes the de-schedule, the HW will only re-schedule
+ *     the head of the FIFO associated with the given tag. This
+ *     means that in many respects, the HW treats this ORDERED
+ *     tag as an ATOMIC tag. Note that in the SWTAG_DESCH
+ *     case (to an ORDERED tag), the HW will do the switch
+ *     before the deschedule whenever it is possible to do
+ *     the switch immediately, so it may often look like
+ *     this case.
+ *   - If there is a pending switch to ORDERED at the time
+ *     the HW executes the de-schedule, the HW will perform
+ *     the switch at the time it re-schedules, and will be
+ *     able to reschedule any/all of the entries with the
+ *     same tag.
+ * Due to this behavior, the RECOMMENDATION to software is
+ * that they have a (next) state of ATOMIC when they
+ * DE-SCHEDULE. If an ORDERED tag is what was really desired,
+ * SW can choose to immediately switch to an ORDERED tag
+ * after the work (that has an ATOMIC tag) is re-scheduled.
+ * Note that since there are never any tag switches pending
+ * when the HW re-schedules, this switch can be IMMEDIATE upon
+ * the reception of the pointer during the re-schedule.
+ *
+ * @param tag      New tag value
+ * @param tag_type New tag type
+ * @param group    New group value
+ * @param no_sched Control whether this work queue entry will be rescheduled.
+ *                 - 1 : don't schedule this work
+ *                 - 0 : allow this work to be scheduled.
+ */
+static inline void cvmx_pow_tag_sw_desched(u32 tag, cvmx_pow_tag_type_t tag_type, u64 group,
+					   u64 no_sched)
+{
+	/* Need to make sure any writes to the work queue entry are complete */
+	CVMX_SYNCWS;
+	/* Ensure that there is not a pending tag switch, as a tag switch cannot be started
+	 * if a previous switch is still pending.  */
+	cvmx_pow_tag_sw_wait();
+	cvmx_pow_tag_sw_desched_nocheck(tag, tag_type, group, no_sched);
+}
+
+/**
+ * Descchedules the current work queue entry.
+ *
+ * @param no_sched no schedule flag value to be set on the work queue entry.
+ *     If this is set the entry will not be rescheduled.
+ */
+static inline void cvmx_pow_desched(u64 no_sched)
+{
+	union cvmx_pow_tag_req_addr ptr;
+	cvmx_pow_tag_req_t tag_req;
+
+	if (CVMX_ENABLE_POW_CHECKS) {
+		cvmx_pow_tag_info_t current_tag;
+
+		__cvmx_pow_warn_if_pending_switch(__func__);
+		current_tag = cvmx_pow_get_current_tag();
+		cvmx_warn_if(current_tag.tag_type == CVMX_POW_TAG_TYPE_NULL_NULL,
+			     "%s called with NULL_NULL tag\n", __func__);
+		cvmx_warn_if(current_tag.tag_type == CVMX_POW_TAG_TYPE_NULL,
+			     "%s called with NULL tag. Deschedule not expected from NULL state\n",
+			     __func__);
+	}
+	/* Need to make sure any writes to the work queue entry are complete */
+	CVMX_SYNCWS;
+
+	tag_req.u64 = 0;
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		tag_req.s_cn78xx_other.op = CVMX_POW_TAG_OP_DESCH;
+		tag_req.s_cn78xx_other.no_sched = no_sched;
+	} else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE)) {
+		tag_req.s_cn68xx_other.op = CVMX_POW_TAG_OP_DESCH;
+		tag_req.s_cn68xx_other.no_sched = no_sched;
+	} else {
+		tag_req.s_cn38xx.op = CVMX_POW_TAG_OP_DESCH;
+		tag_req.s_cn38xx.no_sched = no_sched;
+	}
+	ptr.u64 = 0;
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		ptr.s_cn78xx.mem_region = CVMX_IO_SEG;
+		ptr.s_cn78xx.is_io = 1;
+		ptr.s_cn78xx.did = CVMX_OCT_DID_TAG_TAG3;
+		ptr.s_cn78xx.node = cvmx_get_node_num();
+	} else {
+		ptr.s.mem_region = CVMX_IO_SEG;
+		ptr.s.is_io = 1;
+		ptr.s.did = CVMX_OCT_DID_TAG_TAG3;
+	}
+	cvmx_write_io(ptr.u64, tag_req.u64);
+}
+
+/******************************************************************************/
+/* OCTEON3-specific functions.                                                */
+/******************************************************************************/
+/**
+ * This function sets the the affinity of group to the cores in 78xx.
+ * It sets up all the cores in core_mask to accept work from the specified group.
+ *
+ * @param xgrp	Group to accept work from, 0 - 255.
+ * @param core_mask	Mask of all the cores which will accept work from this group
+ * @param mask_set	Every core has set of 2 masks which can be set to accept work
+ *     from 256 groups. At the time of get_work, cores can choose which mask_set
+ *     to get work from. 'mask_set' values range from 0 to 3, where	each of the
+ *     two bits represents a mask set. Cores will be added to the mask set with
+ *     corresponding bit set, and removed from the mask set with corresponding
+ *     bit clear.
+ * Note: cores can only accept work from SSO groups on the same node,
+ * so the node number for the group is derived from the core number.
+ */
+static inline void cvmx_sso_set_group_core_affinity(cvmx_xgrp_t xgrp,
+						    const struct cvmx_coremask *core_mask,
+						    u8 mask_set)
+{
+	cvmx_sso_ppx_sx_grpmskx_t grp_msk;
+	int core;
+	int grp_index = xgrp.xgrp >> 6;
+	int bit_pos = xgrp.xgrp % 64;
+
+	if (!octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		debug("ERROR: %s is not supported on this chip)\n", __func__);
+		return;
+	}
+	cvmx_coremask_for_each_core(core, core_mask)
+	{
+		unsigned int node, ncore;
+		u64 reg_addr;
+
+		node = cvmx_coremask_core_to_node(core);
+		ncore = cvmx_coremask_core_on_node(core);
+
+		reg_addr = CVMX_SSO_PPX_SX_GRPMSKX(ncore, 0, grp_index);
+		grp_msk.u64 = csr_rd_node(node, reg_addr);
+
+		if (mask_set & 1)
+			grp_msk.s.grp_msk |= (1ull << bit_pos);
+		else
+			grp_msk.s.grp_msk &= ~(1ull << bit_pos);
+
+		csr_wr_node(node, reg_addr, grp_msk.u64);
+
+		reg_addr = CVMX_SSO_PPX_SX_GRPMSKX(ncore, 1, grp_index);
+		grp_msk.u64 = csr_rd_node(node, reg_addr);
+
+		if (mask_set & 2)
+			grp_msk.s.grp_msk |= (1ull << bit_pos);
+		else
+			grp_msk.s.grp_msk &= ~(1ull << bit_pos);
+
+		csr_wr_node(node, reg_addr, grp_msk.u64);
+	}
+}
+
+/**
+ * This function sets the priority and group affinity arbitration for each group.
+ *
+ * @param node		Node number
+ * @param xgrp	Group 0 - 255 to apply mask parameters to
+ * @param priority	Priority of the group relative to other groups
+ *     0x0 - highest priority
+ *     0x7 - lowest priority
+ * @param weight	Cross-group arbitration weight to apply to this group.
+ *     valid values are 1-63
+ *     h/w default is 0x3f
+ * @param affinity	Processor affinity arbitration weight to apply to this group.
+ *     If zero, affinity is disabled.
+ *     valid values are 0-15
+ *     h/w default which is 0xf.
+ * @param modify_mask   mask of the parameters which needs to be modified.
+ *     enum cvmx_sso_group_modify_mask
+ *     to modify only priority -- set bit0
+ *     to modify only weight   -- set bit1
+ *     to modify only affinity -- set bit2
+ */
+static inline void cvmx_sso_set_group_priority(int node, cvmx_xgrp_t xgrp, int priority, int weight,
+					       int affinity,
+					       enum cvmx_sso_group_modify_mask modify_mask)
+{
+	cvmx_sso_grpx_pri_t grp_pri;
+
+	if (!octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		debug("ERROR: %s is not supported on this chip)\n", __func__);
+		return;
+	}
+	if (weight <= 0)
+		weight = 0x3f; /* Force HW default when out of range */
+
+	grp_pri.u64 = csr_rd_node(node, CVMX_SSO_GRPX_PRI(xgrp.xgrp));
+	if (grp_pri.s.weight == 0)
+		grp_pri.s.weight = 0x3f;
+	if (modify_mask & CVMX_SSO_MODIFY_GROUP_PRIORITY)
+		grp_pri.s.pri = priority;
+	if (modify_mask & CVMX_SSO_MODIFY_GROUP_WEIGHT)
+		grp_pri.s.weight = weight;
+	if (modify_mask & CVMX_SSO_MODIFY_GROUP_AFFINITY)
+		grp_pri.s.affinity = affinity;
+	csr_wr_node(node, CVMX_SSO_GRPX_PRI(xgrp.xgrp), grp_pri.u64);
+}
+
+/**
+ * Asynchronous work request.
+ * Only works on CN78XX style SSO.
+ *
+ * Work is requested from the SSO unit, and should later be checked with
+ * function cvmx_pow_work_response_async.
+ * This function does NOT wait for previous tag switches to complete,
+ * so the caller must ensure that there is not a pending tag switch.
+ *
+ * @param scr_addr Scratch memory address that response will be returned to,
+ *     which is either a valid WQE, or a response with the invalid bit set.
+ *     Byte address, must be 8 byte aligned.
+ * @param xgrp  Group to receive work for (0-255).
+ * @param wait
+ *     1 to cause response to wait for work to become available (or timeout)
+ *     0 to cause response to return immediately
+ */
+static inline void cvmx_sso_work_request_grp_async_nocheck(int scr_addr, cvmx_xgrp_t xgrp,
+							   cvmx_pow_wait_t wait)
+{
+	cvmx_pow_iobdma_store_t data;
+	unsigned int node = cvmx_get_node_num();
+
+	if (CVMX_ENABLE_POW_CHECKS) {
+		__cvmx_pow_warn_if_pending_switch(__func__);
+		cvmx_warn_if(!octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE), "Not CN78XX");
+	}
+	/* scr_addr must be 8 byte aligned */
+	data.u64 = 0;
+	data.s_cn78xx.scraddr = scr_addr >> 3;
+	data.s_cn78xx.len = 1;
+	data.s_cn78xx.did = CVMX_OCT_DID_TAG_SWTAG;
+	data.s_cn78xx.grouped = 1;
+	data.s_cn78xx.index_grp_mask = (node << 8) | xgrp.xgrp;
+	data.s_cn78xx.wait = wait;
+	data.s_cn78xx.node = node;
+
+	cvmx_send_single(data.u64);
+}
+
+/**
+ * Synchronous work request from the node-local SSO without verifying
+ * pending tag switch. It requests work from a specific SSO group.
+ *
+ * @param lgrp The local group number (within the SSO of the node of the caller)
+ *     from which to get the work.
+ * @param wait When set, call stalls until work becomes available, or times out.
+ *     If not set, returns immediately.
+ *
+ * @return Returns the WQE pointer from SSO.
+ *     Returns NULL if no work was available.
+ */
+static inline void *cvmx_sso_work_request_grp_sync_nocheck(unsigned int lgrp, cvmx_pow_wait_t wait)
+{
+	cvmx_pow_load_addr_t ptr;
+	cvmx_pow_tag_load_resp_t result;
+	unsigned int node = cvmx_get_node_num() & 3;
+
+	if (CVMX_ENABLE_POW_CHECKS) {
+		__cvmx_pow_warn_if_pending_switch(__func__);
+		cvmx_warn_if(!octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE), "Not CN78XX");
+	}
+	ptr.u64 = 0;
+	ptr.swork_78xx.mem_region = CVMX_IO_SEG;
+	ptr.swork_78xx.is_io = 1;
+	ptr.swork_78xx.did = CVMX_OCT_DID_TAG_SWTAG;
+	ptr.swork_78xx.node = node;
+	ptr.swork_78xx.grouped = 1;
+	ptr.swork_78xx.index = (lgrp & 0xff) | node << 8;
+	ptr.swork_78xx.wait = wait;
+
+	result.u64 = csr_rd(ptr.u64);
+	if (result.s_work.no_work)
+		return NULL;
+	else
+		return cvmx_phys_to_ptr(result.s_work.addr);
+}
+
+/**
+ * Synchronous work request from the node-local SSO.
+ * It requests work from a specific SSO group.
+ * This function waits for any previous tag switch to complete before
+ * requesting the new work.
+ *
+ * @param lgrp The node-local group number from which to get the work.
+ * @param wait When set, call stalls until work becomes available, or times out.
+ *     If not set, returns immediately.
+ *
+ * @return The WQE pointer or NULL, if work is not available.
+ */
+static inline void *cvmx_sso_work_request_grp_sync(unsigned int lgrp, cvmx_pow_wait_t wait)
+{
+	cvmx_pow_tag_sw_wait();
+	return cvmx_sso_work_request_grp_sync_nocheck(lgrp, wait);
+}
+
+/**
+ * This function sets the group mask for a core.  The group mask bits
+ * indicate which groups each core will accept work from.
+ *
+ * @param core_num	Processor core to apply mask to.
+ * @param mask_set	7XXX has 2 sets of masks per core.
+ *     Bit 0 represents the first mask set, bit 1 -- the second.
+ * @param xgrp_mask	Group mask array.
