[U-Boot] [RESEND PATCH V2 3/3] imx: mx7: add system suspend/resume support

Stefan Agner stefan at agner.ch
Wed Aug 8 12:09:09 UTC 2018


On 08.08.2018 11:09, Anson Huang wrote:
> Hi, Stefan
> 
> Anson Huang
> Best Regards!
> 
> 
>> -----Original Message-----
>> From: Stefan Agner [mailto:stefan at agner.ch]
>> Sent: Wednesday, August 8, 2018 5:03 PM
>> To: Anson Huang <anson.huang at nxp.com>
>> Cc: sbabic at denx.de; Fabio Estevam <fabio.estevam at nxp.com>;
>> albert.u.boot at aribaud.net; bryan.odonoghue at linaro.org;
>> uri.mashiach at compulab.co.il; Stefan Agner <stefan.agner at toradex.com>;
>> Peng Fan <peng.fan at nxp.com>; Ye Li <ye.li at nxp.com>; sjg at chromium.org;
>> u-boot at lists.denx.de; dl-linux-imx <linux-imx at nxp.com>
>> Subject: Re: [U-Boot] [RESEND PATCH V2 3/3] imx: mx7: add system
>> suspend/resume support
>>
>> Hi Anson,
>>
>> On 08.08.2018 03:17, Anson Huang wrote:
>> > This patch adds system suspend/resume support, when linux kernel
>> > enters deep sleep mode, SoC will go into below mode:
>> >
>> >  - CA7 platform goes into STOP mode;
>> >  - SoC goes into DSM mode;
>> >  - DDR goes into self-refresh mode;
>> >  - CPU0/SCU will be powered down.
>> >
>> > When wake up event arrives:
>> >
>> >  - SoC DSM mdoe exits;
>> >  - CA7 platform exit STOP mode, SCU/CPU0 power up;
>> >  - Invalidate L1 cache;
>> >  - DDR exit self-refresh mode;
>> >  - Do secure monitor mode related initialization;
>> >  - Jump to linux kernel resume entry.
>> >
>> > Belwo is the log of 1 iteration of system suspend/resume:
>> >
>> > [  338.824862] PM: suspend entry (deep) [  338.828853] PM: Syncing
>> > filesystems ... done.
>> > [  338.834433] Freezing user space processes ... (elapsed 0.001 seconds)
>> done.
>> > [  338.842939] OOM killer disabled.
>> > [  338.846182] Freezing remaining freezable tasks ... (elapsed 0.001
>> > seconds) done.
>> > [  338.869717] PM: suspend devices took 0.010 seconds [  338.877846]
>> > Disabling non-boot CPUs ...
>> > [  338.960301] Retrying again to check for CPU kill [  338.964953]
>> > CPU1 killed.
>> > [  338.968104] Enabling non-boot CPUs ...
>> > [  338.973598] CPU1 is up
>> > [  339.267155] mmc1: queuing unknown CIS tuple 0x80 (2 bytes) [
>> > 339.275833] mmc1: queuing unknown CIS tuple 0x80 (7 bytes) [
>> > 339.284158] mmc1: queuing unknown CIS tuple 0x80 (6 bytes) [
>> > 339.385065] PM: resume devices took 0.400 seconds [  339.389836] OOM
>> > killer enabled.
>> > [  339.392986] Restarting tasks ... done.
>> > [  339.398990] PM: suspend exit
>>
>> Gave this a try with a 3.18-rc6 kernel today on our Colibri iMX7, I could
>> successfully suspend and resume using RTC wakealarm.
>>
>> What is a bit weird is that sometimes the SoC seems to consume more power
>> than other times (sometimes 135mW, sometimes 60mW). Any idea where that
>> could come from?
>>
>> In any case, the system seems to wakeup just fine.
>  
> I ONLY tried the function with latest 4.18 kernel and did NOT measure
> the power, if the power you saw
> is a board level total power, then I have no idea, maybe you can
> disable all peripheral devices
> in dtb to have another try, could be some peripherals' different state
> consuming different
> power, if you can break down the power rails to VDD_ARM, VDD_SOC and VDDA_1P8,
> then I think you should be able to see the benefit, and from each
> power rails' number,
> maybe we can figure out why, but I think this patch set is mainly for
> function level, for
> power number optimization, we can have patch later, I think.

I checked VDD_ARM, it gets turned off properly. But yeah, could
definitely be related to some peripheral. Just wanted to ask you since
you just worked with it.

I agree, this finding should not hold back the merge of this patch set.

> 
> Noted that latest kernel's PCIe driver has resume issue, you have to
> disable PCIe if using 4.18 kernel.

We do not use PCIe by default, so I think it should not have any
influence.

