Skip to contentSkip to conversation

ARM: OMAP2+: omap_device: maintain sane runtime pm status around suspend/resume

PATCH Superseded Nishanth Menon 6 participants 16 messages Nov 12, 2013 – Nov 19, 2013

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 ARM: OMAP2+: omap_device: maintain sane runtime pm status around suspend/resume

OMAP device hooks around suspend|resume_noirq ensures that hwmod devices are forced to idle using omap_device_idle/enable as part of the last stage of suspend activity.

For a device such as i2c who uses autosuspend, it is possible to enter the suspend path with dev->power.runtime_status = RPM_ACTIVE.

As part of the suspend flow, the generic runtime logic would increment it's dev->power.disable_depth to 1. This should prevent further pm_runtime_get_sync from succeeding once the runtime_status has been set to RPM_SUSPENDED.

Now, as part of the suspend_noirq handler in omap_device, we force the
following: if the device status is !suspended, we force the device to idle using omap_device_idle (clocks are cut etc..). This ensures that from a hardware perspective, the device is "suspended". However, runtime_status is left to be active.

*if* an operation is attempted after this point to pm_runtime_get_sync, runtime framework depends on runtime_status to indicate accurately the device status, and since it sees it to be ACTIVE, it assumes the module is functional and returns a non-error value. As a result the user will see pm_runtime_get succeed, however a register access will crash due to the lack of clocks.

To prevent this from happening, we should ensure that runtime_status exactly indicates the device status. As a result of this change any further calls to pm_runtime_get* would return -EACCES (since disable_depth is 1). On resume, we restore the clocks and runtime status exactly as we suspended with.

---
Changes in V2 (resend):
- dropped the unnecessary flag for runtime status restore
- picked up kevin's ack from https://patchwork.kernel.org/patch/3168371/

v1: https://patchwork.kernel.org/patch/3154501/

patch baseline: V3.12 tag (also applies on linux-next next-20131107 tag)

Logs from 3.12 based vendor kernel:
Before: http://pastebin.com/m5KxnB7a
After: http://pastebin.com/8AfX4e7r
The discussion on cpufreq side of the story which triggered this scenario:
http://marc.info/?l=linux-pm&m=138263811321921&w=2

Tested on TI vendor kernel (with dt boot):
AM335x: evm, BBB, sk, BBW
OMAP5uEVM, DRA7-evm

 arch/arm/mach-omap2/omap_device.c |    5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
arch/arm/mach-omap2/omap_device.c +3 -2
index b69dd9a..f97b34b 100644
@@ -621,6 +621,7 @@ static int _od_suspend_noirq(struct device *dev)
 
 	if (!ret && !pm_runtime_status_suspended(dev)) {
 		if (pm_generic_runtime_suspend(dev) == 0) {
+			pm_runtime_set_suspended(dev);
 			omap_device_idle(pdev);
 			od->flags |= OMAP_DEVICE_SUSPENDED;
 		}
@@ -634,10 +635,10 @@ static int _od_resume_noirq(struct device *dev)
 	struct platform_device *pdev = to_platform_device(dev);
 	struct omap_device *od = to_omap_device(pdev);
 
-	if ((od->flags & OMAP_DEVICE_SUSPENDED) &&
-	    !pm_runtime_status_suspended(dev)) {
+	if (od->flags & OMAP_DEVICE_SUSPENDED) {
 		od->flags &= ~OMAP_DEVICE_SUSPENDED;
 		omap_device_enable(pdev);
+		pm_runtime_set_active(dev);
 		pm_generic_runtime_resume(dev);
 	}
 
Signature · 1 line
1.7.9.5
#1 lore raw

Hi,

arch/arm/mach-omap2/omap_device.c static int _od_suspend_noirq(struct device *dev) · line 624

Quoted 9 lines · Nishanth Menon
arch/arm/mach-omap2/omap_device.c +1 -0
index b69dd9a..f97b34b 100644
@@ -621,6 +621,7 @@ static int _od_suspend_noirq(struct device *dev)
 
 	if (!ret && !pm_runtime_status_suspended(dev)) {
 		if (pm_generic_runtime_suspend(dev) == 0) {
+			pm_runtime_set_suspended(dev);

don't you have to disable pm_runtime around status changes ? Or is pm_runtime already disabled by the time we get here ?

arch/arm/mach-omap2/omap_device.c static int _od_resume_noirq(struct device *dev) · line 641

Quoted 10 lines
arch/arm/mach-omap2/omap_device.c +2 -2
@@ -634,10 +635,10 @@ static int _od_resume_noirq(struct device *dev)
 	struct platform_device *pdev = to_platform_device(dev);
 	struct omap_device *od = to_omap_device(pdev);
 
-	if ((od->flags & OMAP_DEVICE_SUSPENDED) &&
-	    !pm_runtime_status_suspended(dev)) {
+	if (od->flags & OMAP_DEVICE_SUSPENDED) {
 		od->flags &= ~OMAP_DEVICE_SUSPENDED;
 		omap_device_enable(pdev);
+		pm_runtime_set_active(dev);

ditto, also pm_runtime_set_active() may fail.

