On Exynos4 USE_DELAYED_RESET_ASSERTION must be set in
ARM_COREx_OPTION register during CPU power down. This is the proper way
of powering down CPU on Exynos4.
Additionally on Exynos4212 without this the CPU clock down feature won't
work after powering down some CPU and the online CPUs will work at full
frequency chosen by CPUfreq governor.
Signed-off-by: Krzysztof Kozlowski <[email protected]>
---
Changes since v1:
1. Use delayed reset assertion on all Exynos4 family and all cores, not
only on core 1 of Exynos4212.
2. Rebase on Tomasz Figa's patch:
ARM: EXYNOS: Fix core ID used by platsmp and hotplug code
http://www.spinics.net/lists/linux-samsung-soc/msg31604.html
Tomasz's patch is currently applied to Kukjin's v3.16-samsung-fixes-4
---
arch/arm/mach-exynos/hotplug.c | 27 +++++++++++++++++++++++++--
arch/arm/mach-exynos/regs-pmu.h | 3 +++
2 files changed, 28 insertions(+), 2 deletions(-)
diff --git a/arch/arm/mach-exynos/hotplug.c b/arch/arm/mach-exynos/hotplug.c
index 920a4baa53cd..e6340b645f9d 100644
--- a/arch/arm/mach-exynos/hotplug.c
+++ b/arch/arm/mach-exynos/hotplug.c
@@ -22,7 +22,7 @@
#include "common.h"
#include "regs-pmu.h"
-static inline void cpu_leave_lowpower(void)
+static inline void cpu_leave_lowpower(u32 core_id)
{
unsigned int v;
@@ -36,6 +36,14 @@ static inline void cpu_leave_lowpower(void)
: "=&r" (v)
: "Ir" (CR_C), "Ir" (0x40)
: "cc");
+
+ if (soc_is_exynos4()) {
+ unsigned int tmp;
+
+ tmp = __raw_readl(EXYNOS_ARM_CORE_OPTION(core_id));
+ tmp &= ~(S5P_USE_DELAYED_RESET_ASSERTION);
+ __raw_writel(tmp, EXYNOS_ARM_CORE_OPTION(core_id));
+ }
}
static inline void platform_do_lowpower(unsigned int cpu, int *spurious)
@@ -47,6 +55,19 @@ static inline void platform_do_lowpower(unsigned int cpu, int *spurious)
/* Turn the CPU off on next WFI instruction. */
exynos_cpu_power_down(core_id);
+ if (soc_is_exynos4()) {
+ unsigned int tmp;
+
+ /*
+ * Exynos 4 SoCs require setting
+ * USE_DELAYED_RESET_ASSERTION so the CPU idle
+ * clock down feature could properly detect
+ * global idle state when CPUx is off.
+ */
+ tmp = __raw_readl(EXYNOS_ARM_CORE_OPTION(core_id));
+ tmp |= S5P_USE_DELAYED_RESET_ASSERTION;
+ __raw_writel(tmp, EXYNOS_ARM_CORE_OPTION(core_id));
+ }
wfi();
@@ -76,6 +97,8 @@ static inline void platform_do_lowpower(unsigned int cpu, int *spurious)
void __ref exynos_cpu_die(unsigned int cpu)
{
int spurious = 0;
+ u32 mpidr = cpu_logical_map(cpu);
+ u32 core_id = MPIDR_AFFINITY_LEVEL(mpidr, 0);
v7_exit_coherency_flush(louis);
@@ -85,7 +108,7 @@ void __ref exynos_cpu_die(unsigned int cpu)
* bring this CPU back into the world of cache
* coherency, and then restore interrupts
*/
- cpu_leave_lowpower();
+ cpu_leave_lowpower(core_id);
if (spurious)
pr_warn("CPU%u: %u spurious wakeup calls\n", cpu, spurious);
diff --git a/arch/arm/mach-exynos/regs-pmu.h b/arch/arm/mach-exynos/regs-pmu.h
index 1d13b08708f0..59bd92112842 100644
--- a/arch/arm/mach-exynos/regs-pmu.h
+++ b/arch/arm/mach-exynos/regs-pmu.h
@@ -25,6 +25,7 @@
#define S5P_USE_STANDBY_WFI0 (1 << 16)
#define S5P_USE_STANDBY_WFE0 (1 << 24)
+#define S5P_USE_DELAYED_RESET_ASSERTION BIT(12)
#define EXYNOS_SWRESET S5P_PMUREG(0x0400)
#define EXYNOS5440_SWRESET S5P_PMUREG(0x00C4)
@@ -111,6 +112,8 @@
(EXYNOS_ARM_CORE0_CONFIGURATION + (0x80 * (_nr)))
#define EXYNOS_ARM_CORE_STATUS(_nr) \
(EXYNOS_ARM_CORE_CONFIGURATION(_nr) + 0x4)
+#define EXYNOS_ARM_CORE_OPTION(_nr) \
+ (EXYNOS_ARM_CORE_CONFIGURATION(_nr) + 0x8)
#define EXYNOS_ARM_COMMON_CONFIGURATION S5P_PMUREG(0x2500)
#define EXYNOS_COMMON_CONFIGURATION(_nr) \
--
1.9.1
Hi Krzysztof,
Please see my comment below.