> I suggest you to break down power rails first.

Yeah will have a closer look to all the individual rails. However, due
to other priorities I probably will do this at a later time.

Since suspend/resume does work stable:

Acked-by: Stefan Agner <stefan at agner.ch>
Tested-by: Stefan Agner <stefan at agner.ch>

--
Stefan

> 
> Anson.
>>
>> --
>> Stefan
>>
>> >
>> > The resume entry function has to initialize stack pointer before
>> > calling C code, otherwise there will be an external abort occur, in
>> > additional, invalidate L1 cache must be done in secure section as
>> > well, so this patch also adds assembly code back and keep it as simple as
>> possible.
>> >
>> > Signed-off-by: Anson Huang <Anson.Huang at nxp.com>
>> > ---
>> >  arch/arm/mach-imx/mx7/Makefile       |   2 +-
>> >  arch/arm/mach-imx/mx7/psci-mx7.c     | 449
>> ++++++++++++++++++++++++++++++++++-
>> >  arch/arm/mach-imx/mx7/psci-suspend.S |  67 ++++++
>> >  3 files changed, 515 insertions(+), 3 deletions(-)  create mode
>> > 100644 arch/arm/mach-imx/mx7/psci-suspend.S
>> >
>> > diff --git a/arch/arm/mach-imx/mx7/Makefile
>> > b/arch/arm/mach-imx/mx7/Makefile index 94a0e64..f1436e2 100644
>> > --- a/arch/arm/mach-imx/mx7/Makefile
>> > +++ b/arch/arm/mach-imx/mx7/Makefile
>> > @@ -4,4 +4,4 @@
>> >  #
>> >
>> >  obj-y	:= soc.o clock.o clock_slice.o ddr.o snvs.o
>> > -obj-$(CONFIG_ARMV7_PSCI)  += psci-mx7.o
>> > +obj-$(CONFIG_ARMV7_PSCI)  += psci-mx7.o psci-suspend.o
>> > diff --git a/arch/arm/mach-imx/mx7/psci-mx7.c
>> > b/arch/arm/mach-imx/mx7/psci-mx7.c
>> > index d6c4263..34ba0a9 100644
>> > --- a/arch/arm/mach-imx/mx7/psci-mx7.c
>> > +++ b/arch/arm/mach-imx/mx7/psci-mx7.c
>> > @@ -8,20 +8,77 @@
>> >  #include <asm/psci.h>
>> >  #include <asm/secure.h>
>> >  #include <asm/arch/imx-regs.h>
>> > +#include <asm/armv7.h>
>> > +#include <asm/gic.h>
>> >  #include <linux/bitops.h>
>> >  #include <common.h>
>> >  #include <fsl_wdog.h>
>> >
>> > -#define GPC_CPU_PGC_SW_PDN_REQ	0xfc
>> > +#define GPC_LPCR_A7_BSC	0x0
>> > +#define GPC_LPCR_A7_AD		0x4
>> > +#define GPC_SLPCR		0x14
>> > +#define GPC_PGC_ACK_SEL_A7	0x24
>> > +#define GPC_IMR1_CORE0		0x30
>> > +#define GPC_SLOT0_CFG		0xb0
>> >  #define GPC_CPU_PGC_SW_PUP_REQ	0xf0
>> > +#define GPC_CPU_PGC_SW_PDN_REQ	0xfc
>> > +#define GPC_PGC_C0		0x800
>> >  #define GPC_PGC_C0		0x800
>> >  #define GPC_PGC_C1		0x840
>> > +#define GPC_PGC_SCU		0x880
>> > +
>> > +#define BM_LPCR_A7_BSC_CPU_CLK_ON_LPM		0x4000
>> > +#define BM_LPCR_A7_BSC_LPM1			0xc
>> > +#define BM_LPCR_A7_BSC_LPM0			0x3
>> > +#define BP_LPCR_A7_BSC_LPM0			0
>> > +#define BM_SLPCR_EN_DSM				0x80000000
>> > +#define BM_SLPCR_RBC_EN				0x40000000
>> > +#define BM_SLPCR_REG_BYPASS_COUNT		0x3f000000
>> > +#define BM_SLPCR_VSTBY				0x4
>> > +#define BM_SLPCR_SBYOS				0x2
>> > +#define BM_SLPCR_BYPASS_PMIC_READY		0x1
>> > +#define BM_LPCR_A7_AD_L2PGE			0x10000
>> > +#define BM_LPCR_A7_AD_EN_C1_PUP			0x800
>> > +#define BM_LPCR_A7_AD_EN_C0_PUP			0x200
>> > +#define BM_LPCR_A7_AD_EN_PLAT_PDN		0x10