Signature · 8 lines
balbi
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 836 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20131113/04bc436c/attachment.sig>
#2 lore raw
Quoted 9 lines · Felipe Balbi

Hi,

Quoted 6 lines · Nishanth Menon
arch/arm/mach-omap2/omap_device.c +0 -0
index b69dd9a..f97b34b 100644
@@ -621,6 +621,7 @@ static int _od_suspend_noirq(struct device *dev)
 
Quoted 6 lines

arch/arm/mach-omap2/omap_device.c static int _od_suspend_noirq(struct device *dev) · line 624

Quoted 3 lines
arch/arm/mach-omap2/omap_device.c +1 -0
 	if (!ret && !pm_runtime_status_suspended(dev)) {
 		if (pm_generic_runtime_suspend(dev) == 0) {
+			pm_runtime_set_suspended(dev);

don't you have to disable pm_runtime around status changes ? Or is pm_runtime already disabled by the time we get here ?

pm_runtime is already disabled by the time no_irq suspend is invoked.

Quoted 5 lines
Quoted 5 lines
arch/arm/mach-omap2/omap_device.c +0 -0
@@ -634,10 +635,10 @@ static int _od_resume_noirq(struct device *dev)
 	struct platform_device *pdev = to_platform_device(dev);
 	struct omap_device *od = to_omap_device(pdev);
 
Quoted 9 lines

arch/arm/mach-omap2/omap_device.c static int _od_resume_noirq(struct device *dev) · line 641

Quoted 6 lines
arch/arm/mach-omap2/omap_device.c +2 -2
-	if ((od->flags & OMAP_DEVICE_SUSPENDED) &&
-	    !pm_runtime_status_suspended(dev)) {
+	if (od->flags & OMAP_DEVICE_SUSPENDED) {
 		od->flags &= ~OMAP_DEVICE_SUSPENDED;
 		omap_device_enable(pdev);
+		pm_runtime_set_active(dev);

ditto, also pm_runtime_set_active() may fail.

again, pm_runtime is not yet active here yet - we just restore the pm runtime state with which we went down with -> and that is not expected to fail either - So, how about just adding a WARN if our expectation of balanced operation was somehow broken in the future with changes to runtime framework?

Signature · 2 lines
Regards,
Nishanth Menon
#3 lore raw
Quoted 33 lines · Nishanth Menon
Quoted 9 lines · Felipe Balbi

Hi,

Quoted 6 lines · Nishanth Menon
arch/arm/mach-omap2/omap_device.c +0 -0
index b69dd9a..f97b34b 100644
@@ -621,6 +621,7 @@ static int _od_suspend_noirq(struct device *dev)
 
Quoted 6 lines

arch/arm/mach-omap2/omap_device.c static int _od_suspend_noirq(struct device *dev) · line 624

Quoted 3 lines
arch/arm/mach-omap2/omap_device.c +1 -0
 	if (!ret && !pm_runtime_status_suspended(dev)) {
 		if (pm_generic_runtime_suspend(dev) == 0) {
+			pm_runtime_set_suspended(dev);

don't you have to disable pm_runtime around status changes ? Or is pm_runtime already disabled by the time we get here ?

pm_runtime is already disabled by the time no_irq suspend is invoked.

Quoted 5 lines
Quoted 5 lines
arch/arm/mach-omap2/omap_device.c +0 -0
@@ -634,10 +635,10 @@ static int _od_resume_noirq(struct device *dev)
 	struct platform_device *pdev = to_platform_device(dev);
 	struct omap_device *od = to_omap_device(pdev);
 
Quoted 9 lines

arch/arm/mach-omap2/omap_device.c static int _od_resume_noirq(struct device *dev) · line 641

Quoted 6 lines
arch/arm/mach-omap2/omap_device.c +2 -2
-	if ((od->flags & OMAP_DEVICE_SUSPENDED) &&
-	    !pm_runtime_status_suspended(dev)) {
+	if (od->flags & OMAP_DEVICE_SUSPENDED) {
 		od->flags &= ~OMAP_DEVICE_SUSPENDED;
 		omap_device_enable(pdev);
+		pm_runtime_set_active(dev);

ditto, also pm_runtime_set_active() may fail.

