diff mbox

[1/4] ARM: exynos4: Add support for AFTR mode cpuidle state

Message ID 1320426228-21746-2-git-send-email-amit.kachhap@linaro.org
State Superseded
Headers show

Commit Message

Amit Daniel Kachhap Nov. 4, 2011, 5:03 p.m. UTC
From: Amit Daniel Kachhap <amit.kachhap@linaro.org>

This patch adds support for AFTR(ARM OFF TOP RUNNING) mode in
cpuidle driver for EXYNOS4210. L2 cache keeps their data in this mode.
This patch ports the code to the latest interfaces to
save/restore CPU state inclusive of CPU PM notifiers, l2
resume and cpu_suspend/resume.

Signed-off-by: Jaecheol Lee <jc.lee@samsung.com>
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Signed-off-by: Amit Daniel Kachhap <amit.kachhap@linaro.org>
---
 arch/arm/mach-exynos4/cpuidle.c          |  152 +++++++++++++++++++++++++++++-
 arch/arm/mach-exynos4/include/mach/pmu.h |    2 +
 2 files changed, 151 insertions(+), 3 deletions(-)

Comments

MyungJoo Ham Nov. 11, 2011, 7:33 a.m. UTC | #1
On Sat, Nov 5, 2011 at 2:03 AM,  <amit.kachhap@linaro.org> wrote:
> From: Amit Daniel Kachhap <amit.kachhap@linaro.org>
>
> This patch adds support for AFTR(ARM OFF TOP RUNNING) mode in
> cpuidle driver for EXYNOS4210. L2 cache keeps their data in this mode.
> This patch ports the code to the latest interfaces to
> save/restore CPU state inclusive of CPU PM notifiers, l2
> resume and cpu_suspend/resume.
>
> Signed-off-by: Jaecheol Lee <jc.lee@samsung.com>
> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
> Signed-off-by: Amit Daniel Kachhap <amit.kachhap@linaro.org>
> ---
[]
> +#define REG_DIRECTGO_ADDR      (samsung_rev() == EXYNOS4210_REV_1_1 ? \
> +                               S5P_INFORM7 : (S5P_VA_SYSRAM + 0x24))
> +#define REG_DIRECTGO_FLAG      (samsung_rev() == EXYNOS4210_REV_1_1 ? \
> +                               S5P_INFORM6 : (S5P_VA_SYSRAM + 0x20))
>
[]
> +
> +       __raw_writel(BSYM(virt_to_phys(s3c_cpu_resume)),
> +                                                REG_DIRECTGO_ADDR);
> +       __raw_writel(0xfcba0d10, REG_DIRECTGO_FLAG);
> +
>        return 0;

Hello,


Why are you using INFORM6 and 7 registers in order to save resume
address and power-mode flags?

INFORM0 and 1 have been used in pm.c for the exactly same purpose.
Please use the same registers in cpuidle.c as well.

The same part in bootloader (IPL) can handle whether it's
suspend-to-RAM or AFTR and the both modes are mutually exclusive and
you only need one value for resume PC.

Therefore, you can keep the value at the same location, which is the
method we have been using.

Besides, the Exynos4210 chipmaker (S.LSI) has told that INFORM6 and 7
registers are used by in-chip code (iROM or iRAM).