>> > +#define BM_LPCR_A7_AD_EN_C1_PDN			0x8
>> > +#define BM_LPCR_A7_AD_EN_C0_PDN			0x2
>> >
>> >  #define BM_CPU_PGC_SW_PDN_PUP_REQ_CORE0_A7	0x1
>> >  #define BM_CPU_PGC_SW_PDN_PUP_REQ_CORE1_A7	0x2
>> >
>> > -/* below is for i.MX7D */
>> > +#define BM_GPC_PGC_ACK_SEL_A7_PD_DUMMY_ACK	0x8000
>> > +#define BM_GPC_PGC_ACK_SEL_A7_PU_DUMMY_ACK	0x80000000
>> > +
>> > +#define MAX_SLOT_NUMBER				10
>> > +#define A7_LPM_WAIT				0x5
>> > +#define A7_LPM_STOP				0xa
>> > +
>> > +#define BM_SYS_COUNTER_CNTCR_FCR1 0x200 #define
>> > +BM_SYS_COUNTER_CNTCR_FCR0 0x100
>> > +
>> > +#define REG_SET		0x4
>> > +#define REG_CLR		0x8
>> > +
>> > +#define ANADIG_ARM_PLL		0x60
>> > +#define ANADIG_DDR_PLL		0x70
>> > +#define ANADIG_SYS_PLL		0xb0
>> > +#define ANADIG_ENET_PLL		0xe0
>> > +#define ANADIG_AUDIO_PLL	0xf0
>> > +#define ANADIG_VIDEO_PLL	0x130
>> > +#define BM_ANATOP_ARM_PLL_OVERRIDE	BIT(20)
>> > +#define BM_ANATOP_DDR_PLL_OVERRIDE	BIT(19)
>> > +#define BM_ANATOP_SYS_PLL_OVERRIDE	(0x1ff << 17)
>> > +#define BM_ANATOP_ENET_PLL_OVERRIDE	BIT(13)
>> > +#define BM_ANATOP_AUDIO_PLL_OVERRIDE	BIT(24)
>> > +#define BM_ANATOP_VIDEO_PLL_OVERRIDE	BIT(24)
>> > +
>> > +#define DDRC_STAT	0x4
>> > +#define DDRC_PWRCTL	0x30
>> > +#define DDRC_PSTAT	0x3fc
>> > +
>> >  #define SRC_GPR1_MX7D		0x074
>> > +#define SRC_GPR2_MX7D		0x078
>> >  #define SRC_A7RCR0		0x004
>> >  #define SRC_A7RCR1		0x008
>> >
>> > @@ -56,6 +113,27 @@ u8 psci_state[IMX7D_PSCI_NR_CPUS]
>> __secure_data = {
>> >  	 PSCI_AFFINITY_LEVEL_ON,
>> >  	 PSCI_AFFINITY_LEVEL_OFF};
>> >
>> > +enum imx_gpc_slot {
>> > +	CORE0_A7,
>> > +	CORE1_A7,
>> > +	SCU_A7,
>> > +	FAST_MEGA_MIX,
>> > +	MIPI_PHY,
>> > +	PCIE_PHY,
>> > +	USB_OTG1_PHY,
>> > +	USB_OTG2_PHY,
>> > +	USB_HSIC_PHY,
>> > +	CORE0_M4,
>> > +};
>> > +
>> > +enum mxc_cpu_pwr_mode {
>> > +	RUN,
>> > +	WAIT,
>> > +	STOP,
>> > +};
>> > +
>> > +extern void psci_system_resume(void);
>> > +
>> >  static inline void psci_set_state(int cpu, u8 state)  {
>> >  	psci_state[cpu] = state;
>> > @@ -237,7 +315,374 @@ __secure s32 psci_features(u32 __always_unused
>> > function_id, u32 psci_fid)
>> >  	case ARM_PSCI_0_2_FN_SYSTEM_OFF:
>> >  	case ARM_PSCI_0_2_FN_SYSTEM_RESET:
>> >  	case ARM_PSCI_1_0_FN_PSCI_FEATURES:
>> > +	case ARM_PSCI_1_0_FN_SYSTEM_SUSPEND:
>> >  		return 0x0;
>> >  	}
>> >  	return ARM_PSCI_RET_NI;
>> >  }
>> > +
>> > +static __secure void imx_gpcv2_set_lpm_mode(enum mxc_cpu_pwr_mode
>> > +mode) {
>> > +	u32 val1, val2, val3;
>> > +
>> > +	val1 = readl(GPC_IPS_BASE_ADDR + GPC_LPCR_A7_BSC);
>> > +	val2 = readl(GPC_IPS_BASE_ADDR + GPC_SLPCR);
>> > +
>> > +	/* all cores' LPM settings must be same */
>> > +	val1 &= ~(BM_LPCR_A7_BSC_LPM0 | BM_LPCR_A7_BSC_LPM1);
>> > +	val1 |= BM_LPCR_A7_BSC_CPU_CLK_ON_LPM;
>> > +
>> > +	val2 &= ~(BM_SLPCR_EN_DSM | BM_SLPCR_VSTBY | BM_SLPCR_RBC_EN
>> |
>> > +		BM_SLPCR_SBYOS | BM_SLPCR_BYPASS_PMIC_READY);
>> > +	/*
>> > +	 * GPC: When improper low-power sequence is used,
>> > +	 * the SoC enters low power mode before the ARM core executes WFI.