Quoted 5 lines

again, pm_runtime is not yet active here yet - we just restore the pm runtime state with which we went down with -> and that is not expected to fail either - So, how about just adding a WARN if our expectation of balanced operation was somehow broken in the future with changes to runtime framework?

you mean:

WARN(pm_runtime_set_active(dev)); ?

sounds good

thanks

Signature · 8 lines
balbi
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 836 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20131113/01bd71b0/attachment.sig>
#4 lore raw
Quoted 40 lines · Felipe Balbi
Quoted 33 lines · Nishanth Menon
Quoted 9 lines · Felipe Balbi

Hi,

Quoted 6 lines · Nishanth Menon
arch/arm/mach-omap2/omap_device.c +0 -0
index b69dd9a..f97b34b 100644
@@ -621,6 +621,7 @@ static int _od_suspend_noirq(struct device *dev)
 
Quoted 6 lines

arch/arm/mach-omap2/omap_device.c static int _od_suspend_noirq(struct device *dev) · line 624

Quoted 3 lines
arch/arm/mach-omap2/omap_device.c +1 -0
 	if (!ret && !pm_runtime_status_suspended(dev)) {
 		if (pm_generic_runtime_suspend(dev) == 0) {
+			pm_runtime_set_suspended(dev);

don't you have to disable pm_runtime around status changes ? Or is pm_runtime already disabled by the time we get here ?

pm_runtime is already disabled by the time no_irq suspend is invoked.

Quoted 5 lines
Quoted 5 lines
arch/arm/mach-omap2/omap_device.c +0 -0
@@ -634,10 +635,10 @@ static int _od_resume_noirq(struct device *dev)
 	struct platform_device *pdev = to_platform_device(dev);
 	struct omap_device *od = to_omap_device(pdev);
 
Quoted 9 lines

arch/arm/mach-omap2/omap_device.c static int _od_resume_noirq(struct device *dev) · line 641

Quoted 6 lines
arch/arm/mach-omap2/omap_device.c +2 -2
-	if ((od->flags & OMAP_DEVICE_SUSPENDED) &&
-	    !pm_runtime_status_suspended(dev)) {
+	if (od->flags & OMAP_DEVICE_SUSPENDED) {
 		od->flags &= ~OMAP_DEVICE_SUSPENDED;
 		omap_device_enable(pdev);
+		pm_runtime_set_active(dev);

ditto, also pm_runtime_set_active() may fail.

Quoted 5 lines

again, pm_runtime is not yet active here yet - we just restore the pm runtime state with which we went down with -> and that is not expected to fail either - So, how about just adding a WARN if our expectation of balanced operation was somehow broken in the future with changes to runtime framework?

Quoted 3 lines

you mean:

WARN(pm_runtime_set_active(dev)); ?

yes

Quoted 2 lines

sounds good

Thanks. Will post a v3 tomorrow morning to give a chance for discussions on further comments if any.

Signature · 2 lines
Regards,
Nishanth Menon
#5 lore raw
In reply to Nishanth Menon
Quoted 33 lines · Nishanth Menon
Quoted 9 lines · Felipe Balbi

Hi,

Quoted 6 lines · Nishanth Menon
arch/arm/mach-omap2/omap_device.c +0 -0
index b69dd9a..f97b34b 100644
@@ -621,6 +621,7 @@ static int _od_suspend_noirq(struct device *dev)
 
Quoted 6 lines

arch/arm/mach-omap2/omap_device.c static int _od_suspend_noirq(struct device *dev) · line 624

Quoted 3 lines
arch/arm/mach-omap2/omap_device.c +1 -0
 	if (!ret && !pm_runtime_status_suspended(dev)) {
 		if (pm_generic_runtime_suspend(dev) == 0) {
+			pm_runtime_set_suspended(dev);

don't you have to disable pm_runtime around status changes ? Or is pm_runtime already disabled by the time we get here ?

pm_runtime is already disabled by the time no_irq suspend is invoked.

Quoted 5 lines
Quoted 5 lines
arch/arm/mach-omap2/omap_device.c +0 -0
@@ -634,10 +635,10 @@ static int _od_resume_noirq(struct device *dev)
 	struct platform_device *pdev = to_platform_device(dev);
 	struct omap_device *od = to_omap_device(pdev);
 
Quoted 9 lines

arch/arm/mach-omap2/omap_device.c static int _od_resume_noirq(struct device *dev) · line 641

Quoted 6 lines
arch/arm/mach-omap2/omap_device.c +2 -2
-	if ((od->flags & OMAP_DEVICE_SUSPENDED) &&
-	    !pm_runtime_status_suspended(dev)) {
+	if (od->flags & OMAP_DEVICE_SUSPENDED) {
 		od->flags &= ~OMAP_DEVICE_SUSPENDED;
 		omap_device_enable(pdev);
+		pm_runtime_set_active(dev);

ditto, also pm_runtime_set_active() may fail.

Quoted 5 lines

again, pm_runtime is not yet active here yet - we just restore the pm runtime state with which we went down with -> and that is not expected to fail either - So, how about just adding a WARN if our expectation of balanced operation was somehow broken in the future with changes to runtime framework?

And also a note in the changelog (or comment at the WARN) about the assumption that runtime PM is disabled at this point.