Cheers!
MyungJoo
Amit Daniel Kachhap Nov. 17, 2011, 11:22 a.m. UTC | #2
On 11 November 2011 13:03, MyungJoo Ham <myungjoo.ham@gmail.com> wrote:
> On Sat, Nov 5, 2011 at 2:03 AM,  <amit.kachhap@linaro.org> wrote:
>> From: Amit Daniel Kachhap <amit.kachhap@linaro.org>
>>
>> This patch adds support for AFTR(ARM OFF TOP RUNNING) mode in
>> cpuidle driver for EXYNOS4210. L2 cache keeps their data in this mode.
>> This patch ports the code to the latest interfaces to
>> save/restore CPU state inclusive of CPU PM notifiers, l2
>> resume and cpu_suspend/resume.
>>
>> Signed-off-by: Jaecheol Lee <jc.lee@samsung.com>
>> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
>> Signed-off-by: Amit Daniel Kachhap <amit.kachhap@linaro.org>
>> ---
> []
>> +#define REG_DIRECTGO_ADDR      (samsung_rev() == EXYNOS4210_REV_1_1 ? \
>> +                               S5P_INFORM7 : (S5P_VA_SYSRAM + 0x24))
>> +#define REG_DIRECTGO_FLAG      (samsung_rev() == EXYNOS4210_REV_1_1 ? \
>> +                               S5P_INFORM6 : (S5P_VA_SYSRAM + 0x20))
>>
> []
>> +
>> +       __raw_writel(BSYM(virt_to_phys(s3c_cpu_resume)),
>> +                                                REG_DIRECTGO_ADDR);
>> +       __raw_writel(0xfcba0d10, REG_DIRECTGO_FLAG);
>> +
>>        return 0;
>
> Hello,
>
>
> Why are you using INFORM6 and 7 registers in order to save resume
> address and power-mode flags?
>
> INFORM0 and 1 have been used in pm.c for the exactly same purpose.
> Please use the same registers in cpuidle.c as well.
>
> The same part in bootloader (IPL) can handle whether it's
> suspend-to-RAM or AFTR and the both modes are mutually exclusive and
> you only need one value for resume PC.
>
> Therefore, you can keep the value at the same location, which is the
> method we have been using.
Hi,

I tried using INFORM0 and INFORM1 in cpuidle and make it same as pm.c.
But this doesnt work. Looks like my irom fused code checks for the
wakeup source and needs INFORM7 and INFORM6 for non sleep wakeups. My
cpu version is S5PV310AH--0AH1113(Origen board).
I suggest adding support for both type of wakeups(directly from irom
and through bootloader).

Thanks,
Amit D
>
> Besides, the Exynos4210 chipmaker (S.LSI) has told that INFORM6 and 7
> registers are used by in-chip code (iROM or iRAM).
>
>
> Cheers!
> MyungJoo
>
>
>
> --
> MyungJoo Ham, Ph.D.
> Mobile Software Platform Lab, DMC Business, Samsung Electronics
>
MyungJoo Ham Nov. 18, 2011, 4:07 a.m. UTC | #3
On Thu, Nov 17, 2011 at 8:22 PM, Amit Kachhap <amit.kachhap@linaro.org> wrote:
> On 11 November 2011 13:03, MyungJoo Ham <myungjoo.ham@gmail.com> wrote:
>> On Sat, Nov 5, 2011 at 2:03 AM,  <amit.kachhap@linaro.org> wrote:
>>> From: Amit Daniel Kachhap <amit.kachhap@linaro.org>
>>>
>>> This patch adds support for AFTR(ARM OFF TOP RUNNING) mode in
>>> cpuidle driver for EXYNOS4210. L2 cache keeps their data in this mode.
>>> This patch ports the code to the latest interfaces to
>>> save/restore CPU state inclusive of CPU PM notifiers, l2
>>> resume and cpu_suspend/resume.
>>>
>>> Signed-off-by: Jaecheol Lee <jc.lee@samsung.com>
>>> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
>>> Signed-off-by: Amit Daniel Kachhap <amit.kachhap@linaro.org>
>>> ---
>> []
>>> +#define REG_DIRECTGO_ADDR      (samsung_rev() == EXYNOS4210_REV_1_1 ? \
>>> +                               S5P_INFORM7 : (S5P_VA_SYSRAM + 0x24))
>>> +#define REG_DIRECTGO_FLAG      (samsung_rev() == EXYNOS4210_REV_1_1 ? \
>>> +                               S5P_INFORM6 : (S5P_VA_SYSRAM + 0x20))
>>>
>> []
>>> +
>>> +       __raw_writel(BSYM(virt_to_phys(s3c_cpu_resume)),
>>> +                                                REG_DIRECTGO_ADDR);
>>> +       __raw_writel(0xfcba0d10, REG_DIRECTGO_FLAG);
>>> +
>>>        return 0;
>>
>> Hello,
>>
>>
>> Why are you using INFORM6 and 7 registers in order to save resume
>> address and power-mode flags?
>>
>> INFORM0 and 1 have been used in pm.c for the exactly same purpose.
>> Please use the same registers in cpuidle.c as well.
>>
>> The same part in bootloader (IPL) can handle whether it's
>> suspend-to-RAM or AFTR and the both modes are mutually exclusive and
>> you only need one value for resume PC.
>>
>> Therefore, you can keep the value at the same location, which is the
>> method we have been using.
> Hi,
>
> I tried using INFORM0 and INFORM1 in cpuidle and make it same as pm.c.
> But this doesnt work. Looks like my irom fused code checks for the
> wakeup source and needs INFORM7 and INFORM6 for non sleep wakeups. My
> cpu version is S5PV310AH--0AH1113(Origen board).
> I suggest adding support for both type of wakeups(directly from irom
> and through bootloader).
>
Hello Amit,

