[PATCH V3 0/3] ARM: tegra114: cpuidle: add power down state

[PATCH V3 0/3] ARM: tegra114: cpuidle: add power down state

[Joseph Lo]

This series introduce CPU core power down state for CPU idle. When CPU go
into this state, it saves it's context and needs a proper configuration
in flow controller to power gate the CPU when CPU runs into WFI
instruction. And the CPU also needs to set the IRQ as CPU power down idle
wake up event in flow controller.

To prevent race conditions and ensure proper interrupt routing on
Cortex-A15 CPUs when they are power-gated, add a CPU PM notifier
call-back to reprogram the GIC CPU interface on PM entry. The
GIC CPU interface will be reset back to its normal state by
the common GIC CPU PM exit callback when the CPU wakes up.

This series depends on the patch of "tick:  Fix
tick_broadcast_pending_mask not cleared".

V3:
* use CPUIDLE_FLAG_TIMER_STOP flag
V2:
* clean up the CPUidle driver to make it more generic

Joseph Lo (3):
  ARM: tegra114: Reprogram GIC CPU interface to bypass IRQ on CPU PM
    entry
  ARM: tegra114: add low level support for CPU idle powered-down mode
  ARM: tegra114: cpuidle: add powered-down state

 arch/arm/mach-tegra/cpuidle-tegra114.c | 48 +++++++++++++++++++++++++++++++++-
 arch/arm/mach-tegra/flowctrl.h         |  2 ++
 arch/arm/mach-tegra/irq.c              | 40 ++++++++++++++++++++++++++++
 arch/arm/mach-tegra/sleep-tegra30.S    |  2 ++
 4 files changed, 91 insertions(+), 1 deletion(-)

-- 
1.8.3.1

[PATCH V3 1/3] ARM: tegra114: Reprogram GIC CPU interface to bypass IRQ on CPU PM entry

[Joseph Lo]

There is a difference between GICv1 and v2 when CPU in power management
mode (aka CPU power down on Tegra). For GICv1, IRQ/FIQ interrupt lines
going to CPU are same lines which are also used for wake-interrupt.
Therefore, we cannot disable the GIC CPU interface if we need to use same
interrupts for CPU wake purpose. This creates a race condition for CPU
power off entry. Also, in GICv1, disabling GICv1 CPU interface puts GICv1
into bypass mode such that incoming legacy IRQ/FIQ are sent to CPU, which
means disabling GIC CPU interface doesn't really disable IRQ/FIQ to CPU.

GICv2 provides a wake IRQ/FIQ (for wake-event purpose), which are not
disabled by GIC CPU interface. This is done by adding a bypass override
capability when the interrupts are disabled at the CPU interface. To
support this, there are four bits about IRQ/FIQ BypassDisable in CPU
interface Control Register. When the IRQ/FIQ not being driver by the
CPU interface, each interrupt output signal can be deasserted rather
than being driven by the legacy interrupt input.

So the wake-event can be used as wakeup signals to SoC (system power
controller).

To prevent race conditions and ensure proper interrupt routing on
Cortex-A15 CPUs when they are power-gated, add a CPU PM notifier
call-back to reprogram the GIC CPU interface on PM entry. The
GIC CPU interface will be reset back to its normal state by
the common GIC CPU PM exit callback when the CPU wakes up.

Based on the work by: Scott Williams <scwilliams@nvidia.com>

Signed-off-by: Joseph Lo <josephl@nvidia.com>
---
V3:
* no change
V2:
* no change
---
 arch/arm/mach-tegra/irq.c | 40 ++++++++++++++++++++++++++++++++++++++++
 1 file changed, 40 insertions(+)

diff --git a/arch/arm/mach-tegra/irq.c b/arch/arm/mach-tegra/irq.c
index 0de4eed..1a74d56 100644
--- a/arch/arm/mach-tegra/irq.c
+++ b/arch/arm/mach-tegra/irq.c
@@ -18,10 +18,12 @@
  */
 
 #include <linux/kernel.h>
+#include <linux/cpu_pm.h>
 #include <linux/interrupt.h>
 #include <linux/irq.h>
 #include <linux/io.h>
 #include <linux/of.h>
+#include <linux/of_address.h>
 #include <linux/irqchip/arm-gic.h>
 #include <linux/syscore_ops.h>
 