+ *     Total number of groups is divided into a number of
+ *     64-bits mask sets. Each bit in the mask, if set, enables
+ *     the core to accept work from the corresponding group.
+ *
+ * NOTE: Each core can be configured to accept work in accordance to both
+ * mask sets, with the first having higher precedence over the second,
+ * or to accept work in accordance to just one of the two mask sets.
+ * The 'core_num' argument represents a processor core on any node
+ * in a coherent multi-chip system.
+ *
+ * If the 'mask_set' argument is 3, both mask sets are configured
+ * with the same value (which is not typically the intention),
+ * so keep in mind the function needs to be called twice
+ * to set a different value into each of the mask sets,
+ * once with 'mask_set=1' and second time with 'mask_set=2'.
+ */
+static inline void cvmx_pow_set_xgrp_mask(u64 core_num, u8 mask_set, const u64 xgrp_mask[])
+{
+	unsigned int grp, node, core;
+	u64 reg_addr;
+
+	if (!octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		debug("ERROR: %s is not supported on this chip)\n", __func__);
+		return;
+	}
+
+	if (CVMX_ENABLE_POW_CHECKS) {
+		cvmx_warn_if(((mask_set < 1) || (mask_set > 3)), "Invalid mask set");
+	}
+
+	if ((mask_set < 1) || (mask_set > 3))
+		mask_set = 3;
+
+	node = cvmx_coremask_core_to_node(core_num);
+	core = cvmx_coremask_core_on_node(core_num);
+
+	for (grp = 0; grp < (cvmx_sso_num_xgrp() >> 6); grp++) {
+		if (mask_set & 1) {
+			reg_addr = CVMX_SSO_PPX_SX_GRPMSKX(core, 0, grp),
+			csr_wr_node(node, reg_addr, xgrp_mask[grp]);
+		}
+		if (mask_set & 2) {
+			reg_addr = CVMX_SSO_PPX_SX_GRPMSKX(core, 1, grp),
+			csr_wr_node(node, reg_addr, xgrp_mask[grp]);
+		}
+	}
+}
+
+/**
+ * This function gets the group mask for a core.  The group mask bits
+ * indicate which groups each core will accept work from.
+ *
+ * @param core_num	Processor core to apply mask to.
+ * @param mask_set	7XXX has 2 sets of masks per core.
+ *     Bit 0 represents the first mask set, bit 1 -- the second.
+ * @param xgrp_mask	Provide pointer to u64 mask[8] output array.
+ *     Total number of groups is divided into a number of
+ *     64-bits mask sets. Each bit in the mask represents
+ *     the core accepts work from the corresponding group.
+ *
+ * NOTE: Each core can be configured to accept work in accordance to both
+ * mask sets, with the first having higher precedence over the second,
+ * or to accept work in accordance to just one of the two mask sets.
+ * The 'core_num' argument represents a processor core on any node
+ * in a coherent multi-chip system.
+ */
+static inline void cvmx_pow_get_xgrp_mask(u64 core_num, u8 mask_set, u64 *xgrp_mask)
+{
+	cvmx_sso_ppx_sx_grpmskx_t grp_msk;
+	unsigned int grp, node, core;
+	u64 reg_addr;
+
+	if (!octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		debug("ERROR: %s is not supported on this chip)\n", __func__);
+		return;
+	}
+
+	if (CVMX_ENABLE_POW_CHECKS) {
+		cvmx_warn_if(mask_set != 1 && mask_set != 2, "Invalid mask set");
+	}
+
+	node = cvmx_coremask_core_to_node(core_num);
+	core = cvmx_coremask_core_on_node(core_num);
+
+	for (grp = 0; grp < cvmx_sso_num_xgrp() >> 6; grp++) {
+		if (mask_set & 1) {
+			reg_addr = CVMX_SSO_PPX_SX_GRPMSKX(core, 0, grp),
+			grp_msk.u64 = csr_rd_node(node, reg_addr);
+			xgrp_mask[grp] = grp_msk.s.grp_msk;
+		}
+		if (mask_set & 2) {
+			reg_addr = CVMX_SSO_PPX_SX_GRPMSKX(core, 1, grp),
+			grp_msk.u64 = csr_rd_node(node, reg_addr);
+			xgrp_mask[grp] = grp_msk.s.grp_msk;
+		}
+	}
+}
+
+/**
+ * Executes SSO SWTAG command.
+ * This is similar to cvmx_pow_tag_sw() function, but uses linear
+ * (vs. integrated group-qos) group index.
+ */
+static inline void cvmx_pow_tag_sw_node(cvmx_wqe_t *wqp, u32 tag, cvmx_pow_tag_type_t tag_type,
+					int node)
+{
+	union cvmx_pow_tag_req_addr ptr;
+	cvmx_pow_tag_req_t tag_req;
+
+	if (cvmx_unlikely(!octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE))) {
+		debug("ERROR: %s is supported on OCTEON3 only\n", __func__);
+		return;
+	}
+	CVMX_SYNCWS;
+	cvmx_pow_tag_sw_wait();
+
+	if (CVMX_ENABLE_POW_CHECKS) {
+		cvmx_pow_tag_info_t current_tag;
+
+		__cvmx_pow_warn_if_pending_switch(__func__);
+		current_tag = cvmx_pow_get_current_tag();
+		cvmx_warn_if(current_tag.tag_type == CVMX_POW_TAG_TYPE_NULL_NULL,
+			     "%s called with NULL_NULL tag\n", __func__);
+		cvmx_warn_if(current_tag.tag_type == CVMX_POW_TAG_TYPE_NULL,
+			     "%s called with NULL tag\n", __func__);
+		cvmx_warn_if((current_tag.tag_type == tag_type) && (current_tag.tag == tag),
+			     "%s called to perform a tag switch to the same tag\n", __func__);
+		cvmx_warn_if(
+			tag_type == CVMX_POW_TAG_TYPE_NULL,
+			"%s called to perform a tag switch to NULL. Use cvmx_pow_tag_sw_null() instead\n",
+			__func__);
+	}
+	wqp->word1.cn78xx.tag = tag;
+	wqp->word1.cn78xx.tag_type = tag_type;
+	CVMX_SYNCWS;
+
+	tag_req.u64 = 0;
+	tag_req.s_cn78xx_other.op = CVMX_POW_TAG_OP_SWTAG;
+	tag_req.s_cn78xx_other.type = tag_type;
+
+	ptr.u64 = 0;
+	ptr.s_cn78xx.mem_region = CVMX_IO_SEG;
+	ptr.s_cn78xx.is_io = 1;
+	ptr.s_cn78xx.did = CVMX_OCT_DID_TAG_SWTAG;
+	ptr.s_cn78xx.node = node;
+	ptr.s_cn78xx.tag = tag;
+	cvmx_write_io(ptr.u64, tag_req.u64);
+}
+
+/**
+ * Executes SSO SWTAG_FULL command.
+ * This is similar to cvmx_pow_tag_sw_full() function, but
+ * uses linear (vs. integrated group-qos) group index.
+ */
+static inline void cvmx_pow_tag_sw_full_node(cvmx_wqe_t *wqp, u32 tag, cvmx_pow_tag_type_t tag_type,
+					     u8 xgrp, int node)
+{
+	union cvmx_pow_tag_req_addr ptr;
+	cvmx_pow_tag_req_t tag_req;
+	u16 gxgrp;
+
+	if (cvmx_unlikely(!octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE))) {
+		debug("ERROR: %s is supported on OCTEON3 only\n", __func__);
+		return;
+	}
+	/* Ensure that there is not a pending tag switch, as a tag switch cannot be
+	 * started, if a previous switch is still pending. */
+	CVMX_SYNCWS;
+	cvmx_pow_tag_sw_wait();
+
+	if (CVMX_ENABLE_POW_CHECKS) {
+		cvmx_pow_tag_info_t current_tag;
+
+		__cvmx_pow_warn_if_pending_switch(__func__);
+		current_tag = cvmx_pow_get_current_tag();
+		cvmx_warn_if(current_tag.tag_type == CVMX_POW_TAG_TYPE_NULL_NULL,
+			     "%s called with NULL_NULL tag\n", __func__);
+		cvmx_warn_if((current_tag.tag_type == tag_type) && (current_tag.tag == tag),
+			     "%s called to perform a tag switch to the same tag\n", __func__);
+		cvmx_warn_if(
+			tag_type == CVMX_POW_TAG_TYPE_NULL,
+			"%s called to perform a tag switch to NULL. Use cvmx_pow_tag_sw_null() instead\n",
+			__func__);
+		if ((wqp != cvmx_phys_to_ptr(0x80)) && cvmx_pow_get_current_wqp())
+			cvmx_warn_if(wqp != cvmx_pow_get_current_wqp(),
+				     "%s passed WQE(%p) doesn't match the address in the POW(%p)\n",
+				     __func__, wqp, cvmx_pow_get_current_wqp());
+	}
+	gxgrp = node;
+	gxgrp = gxgrp << 8 | xgrp;
+	wqp->word1.cn78xx.grp = gxgrp;
+	wqp->word1.cn78xx.tag = tag;
+	wqp->word1.cn78xx.tag_type = tag_type;
+	CVMX_SYNCWS;
+
+	tag_req.u64 = 0;
+	tag_req.s_cn78xx_other.op = CVMX_POW_TAG_OP_SWTAG_FULL;
+	tag_req.s_cn78xx_other.type = tag_type;
+	tag_req.s_cn78xx_other.grp = gxgrp;
+	tag_req.s_cn78xx_other.wqp = cvmx_ptr_to_phys(wqp);
+
+	ptr.u64 = 0;
+	ptr.s_cn78xx.mem_region = CVMX_IO_SEG;
+	ptr.s_cn78xx.is_io = 1;
+	ptr.s_cn78xx.did = CVMX_OCT_DID_TAG_SWTAG;
+	ptr.s_cn78xx.node = node;
+	ptr.s_cn78xx.tag = tag;
+	cvmx_write_io(ptr.u64, tag_req.u64);
+}
+
+/**
+ * Submits work to an SSO group on any OCI node.
+ * This function updates the work queue entry in DRAM to match
+ * the arguments given.
+ * Note that the tag provided is for the work queue entry submitted,
+ * and is unrelated to the tag that the core currently holds.
+ *
+ * @param wqp pointer to work queue entry to submit.
+ * This entry is updated to match the other parameters
+ * @param tag tag value to be assigned to work queue entry
+ * @param tag_type type of tag
+ * @param xgrp native CN78XX group in the range 0..255
+ * @param node The OCI node number for the target group
+ *
+ * When this function is called on a model prior to CN78XX, which does
+ * not support OCI nodes, the 'node' argument is ignored, and the 'xgrp'
+ * parameter is converted into 'qos' (the lower 3 bits) and 'grp' (the higher
+ * 5 bits), following the backward-compatibility scheme of translating
+ * between new and old style group numbers.
+ */
+static inline void cvmx_pow_work_submit_node(cvmx_wqe_t *wqp, u32 tag, cvmx_pow_tag_type_t tag_type,
+					     u8 xgrp, u8 node)
+{
+	union cvmx_pow_tag_req_addr ptr;
+	cvmx_pow_tag_req_t tag_req;
+	u16 group;
+
+	if (cvmx_unlikely(!octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE))) {
+		debug("ERROR: %s is supported on OCTEON3 only\n", __func__);
+		return;
+	}
+	group = node;
+	group = group << 8 | xgrp;
+	wqp->word1.cn78xx.tag = tag;
+	wqp->word1.cn78xx.tag_type = tag_type;
+	wqp->word1.cn78xx.grp = group;
+	CVMX_SYNCWS;
+
+	tag_req.u64 = 0;
+	tag_req.s_cn78xx_other.op = CVMX_POW_TAG_OP_ADDWQ;
+	tag_req.s_cn78xx_other.type = tag_type;
+	tag_req.s_cn78xx_other.wqp = cvmx_ptr_to_phys(wqp);
+	tag_req.s_cn78xx_other.grp = group;
+
+	ptr.u64 = 0;
+	ptr.s_cn78xx.did = 0x66; // CVMX_OCT_DID_TAG_TAG6;
+	ptr.s_cn78xx.mem_region = CVMX_IO_SEG;
+	ptr.s_cn78xx.is_io = 1;
+	ptr.s_cn78xx.node = node;
+	ptr.s_cn78xx.tag = tag;
+
+	/* SYNC write to memory before the work submit.  This is necessary
+	 ** as POW may read values from DRAM at this time */
+	CVMX_SYNCWS;
+	cvmx_write_io(ptr.u64, tag_req.u64);
+}
+
+/**
+ * Executes the SSO SWTAG_DESCHED operation.
+ * This is similar to the cvmx_pow_tag_sw_desched() function, but
+ * uses linear (vs. unified group-qos) group index.
+ */
+static inline void cvmx_pow_tag_sw_desched_node(cvmx_wqe_t *wqe, u32 tag,
+						cvmx_pow_tag_type_t tag_type, u8 xgrp, u64 no_sched,
+						u8 node)
+{
+	union cvmx_pow_tag_req_addr ptr;
+	cvmx_pow_tag_req_t tag_req;
+	u16 group;
+
+	if (cvmx_unlikely(!octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE))) {
+		debug("ERROR: %s is supported on OCTEON3 only\n", __func__);
+		return;
+	}
+	/* Need to make sure any writes to the work queue entry are complete */
+	CVMX_SYNCWS;
+	/*
+	 * Ensure that there is not a pending tag switch, as a tag switch cannot
+	 * be started if a previous switch is still pending.
