[U-Boot] [PATCH v1 2/3] ddr: vybrid: Provide code to perform on-boot calibration
Stefan Agner
stefan at agner.ch
Mon Dec 3 15:53:56 UTC 2018
On 02.12.2018 21:42, Lukasz Majewski wrote:
> This patch provides the code to calibrate the DDR's
> DQS to DQ signals (RDLVL).
>
> It is based on:
> VFxxx Controller Reference Manual, Rev. 0, 10/2016, page 1600
> 10.1.6.16.4.1 "Software Read Leveling in MC Evaluation Mode"
>
> and NXP's community thread:
> "Vybrid: About DDR leveling feature on DDRMC."
> https://community.nxp.com/thread/395323
>
> Signed-off-by: Lukasz Majewski <lukma at denx.de>
Thanks for looking into this! We actually tried to use the official DDRV
tool, but were not able to get really good results with it!
In a quick test the calibration worked here on a Colibri VF50. Will do
some more testing later this week.
Some minor things below:
> ---
>
> arch/arm/mach-imx/Kconfig | 8 +
> arch/arm/mach-imx/Makefile | 1 +
> arch/arm/mach-imx/ddrmc-vf610-calibration.c | 336 ++++++++++++++++++++++++++++
> arch/arm/mach-imx/ddrmc-vf610-calibration.h | 59 +++++
> 4 files changed, 404 insertions(+)
> create mode 100644 arch/arm/mach-imx/ddrmc-vf610-calibration.c
> create mode 100644 arch/arm/mach-imx/ddrmc-vf610-calibration.h
>
> diff --git a/arch/arm/mach-imx/Kconfig b/arch/arm/mach-imx/Kconfig
> index a1566cc2ad..ca10ba683f 100644
> --- a/arch/arm/mach-imx/Kconfig
> +++ b/arch/arm/mach-imx/Kconfig
> @@ -78,3 +78,11 @@ config NXP_BOARD_REVISION
> NXP boards based on i.MX6/7 contain the board revision information
> stored in the fuses. Select this option if you want to be able to
> retrieve the board revision information.
> +
> +config DDRMC_VF610_CALIBRATION
> + bool "Enable DDRMC (DDR3) on-chip calibration"
> + depends on ARCH_VF610
> + help
> + Vybrid (vf610) SoC provides some on-chip facility to tune the DDR3
> + memory parameters. Select this option if you want to calculate them
> + at boot time.
> diff --git a/arch/arm/mach-imx/Makefile b/arch/arm/mach-imx/Makefile
> index 53d9e5f42b..4a07b1ea69 100644
> --- a/arch/arm/mach-imx/Makefile
> +++ b/arch/arm/mach-imx/Makefile
> @@ -51,6 +51,7 @@ obj-$(CONFIG_SECURE_BOOT) += hab.o
> endif
> ifeq ($(SOC),$(filter $(SOC),vf610))
> obj-y += ddrmc-vf610.o
> +obj-$(CONFIG_DDRMC_VF610_CALIBRATION) += ddrmc-vf610-calibration.o
> endif
> ifneq ($(CONFIG_SPL_BUILD),y)
> obj-$(CONFIG_CMD_BMODE) += cmd_bmode.o
> diff --git a/arch/arm/mach-imx/ddrmc-vf610-calibration.c
> b/arch/arm/mach-imx/ddrmc-vf610-calibration.c
> new file mode 100644
> index 0000000000..6ed3f5375b
> --- /dev/null
> +++ b/arch/arm/mach-imx/ddrmc-vf610-calibration.c
> @@ -0,0 +1,336 @@
> +// SPDX-License-Identifier: GPL-2.0+
> +/*
> + * ddrmc DDR3 calibration code for NXP's VF610
> + *
> + * Copyright (C) 2018 DENX Software Engineering
> + * Lukasz Majewski, DENX Software Engineering, lukma at denx.de
> + *
> + */
> +/* #define DEBUG */
> +#include <common.h>
> +#include <asm/io.h>
> +#include <asm/arch/imx-regs.h>
> +#include <linux/bitmap.h>
> +
> +#include "ddrmc-vf610-calibration.h"
> +
> +/*
> + * Documents:
> + *
> + * [1] "Vybrid: About DDR leveling feature on DDRMC."
> + * https://community.nxp.com/thread/395323
> + *
> + * [2] VFxxx Controller Reference Manual, Rev. 0, 10/2016
> + *
> + *
> + * NOTE
> + * ====
> + *
> + * NXP recommends setting 'fixed' parameters instead of performing the
> + * training at each boot.
