Skip to contentSkip to conversation

ACPI: PM: Avoid attaching ACPI PM domain to certain devices

PATCH In review Rafael J. Wysocki 4 participants 6 messages Dec 4, 2019 – Dec 12, 2019 1 message not in the archive ACPI

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ACPI: PM: Avoid attaching ACPI PM domain to certain devices

From: Rafael J. Wysocki <[email protected]>

Certain ACPI-enumerated devices represented as platform devices in Linux, like fans, require special low-level power management handling implemented by their drivers that is not in agreement with the ACPI PM domain behavior. That leads to problems with managing ACPI fans during system-wide suspend and resume.

For this reason, make acpi_dev_pm_attach() skip the affected devices by adding a list of device IDs to avoid to it and putting the IDs of the affected devices into that list.

Fixes e5cc8ef31267 (ACPI / PM: Provide ACPI PM callback routines for subsystems)
Cc 3.10+ <[email protected]> # 3.10+

---

Rui,

Please test this on the machine(s) affected by the fan suspend/resume issues.

I don't really see any cleaner way to address this problem, because the ACPI PM domain should not be used with the devices in question even if the driver that binds to them is not loaded.

Cheers,
Rafael

---
 drivers/acpi/device_pm.c |   12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)
drivers/acpi/device_pm.c +11 -1
Index: linux-pm/drivers/acpi/device_pm.c
===================================================================
@@ -1314,9 +1314,19 @@ static void acpi_dev_pm_detach(struct de
  */
 int acpi_dev_pm_attach(struct device *dev, bool power_on)
 {
+	/*
+	 * Skip devices whose ACPI companions match the device IDs below,
+	 * because they require special power management handling incompatible
+	 * with the generic ACPI PM domain.
+	 */
+	static const struct acpi_device_id special_pm_ids[] = {
+		{"PNP0C0B", }, /* Generic ACPI fan */
+		{"INT3404", }, /* Fan */
+		{}
+	};
 	struct acpi_device *adev = ACPI_COMPANION(dev);
 
-	if (!adev)
+	if (!adev || !acpi_match_device_ids(adev, special_pm_ids))
 		return 0;
 
 	/*
#1 lore raw
Quoted 17 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Certain ACPI-enumerated devices represented as platform devices in Linux, like fans, require special low-level power management handling implemented by their drivers that is not in agreement with the ACPI PM domain behavior. That leads to problems with managing ACPI fans during system-wide suspend and resume.

For this reason, make acpi_dev_pm_attach() skip the affected devices by adding a list of device IDs to avoid to it and putting the IDs of the affected devices into that list.

Fixes e5cc8ef31267 (ACPI / PM: Provide ACPI PM callback routines for

subsystems)

Quoted 6 lines

---

Rui,

Please test this on the machine(s) affected by the fan suspend/resume issues.

Sure, Todd and I will re-run stress test with this patch applied when 5.5-rc1 released.

thanks,
rui

Quoted 44 lines

I don't really see any cleaner way to address this problem, because the
ACPI PM domain should not be used with the devices in question even if
the driver that binds to them is not loaded.

Cheers,
Rafael

---
 drivers/acpi/device_pm.c |   12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

Index: linux-pm/drivers/acpi/device_pm.c
===================================================================

drivers/acpi/device_pm.c +3 -0
@@ -1314,9 +1314,19 @@ static void acpi_dev_pm_detach(struct de
  */
 int acpi_dev_pm_attach(struct device *dev, bool power_on)
 {
+	/*
+	 * Skip devices whose ACPI companions match the device IDs
 below,
+	 * because they require special power management handling

incompatible

drivers/acpi/device_pm.c +8 -1
+	 * with the generic ACPI PM domain.
+	 */
+	static const struct acpi_device_id special_pm_ids[] = {
+		{"PNP0C0B", }, /* Generic ACPI fan */
+		{"INT3404", }, /* Fan */
+		{}
+	};
 	struct acpi_device *adev = ACPI_COMPANION(dev);
 
-	if (!adev)
+	if (!adev || !acpi_match_device_ids(adev, special_pm_ids))
 		return 0;
 
 	/*
#2 lore raw
Quoted 24 lines · Zhang Rui
Quoted 23 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Certain ACPI-enumerated devices represented as platform devices in Linux, like fans, require special low-level power management handling
implemented by their drivers that is not in agreement with the ACPI PM domain behavior. That leads to problems with managing ACPI fans during system-wide suspend and resume.

For this reason, make acpi_dev_pm_attach() skip the affected devices
by adding a list of device IDs to avoid to it and putting the IDs of
the affected devices into that list.