>> > +	 *
>> > +	 * Software workaround:
>> > +	 * 1) Software should trigger IRQ #32 (IOMUX) to be always pending
>> > +	 *    by setting IOMUX_GPR1_IRQ.
>> > +	 * 2) Software should then unmask IRQ #32 in GPC before setting GPC
>> > +	 *    Low-Power mode.
>> > +	 * 3) Software should mask IRQ #32 right after GPC Low-Power mode
>> > +	 *    is set.
>> > +	 */
>> > +	switch (mode) {
>> > +	case RUN:
>> > +		val3 = readl(GPC_IPS_BASE_ADDR + GPC_IMR1_CORE0);
>> > +		val3 &= ~0x1;
>> > +		writel(val3, GPC_IPS_BASE_ADDR + GPC_IMR1_CORE0);
>> > +		break;
>> > +	case WAIT:
>> > +		val1 |= A7_LPM_WAIT << BP_LPCR_A7_BSC_LPM0;
>> > +		val1 &= ~BM_LPCR_A7_BSC_CPU_CLK_ON_LPM;
>> > +		val3 = readl(GPC_IPS_BASE_ADDR + GPC_IMR1_CORE0);
>> > +		val3 &= ~0x1;
>> > +		writel(val3, GPC_IPS_BASE_ADDR + GPC_IMR1_CORE0);
>> > +		break;
>> > +	case STOP:
>> > +		val1 |= A7_LPM_STOP << BP_LPCR_A7_BSC_LPM0;
>> > +		val1 &= ~BM_LPCR_A7_BSC_CPU_CLK_ON_LPM;
>> > +		val2 |= BM_SLPCR_EN_DSM;
>> > +		val2 |= BM_SLPCR_SBYOS;
>> > +		val2 |= BM_SLPCR_VSTBY;
>> > +		val2 |= BM_SLPCR_BYPASS_PMIC_READY;
>> > +		val3 = readl(GPC_IPS_BASE_ADDR + GPC_IMR1_CORE0);
>> > +		val3 |= 0x1;
>> > +		writel(val3, GPC_IPS_BASE_ADDR + GPC_IMR1_CORE0);
>> > +		break;
>> > +	default:
>> > +		return;
>> > +	}
>> > +	writel(val1, GPC_IPS_BASE_ADDR + GPC_LPCR_A7_BSC);
>> > +	writel(val2, GPC_IPS_BASE_ADDR + GPC_SLPCR); }
>> > +
>> > +static __secure void imx_gpcv2_set_plat_power_gate_by_lpm(bool pdn) {
>> > +	u32 val = readl(GPC_IPS_BASE_ADDR + GPC_LPCR_A7_AD);
>> > +
>> > +	val &= ~(BM_LPCR_A7_AD_EN_PLAT_PDN | BM_LPCR_A7_AD_L2PGE);
>> > +	if (pdn)
>> > +		val |= BM_LPCR_A7_AD_EN_PLAT_PDN | BM_LPCR_A7_AD_L2PGE;
>> > +
>> > +	writel(val, GPC_IPS_BASE_ADDR + GPC_LPCR_A7_AD); }
>> > +
>> > +static __secure void imx_gpcv2_set_cpu_power_gate_by_lpm(u32 cpu,
>> > +bool pdn) {
>> > +	u32 val;
>> > +
>> > +	val = readl(GPC_IPS_BASE_ADDR + GPC_LPCR_A7_AD);
>> > +	if (cpu == 0) {
>> > +		if (pdn)
>> > +			val |= BM_LPCR_A7_AD_EN_C0_PDN |
>> > +				BM_LPCR_A7_AD_EN_C0_PUP;
>> > +		else
>> > +			val &= ~(BM_LPCR_A7_AD_EN_C0_PDN |
>> > +				BM_LPCR_A7_AD_EN_C0_PUP);
>> > +	}
>> > +	if (cpu == 1) {
>> > +		if (pdn)
>> > +			val |= BM_LPCR_A7_AD_EN_C1_PDN |
>> > +				BM_LPCR_A7_AD_EN_C1_PUP;
>> > +		else
>> > +			val &= ~(BM_LPCR_A7_AD_EN_C1_PDN |
>> > +				BM_LPCR_A7_AD_EN_C1_PUP);
>> > +	}
>> > +	writel(val, GPC_IPS_BASE_ADDR + GPC_LPCR_A7_AD); }
>> > +
>> > +static __secure void imx_gpcv2_set_slot_ack(u32 index, enum
>> > imx_gpc_slot m_core,
>> > +					    bool mode, bool ack)
>> > +{
>> > +	u32 val;
>> > +
>> > +	if (index >= MAX_SLOT_NUMBER)
>> > +		return;
>> > +
>> > +	/* set slot */
>> > +	writel(readl(GPC_IPS_BASE_ADDR + GPC_SLOT0_CFG + index * 4) |
