Skip to contentSkip to conversation

i2c: mediatek: Move suspend and resume handling to NOIRQ phase

PATCH Superseded [email protected] 3 participants 12 messages Nov 18, 2020 – Dec 23, 2020

Subsystems (3)
lore
Share & tools
lore

[email protected]

author maintainer
#0 lore raw

PATCH v2 1/1 i2c: mediatek: Move suspend and resume handling to NOIRQ phase

From: Qii Wang <[email protected]>

Some i2c device driver indirectly uses I2C driver when it is now being suspended. The i2c devices driver is suspended during the NOIRQ phase and this cannot be changed due to other dependencies. Therefore, we also need to move the suspend handling for the I2C controller driver to the NOIRQ phase as well.

---

Changes in v2:
- fix the wrong spelling medaitek to mediatek

 drivers/i2c/busses/i2c-mt65xx.c | 19 ++++++++++++++++---
 1 file changed, 16 insertions(+), 3 deletions(-)
drivers/i2c/busses/i2c-mt65xx.c +16 -3
index 33de99b..6f61595 100644
@@ -1258,7 +1258,8 @@ static int mtk_i2c_probe(struct platform_device *pdev)
 	mtk_i2c_clock_disable(i2c);
 
 	ret = devm_request_irq(&pdev->dev, irq, mtk_i2c_irq,
-			       IRQF_TRIGGER_NONE, I2C_DRV_NAME, i2c);
+			       IRQF_NO_SUSPEND | IRQF_TRIGGER_NONE,
+			       I2C_DRV_NAME, i2c);
 	if (ret < 0) {
 		dev_err(&pdev->dev,
 			"Request I2C IRQ %d fail\n", irq);
@@ -1285,7 +1286,16 @@ static int mtk_i2c_remove(struct platform_device *pdev)
 }
 
 #ifdef CONFIG_PM_SLEEP
-static int mtk_i2c_resume(struct device *dev)
+static int mtk_i2c_suspend_noirq(struct device *dev)
+{
+	struct mtk_i2c *i2c = dev_get_drvdata(dev);
+
+	i2c_mark_adapter_suspended(&i2c->adap);
+
+	return 0;
+}
+
+static int mtk_i2c_resume_noirq(struct device *dev)
 {
 	int ret;
 	struct mtk_i2c *i2c = dev_get_drvdata(dev);
@@ -1300,12 +1310,15 @@ static int mtk_i2c_resume(struct device *dev)
 
 	mtk_i2c_clock_disable(i2c);
 
+	i2c_mark_adapter_resumed(&i2c->adap);
+
 	return 0;
 }
 #endif
 
 static const struct dev_pm_ops mtk_i2c_pm = {
-	SET_SYSTEM_SLEEP_PM_OPS(NULL, mtk_i2c_resume)
+	SET_NOIRQ_SYSTEM_SLEEP_PM_OPS(mtk_i2c_suspend_noirq,
+				      mtk_i2c_resume_noirq)
 };
 
 static struct platform_driver mtk_i2c_driver = {
Signature · 1 line
1.9.1
#1 lore raw

Hi,

Quoted 7 lines

Some i2c device driver indirectly uses I2C driver when it is now being suspended. The i2c devices driver is suspended during the NOIRQ phase and this cannot be changed due to other dependencies. Therefore, we also need to move the suspend handling for the I2C controller driver to the NOIRQ phase as well.

Is this a bugfix and should go into 5.10? Or can it wait for 5.11?

Thanks,

Wolfram

_______________________________________________
Linux-mediatek mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-mediatek

author maintainer
#2 lore raw
Quoted 8 lines · Wolfram Sang

Hi,

Quoted 6 lines

Some i2c device driver indirectly uses I2C driver when it is now being suspended. The i2c devices driver is suspended during the NOIRQ phase and this cannot be changed due to other dependencies. Therefore, we also need to move the suspend handling for the I2C controller driver to the NOIRQ phase as well.

Quoted 4 lines
Quoted 1 line

Is this a bugfix and should go into 5.10? Or can it wait for 5.11?