Kevin

#6 lore raw
Quoted 35 lines · Kevin Hilman
Quoted 33 lines · Nishanth Menon
Quoted 9 lines · Felipe Balbi

Hi,

Quoted 6 lines · Nishanth Menon
arch/arm/mach-omap2/omap_device.c +0 -0
index b69dd9a..f97b34b 100644
@@ -621,6 +621,7 @@ static int _od_suspend_noirq(struct device *dev)
 
Quoted 6 lines

arch/arm/mach-omap2/omap_device.c static int _od_suspend_noirq(struct device *dev) · line 624

Quoted 3 lines
arch/arm/mach-omap2/omap_device.c +1 -0
    if (!ret && !pm_runtime_status_suspended(dev)) {
            if (pm_generic_runtime_suspend(dev) == 0) {
+                   pm_runtime_set_suspended(dev);

don't you have to disable pm_runtime around status changes ? Or is pm_runtime already disabled by the time we get here ?

pm_runtime is already disabled by the time no_irq suspend is invoked.

Quoted 14 lines

arch/arm/mach-omap2/omap_device.c static int _od_resume_noirq(struct device *dev) · line 641

Quoted 11 lines
arch/arm/mach-omap2/omap_device.c +2 -2
@@ -634,10 +635,10 @@ static int _od_resume_noirq(struct device *dev)
    struct platform_device *pdev = to_platform_device(dev);
    struct omap_device *od = to_omap_device(pdev);
 
-   if ((od->flags & OMAP_DEVICE_SUSPENDED) &&
-       !pm_runtime_status_suspended(dev)) {
+   if (od->flags & OMAP_DEVICE_SUSPENDED) {
            od->flags &= ~OMAP_DEVICE_SUSPENDED;
            omap_device_enable(pdev);
+           pm_runtime_set_active(dev);

ditto, also pm_runtime_set_active() may fail.

Quoted 8 lines
Quoted 5 lines

again, pm_runtime is not yet active here yet - we just restore the pm runtime state with which we went down with -> and that is not expected to fail either - So, how about just adding a WARN if our expectation of balanced operation was somehow broken in the future with changes to runtime framework?

And also a note in the changelog (or comment at the WARN) about the assumption that runtime PM is disabled at this point.

Ofcourse. will do.

--
Regards,
Nishanth Menon

#7 lore raw

PATCH v3 1/1 ARM: OMAP2+: omap_device: maintain sane runtime pm status around suspend/resume

In reply to Nishanth Menon

OMAP device hooks around suspend|resume_noirq ensures that hwmod devices are forced to idle using omap_device_idle/enable as part of the last stage of suspend activity.

For a device such as i2c who uses autosuspend, it is possible to enter the suspend path with dev->power.runtime_status = RPM_ACTIVE.

As part of the suspend flow, the generic runtime logic would increment it's dev->power.disable_depth to 1. This should prevent further pm_runtime_get_sync from succeeding once the runtime_status has been set to RPM_SUSPENDED.

Now, as part of the suspend_noirq handler in omap_device, we force the
following: if the device status is !suspended, we force the device to idle using omap_device_idle (clocks are cut etc..). This ensures that from a hardware perspective, the device is "suspended". However, runtime_status is left to be active.

*if* an operation is attempted after this point to pm_runtime_get_sync, runtime framework depends on runtime_status to indicate accurately the device status, and since it sees it to be ACTIVE, it assumes the module is functional and returns a non-error value. As a result the user will see pm_runtime_get succeed, however a register access will crash due to the lack of clocks.

To prevent this from happening, we should ensure that runtime_status exactly indicates the device status. As a result of this change any further calls to pm_runtime_get* would return -EACCES (since disable_depth is 1). On resume, we restore the clocks and runtime status exactly as we suspended with. These operations are not expected to fail as we update the states after the core runtime framework has suspended itself and restore before the core runtime framework has resumed.

---
Changes in V3:
- Added WARN in case of unexpected failure of runtime pm status restore
v2: https://patchwork.kernel.org/patch/3176501/
v1: https://patchwork.kernel.org/patch/3154501/

patch baseline: V3.12 tag (also applies on linux-next next-20131107 tag)

 arch/arm/mach-omap2/omap_device.c |   13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)
arch/arm/mach-omap2/omap_device.c +11 -2
index b69dd9a..53f0735 100644
@@ -621,6 +621,7 @@ static int _od_suspend_noirq(struct device *dev)
 
 	if (!ret && !pm_runtime_status_suspended(dev)) {
 		if (pm_generic_runtime_suspend(dev) == 0) {
+			pm_runtime_set_suspended(dev);
 			omap_device_idle(pdev);
 			od->flags |= OMAP_DEVICE_SUSPENDED;
 		}
@@ -634,10 +635,18 @@ static int _od_resume_noirq(struct device *dev)
 	struct platform_device *pdev = to_platform_device(dev);
 	struct omap_device *od = to_omap_device(pdev);
 