On 16.07.2014 10:23, Krzysztof Kozlowski wrote:
> On Exynos4 USE_DELAYED_RESET_ASSERTION must be set in
> ARM_COREx_OPTION register during CPU power down. This is the proper way
> of powering down CPU on Exynos4.
>
> Additionally on Exynos4212 without this the CPU clock down feature won't
> work after powering down some CPU and the online CPUs will work at full
> frequency chosen by CPUfreq governor.
>
> Signed-off-by: Krzysztof Kozlowski <[email protected]>
[snip]
> @@ -36,6 +36,14 @@ static inline void cpu_leave_lowpower(void)
> : "=&r" (v)
> : "Ir" (CR_C), "Ir" (0x40)
> : "cc");
> +
> + if (soc_is_exynos4()) {
> + unsigned int tmp;
> +
> + tmp = __raw_readl(EXYNOS_ARM_CORE_OPTION(core_id));
> + tmp &= ~(S5P_USE_DELAYED_RESET_ASSERTION);
> + __raw_writel(tmp, EXYNOS_ARM_CORE_OPTION(core_id));
> + }
This code is executed only if the CPU doesn't manage to power off before
already receiving a wake-up event. Otherwise the context is lost after
wfi(), as the core power is being cut down.
The normal wake-up path goes through platsmp's exynos_boot_secondary()
and I believe that setting of S5P_USE_DELAYED_RESET_ASSERTION should be
taken care there as well.
Best regards,
Tomasz
On śro, 2014-07-16 at 12:48 +0200, Tomasz Figa wrote:
> Hi Krzysztof,
>
> Please see my comment below.
>
> On 16.07.2014 10:23, Krzysztof Kozlowski wrote:
> > On Exynos4 USE_DELAYED_RESET_ASSERTION must be set in
> > ARM_COREx_OPTION register during CPU power down. This is the proper way
> > of powering down CPU on Exynos4.
> >
> > Additionally on Exynos4212 without this the CPU clock down feature won't
> > work after powering down some CPU and the online CPUs will work at full
> > frequency chosen by CPUfreq governor.
> >
> > Signed-off-by: Krzysztof Kozlowski <[email protected]>
>
> [snip]
>
> > @@ -36,6 +36,14 @@ static inline void cpu_leave_lowpower(void)
> > : "=&r" (v)
> > : "Ir" (CR_C), "Ir" (0x40)
> > : "cc");
> > +
> > + if (soc_is_exynos4()) {
> > + unsigned int tmp;
> > +
> > + tmp = __raw_readl(EXYNOS_ARM_CORE_OPTION(core_id));
> > + tmp &= ~(S5P_USE_DELAYED_RESET_ASSERTION);
> > + __raw_writel(tmp, EXYNOS_ARM_CORE_OPTION(core_id));
> > + }
>
> This code is executed only if the CPU doesn't manage to power off before
> already receiving a wake-up event. Otherwise the context is lost after
> wfi(), as the core power is being cut down.
>
> The normal wake-up path goes through platsmp's exynos_boot_secondary()
> and I believe that setting of S5P_USE_DELAYED_RESET_ASSERTION should be
> taken care there as well.
You're right. Thanks for pointing this, I'll fix it.
Best regards,
Krzysztof
>
> Best regards,
> Tomasz