Skip to contentSkip to conversation

panthor: print task pid and comm on gpu errors

PATCH Applied Chia-I Wu 2 participants 5 messages Jul 18, 2025 – Jul 24, 2025

Subsystems (3)
DRM drivers DRM misc Arm Mali panthor DRM driver
lore
Share & tools
lore

Chia-I Wu

author
#0 lore raw

v3 cover panthor: print task pid and comm on gpu errors

This series saves task pid and comm in panthor_group and prints task pid and comm on gpu errors.

v3: fix new kerneldoc warnings
v2: save the task info in panthor_group on panthor_group_create, rather than
in panthor_file on panthor_open, because, when the two differ, we are more interested in the task that created the group.

Chia-I Wu (3):
panthor: set owner field for driver fops
panthor: save task pid and comm in panthor_group
panthor: dump task pid and comm on gpu errors

 drivers/gpu/drm/panthor/panthor_drv.c   | 14 ++------
 drivers/gpu/drm/panthor/panthor_sched.c | 43 ++++++++++++++++++++++---
 2 files changed, 41 insertions(+), 16 deletions(-)
Signature · 1 line
2.50.0.727.gbf7dc18ff4-goog
author
#1 lore raw

PATCH v3 1/3 panthor: set owner field for driver fops

It allows us to get rid of manual try_module_get / module_put.

Signed-off-by Chia-I Wu <[email protected]>
Reviewed-by Boris Brezillon <[email protected]>
Reviewed-by Steven Price <[email protected]>
---
 drivers/gpu/drm/panthor/panthor_drv.c | 14 +++-----------
 1 file changed, 3 insertions(+), 11 deletions(-)
drivers/gpu/drm/panthor/panthor_drv.c +3 -11
index 1116f2d2826ee..775a66c394544 100644
@@ -1400,14 +1400,9 @@ panthor_open(struct drm_device *ddev, struct drm_file *file)
 	struct panthor_file *pfile;
 	int ret;
 
-	if (!try_module_get(THIS_MODULE))
-		return -EINVAL;
-
 	pfile = kzalloc(sizeof(*pfile), GFP_KERNEL);
-	if (!pfile) {
-		ret = -ENOMEM;
-		goto err_put_mod;
-	}
+	if (!pfile)
+		return -ENOMEM;
 
 	pfile->ptdev = ptdev;
 	pfile->user_mmio.offset = DRM_PANTHOR_USER_MMIO_OFFSET;

@@ -1439,9 +1434,6 @@ panthor_open(struct drm_device *ddev, struct drm_file *file)
 
 err_free_file:
 	kfree(pfile);
-
-err_put_mod:
-	module_put(THIS_MODULE);
 	return ret;
 }
 

@@ -1454,7 +1446,6 @@ panthor_postclose(struct drm_device *ddev, struct drm_file *file)
 	panthor_vm_pool_destroy(pfile);
 
 	kfree(pfile);
-	module_put(THIS_MODULE);
 }
 
 static const struct drm_ioctl_desc panthor_drm_driver_ioctls[] = {

@@ -1555,6 +1546,7 @@ static void panthor_show_fdinfo(struct drm_printer *p, struct drm_file *file)
 }
 
 static const struct file_operations panthor_drm_driver_fops = {
+	.owner = THIS_MODULE,
 	.open = drm_open,
 	.release = drm_release,
 	.unlocked_ioctl = drm_ioctl,
Signature · 1 line
2.50.0.727.gbf7dc18ff4-goog
author
#2 lore raw

PATCH v3 2/3 panthor: save task pid and comm in panthor_group

In reply to Chia-I Wu

We would like to report them on gpu errors.

We choose to save the info on panthor_group_create rather than on panthor_open because, when the two differ, we are more interested in the task that created the group.

Signed-off-by Chia-I Wu <[email protected]>
Reviewed-by Steven Price <[email protected]>

---
v3: fix new kerneldoc warnings
v2: save to panthor_group on panthor_group_create rather than to
panthor_file on panthor_open

---
 drivers/gpu/drm/panthor/panthor_sched.c | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)
drivers/gpu/drm/panthor/panthor_sched.c +19 -0
index a2248f692a030..9987aeb4608bc 100644
@@ -641,6 +641,15 @@ struct panthor_group {
 		size_t kbo_sizes;
 	} fdinfo;
 
+	/** @task_info: Info of current->group_leader that created the group. */
+	struct {
+		/** @task_info.pid: pid of current->group_leader */
+		pid_t pid;
+
+		/** @task_info.comm: comm of current->group_leader */
+		char comm[TASK_COMM_LEN];
+	} task_info;
+
 	/** @state: Group state. */
 	enum panthor_group_state state;
 

@@ -3389,6 +3398,14 @@ group_create_queue(struct panthor_group *group,
 	return ERR_PTR(ret);
 }
 