Can you also add this information to the Kconfig? That is what people
usually read to decide whether to enable an option.
> + *
> + * Use those functions to determine those values on new HW and write
> + * default values to registers.
I guess the idea is to add it back to the board specific DDR settings?
Can you be a bit more clear, e.g.
Use those functions to determine those values on new HW, read the
calculated value from registers and add them to the board specific
struct ddrmc_cr_setting.
> + *
> + * SW leveling supported operations - CR93[SW_LVL_MODE]:
> + *
> + * - 0x0 (b'00) - No leveling
> + *
> + * - 0x1 (b'01) - WRLVL_DL_X - It is not recommended to perform this tuning
> + * on HW designs utilizing non-flyback topology
> + * (Single DDR3 with x16).
> + * Instead the WRLVL_DL_0/1 fields shall be set
> + * based on trace length differences from their
> + * layout.
> + * Mismatches up to 25% or tCK (clock period) are
> + * allowed, so the value in the filed doesn’t have
> + * to be very accurate.
> + *
> + * - 0x2 (b'10) - RDLVL_DL_0/1 - refers to adjusting the DQS strobe in relation
> + * to the DQ signals so that the strobe edge is
> + * centered in the window of valid read data.
> + *
> + * - 0x3 (b'11) - RDLVL_GTDL_0/1 - refers to the delay the PHY uses to un-gate
> + * the Read DQS strobe pad from the time that the
> + * PHY enables the pad to input the strobe signal.
> + *
> + */
> +static int ddr_cal_get_first_edge_index(unsigned long *bmap, enum edge e,
> + int samples, int start, int max)
> +{
> + int i, ret = -1;
> +
> + /*
> + * We look only for the first value (and filter out
> + * some wrong data)
> + */
> + switch (e) {
> + case RISING_EDGE:
> + for (i = start; i <= max - samples; i++) {
> + if (test_bit(i, bmap)) {
> + if (!test_bit(i - 1, bmap) &&
> + test_bit(i + 1, bmap) &&
> + test_bit(i + 2, bmap) &&
> + test_bit(i + 3, bmap)) {
> + return i;
> + }
> + }
> + }
> + break;
> + case FALLING_EDGE:
> + for (i = start; i <= max - samples; i++) {
> + if (!test_bit(i, bmap)) {
> + if (test_bit(i - 1, bmap) &&
> + test_bit(i - 2, bmap) &&
> + test_bit(i - 3, bmap)) {
> + return i;
> + }
> + }
> + }
> + }
> +
> + return ret;
> +}
> +
> +static void bitmap_print(unsigned long *bmap, int max)
> +{
> + int i;
> +
> + debug("BITMAP [0x%p]:\n", bmap);
> + for (i = 0; i <= max; i++) {
> + debug("%d ", test_bit(i, bmap) ? 1 : 0);
> + if (i && (i % 32) == (32 - 1))
> + debug("\n");
> + }
> + debug("\n");
> +}
> +
> +#define sw_leveling_op_done \
> + while (!(readl(&ddrmr->cr[94]) & DDRMC_CR94_SWLVL_OP_DONE))
> +
> +#define sw_leveling_load_value \
> + do { clrsetbits_le32(&ddrmr->cr[93], DDRMC_CR93_SWLVL_LOAD, \
> + DDRMC_CR93_SWLVL_LOAD); } while (0)
> +
> +#define sw_leveling_start \
> + do { clrsetbits_le32(&ddrmr->cr[93], DDRMC_CR93_SWLVL_START, \
> + DDRMC_CR93_SWLVL_START); } while (0)
> +
> +#define sw_leveling_exit \
> + do { clrsetbits_le32(&ddrmr->cr[94], DDRMC_CR93_SWLVL_EXIT, \
> + DDRMC_CR93_SWLVL_EXIT); } while (0)
> +
> +/*
> + * RDLVL_DL calibration:
> + *
> + * NXP is _NOT_ recommending performing the leveling at each
> + * boot. Instead - one shall run this procedure on new boards
> + * and then use hardcoded values.
