Skip to contentSkip to conversation

thermal: devfreq_cooling: use local ops instead of global ops

PATCH In review Kant Fan 3 participants 7 messages Mar 25, 2022 – Apr 23, 2022

Subsystems (2)
lore
Share & tools
lore

Kant Fan

author
#0 lore raw

PATCH 1/1 thermal: devfreq_cooling: use local ops instead of global ops

commit 7b62935828266658714f81d4e9176edad808dc70 upstream.

Fix access illegal address problem in following condition:
There are muti devfreq cooling devices in system, some of them register with dfc_power but other does not, power model ops such as state2power will append to global devfreq_cooling_ops when the cooling device with dfc_power register. It makes the cooling device without dfc_power also use devfreq_cooling_ops after appending when register later by of_devfreq_cooling_register_power() or of_devfreq_cooling_register().

IPA governor regards the cooling devices without dfc_power as a power actor because they also have power model ops, and will access illegal address at dfc->power_ops when execute cdev->ops->get_requested_power or cdev->ops->power2state. As the calltrace below shows:

Unable to handle kernel NULL pointer dereference at virtual address
00000008
...
calltrace:
[<c06e5488>] devfreq_cooling_power2state+0x24/0x184
[<c06df420>] power_actor_set_power+0x54/0xa8
[<c06e3774>] power_allocator_throttle+0x770/0x97c
[<c06dd120>] handle_thermal_trip+0x1b4/0x26c
[<c06ddb48>] thermal_zone_device_update+0x154/0x208
[<c014159c>] process_one_work+0x1ec/0x36c
[<c0141c58>] worker_thread+0x204/0x2ec
[<c0146788>] kthread+0x140/0x154
[<c01010e8>] ret_from_fork+0x14/0x2c
Fixes a76caf55e5b35 ("thermal: Add devfreq cooling")
---
 drivers/thermal/devfreq_cooling.c | 25 ++++++++++++++++++-------
 1 file changed, 18 insertions(+), 7 deletions(-)
drivers/thermal/devfreq_cooling.c +18 -7
index dfab49a67252..d36f70513e6a 100644
@@ -462,22 +462,29 @@ of_devfreq_cooling_register_power(struct device_node *np, struct devfreq *df,
 {
 	struct thermal_cooling_device *cdev;
 	struct devfreq_cooling_device *dfc;
+	struct thermal_cooling_device_ops *ops;
 	char dev_name[THERMAL_NAME_LENGTH];
 	int err;
 
-	dfc = kzalloc(sizeof(*dfc), GFP_KERNEL);
-	if (!dfc)
+	ops = kmemdup(&devfreq_cooling_ops, sizeof(*ops), GFP_KERNEL);
+	if (!ops)
 		return ERR_PTR(-ENOMEM);
 
+	dfc = kzalloc(sizeof(*dfc), GFP_KERNEL);
+	if (!dfc) {
+		err = -ENOMEM;
+		goto free_ops;
+	}
+
 	dfc->devfreq = df;
 
 	if (dfc_power) {
 		dfc->power_ops = dfc_power;
 
-		devfreq_cooling_ops.get_requested_power =
+		ops->get_requested_power =
 			devfreq_cooling_get_requested_power;
-		devfreq_cooling_ops.state2power = devfreq_cooling_state2power;
-		devfreq_cooling_ops.power2state = devfreq_cooling_power2state;
+		ops->state2power = devfreq_cooling_state2power;
+		ops->power2state = devfreq_cooling_power2state;
 	}
 
 	err = devfreq_cooling_gen_tables(dfc);
@@ -497,8 +504,7 @@ of_devfreq_cooling_register_power(struct device_node *np, struct devfreq *df,
 
 	snprintf(dev_name, sizeof(dev_name), "thermal-devfreq-%d", dfc->id);
 
