Skip to contentSkip to conversation

i2c: core: Do not enable wakeup by default

RFC In review Mika Westerberg 8 participants 13 messages Feb 7, 2023 – Feb 12, 2023

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

RFC 1/1 i2c: core: Do not enable wakeup by default

After commit b38f2d5d9615 ("i2c: acpi: Use ACPI wake capability bit to set wake_irq") the I2C core has been setting I2C_CLIENT_WAKE for ACPI devices if they announce to be wake capable in their device description. However, on certain systems where audio codec has been connected through I2C this causes system suspend to wake up immediately because power to the codec is turned off which pulls the interrupt line "low" triggering wake up.

Possible reason why the interrupt is marked as wake capable is that some codecs apparently support "Wake on Voice" or similar functionality.

In any case, I don't think we should be enabling wakeup by default on all I2C devices that are wake capable. According to device_init_wakeup() documentation most devices should leave it disabled, with exceptions on devices such as keyboards, power buttons etc. Userspace can enable wakeup as needed by writing to device "power/wakeup" attribute.

---
Hi,

Sending this as RFC because I'm not too familiar with the usage of I2C_CLIENT_WAKE and whether this is something that is expected behaviour in users of I2C devices. On ACPI side I think this is the correct thing to do at least.

 drivers/i2c/i2c-core-base.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/i2c/i2c-core-base.c +1 -1
index 087e480b624c..7046549bdae7 100644
@@ -527,7 +527,7 @@ static int i2c_device_probe(struct device *dev)
 			goto put_sync_adapter;
 		}
 
-		device_init_wakeup(&client->dev, true);
+		device_init_wakeup(&client->dev, false);
 
 		if (wakeirq > 0 && wakeirq != client->irq)
 			status = dev_pm_set_dedicated_wake_irq(dev, wakeirq);
Signature · 1 line
2.39.1
Reviewed-by on v1 1/1
#1 lore raw
Quoted 8 lines · Mika Westerberg

After commit b38f2d5d9615 ("i2c: acpi: Use ACPI wake capability bit to set wake_irq") the I2C core has been setting I2C_CLIENT_WAKE for ACPI devices if they announce to be wake capable in their device description. However, on certain systems where audio codec has been connected through I2C this causes system suspend to wake up immediately because power to the codec is turned off which pulls the interrupt line "low" triggering wake up.

Quoted 8 lines

Possible reason why the interrupt is marked as wake capable is that some codecs apparently support "Wake on Voice" or similar functionality.

In any case, I don't think we should be enabling wakeup by default on all I2C devices that are wake capable. According to device_init_wakeup() documentation most devices should leave it disabled, with exceptions on devices such as keyboards, power buttons etc. Userspace can enable wakeup as needed by writing to device "power/wakeup" attribute.

I agree on the reasoning.

Should we have a Fixes tag?

Otherwise

In any case it would be nice if the initial contributors may have a chance to test this and see how their setup is supposed to work.

Quoted 29 lines

---
Hi,

Sending this as RFC because I'm not too familiar with the usage of I2C_CLIENT_WAKE and whether this is something that is expected behaviour in users of I2C devices. On ACPI side I think this is the correct thing to do at least.

 drivers/i2c/i2c-core-base.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/i2c/i2c-core-base.c +1 -1
index 087e480b624c..7046549bdae7 100644
@@ -527,7 +527,7 @@ static int i2c_device_probe(struct device *dev)
 			goto put_sync_adapter;
 		}
 
-		device_init_wakeup(&client->dev, true);
+		device_init_wakeup(&client->dev, false);
 
 		if (wakeirq > 0 && wakeirq != client->irq)
 			status = dev_pm_set_dedicated_wake_irq(dev, wakeirq);
Signature · 1 line
2.39.1
Signature · 2 lines
With Best Regards,
Andy Shevchenko
#2 lore raw
Quoted 20 lines · Andy Shevchenko
Quoted 8 lines · Mika Westerberg

After commit b38f2d5d9615 ("i2c: acpi: Use ACPI wake capability bit to set wake_irq") the I2C core has been setting I2C_CLIENT_WAKE for ACPI devices if they announce to be wake capable in their device description. However, on certain systems where audio codec has been connected through I2C this causes system suspend to wake up immediately because power to the codec is turned off which pulls the interrupt line "low" triggering wake up.

Quoted 8 lines

Possible reason why the interrupt is marked as wake capable is that some codecs apparently support "Wake on Voice" or similar functionality.

In any case, I don't think we should be enabling wakeup by default on all I2C devices that are wake capable. According to device_init_wakeup() documentation most devices should leave it disabled, with exceptions on devices such as keyboards, power buttons etc. Userspace can enable wakeup as needed by writing to device "power/wakeup" attribute.

I agree on the reasoning.

Quoted 7 lines

Should we have a Fixes tag?

