drm/amdgpu/gfx12: fix EOP interrupt routing for KQ and userq

Try KQ by ring_id first (KCQ and UQ never share a HW slot); fall back
to amdgpu_userq_process_fence_irq() on miss, since KCQ EOPs were
misrouted into the userq fence path when enable_mes is true.

Require a strict (me,pipe,queue) match in the gfx case, then userq gfx
EOPs fall through to amdgpu_userq_process_fence_irq().

Suggested-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Jesse Zhang <jesse.zhang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 6c1f4f7ff08448e0e18cd7fc4e59d6c96a36f25d)
Cc: stable@vger.kernel.org

1 files changed, 29 insertions, 14 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c
index 989c8e2baf6a..3f3b1754c038 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c
@@ -4835,25 +4835,33 @@ static int gfx_v12_0_eop_irq(struct amdgpu_device *adev,
 			     struct amdgpu_iv_entry *entry)
 {
 	u32 doorbell_offset = entry->src_data[0];
-	u8 me_id, pipe_id, queue_id;
-	struct amdgpu_ring *ring;
-	int i;
 
 	DRM_DEBUG("IH: CP EOP\n");
 
-	if (adev->enable_mes && doorbell_offset) {
-		amdgpu_userq_process_fence_irq(adev, doorbell_offset);
-	} else {
-		me_id = (entry->ring_id & 0x0c) >> 2;
-		pipe_id = (entry->ring_id & 0x03) >> 0;
-		queue_id = (entry->ring_id & 0x70) >> 4;
+	if (!adev->gfx.disable_kq) {
+		u8 me_id = (entry->ring_id & 0x0c) >> 2;
+		u8 pipe_id = (entry->ring_id & 0x03) >> 0;
+		u8 queue_id = (entry->ring_id & 0x70) >> 4;
+		struct amdgpu_ring *ring;
+		int i;
 
 		switch (me_id) {
 		case 0:
-			if (pipe_id == 0)
-				amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
-			else
-				amdgpu_fence_process(&adev->gfx.gfx_ring[1]);
+			/*
+			 * MES splits gfx HQDs per (me,pipe): KGQ owns queue=0,
+			 * userq gfx owns queue>=1 (see amdgpu_mes_get_hqd_mask).
+			 * Require a strict (me,pipe,queue) match so userq gfx
+			 * EOPs fall through to amdgpu_userq_process_fence_irq().
+			 */
+			for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
+				ring = &adev->gfx.gfx_ring[i];
+				if ((ring->me == me_id) &&
+				    (ring->pipe == pipe_id) &&
+				    (ring->queue == queue_id)) {
+					amdgpu_fence_process(ring);
+					return 0;
+				}
+			}
 			break;
 		case 1:
 		case 2:
@@ -4865,13 +4873,20 @@ static int gfx_v12_0_eop_irq(struct amdgpu_device *adev,
 				 */
 				if ((ring->me == me_id) &&
 				    (ring->pipe == pipe_id) &&
-				    (ring->queue == queue_id))
+				    (ring->queue == queue_id)) {
 					amdgpu_fence_process(ring);
+					return 0;
+				}
 			}
 			break;
+		default:
+			break;
 		}
 	}
 
+	if (adev->enable_mes && doorbell_offset)
+		amdgpu_userq_process_fence_irq(adev, doorbell_offset);
+
 	return 0;
 }