>> > +		((mode + 1) << (m_core * 2)),
>> > +		GPC_IPS_BASE_ADDR + GPC_SLOT0_CFG + index * 4);
>> > +
>> > +	if (ack) {
>> > +		/* set ack */
>> > +		val = readl(GPC_IPS_BASE_ADDR + GPC_PGC_ACK_SEL_A7);
>> > +		/* clear dummy ack */
>> > +		val &= ~(mode ? BM_GPC_PGC_ACK_SEL_A7_PU_DUMMY_ACK :
>> > +			BM_GPC_PGC_ACK_SEL_A7_PD_DUMMY_ACK);
>> > +		val |= 1 << (m_core + (mode ? 16 : 0));
>> > +		writel(val, GPC_IPS_BASE_ADDR + GPC_PGC_ACK_SEL_A7);
>> > +	}
>> > +}
>> > +
>> > +static __secure void imx_system_counter_resume(void) {
>> > +	u32 val;
>> > +
>> > +	val = readl(SYSCNT_CTRL_IPS_BASE_ADDR);
>> > +	val &= ~BM_SYS_COUNTER_CNTCR_FCR1;
>> > +	val |= BM_SYS_COUNTER_CNTCR_FCR0;
>> > +	writel(val, SYSCNT_CTRL_IPS_BASE_ADDR); }
>> > +
>> > +static __secure void imx_system_counter_suspend(void) {
>> > +	u32 val;
>> > +
>> > +	val = readl(SYSCNT_CTRL_IPS_BASE_ADDR);
>> > +	val &= ~BM_SYS_COUNTER_CNTCR_FCR0;
>> > +	val |= BM_SYS_COUNTER_CNTCR_FCR1;
>> > +	writel(val, SYSCNT_CTRL_IPS_BASE_ADDR); }
>> > +
>> > +static __secure void gic_resume(void) {
>> > +	u32 itlinesnr, i;
>> > +	u32 gic_dist_addr = GIC400_ARB_BASE_ADDR + GIC_DIST_OFFSET;
>> > +
>> > +	/* enable the GIC distributor */
>> > +	writel(readl(gic_dist_addr + GICD_CTLR) | 0x03,
>> > +	       gic_dist_addr + GICD_CTLR);
>> > +
>> > +	/* TYPER[4:0] contains an encoded number of available interrupts */
>> > +	itlinesnr = readl(gic_dist_addr + GICD_TYPER) & 0x1f;
>> > +
>> > +	/* set all bits in the GIC group registers to one to allow access
>> > +	 * from non-secure state. The first 32 interrupts are private per
>> > +	 * CPU and will be set later when enabling the GIC for each core
>> > +	 */
>> > +	for (i = 1; i <= itlinesnr; i++)
>> > +		writel((u32)-1, gic_dist_addr + GICD_IGROUPRn + 4 * i); }
>> > +
>> > +static inline void imx_pll_suspend(void) {
>> > +	writel(BM_ANATOP_ARM_PLL_OVERRIDE,
>> > +	       ANATOP_BASE_ADDR + ANADIG_ARM_PLL + REG_SET);
>> > +	writel(BM_ANATOP_DDR_PLL_OVERRIDE,
>> > +	       ANATOP_BASE_ADDR + ANADIG_DDR_PLL + REG_SET);
>> > +	writel(BM_ANATOP_SYS_PLL_OVERRIDE,
>> > +	       ANATOP_BASE_ADDR + ANADIG_SYS_PLL + REG_SET);
>> > +	writel(BM_ANATOP_ENET_PLL_OVERRIDE,
>> > +	       ANATOP_BASE_ADDR + ANADIG_ENET_PLL + REG_SET);
>> > +	writel(BM_ANATOP_AUDIO_PLL_OVERRIDE,
>> > +	       ANATOP_BASE_ADDR + ANADIG_AUDIO_PLL + REG_SET);
>> > +	writel(BM_ANATOP_VIDEO_PLL_OVERRIDE,
>> > +	       ANATOP_BASE_ADDR + ANADIG_VIDEO_PLL + REG_SET); }
>> > +
>> > +static inline void imx_pll_resume(void) {
>> > +	writel(BM_ANATOP_ARM_PLL_OVERRIDE,
>> > +	       ANATOP_BASE_ADDR + ANADIG_ARM_PLL + REG_CLR);
>> > +	writel(BM_ANATOP_DDR_PLL_OVERRIDE,
>> > +	       ANATOP_BASE_ADDR + ANADIG_DDR_PLL + REG_CLR);