+	 */
+	cvmx_pow_tag_sw_wait();
+
+	if (CVMX_ENABLE_POW_CHECKS) {
+		cvmx_pow_tag_info_t current_tag;
+
+		__cvmx_pow_warn_if_pending_switch(__func__);
+		current_tag = cvmx_pow_get_current_tag();
+		cvmx_warn_if(current_tag.tag_type == CVMX_POW_TAG_TYPE_NULL_NULL,
+			     "%s called with NULL_NULL tag\n", __func__);
+		cvmx_warn_if(current_tag.tag_type == CVMX_POW_TAG_TYPE_NULL,
+			     "%s called with NULL tag. Deschedule not allowed from NULL state\n",
+			     __func__);
+		cvmx_warn_if((current_tag.tag_type != CVMX_POW_TAG_TYPE_ATOMIC) &&
+			     (tag_type != CVMX_POW_TAG_TYPE_ATOMIC),
+			     "%s called where neither the before or after tag is ATOMIC\n",
+			     __func__);
+	}
+	group = node;
+	group = group << 8 | xgrp;
+	wqe->word1.cn78xx.tag = tag;
+	wqe->word1.cn78xx.tag_type = tag_type;
+	wqe->word1.cn78xx.grp = group;
+	CVMX_SYNCWS;
+
+	tag_req.u64 = 0;
+	tag_req.s_cn78xx_other.op = CVMX_POW_TAG_OP_SWTAG_DESCH;
+	tag_req.s_cn78xx_other.type = tag_type;
+	tag_req.s_cn78xx_other.grp = group;
+	tag_req.s_cn78xx_other.no_sched = no_sched;
+
+	ptr.u64 = 0;
+	ptr.s.mem_region = CVMX_IO_SEG;
+	ptr.s.is_io = 1;
+	ptr.s.did = CVMX_OCT_DID_TAG_TAG3;
+	ptr.s_cn78xx.node = node;
+	ptr.s_cn78xx.tag = tag;
+	cvmx_write_io(ptr.u64, tag_req.u64);
+}
+
+/* Executes the UPD_WQP_GRP SSO operation.
+ *
+ * @param wqp  Pointer to the new work queue entry to switch to.
+ * @param xgrp SSO group in the range 0..255
+ *
+ * NOTE: The operation can be performed only on the local node.
+ */
+static inline void cvmx_sso_update_wqp_group(cvmx_wqe_t *wqp, u8 xgrp)
+{
+	union cvmx_pow_tag_req_addr addr;
+	cvmx_pow_tag_req_t data;
+	int node = cvmx_get_node_num();
+	int group = node << 8 | xgrp;
+
+	if (!octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		debug("ERROR: %s is not supported on this chip)\n", __func__);
+		return;
+	}
+	wqp->word1.cn78xx.grp = group;
+	CVMX_SYNCWS;
+
+	data.u64 = 0;
+	data.s_cn78xx_other.op = CVMX_POW_TAG_OP_UPDATE_WQP_GRP;
+	data.s_cn78xx_other.grp = group;
+	data.s_cn78xx_other.wqp = cvmx_ptr_to_phys(wqp);
+
+	addr.u64 = 0;
+	addr.s_cn78xx.mem_region = CVMX_IO_SEG;
+	addr.s_cn78xx.is_io = 1;
+	addr.s_cn78xx.did = CVMX_OCT_DID_TAG_TAG1;
+	addr.s_cn78xx.node = node;
+	cvmx_write_io(addr.u64, data.u64);
+}
+
+/******************************************************************************/
+/* Define usage of bits within the 32 bit tag values.                         */
+/******************************************************************************/
+/*
+ * Number of bits of the tag used by software.  The SW bits
+ * are always a contiguous block of the high starting at bit 31.
+ * The hardware bits are always the low bits.  By default, the top 8 bits
+ * of the tag are reserved for software, and the low 24 are set by the IPD unit.
+ */
+#define CVMX_TAG_SW_BITS  (8)
+#define CVMX_TAG_SW_SHIFT (32 - CVMX_TAG_SW_BITS)
+
+/* Below is the list of values for the top 8 bits of the tag. */
+/*
+ * Tag values with top byte of this value are reserved for internal executive
+ * uses
+ */
+#define CVMX_TAG_SW_BITS_INTERNAL 0x1
+
+/*
+ * The executive divides the remaining 24 bits as follows:
+ * the upper 8 bits (bits 23 - 16 of the tag) define a subgroup
+ * the lower 16 bits (bits 15 - 0 of the tag) define are the value with
+ * the subgroup. Note that this section describes the format of tags generated
+ * by software - refer to the hardware documentation for a description of the
+ * tags values generated by the packet input hardware.
+ * Subgroups are defined here
+ */
+
+/* Mask for the value portion of the tag */
+#define CVMX_TAG_SUBGROUP_MASK	0xFFFF
+#define CVMX_TAG_SUBGROUP_SHIFT 16
+#define CVMX_TAG_SUBGROUP_PKO	0x1
+
+/* End of executive tag subgroup definitions */
+
+/* The remaining values software bit values 0x2 - 0xff are available
+ * for application use */
+
+/**
+ * This function creates a 32 bit tag value from the two values provided.
+ *
+ * @param sw_bits The upper bits (number depends on configuration) are set
+ *     to this value.  The remainder of bits are set by the hw_bits parameter.
+ * @param hw_bits The lower bits (number depends on configuration) are set
+ *     to this value.  The remainder of bits are set by the sw_bits parameter.
+ *
+ * @return 32 bit value of the combined hw and sw bits.
+ */
+static inline u32 cvmx_pow_tag_compose(u64 sw_bits, u64 hw_bits)
+{
+	return (((sw_bits & cvmx_build_mask(CVMX_TAG_SW_BITS)) << CVMX_TAG_SW_SHIFT) |
+		(hw_bits & cvmx_build_mask(32 - CVMX_TAG_SW_BITS)));
+}
+
+/**
+ * Extracts the bits allocated for software use from the tag
+ *
+ * @param tag    32 bit tag value
+ *
+ * @return N bit software tag value, where N is configurable with
+ *     the CVMX_TAG_SW_BITS define
+ */
+static inline u32 cvmx_pow_tag_get_sw_bits(u64 tag)
+{
+	return ((tag >> (32 - CVMX_TAG_SW_BITS)) & cvmx_build_mask(CVMX_TAG_SW_BITS));
+}
+
+/**
+ *
+ * Extracts the bits allocated for hardware use from the tag
+ *
+ * @param tag    32 bit tag value
+ *
+ * @return (32 - N) bit software tag value, where N is configurable with
+ *     the CVMX_TAG_SW_BITS define
+ */
+static inline u32 cvmx_pow_tag_get_hw_bits(u64 tag)
+{
+	return (tag & cvmx_build_mask(32 - CVMX_TAG_SW_BITS));
+}
+
+static inline u64 cvmx_sso3_get_wqe_count(int node)
+{
+	cvmx_sso_grpx_aq_cnt_t aq_cnt;
+	unsigned int grp = 0;
+	u64 cnt = 0;
+
+	for (grp = 0; grp < cvmx_sso_num_xgrp(); grp++) {
+		aq_cnt.u64 = csr_rd_node(node, CVMX_SSO_GRPX_AQ_CNT(grp));
+		cnt += aq_cnt.s.aq_cnt;
+	}
+	return cnt;
+}
+
+static inline u64 cvmx_sso_get_total_wqe_count(void)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		int node = cvmx_get_node_num();
+
+		return cvmx_sso3_get_wqe_count(node);
+	} else if (OCTEON_IS_MODEL(OCTEON_CN68XX)) {
+		cvmx_sso_iq_com_cnt_t sso_iq_com_cnt;
+
+		sso_iq_com_cnt.u64 = csr_rd(CVMX_SSO_IQ_COM_CNT);
+		return (sso_iq_com_cnt.s.iq_cnt);
+	} else {
+		cvmx_pow_iq_com_cnt_t pow_iq_com_cnt;
+
+		pow_iq_com_cnt.u64 = csr_rd(CVMX_POW_IQ_COM_CNT);
+		return (pow_iq_com_cnt.s.iq_cnt);
+	}
+}
+
+/**
+ * Store the current POW internal state into the supplied
+ * buffer. It is recommended that you pass a buffer of at least
+ * 128KB. The format of the capture may change based on SDK
+ * version and Octeon chip.
+ *
+ * @param buffer Buffer to store capture into
+ * @param buffer_size The size of the supplied buffer
+ *
+ * @return Zero on success, negative on failure
+ */
+int cvmx_pow_capture(void *buffer, int buffer_size);
+
+/**
+ * Dump a POW capture to the console in a human readable format.
+ *
+ * @param buffer POW capture from cvmx_pow_capture()
+ * @param buffer_size Size of the buffer
+ */
+void cvmx_pow_display(void *buffer, int buffer_size);
+
+/**
+ * Return the number of POW entries supported by this chip
+ *
+ * @return Number of POW entries
+ */
+int cvmx_pow_get_num_entries(void);
+int cvmx_pow_get_dump_size(void);
+
+/**
+ * This will allocate count number of SSO groups on the specified node to the
+ * calling application. These groups will be for exclusive use of the
+ * application until they are freed.
+ * @param node The numa node for the allocation.
+ * @param base_group Pointer to the initial group, -1 to allocate anywhere.
+ * @param count  The number of consecutive groups to allocate.
+ * @return 0 on success and -1 on failure.
+ */
+int cvmx_sso_reserve_group_range(int node, int *base_group, int count);
+#define cvmx_sso_allocate_group_range cvmx_sso_reserve_group_range
+int cvmx_sso_reserve_group(int node);
+#define cvmx_sso_allocate_group cvmx_sso_reserve_group
+int cvmx_sso_release_group_range(int node, int base_group, int count);
+int cvmx_sso_release_group(int node, int group);
+
+/**
+ * Show integrated SSO configuration.
+ *
+ * @param node	   node number
+ */
+int cvmx_sso_config_dump(unsigned int node);
+
+/**
+ * Show integrated SSO statistics.
+ *
+ * @param node	   node number
+ */
+int cvmx_sso_stats_dump(unsigned int node);
+
+/**
+ * Clear integrated SSO statistics.
+ *
+ * @param node	   node number
+ */
+int cvmx_sso_stats_clear(unsigned int node);
+
+/**
+ * Show SSO core-group affinity and priority per node (multi-node systems)
+ */
+void cvmx_pow_mask_priority_dump_node(unsigned int node, struct cvmx_coremask *avail_coremask);
+
+/**
+ * Show POW/SSO core-group affinity and priority (legacy, single-node systems)
+ */
+static inline void cvmx_pow_mask_priority_dump(struct cvmx_coremask *avail_coremask)
+{
+	cvmx_pow_mask_priority_dump_node(0 /*node */, avail_coremask);
+}
+
+/**
+ * Show SSO performance counters (multi-node systems)
+ */
+void cvmx_pow_show_perf_counters_node(unsigned int node);
+
+/**
+ * Show POW/SSO performance counters (legacy, single-node systems)
+ */
+static inline void cvmx_pow_show_perf_counters(void)
+{
+	cvmx_pow_show_perf_counters_node(0 /*node */);
+}
+
+#endif /* __CVMX_POW_H__ */
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-qlm.h b/arch/mips/mach-octeon/include/mach/cvmx-qlm.h
new file mode 100644
index 0000000000..19915eb82c
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-qlm.h
@@ -0,0 +1,304 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ */
+
+#ifndef __CVMX_QLM_H__
+#define __CVMX_QLM_H__
+
+/*
+ * Interface 0 on the 78xx can be connected to qlm 0 or qlm 2. When interface
+ * 0 is connected to qlm 0, this macro must be set to 0. When interface 0 is
+ * connected to qlm 2, this macro must be set to 1.
+ */
+#define MUX_78XX_IFACE0 0
+
+/*
+ * Interface 1 on the 78xx can be connected to qlm 1 or qlm 3. When interface
+ * 1 is connected to qlm 1, this macro must be set to 0. When interface 1 is
+ * connected to qlm 3, this macro must be set to 1.
+ */
+#define MUX_78XX_IFACE1 0
+
+/* Uncomment this line to print QLM JTAG state */
+/* #define CVMX_QLM_DUMP_STATE 1 */
+
+typedef struct {
+	const char *name;
+	int stop_bit;
+	int start_bit;
+} __cvmx_qlm_jtag_field_t;
+
+/**
+ * Return the number of QLMs supported by the chip
+ *
+ * @return  Number of QLMs
+ */
+int cvmx_qlm_get_num(void);
+
+/**
+ * Return the qlm number based on the interface
+ *
+ * @param xiface  Interface to look
+ */
+int cvmx_qlm_interface(int xiface);
+
+/**
+ * Return the qlm number based for a port in the interface
+ *
+ * @param xiface  interface to look up
+ * @param index  index in an interface
+ *
+ * @return the qlm number based on the xiface
+ */
+int cvmx_qlm_lmac(int xiface, int index);
+
+/**
+ * Return if only DLM5/DLM6/DLM5+DLM6 is used by BGX
+ *
+ * @param BGX  BGX to search for.
+ *
+ * @return muxes used 0 = DLM5+DLM6, 1 = DLM5, 2 = DLM6.
+ */
+int cvmx_qlm_mux_interface(int bgx);
+
+/**
+ * Return number of lanes for a given qlm
+ *
+ * @param qlm QLM block to query
+ *
+ * @return  Number of lanes
+ */
+int cvmx_qlm_get_lanes(int qlm);
+
+/**
+ * Get the QLM JTAG fields based on Octeon model on the supported chips.
