Skip to contentSkip to conversation

Revert "venus: pm_helpers: Fix error check in vcodec_domains_get()"

PATCH In review Dan Carpenter 2 participants 5 messages Jul 3, 2023 – Jul 10, 2023

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Revert "venus: pm_helpers: Fix error check in vcodec_domains_get()"

This reverts commit 0f6e8d8c94a82e85e1b9b62a7671990740dc6f70.

The reverted commit was based on static analysis and a misunderstanding of how PTR_ERR() and NULLs are supposed to work. When a function returns both pointer errors and NULL then normally the NULL means "continue operating without a feature because it was deliberately turned off". The NULL should not be treated as a failure. If a driver cannot work when that feature is disabled then the KConfig should enforce that the function cannot return NULL. We should not need to test for it.

In this code, the patch breaks the venus driver if CONFIG_PM is disabled.

---
This patch is also based on static analysis and review so probably best to be cautious. My guess is that very few people disable CONFIG_PM these days so that's why the bug wasn't caught.

 drivers/media/platform/qcom/venus/pm_helpers.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/media/platform/qcom/venus/pm_helpers.c +2 -2
index 48c9084bb4db..c93d2906e4c7 100644
@@ -869,8 +869,8 @@ static int vcodec_domains_get(struct venus_core *core)
 	for (i = 0; i < res->vcodec_pmdomains_num; i++) {
 		pd = dev_pm_domain_attach_by_name(dev,
 						  res->vcodec_pmdomains[i]);
-		if (IS_ERR_OR_NULL(pd))
-			return PTR_ERR(pd) ? : -ENODATA;
+		if (IS_ERR(pd))
+			return PTR_ERR(pd);
 		core->pmdomains[i] = pd;
 	}
 
Signature · 1 line
2.39.2
#1 lore raw

Hi Dan,

Quoted 28 lines · Dan Carpenter

This reverts commit 0f6e8d8c94a82e85e1b9b62a7671990740dc6f70.

The reverted commit was based on static analysis and a misunderstanding of how PTR_ERR() and NULLs are supposed to work. When a function returns both pointer errors and NULL then normally the NULL means "continue operating without a feature because it was deliberately turned off". The NULL should not be treated as a failure. If a driver cannot work when that feature is disabled then the KConfig should enforce that the function cannot return NULL. We should not need to test for it.

In this code, the patch breaks the venus driver if CONFIG_PM is disabled.

---
This patch is also based on static analysis and review so probably best to be cautious. My guess is that very few people disable CONFIG_PM these days so that's why the bug wasn't caught.

 drivers/media/platform/qcom/venus/pm_helpers.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/media/platform/qcom/venus/pm_helpers.c +0 -0
index 48c9084bb4db..c93d2906e4c7 100644

drivers/media/platform/qcom/venus/pm_helpers.c static int vcodec_domains_get(struct venus_core *core) · line 873

Quoted 7 lines
drivers/media/platform/qcom/venus/pm_helpers.c +2 -2
@@ -869,8 +869,8 @@ static int vcodec_domains_get(struct venus_core *core)
 	for (i = 0; i < res->vcodec_pmdomains_num; i++) {
 		pd = dev_pm_domain_attach_by_name(dev,
 						  res->vcodec_pmdomains[i]);
-		if (IS_ERR_OR_NULL(pd))
-			return PTR_ERR(pd) ? : -ENODATA;
+		if (IS_ERR(pd))
+			return PTR_ERR(pd);

Trying to understand if pm domains for Venus (pd) is returned NULL here, how would the driver be able to enable and disable those power domains at [1]. And without that, video usecase would be functional ?

[1]
https://elixir.bootlin.com/linux/latest/source/drivers/media/platform/qcom/venus/pm_helpers.c#L1043

Quoted 3 lines
drivers/media/platform/qcom/venus/pm_helpers.c +0 -0
 		core->pmdomains[i] = pd;
 	}
 
#2 lore raw
Quoted 42 lines · Vikash Garodia

Hi Dan,

Quoted 28 lines · Dan Carpenter

This reverts commit 0f6e8d8c94a82e85e1b9b62a7671990740dc6f70.

The reverted commit was based on static analysis and a misunderstanding of how PTR_ERR() and NULLs are supposed to work. When a function returns both pointer errors and NULL then normally the NULL means "continue operating without a feature because it was deliberately turned off". The NULL should not be treated as a failure. If a driver cannot work when that feature is disabled then the KConfig should enforce that the function cannot return NULL. We should not need to test for it.

In this code, the patch breaks the venus driver if CONFIG_PM is disabled.

---
This patch is also based on static analysis and review so probably best to be cautious. My guess is that very few people disable CONFIG_PM these days so that's why the bug wasn't caught.

 drivers/media/platform/qcom/venus/pm_helpers.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/media/platform/qcom/venus/pm_helpers.c +0 -0
index 48c9084bb4db..c93d2906e4c7 100644

drivers/media/platform/qcom/venus/pm_helpers.c static int vcodec_domains_get(struct venus_core *core) · line 873

Quoted 7 lines
drivers/media/platform/qcom/venus/pm_helpers.c +2 -2
@@ -869,8 +869,8 @@ static int vcodec_domains_get(struct venus_core *core)
 	for (i = 0; i < res->vcodec_pmdomains_num; i++) {
 		pd = dev_pm_domain_attach_by_name(dev,
 						  res->vcodec_pmdomains[i]);
-		if (IS_ERR_OR_NULL(pd))
-			return PTR_ERR(pd) ? : -ENODATA;
+		if (IS_ERR(pd))
+			return PTR_ERR(pd);

Trying to understand if pm domains for Venus (pd) is returned NULL here, how would the driver be able to enable and disable those power domains at [1]. And without that, video usecase would be functional ?

