[PATCH v5 2/3] arm: exynos5: Add Suspend-to-RAM support for 5420
Tomasz Figa
t.figa at samsung.com
Mon Jun 30 10:57:25 PDT 2014
Hi Vikas, Abhilash,
Please see my comments inline.
On 26.06.2014 13:12, Vikas Sajjan wrote:
> From: Abhilash Kesavan <a.kesavan at samsung.com>
>
> Adds Suspend-to-RAM support for EXYNOS5420
>
> Signed-off-by: Abhilash Kesavan <a.kesavan at samsung.com>
> Signed-off-by: Vikas Sajjan <vikas.sajjan at samsung.com>
> ---
> arch/arm/mach-exynos/pm.c | 150 ++++++++++++++++++++++++++++++++++++++++-----
> 1 file changed, 134 insertions(+), 16 deletions(-)
>
> diff --git a/arch/arm/mach-exynos/pm.c b/arch/arm/mach-exynos/pm.c
> index de61d48..bf8564a 100644
> --- a/arch/arm/mach-exynos/pm.c
> +++ b/arch/arm/mach-exynos/pm.c
> @@ -38,9 +38,13 @@
> #include "regs-pmu.h"
> #include "regs-sys.h"
>
> +#define EXYNOS5420_CPU_STATE 0x28
> +
> #define pmu_raw_writel(val, offset) \
> __raw_writel(val, pmu_base_addr + offset)
>
> +static int exynos5420_cpu_state;
> +
> /**
> * struct exynos_wkup_irq - Exynos GIC to PMU IRQ mapping
> * @hwirq: Hardware IRQ signal of the GIC
> @@ -64,6 +68,10 @@ static struct sleep_save exynos_core_save[] = {
> SAVE_ITEM(S5P_SROM_BC3),
> };
>
> +static struct sleep_save exynos5420_pmu_reg_save[] = {
> + SAVE_ITEM((void __iomem *)S5P_PMU_SPARE3),
> +};
Do you need a whole array for this single register?
> +
> /*
> * GIC wake-up support
> */
> @@ -86,7 +94,7 @@ static int exynos_irq_set_wake(struct irq_data *data, unsigned int state)
> {
> const struct exynos_wkup_irq *wkup_irq;
>
> - if (soc_is_exynos5250())
> + if (soc_is_exynos5250() || soc_is_exynos5420())
You should rework this to eliminate the need to use any SoC-specific
checks. For example:
1) create a struct where any SoC/family specific data are stored, e.g.
struct exynos_pm_data {
const struct exynos_wkup_irq *wkup_irq;
/* arrays, flags, values, function pointers, etc. */
};
2) describe supported variants using instances of this struct, e.g.
static const struct exynos_pm_data exynos5250_pm_data {
.wkup_irq = exynos5250_wkup_irq,
/* ... */
};
static const struct exynos_pm_data exynos5420_pm_data {
.wkup_irq = exynos5250_wkup_irq,
/* ... */
};
3) put pointers to those structs in DT match table:
static const struct of_device_id exynos_pm_matches[] = {
{ .compatible = "samsung,exynos5250-pmu",
.data = &exynos5250_pm_data },
{ .compatible = "samsung,exynos5420-pmu",
.data = &exynos5420_pm_data },
};
4) find a matching node in DT and use the struct pointed by match data.
Also certain checks could probably be replaced with non-SoC-specific
checks, e.g. by CPU part checks.
> wkup_irq = exynos5250_wkup_irq;
> else
> wkup_irq = exynos4_wkup_irq;
> @@ -250,7 +258,16 @@ static int exynos_cpu_suspend(unsigned long arg)
> outer_flush_all();
> #endif
>
> - if (soc_is_exynos5250())
> + /*
> + * Clear sysram register for cpu state so that primary CPU does
> + * not enter low power start in U-Boot.
> + * This is specific to exynos5420 SoC only.