+ *
+ * @return  qlm_jtag_field_t structure
+ */
+const __cvmx_qlm_jtag_field_t *cvmx_qlm_jtag_get_field(void);
+
+/**
+ * Get the QLM JTAG length by going through qlm_jtag_field for each
+ * Octeon model that is supported
+ *
+ * @return return the length.
+ */
+int cvmx_qlm_jtag_get_length(void);
+
+/**
+ * Initialize the QLM layer
+ */
+void cvmx_qlm_init(void);
+
+/**
+ * Get a field in a QLM JTAG chain
+ *
+ * @param qlm    QLM to get
+ * @param lane   Lane in QLM to get
+ * @param name   String name of field
+ *
+ * @return JTAG field value
+ */
+u64 cvmx_qlm_jtag_get(int qlm, int lane, const char *name);
+
+/**
+ * Set a field in a QLM JTAG chain
+ *
+ * @param qlm    QLM to set
+ * @param lane   Lane in QLM to set, or -1 for all lanes
+ * @param name   String name of field
+ * @param value  Value of the field
+ */
+void cvmx_qlm_jtag_set(int qlm, int lane, const char *name, u64 value);
+
+/**
+ * Errata G-16094: QLM Gen2 Equalizer Default Setting Change.
+ * CN68XX pass 1.x and CN66XX pass 1.x QLM tweak. This function tweaks the
+ * JTAG setting for a QLMs to run better at 5 and 6.25Ghz.
+ */
+void __cvmx_qlm_speed_tweak(void);
+
+/**
+ * Errata G-16174: QLM Gen2 PCIe IDLE DAC change.
+ * CN68XX pass 1.x, CN66XX pass 1.x and CN63XX pass 1.0-2.2 QLM tweak.
+ * This function tweaks the JTAG setting for a QLMs for PCIe to run better.
+ */
+void __cvmx_qlm_pcie_idle_dac_tweak(void);
+
+void __cvmx_qlm_pcie_cfg_rxd_set_tweak(int qlm, int lane);
+
+/**
+ * Get the speed (Gbaud) of the QLM in Mhz.
+ *
+ * @param qlm    QLM to examine
+ *
+ * @return Speed in Mhz
+ */
+int cvmx_qlm_get_gbaud_mhz(int qlm);
+/**
+ * Get the speed (Gbaud) of the QLM in Mhz on specific node.
+ *
+ * @param node   Target QLM node
+ * @param qlm    QLM to examine
+ *
+ * @return Speed in Mhz
+ */
+int cvmx_qlm_get_gbaud_mhz_node(int node, int qlm);
+
+enum cvmx_qlm_mode {
+	CVMX_QLM_MODE_DISABLED = -1,
+	CVMX_QLM_MODE_SGMII = 1,
+	CVMX_QLM_MODE_XAUI,
+	CVMX_QLM_MODE_RXAUI,
+	CVMX_QLM_MODE_PCIE,	/* gen3 / gen2 / gen1 */
+	CVMX_QLM_MODE_PCIE_1X2, /* 1x2 gen2 / gen1 */
+	CVMX_QLM_MODE_PCIE_2X1, /* 2x1 gen2 / gen1 */
+	CVMX_QLM_MODE_PCIE_1X1, /* 1x1 gen2 / gen1 */
+	CVMX_QLM_MODE_SRIO_1X4, /* 1x4 short / long */
+	CVMX_QLM_MODE_SRIO_2X2, /* 2x2 short / long */
+	CVMX_QLM_MODE_SRIO_4X1, /* 4x1 short / long */
+	CVMX_QLM_MODE_ILK,
+	CVMX_QLM_MODE_QSGMII,
+	CVMX_QLM_MODE_SGMII_SGMII,
+	CVMX_QLM_MODE_SGMII_DISABLED,
+	CVMX_QLM_MODE_DISABLED_SGMII,
+	CVMX_QLM_MODE_SGMII_QSGMII,
+	CVMX_QLM_MODE_QSGMII_QSGMII,
+	CVMX_QLM_MODE_QSGMII_DISABLED,
+	CVMX_QLM_MODE_DISABLED_QSGMII,
+	CVMX_QLM_MODE_QSGMII_SGMII,
+	CVMX_QLM_MODE_RXAUI_1X2,
+	CVMX_QLM_MODE_SATA_2X1,
+	CVMX_QLM_MODE_XLAUI,
+	CVMX_QLM_MODE_XFI,
+	CVMX_QLM_MODE_10G_KR,
+	CVMX_QLM_MODE_40G_KR4,
+	CVMX_QLM_MODE_PCIE_1X8, /* 1x8 gen3 / gen2 / gen1 */
+	CVMX_QLM_MODE_RGMII_SGMII,
+	CVMX_QLM_MODE_RGMII_XFI,
+	CVMX_QLM_MODE_RGMII_10G_KR,
+	CVMX_QLM_MODE_RGMII_RXAUI,
+	CVMX_QLM_MODE_RGMII_XAUI,
+	CVMX_QLM_MODE_RGMII_XLAUI,
+	CVMX_QLM_MODE_RGMII_40G_KR4,
+	CVMX_QLM_MODE_MIXED,		/* BGX2 is mixed mode, DLM5(SGMII) & DLM6(XFI) */
+	CVMX_QLM_MODE_SGMII_2X1,	/* Configure BGX2 separate for DLM5 & DLM6 */
+	CVMX_QLM_MODE_10G_KR_1X2,	/* Configure BGX2 separate for DLM5 & DLM6 */
+	CVMX_QLM_MODE_XFI_1X2,		/* Configure BGX2 separate for DLM5 & DLM6 */
+	CVMX_QLM_MODE_RGMII_SGMII_1X1,	/* Configure BGX2, applies to DLM5 */
+	CVMX_QLM_MODE_RGMII_SGMII_2X1,	/* Configure BGX2, applies to DLM6 */
+	CVMX_QLM_MODE_RGMII_10G_KR_1X1, /* Configure BGX2, applies to DLM6 */
+	CVMX_QLM_MODE_RGMII_XFI_1X1,	/* Configure BGX2, applies to DLM6 */
+	CVMX_QLM_MODE_SDL,		/* RMAC Pipe */
+	CVMX_QLM_MODE_CPRI,		/* RMAC */
+	CVMX_QLM_MODE_OCI
+};
+
+enum cvmx_gmx_inf_mode {
+	CVMX_GMX_INF_MODE_DISABLED = 0,
+	CVMX_GMX_INF_MODE_SGMII = 1,  /* Other interface can be SGMII or QSGMII */
+	CVMX_GMX_INF_MODE_QSGMII = 2, /* Other interface can be SGMII or QSGMII */
+	CVMX_GMX_INF_MODE_RXAUI = 3,  /* Only interface 0, interface 1 must be DISABLED */
+};
+
+/**
+ * Eye diagram captures are stored in the following structure
+ */
+typedef struct {
+	int width;	   /* Width in the x direction (time) */
+	int height;	   /* Height in the y direction (voltage) */
+	u32 data[64][128]; /* Error count at location, saturates as max */
+} cvmx_qlm_eye_t;
+
+/**
+ * These apply to DLM1 and DLM2 if its not in SATA mode
+ * Manual refers to lanes as follows:
+ *  DML 0 lane 0 == GSER0 lane 0
+ *  DML 0 lane 1 == GSER0 lane 1
+ *  DML 1 lane 2 == GSER1 lane 0
+ *  DML 1 lane 3 == GSER1 lane 1
+ *  DML 2 lane 4 == GSER2 lane 0
+ *  DML 2 lane 5 == GSER2 lane 1
+ */
+enum cvmx_pemx_cfg_mode {
+	CVMX_PEM_MD_GEN2_2LANE = 0, /* Valid for PEM0(DLM1), PEM1(DLM2) */
+	CVMX_PEM_MD_GEN2_1LANE = 1, /* Valid for PEM0(DLM1.0), PEM1(DLM1.1,DLM2.0), PEM2(DLM2.1) */
+	CVMX_PEM_MD_GEN2_4LANE = 2, /* Valid for PEM0(DLM1-2) */
+	/* Reserved */
+	CVMX_PEM_MD_GEN1_2LANE = 4, /* Valid for PEM0(DLM1), PEM1(DLM2) */
+	CVMX_PEM_MD_GEN1_1LANE = 5, /* Valid for PEM0(DLM1.0), PEM1(DLM1.1,DLM2.0), PEM2(DLM2.1) */
+	CVMX_PEM_MD_GEN1_4LANE = 6, /* Valid for PEM0(DLM1-2) */
+	/* Reserved */
+};
+
+/*
+ * Read QLM and return mode.
+ */
+enum cvmx_qlm_mode cvmx_qlm_get_mode(int qlm);
+enum cvmx_qlm_mode cvmx_qlm_get_mode_cn78xx(int node, int qlm);
+enum cvmx_qlm_mode cvmx_qlm_get_dlm_mode(int dlm_mode, int interface);
+void __cvmx_qlm_set_mult(int qlm, int baud_mhz, int old_multiplier);
+
+void cvmx_qlm_display_registers(int qlm);
+
+int cvmx_qlm_measure_clock(int qlm);
+
+/**
+ * Measure the reference clock of a QLM on a multi-node setup
+ *
+ * @param node   node to measure
+ * @param qlm    QLM to measure
+ *
+ * @return Clock rate in Hz
+ */
+int cvmx_qlm_measure_clock_node(int node, int qlm);
+
+/*
+ * Perform RX equalization on a QLM
+ *
+ * @param node	Node the QLM is on
+ * @param qlm	QLM to perform RX equalization on
+ * @param lane	Lane to use, or -1 for all lanes
+ *
+ * @return Zero on success, negative if any lane failed RX equalization
+ */
+int __cvmx_qlm_rx_equalization(int node, int qlm, int lane);
+
+/**
+ * Errata GSER-27882 -GSER 10GBASE-KR Transmit Equalizer
+ * Training may not update PHY Tx Taps. This function is not static
+ * so we can share it with BGX KR
+ *
+ * @param node	Node to apply errata workaround
+ * @param qlm	QLM to apply errata workaround
+ * @param lane	Lane to apply the errata
+ */
+int cvmx_qlm_gser_errata_27882(int node, int qlm, int lane);
+
+void cvmx_qlm_gser_errata_25992(int node, int qlm);
+
+#ifdef CVMX_DUMP_GSER
+/**
+ * Dump GSER configuration for node 0
+ */
+int cvmx_dump_gser_config(unsigned int gser);
+/**
+ * Dump GSER status for node 0
+ */
+int cvmx_dump_gser_status(unsigned int gser);
+/**
+ * Dump GSER configuration
+ */
+int cvmx_dump_gser_config_node(unsigned int node, unsigned int gser);
+/**
+ * Dump GSER status
+ */
+int cvmx_dump_gser_status_node(unsigned int node, unsigned int gser);
+#endif
+
+int cvmx_qlm_eye_display(int node, int qlm, int qlm_lane, int format, const cvmx_qlm_eye_t *eye);
+
+void cvmx_prbs_process_cmd(int node, int qlm, int mode);
+
+#endif /* __CVMX_QLM_H__ */
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-scratch.h b/arch/mips/mach-octeon/include/mach/cvmx-scratch.h
new file mode 100644
index 0000000000..d567a8453b
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-scratch.h
@@ -0,0 +1,113 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ *
+ * This file provides support for the processor local scratch memory.
+ * Scratch memory is byte addressable - all addresses are byte addresses.
+ */
+
+#ifndef __CVMX_SCRATCH_H__
+#define __CVMX_SCRATCH_H__
+
+/* Note: This define must be a long, not a long long in order to compile
+	without warnings for both 32bit and 64bit. */
+#define CVMX_SCRATCH_BASE (-32768l) /* 0xffffffffffff8000 */
+
+/* Scratch line for LMTST/LMTDMA on Octeon3 models */
+#ifdef CVMX_CAVIUM_OCTEON3
+#define CVMX_PKO_LMTLINE 2ull
+#endif
+
+/**
+ * Reads an 8 bit value from the processor local scratchpad memory.
+ *
+ * @param address byte address to read from
+ *
+ * @return value read
+ */
+static inline u8 cvmx_scratch_read8(u64 address)
+{
+	return *CASTPTR(volatile u8, CVMX_SCRATCH_BASE + address);
+}
+
+/**
+ * Reads a 16 bit value from the processor local scratchpad memory.
+ *
+ * @param address byte address to read from
+ *
+ * @return value read
+ */
+static inline u16 cvmx_scratch_read16(u64 address)
+{
+	return *CASTPTR(volatile u16, CVMX_SCRATCH_BASE + address);
+}
+
+/**
+ * Reads a 32 bit value from the processor local scratchpad memory.
+ *
+ * @param address byte address to read from
+ *
+ * @return value read
+ */
+static inline u32 cvmx_scratch_read32(u64 address)
+{
+	return *CASTPTR(volatile u32, CVMX_SCRATCH_BASE + address);
+}
+
+/**
+ * Reads a 64 bit value from the processor local scratchpad memory.
+ *
+ * @param address byte address to read from
+ *
+ * @return value read
+ */
+static inline u64 cvmx_scratch_read64(u64 address)
+{
+	return *CASTPTR(volatile u64, CVMX_SCRATCH_BASE + address);
+}
+
+/**
+ * Writes an 8 bit value to the processor local scratchpad memory.