@@ -65,6 +67,7 @@ static u32 cpu_ier[TEGRA_MAX_NUM_ICTLRS];
 static u32 cpu_iep[TEGRA_MAX_NUM_ICTLRS];
 
 static u32 ictlr_wake_mask[TEGRA_MAX_NUM_ICTLRS];
+static void __iomem *tegra_gic_cpu_base;
 #endif
 
 bool tegra_pending_sgi(void)
@@ -213,8 +216,43 @@ int tegra_legacy_irq_syscore_init(void)
 
 	return 0;
 }
+
+static int tegra_gic_notifier(struct notifier_block *self,
+			      unsigned long cmd, void *v)
+{
+	switch (cmd) {
+	case CPU_PM_ENTER:
+		writel_relaxed(0x1E0, tegra_gic_cpu_base + GIC_CPU_CTRL);
+		break;
+	}
+
+	return NOTIFY_OK;
+}
+
+static struct notifier_block tegra_gic_notifier_block = {
+	.notifier_call = tegra_gic_notifier,
+};
+
+static const struct of_device_id tegra114_dt_gic_match[] __initconst = {
+	{ .compatible = "arm,cortex-a15-gic" },
+	{ }
+};
+
+static void tegra114_gic_cpu_pm_registration(void)
+{
+	struct device_node *dn;
+
+	dn = of_find_matching_node(NULL, tegra114_dt_gic_match);
+	if (!dn)
+		return;
+
+	tegra_gic_cpu_base = of_iomap(dn, 1);
+
+	cpu_pm_register_notifier(&tegra_gic_notifier_block);
+}
 #else
 #define tegra_set_wake NULL
+static void tegra114_gic_cpu_pm_registration(void) { }
 #endif
 
 void __init tegra_init_irq(void)
@@ -252,4 +290,6 @@ void __init tegra_init_irq(void)
 	if (!of_have_populated_dt())
 		gic_init(0, 29, distbase,
 			IO_ADDRESS(TEGRA_ARM_PERIF_BASE + 0x100));
+
+	tegra114_gic_cpu_pm_registration();
 }
-- 
1.8.3.1

[PATCH V3 2/3] ARM: tegra114: add low level support for CPU idle powered-down mode

[Joseph Lo]

The flow controller would take care the power sequence when CPU idle in
powered-down mode. It powered gate the CPU when CPU runs into WFI
instruction. And wake up the CPU when event be triggered.

The sequence is below.
* setting wfi bitmap for the CPU as the halt event in the
  FLOW_CTRL_CPU_HALT_REG to monitor the CPU running into WFI,then power
  gate it
* setting IRQ and FIQ as wake up event to wake up CPU when event triggered

Signed-off-by: Joseph Lo <josephl@nvidia.com>
---
V3:
* no change
V2:
* no change
---
 arch/arm/mach-tegra/flowctrl.h      | 2 ++
 arch/arm/mach-tegra/sleep-tegra30.S | 2 ++
 2 files changed, 4 insertions(+)

diff --git a/arch/arm/mach-tegra/flowctrl.h b/arch/arm/mach-tegra/flowctrl.h
index 7a29bae..e56a950 100644
--- a/arch/arm/mach-tegra/flowctrl.h
+++ b/arch/arm/mach-tegra/flowctrl.h
@@ -28,6 +28,8 @@
 #define FLOW_CTRL_SCLK_RESUME		(1 << 27)
 #define FLOW_CTRL_HALT_CPU_IRQ		(1 << 10)
 #define	FLOW_CTRL_HALT_CPU_FIQ		(1 << 8)
+#define FLOW_CTRL_HALT_GIC_IRQ		(1 << 9)
+#define FLOW_CTRL_HALT_GIC_FIQ		(1 << 8)
 #define FLOW_CTRL_CPU0_CSR		0x8
 #define	FLOW_CTRL_CSR_INTR_FLAG		(1 << 15)
 #define FLOW_CTRL_CSR_EVENT_FLAG	(1 << 14)
diff --git a/arch/arm/mach-tegra/sleep-tegra30.S b/arch/arm/mach-tegra/sleep-tegra30.S
index ada8821..5877f26 100644
--- a/arch/arm/mach-tegra/sleep-tegra30.S
+++ b/arch/arm/mach-tegra/sleep-tegra30.S
@@ -99,6 +99,8 @@ flow_ctrl_setting_for_lp2:
 	cmp	r10, #TEGRA30
 	moveq   r3, #FLOW_CTRL_WAIT_FOR_INTERRUPT	@ For LP2
 	movne	r3, #FLOW_CTRL_WAITEVENT
+	orrne	r3, r3, #FLOW_CTRL_HALT_GIC_IRQ
+	orrne	r3, r3, #FLOW_CTRL_HALT_GIC_FIQ
 flow_ctrl_done:
 	cmp	r10, #TEGRA30
 	str	r3, [r2]
-- 
1.8.3.1