Have you checked the part that checks flags for AFTR mode and
Suspend-to-RAM in IPL?

If your code is checking AFTR flags with INFORM6/7 and PM flags with
INFORM0/1, that is terribly wrong as those two modes are mutually
exclusive and are very similar in terms of IPL codes.

Also, the IPL-bootloader code runs in IRAM where CPU determines
whether to continue to boot or jump to resume address of
suspend-to-RAM or deepidle (AFTR/LPA/...).

If yours works with INFORM0/1 for PM, it should work with INFORM0/1 in
AFTR as well if your (S/W loadable) IPL code is correct and INFORM6/7
is not touched regardless of how your IROM code is written. Please
check your IPL code and try to let AFTR use the same address with PM.
There is no reason to use another INFORM registers for AFTR only.
Please note that pm-related IPL code is not in ROM, but loaded by ROM
part of IPL to RAM from a storage.



To jc.lee,

Could you please check whether your IPL code checks for INFORM0/1 for
both PM and AFTR or INFORM0/1 for PM and INFORM6/7 for AFTR?
If the latter is what your IPL does, then, how does it react when both
INFORM0/1 and INFORM6/7 are set for both PM and AFTR?
The IPL code we have uses INFORM0/1 for both PM and AFTR/LPA.


Cheers!
MyungJoo

> Thanks,
> Amit D
>>
>> Besides, the Exynos4210 chipmaker (S.LSI) has told that INFORM6 and 7
>> registers are used by in-chip code (iROM or iRAM).
>>
>>
>> Cheers!
>> MyungJoo
>>
>>
>>
>> --
>> MyungJoo Ham, Ph.D.
>> Mobile Software Platform Lab, DMC Business, Samsung Electronics
>>
>
diff mbox

Patch

diff --git a/arch/arm/mach-exynos4/cpuidle.c b/arch/arm/mach-exynos4/cpuidle.c
index bf7e96f..111ccc3 100644
--- a/arch/arm/mach-exynos4/cpuidle.c
+++ b/arch/arm/mach-exynos4/cpuidle.c
@@ -11,22 +11,48 @@ 
 #include <linux/kernel.h>
 #include <linux/init.h>
 #include <linux/cpuidle.h>
+#include <linux/cpu_pm.h>
 #include <linux/io.h>
-
+#include <linux/suspend.h>
+#include <linux/err.h>
 #include <asm/proc-fns.h>
+#include <asm/smp_scu.h>
+#include <asm/suspend.h>
+#include <asm/unified.h>
+#include <mach/regs-pmu.h>
+#include <mach/pmu.h>
+
+#include <plat/exynos4.h>
+#include <plat/cpu.h>
+
+#define REG_DIRECTGO_ADDR	(samsung_rev() == EXYNOS4210_REV_1_1 ? \
+				S5P_INFORM7 : (S5P_VA_SYSRAM + 0x24))
+#define REG_DIRECTGO_FLAG	(samsung_rev() == EXYNOS4210_REV_1_1 ? \
+				S5P_INFORM6 : (S5P_VA_SYSRAM + 0x20))
 
 static int exynos4_enter_idle(struct cpuidle_device *dev,
 			      struct cpuidle_state *state);
 