Otherwise

In any case it would be nice if the initial contributors may have a chance to test this and see how their setup is supposed to work.

Hello,

Thanks for the patch Mika and your input Andy. While neither I nor Amadeusz contributed to the initial change, we will be testing this one too - reverting the revert of the offending change that we had internally and applying this patch instead. Tests should be concluded by Thursday/Friday this week.

Regards,
Czarek

#3 lore raw
In reply to Mika Westerberg

Sorry, resending in plain text mode.

Quoted 11 lines · Mika Westerberg

After commit b38f2d5d9615 ("i2c: acpi: Use ACPI wake capability bit to set wake_irq") the I2C core has been setting I2C_CLIENT_WAKE for ACPI devices if they announce to be wake capable in their device description. However, on certain systems where audio codec has been connected through I2C this causes system suspend to wake up immediately because power to the codec is turned off which pulls the interrupt line "low" triggering wake up.

Possible reason why the interrupt is marked as wake capable is that some codecs apparently support "Wake on Voice" or similar functionality.

That's generally a bug in the ACPI tables. The wake bit shouldn't be set if the power domain for the device is powered off on suspend. The best thing is to fix the ACPI tables, but if you can't, then you can set the ignore_wake flag for the device:
https://github.com/torvalds/linux/blob/master/drivers/gpio/gpiolib-acpi.c#L31.
If that works we can add a quirk for the device:
https://github.com/torvalds/linux/blob/master/drivers/gpio/gpiolib-acpi.c#L1633.

Quoted 6 lines

In any case, I don't think we should be enabling wakeup by default on all I2C devices that are wake capable. According to device_init_wakeup() documentation most devices should leave it disabled, with exceptions on devices such as keyboards, power buttons etc. Userspace can enable wakeup as needed by writing to device "power/wakeup" attribute.

Enabling wake by default was an unintended side-effect. I didn't catch this when I wrote the patch :/ It's been exposing all the incorrect ACPI configurations for better or worse. Mario pushed a patch up earlier to disable thes Wake GPIOs when using S3:
https://github.com/torvalds/linux/commit/d63f11c02b8d3e54bdb65d8c309f73b7f474aec4.
Are you having problems with S3 or S0iX?

Quoted 19 lines

---
Hi,

Sending this as RFC because I'm not too familiar with the usage of I2C_CLIENT_WAKE and whether this is something that is expected behaviour in users of I2C devices. On ACPI side I think this is the correct thing to do at least.

 drivers/i2c/i2c-core-base.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/i2c/i2c-core-base.c +0 -0
index 087e480b624c..7046549bdae7 100644

drivers/i2c/i2c-core-base.c static int i2c_device_probe(struct device *dev) · line 530

Quoted 5 lines
drivers/i2c/i2c-core-base.c +1 -1
@@ -527,7 +527,7 @@ static int i2c_device_probe(struct device *dev)
                        goto put_sync_adapter;
                }
 
-               device_init_wakeup(&client->dev, true);
+               device_init_wakeup(&client->dev, false);

This would be a change in behavior for Device Tree. Maybe you can declare a bool enable_wake = true, then in the ACPI branch
(https://github.com/torvalds/linux/blob/master/drivers/i2c/i2c-core-base.c#L495)
set enable_wake = false. This would keep wakes enabled by default on device tree and disabled for ACPI. This matches the original behavior before my patch.

Quoted 6 lines
drivers/i2c/i2c-core-base.c +0 -1
                if (wakeirq > 0 && wakeirq != client->irq)
                        status = dev_pm_set_dedicated_wake_irq(dev, wakeirq);
--

2.39.1

#4 lore raw

Hi,

Quoted 16 lines · Raul Rangel

Sorry, resending in plain text mode.

Quoted 11 lines · Mika Westerberg

After commit b38f2d5d9615 ("i2c: acpi: Use ACPI wake capability bit to set wake_irq") the I2C core has been setting I2C_CLIENT_WAKE for ACPI devices if they announce to be wake capable in their device description. However, on certain systems where audio codec has been connected through I2C this causes system suspend to wake up immediately because power to the codec is turned off which pulls the interrupt line "low" triggering wake up.

Possible reason why the interrupt is marked as wake capable is that some codecs apparently support "Wake on Voice" or similar functionality.

Quoted 7 lines

That's generally a bug in the ACPI tables. The wake bit shouldn't be set if the power domain for the device is powered off on suspend. The best thing is to fix the ACPI tables, but if you can't, then you can set the ignore_wake flag for the device:
https://github.com/torvalds/linux/blob/master/drivers/gpio/gpiolib-acpi.c#L31.
If that works we can add a quirk for the device:
https://github.com/torvalds/linux/blob/master/drivers/gpio/gpiolib-acpi.c#L1633.

I think (hope) these systems are not yet available for public so there is a chance that the tables can still be fixed, without need to add any quirks.