[PATCH V3 3/3] ARM: tegra114: cpuidle: add powered-down state

[Joseph Lo]

This supports CPU core power down on each CPU when CPU idle. When CPU go
into this state, it saves it's context and needs a proper configuration
in flow controller to power gate the CPU when CPU runs into WFI
instruction. And the CPU also needs to set the IRQ as CPU power down idle
wake up event in flow controller.

Signed-off-by: Joseph Lo <josephl@nvidia.com>
---
V3:
* use CPUIDLE_FLAG_TIMER_STOP flag
V2:
* remove some redundant code of memory barrier
* remove the function declaration by rearranging the coding sequence 
---
 arch/arm/mach-tegra/cpuidle-tegra114.c | 48 +++++++++++++++++++++++++++++++++-
 1 file changed, 47 insertions(+), 1 deletion(-)

diff --git a/arch/arm/mach-tegra/cpuidle-tegra114.c b/arch/arm/mach-tegra/cpuidle-tegra114.c
index 1d1c602..658b205 100644
--- a/arch/arm/mach-tegra/cpuidle-tegra114.c
+++ b/arch/arm/mach-tegra/cpuidle-tegra114.c
@@ -17,15 +17,61 @@
 #include <linux/kernel.h>
 #include <linux/module.h>
 #include <linux/cpuidle.h>
+#include <linux/cpu_pm.h>
+#include <linux/clockchips.h>
 
 #include <asm/cpuidle.h>
+#include <asm/suspend.h>
+#include <asm/smp_plat.h>
+
+#include "pm.h"
+#include "sleep.h"
+
+#ifdef CONFIG_PM_SLEEP
+#define TEGRA114_MAX_STATES 2
+#else
+#define TEGRA114_MAX_STATES 1
+#endif
+
+#ifdef CONFIG_PM_SLEEP
+static int tegra114_idle_power_down(struct cpuidle_device *dev,
+				    struct cpuidle_driver *drv,
+				    int index)
+{
+	local_fiq_disable();
+
+	tegra_set_cpu_in_lp2();
+	cpu_pm_enter();
+
+	cpu_suspend(0, tegra30_sleep_cpu_secondary_finish);
+
+	cpu_pm_exit();
+	tegra_clear_cpu_in_lp2();
+
+	local_fiq_enable();
+
+	return index;
+}
+#endif
 
 static struct cpuidle_driver tegra_idle_driver = {
 	.name = "tegra_idle",
 	.owner = THIS_MODULE,
-	.state_count = 1,
+	.state_count = TEGRA114_MAX_STATES,
 	.states = {
 		[0] = ARM_CPUIDLE_WFI_STATE_PWR(600),
+#ifdef CONFIG_PM_SLEEP
+		[1] = {
+			.enter			= tegra114_idle_power_down,
+			.exit_latency		= 500,
+			.target_residency	= 1000,
+			.power_usage		= 0,
+			.flags			= CPUIDLE_FLAG_TIME_VALID |
+						  CPUIDLE_FLAG_TIMER_STOP,
+			.name			= "powered-down",
+			.desc			= "CPU power gated",
+		},
+#endif
 	},
 };
 
-- 
1.8.3.1

[PATCH V3 1/3] ARM: tegra114: Reprogram GIC CPU interface to bypass IRQ on CPU PM entry

[Stephen Warren]