+static int exynos4_enter_lowpower(struct cpuidle_device *dev,
+				  struct cpuidle_state *state);
+
 static struct cpuidle_state exynos4_cpuidle_set[] = {
 	[0] = {
 		.enter			= exynos4_enter_idle,
 		.exit_latency		= 1,
 		.target_residency	= 100000,
 		.flags			= CPUIDLE_FLAG_TIME_VALID,
-		.name			= "IDLE",
+		.name			= "C0",
 		.desc			= "ARM clock gating(WFI)",
 	},
+	[1] = {
+		.enter			= exynos4_enter_lowpower,
+		.exit_latency		= 300,
+		.target_residency	= 100000,
+		.flags			= CPUIDLE_FLAG_TIME_VALID,
+		.name			= "C1",
+		.desc			= "ARM power down",
+	},
 };
 
 static DEFINE_PER_CPU(struct cpuidle_device, exynos4_cpuidle_device);
@@ -36,6 +62,96 @@  static struct cpuidle_driver exynos4_idle_driver = {
 	.owner		= THIS_MODULE,
 };
 
+/* Ext-GIC nIRQ/nFIQ is the only wakeup source in AFTR */
+static void exynos4_set_wakeupmask(void)
+{
+	__raw_writel(0x0000ff3e, S5P_WAKEUP_MASK);
+}
+
+static unsigned int g_pwr_ctrl, g_diag_reg;
+
+static void save_cpu_arch_register(void)
+{
+	/*read power control register*/
+	asm("mrc p15, 0, %0, c15, c0, 0" : "=r"(g_pwr_ctrl) : : "cc");
+	/*read diagnostic register*/
+	asm("mrc p15, 0, %0, c15, c0, 1" : "=r"(g_diag_reg) : : "cc");
+	return;
+}
+
+static void restore_cpu_arch_register(void)
+{
+	/*write power control register*/
+	asm("mcr p15, 0, %0, c15, c0, 0" : : "r"(g_pwr_ctrl) : "cc");
+	/*write diagnostic register*/
+	asm("mcr p15, 0, %0, c15, c0, 1" : : "r"(g_diag_reg) : "cc");
+	return;
+}
+
+static int idle_finisher(unsigned long flags)
+{
+	cpu_do_idle();
+	return 1;
+}
+
+static int exynos4_enter_core0_aftr(struct cpuidle_device *dev,
+				    struct cpuidle_state *state)
+{
+	struct timeval before, after;
+	int idle_time;
+	unsigned long tmp;
+
+	local_irq_disable();
+	do_gettimeofday(&before);
+
+	exynos4_set_wakeupmask();
+
+	/* Set value of power down register for aftr mode */
+	exynos4_sys_powerdown_conf(SYS_AFTR);
+
+	save_cpu_arch_register();
+
+	/* Setting Central Sequence Register for power down mode */
+	tmp = __raw_readl(S5P_CENTRAL_SEQ_CONFIGURATION);
+	tmp &= ~S5P_CENTRAL_LOWPWR_CFG;
+	__raw_writel(tmp, S5P_CENTRAL_SEQ_CONFIGURATION);
+
+	cpu_pm_enter();
+	cpu_cluster_pm_enter();
+
+	cpu_suspend(0, idle_finisher);
+
+	scu_enable(S5P_VA_SCU);
+
+	cpu_cluster_pm_exit();
+	cpu_pm_exit();
+
+	restore_cpu_arch_register();
+
+	/*
+	 * If PMU failed while entering sleep mode, WFI will be
+	 * ignored by PMU and then exiting cpu_do_idle().
+	 * S5P_CENTRAL_LOWPWR_CFG bit will not be set automatically
+	 * in this situation.
+	 */
+	tmp = __raw_readl(S5P_CENTRAL_SEQ_CONFIGURATION);
+	if (!(tmp & S5P_CENTRAL_LOWPWR_CFG)) {
+		tmp |= S5P_CENTRAL_LOWPWR_CFG;
+		__raw_writel(tmp, S5P_CENTRAL_SEQ_CONFIGURATION);
+	}
+
+	/* Clear wakeup state register */
+	__raw_writel(0x0, S5P_WAKEUP_STAT);
+
+	do_gettimeofday(&after);
+
+	local_irq_enable();
+	idle_time = (after.tv_sec - before.tv_sec) * USEC_PER_SEC +
+		    (after.tv_usec - before.tv_usec);
+
+	return idle_time;
+}
+
 static int exynos4_enter_idle(struct cpuidle_device *dev,
 			      struct cpuidle_state *state)
 {
@@ -55,6 +171,26 @@  static int exynos4_enter_idle(struct cpuidle_device *dev,
 	return idle_time;
 }
 