@Amadeusz, @Cezary, if that's the case I suggest filing a bug against the BIOS.

Quoted 6 lines
Quoted 5 lines

In any case, I don't think we should be enabling wakeup by default on all I2C devices that are wake capable. According to device_init_wakeup() documentation most devices should leave it disabled, with exceptions on devices such as keyboards, power buttons etc. Userspace can enable wakeup as needed by writing to device "power/wakeup" attribute.

Quoted 6 lines

Enabling wake by default was an unintended side-effect. I didn't catch this when I wrote the patch :/ It's been exposing all the incorrect ACPI configurations for better or worse. Mario pushed a patch up earlier to disable thes Wake GPIOs when using S3:
https://github.com/torvalds/linux/commit/d63f11c02b8d3e54bdb65d8c309f73b7f474aec4.
Are you having problems with S3 or S0iX?

I think this case is S0ix.

Quoted 21 lines
Quoted 21 lines

---
Hi,

Sending this as RFC because I'm not too familiar with the usage of I2C_CLIENT_WAKE and whether this is something that is expected behaviour in users of I2C devices. On ACPI side I think this is the correct thing to do at least.

 drivers/i2c/i2c-core-base.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/i2c/i2c-core-base.c +0 -0
index 087e480b624c..7046549bdae7 100644
@@ -527,7 +527,7 @@ static int i2c_device_probe(struct device *dev)
                        goto put_sync_adapter;
                }
 
Quoted 9 lines

drivers/i2c/i2c-core-base.c static int i2c_device_probe(struct device *dev) · line 530

Quoted 2 lines
drivers/i2c/i2c-core-base.c +1 -1
-               device_init_wakeup(&client->dev, true);
+               device_init_wakeup(&client->dev, false);

This would be a change in behavior for Device Tree. Maybe you can declare a bool enable_wake = true, then in the ACPI branch
(https://github.com/torvalds/linux/blob/master/drivers/i2c/i2c-core-base.c#L495)
set enable_wake = false. This would keep wakes enabled by default on device tree and disabled for ACPI. This matches the original behavior before my patch.

I don't think it's a good idea to make the behaviour different. Drivers in general do not need to know whether the device was enumerated on ACPI or DT or whatnot. Same goes for users who should expect similar behaviour on the same device.

I wonder what is the reason why I2C bus does this for all wake capable devices in the first place? Typically it should be up to the user to enable them not the opposite.

#5 lore raw
Quoted 26 lines · Mika Westerberg

Hi,

Quoted 23 lines · Raul Rangel

Sorry, resending in plain text mode.

Quoted 11 lines · Mika Westerberg

After commit b38f2d5d9615 ("i2c: acpi: Use ACPI wake capability bit to set wake_irq") the I2C core has been setting I2C_CLIENT_WAKE for ACPI devices if they announce to be wake capable in their device description. However, on certain systems where audio codec has been connected through I2C this causes system suspend to wake up immediately because power to the codec is turned off which pulls the interrupt line "low" triggering wake up.

Possible reason why the interrupt is marked as wake capable is that some codecs apparently support "Wake on Voice" or similar functionality.

That's generally a bug in the ACPI tables. The wake bit shouldn't be set if the power domain for the device is powered off on suspend. The best thing is to fix the ACPI tables, but if you can't, then you can set the ignore_wake flag for the device:
https://github.com/torvalds/linux/blob/master/drivers/gpio/gpiolib-acpi.c#L31.
If that works we can add a quirk for the device:
https://github.com/torvalds/linux/blob/master/drivers/gpio/gpiolib-acpi.c#L1633.

I've seen this one already and also tried to use it, but it didn't work. Also when I was reading code I wasn't really convinced that it is linked to i2c in any straightforward way. I mean i2c decides in different places that it has wake support (I even added some prints to make sure ;). The code you pointed out decides in https://github.com/torvalds/linux/blob/master/drivers/gpio/gpiolib-acpi.c#L387 but i2c code seems to decide in https://github.com/torvalds/linux/blob/master/drivers/i2c/i2c-core-acpi.c#L176 where it just checks if irq flags has wake_capable flag set. When I looked at it previously I was pretty sure it comes straight from BIOS and passes the quirk code you mentioned, still I may have missed something.

Quoted 8 lines

I think (hope) these systems are not yet available for public so there is a chance that the tables can still be fixed, without need to add any quirks.

@Amadeusz, @Cezary, if that's the case I suggest filing a bug against the BIOS.

Well, I tried custom DSDT and had problems, but I just remembered that I probably need to pass "revision+1" in file, so kernel sees it as a newer version, let me try again. Is it enough to replace "ExclusiveAndWake" with "Exclusive"?