+ *
+ * @param address byte address to write to
+ * @param value   value to write
+ */
+static inline void cvmx_scratch_write8(u64 address, u64 value)
+{
+	*CASTPTR(volatile u8, CVMX_SCRATCH_BASE + address) = (u8)value;
+}
+
+/**
+ * Writes a 32 bit value to the processor local scratchpad memory.
+ *
+ * @param address byte address to write to
+ * @param value   value to write
+ */
+static inline void cvmx_scratch_write16(u64 address, u64 value)
+{
+	*CASTPTR(volatile u16, CVMX_SCRATCH_BASE + address) = (u16)value;
+}
+
+/**
+ * Writes a 16 bit value to the processor local scratchpad memory.
+ *
+ * @param address byte address to write to
+ * @param value   value to write
+ */
+static inline void cvmx_scratch_write32(u64 address, u64 value)
+{
+	*CASTPTR(volatile u32, CVMX_SCRATCH_BASE + address) = (u32)value;
+}
+
+/**
+ * Writes a 64 bit value to the processor local scratchpad memory.
+ *
+ * @param address byte address to write to
+ * @param value   value to write
+ */
+static inline void cvmx_scratch_write64(u64 address, u64 value)
+{
+	*CASTPTR(volatile u64, CVMX_SCRATCH_BASE + address) = value;
+}
+
+#endif /* __CVMX_SCRATCH_H__ */
diff --git a/arch/mips/mach-octeon/include/mach/cvmx-wqe.h b/arch/mips/mach-octeon/include/mach/cvmx-wqe.h
new file mode 100644
index 0000000000..c9e3c8312a
--- /dev/null
+++ b/arch/mips/mach-octeon/include/mach/cvmx-wqe.h
@@ -0,0 +1,1462 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Marvell International Ltd.
+ *
+ * This header file defines the work queue entry (wqe) data structure.
+ * Since this is a commonly used structure that depends on structures
+ * from several hardware blocks, those definitions have been placed
+ * in this file to create a single point of definition of the wqe
+ * format.
+ * Data structures are still named according to the block that they
+ * relate to.
+ */
+
+#ifndef __CVMX_WQE_H__
+#define __CVMX_WQE_H__
+
+#include "cvmx-packet.h"
+#include "cvmx-csr-enums.h"
+#include "cvmx-pki-defs.h"
+#include "cvmx-pip-defs.h"
+#include "octeon-feature.h"
+
+#define OCT_TAG_TYPE_STRING(x)						\
+	(((x) == CVMX_POW_TAG_TYPE_ORDERED) ?				\
+	 "ORDERED" :							\
+	 (((x) == CVMX_POW_TAG_TYPE_ATOMIC) ?				\
+	  "ATOMIC" :							\
+	  (((x) == CVMX_POW_TAG_TYPE_NULL) ? "NULL" : "NULL_NULL")))
+
+/* Error levels in WQE WORD2 (ERRLEV).*/
+#define PKI_ERRLEV_E__RE_M 0x0
+#define PKI_ERRLEV_E__LA_M 0x1
+#define PKI_ERRLEV_E__LB_M 0x2
+#define PKI_ERRLEV_E__LC_M 0x3
+#define PKI_ERRLEV_E__LD_M 0x4
+#define PKI_ERRLEV_E__LE_M 0x5
+#define PKI_ERRLEV_E__LF_M 0x6
+#define PKI_ERRLEV_E__LG_M 0x7
+
+enum cvmx_pki_errlevel {
+	CVMX_PKI_ERRLEV_E_RE = PKI_ERRLEV_E__RE_M,
+	CVMX_PKI_ERRLEV_E_LA = PKI_ERRLEV_E__LA_M,
+	CVMX_PKI_ERRLEV_E_LB = PKI_ERRLEV_E__LB_M,
+	CVMX_PKI_ERRLEV_E_LC = PKI_ERRLEV_E__LC_M,
+	CVMX_PKI_ERRLEV_E_LD = PKI_ERRLEV_E__LD_M,
+	CVMX_PKI_ERRLEV_E_LE = PKI_ERRLEV_E__LE_M,
+	CVMX_PKI_ERRLEV_E_LF = PKI_ERRLEV_E__LF_M,
+	CVMX_PKI_ERRLEV_E_LG = PKI_ERRLEV_E__LG_M
+};
+
+#define CVMX_PKI_ERRLEV_MAX BIT(3) /* The size of WORD2:ERRLEV field.*/
+
+/* Error code in WQE WORD2 (OPCODE).*/
+#define CVMX_PKI_OPCODE_RE_NONE	      0x0
+#define CVMX_PKI_OPCODE_RE_PARTIAL    0x1
+#define CVMX_PKI_OPCODE_RE_JABBER     0x2
+#define CVMX_PKI_OPCODE_RE_FCS	      0x7
+#define CVMX_PKI_OPCODE_RE_FCS_RCV    0x8
+#define CVMX_PKI_OPCODE_RE_TERMINATE  0x9
+#define CVMX_PKI_OPCODE_RE_RX_CTL     0xb
+#define CVMX_PKI_OPCODE_RE_SKIP	      0xc
+#define CVMX_PKI_OPCODE_RE_DMAPKT     0xf
+#define CVMX_PKI_OPCODE_RE_PKIPAR     0x13
+#define CVMX_PKI_OPCODE_RE_PKIPCAM    0x14
+#define CVMX_PKI_OPCODE_RE_MEMOUT     0x15
+#define CVMX_PKI_OPCODE_RE_BUFS_OFLOW 0x16
+#define CVMX_PKI_OPCODE_L2_FRAGMENT   0x20
+#define CVMX_PKI_OPCODE_L2_OVERRUN    0x21
+#define CVMX_PKI_OPCODE_L2_PFCS	      0x22
+#define CVMX_PKI_OPCODE_L2_PUNY	      0x23
+#define CVMX_PKI_OPCODE_L2_MAL	      0x24
+#define CVMX_PKI_OPCODE_L2_OVERSIZE   0x25
+#define CVMX_PKI_OPCODE_L2_UNDERSIZE  0x26
+#define CVMX_PKI_OPCODE_L2_LENMISM    0x27
+#define CVMX_PKI_OPCODE_IP_NOT	      0x41
+#define CVMX_PKI_OPCODE_IP_CHK	      0x42
+#define CVMX_PKI_OPCODE_IP_MAL	      0x43
+#define CVMX_PKI_OPCODE_IP_MALD	      0x44
+#define CVMX_PKI_OPCODE_IP_HOP	      0x45
+#define CVMX_PKI_OPCODE_L4_MAL	      0x61
+#define CVMX_PKI_OPCODE_L4_CHK	      0x62
+#define CVMX_PKI_OPCODE_L4_LEN	      0x63
+#define CVMX_PKI_OPCODE_L4_PORT	      0x64
+#define CVMX_PKI_OPCODE_TCP_FLAG      0x65
+
+#define CVMX_PKI_OPCODE_MAX BIT(8) /* The size of WORD2:OPCODE field.*/
+
+/* Layer types in pki */
+#define CVMX_PKI_LTYPE_E_NONE_M	      0x0
+#define CVMX_PKI_LTYPE_E_ENET_M	      0x1
+#define CVMX_PKI_LTYPE_E_VLAN_M	      0x2
+#define CVMX_PKI_LTYPE_E_SNAP_PAYLD_M 0x5
+#define CVMX_PKI_LTYPE_E_ARP_M	      0x6
+#define CVMX_PKI_LTYPE_E_RARP_M	      0x7
+#define CVMX_PKI_LTYPE_E_IP4_M	      0x8
+#define CVMX_PKI_LTYPE_E_IP4_OPT_M    0x9
+#define CVMX_PKI_LTYPE_E_IP6_M	      0xA
+#define CVMX_PKI_LTYPE_E_IP6_OPT_M    0xB
+#define CVMX_PKI_LTYPE_E_IPSEC_ESP_M  0xC
+#define CVMX_PKI_LTYPE_E_IPFRAG_M     0xD
+#define CVMX_PKI_LTYPE_E_IPCOMP_M     0xE
+#define CVMX_PKI_LTYPE_E_TCP_M	      0x10
+#define CVMX_PKI_LTYPE_E_UDP_M	      0x11
+#define CVMX_PKI_LTYPE_E_SCTP_M	      0x12
+#define CVMX_PKI_LTYPE_E_UDP_VXLAN_M  0x13
+#define CVMX_PKI_LTYPE_E_GRE_M	      0x14
+#define CVMX_PKI_LTYPE_E_NVGRE_M      0x15
+#define CVMX_PKI_LTYPE_E_GTP_M	      0x16
+#define CVMX_PKI_LTYPE_E_SW28_M	      0x1C
+#define CVMX_PKI_LTYPE_E_SW29_M	      0x1D
+#define CVMX_PKI_LTYPE_E_SW30_M	      0x1E
+#define CVMX_PKI_LTYPE_E_SW31_M	      0x1F
+
+enum cvmx_pki_layer_type {
+	CVMX_PKI_LTYPE_E_NONE = CVMX_PKI_LTYPE_E_NONE_M,
+	CVMX_PKI_LTYPE_E_ENET = CVMX_PKI_LTYPE_E_ENET_M,
+	CVMX_PKI_LTYPE_E_VLAN = CVMX_PKI_LTYPE_E_VLAN_M,
+	CVMX_PKI_LTYPE_E_SNAP_PAYLD = CVMX_PKI_LTYPE_E_SNAP_PAYLD_M,
+	CVMX_PKI_LTYPE_E_ARP = CVMX_PKI_LTYPE_E_ARP_M,
+	CVMX_PKI_LTYPE_E_RARP = CVMX_PKI_LTYPE_E_RARP_M,
+	CVMX_PKI_LTYPE_E_IP4 = CVMX_PKI_LTYPE_E_IP4_M,
+	CVMX_PKI_LTYPE_E_IP4_OPT = CVMX_PKI_LTYPE_E_IP4_OPT_M,
+	CVMX_PKI_LTYPE_E_IP6 = CVMX_PKI_LTYPE_E_IP6_M,
+	CVMX_PKI_LTYPE_E_IP6_OPT = CVMX_PKI_LTYPE_E_IP6_OPT_M,
+	CVMX_PKI_LTYPE_E_IPSEC_ESP = CVMX_PKI_LTYPE_E_IPSEC_ESP_M,
+	CVMX_PKI_LTYPE_E_IPFRAG = CVMX_PKI_LTYPE_E_IPFRAG_M,
+	CVMX_PKI_LTYPE_E_IPCOMP = CVMX_PKI_LTYPE_E_IPCOMP_M,
+	CVMX_PKI_LTYPE_E_TCP = CVMX_PKI_LTYPE_E_TCP_M,
+	CVMX_PKI_LTYPE_E_UDP = CVMX_PKI_LTYPE_E_UDP_M,
+	CVMX_PKI_LTYPE_E_SCTP = CVMX_PKI_LTYPE_E_SCTP_M,
+	CVMX_PKI_LTYPE_E_UDP_VXLAN = CVMX_PKI_LTYPE_E_UDP_VXLAN_M,
+	CVMX_PKI_LTYPE_E_GRE = CVMX_PKI_LTYPE_E_GRE_M,
+	CVMX_PKI_LTYPE_E_NVGRE = CVMX_PKI_LTYPE_E_NVGRE_M,
+	CVMX_PKI_LTYPE_E_GTP = CVMX_PKI_LTYPE_E_GTP_M,
+	CVMX_PKI_LTYPE_E_SW28 = CVMX_PKI_LTYPE_E_SW28_M,
+	CVMX_PKI_LTYPE_E_SW29 = CVMX_PKI_LTYPE_E_SW29_M,
+	CVMX_PKI_LTYPE_E_SW30 = CVMX_PKI_LTYPE_E_SW30_M,
+	CVMX_PKI_LTYPE_E_SW31 = CVMX_PKI_LTYPE_E_SW31_M,
+	CVMX_PKI_LTYPE_E_MAX = CVMX_PKI_LTYPE_E_SW31
+};
+
+typedef union {
+	u64 u64;
+	struct {
+		u64 ptr_vlan : 8;
+		u64 ptr_layer_g : 8;
+		u64 ptr_layer_f : 8;
+		u64 ptr_layer_e : 8;
+		u64 ptr_layer_d : 8;
+		u64 ptr_layer_c : 8;
+		u64 ptr_layer_b : 8;
+		u64 ptr_layer_a : 8;
+	};
+} cvmx_pki_wqe_word4_t;
+
+/**
+ * HW decode / err_code in work queue entry
+ */
+typedef union {
+	u64 u64;
+	struct {
+		u64 bufs : 8;
+		u64 ip_offset : 8;
+		u64 vlan_valid : 1;
+		u64 vlan_stacked : 1;
+		u64 unassigned : 1;
+		u64 vlan_cfi : 1;
+		u64 vlan_id : 12;
+		u64 varies : 12;
+		u64 dec_ipcomp : 1;
+		u64 tcp_or_udp : 1;
+		u64 dec_ipsec : 1;
+		u64 is_v6 : 1;
+		u64 software : 1;
+		u64 L4_error : 1;
+		u64 is_frag : 1;
+		u64 IP_exc : 1;
+		u64 is_bcast : 1;
+		u64 is_mcast : 1;
+		u64 not_IP : 1;
+		u64 rcv_error : 1;
+		u64 err_code : 8;
+	} s;
+	struct {
+		u64 bufs : 8;
+		u64 ip_offset : 8;
+		u64 vlan_valid : 1;
+		u64 vlan_stacked : 1;
+		u64 unassigned : 1;
+		u64 vlan_cfi : 1;
+		u64 vlan_id : 12;
+		u64 port : 12;
+		u64 dec_ipcomp : 1;
+		u64 tcp_or_udp : 1;
+		u64 dec_ipsec : 1;
+		u64 is_v6 : 1;
+		u64 software : 1;
+		u64 L4_error : 1;
+		u64 is_frag : 1;
+		u64 IP_exc : 1;
+		u64 is_bcast : 1;
+		u64 is_mcast : 1;
+		u64 not_IP : 1;
+		u64 rcv_error : 1;
+		u64 err_code : 8;
+	} s_cn68xx;
+	struct {
+		u64 bufs : 8;
+		u64 ip_offset : 8;
+		u64 vlan_valid : 1;
+		u64 vlan_stacked : 1;
+		u64 unassigned : 1;
+		u64 vlan_cfi : 1;
+		u64 vlan_id : 12;
+		u64 pr : 4;
+		u64 unassigned2a : 4;
+		u64 unassigned2 : 4;
+		u64 dec_ipcomp : 1;
+		u64 tcp_or_udp : 1;
+		u64 dec_ipsec : 1;
+		u64 is_v6 : 1;
+		u64 software : 1;
+		u64 L4_error : 1;
+		u64 is_frag : 1;
+		u64 IP_exc : 1;
+		u64 is_bcast : 1;
+		u64 is_mcast : 1;
+		u64 not_IP : 1;
+		u64 rcv_error : 1;
+		u64 err_code : 8;
+	} s_cn38xx;
+	struct {
+		u64 unused1 : 16;
+		u64 vlan : 16;
+		u64 unused2 : 32;
+	} svlan;
+	struct {
+		u64 bufs : 8;
+		u64 unused : 8;
+		u64 vlan_valid : 1;
+		u64 vlan_stacked : 1;