> + */
> + if (soc_is_exynos5420())
> + __raw_writel(0x0,
> + sysram_base_addr + EXYNOS5420_CPU_STATE);
> +
> + if (soc_is_exynos5250() || soc_is_exynos5420())
This probably can be replaced with a check for Cortex A15 or A7.
> flush_cache_all();
>
> /* issue the standby signal into the pm unit. */
> @@ -276,6 +293,22 @@ static void exynos_pm_prepare(void)
> tmp = __raw_readl(pmu_base_addr + EXYNOS5_JPEG_MEM_OPTION);
> tmp &= ~EXYNOS5_OPTION_USE_RETENTION;
> pmu_raw_writel(tmp, EXYNOS5_JPEG_MEM_OPTION);
> + } else if (soc_is_exynos5420()) {
> + unsigned int i;
> +
> + for (i = 0; i < ARRAY_SIZE(exynos5420_pmu_reg_save); i++)
> + exynos5420_pmu_reg_save[i].val =
> + __raw_readl(pmu_base_addr +
> + (unsigned int)exynos5420_pmu_reg_save[i].reg);
> + /*
> + * The cpu state needs to be saved and restored so that the
> + * secondary CPUs will enter low power start. Though the U-Boot
> + * is setting the cpu state with low power flag, the kernel
> + * needs to restore it back in case, the primary cpu fails to
> + * suspend for any reason.
> + */
> + exynos5420_cpu_state =
> + __raw_readl(sysram_base_addr + EXYNOS5420_CPU_STATE);
> }
>
> /* Set value of power down register for sleep mode */
> @@ -286,6 +319,27 @@ static void exynos_pm_prepare(void)
> /* ensure at least INFORM0 has the resume address */
>
> pmu_raw_writel(virt_to_phys(exynos_cpu_resume), S5P_INFORM0);
> +
> + if (soc_is_exynos5420()) {
> + tmp = __raw_readl(pmu_base_addr + EXYNOS5_ARM_L2_OPTION);
> + tmp &= ~EXYNOS5_USE_RETENTION;
> + pmu_raw_writel(tmp, EXYNOS5_ARM_L2_OPTION);
> +
> + tmp = __raw_readl(pmu_base_addr + EXYNOS5420_SFR_AXI_CGDIS1);
> + tmp |= EXYNOS5420_UFS;
> + pmu_raw_writel(tmp, EXYNOS5420_SFR_AXI_CGDIS1);
> +
> + tmp = __raw_readl(pmu_base_addr +
> + EXYNOS5420_ARM_COMMON_OPTION);
> + tmp &= ~EXYNOS5420_L2RSTDISABLE_VALUE;
> + pmu_raw_writel(tmp, EXYNOS5420_ARM_COMMON_OPTION);
nit: A blank line here could improve readability.
> + tmp = __raw_readl(pmu_base_addr + EXYNOS5420_FSYS2_OPTION);
> + tmp |= EXYNOS5420_EMULATION;
> + pmu_raw_writel(tmp, EXYNOS5420_FSYS2_OPTION);
nit: A blank line here could improve readability.
> + tmp = __raw_readl(pmu_base_addr + EXYNOS5420_PSGEN_OPTION);
> + tmp |= EXYNOS5420_EMULATION;
> + pmu_raw_writel(tmp, EXYNOS5420_PSGEN_OPTION);
> + }
> }
>
> static void exynos_pm_central_suspend(void)
> @@ -301,13 +355,24 @@ static void exynos_pm_central_suspend(void)
> static int exynos_pm_suspend(void)
> {
> unsigned long tmp;
> + unsigned int this_cluster;
nit: Two spaces between "int" and "this_cluster". Also it might be a
better idea to use explicitly sized types with same size as register width.