Yes, Can you help to apply it into 5.10? Thanks

Quoted 4 lines

Thanks,

Wolfram

#3 lore raw
Quoted 9 lines · Qii Wang
Quoted 8 lines · Wolfram Sang

Hi,

Quoted 6 lines

Some i2c device driver indirectly uses I2C driver when it is now being suspended. The i2c devices driver is suspended during the NOIRQ phase and this cannot be changed due to other dependencies. Therefore, we also need to move the suspend handling for the I2C controller driver to the NOIRQ phase as well.

Quoted 6 lines
Quoted 4 lines
Quoted 1 line

Is this a bugfix and should go into 5.10? Or can it wait for 5.11?

Yes, Can you help to apply it into 5.10? Thanks

To be honest if you still do have any i2c device which accessing i2c buss after _noirq stage and your driver does not implement .master_xfer_atomic() - you definitely have a bigger problem. So adding IRQF_NO_SUSPEND sound like a hack and probably works just by luck.

Signature · 7 lines
Best regards,
grygorii

_______________________________________________
Linux-mediatek mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-mediatek
author maintainer
#4 lore raw

Hi:
Thank you very much for your patience review.
There are two main purposes of this patch:
1.i2c_mark_adapter_suspended&i2c_mark_adapter_resumed
Avoid accessing the adapter while it is suspended by marking it suspended during suspend. This allows the I2C core to catch this, and print a warning.
https://patchwork.kernel.org/project/linux-arm-kernel/patch/[email protected]/

2. IRQF_NO_SUSPEND.
Having interrupts disabled means not only that an interrupt will not occur at an awkward time, but also that using any functionality that requires interrupts will not work. So if the driver uses an I2C bus or similar to tell the device to turn off, and if the I2C bus uses interrupts to indicate completion (which is normal), then either the device must be powered-off in suspend_late, so the I2C interrupt must be marked IRQF_NO_SUSPEND.
https://patchwork.kernel.org/project/linux-acpi/patch/[email protected]/

Quoted 13 lines · Grygorii Strashko
Quoted 11 lines · Qii Wang
Quoted 10 lines · Wolfram Sang

Hi,

Quoted 7 lines

Some i2c device driver indirectly uses I2C driver when it is now being suspended. The i2c devices driver is suspended during the NOIRQ phase and this cannot be changed due to other dependencies. Therefore, we also need to move the suspend handling for the I2C controller driver to the NOIRQ phase as well.

Quoted 9 lines
Quoted 4 lines
Quoted 2 lines

Is this a bugfix and should go into 5.10? Or can it wait for 5.11?

Yes, Can you help to apply it into 5.10? Thanks

To be honest if you still do have any i2c device which accessing i2c buss after _noirq stage and your driver does not implement .master_xfer_atomic() - you definitely have a bigger problem. So adding IRQF_NO_SUSPEND sound like a hack and probably works just by luck.

At present, it is only a problem caused by missing interrupts, and .master_xfer_atomic() just a implement in polling mode. Why not set the interrupt to a state that can always be triggered?

#5 lore raw
Quoted 9 lines · Qii Wang

Hi:
Thank you very much for your patience review.
There are two main purposes of this patch:
1.i2c_mark_adapter_suspended&i2c_mark_adapter_resumed
Avoid accessing the adapter while it is suspended by marking it suspended during suspend. This allows the I2C core to catch this, and print a warning.
https://patchwork.kernel.org/project/linux-arm-kernel/patch/[email protected]/

Quoted 10 lines

2. IRQF_NO_SUSPEND.
Having interrupts disabled means not only that an interrupt will not occur at an awkward time, but also that using any functionality that requires interrupts will not work. So if the driver uses an I2C bus or similar to tell the device to turn off, and if the I2C bus uses interrupts to indicate completion (which is normal), then either the device must be powered-off in suspend_late, so the I2C interrupt must be marked IRQF_NO_SUSPEND.
https://patchwork.kernel.org/project/linux-acpi/patch/[email protected]/

Pls, do not top post.

Quoted 20 lines · Grygorii Strashko
Quoted 18 lines · Qii Wang
Quoted 15 lines · Wolfram Sang
Quoted 8 lines

Hi,

Quoted 6 lines

Some i2c device driver indirectly uses I2C driver when it is now being suspended. The i2c devices driver is suspended during the NOIRQ phase and this cannot be changed due to other dependencies. Therefore, we also need to move the suspend handling for the I2C controller driver to the NOIRQ phase as well.