-	cdev = thermal_of_cooling_device_register(np, dev_name, dfc,
-						  &devfreq_cooling_ops);
+	cdev = thermal_of_cooling_device_register(np, dev_name, dfc, ops);
 	if (IS_ERR(cdev)) {
 		err = PTR_ERR(cdev);
 		dev_err(df->dev.parent,
@@ -522,6 +528,8 @@ of_devfreq_cooling_register_power(struct device_node *np, struct devfreq *df,
 	kfree(dfc->freq_table);
 free_dfc:
 	kfree(dfc);
+free_ops:
+	kfree(ops);
 
 	return ERR_PTR(err);
 }
@@ -557,10 +565,12 @@ EXPORT_SYMBOL_GPL(devfreq_cooling_register);
 void devfreq_cooling_unregister(struct thermal_cooling_device *cdev)
 {
 	struct devfreq_cooling_device *dfc;
+	const struct thermal_cooling_device_ops *ops;
 
 	if (!cdev)
 		return;
 
+	ops = cdev->ops;
 	dfc = cdev->devdata;
 
 	thermal_cooling_device_unregister(dfc->cdev);
@@ -570,5 +580,6 @@ void devfreq_cooling_unregister(struct thermal_cooling_device *cdev)
 	kfree(dfc->freq_table);
 
 	kfree(dfc);
+	kfree(ops);
 }
 EXPORT_SYMBOL_GPL(devfreq_cooling_unregister);
Signature · 1 line
2.29.0
reviewer
#1 lore raw
Quoted 29 lines · Kant Fan

commit 7b62935828266658714f81d4e9176edad808dc70 upstream.

Fix access illegal address problem in following condition:
There are muti devfreq cooling devices in system, some of them register with dfc_power but other does not, power model ops such as state2power will append to global devfreq_cooling_ops when the cooling device with dfc_power register. It makes the cooling device without dfc_power also use devfreq_cooling_ops after appending when register later by of_devfreq_cooling_register_power() or of_devfreq_cooling_register().

IPA governor regards the cooling devices without dfc_power as a power actor because they also have power model ops, and will access illegal address at dfc->power_ops when execute cdev->ops->get_requested_power or cdev->ops->power2state. As the calltrace below shows:

Unable to handle kernel NULL pointer dereference at virtual address
00000008
...
calltrace:
[<c06e5488>] devfreq_cooling_power2state+0x24/0x184
[<c06df420>] power_actor_set_power+0x54/0xa8
[<c06e3774>] power_allocator_throttle+0x770/0x97c
[<c06dd120>] handle_thermal_trip+0x1b4/0x26c
[<c06ddb48>] thermal_zone_device_update+0x154/0x208
[<c014159c>] process_one_work+0x1ec/0x36c
[<c0141c58>] worker_thread+0x204/0x2ec
[<c0146788>] kthread+0x140/0x154
[<c01010e8>] ret_from_fork+0x14/0x2c
Quoted 7 lines
Fixes a76caf55e5b35 ("thermal: Add devfreq cooling")
---
  drivers/thermal/devfreq_cooling.c | 25 ++++++++++++++++++-------
  1 file changed, 18 insertions(+), 7 deletions(-)

Looks good. So this patch should be applied for all stable kernels starting from v4.4 to v5.12 (the v5.13 and later need other patch).

Next time you might use in the subject something like:
[PATCH 4.4] thermal: devfreq_cooling: use local ops instead of global ops It would be better distinguished from your other patch with the same subject, which was for mainline and v5.13+

author
#2 lore raw
Quoted 40 lines · Lukasz Luba
Quoted 29 lines · Kant Fan

commit 7b62935828266658714f81d4e9176edad808dc70 upstream.

Fix access illegal address problem in following condition:
There are muti devfreq cooling devices in system, some of them register with dfc_power but other does not, power model ops such as state2power will append to global devfreq_cooling_ops when the cooling device with dfc_power register. It makes the cooling device without dfc_power also use devfreq_cooling_ops after appending when register later by of_devfreq_cooling_register_power() or of_devfreq_cooling_register().