-	if ((od->flags & OMAP_DEVICE_SUSPENDED) &&
-	    !pm_runtime_status_suspended(dev)) {
+	if (od->flags & OMAP_DEVICE_SUSPENDED) {
 		od->flags &= ~OMAP_DEVICE_SUSPENDED;
 		omap_device_enable(pdev);
+		/*
+		 * XXX: we run before core runtime pm has resumed itself. At
+		 * this point in time, we just restore the runtime pm state and
+		 * considering symmetric operations in resume, we donot expect
+		 * to fail. If we failed, something changed in core runtime_pm
+		 * framework OR some device driver messed things up, hence, WARN
+		 */
+		WARN(pm_runtime_set_active(dev),
+		     "Could not set %s runtime state active\n", dev_name(dev));
 		pm_generic_runtime_resume(dev);
 	}
 
Signature · 1 line
1.7.9.5
Reviewed-by on v3 1/1
#8 lore raw

Hi,

Quoted 69 lines · Nishanth Menon

OMAP device hooks around suspend|resume_noirq ensures that hwmod devices are forced to idle using omap_device_idle/enable as part of the last stage of suspend activity.

For a device such as i2c who uses autosuspend, it is possible to enter the suspend path with dev->power.runtime_status = RPM_ACTIVE.

As part of the suspend flow, the generic runtime logic would increment it's dev->power.disable_depth to 1. This should prevent further pm_runtime_get_sync from succeeding once the runtime_status has been set to RPM_SUSPENDED.

Now, as part of the suspend_noirq handler in omap_device, we force the
following: if the device status is !suspended, we force the device to idle using omap_device_idle (clocks are cut etc..). This ensures that from a hardware perspective, the device is "suspended". However, runtime_status is left to be active.

*if* an operation is attempted after this point to pm_runtime_get_sync, runtime framework depends on runtime_status to indicate accurately the device status, and since it sees it to be ACTIVE, it assumes the module is functional and returns a non-error value. As a result the user will see pm_runtime_get succeed, however a register access will crash due to the lack of clocks.

To prevent this from happening, we should ensure that runtime_status exactly indicates the device status. As a result of this change any further calls to pm_runtime_get* would return -EACCES (since disable_depth is 1). On resume, we restore the clocks and runtime status exactly as we suspended with. These operations are not expected to fail as we update the states after the core runtime framework has suspended itself and restore before the core runtime framework has resumed.

---
Changes in V3:
- Added WARN in case of unexpected failure of runtime pm status restore
v2: https://patchwork.kernel.org/patch/3176501/
v1: https://patchwork.kernel.org/patch/3154501/

patch baseline: V3.12 tag (also applies on linux-next next-20131107 tag)

 arch/arm/mach-omap2/omap_device.c |   13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)
arch/arm/mach-omap2/omap_device.c +2 -2
index b69dd9a..53f0735 100644
@@ -621,6 +621,7 @@ static int _od_suspend_noirq(struct device *dev)
 
 	if (!ret && !pm_runtime_status_suspended(dev)) {
 		if (pm_generic_runtime_suspend(dev) == 0) {
+			pm_runtime_set_suspended(dev);
 			omap_device_idle(pdev);
 			od->flags |= OMAP_DEVICE_SUSPENDED;
 		}
@@ -634,10 +635,18 @@ static int _od_resume_noirq(struct device *dev)
 	struct platform_device *pdev = to_platform_device(dev);
 	struct omap_device *od = to_omap_device(pdev);
 
-	if ((od->flags & OMAP_DEVICE_SUSPENDED) &&
-	    !pm_runtime_status_suspended(dev)) {
+	if (od->flags & OMAP_DEVICE_SUSPENDED) {
 		od->flags &= ~OMAP_DEVICE_SUSPENDED;

arch/arm/mach-omap2/omap_device.c static int _od_resume_noirq(struct device *dev) · line 649

Quoted 10 lines
arch/arm/mach-omap2/omap_device.c +9 -0
 		omap_device_enable(pdev);
+		/*
+		 * XXX: we run before core runtime pm has resumed itself. At
+		 * this point in time, we just restore the runtime pm state and
+		 * considering symmetric operations in resume, we donot expect
+		 * to fail. If we failed, something changed in core runtime_pm
+		 * framework OR some device driver messed things up, hence, WARN
+		 */
+		WARN(pm_runtime_set_active(dev),
+		     "Could not set %s runtime state active\n", dev_name(dev));

if you want to print the device name, how about dev_WARN() ?

no strong feelings though:

Signature · 8 lines
balbi
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 836 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20131114/843ad991/attachment-0001.sig>
#9 lore raw
Quoted 85 lines · Felipe Balbi

Hi,

Quoted 69 lines · Nishanth Menon

OMAP device hooks around suspend|resume_noirq ensures that hwmod devices are forced to idle using omap_device_idle/enable as part of the last stage of suspend activity.