> + *
> + */
> +static int ddrmc_cal_dqs_to_dq(struct ddrmr_regs *ddrmr)
> +{
> + DECLARE_BITMAP(rdlvl_rsp, DDRMC_DQS_DQ_MAX_DELAY + 1);
> + int rdlvl_dl_0_min = -1, rdlvl_dl_0_max = -1;
> + int rdlvl_dl_1_min = -1, rdlvl_dl_1_max = -1;
> + int rdlvl_dl_0, rdlvl_dl_1;
> + u8 swlvl_rsp;
> + u32 tmp;
> + int i;
> +
> + /* Read defaults */
> + u16 rdlvl_dl_0_def =
> + (readl(&ddrmr->cr[105]) >> RDLVL_DL_O_OFF) & 0xFFFF;
> + u16 rdlvl_dl_1_def = readl(&ddrmr->cr[110]) & 0xFFFF;
> +
> + debug("\nRDLVL: ======================\n");
> + debug("RDLVL: DQS to DQ (RDLVL)\n");
> +
> + debug("RDLVL: RDLVL_DL_0_DFL:\t 0x%x\n", rdlvl_dl_0_def);
> + debug("RDLVL: RDLVL_DL_1_DFL:\t 0x%x\n", rdlvl_dl_1_def);
> +
> + /*
> + * Set/Read setup for calibration
> + *
> + * Values necessary for leveling from Vybrid RM [2] - page 1600
> + */
> + writel(0x40703030, &ddrmr->cr[144]);
> + writel(0x40, &ddrmr->cr[145]);
> + writel(0x40, &ddrmr->cr[146]);
> +
> + tmp = readl(&ddrmr->cr[144]);
> + debug("RDLVL: PHY_RDLVL_RES:\t 0x%x\n", (tmp >> 24) & 0xFF);// set 0x40
> + debug("RDLVL: PHY_RDLV_LOAD:\t 0x%x\n", (tmp >> 16) & 0xFF);// set 0x70
> + debug("RDLVL: PHY_RDLV_DLL:\t 0x%x\n", (tmp >> 8) & 0xFF); // set 0x30
> + debug("RDLVL: PHY_RDLV_EN:\t 0x%x\n", tmp & 0xFF); //set 0x30
> +
> + tmp = readl(&ddrmr->cr[145]);
> + debug("RDLVL: PHY_RDLV_RR:\t 0x%x\n", tmp & 0x3FF); //set 0x40
> +
> + tmp = readl(&ddrmr->cr[146]);
> + debug("RDLVL: PHY_RDLV_RESP:\t 0x%x\n", tmp); //set 0x40
> +
> + /*
> + * Program/read the leveling edge RDLVL_EDGE = 0
> + *
> + * 0x00 is the correct output on SWLVL_RSP_X
> + * If by any chance 1s are visible -> wrong number read
> + */
> + clrbits_le32(&ddrmr->cr[101], DDRMC_CR101_PHY_RDLVL_EDGE);
> +
> + tmp = readl(&ddrmr->cr[101]);
> + debug("RDLVL: PHY_RDLVL_EDGE:\t 0x%x\n",
> + (tmp >> PHY_RDLVL_EDGE) & 0x1); //set 0
> +
> + /* Program Leveling mode - CR93[SW_LVL_MODE] to ’b10 */
> + clrsetbits_le32(&ddrmr->cr[93], DDRMC_CR93_SW_LVL_MODE(0x3),
> + DDRMC_CR93_SW_LVL_MODE(0x2));
> + tmp = readl(&ddrmr->cr[93]);
> + debug("RDLVL: SW_LVL_MODE:\t 0x%x\n", (tmp >> SW_LVL_MODE) & 0x3);
> +
> + /* Start procedure - CR93[SWLVL_START] to ’b1 */
> + sw_leveling_start;
> +
> + /* Poll CR94[SWLVL_OP_DONE] */
> + sw_leveling_op_done;
> +
> + /*
> + * Program delays for RDLVL_DL_0
> + *
> + * The procedure is to increase the delay values from 0 to 0xFF
> + * and read the response from the DDRMC
> + */
> + debug("\nRDLVL: ---> RDLVL_DL_0\n");
> + bitmap_zero(rdlvl_rsp, DDRMC_DQS_DQ_MAX_DELAY + 1);
> +
> + for (i = 0; i <= DDRMC_DQS_DQ_MAX_DELAY; i++) {
> + clrsetbits_le32(&ddrmr->cr[105], 0xFFFF << RDLVL_DL_O_OFF,
> + i << RDLVL_DL_O_OFF);
> +
> + /* Load values CR93[SWLVL_LOAD] to ’b1 */
> + sw_leveling_load_value;
> +