Quoted 4 lines
Quoted 1 line

Is this a bugfix and should go into 5.10? Or can it wait for 5.11?

Yes, Can you help to apply it into 5.10? Thanks

Quoted 10 lines
Quoted 4 lines

To be honest if you still do have any i2c device which accessing i2c buss after _noirq stage and your driver does not implement .master_xfer_atomic() - you definitely have a bigger problem. So adding IRQF_NO_SUSPEND sound like a hack and probably works just by luck.

At present, it is only a problem caused by missing interrupts, and .master_xfer_atomic() just a implement in polling mode. Why not set the interrupt to a state that can always be triggered?

Because you must not use any IRQ driven operations after _noirq suspend state as it might (and most probably will) cause unpredictable behavior later in suspend_enter():

	arch_suspend_disable_irqs();
	BUG_ON(!irqs_disabled());

^after this point any IRQ driven I2C transfer will cause IRQ to be re-enabled

if you need turn off device from platform callbacks - .master_xfer_atomic() has to be implemented and used.

Signature · 7 lines
Best regards,
grygorii

_______________________________________________
Linux-mediatek mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-mediatek
author maintainer
#6 lore raw
Quoted 32 lines · Grygorii Strashko
Quoted 21 lines · Grygorii Strashko
Quoted 18 lines · Qii Wang
Quoted 15 lines · Wolfram Sang
Quoted 8 lines

Hi,

Quoted 6 lines

Some i2c device driver indirectly uses I2C driver when it is now being suspended. The i2c devices driver is suspended during the NOIRQ phase and this cannot be changed due to other dependencies. Therefore, we also need to move the suspend handling for the I2C controller driver to the NOIRQ phase as well.

Quoted 4 lines
Quoted 1 line

Is this a bugfix and should go into 5.10? Or can it wait for 5.11?

Yes, Can you help to apply it into 5.10? Thanks

Quoted 10 lines
Quoted 4 lines

To be honest if you still do have any i2c device which accessing i2c buss after _noirq stage and your driver does not implement .master_xfer_atomic() - you definitely have a bigger problem. So adding IRQF_NO_SUSPEND sound like a hack and probably works just by luck.

At present, it is only a problem caused by missing interrupts, and .master_xfer_atomic() just a implement in polling mode. Why not set the interrupt to a state that can always be triggered?

Quoted 9 lines

Because you must not use any IRQ driven operations after _noirq suspend state as it might (and most probably will) cause unpredictable behavior later in suspend_enter():

	arch_suspend_disable_irqs();
	BUG_ON(!irqs_disabled());

^after this point any IRQ driven I2C transfer will cause IRQ to be re-enabled

if you need turn off device from platform callbacks - .master_xfer_atomic() has to be implemented and used.

Maybe my comment is a bit disturbing.Our purpose is not to call i2c and use interrupts after _noirq pauses.So We use i2c_mark_adapter_suspended&i2c_mark_adapter_resumed to block these i2c transfers, There will not have any IRQ driven I2C transfer after this point:

        arch_suspend_disable_irqs();
        BUG_ON(!irqs_disabled());
But some device driver will do i2c transfer after
dpm_noirq_resume_devices in dpm_resume_noirq(PMSG_RESUME) when our
driver irq hasn't resume.
	void dpm_resume_noirq(pm_message_t state)
	{
        	dpm_noirq_resume_devices(state);
        	resume_device_irqs();
        	device_wakeup_disarm_wake_irqs();
        	cpuidle_resume();
	}

.master_xfer_atomic() seems to be invalid for this question at this time?

#7 lore raw
Quoted 47 lines · Qii Wang
Quoted 32 lines · Grygorii Strashko
Quoted 21 lines · Grygorii Strashko
Quoted 18 lines · Qii Wang
Quoted 15 lines · Wolfram Sang
Quoted 8 lines

Hi,

Quoted 6 lines

Some i2c device driver indirectly uses I2C driver when it is now being suspended. The i2c devices driver is suspended during the NOIRQ phase and this cannot be changed due to other dependencies. Therefore, we also need to move the suspend handling for the I2C controller driver to the NOIRQ phase as well.