For a device such as i2c who uses autosuspend, it is possible to enter the suspend path with dev->power.runtime_status = RPM_ACTIVE.

As part of the suspend flow, the generic runtime logic would increment it's dev->power.disable_depth to 1. This should prevent further pm_runtime_get_sync from succeeding once the runtime_status has been set to RPM_SUSPENDED.

Now, as part of the suspend_noirq handler in omap_device, we force the
following: if the device status is !suspended, we force the device to idle using omap_device_idle (clocks are cut etc..). This ensures that from a hardware perspective, the device is "suspended". However, runtime_status is left to be active.

*if* an operation is attempted after this point to pm_runtime_get_sync, runtime framework depends on runtime_status to indicate accurately the device status, and since it sees it to be ACTIVE, it assumes the module is functional and returns a non-error value. As a result the user will see pm_runtime_get succeed, however a register access will crash due to the lack of clocks.

To prevent this from happening, we should ensure that runtime_status exactly indicates the device status. As a result of this change any further calls to pm_runtime_get* would return -EACCES (since disable_depth is 1). On resume, we restore the clocks and runtime status exactly as we suspended with. These operations are not expected to fail as we update the states after the core runtime framework has suspended itself and restore before the core runtime framework has resumed.

---
Changes in V3:
- Added WARN in case of unexpected failure of runtime pm status restore
v2: https://patchwork.kernel.org/patch/3176501/
v1: https://patchwork.kernel.org/patch/3154501/

patch baseline: V3.12 tag (also applies on linux-next next-20131107 tag)

 arch/arm/mach-omap2/omap_device.c |   13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)
arch/arm/mach-omap2/omap_device.c +2 -2
index b69dd9a..53f0735 100644
@@ -621,6 +621,7 @@ static int _od_suspend_noirq(struct device *dev)
 
 	if (!ret && !pm_runtime_status_suspended(dev)) {
 		if (pm_generic_runtime_suspend(dev) == 0) {
+			pm_runtime_set_suspended(dev);
 			omap_device_idle(pdev);
 			od->flags |= OMAP_DEVICE_SUSPENDED;
 		}
@@ -634,10 +635,18 @@ static int _od_resume_noirq(struct device *dev)
 	struct platform_device *pdev = to_platform_device(dev);
 	struct omap_device *od = to_omap_device(pdev);
 
-	if ((od->flags & OMAP_DEVICE_SUSPENDED) &&
-	    !pm_runtime_status_suspended(dev)) {
+	if (od->flags & OMAP_DEVICE_SUSPENDED) {
 		od->flags &= ~OMAP_DEVICE_SUSPENDED;

arch/arm/mach-omap2/omap_device.c static int _od_resume_noirq(struct device *dev) · line 649

Quoted 10 lines
arch/arm/mach-omap2/omap_device.c +9 -0
 		omap_device_enable(pdev);
+		/*
+		 * XXX: we run before core runtime pm has resumed itself. At
+		 * this point in time, we just restore the runtime pm state and
+		 * considering symmetric operations in resume, we donot expect
+		 * to fail. If we failed, something changed in core runtime_pm
+		 * framework OR some device driver messed things up, hence, WARN
+		 */
+		WARN(pm_runtime_set_active(dev),
+		     "Could not set %s runtime state active\n", dev_name(dev));

if you want to print the device name, how about dev_WARN() ?

Quoted 1 line

no strong feelings though:

I would like to stick with WARN as dev_WARN would probably need an condition check.. unless someone has a strong opinion, I dont see it adding value here.

Quoted 3 lines
Signature · 2 lines
Regards,
Nishanth Menon
#10 lore raw
In reply to Nishanth Menon
Quoted 34 lines · Nishanth Menon

OMAP device hooks around suspend|resume_noirq ensures that hwmod devices are forced to idle using omap_device_idle/enable as part of the last stage of suspend activity.

For a device such as i2c who uses autosuspend, it is possible to enter the suspend path with dev->power.runtime_status = RPM_ACTIVE.

As part of the suspend flow, the generic runtime logic would increment it's dev->power.disable_depth to 1. This should prevent further pm_runtime_get_sync from succeeding once the runtime_status has been set to RPM_SUSPENDED.

Now, as part of the suspend_noirq handler in omap_device, we force the
following: if the device status is !suspended, we force the device to idle using omap_device_idle (clocks are cut etc..). This ensures that from a hardware perspective, the device is "suspended". However, runtime_status is left to be active.

*if* an operation is attempted after this point to pm_runtime_get_sync, runtime framework depends on runtime_status to indicate accurately the device status, and since it sees it to be ACTIVE, it assumes the module is functional and returns a non-error value. As a result the user will see pm_runtime_get succeed, however a register access will crash due to the lack of clocks.