Fixes e5cc8ef31267 (ACPI / PM: Provide ACPI PM callback routines

for
subsystems)

---

Quoted 8 lines
Quoted 5 lines

Rui,

Please test this on the machine(s) affected by the fan suspend/resume
issues.

Sure, Todd and I will re-run stress test with this patch applied when 5.5-rc1 released.

I've applied it 5.4.0 and will do a full stress test run this weekend in the lab (where 7 out of 20 machines have this issue). The kernel will be called "5.4.0-acpifanfix", and the data should be ready Sunday Dec 8.

This is the issue I'll test for:
https://bugzilla.kernel.org/show_bug.cgi?id=204321

Quoted 50 lines

thanks,
rui

Quoted 44 lines

I don't really see any cleaner way to address this problem, because the
ACPI PM domain should not be used with the devices in question even if
the driver that binds to them is not loaded.

Cheers,
Rafael

---
 drivers/acpi/device_pm.c |   12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

Index: linux-pm/drivers/acpi/device_pm.c
===================================================================

drivers/acpi/device_pm.c +3 -0
@@ -1314,9 +1314,19 @@ static void acpi_dev_pm_detach(struct de
  */
 int acpi_dev_pm_attach(struct device *dev, bool power_on)
 {
+	/*
+	 * Skip devices whose ACPI companions match the device IDs
 below,
+	 * because they require special power management handling

incompatible

drivers/acpi/device_pm.c +8 -1
+	 * with the generic ACPI PM domain.
+	 */
+	static const struct acpi_device_id special_pm_ids[] = {
+		{"PNP0C0B", }, /* Generic ACPI fan */
+		{"INT3404", }, /* Fan */
+		{}
+	};
 	struct acpi_device *adev = ACPI_COMPANION(dev);
 
-	if (!adev)
+	if (!adev || !acpi_match_device_ids(adev, special_pm_ids))
 		return 0;
 
 	/*
#3 lore raw
Quoted 38 lines · Todd Brandt
Quoted 29 lines · Zhang Rui
Quoted 28 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Certain ACPI-enumerated devices represented as platform devices in
Linux, like fans, require special low-level power management handling
implemented by their drivers that is not in agreement with the ACPI
PM domain behavior. That leads to problems with managing ACPI fans
during system-wide suspend and resume.

For this reason, make acpi_dev_pm_attach() skip the affected devices
by adding a list of device IDs to avoid to it and putting the IDs of
the affected devices into that list.

Fixes e5cc8ef31267 (ACPI / PM: Provide ACPI PM callback routines

for
subsystems)

---

Rui,

Quoted 7 lines
Quoted 3 lines

Please test this on the machine(s) affected by the fan suspend/resume
issues.

Sure, Todd and I will re-run stress test with this patch applied when
5.5-rc1 released.

Quoted 9 lines

I've applied it 5.4.0 and will do a full stress test run this weekend in the lab (where 7 out of 20 machines have this issue). The kernel will be called "5.4.0-acpifanfix", and the data should be ready Sunday
Dec 8.

This is the issue I'll test for:
https://bugzilla.kernel.org/show_bug.cgi?id=204321

The test data is in and I'm very happy to report that the patch works extremely well. Here's the data gathered on the bugzilla issue 204321:
ACPI fan resume is too long.

[With the patch]

Kernel              Host                      Test Run    Count Rate
5.4.0-acpifanfix    otcpl-whl-u-clear         mem-x2351    1    0.04%
5.4.0-acpifanfix    otcpl-z170x-ud5           mem-x2369    0    0%
5.4.0-acpifanfix    otcpl-z170x-ud5           freeze-x2492 0    0%
5.4.0-acpifanfix    otcpl-whl-u-clear         freeze-x2411 0    0%
5.4.0-acpifanfix    otcpl-whl-u               mem-x2289    0    0%
5.4.0-acpifanfix    otcpl-whl-u               freeze-x2389 0    0%
5.4.0-acpifanfix    otcpl-icl-u-2             mem-x2273    0    0%
5.4.0-acpifanfix    otcpl-icl-u-2             freeze-x817  0    0%
5.4.0-acpifanfix    otcpl-glk-rvp-1           freeze-x122  0    0%
5.4.0-acpifanfix    otcpl-dell-p5510-xeon-1   mem-x1851    0    0%
5.4.0-acpifanfix    otcpl-dell-p5510-xeon-1   freeze-x1994 0    0%
5.4.0-acpifanfix    otcpl-dell-inspiron-3493  mem-x1722    0    0%
5.4.0-acpifanfix    otcpl-dell-inspiron-3493  freeze-x2102 0    0%
5.4.0-acpifanfix    otcpl-cfl-u-01            mem-x39      0    0%
5.4.0-acpifanfix    otcpl-cfl-u-01            freeze-x2091 0    0%
5.4.0-acpifanfix    otcpl-cfl-h               mem-x415     0    0%
5.4.0-acpifanfix    otcpl-cfl-h               freeze-x2265 0    0%
5.4.0-acpifanfix    otcpl-aml-y               mem-x3126    0    0%
5.4.0-acpifanfix    otcpl-aml-y               freeze-x2288 0    0%