IPA governor regards the cooling devices without dfc_power as a power actor because they also have power model ops, and will access illegal address at dfc->power_ops when execute cdev->ops->get_requested_power or cdev->ops->power2state. As the calltrace below shows:

Unable to handle kernel NULL pointer dereference at virtual address
00000008
...
calltrace:
[<c06e5488>] devfreq_cooling_power2state+0x24/0x184
[<c06df420>] power_actor_set_power+0x54/0xa8
[<c06e3774>] power_allocator_throttle+0x770/0x97c
[<c06dd120>] handle_thermal_trip+0x1b4/0x26c
[<c06ddb48>] thermal_zone_device_update+0x154/0x208
[<c014159c>] process_one_work+0x1ec/0x36c
[<c0141c58>] worker_thread+0x204/0x2ec
[<c0146788>] kthread+0x140/0x154
[<c01010e8>] ret_from_fork+0x14/0x2c
Quoted 7 lines
Fixes a76caf55e5b35 ("thermal: Add devfreq cooling")

---
  drivers/thermal/devfreq_cooling.c | 25 ++++++++++++++++++-------

  1 file changed, 18 insertions(+), 7 deletions(-)
Quoted 8 lines

Looks good. So this patch should be applied for all stable kernels starting from v4.4 to v5.12 (the v5.13 and later need other patch).

Next time you might use in the subject something like:
[PATCH 4.4] thermal: devfreq_cooling: use local ops instead of global ops It would be better distinguished from your other patch with the same subject, which was for mainline and v5.13+

Hi Lukasz,
Thank you for the guidance. I want to know if I'm understanding you in a right way. Could you confirm the following information?

1. The stable patches
After the patch is merged into mainline later, I'll submit the following patches individually for v4.4 ~ v5.12:

[PATCH 4.4] thermal: devfreq_cooling: use local ops instead of global ops [PATCH 4.5] thermal: devfreq_cooling: use local ops instead of global ops
...
[PATCH 5.12] thermal: devfreq_cooling: use local ops instead of global ops

And also the following patches individually for v5.13+ :
[PATCH 5.13] thermal: devfreq_cooling: use local ops instead of global ops [PATCH 5.14] thermal: devfreq_cooling: use local ops instead of global ops
...
[PATCH 5.17] thermal: devfreq_cooling: use local ops instead of global ops

2. The mainline patch
I saw your mail with Rafael, seems there are conflicts... I wonder if there's anything wrong with my patch, or anything I can help?

Signature · 2 lines
Best Regards,
Kant Fan
reviewer
#3 lore raw

Hi Kant,

Quoted 50 lines · Kant Fan
Quoted 40 lines · Lukasz Luba
Quoted 29 lines · Kant Fan

commit 7b62935828266658714f81d4e9176edad808dc70 upstream.

Fix access illegal address problem in following condition:
There are muti devfreq cooling devices in system, some of them register with dfc_power but other does not, power model ops such as state2power will append to global devfreq_cooling_ops when the cooling device with dfc_power register. It makes the cooling device without dfc_power also use devfreq_cooling_ops after appending when register later by of_devfreq_cooling_register_power() or of_devfreq_cooling_register().

IPA governor regards the cooling devices without dfc_power as a power actor because they also have power model ops, and will access illegal address at dfc->power_ops when execute cdev->ops->get_requested_power or cdev->ops->power2state. As the calltrace below shows:

Unable to handle kernel NULL pointer dereference at virtual address
00000008
...
calltrace:
[<c06e5488>] devfreq_cooling_power2state+0x24/0x184
[<c06df420>] power_actor_set_power+0x54/0xa8
[<c06e3774>] power_allocator_throttle+0x770/0x97c
[<c06dd120>] handle_thermal_trip+0x1b4/0x26c
[<c06ddb48>] thermal_zone_device_update+0x154/0x208
[<c014159c>] process_one_work+0x1ec/0x36c
[<c0141c58>] worker_thread+0x204/0x2ec
[<c0146788>] kthread+0x140/0x154
[<c01010e8>] ret_from_fork+0x14/0x2c
Quoted 7 lines
Fixes a76caf55e5b35 ("thermal: Add devfreq cooling")

---
  drivers/thermal/devfreq_cooling.c | 25 ++++++++++++++++++-------

  1 file changed, 18 insertions(+), 7 deletions(-)
Quoted 8 lines

Looks good. So this patch should be applied for all stable kernels starting from v4.4 to v5.12 (the v5.13 and later need other patch).

