Receiving an MMU fault currently disables the AS, so each of the groups
is marked with the appropriate faults. Since no further submissions
can occur on a fatal fault for that group, the fault information
does not have to be cleared until the group is terminated.

Signed-off-by: Lukas Zapolskas <lukas.zapolskas@arm.com>
---
 drivers/gpu/drm/panthor/panthor_mmu.c   |  8 ++++++++
 drivers/gpu/drm/panthor/panthor_mmu.h   |  2 ++
 drivers/gpu/drm/panthor/panthor_sched.c | 13 +++++++++++++
 3 files changed, 23 insertions(+)

diff --git a/drivers/gpu/drm/panthor/panthor_mmu.c b/drivers/gpu/drm/panthor/panthor_mmu.c
index 10a7418eecda..9e78b0509f1a 100644
--- a/drivers/gpu/drm/panthor/panthor_mmu.c
+++ b/drivers/gpu/drm/panthor/panthor_mmu.c
@@ -2895,3 +2895,11 @@ void panthor_mmu_pt_cache_fini(void)
 {
 	kmem_cache_destroy(pt_cache);
 }
+
+struct panthor_vm_fault *panthor_vm_get_fault(struct panthor_vm *vm)
+{
+	if (!vm)
+		return NULL;
+
+	return &vm->fault;
+}
diff --git a/drivers/gpu/drm/panthor/panthor_mmu.h b/drivers/gpu/drm/panthor/panthor_mmu.h
index 023fdc79c231..d69b4000a39e 100644
--- a/drivers/gpu/drm/panthor/panthor_mmu.h
+++ b/drivers/gpu/drm/panthor/panthor_mmu.h
@@ -123,4 +123,6 @@ void panthor_mmu_pt_cache_fini(void);
 void panthor_mmu_debugfs_init(struct drm_minor *minor);
 #endif
 
+struct panthor_vm_fault *panthor_vm_get_fault(struct panthor_vm *vm);
+
 #endif
diff --git a/drivers/gpu/drm/panthor/panthor_sched.c b/drivers/gpu/drm/panthor/panthor_sched.c
index a77399e95620..9ea0d2b27114 100644
--- a/drivers/gpu/drm/panthor/panthor_sched.c
+++ b/drivers/gpu/drm/panthor/panthor_sched.c
@@ -722,6 +722,11 @@ struct panthor_group {
 	 * panthor_group::groups::waiting list.
 	 */
 	struct list_head wait_node;
+
+	/**
+	 * @fatal: VM-level fault that caused a fatal error on the group.
+	 */
+	struct panthor_vm_fault fatal;
 };
 
 struct panthor_job_profiling_data {
@@ -1575,6 +1580,14 @@ cs_slot_process_fatal_event_locked(struct panthor_device *ptdev,
 			 group->task_info.pid, group->task_info.comm);
 
 		group->fatal_queues |= BIT(cs_id);
+
+		if (panthor_vm_has_unhandled_faults(group->vm)) {
+			struct panthor_vm_fault *fault;
+
+			fault = panthor_vm_get_fault(group->vm);
+			if (fault)
+				group->fatal = *fault;
+		}
 	}
 
 	if (CS_EXCEPTION_TYPE(fatal) == DRM_PANTHOR_EXCEPTION_CS_UNRECOVERABLE) {
-- 
2.33.0.dirty

On Mon, 15 Dec 2025 11:54:56 +0000
Lukas Zapolskas <lukas.zapolskas@arm.com> wrote:

>  			 group->task_info.pid, group->task_info.comm);
>  
>  		group->fatal_queues |= BIT(cs_id);
> +
> +		if (panthor_vm_has_unhandled_faults(group->vm)) {
> +			struct panthor_vm_fault *fault;
> +
> +			fault = panthor_vm_get_fault(group->vm);
> +			if (fault)
> +				group->fatal = *fault;

group->vm can't be NULL, meaning fault can't be NULL either.