+		u64 unassigned : 1;
+		u64 vlan_cfi : 1;
+		u64 vlan_id : 12;
+		u64 varies : 12;
+		u64 unassigned2 : 4;
+		u64 software : 1;
+		u64 unassigned3 : 1;
+		u64 is_rarp : 1;
+		u64 is_arp : 1;
+		u64 is_bcast : 1;
+		u64 is_mcast : 1;
+		u64 not_IP : 1;
+		u64 rcv_error : 1;
+		u64 err_code : 8;
+	} snoip;
+	struct {
+		u64 bufs : 8;
+		u64 unused : 8;
+		u64 vlan_valid : 1;
+		u64 vlan_stacked : 1;
+		u64 unassigned : 1;
+		u64 vlan_cfi : 1;
+		u64 vlan_id : 12;
+		u64 port : 12;
+		u64 unassigned2 : 4;
+		u64 software : 1;
+		u64 unassigned3 : 1;
+		u64 is_rarp : 1;
+		u64 is_arp : 1;
+		u64 is_bcast : 1;
+		u64 is_mcast : 1;
+		u64 not_IP : 1;
+		u64 rcv_error : 1;
+		u64 err_code : 8;
+	} snoip_cn68xx;
+	struct {
+		u64 bufs : 8;
+		u64 unused : 8;
+		u64 vlan_valid : 1;
+		u64 vlan_stacked : 1;
+		u64 unassigned : 1;
+		u64 vlan_cfi : 1;
+		u64 vlan_id : 12;
+		u64 pr : 4;
+		u64 unassigned2a : 8;
+		u64 unassigned2 : 4;
+		u64 software : 1;
+		u64 unassigned3 : 1;
+		u64 is_rarp : 1;
+		u64 is_arp : 1;
+		u64 is_bcast : 1;
+		u64 is_mcast : 1;
+		u64 not_IP : 1;
+		u64 rcv_error : 1;
+		u64 err_code : 8;
+	} snoip_cn38xx;
+} cvmx_pip_wqe_word2_t;
+
+typedef union {
+	u64 u64;
+	struct {
+		u64 software : 1;
+		u64 lg_hdr_type : 5;
+		u64 lf_hdr_type : 5;
+		u64 le_hdr_type : 5;
+		u64 ld_hdr_type : 5;
+		u64 lc_hdr_type : 5;
+		u64 lb_hdr_type : 5;
+		u64 is_la_ether : 1;
+		u64 rsvd_0 : 8;
+		u64 vlan_valid : 1;
+		u64 vlan_stacked : 1;
+		u64 stat_inc : 1;
+		u64 pcam_flag4 : 1;
+		u64 pcam_flag3 : 1;
+		u64 pcam_flag2 : 1;
+		u64 pcam_flag1 : 1;
+		u64 is_frag : 1;
+		u64 is_l3_bcast : 1;
+		u64 is_l3_mcast : 1;
+		u64 is_l2_bcast : 1;
+		u64 is_l2_mcast : 1;
+		u64 is_raw : 1;
+		u64 err_level : 3;
+		u64 err_code : 8;
+	};
+} cvmx_pki_wqe_word2_t;
+
+typedef union {
+	u64 u64;
+	cvmx_pki_wqe_word2_t pki;
+	cvmx_pip_wqe_word2_t pip;
+} cvmx_wqe_word2_t;
+
+typedef union {
+	u64 u64;
+	struct {
+		u16 hw_chksum;
+		u8 unused;
+		u64 next_ptr : 40;
+	} cn38xx;
+	struct {
+		u64 l4ptr : 8;	  /* 56..63 */
+		u64 unused0 : 8;  /* 48..55 */
+		u64 l3ptr : 8;	  /* 40..47 */
+		u64 l2ptr : 8;	  /* 32..39 */
+		u64 unused1 : 18; /* 14..31 */
+		u64 bpid : 6;	  /* 8..13 */
+		u64 unused2 : 2;  /* 6..7 */
+		u64 pknd : 6;	  /* 0..5 */
+	} cn68xx;
+} cvmx_pip_wqe_word0_t;
+
+typedef union {
+	u64 u64;
+	struct {
+		u64 rsvd_0 : 4;
+		u64 aura : 12;
+		u64 rsvd_1 : 1;
+		u64 apad : 3;
+		u64 channel : 12;
+		u64 bufs : 8;
+		u64 style : 8;
+		u64 rsvd_2 : 10;
+		u64 pknd : 6;
+	};
+} cvmx_pki_wqe_word0_t;
+
+/* Use reserved bit, set by HW to 0, to indicate buf_ptr legacy translation*/
+#define pki_wqe_translated word0.rsvd_1
+
+typedef union {
+	u64 u64;
+	cvmx_pip_wqe_word0_t pip;
+	cvmx_pki_wqe_word0_t pki;
+	struct {
+		u64 unused : 24;
+		u64 next_ptr : 40; /* On cn68xx this is unused as well */
+	} raw;
+} cvmx_wqe_word0_t;
+
+typedef union {
+	u64 u64;
+	struct {
+		u64 len : 16;
+		u64 rsvd_0 : 2;
+		u64 rsvd_1 : 2;
+		u64 grp : 10;
+		cvmx_pow_tag_type_t tag_type : 2;
+		u64 tag : 32;
+	};
+} cvmx_pki_wqe_word1_t;
+
+#define pki_errata20776 word1.rsvd_0
+
+typedef union {
+	u64 u64;
+	struct {
+		u64 len : 16;
+		u64 varies : 14;
+		cvmx_pow_tag_type_t tag_type : 2;
+		u64 tag : 32;
+	};
+	cvmx_pki_wqe_word1_t cn78xx;
+	struct {
+		u64 len : 16;
+		u64 zero_0 : 1;
+		u64 qos : 3;
+		u64 zero_1 : 1;
+		u64 grp : 6;
+		u64 zero_2 : 3;
+		cvmx_pow_tag_type_t tag_type : 2;
+		u64 tag : 32;
+	} cn68xx;
+	struct {
+		u64 len : 16;
+		u64 ipprt : 6;
+		u64 qos : 3;
+		u64 grp : 4;
+		u64 zero_2 : 1;
+		cvmx_pow_tag_type_t tag_type : 2;
+		u64 tag : 32;
+	} cn38xx;
+} cvmx_wqe_word1_t;
+
+typedef union {
+	u64 u64;
+	struct {
+		u64 rsvd_0 : 8;
+		u64 hwerr : 8;
+		u64 rsvd_1 : 24;
+		u64 sqid : 8;
+		u64 rsvd_2 : 4;
+		u64 vfnum : 12;
+	};
+} cvmx_wqe_word3_t;
+
+typedef union {
+	u64 u64;
+	struct {
+		u64 rsvd_0 : 21;
+		u64 sqfc : 11;
+		u64 rsvd_1 : 5;
+		u64 sqtail : 11;
+		u64 rsvd_2 : 3;
+		u64 sqhead : 13;
+	};
+} cvmx_wqe_word4_t;
+
+/**
+ * Work queue entry format.
+ * Must be 8-byte aligned.
+ */
+typedef struct cvmx_wqe_s {
+	/*-------------------------------------------------------------------*/
+	/* WORD 0                                                            */
+	/*-------------------------------------------------------------------*/
+	/* HW WRITE: the following 64 bits are filled by HW when a packet
+	 * arrives.
+	 */
+	cvmx_wqe_word0_t word0;
+
+	/*-------------------------------------------------------------------*/
+	/* WORD 1                                                            */
+	/*-------------------------------------------------------------------*/
+	/* HW WRITE: the following 64 bits are filled by HW when a packet
+	 * arrives.
+	 */
+	cvmx_wqe_word1_t word1;
+
+	/*-------------------------------------------------------------------*/
+	/* WORD 2                                                            */
+	/*-------------------------------------------------------------------*/
+	/* HW WRITE: the following 64-bits are filled in by hardware when a
+	 * packet arrives. This indicates a variety of status and error
+	 *conditions.
+	 */
+	cvmx_pip_wqe_word2_t word2;
+
+	/* Pointer to the first segment of the packet. */
+	cvmx_buf_ptr_t packet_ptr;
+
+	/* HW WRITE: OCTEON will fill in a programmable amount from the packet,
+	 * up to (at most, but perhaps less) the amount needed to fill the work
+	 * queue entry to 128 bytes. If the packet is recognized to be IP, the
+	 * hardware starts (except that the IPv4 header is padded for
+	 * appropriate alignment) writing here where the IP header starts.
+	 * If the packet is not recognized to be IP, the hardware starts
+	 * writing the beginning of the packet here.
+	 */
+	u8 packet_data[96];
+
+	/* If desired, SW can make the work Q entry any length. For the purposes
+	 * of discussion here, Assume 128B always, as this is all that the hardware
+	 * deals with.
+	 */
+} CVMX_CACHE_LINE_ALIGNED cvmx_wqe_t;
+
+/**
+ * Work queue entry format for NQM
+ * Must be 8-byte aligned
+ */
+typedef struct cvmx_wqe_nqm_s {
+	/*-------------------------------------------------------------------*/
+	/* WORD 0                                                            */
+	/*-------------------------------------------------------------------*/
+	/* HW WRITE: the following 64 bits are filled by HW when a packet
+	 * arrives.
+	 */
+	cvmx_wqe_word0_t word0;
+
+	/*-------------------------------------------------------------------*/
+	/* WORD 1                                                            */
+	/*-------------------------------------------------------------------*/
+	/* HW WRITE: the following 64 bits are filled by HW when a packet
+	 * arrives.
+	 */
+	cvmx_wqe_word1_t word1;
+
+	/*-------------------------------------------------------------------*/
+	/* WORD 2                                                            */
+	/*-------------------------------------------------------------------*/
+	/* Reserved */
+	u64 word2;
+
+	/*-------------------------------------------------------------------*/
+	/* WORD 3                                                            */
+	/*-------------------------------------------------------------------*/
+	/* NVMe specific information.*/
+	cvmx_wqe_word3_t word3;
+
+	/*-------------------------------------------------------------------*/
+	/* WORD 4                                                            */
+	/*-------------------------------------------------------------------*/
+	/* NVMe specific information.*/
+	cvmx_wqe_word4_t word4;
+
+	/* HW WRITE: OCTEON will fill in a programmable amount from the packet,
+	 * up to (at most, but perhaps less) the amount needed to fill the work
+	 * queue entry to 128 bytes. If the packet is recognized to be IP, the
+	 * hardware starts (except that the IPv4 header is padded for
+	 * appropriate alignment) writing here where the IP header starts.
+	 * If the packet is not recognized to be IP, the hardware starts
+	 * writing the beginning of the packet here.
+	 */
+	u8 packet_data[88];
+
+	/* If desired, SW can make the work Q entry any length.
+	 * For the purposes of discussion here, assume 128B always, as this is
+	 * all that the hardware deals with.
+	 */
+} CVMX_CACHE_LINE_ALIGNED cvmx_wqe_nqm_t;
+
+/**
+ * Work queue entry format for 78XX.
+ * In 78XX packet data always resides in WQE buffer unless option
+ * DIS_WQ_DAT=1 in PKI_STYLE_BUF, which causes packet data to use separate buffer.
+ *
+ * Must be 8-byte aligned.
+ */
+typedef struct {
+	/*-------------------------------------------------------------------*/
+	/* WORD 0                                                            */
+	/*-------------------------------------------------------------------*/
+	/* HW WRITE: the following 64 bits are filled by HW when a packet
+	 * arrives.