Quoted 6 lines
Quoted 6 lines
Quoted 5 lines

In any case, I don't think we should be enabling wakeup by default on all I2C devices that are wake capable. According to device_init_wakeup() documentation most devices should leave it disabled, with exceptions on devices such as keyboards, power buttons etc. Userspace can enable wakeup as needed by writing to device "power/wakeup" attribute.

Quoted 8 lines
Quoted 6 lines

Enabling wake by default was an unintended side-effect. I didn't catch this when I wrote the patch :/ It's been exposing all the incorrect ACPI configurations for better or worse. Mario pushed a patch up earlier to disable thes Wake GPIOs when using S3:
https://github.com/torvalds/linux/commit/d63f11c02b8d3e54bdb65d8c309f73b7f474aec4.
Are you having problems with S3 or S0iX?

I think this case is S0ix.

We test both cases in our setups.

Quoted 40 lines
Quoted 22 lines
Quoted 22 lines

---
Hi,

Sending this as RFC because I'm not too familiar with the usage of I2C_CLIENT_WAKE and whether this is something that is expected behaviour in users of I2C devices. On ACPI side I think this is the correct thing to do at least.

  drivers/i2c/i2c-core-base.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
drivers/i2c/i2c-core-base.c +0 -0
index 087e480b624c..7046549bdae7 100644
@@ -527,7 +527,7 @@ static int i2c_device_probe(struct device *dev)
                         goto put_sync_adapter;
                 }
 
Quoted 9 lines

drivers/i2c/i2c-core-base.c static int i2c_device_probe(struct device *dev) · line 530

Quoted 2 lines
drivers/i2c/i2c-core-base.c +1 -1
-               device_init_wakeup(&client->dev, true);
+               device_init_wakeup(&client->dev, false);

This would be a change in behavior for Device Tree. Maybe you can declare a bool enable_wake = true, then in the ACPI branch
(https://github.com/torvalds/linux/blob/master/drivers/i2c/i2c-core-base.c#L495)
set enable_wake = false. This would keep wakes enabled by default on device tree and disabled for ACPI. This matches the original behavior before my patch.

I don't think it's a good idea to make the behaviour different. Drivers in general do not need to know whether the device was enumerated on ACPI or DT or whatnot. Same goes for users who should expect similar behaviour on the same device.

I wonder what is the reason why I2C bus does this for all wake capable devices in the first place? Typically it should be up to the user to enable them not the opposite.

#6 lore raw

Hi,

Quoted 43 lines · Amadeusz Sławiński
Quoted 26 lines · Mika Westerberg

Hi,

Quoted 23 lines · Raul Rangel

Sorry, resending in plain text mode.

Quoted 11 lines · Mika Westerberg

After commit b38f2d5d9615 ("i2c: acpi: Use ACPI wake capability bit to set wake_irq") the I2C core has been setting I2C_CLIENT_WAKE for ACPI devices if they announce to be wake capable in their device description. However, on certain systems where audio codec has been connected through I2C this causes system suspend to wake up immediately because power to the codec is turned off which pulls the interrupt line "low" triggering wake up.

Possible reason why the interrupt is marked as wake capable is that some codecs apparently support "Wake on Voice" or similar functionality.

That's generally a bug in the ACPI tables. The wake bit shouldn't be set if the power domain for the device is powered off on suspend. The best thing is to fix the ACPI tables, but if you can't, then you can set the ignore_wake flag for the device:
https://github.com/torvalds/linux/blob/master/drivers/gpio/gpiolib-acpi.c#L31.
If that works we can add a quirk for the device:
https://github.com/torvalds/linux/blob/master/drivers/gpio/gpiolib-acpi.c#L1633.

I've seen this one already and also tried to use it, but it didn't work.
Also when I was reading code I wasn't really convinced that it is linked to
i2c in any straightforward way. I mean i2c decides in different places that
it has wake support (I even added some prints to make sure ;). The code you
pointed out decides in https://github.com/torvalds/linux/blob/master/drivers/gpio/gpiolib-acpi.c#L387
but i2c code seems to decide in https://github.com/torvalds/linux/blob/master/drivers/i2c/i2c-core-acpi.c#L176
where it just checks if irq flags has wake_capable flag set. When I looked
at it previously I was pretty sure it comes straight from BIOS and passes
the quirk code you mentioned, still I may have missed something.

Quoted 5 lines

I think (hope) these systems are not yet available for public so there is a chance that the tables can still be fixed, without need to add any quirks.

Quoted 8 lines
Quoted 3 lines

@Amadeusz, @Cezary, if that's the case I suggest filing a bug against the BIOS.

Well, I tried custom DSDT and had problems, but I just remembered that I probably need to pass "revision+1" in file, so kernel sees it as a newer version, let me try again. Is it enough to replace "ExclusiveAndWake" with "Exclusive"?

Yes, I think that should be enough.