> + /* Poll CR94[SWLVL_OP_DONE] */
> + sw_leveling_op_done;
> +
> + /*
> + * Read Responses - SWLVL_RESP_0
> + *
> + * The 0x00 (correct response when PHY_RDLVL_EDGE = 0)
> + * -> 1 in the bit vector
> + */
> + swlvl_rsp = (readl(&ddrmr->cr[94]) >> SWLVL_RESP_0) & 0xF;
> + if (swlvl_rsp == 0)
> + generic_set_bit(i, rdlvl_rsp);
> + }
> +
> + bitmap_print(rdlvl_rsp, DDRMC_DQS_DQ_MAX_DELAY);
> +
> + /*
> + * First test for rising edge 0x0 -> 0x1 in bitmap
> + */
> + rdlvl_dl_0_min = ddr_cal_get_first_edge_index(rdlvl_rsp, RISING_EDGE,
> + N_SAMPLES, N_SAMPLES,
> + DDRMC_DQS_DQ_MAX_DELAY);
> +
> + /*
> + * Secondly test for falling edge 0x1 -> 0x0 in bitmap
> + */
> + rdlvl_dl_0_max = ddr_cal_get_first_edge_index(rdlvl_rsp, FALLING_EDGE,
> + N_SAMPLES, rdlvl_dl_0_min,
> + DDRMC_DQS_DQ_MAX_DELAY);
> +
> + debug("RDLVL: DL_0 min: %d [0x%x] DL_0 max: %d [0x%x]\n",
> + rdlvl_dl_0_min, rdlvl_dl_0_min, rdlvl_dl_0_max, rdlvl_dl_0_max);
> + rdlvl_dl_0 = (rdlvl_dl_0_max - rdlvl_dl_0_min) / 2;
> +
> + if (rdlvl_dl_0_max == -1 || rdlvl_dl_0_min == -1 || rdlvl_dl_0 <= 0) {
> + debug("RDLVL: The DQS to DQ delay cannot be found!\n");
> + debug("RDLVL: Using default - slice 0: %d!\n", rdlvl_dl_0_def);
> + rdlvl_dl_0 = rdlvl_dl_0_def;
> + }
> +
> + debug("\nRDLVL: ---> RDLVL_DL_1\n");
> + bitmap_zero(rdlvl_rsp, DDRMC_DQS_DQ_MAX_DELAY + 1);
> +
> + for (i = 0; i <= DDRMC_DQS_DQ_MAX_DELAY; i++) {
> + clrsetbits_le32(&ddrmr->cr[110], 0xFFFF << RDLVL_DL_1_OFF,
> + i << RDLVL_DL_1_OFF);
> +
> + /* Load values CR93[SWLVL_LOAD] to ’b1 */
> + sw_leveling_load_value;
> +
> + /* Poll CR94[SWLVL_OP_DONE] */
> + sw_leveling_op_done;
> +
> + /*
> + * Read Responses - SWLVL_RESP_1
> + *
> + * The 0x00 (correct response when PHY_RDLVL_EDGE = 0)
> + * -> 1 in the bit vector
> + */
> + swlvl_rsp = (readl(&ddrmr->cr[95]) >> SWLVL_RESP_1) & 0xF;
> + if (swlvl_rsp == 0)
> + generic_set_bit(i, rdlvl_rsp);
> + }
> +
> + bitmap_print(rdlvl_rsp, DDRMC_DQS_DQ_MAX_DELAY);
> +
> + /*
> + * First test for rising edge 0x0 -> 0x1 in bitmap
> + */
> + rdlvl_dl_1_min = ddr_cal_get_first_edge_index(rdlvl_rsp, RISING_EDGE,
> + N_SAMPLES, N_SAMPLES,
> + DDRMC_DQS_DQ_MAX_DELAY);
> +
> + /*
> + * Secondly test for falling edge 0x1 -> 0x0 in bitmap
> + */
> + rdlvl_dl_1_max = ddr_cal_get_first_edge_index(rdlvl_rsp, FALLING_EDGE,
> + N_SAMPLES, rdlvl_dl_1_min,
> + DDRMC_DQS_DQ_MAX_DELAY);
> +
> + debug("RDLVL: DL_1 min: %d [0x%x] DL_1 max: %d [0x%x]\n",
> + rdlvl_dl_1_min, rdlvl_dl_1_min, rdlvl_dl_1_max, rdlvl_dl_1_max);
> + rdlvl_dl_1 = (rdlvl_dl_1_max - rdlvl_dl_1_min) / 2;
> +
> + if (rdlvl_dl_1_max == -1 || rdlvl_dl_1_min == -1 || rdlvl_dl_1 <= 0) {
> + debug("RDLVL: The DQS to DQ delay cannot be found!\n");
> + debug("RDLVL: Using default - slice 1: %d!\n", rdlvl_dl_1_def);