Quoted 2 lines

[1]
https://elixir.bootlin.com/linux/latest/source/drivers/media/platform/qcom/venus/pm_helpers.c#L1043

I don't understand. The CONFIG_PM is Power Management. If you deliberately disable power management then then, obviously, power domains are not going to be enabled. The __pm_runtime_resume() function turns into:

include/linux/pm_runtime.h
268 static inline int __pm_runtime_resume(struct device *dev, int rpmflags) 269 {
270 return 1;
271 }

In real life, most people are going to want power management.

This is a video accelerator. I would expect that it could still work without power management. If it really can't then that should be enforced in the Kconfig. Having this code here which says "We can't load without CONFIG_PM" is wrong, it should be "We can't compile without CONFIG_PM".

regards,
dan carpenter

#3 lore raw
Quoted 53 lines · Dan Carpenter
Quoted 42 lines · Vikash Garodia

Hi Dan,

Quoted 28 lines · Dan Carpenter

This reverts commit 0f6e8d8c94a82e85e1b9b62a7671990740dc6f70.

The reverted commit was based on static analysis and a misunderstanding of how PTR_ERR() and NULLs are supposed to work. When a function returns both pointer errors and NULL then normally the NULL means "continue operating without a feature because it was deliberately turned off". The NULL should not be treated as a failure. If a driver cannot work when that feature is disabled then the KConfig should enforce that the function cannot return NULL. We should not need to test for it.

In this code, the patch breaks the venus driver if CONFIG_PM is disabled.

---
This patch is also based on static analysis and review so probably best to be cautious. My guess is that very few people disable CONFIG_PM these days so that's why the bug wasn't caught.

 drivers/media/platform/qcom/venus/pm_helpers.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/media/platform/qcom/venus/pm_helpers.c +0 -0
index 48c9084bb4db..c93d2906e4c7 100644

drivers/media/platform/qcom/venus/pm_helpers.c static int vcodec_domains_get(struct venus_core *core) · line 873

Quoted 7 lines
drivers/media/platform/qcom/venus/pm_helpers.c +2 -2
@@ -869,8 +869,8 @@ static int vcodec_domains_get(struct venus_core *core)
 	for (i = 0; i < res->vcodec_pmdomains_num; i++) {
 		pd = dev_pm_domain_attach_by_name(dev,
 						  res->vcodec_pmdomains[i]);
-		if (IS_ERR_OR_NULL(pd))
-			return PTR_ERR(pd) ? : -ENODATA;
+		if (IS_ERR(pd))
+			return PTR_ERR(pd);

Trying to understand if pm domains for Venus (pd) is returned NULL here, how would the driver be able to enable and disable those power domains at [1]. And without that, video usecase would be functional ?

Quoted 2 lines

[1]
https://elixir.bootlin.com/linux/latest/source/drivers/media/platform/qcom/venus/pm_helpers.c#L1043

I don't understand. The CONFIG_PM is Power Management. If you deliberately disable power management then then, obviously, power domains are not going to be enabled. The __pm_runtime_resume() function turns into:

include/linux/pm_runtime.h
268 static inline int __pm_runtime_resume(struct device *dev, int rpmflags)

Quoted 8 lines
   269  {
   270          return 1;
   271  }

In real life, most people are going to want power management.

This is a video accelerator. I would expect that it could still work without power management.

It won't, without those 2 power domains enabled for Venus. Does it work for you when you disabled the config ?

If it really can't then that should be

Quoted 3 lines

enforced in the Kconfig. Having this code here which says "We can't load without CONFIG_PM" is wrong, it should be "We can't compile without CONFIG_PM".

Its better enforced in Kconfig. By allowing NULL for dev_pm_domain_attach_by_name does not still indicate that the functionality is dependent on CONFIG_PM to be enabled.

Thanks,
Vikash

#4 lore raw
Quoted 4 lines · Vikash Garodia
Quoted 2 lines

This is a video accelerator. I would expect that it could still work without power management.

It won't, without those 2 power domains enabled for Venus. Does it work for you when you disabled the config ?

I have not tested this, I'm reviewing static checker bugs, as I explained here:

Quoted 8 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

This patch is also based on static analysis and review so probably best to be cautious. My guess is that very few people disable CONFIG_PM these days so that's why the bug wasn't caught.

Its better enforced in Kconfig. By allowing NULL for dev_pm_domain_attach_by_name does not still indicate that the functionality is dependent on CONFIG_PM to be enabled.

The way I'm writing it is the correct way. I explain this better in my blog.

https://staticthinking.wordpress.com/2022/08/01/mixing-error-pointers-and-null/

Currently it looks like the Kconfig does not have a dependency on CONFIG_PM. If we fix the the Kconfig, but leave the "harmless" bug here then it will still show up as a static checker warning when COMPILE_TEST is enabled.

We should just do it in the correct way even after we fix the Kconfig.

regards,
dan carpenter