Next time you might use in the subject something like:
[PATCH 4.4] thermal: devfreq_cooling: use local ops instead of global ops It would be better distinguished from your other patch with the same subject, which was for mainline and v5.13+

Quoted 5 lines

Hi Lukasz,
Thank you for the guidance. I want to know if I'm understanding you in a right way. Could you confirm the following information?

1. The stable patches
After the patch is merged into mainline later, I'll submit the following patches individually for v4.4 ~ v5.12:

Correct, after it gets mainline you can point to that commit hash and process with those patches. I don't now which of those older stable kernels are still maintained, since some of them have longer support and the rest had shorter and might already ended. You can check the end of life for those 'Longterm' here [1]. AFAICS the 4.4 is not in that table, so you can start from 4.9, should be OK.
So the list of needed patches would be for those stable kernels:
4.9, 4.14, 4.19, 5.4, 5.10
I can see that last release for 5.11.x was in May 2021, so it's probably ended, similar for 5.12.x (Jul 2021). That's why I suggested that list for the long support kernels.

Quoted 7 lines

[PATCH 4.4] thermal: devfreq_cooling: use local ops instead of global ops [PATCH 4.5] thermal: devfreq_cooling: use local ops instead of global ops
...
[PATCH 5.12] thermal: devfreq_cooling: use local ops instead of global ops

And also the following patches individually for v5.13+ :

For this, you probably don't have to. You have added 'v5.13+' in the original patch v2, so it will be picked correctly. It should apply on those stable kernels w/o issues. If there will be, stable kernel engineers will ping us.

Quoted 8 lines

[PATCH 5.13] thermal: devfreq_cooling: use local ops instead of global ops [PATCH 5.14] thermal: devfreq_cooling: use local ops instead of global ops
...
[PATCH 5.17] thermal: devfreq_cooling: use local ops instead of global ops

2. The mainline patch
I saw your mail with Rafael, seems there are conflicts... I wonder if there's anything wrong with my patch, or anything I can help?

Thank you for offering help. Rafael solved that correctly, so it doesn't need any more work.

Thank you for doing that work!

Regards,
Lukasz

[1] https://www.kernel.org/category/releases.html

author
#4 lore raw
Quoted 59 lines · Lukasz Luba

Hi Kant,

Quoted 50 lines · Kant Fan
Quoted 40 lines · Lukasz Luba
Quoted 29 lines · Kant Fan

commit 7b62935828266658714f81d4e9176edad808dc70 upstream.

Fix access illegal address problem in following condition:
There are muti devfreq cooling devices in system, some of them register with dfc_power but other does not, power model ops such as state2power will append to global devfreq_cooling_ops when the cooling device with dfc_power register. It makes the cooling device without dfc_power also use devfreq_cooling_ops after appending when register later by of_devfreq_cooling_register_power() or of_devfreq_cooling_register().