>> > +	writel(BM_ANATOP_SYS_PLL_OVERRIDE,
>> > +	       ANATOP_BASE_ADDR + ANADIG_SYS_PLL + REG_CLR);
>> > +	writel(BM_ANATOP_ENET_PLL_OVERRIDE,
>> > +	       ANATOP_BASE_ADDR + ANADIG_ENET_PLL + REG_CLR);
>> > +	writel(BM_ANATOP_AUDIO_PLL_OVERRIDE,
>> > +	       ANATOP_BASE_ADDR + ANADIG_AUDIO_PLL + REG_CLR);
>> > +	writel(BM_ANATOP_VIDEO_PLL_OVERRIDE,
>> > +	       ANATOP_BASE_ADDR + ANADIG_VIDEO_PLL + REG_CLR); }
>> > +
>> > +static inline void imx_udelay(u32 usec) {
>> > +	u32 freq;
>> > +	u64 start, end;
>> > +
>> > +	asm volatile("mrc p15, 0, %0, c14, c0, 0" : "=r" (freq));
>> > +	asm volatile("mrrc p15, 0, %Q0, %R0, c14" : "=r" (start));
>> > +	do {
>> > +		asm volatile("mrrc p15, 0, %Q0, %R0, c14" : "=r" (end));
>> > +		if ((end - start) > usec * (freq / 1000000))
>> > +			break;
>> > +	} while (1);
>> > +}
>> > +
>> > +static inline void imx_ddrc_enter_self_refresh(void) {
>> > +	writel(0, DDRC_IPS_BASE_ADDR + DDRC_PWRCTL);
>> > +	while (readl(DDRC_IPS_BASE_ADDR + DDRC_PSTAT) & 0x10001)
>> > +		;
>> > +
>> > +	writel(0x20, DDRC_IPS_BASE_ADDR + DDRC_PWRCTL);
>> > +	while ((readl(DDRC_IPS_BASE_ADDR + DDRC_STAT) & 0x23) != 0x23)
>> > +		;
>> > +	writel(readl(DDRC_IPS_BASE_ADDR + DDRC_PWRCTL) | 0x8,
>> > +	       DDRC_IPS_BASE_ADDR + DDRC_PWRCTL); }
>> > +
>> > +static inline void imx_ddrc_exit_self_refresh(void) {
>> > +	writel(0, DDRC_IPS_BASE_ADDR + DDRC_PWRCTL);
>> > +	while ((readl(DDRC_IPS_BASE_ADDR + DDRC_STAT) & 0x3) == 0x3)
>> > +		;
>> > +	writel(readl(DDRC_IPS_BASE_ADDR + DDRC_PWRCTL) | 0x1,
>> > +	       DDRC_IPS_BASE_ADDR + DDRC_PWRCTL); }
>> > +
>> > +__secure void imx_system_resume(void) {
>> > +	unsigned int i, val, imr[4], entry;
>> > +
>> > +	entry = psci_get_target_pc(0);
>> > +	imx_ddrc_exit_self_refresh();
>> > +	imx_system_counter_resume();
>> > +	imx_gpcv2_set_lpm_mode(RUN);
>> > +	imx_gpcv2_set_cpu_power_gate_by_lpm(0, false);
>> > +	imx_gpcv2_set_plat_power_gate_by_lpm(false);
>> > +	imx_gpcv2_set_m_core_pgc(false, GPC_PGC_C0);
>> > +	imx_gpcv2_set_m_core_pgc(false, GPC_PGC_SCU);
>> > +
>> > +	/*
>> > +	 * need to mask all interrupts in GPC before
>> > +	 * operating RBC configurations
>> > +	 */
>> > +	for (i = 0; i < 4; i++) {
>> > +		imr[i] = readl(GPC_IPS_BASE_ADDR + GPC_IMR1_CORE0 + i * 4);
>> > +		writel(~0, GPC_IPS_BASE_ADDR + GPC_IMR1_CORE0 + i * 4);
>> > +	}
>> > +
>> > +	/* configure RBC enable bit */
>> > +	val = readl(GPC_IPS_BASE_ADDR + GPC_SLPCR);
>> > +	val &= ~BM_SLPCR_RBC_EN;
>> > +	writel(val, GPC_IPS_BASE_ADDR + GPC_SLPCR);
>> > +
>> > +	/* configure RBC count */
>> > +	val = readl(GPC_IPS_BASE_ADDR + GPC_SLPCR);
>> > +	val &= ~BM_SLPCR_REG_BYPASS_COUNT;
>> > +	writel(val, GPC_IPS_BASE_ADDR + GPC_SLPCR);
>> > +