Quoted 14 lines
Quoted 14 lines
Quoted 13 lines
Quoted 6 lines

In any case, I don't think we should be enabling wakeup by default on all I2C devices that are wake capable. According to device_init_wakeup() documentation most devices should leave it disabled, with exceptions on devices such as keyboards, power buttons etc. Userspace can enable wakeup as needed by writing to device "power/wakeup" attribute.

Enabling wake by default was an unintended side-effect. I didn't catch this when I wrote the patch :/ It's been exposing all the incorrect ACPI configurations for better or worse. Mario pushed a patch up earlier to disable thes Wake GPIOs when using S3:
https://github.com/torvalds/linux/commit/d63f11c02b8d3e54bdb65d8c309f73b7f474aec4.
Are you having problems with S3 or S0iX?

Quoted 3 lines
Quoted 1 line

I think this case is S0ix.

We test both cases in our setups.

Thanks for the clarification!

#7 lore raw
Quoted 50 lines · Mika Westerberg

Hi,

Quoted 47 lines · Amadeusz Sławiński
Quoted 26 lines · Mika Westerberg

Hi,

Quoted 23 lines · Raul Rangel

Sorry, resending in plain text mode.

Quoted 11 lines · Mika Westerberg

After commit b38f2d5d9615 ("i2c: acpi: Use ACPI wake capability bit to set wake_irq") the I2C core has been setting I2C_CLIENT_WAKE for ACPI devices if they announce to be wake capable in their device description. However, on certain systems where audio codec has been connected through I2C this causes system suspend to wake up immediately because power to the codec is turned off which pulls the interrupt line "low" triggering wake up.

Possible reason why the interrupt is marked as wake capable is that some codecs apparently support "Wake on Voice" or similar functionality.

That's generally a bug in the ACPI tables. The wake bit shouldn't be set if the power domain for the device is powered off on suspend. The best thing is to fix the ACPI tables, but if you can't, then you can set the ignore_wake flag for the device:
https://github.com/torvalds/linux/blob/master/drivers/gpio/gpiolib-acpi.c#L31.
If that works we can add a quirk for the device:
https://github.com/torvalds/linux/blob/master/drivers/gpio/gpiolib-acpi.c#L1633.

I've seen this one already and also tried to use it, but it didn't work.
Also when I was reading code I wasn't really convinced that it is linked to
i2c in any straightforward way. I mean i2c decides in different places that
it has wake support (I even added some prints to make sure ;). The code you
pointed out decides in https://github.com/torvalds/linux/blob/master/drivers/gpio/gpiolib-acpi.c#L387
but i2c code seems to decide in https://github.com/torvalds/linux/blob/master/drivers/i2c/i2c-core-acpi.c#L176
where it just checks if irq flags has wake_capable flag set. When I looked
at it previously I was pretty sure it comes straight from BIOS and passes
the quirk code you mentioned, still I may have missed something.

Quoted 8 lines

I think (hope) these systems are not yet available for public so there is a chance that the tables can still be fixed, without need to add any quirks.

@Amadeusz, @Cezary, if that's the case I suggest filing a bug against the BIOS.

Quoted 7 lines
Quoted 4 lines

Well, I tried custom DSDT and had problems, but I just remembered that I probably need to pass "revision+1" in file, so kernel sees it as a newer version, let me try again. Is it enough to replace "ExclusiveAndWake" with "Exclusive"?

Yes, I think that should be enough.

And yes, it seems to work when I bump revision. So will use it as workaround for now and see about fixing BIOS.

#8 lore raw

Hi,

Quoted 9 lines · Amadeusz Sławiński
Quoted 6 lines
Quoted 4 lines

Well, I tried custom DSDT and had problems, but I just remembered that I probably need to pass "revision+1" in file, so kernel sees it as a newer version, let me try again. Is it enough to replace "ExclusiveAndWake" with "Exclusive"?

Yes, I think that should be enough.

And yes, it seems to work when I bump revision. So will use it as workaround for now and see about fixing BIOS.

Okay good to know, then I think we can forget this patch.

#9 lore raw
Quoted 2 lines

Okay good to know, then I think we can forget this patch.

Based on this comment, I marked the patch as "Rejected".

#10 lore raw
In reply to Amadeusz Sławiński
Quoted 30 lines · Amadeusz Sławiński
Quoted 26 lines · Mika Westerberg

Hi,

Quoted 23 lines · Raul Rangel

Sorry, resending in plain text mode.

Quoted 11 lines · Mika Westerberg

After commit b38f2d5d9615 ("i2c: acpi: Use ACPI wake capability bit to set wake_irq") the I2C core has been setting I2C_CLIENT_WAKE for ACPI devices if they announce to be wake capable in their device description. However, on certain systems where audio codec has been connected through I2C this causes system suspend to wake up immediately because power to the codec is turned off which pulls the interrupt line "low" triggering wake up.