+	 */
+	cvmx_pki_wqe_word0_t word0;
+
+	/*-------------------------------------------------------------------*/
+	/* WORD 1                                                            */
+	/*-------------------------------------------------------------------*/
+	/* HW WRITE: the following 64 bits are filled by HW when a packet
+	 * arrives.
+	 */
+	cvmx_pki_wqe_word1_t word1;
+
+	/*-------------------------------------------------------------------*/
+	/* WORD 2                                                            */
+	/*-------------------------------------------------------------------*/
+	/* HW WRITE: the following 64-bits are filled in by hardware when a
+	 * packet arrives. This indicates a variety of status and error
+	 * conditions.
+	 */
+	cvmx_pki_wqe_word2_t word2;
+
+	/*-------------------------------------------------------------------*/
+	/* WORD 3                                                            */
+	/*-------------------------------------------------------------------*/
+	/* Pointer to the first segment of the packet.*/
+	cvmx_buf_ptr_pki_t packet_ptr;
+
+	/*-------------------------------------------------------------------*/
+	/* WORD 4                                                            */
+	/*-------------------------------------------------------------------*/
+	/* HW WRITE: the following 64-bits are filled in by hardware when a
+	 * packet arrives contains a byte pointer to the start of Layer
+	 * A/B/C/D/E/F/G relative of start of packet.
+	 */
+	cvmx_pki_wqe_word4_t word4;
+
+	/*-------------------------------------------------------------------*/
+	/* WORDs 5/6/7 may be extended there, if WQE_HSZ is set.             */
+	/*-------------------------------------------------------------------*/
+	u64 wqe_data[11];
+
+} CVMX_CACHE_LINE_ALIGNED cvmx_wqe_78xx_t;
+
+/* Node LS-bit position in the WQE[grp] or PKI_QPG_TBL[grp_ok].*/
+#define CVMX_WQE_GRP_NODE_SHIFT 8
+
+/*
+ * This is an accessor function into the WQE that retreives the
+ * ingress port number, which can also be used as a destination
+ * port number for the same port.
+ *
+ * @param work - Work Queue Entrey pointer
+ * @returns returns the normalized port number, also known as "ipd" port
+ */
+static inline int cvmx_wqe_get_port(cvmx_wqe_t *work)
+{
+	int port;
+
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		/* In 78xx wqe entry has channel number not port*/
+		port = work->word0.pki.channel;
+		/* For BGX interfaces (0x800 - 0xdff) the 4 LSBs indicate
+		 * the PFC channel, must be cleared to normalize to "ipd"
+		 */
+		if (port & 0x800)
+			port &= 0xff0;
+		/* Node number is in AURA field, make it part of port # */
+		port |= (work->word0.pki.aura >> 10) << 12;
+	} else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE)) {
+		port = work->word2.s_cn68xx.port;
+	} else {
+		port = work->word1.cn38xx.ipprt;
+	}
+
+	return port;
+}
+
+static inline void cvmx_wqe_set_port(cvmx_wqe_t *work, int port)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE))
+		work->word0.pki.channel = port;
+	else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE))
+		work->word2.s_cn68xx.port = port;
+	else
+		work->word1.cn38xx.ipprt = port;
+}
+
+static inline int cvmx_wqe_get_grp(cvmx_wqe_t *work)
+{
+	int grp;
+
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE))
+		/* legacy: GRP[0..2] :=QOS */
+		grp = (0xff & work->word1.cn78xx.grp) >> 3;
+	else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE))
+		grp = work->word1.cn68xx.grp;
+	else
+		grp = work->word1.cn38xx.grp;
+
+	return grp;
+}
+
+static inline void cvmx_wqe_set_xgrp(cvmx_wqe_t *work, int grp)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE))
+		work->word1.cn78xx.grp = grp;
+	else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE))
+		work->word1.cn68xx.grp = grp;
+	else
+		work->word1.cn38xx.grp = grp;
+}
+
+static inline int cvmx_wqe_get_xgrp(cvmx_wqe_t *work)
+{
+	int grp;
+
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE))
+		grp = work->word1.cn78xx.grp;
+	else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE))
+		grp = work->word1.cn68xx.grp;
+	else
+		grp = work->word1.cn38xx.grp;
+
+	return grp;
+}
+
+static inline void cvmx_wqe_set_grp(cvmx_wqe_t *work, int grp)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		unsigned int node = cvmx_get_node_num();
+		/* Legacy: GRP[0..2] :=QOS */
+		work->word1.cn78xx.grp &= 0x7;
+		work->word1.cn78xx.grp |= 0xff & (grp << 3);
+		work->word1.cn78xx.grp |= (node << 8);
+	} else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE)) {
+		work->word1.cn68xx.grp = grp;
+	} else {
+		work->word1.cn38xx.grp = grp;
+	}
+}
+
+static inline int cvmx_wqe_get_qos(cvmx_wqe_t *work)
+{
+	int qos;
+
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		/* Legacy: GRP[0..2] :=QOS */
+		qos = work->word1.cn78xx.grp & 0x7;
+	} else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE)) {
+		qos = work->word1.cn68xx.qos;
+	} else {
+		qos = work->word1.cn38xx.qos;
+	}
+
+	return qos;
+}
+
+static inline void cvmx_wqe_set_qos(cvmx_wqe_t *work, int qos)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		/* legacy: GRP[0..2] :=QOS */
+		work->word1.cn78xx.grp &= ~0x7;
+		work->word1.cn78xx.grp |= qos & 0x7;
+	} else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE)) {
+		work->word1.cn68xx.qos = qos;
+	} else {
+		work->word1.cn38xx.qos = qos;
+	}
+}
+
+static inline int cvmx_wqe_get_len(cvmx_wqe_t *work)
+{
+	int len;
+
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE))
+		len = work->word1.cn78xx.len;
+	else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE))
+		len = work->word1.cn68xx.len;
+	else
+		len = work->word1.cn38xx.len;
+
+	return len;
+}
+
+static inline void cvmx_wqe_set_len(cvmx_wqe_t *work, int len)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE))
+		work->word1.cn78xx.len = len;
+	else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE))
+		work->word1.cn68xx.len = len;
+	else
+		work->word1.cn38xx.len = len;
+}
+
+/**
+ * This function returns, if there was L2/L1 errors detected in packet.
+ *
+ * @param work	pointer to work queue entry
+ *
+ * @return	0 if packet had no error, non-zero to indicate error code.
+ *
+ * Please refer to HRM for the specific model for full enumaration of error codes.
+ * With Octeon1/Octeon2 models, the returned code indicates L1/L2 errors.
+ * On CN73XX/CN78XX, the return code is the value of PKI_OPCODE_E,
+ * if it is non-zero, otherwise the returned code will be derived from
+ * PKI_ERRLEV_E such that an error indicated in LayerA will return 0x20,
+ * LayerB - 0x30, LayerC - 0x40 and so forth.
+ */
+static inline int cvmx_wqe_get_rcv_err(cvmx_wqe_t *work)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		if (wqe->word2.err_level == CVMX_PKI_ERRLEV_E_RE || wqe->word2.err_code != 0)
+			return wqe->word2.err_code;
+		else
+			return (wqe->word2.err_level << 4) + 0x10;
+	} else if (work->word2.snoip.rcv_error) {
+		return work->word2.snoip.err_code;
+	}
+
+	return 0;
+}
+
+static inline u32 cvmx_wqe_get_tag(cvmx_wqe_t *work)
+{
+	return work->word1.tag;
+}
+
+static inline void cvmx_wqe_set_tag(cvmx_wqe_t *work, u32 tag)
+{
+	work->word1.tag = tag;
+}
+
+static inline int cvmx_wqe_get_tt(cvmx_wqe_t *work)
+{
+	return work->word1.tag_type;
+}
+
+static inline void cvmx_wqe_set_tt(cvmx_wqe_t *work, int tt)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		work->word1.cn78xx.tag_type = (cvmx_pow_tag_type_t)tt;
+	} else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE)) {
+		work->word1.cn68xx.tag_type = (cvmx_pow_tag_type_t)tt;
+		work->word1.cn68xx.zero_2 = 0;
+	} else {
+		work->word1.cn38xx.tag_type = (cvmx_pow_tag_type_t)tt;
+		work->word1.cn38xx.zero_2 = 0;
+	}
+}
+
+static inline u8 cvmx_wqe_get_unused8(cvmx_wqe_t *work)
+{
+	u8 bits;
+
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		bits = wqe->word2.rsvd_0;
+	} else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE)) {
+		bits = work->word0.pip.cn68xx.unused1;
+	} else {
+		bits = work->word0.pip.cn38xx.unused;
+	}
+
+	return bits;
+}
+
+static inline void cvmx_wqe_set_unused8(cvmx_wqe_t *work, u8 v)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		wqe->word2.rsvd_0 = v;
+	} else if (octeon_has_feature(OCTEON_FEATURE_CN68XX_WQE)) {
+		work->word0.pip.cn68xx.unused1 = v;
+	} else {
+		work->word0.pip.cn38xx.unused = v;
+	}
+}
+
+static inline u8 cvmx_wqe_get_user_flags(cvmx_wqe_t *work)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE))
+		return work->word0.pki.rsvd_2;
+	else
+		return 0;
+}
+
+static inline void cvmx_wqe_set_user_flags(cvmx_wqe_t *work, u8 v)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE))
+		work->word0.pki.rsvd_2 = v;
+}
+
+static inline int cvmx_wqe_get_channel(cvmx_wqe_t *work)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE))
+		return (work->word0.pki.channel);
+	else
+		return cvmx_wqe_get_port(work);
+}
+
+static inline void cvmx_wqe_set_channel(cvmx_wqe_t *work, int channel)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE))
+		work->word0.pki.channel = channel;
+	else
+		debug("%s: ERROR: not supported for model\n", __func__);
+}
+
+static inline int cvmx_wqe_get_aura(cvmx_wqe_t *work)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE))
+		return (work->word0.pki.aura);
+	else
+		return (work->packet_ptr.s.pool);
+}
+
+static inline void cvmx_wqe_set_aura(cvmx_wqe_t *work, int aura)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE))
+		work->word0.pki.aura = aura;
+	else
+		work->packet_ptr.s.pool = aura;
+}
+
+static inline int cvmx_wqe_get_style(cvmx_wqe_t *work)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE))
+		return (work->word0.pki.style);
+	return 0;
+}
+
+static inline void cvmx_wqe_set_style(cvmx_wqe_t *work, int style)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE))
+		work->word0.pki.style = style;
+}
+
+static inline int cvmx_wqe_is_l3_ip(cvmx_wqe_t *work)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+		/* Match all 4 values for v4/v6 with.without options */
+		if ((wqe->word2.lc_hdr_type & 0x1c) == CVMX_PKI_LTYPE_E_IP4)
+			return 1;
+		if ((wqe->word2.le_hdr_type & 0x1c) == CVMX_PKI_LTYPE_E_IP4)
+			return 1;
+		return 0;
+	} else {
+		return !work->word2.s_cn38xx.not_IP;
+	}
+}
+
+static inline int cvmx_wqe_is_l3_ipv4(cvmx_wqe_t *work)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+		/* Match 2 values - with/wotuout options */
+		if ((wqe->word2.lc_hdr_type & 0x1e) == CVMX_PKI_LTYPE_E_IP4)
+			return 1;
+		if ((wqe->word2.le_hdr_type & 0x1e) == CVMX_PKI_LTYPE_E_IP4)
+			return 1;
+		return 0;
+	} else {
+		return (!work->word2.s_cn38xx.not_IP &&
+			!work->word2.s_cn38xx.is_v6);
+	}
+}
+
+static inline int cvmx_wqe_is_l3_ipv6(cvmx_wqe_t *work)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+		/* Match 2 values - with/wotuout options */
+		if ((wqe->word2.lc_hdr_type & 0x1e) == CVMX_PKI_LTYPE_E_IP6)
+			return 1;
+		if ((wqe->word2.le_hdr_type & 0x1e) == CVMX_PKI_LTYPE_E_IP6)
+			return 1;
+		return 0;
+	} else {
+		return (!work->word2.s_cn38xx.not_IP &&
+			work->word2.s_cn38xx.is_v6);
+	}
+}
+
+static inline bool cvmx_wqe_is_l4_udp_or_tcp(cvmx_wqe_t *work)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		if (wqe->word2.lf_hdr_type == CVMX_PKI_LTYPE_E_TCP)
+			return true;
+		if (wqe->word2.lf_hdr_type == CVMX_PKI_LTYPE_E_UDP)
+			return true;
+		return false;
+	}
+
+	if (work->word2.s_cn38xx.not_IP)
+		return false;
+
+	return (work->word2.s_cn38xx.tcp_or_udp != 0);
+}
+
+static inline int cvmx_wqe_is_l2_bcast(cvmx_wqe_t *work)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		return wqe->word2.is_l2_bcast;
+	} else {
+		return work->word2.s_cn38xx.is_bcast;
+	}
+}
+
+static inline int cvmx_wqe_is_l2_mcast(cvmx_wqe_t *work)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		return wqe->word2.is_l2_mcast;
+	} else {
+		return work->word2.s_cn38xx.is_mcast;
+	}
+}
+
+static inline void cvmx_wqe_set_l2_bcast(cvmx_wqe_t *work, bool bcast)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		wqe->word2.is_l2_bcast = bcast;
+	} else {
+		work->word2.s_cn38xx.is_bcast = bcast;
+	}
+}
+
+static inline void cvmx_wqe_set_l2_mcast(cvmx_wqe_t *work, bool mcast)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		wqe->word2.is_l2_mcast = mcast;
+	} else {
+		work->word2.s_cn38xx.is_mcast = mcast;
+	}
+}
+
+static inline int cvmx_wqe_is_l3_bcast(cvmx_wqe_t *work)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		return wqe->word2.is_l3_bcast;
+	}
+	debug("%s: ERROR: not supported for model\n", __func__);
+	return 0;
+}
+
+static inline int cvmx_wqe_is_l3_mcast(cvmx_wqe_t *work)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		return wqe->word2.is_l3_mcast;
+	}
+	debug("%s: ERROR: not supported for model\n", __func__);
+	return 0;
+}
+
+/**
+ * This function returns is there was IP error detected in packet.