> + rdlvl_dl_1 = rdlvl_dl_1_def;
> + }
> +
> + debug("RDLVL: CALIBRATED: rdlvl_dl_0: 0x%x\t rdlvl_dl_1: 0x%x\n",
> + rdlvl_dl_0, rdlvl_dl_1);
> +
> + /* Write new delay values */
> + writel((rdlvl_dl_0 << RDLVL_DL_O_OFF), &ddrmr->cr[105]);
> + writel((rdlvl_dl_1 << RDLVL_DL_1_OFF), &ddrmr->cr[110]);
> +
> + sw_leveling_load_value;
> + sw_leveling_op_done;
> +
> + /* Exit procedure - CR94[SWLVL_EXIT] to ’b1 */
> + sw_leveling_exit;
> +
> + /* Poll CR94[SWLVL_OP_DONE] */
> + sw_leveling_op_done;
> +
> + return 0;
> +}
> +
> +/*
> + * WRLVL_DL calibration:
> + *
> + * For non-flyback memory architecture - where one have a single DDR3 x16
> + * memory - it is NOT necessary to perform "Write Leveling"
> + * [3] 'Vybrid DDR3 write leveling' https://community.nxp.com/thread/429362
> + *
> + */
> +
> +int ddrmc_calibration(struct ddrmr_regs *ddrmr)
> +{
> + ddrmc_cal_dqs_to_dq(ddrmr);
> +
> + return 0;
> +}
> diff --git a/arch/arm/mach-imx/ddrmc-vf610-calibration.h
> b/arch/arm/mach-imx/ddrmc-vf610-calibration.h
> new file mode 100644
> index 0000000000..7bc53236be
> --- /dev/null
> +++ b/arch/arm/mach-imx/ddrmc-vf610-calibration.h
> @@ -0,0 +1,59 @@
> +/* SPDX-License-Identifier: GPL-2.0+ */
> +/*
> + * ddrmc DDR3 calibration code for NXP's VF610
> + *
> + * Copyright (C) 2018 DENX Software Engineering
> + * Lukasz Majewski, DENX Software Engineering, lukma at denx.de
> + *
> + */
> +
> +#ifndef __DDRMC_VF610_CALIBRATOIN_H_
> +#define __DDRMC_VF610_CALIBRATOIN_H_
> +
> +/*
> + * Number of "samples" in the calibration bitmap
> + * to be considered during calibration.
> + */
> +#define N_SAMPLES 3
> +
> +/*
> + * Constants to indicate if we are looking for a rising or
> + * falling edge in the calibration bitmap
> + */
> +enum edge {
> + FALLING_EDGE = 1,
> + RISING_EDGE
> +};
> +
> +/*
> + * The max number of delay elements when DQS to DQ setting
> + */
> +#define DDRMC_DQS_DQ_MAX_DELAY 0xFF
> +
> +/* Bits offsets for DDRMC_CR registers */
> +/* CR101 */
> +#define PHY_RDLVL_EDGE 24
> +/* CR93 */
> +#define SW_LVL_MODE 8
> +/* CR94 */
> +#define SWLVL_RESP_0 24
> +/* CR95 */
> +#define SWLVL_RESP_1 0
> +/* CR105 */
> +#define RDLVL_DL_O_OFF 8
> +/* CR110 */
> +#define RDLVL_DL_1_OFF 0
Should we not move those defines to
arch/arm/include/asm/arch-vf610/imx-regs.h too?
--
Stefan
> +
> +/**
> + * ddrmc_calibration - Vybrid's (VF610) DDR3 calibration code
> + *
> + * This function is calculating proper memory controller values
> + * during run time.
> + *
> + * @param ddrmr_regs - memory controller registers
> + *
> + * @return 0 on success, otherwise error code
> + */
> +int ddrmc_calibration(struct ddrmr_regs *ddrmr);
> +
> +#endif /* __DDRMC_VF610_CALIBRATOIN_H_ */
More information about the U-Boot
mailing list