Possible reason why the interrupt is marked as wake capable is that some codecs apparently support "Wake on Voice" or similar functionality.

That's generally a bug in the ACPI tables. The wake bit shouldn't be set if the power domain for the device is powered off on suspend. The best thing is to fix the ACPI tables, but if you can't, then you can set the ignore_wake flag for the device:
https://github.com/torvalds/linux/blob/master/drivers/gpio/gpiolib-acpi.c#L31.
If that works we can add a quirk for the device:
https://github.com/torvalds/linux/blob/master/drivers/gpio/gpiolib-acpi.c#L1633.

Quoted 11 lines

I've seen this one already and also tried to use it, but it didn't work. Also when I was reading code I wasn't really convinced that it is linked to i2c in any straightforward way. I mean i2c decides in different places that it has wake support (I even added some prints to make sure ;). The code you pointed out decides in
https://github.com/torvalds/linux/blob/master/drivers/gpio/gpiolib-acpi.c#L387
but i2c code seems to decide in
https://github.com/torvalds/linux/blob/master/drivers/i2c/i2c-core-acpi.c#L176
where it just checks if irq flags has wake_capable flag set. When I looked at it previously I was pretty sure it comes straight from BIOS and passes the quirk code you mentioned, still I may have missed something.

You also need the following patch
https://github.com/torvalds/linux/commit/0e3b175f079247f0d40d2ab695999c309d3a7498,
otherwise the ignore flag only applies to _AEI GPIOs.

Quoted 28 lines
Quoted 9 lines

I think (hope) these systems are not yet available for public so there is a chance that the tables can still be fixed, without need to add any quirks.

@Amadeusz, @Cezary, if that's the case I suggest filing a bug against the BIOS.

Well, I tried custom DSDT and had problems, but I just remembered that I probably need to pass "revision+1" in file, so kernel sees it as a newer version, let me try again. Is it enough to replace "ExclusiveAndWake" with "Exclusive"?

Quoted 13 lines
Quoted 12 lines
Quoted 5 lines

In any case, I don't think we should be enabling wakeup by default on all I2C devices that are wake capable. According to device_init_wakeup() documentation most devices should leave it disabled, with exceptions on devices such as keyboards, power buttons etc. Userspace can enable wakeup as needed by writing to device "power/wakeup" attribute.

Enabling wake by default was an unintended side-effect. I didn't catch this when I wrote the patch :/ It's been exposing all the incorrect ACPI configurations for better or worse. Mario pushed a patch up earlier to disable thes Wake GPIOs when using S3:
https://github.com/torvalds/linux/commit/d63f11c02b8d3e54bdb65d8c309f73b7f474aec4.
Are you having problems with S3 or S0iX?

Quoted 3 lines
Quoted 1 line

I think this case is S0ix.

We test both cases in our setups.

IMO if a device needs to support wake from S3 the ACPI table needs to define a _PRW and define the proper power resources to keep the device functional during S3.

Quoted 42 lines
Quoted 41 lines
Quoted 23 lines
Quoted 23 lines

---
Hi,

Sending this as RFC because I'm not too familiar with the usage of I2C_CLIENT_WAKE and whether this is something that is expected behaviour in users of I2C devices. On ACPI side I think this is the correct thing to do at least.

  drivers/i2c/i2c-core-base.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
drivers/i2c/i2c-core-base.c +0 -0
index 087e480b624c..7046549bdae7 100644
@@ -527,7 +527,7 @@ static int i2c_device_probe(struct device *dev)
                         goto put_sync_adapter;
                 }
 
Quoted 9 lines

drivers/i2c/i2c-core-base.c static int i2c_device_probe(struct device *dev) · line 530

Quoted 2 lines
drivers/i2c/i2c-core-base.c +1 -1
-               device_init_wakeup(&client->dev, true);
+               device_init_wakeup(&client->dev, false);

This would be a change in behavior for Device Tree. Maybe you can declare a bool enable_wake = true, then in the ACPI branch
(https://github.com/torvalds/linux/blob/master/drivers/i2c/i2c-core-base.c#L495)
set enable_wake = false. This would keep wakes enabled by default on device tree and disabled for ACPI. This matches the original behavior before my patch.

I don't think it's a good idea to make the behaviour different. Drivers in general do not need to know whether the device was enumerated on ACPI or DT or whatnot. Same goes for users who should expect similar behaviour on the same device.

I wonder what is the reason why I2C bus does this for all wake capable devices in the first place? Typically it should be up to the user to enable them not the opposite.

#11 lore raw
Quoted 44 lines · Raul Rangel
Quoted 30 lines · Amadeusz Sławiński
Quoted 26 lines · Mika Westerberg

Hi,

Quoted 23 lines · Raul Rangel

Sorry, resending in plain text mode.