+ * For 78XX it does not flag ipv4 options and ipv6 extensions.
+ * For older chips if PIP_GBL_CTL was proviosned to flag ip4_otions and
+ * ipv6 extension, it will be flag them.
+ * @param work	pointer to work queue entry
+ * @return	1 -- If IP error was found in packet
+ *          0 -- If no IP error was found in packet.
+ */
+static inline int cvmx_wqe_is_ip_exception(cvmx_wqe_t *work)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		if (wqe->word2.err_level == CVMX_PKI_ERRLEV_E_LC)
+			return 1;
+		else
+			return 0;
+	}
+
+	return work->word2.s.IP_exc;
+}
+
+static inline int cvmx_wqe_is_l4_error(cvmx_wqe_t *work)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		if (wqe->word2.err_level == CVMX_PKI_ERRLEV_E_LF)
+			return 1;
+		else
+			return 0;
+	} else {
+		return work->word2.s.L4_error;
+	}
+}
+
+static inline void cvmx_wqe_set_vlan(cvmx_wqe_t *work, bool set)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		wqe->word2.vlan_valid = set;
+	} else {
+		work->word2.s.vlan_valid = set;
+	}
+}
+
+static inline int cvmx_wqe_is_vlan(cvmx_wqe_t *work)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		return wqe->word2.vlan_valid;
+	} else {
+		return work->word2.s.vlan_valid;
+	}
+}
+
+static inline int cvmx_wqe_is_vlan_stacked(cvmx_wqe_t *work)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		return wqe->word2.vlan_stacked;
+	} else {
+		return work->word2.s.vlan_stacked;
+	}
+}
+
+/**
+ * Extract packet data buffer pointer from work queue entry.
+ *
+ * Returns the legacy (Octeon1/Octeon2) buffer pointer structure
+ * for the linked buffer list.
+ * On CN78XX, the native buffer pointer structure is converted into
+ * the legacy format.
+ * The legacy buf_ptr is then stored in the WQE, and word0 reserved
+ * field is set to indicate that the buffer pointers were translated.
+ * If the packet data is only found inside the work queue entry,
+ * a standard buffer pointer structure is created for it.
+ */
+cvmx_buf_ptr_t cvmx_wqe_get_packet_ptr(cvmx_wqe_t *work);
+
+static inline int cvmx_wqe_get_bufs(cvmx_wqe_t *work)
+{
+	int bufs;
+
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		bufs = work->word0.pki.bufs;
+	} else {
+		/* Adjust for packet-in-WQE cases */
+		if (cvmx_unlikely(work->word2.s_cn38xx.bufs == 0 && !work->word2.s.software))
+			(void)cvmx_wqe_get_packet_ptr(work);
+		bufs = work->word2.s_cn38xx.bufs;
+	}
+	return bufs;
+}
+
+/**
+ * Free Work Queue Entry memory
+ *
+ * Will return the WQE buffer to its pool, unless the WQE contains
+ * non-redundant packet data.
+ * This function is intended to be called AFTER the packet data
+ * has been passed along to PKO for transmission and release.
+ * It can also follow a call to cvmx_helper_free_packet_data()
+ * to release the WQE after associated data was released.
+ */
+void cvmx_wqe_free(cvmx_wqe_t *work);
+
+/**
+ * Check if a work entry has been intiated by software
+ *
+ */
+static inline bool cvmx_wqe_is_soft(cvmx_wqe_t *work)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		return wqe->word2.software;
+	} else {
+		return work->word2.s.software;
+	}
+}
+
+/**
+ * Allocate a work-queue entry for delivering software-initiated
+ * event notifications.
+ * The application data is copied into the work-queue entry,
+ * if the space is sufficient.
+ */
+cvmx_wqe_t *cvmx_wqe_soft_create(void *data_p, unsigned int data_sz);
+
+/* Errata (PKI-20776) PKI_BUFLINK_S's are endian-swapped
+ * CN78XX pass 1.x has a bug where the packet pointer in each segment is
+ * written in the opposite endianness of the configured mode. Fix these here.
+ */
+static inline void cvmx_wqe_pki_errata_20776(cvmx_wqe_t *work)
+{
+	cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+	if (OCTEON_IS_MODEL(OCTEON_CN78XX_PASS1_X) && !wqe->pki_errata20776) {
+		u64 bufs;
+		cvmx_buf_ptr_pki_t buffer_next;
+
+		bufs = wqe->word0.bufs;
+		buffer_next = wqe->packet_ptr;
+		while (bufs > 1) {
+			cvmx_buf_ptr_pki_t next;
+			void *nextaddr = cvmx_phys_to_ptr(buffer_next.addr - 8);
+
+			memcpy(&next, nextaddr, sizeof(next));
+			next.u64 = __builtin_bswap64(next.u64);
+			memcpy(nextaddr, &next, sizeof(next));
+			buffer_next = next;
+			bufs--;
+		}
+		wqe->pki_errata20776 = 1;
+	}
+}
+
+/**
+ * @INTERNAL
+ *
+ * Extract the native PKI-specific buffer pointer from WQE.
+ *
+ * NOTE: Provisional, may be superceded.
+ */
+static inline cvmx_buf_ptr_pki_t cvmx_wqe_get_pki_pkt_ptr(cvmx_wqe_t *work)
+{
+	cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+	if (!octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_buf_ptr_pki_t x = { 0 };
+		return x;
+	}
+
+	cvmx_wqe_pki_errata_20776(work);
+	return wqe->packet_ptr;
+}
+
+/**
+ * Set the buffer segment count for a packet.
+ *
+ * @return Returns the actual resulting value in the WQE fielda
+ *
+ */
+static inline unsigned int cvmx_wqe_set_bufs(cvmx_wqe_t *work, unsigned int bufs)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		work->word0.pki.bufs = bufs;
+		return work->word0.pki.bufs;
+	}
+
+	work->word2.s.bufs = bufs;
+	return work->word2.s.bufs;
+}
+
+/**
+ * Get the offset of Layer-3 header,
+ * only supported when Layer-3 protocol is IPv4 or IPv6.
+ *
+ * @return Returns the offset, or 0 if the offset is not known or unsupported.
+ *
+ * FIXME: Assuming word4 is present.
+ */
+static inline unsigned int cvmx_wqe_get_l3_offset(cvmx_wqe_t *work)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+		/* Match 4 values: IPv4/v6 w/wo options */
+		if ((wqe->word2.lc_hdr_type & 0x1c) == CVMX_PKI_LTYPE_E_IP4)
+			return wqe->word4.ptr_layer_c;
+	} else {
+		return work->word2.s.ip_offset;
+	}
+
+	return 0;
+}
+
+/**
+ * Set the offset of Layer-3 header in a packet.
+ * Typically used when an IP packet is generated by software
+ * or when the Layer-2 header length is modified, and
+ * a subsequent recalculation of checksums is anticipated.
+ *
+ * @return Returns the actual value of the work entry offset field.
+ *
+ * FIXME: Assuming word4 is present.
+ */
+static inline unsigned int cvmx_wqe_set_l3_offset(cvmx_wqe_t *work, unsigned int ip_off)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+		/* Match 4 values: IPv4/v6 w/wo options */
+		if ((wqe->word2.lc_hdr_type & 0x1c) == CVMX_PKI_LTYPE_E_IP4)
+			wqe->word4.ptr_layer_c = ip_off;
+	} else {
+		work->word2.s.ip_offset = ip_off;
+	}
+
+	return cvmx_wqe_get_l3_offset(work);
+}
+
+/**
+ * Set the indication that the packet contains a IPv4 Layer-3 * header.
+ * Use 'cvmx_wqe_set_l3_ipv6()' if the protocol is IPv6.
+ * When 'set' is false, the call will result in an indication
+ * that the Layer-3 protocol is neither IPv4 nor IPv6.
+ *
+ * FIXME: Add IPV4_OPT handling based on L3 header length.
+ */
+static inline void cvmx_wqe_set_l3_ipv4(cvmx_wqe_t *work, bool set)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		if (set)
+			wqe->word2.lc_hdr_type = CVMX_PKI_LTYPE_E_IP4;
+		else
+			wqe->word2.lc_hdr_type = CVMX_PKI_LTYPE_E_NONE;
+	} else {
+		work->word2.s.not_IP = !set;
+		if (set)
+			work->word2.s_cn38xx.is_v6 = 0;
+	}
+}
+
+/**
+ * Set packet Layer-3 protocol to IPv6.
+ *
+ * FIXME: Add IPV6_OPT handling based on presence of extended headers.
+ */
+static inline void cvmx_wqe_set_l3_ipv6(cvmx_wqe_t *work, bool set)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		if (set)
+			wqe->word2.lc_hdr_type = CVMX_PKI_LTYPE_E_IP6;
+		else
+			wqe->word2.lc_hdr_type = CVMX_PKI_LTYPE_E_NONE;
+	} else {
+		work->word2.s_cn38xx.not_IP = !set;
+		if (set)
+			work->word2.s_cn38xx.is_v6 = 1;
+	}
+}
+
+/**
+ * Set a packet Layer-4 protocol type to UDP.
+ */
+static inline void cvmx_wqe_set_l4_udp(cvmx_wqe_t *work, bool set)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		if (set)
+			wqe->word2.lf_hdr_type = CVMX_PKI_LTYPE_E_UDP;
+		else
+			wqe->word2.lf_hdr_type = CVMX_PKI_LTYPE_E_NONE;
+	} else {
+		if (!work->word2.s_cn38xx.not_IP)
+			work->word2.s_cn38xx.tcp_or_udp = set;
+	}
+}
+
+/**
+ * Set a packet Layer-4 protocol type to TCP.
+ */
+static inline void cvmx_wqe_set_l4_tcp(cvmx_wqe_t *work, bool set)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		if (set)
+			wqe->word2.lf_hdr_type = CVMX_PKI_LTYPE_E_TCP;
+		else
+			wqe->word2.lf_hdr_type = CVMX_PKI_LTYPE_E_NONE;
+	} else {
+		if (!work->word2.s_cn38xx.not_IP)
+			work->word2.s_cn38xx.tcp_or_udp = set;
+	}
+}
+
+/**
+ * Set the "software" flag in a work entry.
+ */
+static inline void cvmx_wqe_set_soft(cvmx_wqe_t *work, bool set)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		wqe->word2.software = set;
+	} else {
+		work->word2.s.software = set;
+	}
+}
+
+/**
+ * Return true if the packet is an IP fragment.
+ */
+static inline bool cvmx_wqe_is_l3_frag(cvmx_wqe_t *work)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		return (wqe->word2.is_frag != 0);
+	}
+
+	if (!work->word2.s_cn38xx.not_IP)
+		return (work->word2.s.is_frag != 0);
+
+	return false;
+}
+
+/**
+ * Set the indicator that the packet is an fragmented IP packet.
+ */
+static inline void cvmx_wqe_set_l3_frag(cvmx_wqe_t *work, bool set)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		wqe->word2.is_frag = set;
+	} else {
+		if (!work->word2.s_cn38xx.not_IP)
+			work->word2.s.is_frag = set;
+	}
+}
+
+/**
+ * Set the packet Layer-3 protocol to RARP.
+ */
+static inline void cvmx_wqe_set_l3_rarp(cvmx_wqe_t *work, bool set)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		if (set)
+			wqe->word2.lc_hdr_type = CVMX_PKI_LTYPE_E_RARP;
+		else
+			wqe->word2.lc_hdr_type = CVMX_PKI_LTYPE_E_NONE;
+	} else {
+		work->word2.snoip.is_rarp = set;
+	}
+}
+
+/**
+ * Set the packet Layer-3 protocol to ARP.
+ */
+static inline void cvmx_wqe_set_l3_arp(cvmx_wqe_t *work, bool set)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		if (set)
+			wqe->word2.lc_hdr_type = CVMX_PKI_LTYPE_E_ARP;
+		else
+			wqe->word2.lc_hdr_type = CVMX_PKI_LTYPE_E_NONE;
+	} else {
+		work->word2.snoip.is_arp = set;
+	}
+}
+
+/**
+ * Return true if the packet Layer-3 protocol is ARP.
+ */
+static inline bool cvmx_wqe_is_l3_arp(cvmx_wqe_t *work)
+{
+	if (octeon_has_feature(OCTEON_FEATURE_CN78XX_WQE)) {
+		cvmx_wqe_78xx_t *wqe = (cvmx_wqe_78xx_t *)work;
+
+		return (wqe->word2.lc_hdr_type == CVMX_PKI_LTYPE_E_ARP);
+	}
+
+	if (work->word2.s_cn38xx.not_IP)
+		return (work->word2.snoip.is_arp != 0);
+
+	return false;
+}
+
+#endif /* __CVMX_WQE_H__ */
-- 
2.29.2



More information about the U-Boot mailing list