On 06/25/2013 03:23 AM, Joseph Lo wrote:
> There is a difference between GICv1 and v2 when CPU in power management
> mode (aka CPU power down on Tegra). For GICv1, IRQ/FIQ interrupt lines
> going to CPU are same lines which are also used for wake-interrupt.
> Therefore, we cannot disable the GIC CPU interface if we need to use same
> interrupts for CPU wake purpose. This creates a race condition for CPU
> power off entry. Also, in GICv1, disabling GICv1 CPU interface puts GICv1
> into bypass mode such that incoming legacy IRQ/FIQ are sent to CPU, which
> means disabling GIC CPU interface doesn't really disable IRQ/FIQ to CPU.
> 
> GICv2 provides a wake IRQ/FIQ (for wake-event purpose), which are not
> disabled by GIC CPU interface. This is done by adding a bypass override
> capability when the interrupts are disabled at the CPU interface. To
> support this, there are four bits about IRQ/FIQ BypassDisable in CPU
> interface Control Register. When the IRQ/FIQ not being driver by the
> CPU interface, each interrupt output signal can be deasserted rather
> than being driven by the legacy interrupt input.
> 
...
> diff --git a/arch/arm/mach-tegra/irq.c b/arch/arm/mach-tegra/irq.c

> +static int tegra_gic_notifier(struct notifier_block *self,
> +			      unsigned long cmd, void *v)
> +{
> +	switch (cmd) {
> +	case CPU_PM_ENTER:
> +		writel_relaxed(0x1E0, tegra_gic_cpu_base + GIC_CPU_CTRL);

I assume that 0x1e0 is the "four bits" for IRQ/FIQ bypass disable that
are mentioned in the commit description. are there #defines that can be
used instead of literal 0x1e0 here?

[PATCH V3 1/3] ARM: tegra114: Reprogram GIC CPU interface to bypass IRQ on CPU PM entry

[Joseph Lo]

On Wed, 2013-06-26 at 07:02 +0800, Stephen Warren wrote:
> On 06/25/2013 03:23 AM, Joseph Lo wrote:
> > There is a difference between GICv1 and v2 when CPU in power management
> > mode (aka CPU power down on Tegra). For GICv1, IRQ/FIQ interrupt lines
> > going to CPU are same lines which are also used for wake-interrupt.
> > Therefore, we cannot disable the GIC CPU interface if we need to use same
> > interrupts for CPU wake purpose. This creates a race condition for CPU
> > power off entry. Also, in GICv1, disabling GICv1 CPU interface puts GICv1
> > into bypass mode such that incoming legacy IRQ/FIQ are sent to CPU, which
> > means disabling GIC CPU interface doesn't really disable IRQ/FIQ to CPU.
> > 
> > GICv2 provides a wake IRQ/FIQ (for wake-event purpose), which are not
> > disabled by GIC CPU interface. This is done by adding a bypass override
> > capability when the interrupts are disabled at the CPU interface. To
> > support this, there are four bits about IRQ/FIQ BypassDisable in CPU
> > interface Control Register. When the IRQ/FIQ not being driver by the
> > CPU interface, each interrupt output signal can be deasserted rather
> > than being driven by the legacy interrupt input.
> > 
> ...
> > diff --git a/arch/arm/mach-tegra/irq.c b/arch/arm/mach-tegra/irq.c
> 
> > +static int tegra_gic_notifier(struct notifier_block *self,
> > +			      unsigned long cmd, void *v)
> > +{
> > +	switch (cmd) {
> > +	case CPU_PM_ENTER:
> > +		writel_relaxed(0x1E0, tegra_gic_cpu_base + GIC_CPU_CTRL);
> 
> I assume that 0x1e0 is the "four bits" for IRQ/FIQ bypass disable that
> are mentioned in the commit description. are there #defines that can be
> used instead of literal 0x1e0 here?

The common "arm-gic.h" didn't have the new define for GICv2 yet. So I
didn't define something like GIC_CPU_{FIQ,IRQ}BypDisGrp{0,1} for Tegra.