Quoted 11 lines · Mika Westerberg

After commit b38f2d5d9615 ("i2c: acpi: Use ACPI wake capability bit to set wake_irq") the I2C core has been setting I2C_CLIENT_WAKE for ACPI devices if they announce to be wake capable in their device description. However, on certain systems where audio codec has been connected through I2C this causes system suspend to wake up immediately because power to the codec is turned off which pulls the interrupt line "low" triggering wake up.

Possible reason why the interrupt is marked as wake capable is that some codecs apparently support "Wake on Voice" or similar functionality.

That's generally a bug in the ACPI tables. The wake bit shouldn't be set if the power domain for the device is powered off on suspend. The best thing is to fix the ACPI tables, but if you can't, then you can set the ignore_wake flag for the device:
https://github.com/torvalds/linux/blob/master/drivers/gpio/gpiolib-acpi.c#L31.
If that works we can add a quirk for the device:
https://github.com/torvalds/linux/blob/master/drivers/gpio/gpiolib-acpi.c#L1633.

Quoted 11 lines

I've seen this one already and also tried to use it, but it didn't work. Also when I was reading code I wasn't really convinced that it is linked to i2c in any straightforward way. I mean i2c decides in different places that it has wake support (I even added some prints to make sure ;). The code you pointed out decides in
https://github.com/torvalds/linux/blob/master/drivers/gpio/gpiolib-acpi.c#L387
but i2c code seems to decide in
https://github.com/torvalds/linux/blob/master/drivers/i2c/i2c-core-acpi.c#L176
where it just checks if irq flags has wake_capable flag set. When I looked at it previously I was pretty sure it comes straight from BIOS and passes the quirk code you mentioned, still I may have missed something.

Quoted 4 lines

You also need the following patch
https://github.com/torvalds/linux/commit/0e3b175f079247f0d40d2ab695999c309d3a7498,
otherwise the ignore flag only applies to _AEI GPIOs.

I want to point out a non-obvious nuance to that as well. When it's not triggered by an _AEI GPIO then it will not have the parent controller for the quirk but rather the string used for the ACPI device that declared the GPIO.

That's why the quirk that we applied for the Clevo BIOS bug was not AMDI0030 but was instead the string used for the ACPI device.

So if you're experimenting with this make sure you're using the right values, and explicitly look for the string in your kernel log that it's in use.

Quoted 78 lines
Quoted 28 lines
Quoted 9 lines

I think (hope) these systems are not yet available for public so there is a chance that the tables can still be fixed, without need to add any quirks.

@Amadeusz, @Cezary, if that's the case I suggest filing a bug against the BIOS.

Well, I tried custom DSDT and had problems, but I just remembered that I probably need to pass "revision+1" in file, so kernel sees it as a newer version, let me try again. Is it enough to replace "ExclusiveAndWake" with "Exclusive"?

Quoted 13 lines
Quoted 12 lines
Quoted 5 lines

In any case, I don't think we should be enabling wakeup by default on all I2C devices that are wake capable. According to device_init_wakeup() documentation most devices should leave it disabled, with exceptions on devices such as keyboards, power buttons etc. Userspace can enable wakeup as needed by writing to device "power/wakeup" attribute.

Enabling wake by default was an unintended side-effect. I didn't catch this when I wrote the patch :/ It's been exposing all the incorrect ACPI configurations for better or worse. Mario pushed a patch up earlier to disable thes Wake GPIOs when using S3:
https://github.com/torvalds/linux/commit/d63f11c02b8d3e54bdb65d8c309f73b7f474aec4.
Are you having problems with S3 or S0iX?

Quoted 3 lines
Quoted 1 line

I think this case is S0ix.

We test both cases in our setups.

IMO if a device needs to support wake from S3 the ACPI table needs to define a _PRW and define the proper power resources to keep the device functional during S3.

Quoted 42 lines
Quoted 41 lines
Quoted 23 lines
Quoted 23 lines

---
Hi,

Sending this as RFC because I'm not too familiar with the usage of I2C_CLIENT_WAKE and whether this is something that is expected behaviour in users of I2C devices. On ACPI side I think this is the correct thing to do at least.

   drivers/i2c/i2c-core-base.c | 2 +-
   1 file changed, 1 insertion(+), 1 deletion(-)
drivers/i2c/i2c-core-base.c +0 -0
index 087e480b624c..7046549bdae7 100644
@@ -527,7 +527,7 @@ static int i2c_device_probe(struct device *dev)
                          goto put_sync_adapter;
                  }
 
Quoted 9 lines

drivers/i2c/i2c-core-base.c static int i2c_device_probe(struct device *dev) · line 530

Quoted 2 lines
drivers/i2c/i2c-core-base.c +1 -1
-               device_init_wakeup(&client->dev, true);
+               device_init_wakeup(&client->dev, false);

This would be a change in behavior for Device Tree. Maybe you can declare a bool enable_wake = true, then in the ACPI branch
(https://github.com/torvalds/linux/blob/master/drivers/i2c/i2c-core-base.c#L495)
set enable_wake = false. This would keep wakes enabled by default on device tree and disabled for ACPI. This matches the original behavior before my patch.