To prevent this from happening, we should ensure that runtime_status exactly indicates the device status. As a result of this change any further calls to pm_runtime_get* would return -EACCES (since disable_depth is 1). On resume, we restore the clocks and runtime status exactly as we suspended with. These operations are not expected to fail as we update the states after the core runtime framework has suspended itself and restore before the core runtime framework has resumed.

Quoted 4 lines

Looks reasonable to me. Looks like this should be considered for -stable
- Nishanth, what do you think?

Tony or Kevin, do you want to take this one, or want me to?

- Paul

#11 lore raw
Quoted 36 lines · Paul Walmsley
Quoted 34 lines · Nishanth Menon

OMAP device hooks around suspend|resume_noirq ensures that hwmod devices are forced to idle using omap_device_idle/enable as part of the last stage of suspend activity.

For a device such as i2c who uses autosuspend, it is possible to enter the suspend path with dev->power.runtime_status = RPM_ACTIVE.

As part of the suspend flow, the generic runtime logic would increment it's dev->power.disable_depth to 1. This should prevent further pm_runtime_get_sync from succeeding once the runtime_status has been set to RPM_SUSPENDED.

Now, as part of the suspend_noirq handler in omap_device, we force the
following: if the device status is !suspended, we force the device to idle using omap_device_idle (clocks are cut etc..). This ensures that from a hardware perspective, the device is "suspended". However, runtime_status is left to be active.

*if* an operation is attempted after this point to pm_runtime_get_sync, runtime framework depends on runtime_status to indicate accurately the device status, and since it sees it to be ACTIVE, it assumes the module is functional and returns a non-error value. As a result the user will see pm_runtime_get succeed, however a register access will crash due to the lack of clocks.

To prevent this from happening, we should ensure that runtime_status exactly indicates the device status. As a result of this change any further calls to pm_runtime_get* would return -EACCES (since disable_depth is 1). On resume, we restore the clocks and runtime status exactly as we suspended with. These operations are not expected to fail as we update the states after the core runtime framework has suspended itself and restore before the core runtime framework has resumed.

Quoted 7 lines
Quoted 4 lines

Looks reasonable to me. Looks like this should be considered for -stable
- Nishanth, what do you think?

Every product kernel since 3.4 needed to be hacked (we have hacked in different ways so far) to work around this (since we never spend time digging deeper :( ), So, I do agree with your view that a -stable tag will be most beneficial.

Quoted 6 lines

Tony or Kevin, do you want to take this one, or want me to?

- Paul

Signature · 2 lines
Regards,
Nishanth Menon
maintainer
#12 lore raw
Quoted 42 lines · Nishanth Menon
Quoted 41 lines · Paul Walmsley
Quoted 38 lines · Nishanth Menon

OMAP device hooks around suspend|resume_noirq ensures that hwmod devices are forced to idle using omap_device_idle/enable as part of the last stage of suspend activity.

For a device such as i2c who uses autosuspend, it is possible to enter the suspend path with dev->power.runtime_status = RPM_ACTIVE.

As part of the suspend flow, the generic runtime logic would increment it's dev->power.disable_depth to 1. This should prevent further pm_runtime_get_sync from succeeding once the runtime_status has been set to RPM_SUSPENDED.

Now, as part of the suspend_noirq handler in omap_device, we force the
following: if the device status is !suspended, we force the device to idle using omap_device_idle (clocks are cut etc..). This ensures that from a hardware perspective, the device is "suspended". However, runtime_status is left to be active.

*if* an operation is attempted after this point to pm_runtime_get_sync, runtime framework depends on runtime_status to indicate accurately the device status, and since it sees it to be ACTIVE, it assumes the module is functional and returns a non-error value. As a result the user will see pm_runtime_get succeed, however a register access will crash due to the lack of clocks.

To prevent this from happening, we should ensure that runtime_status exactly indicates the device status. As a result of this change any further calls to pm_runtime_get* would return -EACCES (since disable_depth is 1). On resume, we restore the clocks and runtime status exactly as we suspended with. These operations are not expected to fail as we update the states after the core runtime framework has suspended itself and restore before the core runtime framework has resumed.

Quoted 10 lines
Quoted 2 lines

Looks reasonable to me. Looks like this should be considered for -stable
- Nishanth, what do you think?

Every product kernel since 3.4 needed to be hacked (we have hacked in different ways so far) to work around this (since we never spend time digging deeper :( ), So, I do agree with your view that a -stable tag will be most beneficial.

Quoted 2 lines

Tony or Kevin, do you want to take this one, or want me to?

I can take it unless you have other fixes pending right now.