>> > +	/*
>> > +	 * need to delay at least 2 cycles of CKIL(32K)
>> > +	 * due to hardware design requirement, which is
>> > +	 * ~61us, here we use 65us for safe
>> > +	 */
>> > +	imx_udelay(65);
>> > +
>> > +	/* restore GPC interrupt mask settings */
>> > +	for (i = 0; i < 4; i++)
>> > +		writel(imr[i], GPC_IPS_BASE_ADDR + GPC_IMR1_CORE0 + i * 4);
>> > +
>> > +	/* initialize gic distributor */
>> > +	gic_resume();
>> > +	_nonsec_init();
>> > +
>> > +	/* save cpu0 entry */
>> > +	psci_save(0, entry, 0);
>> > +	psci_cpu_entry();
>> > +}
>> > +
>> > +__secure void psci_system_suspend(u32 __always_unused function_id,
>> > +				  u32 ep, u32 context_id)
>> > +{
>> > +	u32 gpc_mask[4];
>> > +	u32 i, val;
>> > +
>> > +	psci_save(0, ep, context_id);
>> > +	/* overwrite PLL to be controlled by low power mode */
>> > +	imx_pll_suspend();
>> > +	imx_system_counter_suspend();
>> > +	/* set CA7 platform to enter STOP mode */
>> > +	imx_gpcv2_set_lpm_mode(STOP);
>> > +	/* enable core0/scu power down/up with low power mode */
>> > +	imx_gpcv2_set_cpu_power_gate_by_lpm(0, true);
>> > +	imx_gpcv2_set_plat_power_gate_by_lpm(true);
>> > +	/* time slot settings for core0 and scu */
>> > +	imx_gpcv2_set_slot_ack(0, CORE0_A7, false, false);
>> > +	imx_gpcv2_set_slot_ack(1, SCU_A7, false, true);
>> > +	imx_gpcv2_set_slot_ack(5, SCU_A7, true, false);
>> > +	imx_gpcv2_set_slot_ack(6, CORE0_A7, true, true);
>> > +	imx_gpcv2_set_m_core_pgc(true, GPC_PGC_C0);
>> > +	imx_gpcv2_set_m_core_pgc(true, GPC_PGC_SCU);
>> > +	psci_v7_flush_dcache_all();
>> > +
>> > +	imx_ddrc_enter_self_refresh();
>> > +
>> > +	/*
>> > +	 * e10133: ARM: Boot failure after A7 enters into
>> > +	 * low-power idle mode
>> > +	 *
>> > +	 * Workaround:
>> > +	 * If both CPU0/CPU1 are IDLE, the last IDLE CPU should
>> > +	 * disable GIC first, then REG_BYPASS_COUNTER is used
>> > +	 * to mask wakeup INT, and then execute “wfi” is used to
>> > +	 * bring the system into power down processing safely.
>> > +	 * The counter must be enabled as close to the “wfi” state
>> > +	 * as possible. The following equation can be used to
>> > +	 * determine the RBC counter value:
>> > +	 * RBC_COUNT * (1/32K RTC frequency) >=
>> > +	 * (46 + PDNSCR_SW + PDNSCR_SW2ISO ) ( 1/IPG_CLK frequency ).
>> > +	 */
>> > +
>> > +	/* disable GIC distributor */
>> > +	writel(0, GIC400_ARB_BASE_ADDR + GIC_DIST_OFFSET);
>> > +
>> > +	for (i = 0; i < 4; i++)
>> > +		gpc_mask[i] = readl(GPC_IPS_BASE_ADDR + GPC_IMR1_CORE0 + i *
>> 4);
>> > +
>> > +	/*
>> > +	 * enable the RBC bypass counter here
>> > +	 * to hold off the interrupts. RBC counter
>> > +	 * = 8 (240us). With this setting, the latency
>> > +	 * from wakeup interrupt to ARM power up
>> > +	 * is ~250uS.
