Skip to contentSkip to conversation

acpi: apei: check for pending errors when probing HED type GHES entries

PATCH Superseded Tyler Baicar 2 participants 3 messages Mar 29, 2017 – Mar 31, 2017

Subsystems (2)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 acpi: apei: check for pending errors when probing HED type GHES entries

If a HED type error occurs prior to GHES probing, the kernel will never report the error. The HED driver will see that no notifiers are registered, and clear the interrupt.

This becomes a more serious problem with firmware that supports GHESv2 acknowledgements from the kernel. The firmware will populate the error and wait for the kernel ack. But since the kernel will never process the error we get into a state that the firmware will not send any more errors and the kernel will never see or ack the original error.

Check for pending errors when probing HED type GHES entries to avoid the above situation.

This patch is based on Shiju's patch that adds support for GSIV and GPIO notification types:
https://patchwork.kernel.org/patch/9628817/

---
 drivers/acpi/apei/ghes.c | 1 +
 1 file changed, 1 insertion(+)
drivers/acpi/apei/ghes.c +1 -0
index fd39929..cf5e938 100644
@@ -1035,6 +1035,7 @@ static int ghes_probe(struct platform_device *ghes_dev)
 			register_acpi_hed_notifier(&ghes_notifier_hed);
 		list_add_rcu(&ghes->list, &ghes_hed);
 		mutex_unlock(&ghes_list_mutex);
+		ghes_proc(ghes);
 		break;
 	case ACPI_HEST_NOTIFY_NMI:
 		ghes_nmi_add(ghes);
Signature · 3 lines
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project.
#1 lore raw

Hi Tyler,

Quoted 11 lines · Tyler Baicar

If a HED type error occurs prior to GHES probing, the kernel will never report the error. The HED driver will see that no notifiers are registered, and clear the interrupt.

This becomes a more serious problem with firmware that supports GHESv2 acknowledgements from the kernel. The firmware will populate the error and wait for the kernel ack. But since the kernel will never process the error we get into a state that the firmware will not send any more errors and the kernel will never see or ack the original error.

Quoted 2 lines

Check for pending errors when probing HED type GHES entries to avoid the above situation.

Isn't this a problem for the other notification types too?

It looks like SEI can indicate the notification is non-fatal even if we haven't done the ghes_probe() yet and fail to find the CPER records.

Would moving the OSC call to set the APEI bit later solve this, or is it specific to the way AMLs Notify() works?

Thanks,

James

Quoted 23 lines

This patch is based on Shiju's patch that adds support for GSIV and GPIO notification types:
https://patchwork.kernel.org/patch/9628817/

---
 drivers/acpi/apei/ghes.c | 1 +
 1 file changed, 1 insertion(+)
drivers/acpi/apei/ghes.c +1 -0
index fd39929..cf5e938 100644
@@ -1035,6 +1035,7 @@ static int ghes_probe(struct platform_device *ghes_dev)
 			register_acpi_hed_notifier(&ghes_notifier_hed);
 		list_add_rcu(&ghes->list, &ghes_hed);
 		mutex_unlock(&ghes_list_mutex);
+		ghes_proc(ghes);
 		break;
 	case ACPI_HEST_NOTIFY_NMI:
 		ghes_nmi_add(ghes);
#2 lore raw

Hello James,

Quoted 14 lines · James Morse
Quoted 11 lines · Tyler Baicar

If a HED type error occurs prior to GHES probing, the kernel will never report the error. The HED driver will see that no notifiers are registered, and clear the interrupt.

This becomes a more serious problem with firmware that supports GHESv2 acknowledgements from the kernel. The firmware will populate the error and wait for the kernel ack. But since the kernel will never process the error we get into a state that the firmware will not send any more errors and the kernel will never see or ack the original error.

Quoted 2 lines

Check for pending errors when probing HED type GHES entries to avoid the above situation.

Quoted 7 lines

Isn't this a problem for the other notification types too?

It looks like SEI can indicate the notification is non-fatal even if we haven't done the ghes_probe() yet and fail to find the CPER records.

Would moving the OSC call to set the APEI bit later solve this, or is it specific to the way AMLs Notify() works?

For SEAs, the kernel would have already tried handling the SEA and gone through the GHES code unsuccessfully. If the SEA was non-fatal then we could print the GHES/CPER info here.

SEIs are probably similar to the SEA case I would imagine.

The notification interface I believe just sends the notification to anyone registered for that event and then clears the interrupt, so if no one is registered, the interrupt will be cleared.

It may be more practical to add this call to ghes_proc() at the end of the switch statement so that all notification types check for pending errors right after probing. Adding this call shouldn't be an issue since if there is no error pending it will just return back. The polled notification type calls ghes_proc() every time it's timer expires to check for errors.

Thanks,
Tyler

Quoted 22 lines
Quoted 22 lines

This patch is based on Shiju's patch that adds support for GSIV and GPIO notification types:
https://patchwork.kernel.org/patch/9628817/

---
  drivers/acpi/apei/ghes.c | 1 +
  1 file changed, 1 insertion(+)
drivers/acpi/apei/ghes.c +1 -0
index fd39929..cf5e938 100644
@@ -1035,6 +1035,7 @@ static int ghes_probe(struct platform_device *ghes_dev)
  			register_acpi_hed_notifier(&ghes_notifier_hed);
  		list_add_rcu(&ghes->list, &ghes_hed);
  		mutex_unlock(&ghes_list_mutex);
+		ghes_proc(ghes);
  		break;
  	case ACPI_HEST_NOTIFY_NMI:
  		ghes_nmi_add(ghes);
Signature · 3 lines
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project.