Quoted 4 lines
Quoted 1 line

Is this a bugfix and should go into 5.10? Or can it wait for 5.11?

Yes, Can you help to apply it into 5.10? Thanks

Quoted 10 lines
Quoted 4 lines

To be honest if you still do have any i2c device which accessing i2c buss after _noirq stage and your driver does not implement .master_xfer_atomic() - you definitely have a bigger problem. So adding IRQF_NO_SUSPEND sound like a hack and probably works just by luck.

At present, it is only a problem caused by missing interrupts, and .master_xfer_atomic() just a implement in polling mode. Why not set the interrupt to a state that can always be triggered?

Quoted 9 lines

Because you must not use any IRQ driven operations after _noirq suspend state as it might (and most probably will) cause unpredictable behavior later in suspend_enter():

	arch_suspend_disable_irqs();
	BUG_ON(!irqs_disabled());

^after this point any IRQ driven I2C transfer will cause IRQ to be re-enabled

if you need turn off device from platform callbacks - .master_xfer_atomic() has to be implemented and used.

Maybe my comment is a bit disturbing.Our purpose is not to call i2c and use interrupts after _noirq pauses.So We use i2c_mark_adapter_suspended&i2c_mark_adapter_resumed to block these i2c transfers, There will not have any IRQ driven I2C transfer after this point:

Quoted 8 lines
         arch_suspend_disable_irqs();
         BUG_ON(!irqs_disabled());
But some device driver will do i2c transfer after
dpm_noirq_resume_devices in dpm_resume_noirq(PMSG_RESUME) when our
driver irq hasn't resume.
	void dpm_resume_noirq(pm_message_t state)
	{
         	dpm_noirq_resume_devices(state);

Just to clarify. You have resume sequence in dpm_noirq_resume_devices
dpm_noirq_resume_devices -> resume I2C -> resume some device -> do i2c transfer after?

Is "some device" in Kernel mainline?

Quoted 7 lines
         	resume_device_irqs();
         	device_wakeup_disarm_wake_irqs();
         	cpuidle_resume();
	}

.master_xfer_atomic() seems to be invalid for this question at this time?

Signature · 7 lines
Best regards,
grygorii

_______________________________________________
Linux-mediatek mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-mediatek
author maintainer
#8 lore raw
Quoted 58 lines · Grygorii Strashko
Quoted 47 lines · Qii Wang
Quoted 32 lines · Grygorii Strashko
Quoted 21 lines · Grygorii Strashko
Quoted 18 lines · Qii Wang
Quoted 15 lines · Wolfram Sang
Quoted 8 lines

Hi,

Quoted 6 lines

Some i2c device driver indirectly uses I2C driver when it is now being suspended. The i2c devices driver is suspended during the NOIRQ phase and this cannot be changed due to other dependencies. Therefore, we also need to move the suspend handling for the I2C controller driver to the NOIRQ phase as well.

Quoted 4 lines
Quoted 1 line

Is this a bugfix and should go into 5.10? Or can it wait for 5.11?

Yes, Can you help to apply it into 5.10? Thanks

Quoted 10 lines
Quoted 4 lines

To be honest if you still do have any i2c device which accessing i2c buss after _noirq stage and your driver does not implement .master_xfer_atomic() - you definitely have a bigger problem. So adding IRQF_NO_SUSPEND sound like a hack and probably works just by luck.

At present, it is only a problem caused by missing interrupts, and .master_xfer_atomic() just a implement in polling mode. Why not set the interrupt to a state that can always be triggered?

Quoted 9 lines

Because you must not use any IRQ driven operations after _noirq suspend state as it might (and most probably will) cause unpredictable behavior later in suspend_enter():

	arch_suspend_disable_irqs();
	BUG_ON(!irqs_disabled());

^after this point any IRQ driven I2C transfer will cause IRQ to be re-enabled

if you need turn off device from platform callbacks - .master_xfer_atomic() has to be implemented and used.