+static void group_init_task_info(struct panthor_group *group)
+{
+	struct task_struct *task = current->group_leader;
+
+	group->task_info.pid = task->pid;
+	get_task_comm(group->task_info.comm, task);
+}
+
 static void add_group_kbo_sizes(struct panthor_device *ptdev,
 				struct panthor_group *group)
 {

@@ -3540,6 +3557,8 @@ int panthor_group_create(struct panthor_file *pfile,
 	add_group_kbo_sizes(group->ptdev, group);
 	spin_lock_init(&group->fdinfo.lock);
 
+	group_init_task_info(group);
+
 	return gid;
 
 err_put_group:
Signature · 1 line
2.50.0.727.gbf7dc18ff4-goog
author
#3 lore raw

PATCH v3 3/3 panthor: dump task pid and comm on gpu errors

In reply to Chia-I Wu

It is useful to know which tasks cause gpu errors.

Signed-off-by Chia-I Wu <[email protected]>
Reviewed-by Steven Price <[email protected]>
---
 drivers/gpu/drm/panthor/panthor_sched.c | 24 +++++++++++++++++++-----
 1 file changed, 19 insertions(+), 5 deletions(-)
drivers/gpu/drm/panthor/panthor_sched.c +19 -5
index 9987aeb4608bc..3a7472baa09ac 100644
@@ -1364,8 +1364,12 @@ cs_slot_process_fatal_event_locked(struct panthor_device *ptdev,
 	fatal = cs_iface->output->fatal;
 	info = cs_iface->output->fatal_info;
 
-	if (group)
+	if (group) {
+		drm_warn(&ptdev->base, "CS_FATAL: pid=%d, comm=%s\n",
+			 group->task_info.pid, group->task_info.comm);
+
 		group->fatal_queues |= BIT(cs_id);
+	}
 
 	if (CS_EXCEPTION_TYPE(fatal) == DRM_PANTHOR_EXCEPTION_CS_UNRECOVERABLE) {
 		/* If this exception is unrecoverable, queue a reset, and make

@@ -1425,6 +1429,11 @@ cs_slot_process_fault_event_locked(struct panthor_device *ptdev,
 		spin_unlock(&queue->fence_ctx.lock);
 	}
 
+	if (group) {
+		drm_warn(&ptdev->base, "CS_FAULT: pid=%d, comm=%s\n",
+			 group->task_info.pid, group->task_info.comm);
+	}
+
 	drm_warn(&ptdev->base,
 		 "CSG slot %d CS slot: %d\n"
 		 "CS_FAULT.EXCEPTION_TYPE: 0x%x (%s)\n"

@@ -1641,11 +1650,15 @@ csg_slot_process_progress_timer_event_locked(struct panthor_device *ptdev, u32 c
 
 	lockdep_assert_held(&sched->lock);
 
-	drm_warn(&ptdev->base, "CSG slot %d progress timeout\n", csg_id);
-
 	group = csg_slot->group;
-	if (!drm_WARN_ON(&ptdev->base, !group))
+	if (!drm_WARN_ON(&ptdev->base, !group)) {
+		drm_warn(&ptdev->base, "CSG_PROGRESS_TIMER_EVENT: pid=%d, comm=%s\n",
+			 group->task_info.pid, group->task_info.comm);
+
 		group->timedout = true;
+	}
+
+	drm_warn(&ptdev->base, "CSG slot %d progress timeout\n", csg_id);
 
 	sched_queue_delayed_work(sched, tick, 0);
 }

@@ -3227,7 +3240,8 @@ queue_timedout_job(struct drm_sched_job *sched_job)
 	struct panthor_scheduler *sched = ptdev->scheduler;
 	struct panthor_queue *queue = group->queues[job->queue_idx];
 
-	drm_warn(&ptdev->base, "job timeout\n");
+	drm_warn(&ptdev->base, "job timeout: pid=%d, comm=%s, seqno=%llu\n",
+		 group->task_info.pid, group->task_info.comm, job->done_fence->seqno);
 
 	drm_WARN_ON(&ptdev->base, atomic_read(&sched->reset.in_progress));
Signature · 1 line
2.50.0.727.gbf7dc18ff4-goog
maintainer Applied
#4 lore raw
In reply to Chia-I Wu
Quoted 8 lines · Chia-I Wu

This series saves task pid and comm in panthor_group and prints task pid and comm on gpu errors.

v3: fix new kerneldoc warnings
v2: save the task info in panthor_group on panthor_group_create, rather than
in panthor_file on panthor_open, because, when the two differ, we are more interested in the task that created the group.

Quoted 9 lines

Chia-I Wu (3):
panthor: set owner field for driver fops
panthor: save task pid and comm in panthor_group
panthor: dump task pid and comm on gpu errors

 drivers/gpu/drm/panthor/panthor_drv.c   | 14 ++------
 drivers/gpu/drm/panthor/panthor_sched.c | 43 ++++++++++++++++++++++---
 2 files changed, 41 insertions(+), 16 deletions(-)

Applied to drm-misc-next.

Thanks,
Steve