> +		}
>  	}

On Mon, 15 Dec 2025 11:54:56 +0000
Lukas Zapolskas <lukas.zapolskas@arm.com> wrote:

> Receiving an MMU fault currently disables the AS, so each of the groups
> is marked with the appropriate faults. Since no further submissions
> can occur on a fatal fault for that group, the fault information
> does not have to be cleared until the group is terminated.
> 
> Signed-off-by: Lukas Zapolskas <lukas.zapolskas@arm.com>
> ---
>  drivers/gpu/drm/panthor/panthor_mmu.c   |  8 ++++++++
>  drivers/gpu/drm/panthor/panthor_mmu.h   |  2 ++
>  drivers/gpu/drm/panthor/panthor_sched.c | 13 +++++++++++++
>  3 files changed, 23 insertions(+)
> 
> diff --git a/drivers/gpu/drm/panthor/panthor_mmu.c b/drivers/gpu/drm/panthor/panthor_mmu.c
> index 10a7418eecda..9e78b0509f1a 100644
> --- a/drivers/gpu/drm/panthor/panthor_mmu.c
> +++ b/drivers/gpu/drm/panthor/panthor_mmu.c
> @@ -2895,3 +2895,11 @@ void panthor_mmu_pt_cache_fini(void)
>  {
>  	kmem_cache_destroy(pt_cache);
>  }
> +
> +struct panthor_vm_fault *panthor_vm_get_fault(struct panthor_vm *vm)

const struct panthor_vm_fault *
panthor_vm_get_fault(struct panthor_vm *vm)

or

struct panthor_vm_fault
panthor_vm_get_fault(struct panthor_vm *vm)

or

void
panthor_vm_get_fault(struct panthor_vm *vm,
		     struct panthor_vm_fault *fault)

but you shouldn't let the caller with a writable pointer to vm->fault.

> +{
> +	if (!vm)
> +		return NULL;

I don't see a valid case where panthor_vm_get_fault() would be called
with a NULL pointer.

> +
> +	return &vm->fault;
> +}
> diff --git a/drivers/gpu/drm/panthor/panthor_mmu.h b/drivers/gpu/drm/panthor/panthor_mmu.h
> index 023fdc79c231..d69b4000a39e 100644
> --- a/drivers/gpu/drm/panthor/panthor_mmu.h
> +++ b/drivers/gpu/drm/panthor/panthor_mmu.h
> @@ -123,4 +123,6 @@ void panthor_mmu_pt_cache_fini(void);
>  void panthor_mmu_debugfs_init(struct drm_minor *minor);
>  #endif
>  
> +struct panthor_vm_fault *panthor_vm_get_fault(struct panthor_vm *vm);
> +
>  #endif
> diff --git a/drivers/gpu/drm/panthor/panthor_sched.c b/drivers/gpu/drm/panthor/panthor_sched.c
> index a77399e95620..9ea0d2b27114 100644
> --- a/drivers/gpu/drm/panthor/panthor_sched.c
> +++ b/drivers/gpu/drm/panthor/panthor_sched.c
> @@ -722,6 +722,11 @@ struct panthor_group {
>  	 * panthor_group::groups::waiting list.
>  	 */
>  	struct list_head wait_node;
> +
> +	/**
> +	 * @fatal: VM-level fault that caused a fatal error on the group.
> +	 */
> +	struct panthor_vm_fault fatal;
>  };
>  
>  struct panthor_job_profiling_data {
> @@ -1575,6 +1580,14 @@ cs_slot_process_fatal_event_locked(struct panthor_device *ptdev,
>  			 group->task_info.pid, group->task_info.comm);
>  
>  		group->fatal_queues |= BIT(cs_id);
> +
> +		if (panthor_vm_has_unhandled_faults(group->vm)) {
> +			struct panthor_vm_fault *fault;
> +
> +			fault = panthor_vm_get_fault(group->vm);
> +			if (fault)
> +				group->fatal = *fault;
> +		}
>  	}
>  
>  	if (CS_EXCEPTION_TYPE(fatal) == DRM_PANTHOR_EXCEPTION_CS_UNRECOVERABLE) {