drm/amdgpu/gfx_9.4.3: wait for reset done before remap

There is a racing condition that cp firmware modifies
MQD in reset sequence after driver updates it for
remapping. We have to wait till CP_HQD_ACTIVE becoming
false then remap the queue.

v2: fix KIQ locking (Alex)
v3: fix KIQ locking harder

Acked-by: Vitaly Prosyak <[email protected]>
Signed-off-by: Jiadong Zhu <[email protected]>
Reviewed-by: Alex Deucher <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>

1 files changed, 33 insertions, 6 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c
index 44c6e2d44722..9a740020243d 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c
@@ -3433,7 +3433,7 @@ static int gfx_v9_4_3_reset_kcq(struct amdgpu_ring *ring,
 	struct amdgpu_kiq *kiq = &adev->gfx.kiq[ring->xcc_id];
 	struct amdgpu_ring *kiq_ring = &kiq->ring;
 	unsigned long flags;
-	int r;
+	int r, i;
 
 	if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues)
 		return -EINVAL;
@@ -3455,9 +3455,28 @@ static int gfx_v9_4_3_reset_kcq(struct amdgpu_ring *ring,
 	if (r)
 		return r;
 
+	/* make sure dequeue is complete*/
+	gfx_v9_4_3_xcc_set_safe_mode(adev, ring->xcc_id);
+	mutex_lock(&adev->srbm_mutex);
+	soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, GET_INST(GC, ring->xcc_id));
+	for (i = 0; i < adev->usec_timeout; i++) {
+		if (!(RREG32_SOC15(GC, 0, regCP_HQD_ACTIVE) & 1))
+			break;
+		udelay(1);
+	}
+	if (i >= adev->usec_timeout)
+		r = -ETIMEDOUT;
+	soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, ring->xcc_id));
+	mutex_unlock(&adev->srbm_mutex);
+	gfx_v9_4_3_xcc_unset_safe_mode(adev, ring->xcc_id);
+	if (r) {
+		dev_err(adev->dev, "fail to wait on hqd deactive\n");
+		return r;
+	}
+
 	r = amdgpu_bo_reserve(ring->mqd_obj, false);
 	if (unlikely(r != 0)){
-		DRM_ERROR("fail to resv mqd_obj\n");
+		dev_err(adev->dev, "fail to resv mqd_obj\n");
 		return r;
 	}
 	r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
@@ -3467,15 +3486,23 @@ static int gfx_v9_4_3_reset_kcq(struct amdgpu_ring *ring,
 		ring->mqd_ptr = NULL;
 	}
 	amdgpu_bo_unreserve(ring->mqd_obj);
-	if (r){
-		DRM_ERROR("fail to unresv mqd_obj\n");
+	if (r) {
+		dev_err(adev->dev, "fail to unresv mqd_obj\n");
 		return r;
 	}
+	spin_lock_irqsave(&kiq->ring_lock, flags);
 	r = amdgpu_ring_alloc(kiq_ring, kiq->pmf->map_queues_size);
+	if (r) {
+		spin_unlock_irqrestore(&kiq->ring_lock, flags);
+		return -ENOMEM;
+	}
 	kiq->pmf->kiq_map_queues(kiq_ring, ring);
+	amdgpu_ring_commit(kiq_ring);
+	spin_unlock_irqrestore(&kiq->ring_lock, flags);
+
 	r = amdgpu_ring_test_ring(kiq_ring);
-	if (r){
-		DRM_ERROR("fail to remap queue\n");
+	if (r) {
+		dev_err(adev->dev, "fail to remap queue\n");
 		return r;
 	}
 	return amdgpu_ring_test_ring(ring);