[Without the patch]

Kernel    Host                      Test Run    Count Rate
5.3.0+    otcpl-icl-u-2             mem-x2571    2571 100.00%
5.3.0+    otcpl-whl-u               mem-x2497    2497 100.00%
5.3.0+    otcpl-cfl-h               mem-x2068    2068 100.00%
5.3.0+    otcpl-glk-rvp-1           mem-x75      74   98.67%
5.3.0+    otcpl-cfl-u-01            mem-x1074    1050 97.77%
5.3.0+    otcpl-glk-rvp-1           freeze-x45   12   26.67%
5.3.0+    otcpl-whl-u               freeze-x2649 428  16.16%
5.3.0+    otcpl-aml-y               freeze-x2434 373  15.32%
5.3.0+    otcpl-cfl-u-01            freeze-x2419 123  5.08%
5.3.0+    otcpl-latexo-ivb-cpt      freeze-x1914 97   5.07%
5.3.0+    otcpl-whl-u-clear         mem-x2640    69   2.61%
5.3.0+    otcpl-icl-u-2             freeze-x2757 59   2.14%
5.3.0+    otcpl-whl-u-clear         freeze-x2830 53   1.87%
5.3.0+    otcpl-tgl-rvp             freeze-x2086 20   0.96%
5.3.0+    otcpl-cfl-h               freeze-x2457 8    0.33%
5.3.0+    otcpl-latexo-ivb-cpt      mem-x2000    4    0.20%
5.3.0+    otcpl-aml-y               mem-x2727    2    0.07%
5.3.0+    otcpl-z170x-ud5           mem-x2669    0    0%    
5.3.0+    otcpl-z170x-ud5           freeze-x2881 0    0%    
5.3.0+    otcpl-lenovo-ideapad-130  mem-x2000    0    0%    
5.3.0+    otcpl-lenovo-ideapad-130  freeze-x2000 0    0%    
5.3.0+    otcpl-dell-p5510-xeon-2   mem-x570     0    0%    
5.3.0+    otcpl-dell-p5510-xeon-2   freeze-x1093 0    0%    
5.3.0+    otcpl-dell-p5510-xeon-1   mem-x2209    0    0%    
5.3.0+    otcpl-dell-p5510-xeon-1   freeze-x2431 0    0%    
5.3.0+    otcpl-dell-inspiron-3493  freeze-x1170 0    0%    
5.3.0+    otcpl-chromebook-hsw      freeze-x1305 0    0%    
5.3.0+    otcpl-chromebook-hsw      freeze-x368  0    0%    
Quoted 53 lines
Quoted 53 lines

thanks,
rui

Quoted 47 lines

I don't really see any cleaner way to address this problem, because
the
ACPI PM domain should not be used with the devices in question even
if
the driver that binds to them is not loaded.

Cheers,
Rafael

---
 drivers/acpi/device_pm.c |   12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

Index: linux-pm/drivers/acpi/device_pm.c
=================================================================
==

drivers/acpi/device_pm.c +3 -0
@@ -1314,9 +1314,19 @@ static void acpi_dev_pm_detach(struct de
  */
 int acpi_dev_pm_attach(struct device *dev, bool power_on)
 {
+	/*
+	 * Skip devices whose ACPI companions match the device IDs
 below,
+	 * because they require special power management handling

incompatible

drivers/acpi/device_pm.c +8 -1
+	 * with the generic ACPI PM domain.
+	 */
+	static const struct acpi_device_id special_pm_ids[] = {
+		{"PNP0C0B", }, /* Generic ACPI fan */
+		{"INT3404", }, /* Fan */
+		{}
+	};
 	struct acpi_device *adev = ACPI_COMPANION(dev);
 
-	if (!adev)
+	if (!adev || !acpi_match_device_ids(adev, special_pm_ids))
 		return 0;
 
 	/*
#4 lore raw
Quoted 100 lines · Todd Brandt
Quoted 38 lines · Todd Brandt
Quoted 29 lines · Zhang Rui
Quoted 28 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Certain ACPI-enumerated devices represented as platform devices in
Linux, like fans, require special low-level power management handling
implemented by their drivers that is not in agreement with the ACPI
PM domain behavior. That leads to problems with managing ACPI fans
during system-wide suspend and resume.