>
> exynos_pm_central_suspend();
>
> /* Setting SEQ_OPTION register */
>
> - tmp = (S5P_USE_STANDBY_WFI0 | S5P_USE_STANDBY_WFE0);
> - pmu_raw_writel(tmp, S5P_CENTRAL_SEQ_OPTION);
> + if (soc_is_exynos5420()) {
I believe this is not Exynos5420-specific, but rather specific to any
multi-cluster Exynos SoC. I think there might be a way to check the
number of clusters.
> + this_cluster = MPIDR_AFFINITY_LEVEL(read_cpuid_mpidr(), 1);
> + if (!this_cluster)
> + pmu_raw_writel(EXYNOS5420_ARM_USE_STANDBY_WFI0,
> + S5P_CENTRAL_SEQ_OPTION);
> + else
> + pmu_raw_writel(EXYNOS5420_KFC_USE_STANDBY_WFI0,
> + S5P_CENTRAL_SEQ_OPTION);
By the way, is it even possible to boot this SoC using other CPU than
first Cortex A15?
> + } else {
> + tmp = (S5P_USE_STANDBY_WFI0 | S5P_USE_STANDBY_WFE0);
> + pmu_raw_writel(tmp, S5P_CENTRAL_SEQ_OPTION);
> + }
>
> if (read_cpuid_part_number() == ARM_CPU_PART_CORTEX_A9)
> exynos_cpu_save_register();
> @@ -340,6 +405,17 @@ static int exynos_pm_central_resume(void)
>
> static void exynos_pm_resume(void)
> {
> + unsigned int tmp;
> +
> + if (soc_is_exynos5420()) {
> + /* Restore the sysram cpu state register */
> + __raw_writel(exynos5420_cpu_state,
> + sysram_base_addr + EXYNOS5420_CPU_STATE);
> +
> + pmu_raw_writel(EXYNOS5420_USE_STANDBY_WFI_ALL,
> + S5P_CENTRAL_SEQ_OPTION);
> + }
> +
> if (exynos_pm_central_resume())
> goto early_wakeup;
>
> @@ -347,18 +423,42 @@ static void exynos_pm_resume(void)
> exynos_cpu_restore_register();
>
> /* For release retention */
> + if (soc_is_exynos5420()) {
> + pmu_raw_writel((1 << 28), EXYNOS_PAD_RET_DRAM_OPTION);
> + pmu_raw_writel((1 << 28), EXYNOS_PAD_RET_MAUDIO_OPTION);
> + pmu_raw_writel((1 << 28), EXYNOS_PAD_RET_JTAG_OPTION);
> + pmu_raw_writel((1 << 28), EXYNOS5420_PAD_RET_GPIO_OPTION);
> + pmu_raw_writel((1 << 28), EXYNOS5420_PAD_RET_UART_OPTION);
> + pmu_raw_writel((1 << 28), EXYNOS5420_PAD_RET_MMCA_OPTION);
> + pmu_raw_writel((1 << 28), EXYNOS5420_PAD_RET_MMCB_OPTION);
> + pmu_raw_writel((1 << 28), EXYNOS5420_PAD_RET_MMCC_OPTION);
> + pmu_raw_writel((1 << 28), EXYNOS5420_PAD_RET_HSI_OPTION);
> + pmu_raw_writel((1 << 28), EXYNOS_PAD_RET_EBIA_OPTION);
> + pmu_raw_writel((1 << 28), EXYNOS_PAD_RET_EBIB_OPTION);
> + pmu_raw_writel((1 << 28), EXYNOS5420_PAD_RET_SPI_OPTION);
> + pmu_raw_writel((1 << 28),
> + EXYNOS5420_PAD_RET_DRAM_COREBLK_OPTION);
This could be replaced with an array of registers that is specified in
the exynos_pm_data struct. Also while at it, the (1 << 28) could be
replaced with proper macro.