Thanks,
Joseph

[PATCH V3 0/3] ARM: tegra114: cpuidle: add power down state

[Stephen Warren]

On 06/25/2013 03:23 AM, Joseph Lo wrote:
> This series introduce CPU core power down state for CPU idle. When CPU go
> into this state, it saves it's context and needs a proper configuration
> in flow controller to power gate the CPU when CPU runs into WFI
> instruction. And the CPU also needs to set the IRQ as CPU power down idle
> wake up event in flow controller.
> 
> To prevent race conditions and ensure proper interrupt routing on
> Cortex-A15 CPUs when they are power-gated, add a CPU PM notifier
> call-back to reprogram the GIC CPU interface on PM entry. The
> GIC CPU interface will be reset back to its normal state by
> the common GIC CPU PM exit callback when the CPU wakes up.
> 
> This series depends on the patch of "tick:  Fix
> tick_broadcast_pending_mask not cleared".

Joseph, I applied all 3 of your patches/series on top of next-20130701,
which does include dependency "tick:  Fix tick_broadcast_pending_mask
not cleared":

* ARM: tegra: cpuidle: use CPUIDLE_FLAG_TIMER_STOP flag
* This series
* ARM: tegra114: add support for system suspend

I find that at least on on Tegra114/Dalmore, CPU hotplug doesn't work
correctly.

If I unplug CPU0, I get an immediate system hang:

root at localhost:~# ./cpuonline.py
echo 0 > /sys/devices/system/cpu/cpu0/online
[   73.555696] IRQ73 no longer affine to CPU0
[   73.561965] CPU0: shutdown

If I adjust my script not to hotplug CPU0, I get a hang when plugging a
CPU back in:

root at localhost:~# ./cpuonline.py
echo 0 > /sys/devices/system/cpu/cpu2/online
[   87.141128] CPU2: shutdown
echo 0 > /sys/devices/system/cpu/cpu1/online
[   89.227597] CPU1: shutdown
echo 0 > /sys/devices/system/cpu/cpu3/online
[   91.253870] CPU3: shutdown
echo 1 > /sys/devices/system/cpu/cpu1/online

or simpler:

root at localhost:~# echo 0 > /sys/devices/system/cpu/cpu1/online
[   80.767825] CPU1: shutdown
root at localhost:~# echo 1 > /sys/devices/system/cpu/cpu1/online

Finally, if I boot afresh and enter system suspend, I see the system
suspend OK, but pressing the power button on either the board itself or
the attached PM342 debug board will not wake the system from sleep.

[PATCH V3 0/3] ARM: tegra114: cpuidle: add power down state

[Joseph Lo]

On Tue, 2013-07-02 at 01:38 +0800, Stephen Warren wrote:
> On 06/25/2013 03:23 AM, Joseph Lo wrote:
> > This series introduce CPU core power down state for CPU idle. When CPU go
> > into this state, it saves it's context and needs a proper configuration
> > in flow controller to power gate the CPU when CPU runs into WFI
> > instruction. And the CPU also needs to set the IRQ as CPU power down idle
> > wake up event in flow controller.
> > 
> > To prevent race conditions and ensure proper interrupt routing on
> > Cortex-A15 CPUs when they are power-gated, add a CPU PM notifier
> > call-back to reprogram the GIC CPU interface on PM entry. The
> > GIC CPU interface will be reset back to its normal state by
> > the common GIC CPU PM exit callback when the CPU wakes up.
> > 
> > This series depends on the patch of "tick:  Fix
> > tick_broadcast_pending_mask not cleared".
> 
> Joseph, I applied all 3 of your patches/series on top of next-20130701,
> which does include dependency "tick:  Fix tick_broadcast_pending_mask
> not cleared":
> 
> * ARM: tegra: cpuidle: use CPUIDLE_FLAG_TIMER_STOP flag
> * This series
> * ARM: tegra114: add support for system suspend
> 
> I find that at least on on Tegra114/Dalmore, CPU hotplug doesn't work
> correctly.
> 
Yes, it's due to the CPU hotplug function broken in next-20130701. The
last working version was next-20130624 that I had tested and verified.
It also breaks the resume function when enable non-boot CPUs. You can
verify the suspend function by un-plug secondary CPUs first. I am
checking what cause the CPU hotplug function broken on next-20130701.

Thanks,
Joseph