drm/panthor: Be robust against resume failures

When the runtime PM resume callback returns an error, it puts the device
in a state where it can't be resumed anymore. Make sure we can recover
from such transient failures by calling pm_runtime_set_suspended()
explicitly after a pm_runtime_resume_and_get() failure.

v3:
- Add R-b/A-b

v2:
- Add a comment explaining potential races in
  panthor_device_resume_and_get()

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Adrian Larumbe <adrian.larumbe@collabora.com>
Acked-by: Steven Price <steven.price@arm.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20241211075419.2333731-5-boris.brezillon@collabora.com

4 files changed, 30 insertions, 3 deletions

diff --git a/drivers/gpu/drm/panthor/panthor_device.c b/drivers/gpu/drm/panthor/panthor_device.c
index 4faca2684324..5f9602bddc53 100644
--- a/drivers/gpu/drm/panthor/panthor_device.c
+++ b/drivers/gpu/drm/panthor/panthor_device.c
@@ -504,6 +504,7 @@ err_disable_core_clk:
 
 err_set_suspended:
 	atomic_set(&ptdev->pm.state, PANTHOR_DEVICE_PM_STATE_SUSPENDED);
+	atomic_set(&ptdev->pm.recovery_needed, 1);
 	return ret;
 }
 
diff --git a/drivers/gpu/drm/panthor/panthor_device.h b/drivers/gpu/drm/panthor/panthor_device.h
index 0e68f5a70d20..b6c4f25a5d6e 100644
--- a/drivers/gpu/drm/panthor/panthor_device.h
+++ b/drivers/gpu/drm/panthor/panthor_device.h
@@ -9,6 +9,7 @@
 #include <linux/atomic.h>
 #include <linux/io-pgtable.h>
 #include <linux/regulator/consumer.h>
+#include <linux/pm_runtime.h>
 #include <linux/sched.h>
 #include <linux/spinlock.h>
 
@@ -180,6 +181,9 @@ struct panthor_device {
 		 * is suspended.
 		 */
 		struct page *dummy_latest_flush;
+
+		/** @recovery_needed: True when a resume attempt failed. */
+		atomic_t recovery_needed;
 	} pm;
 
 	/** @profile_mask: User-set profiling flags for job accounting. */
@@ -243,6 +247,28 @@ int panthor_device_mmap_io(struct panthor_device *ptdev,
 int panthor_device_resume(struct device *dev);
 int panthor_device_suspend(struct device *dev);
 
+static inline int panthor_device_resume_and_get(struct panthor_device *ptdev)
+{
+	int ret = pm_runtime_resume_and_get(ptdev->base.dev);
+
+	/* If the resume failed, we need to clear the runtime_error, which
+	 * can done by forcing the RPM state to suspended. If multiple
+	 * threads called panthor_device_resume_and_get(), we only want
+	 * one of them to update the state, hence the cmpxchg. Note that a
+	 * thread might enter panthor_device_resume_and_get() and call
+	 * pm_runtime_resume_and_get() after another thread had attempted
+	 * to resume and failed. This means we will end up with an error
+	 * without even attempting a resume ourselves. The only risk here
+	 * is to report an error when the second resume attempt might have
+	 * succeeded. Given resume errors are not expected, this is probably
+	 * something we can live with.
+	 */
+	if (ret && atomic_cmpxchg(&ptdev->pm.recovery_needed, 1, 0) == 1)
+		pm_runtime_set_suspended(ptdev->base.dev);
+
+	return ret;
+}
+
 enum drm_panthor_exception_type {
 	DRM_PANTHOR_EXCEPTION_OK = 0x00,
 	DRM_PANTHOR_EXCEPTION_TERMINATED = 0x04,
diff --git a/drivers/gpu/drm/panthor/panthor_drv.c b/drivers/gpu/drm/panthor/panthor_drv.c
index 32421bad3097..33cfba877525 100644
--- a/drivers/gpu/drm/panthor/panthor_drv.c
+++ b/drivers/gpu/drm/panthor/panthor_drv.c
@@ -763,7 +763,7 @@ static int panthor_query_timestamp_info(struct panthor_device *ptdev,
 {
 	int ret;
 
-	ret = pm_runtime_resume_and_get(ptdev->base.dev);
+	ret = panthor_device_resume_and_get(ptdev);
 	if (ret)
 		return ret;
 
diff --git a/drivers/gpu/drm/panthor/panthor_sched.c b/drivers/gpu/drm/panthor/panthor_sched.c
index ef4bec7ff9c7..58200d299fbb 100644
--- a/drivers/gpu/drm/panthor/panthor_sched.c
+++ b/drivers/gpu/drm/panthor/panthor_sched.c
@@ -2354,7 +2354,7 @@ static void tick_work(struct work_struct *work)
 	if (!drm_dev_enter(&ptdev->base, &cookie))
 		return;
 
-	ret = pm_runtime_resume_and_get(ptdev->base.dev);
+	ret = panthor_device_resume_and_get(ptdev);
 	if (drm_WARN_ON(&ptdev->base, ret))
 		goto out_dev_exit;
 
@@ -3115,7 +3115,7 @@ queue_run_job(struct drm_sched_job *sched_job)
 		return dma_fence_get(job->done_fence);
 	}
 
-	ret = pm_runtime_resume_and_get(ptdev->base.dev);
+	ret = panthor_device_resume_and_get(ptdev);
 	if (drm_WARN_ON(&ptdev->base, ret))
 		return ERR_PTR(ret);