For this reason, make acpi_dev_pm_attach() skip the affected devices
by adding a list of device IDs to avoid to it and putting the IDs of
the affected devices into that list.

Fixes e5cc8ef31267 (ACPI / PM: Provide ACPI PM callback routines

for
subsystems)

---

Rui,

Quoted 7 lines
Quoted 3 lines

Please test this on the machine(s) affected by the fan suspend/resume
issues.

Sure, Todd and I will re-run stress test with this patch applied when
5.5-rc1 released.

Quoted 9 lines

I've applied it 5.4.0 and will do a full stress test run this weekend in the lab (where 7 out of 20 machines have this issue). The kernel will be called "5.4.0-acpifanfix", and the data should be ready Sunday
Dec 8.

This is the issue I'll test for:
https://bugzilla.kernel.org/show_bug.cgi?id=204321

The test data is in and I'm very happy to report that the patch works extremely well. Here's the data gathered on the bugzilla issue 204321:
ACPI fan resume is too long.

[With the patch]

Kernel              Host                      Test Run    Count Rate
5.4.0-acpifanfix    otcpl-whl-u-clear         mem-x2351    1    0.04%
5.4.0-acpifanfix    otcpl-z170x-ud5           mem-x2369    0    0%
5.4.0-acpifanfix    otcpl-z170x-ud5           freeze-x2492 0    0%
5.4.0-acpifanfix    otcpl-whl-u-clear         freeze-x2411 0    0%
5.4.0-acpifanfix    otcpl-whl-u               mem-x2289    0    0%
5.4.0-acpifanfix    otcpl-whl-u               freeze-x2389 0    0%
5.4.0-acpifanfix    otcpl-icl-u-2             mem-x2273    0    0%
5.4.0-acpifanfix    otcpl-icl-u-2             freeze-x817  0    0%
5.4.0-acpifanfix    otcpl-glk-rvp-1           freeze-x122  0    0%
5.4.0-acpifanfix    otcpl-dell-p5510-xeon-1   mem-x1851    0    0%
5.4.0-acpifanfix    otcpl-dell-p5510-xeon-1   freeze-x1994 0    0%
5.4.0-acpifanfix    otcpl-dell-inspiron-3493  mem-x1722    0    0%
5.4.0-acpifanfix    otcpl-dell-inspiron-3493  freeze-x2102 0    0%
5.4.0-acpifanfix    otcpl-cfl-u-01            mem-x39      0    0%
5.4.0-acpifanfix    otcpl-cfl-u-01            freeze-x2091 0    0%
5.4.0-acpifanfix    otcpl-cfl-h               mem-x415     0    0%
5.4.0-acpifanfix    otcpl-cfl-h               freeze-x2265 0    0%
5.4.0-acpifanfix    otcpl-aml-y               mem-x3126    0    0%
5.4.0-acpifanfix    otcpl-aml-y               freeze-x2288 0    0%

[Without the patch]

Kernel    Host                      Test Run    Count Rate
5.3.0+    otcpl-icl-u-2             mem-x2571    2571 100.00%
5.3.0+    otcpl-whl-u               mem-x2497    2497 100.00%
5.3.0+    otcpl-cfl-h               mem-x2068    2068 100.00%
5.3.0+    otcpl-glk-rvp-1           mem-x75      74   98.67%
5.3.0+    otcpl-cfl-u-01            mem-x1074    1050 97.77%
5.3.0+    otcpl-glk-rvp-1           freeze-x45   12   26.67%
5.3.0+    otcpl-whl-u               freeze-x2649 428  16.16%
5.3.0+    otcpl-aml-y               freeze-x2434 373  15.32%
5.3.0+    otcpl-cfl-u-01            freeze-x2419 123  5.08%
5.3.0+    otcpl-latexo-ivb-cpt      freeze-x1914 97   5.07%
5.3.0+    otcpl-whl-u-clear         mem-x2640    69   2.61%
5.3.0+    otcpl-icl-u-2             freeze-x2757 59   2.14%
5.3.0+    otcpl-whl-u-clear         freeze-x2830 53   1.87%
5.3.0+    otcpl-tgl-rvp             freeze-x2086 20   0.96%
5.3.0+    otcpl-cfl-h               freeze-x2457 8    0.33%
5.3.0+    otcpl-latexo-ivb-cpt      mem-x2000    4    0.20%
5.3.0+    otcpl-aml-y               mem-x2727    2    0.07%
5.3.0+    otcpl-z170x-ud5           mem-x2669    0    0%
5.3.0+    otcpl-z170x-ud5           freeze-x2881 0    0%
5.3.0+    otcpl-lenovo-ideapad-130  mem-x2000    0    0%
Quoted 8 lines
5.3.0+    otcpl-lenovo-ideapad-130  freeze-x2000 0    0%
5.3.0+    otcpl-dell-p5510-xeon-2   mem-x570     0    0%
5.3.0+    otcpl-dell-p5510-xeon-2   freeze-x1093 0    0%
5.3.0+    otcpl-dell-p5510-xeon-1   mem-x2209    0    0%
5.3.0+    otcpl-dell-p5510-xeon-1   freeze-x2431 0    0%
5.3.0+    otcpl-dell-inspiron-3493  freeze-x1170 0    0%
5.3.0+    otcpl-chromebook-hsw      freeze-x1305 0    0%
5.3.0+    otcpl-chromebook-hsw      freeze-x368  0    0%