I don't think it's a good idea to make the behaviour different. Drivers in general do not need to know whether the device was enumerated on ACPI or DT or whatnot. Same goes for users who should expect similar behaviour on the same device.

I wonder what is the reason why I2C bus does this for all wake capable devices in the first place? Typically it should be up to the user to enable them not the opposite.

#12 lore raw
In reply to Raul Rangel
Quoted 77 lines · Raul Rangel
Quoted 30 lines · Amadeusz Sławiński
Quoted 26 lines · Mika Westerberg

Hi,

Quoted 23 lines · Raul Rangel

Sorry, resending in plain text mode.

Quoted 11 lines · Mika Westerberg

After commit b38f2d5d9615 ("i2c: acpi: Use ACPI wake capability bit to set wake_irq") the I2C core has been setting I2C_CLIENT_WAKE for ACPI devices if they announce to be wake capable in their device description. However, on certain systems where audio codec has been connected through I2C this causes system suspend to wake up immediately because power to the codec is turned off which pulls the interrupt line "low" triggering wake up.

Possible reason why the interrupt is marked as wake capable is that some codecs apparently support "Wake on Voice" or similar functionality.

That's generally a bug in the ACPI tables. The wake bit shouldn't be set if the power domain for the device is powered off on suspend. The best thing is to fix the ACPI tables, but if you can't, then you can set the ignore_wake flag for the device:
https://github.com/torvalds/linux/blob/master/drivers/gpio/gpiolib-acpi.c#L31.
If that works we can add a quirk for the device:
https://github.com/torvalds/linux/blob/master/drivers/gpio/gpiolib-acpi.c#L1633.

Quoted 11 lines

I've seen this one already and also tried to use it, but it didn't work. Also when I was reading code I wasn't really convinced that it is linked to i2c in any straightforward way. I mean i2c decides in different places that it has wake support (I even added some prints to make sure ;). The code you pointed out decides in
https://github.com/torvalds/linux/blob/master/drivers/gpio/gpiolib-acpi.c#L387
but i2c code seems to decide in
https://github.com/torvalds/linux/blob/master/drivers/i2c/i2c-core-acpi.c#L176
where it just checks if irq flags has wake_capable flag set. When I looked at it previously I was pretty sure it comes straight from BIOS and passes the quirk code you mentioned, still I may have missed something.

You also need the following patch
https://github.com/torvalds/linux/commit/0e3b175f079247f0d40d2ab695999c309d3a7498,
otherwise the ignore flag only applies to _AEI GPIOs.

Quoted 28 lines
Quoted 9 lines

I think (hope) these systems are not yet available for public so there is a chance that the tables can still be fixed, without need to add any quirks.

@Amadeusz, @Cezary, if that's the case I suggest filing a bug against the BIOS.

Well, I tried custom DSDT and had problems, but I just remembered that I probably need to pass "revision+1" in file, so kernel sees it as a newer version, let me try again. Is it enough to replace "ExclusiveAndWake" with "Exclusive"?

Quoted 13 lines
Quoted 12 lines
Quoted 5 lines

In any case, I don't think we should be enabling wakeup by default on all I2C devices that are wake capable. According to device_init_wakeup() documentation most devices should leave it disabled, with exceptions on devices such as keyboards, power buttons etc. Userspace can enable wakeup as needed by writing to device "power/wakeup" attribute.

Enabling wake by default was an unintended side-effect. I didn't catch this when I wrote the patch :/ It's been exposing all the incorrect ACPI configurations for better or worse. Mario pushed a patch up earlier to disable thes Wake GPIOs when using S3:
https://github.com/torvalds/linux/commit/d63f11c02b8d3e54bdb65d8c309f73b7f474aec4.
Are you having problems with S3 or S0iX?

Quoted 7 lines
Quoted 3 lines
Quoted 1 line

I think this case is S0ix.

We test both cases in our setups.

IMO if a device needs to support wake from S3 the ACPI table needs to define a _PRW and define the proper power resources to keep the device functional during S3.

Yes, it should, but there's more to it than this.

First of all, the interrupt in question needs to be enabled prior to S3 entry. That will happen if device_may_wakeup() is true for the given device AFAICS.

Second, the device should be in a suitable power state which _S3W is about. That should also happen via ACPI PM.

Next, _PRW should return the list of power resources that need to be _ON prior to S3 entry for the device to be able to wake the system.

However, _PRW also provides the information on which sleep states the system can be woken up from by the given device and S3 need not be one of them.

The kernel will not prevent S3 from being entered if that is the case, so it doesn't follow the spec literally in that respect.