Tony

Reviewed-by
#14 lore raw
In reply to Tony Lindgren
Quoted 52 lines · Tony Lindgren
Quoted 51 lines · Nishanth Menon
Quoted 36 lines · Paul Walmsley
Quoted 34 lines · Nishanth Menon

OMAP device hooks around suspend|resume_noirq ensures that hwmod devices are forced to idle using omap_device_idle/enable as part of the last stage of suspend activity.

For a device such as i2c who uses autosuspend, it is possible to enter the suspend path with dev->power.runtime_status = RPM_ACTIVE.

As part of the suspend flow, the generic runtime logic would increment it's dev->power.disable_depth to 1. This should prevent further pm_runtime_get_sync from succeeding once the runtime_status has been set to RPM_SUSPENDED.

Now, as part of the suspend_noirq handler in omap_device, we force the
following: if the device status is !suspended, we force the device to idle using omap_device_idle (clocks are cut etc..). This ensures that from a hardware perspective, the device is "suspended". However, runtime_status is left to be active.

*if* an operation is attempted after this point to pm_runtime_get_sync, runtime framework depends on runtime_status to indicate accurately the device status, and since it sees it to be ACTIVE, it assumes the module is functional and returns a non-error value. As a result the user will see pm_runtime_get succeed, however a register access will crash due to the lack of clocks.

To prevent this from happening, we should ensure that runtime_status exactly indicates the device status. As a result of this change any further calls to pm_runtime_get* would return -EACCES (since disable_depth is 1). On resume, we restore the clocks and runtime status exactly as we suspended with. These operations are not expected to fail as we update the states after the core runtime framework has suspended itself and restore before the core runtime framework has resumed.

Quoted 7 lines
Quoted 4 lines

Looks reasonable to me. Looks like this should be considered for -stable
- Nishanth, what do you think?

Every product kernel since 3.4 needed to be hacked (we have hacked in different ways so far) to work around this (since we never spend time digging deeper :( ), So, I do agree with your view that a -stable tag will be most beneficial.

Quoted 1 line
Quoted 3 lines
Quoted 1 line
Quoted 1 line

Tony or Kevin, do you want to take this one, or want me to?

I can take it unless you have other fixes pending right now.

This version looks good to me.

#15 lore raw
Quoted 40 lines · Kevin Hilman
Quoted 38 lines · Tony Lindgren
Quoted 37 lines · Nishanth Menon
Quoted 36 lines · Paul Walmsley
Quoted 34 lines · Nishanth Menon

OMAP device hooks around suspend|resume_noirq ensures that hwmod devices are forced to idle using omap_device_idle/enable as part of the last stage of suspend activity.

For a device such as i2c who uses autosuspend, it is possible to enter the suspend path with dev->power.runtime_status = RPM_ACTIVE.

As part of the suspend flow, the generic runtime logic would increment it's dev->power.disable_depth to 1. This should prevent further pm_runtime_get_sync from succeeding once the runtime_status has been set to RPM_SUSPENDED.

Now, as part of the suspend_noirq handler in omap_device, we force the
following: if the device status is !suspended, we force the device to idle using omap_device_idle (clocks are cut etc..). This ensures that from a hardware perspective, the device is "suspended". However, runtime_status is left to be active.

*if* an operation is attempted after this point to pm_runtime_get_sync, runtime framework depends on runtime_status to indicate accurately the device status, and since it sees it to be ACTIVE, it assumes the module is functional and returns a non-error value. As a result the user will see pm_runtime_get succeed, however a register access will crash due to the lack of clocks.

To prevent this from happening, we should ensure that runtime_status exactly indicates the device status. As a result of this change any further calls to pm_runtime_get* would return -EACCES (since disable_depth is 1). On resume, we restore the clocks and runtime status exactly as we suspended with. These operations are not expected to fail as we update the states after the core runtime framework has suspended itself and restore before the core runtime framework has resumed.

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

Hi Kevin,

I am wondering if OMAP would benefit from letting the PM core allow runtime suspends during system suspend, which is not the case as of now.

My impression is that it could simplify OMAP drivers and the power domain, but it would be interesting to hear your opinion in this. It somewhat touches this patch. I guess.

The reason for bringing this up here, is that I wanted to highlight that we are at the moment discussing the following RFC on the linux-pm mailing list:

"[RFC PATCH] PM / Runtime: Allow to inactivate devices during system suspend".

Kind regards
Ulf Hansson

Quoted 23 lines
Quoted 10 lines
Quoted 10 lines
Quoted 3 lines

Looks reasonable to me. Looks like this should be considered for -stable
- Nishanth, what do you think?

Every product kernel since 3.4 needed to be hacked (we have hacked in different ways so far) to work around this (since we never spend time digging deeper :( ), So, I do agree with your view that a -stable tag will be most beneficial.

Quoted 1 line
Quoted 3 lines
Quoted 1 line
Quoted 1 line

Tony or Kevin, do you want to take this one, or want me to?

I can take it unless you have other fixes pending right now.

This version looks good to me.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo at vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/