>> > +	 */
>> > +	val = readl(GPC_IPS_BASE_ADDR + GPC_SLPCR);
>> > +	val &= ~(0x3f << 24);
>> > +	val |= (0x8 << 24);
>> > +	writel(val, GPC_IPS_BASE_ADDR + GPC_SLPCR);
>> > +
>> > +	/* enable the counter. */
>> > +	val = readl(GPC_IPS_BASE_ADDR + GPC_SLPCR);
>> > +	val |= (1 << 30);
>> > +	writel(val, GPC_IPS_BASE_ADDR + GPC_SLPCR);
>> > +
>> > +	/* unmask all the GPC interrupts. */
>> > +	for (i = 0; i < 4; i++)
>> > +		writel(gpc_mask[i], GPC_IPS_BASE_ADDR + GPC_IMR1_CORE0 + i *
>> 4);
>> > +
>> > +	/*
>> > +	 * now delay for a short while (3usec)
>> > +	 * ARM is at 1GHz at this point
>> > +	 * so a short loop should be enough.
>> > +	 * this delay is required to ensure that
>> > +	 * the RBC counter can start counting in
>> > +	 * case an interrupt is already pending
>> > +	 * or in case an interrupt arrives just
>> > +	 * as ARM is about to assert DSM_request.
>> > +	 */
>> > +	imx_udelay(3);
>> > +
>> > +	/* save resume entry and sp in CPU0 GPR registers */
>> > +	asm volatile("mov %0, sp" : "=r" (val));
>> > +	writel((u32)psci_system_resume, SRC_BASE_ADDR + SRC_GPR1_MX7D);
>> > +	writel(val, SRC_BASE_ADDR + SRC_GPR2_MX7D);
>> > +
>> > +	/* sleep */
>> > +	while (1)
>> > +		wfi();
>> > +}
>> > diff --git a/arch/arm/mach-imx/mx7/psci-suspend.S
>> > b/arch/arm/mach-imx/mx7/psci-suspend.S
>> > new file mode 100644
>> > index 0000000..a21403f
>> > --- /dev/null
>> > +++ b/arch/arm/mach-imx/mx7/psci-suspend.S
>> > @@ -0,0 +1,67 @@
>> > +/* SPDX-License-Identifier: GPL-2.0+ */
>> > +/*
>> > + * Copyright 2018 NXP
>> > + */
>> > +
>> > +#include <config.h>
>> > +#include <linux/linkage.h>
>> > +
>> > +#include <asm/armv7.h>
>> > +#include <asm/psci.h>
>> > +
>> > +	.pushsection ._secure.text, "ax"
>> > +
>> > +	.arch_extension sec
>> > +
>> > +.globl v7_invalidate_l1
>> > +v7_invalidate_l1:
>> > +	mov	r0, #0
>> > +	mcr	p15, 2, r0, c0, c0, 0
>> > +	mrc	p15, 1, r0, c0, c0, 0
>> > +
>> > +	movw	r1, #0x7fff
>> > +	and	r2, r1, r0, lsr #13
>> > +
>> > +	movw	r1, #0x3ff
>> > +
>> > +	and	r3, r1, r0, lsr #3      @ NumWays - 1
>> > +	add	r2, r2, #1              @ NumSets
>> > +
>> > +	and	r0, r0, #0x7
>> > +	add	r0, r0, #4      @ SetShift
>> > +
>> > +	clz	r1, r3          @ WayShift
>> > +	add	r4, r3, #1      @ NumWays
>> > +1:
>> > +	sub	r2, r2, #1      @ NumSets--
>> > +	mov	r3, r4          @ Temp = NumWays
>> > +2:
>> > +	subs	r3, r3, #1      @ Temp--
>> > +	mov	r5, r3, lsl r1
>> > +	mov	r6, r2, lsl r0
>> > +	orr	r5, r5, r6      @ Reg = (Temp<<WayShift)|(NumSets<<SetShift)
>> > +	mcr	p15, 0, r5, c7, c6, 2
>> > +	bgt	2b
>> > +	cmp	r2, #0
>> > +	bgt	1b
>> > +	dsb	st
>> > +	isb
>> > +	mov	pc, lr
>> > +
>> > +.globl psci_system_resume
>> > +psci_system_resume:
>> > +	mov	sp, r0
>> > +
>> > +	/* invalidate L1 I-cache first */
>> > +	mov	r6, #0x0
>> > +	mcr	p15, 0, r6, c7, c5, 0
>> > +	mcr	p15, 0, r6, c7, c5, 6
>> > +	/* enable the Icache and branch prediction */
>> > +	mov	r6, #0x1800
>> > +	mcr	p15, 0, r6, c1, c0, 0
>> > +	isb
>> > +
>> > +	bl	v7_invalidate_l1
>> > +	b	imx_system_resume
>> > +
>> > +	.popsection


More information about the U-Boot mailing list