+static int exynos4_enter_lowpower(struct cpuidle_device *dev,
+				  struct cpuidle_state *state)
+{
+	struct cpuidle_state *new_state = state;
+
+	/* This mode only can be entered when Core1 is offline */
+	if (num_online_cpus() > 1) {
+		BUG_ON(!dev->safe_state);
+		new_state = dev->safe_state;
+	}
+	dev->last_state = new_state;
+
+	if (new_state == &dev->states[0])
+		return exynos4_enter_idle(dev, new_state);
+	else
+		return exynos4_enter_core0_aftr(dev, new_state);
+
+	return exynos4_enter_idle(dev, new_state);
+}
+
 static int __init exynos4_init_cpuidle(void)
 {
 	int i, max_cpuidle_state, cpu_id;
@@ -66,8 +202,11 @@  static int __init exynos4_init_cpuidle(void)
 		device = &per_cpu(exynos4_cpuidle_device, cpu_id);
 		device->cpu = cpu_id;
 
-		device->state_count = (sizeof(exynos4_cpuidle_set) /
+		if (cpu_id == 0)
+			device->state_count = (sizeof(exynos4_cpuidle_set) /
 					       sizeof(struct cpuidle_state));
+		else
+			device->state_count = 1;	/* Support IDLE only */
 
 		max_cpuidle_state = device->state_count;
 
@@ -76,11 +215,18 @@  static int __init exynos4_init_cpuidle(void)
 					sizeof(struct cpuidle_state));
 		}
 
+		device->safe_state = &device->states[0];
+
 		if (cpuidle_register_device(device)) {
 			printk(KERN_ERR "CPUidle register device failed\n,");
 			return -EIO;
 		}
 	}
+
+	__raw_writel(BSYM(virt_to_phys(s3c_cpu_resume)),
+						 REG_DIRECTGO_ADDR);
+	__raw_writel(0xfcba0d10, REG_DIRECTGO_FLAG);
+
 	return 0;
 }
 device_initcall(exynos4_init_cpuidle);
diff --git a/arch/arm/mach-exynos4/include/mach/pmu.h b/arch/arm/mach-exynos4/include/mach/pmu.h
index a952904..31d1ed8 100644
--- a/arch/arm/mach-exynos4/include/mach/pmu.h
+++ b/arch/arm/mach-exynos4/include/mach/pmu.h
@@ -20,6 +20,8 @@  enum sys_powerdown {
 	NUM_SYS_POWERDOWN,
 };
 
+extern unsigned long l2x0_regs_phys;
 extern void exynos4_sys_powerdown_conf(enum sys_powerdown mode);
+extern void s3c_cpu_resume(void);
 
 #endif /* __ASM_ARCH_PMU_H */