Skip to contentSkip to conversation

drm/amdgpu: Fix false error return to non-KCQ

PATCH In review Amber Lin 2 participants 2 messages Jun 26

Subsystems (2)
lore
Share & tools
lore

Amber Lin

author
#0 lore raw

PATCH 1/1 drm/amdgpu: Fix false error return to non-KCQ

amdgpu_gfx_reset_mes_compute is used to coordinate suspend_all, reset, and resume_all between KCQ and compute user queues. When a hung queue comes from the computer user queues and the reset is successful, the KCQ failure after reset should be sent to KCQ only and not the compute user queues. Compute user queues can operate after a successful reset without a mode reset.

Fixes 2dcf9a5a2352 ("drm/amdgpu/gfx: defer per-queue helper_end until after MES resume")
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c +11 -4
index 982b41606d48..52dd26dd82d2 100644
@@ -2282,6 +2282,7 @@ int amdgpu_gfx_reset_mes_compute(struct amdgpu_device *adev,
 	struct mes_remove_queue_input *queue_input = (struct mes_remove_queue_input *)faulty_queue_input;
 	struct amdgpu_gfx_deferred_entry deferred_end[AMDGPU_MAX_COMPUTE_RINGS + 1];
 	int n_deferred = 0;
+	int ring_err;
 
 	guard(mutex)(&adev->gfx.mec.reset_mutex);
 	/* stop the drm schedulers for all compute queues */
@@ -2375,17 +2376,23 @@ int amdgpu_gfx_reset_mes_compute(struct amdgpu_device *adev,
 	/* Now CP is running again — replay backed-up commands and ring
 	 * doorbells on each reset queue.
 	 */
+	ring_err = r;
 	for (i = 0; i < n_deferred; i++) {
 		int er = amdgpu_ring_reset_helper_end(deferred_end[i].ring,
 						      deferred_end[i].fence);
-		if (er && !r)
-			r = er;
+
+		if (er && !ring_err)
+			ring_err = er;
 	}
 
-	if (!r)
+	if (!ring_err)
 		amdgpu_gfx_reset_start_compute_scheds(adev, ring);
 
-	return r;
+	/* If this reset is triggered by non-KCQ, the KCQ result after resume should
+	 * not override the reset result. It can return a false reset failure to the
+	 * non-KCQ caller
+	 */
+	return ring? ring_err : r;
 }
 
 int amdgpu_gfx_cleaner_shader_sw_init(struct amdgpu_device *adev,
Signature · 1 line
2.43.0
Acked-by on v1 1/1
#1 lore raw

AMD General

Quoted 66 lines

-----Original Message-----
From: Lin, Amber <[email protected]>
Sent: Friday, June 26, 2026 11:11 AM
To: [email protected]; Zhang, Jesse(Jie) <[email protected]>
Cc: Lin, Amber <[email protected]>
Subject: [PATCH] drm/amdgpu: Fix false error return to non-KCQ

amdgpu_gfx_reset_mes_compute is used to coordinate suspend_all, reset, and resume_all between KCQ and compute user queues. When a hung queue comes from the computer user queues and the reset is successful, the KCQ failure after reset should be sent to KCQ only and not the compute user queues. Compute user queues can operate after a successful reset without a mode reset.

Fixes 2dcf9a5a2352 ("drm/amdgpu/gfx: defer per-queue helper_end until after MES

resume")

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c +1 -0
index 982b41606d48..52dd26dd82d2 100644
@@ -2282,6 +2282,7 @@ int amdgpu_gfx_reset_mes_compute(struct
 amdgpu_device *adev,
      struct mes_remove_queue_input *queue_input = (struct
 mes_remove_queue_input *)faulty_queue_input;
      struct amdgpu_gfx_deferred_entry
 deferred_end[AMDGPU_MAX_COMPUTE_RINGS + 1];
      int n_deferred = 0;
+     int ring_err;
      guard(mutex)(&adev->gfx.mec.reset_mutex);
      /* stop the drm schedulers for all compute queues */ @@ -2375,17
+2376,23 @@ int amdgpu_gfx_reset_mes_compute(struct amdgpu_device *adev,
      /* Now CP is running again — replay backed-up commands and ring
       * doorbells on each reset queue.
       */
drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c +10 -5
+     ring_err = r;
      for (i = 0; i < n_deferred; i++) {
              int er = amdgpu_ring_reset_helper_end(deferred_end[i].ring,
                                                    deferred_end[i].fence);
-             if (er && !r)
-                     r = er;
+
+             if (er && !ring_err)
+                     ring_err = er;
      }
 
-     if (!r)
+     if (!ring_err)
              amdgpu_gfx_reset_start_compute_scheds(adev, ring);
 
-     return r;
+     /* If this reset is triggered by non-KCQ, the KCQ result after resume should
+      * not override the reset result. It can return a false reset failure to the
+      * non-KCQ caller
+      */
+     return ring? ring_err : r;
 }
 
 int amdgpu_gfx_cleaner_shader_sw_init(struct amdgpu_device *adev,
--

2.43.0