Maybe my comment is a bit disturbing.Our purpose is not to call i2c and use interrupts after _noirq pauses.So We use i2c_mark_adapter_suspended&i2c_mark_adapter_resumed to block these i2c transfers, There will not have any IRQ driven I2C transfer after this point:

Quoted 8 lines
         arch_suspend_disable_irqs();
         BUG_ON(!irqs_disabled());
But some device driver will do i2c transfer after
dpm_noirq_resume_devices in dpm_resume_noirq(PMSG_RESUME) when our
driver irq hasn't resume.
	void dpm_resume_noirq(pm_message_t state)
	{
         	dpm_noirq_resume_devices(state);
Quoted 3 lines

Just to clarify. You have resume sequence in dpm_noirq_resume_devices
dpm_noirq_resume_devices -> resume I2C -> resume some device -> do i2c transfer after?

Yes.

Quoted 2 lines

Is "some device" in Kernel mainline?

The problematic device driver is drivers/regulator/da9211-regulator.c in Kernel mainline.

Quoted 8 lines
Quoted 7 lines
         	resume_device_irqs();
         	device_wakeup_disarm_wake_irqs();
         	cpuidle_resume();
	}

.master_xfer_atomic() seems to be invalid for this question at this time?

_______________________________________________
Linux-mediatek mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-mediatek

#9 lore raw
Quoted 59 lines · Qii Wang
Quoted 58 lines · Grygorii Strashko
Quoted 55 lines · Qii Wang
Quoted 32 lines · Grygorii Strashko
Quoted 21 lines · Grygorii Strashko
Quoted 18 lines · Qii Wang
Quoted 15 lines · Wolfram Sang
Quoted 8 lines

Hi,

Quoted 6 lines

Some i2c device driver indirectly uses I2C driver when it is now being suspended. The i2c devices driver is suspended during the NOIRQ phase and this cannot be changed due to other dependencies. Therefore, we also need to move the suspend handling for the I2C controller driver to the NOIRQ phase as well.

Quoted 4 lines
Quoted 1 line

Is this a bugfix and should go into 5.10? Or can it wait for 5.11?

Yes, Can you help to apply it into 5.10? Thanks

Quoted 10 lines
Quoted 4 lines

To be honest if you still do have any i2c device which accessing i2c buss after _noirq stage and your driver does not implement .master_xfer_atomic() - you definitely have a bigger problem. So adding IRQF_NO_SUSPEND sound like a hack and probably works just by luck.

At present, it is only a problem caused by missing interrupts, and .master_xfer_atomic() just a implement in polling mode. Why not set the interrupt to a state that can always be triggered?

Quoted 9 lines

Because you must not use any IRQ driven operations after _noirq suspend state as it might (and most probably will) cause unpredictable behavior later in suspend_enter():

	arch_suspend_disable_irqs();
	BUG_ON(!irqs_disabled());

^after this point any IRQ driven I2C transfer will cause IRQ to be re-enabled

if you need turn off device from platform callbacks - .master_xfer_atomic() has to be implemented and used.

Maybe my comment is a bit disturbing.Our purpose is not to call i2c and use interrupts after _noirq pauses.So We use i2c_mark_adapter_suspended&i2c_mark_adapter_resumed to block these i2c transfers, There will not have any IRQ driven I2C transfer after this point:

          arch_suspend_disable_irqs();
          BUG_ON(!irqs_disabled());
But some device driver will do i2c transfer after
dpm_noirq_resume_devices in dpm_resume_noirq(PMSG_RESUME) when our
driver irq hasn't resume.
	void dpm_resume_noirq(pm_message_t state)
	{
          	dpm_noirq_resume_devices(state);
Quoted 5 lines
Quoted 3 lines

Just to clarify. You have resume sequence in dpm_noirq_resume_devices
dpm_noirq_resume_devices -> resume I2C -> resume some device -> do i2c transfer after?

Yes.

huh. First consider IRQF_EARLY_RESUME - it's better, but still will be a hack

Quoted 6 lines
Quoted 3 lines

Is "some device" in Kernel mainline?

The problematic device driver is drivers/regulator/da9211-regulator.c in Kernel mainline.

regulator is passive device, somebody should call it !?