Great, thanks for the verification!

I will queue up the patch for the next ACPI pull request, then.

Thanks!

A message that is not in the archive (sent off-list or not yet received): [email protected]

#6 lore raw
Quoted 48 lines · CET Sasha Levin

Hi,

[This is an automated email]

This commit has been processed because it contains a "Fixes:" tag,
fixing commit: e5cc8ef31267 ("ACPI / PM: Provide ACPI PM callback routines for subsystems").

The bot has tested the following trees: v5.4.2, v5.3.15, v4.19.88, v4.14.158, v4.9.206, v4.4.206.

v5.4.2: Build OK!
v5.3.15: Build OK!
v4.19.88: Build OK!
v4.14.158: Failed to apply! Possible dependencies:
001d50c9a14f ("PM / Domains: Remove obsolete "samsung,power-domain" check") 919b7308fcc4 ("PM / Domains: Allow a better error handling of dev_pm_domain_attach()")

v4.9.206: Failed to apply! Possible dependencies:
001d50c9a14f ("PM / Domains: Remove obsolete "samsung,power-domain" check") 34994692216b ("PM / Domains: Do not print PM domain add error message if EPROBE_DEFER") 35241d12f750 ("PM / Domains: Abstract genpd locking")
59d65b73a23c ("PM / Domains: Make genpd state allocation dynamic") 72038df3c580 ("PM / Domains: Fix error path during attach in genpd") 919b7308fcc4 ("PM / Domains: Allow a better error handling of dev_pm_domain_attach()")

v4.4.206: Failed to apply! Possible dependencies:
001d50c9a14f ("PM / Domains: Remove obsolete "samsung,power-domain" check") 0106ef5146f9 ("PM / domains: fix lockdep issue for all subdomains") 076395cae203 ("PM / Domains: Propagate start and restore errors during runtime resume") 1f7371b2a5fa ("drm/amd/powerplay: add basic powerplay framework") 25030321ba28 ("drm/amd: add pm domain for ACP IP sub blocks")
288912cb95d1 ("drm/amdgpu: use $(src) in Makefile (v2)")
34994692216b ("PM / Domains: Do not print PM domain add error message if EPROBE_DEFER") 35241d12f750 ("PM / Domains: Abstract genpd locking")
39dd0f234fc3 ("PM / Domains: Allow genpd to power on during system PM phases") 3fe577107ccf ("PM / Domains: Add support for removing PM domains") 4d23a5e84806 ("PM / Domains: Allow runtime PM during system PM phases") 53af16f79fbc ("PM / Domains: Silence compiler warning for an unused function") 59d65b73a23c ("PM / Domains: Make genpd state allocation dynamic") 624c8df7d282 ("PM / Domains: Remove redundant pm_runtime_get|put*() in pm_genpd_prepare()") 72038df3c580 ("PM / Domains: Fix error path during attach in genpd") 7eb231c337e0 ("PM / Domains: Convert pm_genpd_init() to return an error code") 8f9945558785 ("drm/amdgpu/acp: fix resume on CZ systems with AZ audio") 919b7308fcc4 ("PM / Domains: Allow a better error handling of dev_pm_domain_attach()") a8fe58cec351 ("drm/amd: add ACP driver support")
cdb300a041f5 ("PM / Domains: Fix potential deadlock while adding/removing subdomains") fc5cbf0c94b6 ("PM / Domains: Support for multiple states")

Quoted 3 lines

NOTE: The patch will not be queued to stable trees until it is upstream.

How should we proceed with this patch?

This has been present in linux-next for some time and the commit carries a Cc: stable tag. No need to do anything special AFAICS.

Thanks!