IPA governor regards the cooling devices without dfc_power as a power actor because they also have power model ops, and will access illegal address at dfc->power_ops when execute cdev->ops->get_requested_power or cdev->ops->power2state. As the calltrace below shows:

Unable to handle kernel NULL pointer dereference at virtual address
00000008
...
calltrace:
[<c06e5488>] devfreq_cooling_power2state+0x24/0x184
[<c06df420>] power_actor_set_power+0x54/0xa8
[<c06e3774>] power_allocator_throttle+0x770/0x97c
[<c06dd120>] handle_thermal_trip+0x1b4/0x26c
[<c06ddb48>] thermal_zone_device_update+0x154/0x208
[<c014159c>] process_one_work+0x1ec/0x36c
[<c0141c58>] worker_thread+0x204/0x2ec
[<c0146788>] kthread+0x140/0x154
[<c01010e8>] ret_from_fork+0x14/0x2c
Quoted 7 lines
Fixes a76caf55e5b35 ("thermal: Add devfreq cooling")

---
  drivers/thermal/devfreq_cooling.c | 25 ++++++++++++++++++-------

  1 file changed, 18 insertions(+), 7 deletions(-)
Quoted 8 lines

Looks good. So this patch should be applied for all stable kernels starting from v4.4 to v5.12 (the v5.13 and later need other patch).

Next time you might use in the subject something like:
[PATCH 4.4] thermal: devfreq_cooling: use local ops instead of global ops It would be better distinguished from your other patch with the same subject, which was for mainline and v5.13+

Quoted 5 lines

Hi Lukasz,
Thank you for the guidance. I want to know if I'm understanding you in a right way. Could you confirm the following information?

1. The stable patches
After the patch is merged into mainline later, I'll submit the following patches individually for v4.4 ~ v5.12:

Quoted 12 lines

Correct, after it gets mainline you can point to that commit hash and process with those patches. I don't now which of those older stable kernels are still maintained, since some of them have longer support and the rest had shorter and might already ended. You can check the end of life for those 'Longterm' here [1]. AFAICS the 4.4 is not in that table, so you can start from 4.9, should be OK.
So the list of needed patches would be for those stable kernels:
4.9, 4.14, 4.19, 5.4, 5.10
I can see that last release for 5.11.x was in May 2021, so it's probably ended, similar for 5.12.x (Jul 2021). That's why I suggested that list for the long support kernels.

Hi Lukasz,
Thanks for figuring it out. I'll check the stable versions carefully.

Quoted 25 lines
Quoted 7 lines

[PATCH 4.4] thermal: devfreq_cooling: use local ops instead of global ops [PATCH 4.5] thermal: devfreq_cooling: use local ops instead of global ops
...
[PATCH 5.12] thermal: devfreq_cooling: use local ops instead of global ops

And also the following patches individually for v5.13+ :

For this, you probably don't have to. You have added 'v5.13+' in the original patch v2, so it will be picked correctly. It should apply on those stable kernels w/o issues. If there will be, stable kernel engineers will ping us.

Quoted 8 lines

[PATCH 5.13] thermal: devfreq_cooling: use local ops instead of global ops [PATCH 5.14] thermal: devfreq_cooling: use local ops instead of global ops
...
[PATCH 5.17] thermal: devfreq_cooling: use local ops instead of global ops

2. The mainline patch
I saw your mail with Rafael, seems there are conflicts... I wonder if there's anything wrong with my patch, or anything I can help?

Thank you for offering help. Rafael solved that correctly, so it doesn't need any more work.

Quoted 6 lines

Thank you for doing that work!

Regards,
Lukasz

[1] https://www.kernel.org/category/releases.html

No problem. I'll submit the stable patches after the mainline patch is merged.

Signature · 2 lines
Best Regards,
Kant Fan
maintainer
#5 lore raw
In reply to Kant Fan
Quoted 1 line · Kant Fan

commit 7b62935828266658714f81d4e9176edad808dc70 upstream.

I do not see this commit in Linus's tree :(

confused,

greg k-h

reviewer
#6 lore raw
Quoted 8 lines · Greg KH
Quoted 1 line · Kant Fan

commit 7b62935828266658714f81d4e9176edad808dc70 upstream.

I do not see this commit in Linus's tree :(

confused,

greg k-h

My apologies Greg, I missed that.

We need to first get the patch [1] into upstream.
Kant would resend this patch after that happen.

[1] https://lore.kernel.org/linux-pm/[email protected]/