And da9211-regulator IRQ handler should remain disabled till resume_device_irqs() call.

note. regulator_class implements only

static const struct dev_pm_ops __maybe_unused regulator_pm_ops = {
.suspend = regulator_suspend,
.resume = regulator_resume,
};

Quoted 10 lines
Quoted 9 lines
Quoted 8 lines
          	resume_device_irqs();
          	device_wakeup_disarm_wake_irqs();
          	cpuidle_resume();
	}

.master_xfer_atomic() seems to be invalid for this question at this time?

Signature · 7 lines
Best regards,
grygorii

_______________________________________________
Linux-mediatek mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-mediatek
author maintainer
#10 lore raw
Quoted 61 lines · Grygorii Strashko
Quoted 59 lines · Qii Wang
Quoted 58 lines · Grygorii Strashko
Quoted 55 lines · Qii Wang
Quoted 32 lines · Grygorii Strashko
Quoted 21 lines · Grygorii Strashko
Quoted 18 lines · Qii Wang
Quoted 15 lines · Wolfram Sang
Quoted 8 lines

Hi,

Quoted 6 lines

Some i2c device driver indirectly uses I2C driver when it is now being suspended. The i2c devices driver is suspended during the NOIRQ phase and this cannot be changed due to other dependencies. Therefore, we also need to move the suspend handling for the I2C controller driver to the NOIRQ phase as well.

Quoted 4 lines
Quoted 1 line

Is this a bugfix and should go into 5.10? Or can it wait for 5.11?

Yes, Can you help to apply it into 5.10? Thanks

Quoted 10 lines
Quoted 4 lines

To be honest if you still do have any i2c device which accessing i2c buss after _noirq stage and your driver does not implement .master_xfer_atomic() - you definitely have a bigger problem. So adding IRQF_NO_SUSPEND sound like a hack and probably works just by luck.

At present, it is only a problem caused by missing interrupts, and .master_xfer_atomic() just a implement in polling mode. Why not set the interrupt to a state that can always be triggered?

Quoted 9 lines

Because you must not use any IRQ driven operations after _noirq suspend state as it might (and most probably will) cause unpredictable behavior later in suspend_enter():

	arch_suspend_disable_irqs();
	BUG_ON(!irqs_disabled());

^after this point any IRQ driven I2C transfer will cause IRQ to be re-enabled

if you need turn off device from platform callbacks - .master_xfer_atomic() has to be implemented and used.

Maybe my comment is a bit disturbing.Our purpose is not to call i2c and use interrupts after _noirq pauses.So We use i2c_mark_adapter_suspended&i2c_mark_adapter_resumed to block these i2c transfers, There will not have any IRQ driven I2C transfer after this point:

          arch_suspend_disable_irqs();
          BUG_ON(!irqs_disabled());
But some device driver will do i2c transfer after
dpm_noirq_resume_devices in dpm_resume_noirq(PMSG_RESUME) when our
driver irq hasn't resume.
	void dpm_resume_noirq(pm_message_t state)
	{
          	dpm_noirq_resume_devices(state);
Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

Just to clarify. You have resume sequence in dpm_noirq_resume_devices
dpm_noirq_resume_devices -> resume I2C -> resume some device -> do i2c transfer after?

Yes.

huh. First consider IRQF_EARLY_RESUME - it's better, but still will be a hack

There should be the same problem during the suspend process, So IRQF_EARLY_RESUME should not be able to solve the problem.

Quoted 11 lines
Quoted 6 lines
Quoted 3 lines

Is "some device" in Kernel mainline?

The problematic device driver is drivers/regulator/da9211-regulator.c in Kernel mainline.

regulator is passive device, somebody should call it !?

And da9211-regulator IRQ handler should remain disabled till resume_device_irqs() call.

Not only will i2c transfer be called in da9211-regulator IRQ handler, but also other drivers will call da9211_buck_ops which containing i2c transfers.

Quoted 19 lines

note. regulator_class implements only

static const struct dev_pm_ops __maybe_unused regulator_pm_ops = {
.suspend = regulator_suspend,
.resume = regulator_resume,
};

Quoted 10 lines
Quoted 9 lines
Quoted 8 lines
          	resume_device_irqs();
          	device_wakeup_disarm_wake_irqs();
          	cpuidle_resume();
	}

.master_xfer_atomic() seems to be invalid for this question at this time?