> + } else {
> + pmu_raw_writel((1 << 28), S5P_PAD_RET_MAUDIO_OPTION);
> + pmu_raw_writel((1 << 28), S5P_PAD_RET_GPIO_OPTION);
> + pmu_raw_writel((1 << 28), S5P_PAD_RET_UART_OPTION);
> + pmu_raw_writel((1 << 28), S5P_PAD_RET_MMCA_OPTION);
> + pmu_raw_writel((1 << 28), S5P_PAD_RET_MMCB_OPTION);
> + pmu_raw_writel((1 << 28), S5P_PAD_RET_EBIA_OPTION);
> + pmu_raw_writel((1 << 28), S5P_PAD_RET_EBIB_OPTION);
> + }
>
> - pmu_raw_writel((1 << 28), S5P_PAD_RET_MAUDIO_OPTION);
> - pmu_raw_writel((1 << 28), S5P_PAD_RET_GPIO_OPTION);
> - pmu_raw_writel((1 << 28), S5P_PAD_RET_UART_OPTION);
> - pmu_raw_writel((1 << 28), S5P_PAD_RET_MMCA_OPTION);
> - pmu_raw_writel((1 << 28), S5P_PAD_RET_MMCB_OPTION);
> - pmu_raw_writel((1 << 28), S5P_PAD_RET_EBIA_OPTION);
> - pmu_raw_writel((1 << 28), S5P_PAD_RET_EBIB_OPTION);
> -
> - if (soc_is_exynos5250())
> + if (soc_is_exynos5250()) {
> s3c_pm_do_restore(exynos5_sys_save,
> ARRAY_SIZE(exynos5_sys_save));
> + } else if (soc_is_exynos5420()) {
> + unsigned int i;
> +
> + for (i = 0; i < ARRAY_SIZE(exynos5420_pmu_reg_save); i++)
> + pmu_raw_writel(
> + (unsigned int)exynos5420_pmu_reg_save[i].val,
> + (unsigned int)exynos5420_pmu_reg_save[i].reg);
> + }
>
> s3c_pm_do_restore_core(exynos_core_save, ARRAY_SIZE(exynos_core_save));
>
> @@ -367,6 +467,18 @@ static void exynos_pm_resume(void)
>
> early_wakeup:
>
> + if (soc_is_exynos5420()) {
> + tmp = __raw_readl(pmu_base_addr + EXYNOS5420_SFR_AXI_CGDIS1);
> + tmp &= ~EXYNOS5420_UFS;
> + pmu_raw_writel(tmp, EXYNOS5420_SFR_AXI_CGDIS1);
nit: Spacing here would be nice.
> + tmp = __raw_readl(pmu_base_addr + EXYNOS5420_FSYS2_OPTION);
> + tmp &= ~EXYNOS5420_EMULATION;
> + pmu_raw_writel(tmp, EXYNOS5420_FSYS2_OPTION);
Ditto.
> + tmp = __raw_readl(pmu_base_addr + EXYNOS5420_PSGEN_OPTION);
> + tmp &= ~EXYNOS5420_EMULATION;
> + pmu_raw_writel(tmp, EXYNOS5420_PSGEN_OPTION);
> + }
> +
> /* Clear SLEEP mode set in INFORM1 */
> pmu_raw_writel(0x0, S5P_INFORM1);
>
> @@ -483,9 +595,15 @@ void __init exynos_pm_init(void)
> gic_arch_extn.irq_set_wake = exynos_irq_set_wake;
>
> /* All wakeup disable */
> - tmp = __raw_readl(pmu_base_addr + S5P_WAKEUP_MASK);
> - tmp |= ((0xFF << 8) | (0x1F << 1));
> - pmu_raw_writel(tmp, S5P_WAKEUP_MASK);
> + if (soc_is_exynos5420()) {
> + tmp = __raw_readl(pmu_base_addr + S5P_WAKEUP_MASK);
> + tmp |= ((0x7F << 7) | (0x1F << 1));
This mask could be also moved to the struct.
Best regards,
Tomasz
More information about the linux-arm-kernel
mailing list