drm/amdgpu: move reset support type checks into the caller
authorAlex Deucher <alexander.deucher@amd.com>
Tue, 15 Jul 2025 15:55:05 +0000 (11:55 -0400)
committerAlex Deucher <alexander.deucher@amd.com>
Thu, 17 Jul 2025 16:36:56 +0000 (12:36 -0400)
Rather than checking in the callbacks, check if the reset
type is supported in the caller.

Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
25 files changed:
drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c
drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c
drivers/gpu/drm/amd/amdgpu/jpeg_v2_0.c
drivers/gpu/drm/amd/amdgpu/jpeg_v2_5.c
drivers/gpu/drm/amd/amdgpu/jpeg_v3_0.c
drivers/gpu/drm/amd/amdgpu/jpeg_v4_0.c
drivers/gpu/drm/amd/amdgpu/jpeg_v4_0_5.c
drivers/gpu/drm/amd/amdgpu/jpeg_v5_0_0.c
drivers/gpu/drm/amd/amdgpu/jpeg_v5_0_1.c
drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c
drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c
drivers/gpu/drm/amd/amdgpu/sdma_v6_0.c
drivers/gpu/drm/amd/amdgpu/sdma_v7_0.c
drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c
drivers/gpu/drm/amd/amdgpu/vcn_v4_0_3.c
drivers/gpu/drm/amd/amdgpu/vcn_v4_0_5.c
drivers/gpu/drm/amd/amdgpu/vcn_v5_0_0.c

index 5cb7bf9..32a0852 100644 (file)
@@ -112,6 +112,7 @@ static enum drm_gpu_sched_stat amdgpu_job_timedout(struct drm_sched_job *s_job)
                amdgpu_job_core_dump(adev, job);
 
        if (amdgpu_gpu_recovery &&
+           amdgpu_ring_is_reset_type_supported(ring, AMDGPU_RESET_TYPE_SOFT_RESET) &&
            amdgpu_ring_soft_recovery(ring, job->vmid, s_job->s_fence->parent)) {
                dev_err(adev->dev, "ring %s timeout, but soft recovered\n",
                        s_job->sched->name);
@@ -131,7 +132,9 @@ static enum drm_gpu_sched_stat amdgpu_job_timedout(struct drm_sched_job *s_job)
        /* attempt a per ring reset */
        if (unlikely(adev->debug_disable_gpu_ring_reset)) {
                dev_err(adev->dev, "Ring reset disabled by debug mask\n");
-       } else if (amdgpu_gpu_recovery && ring->funcs->reset) {
+       } else if (amdgpu_gpu_recovery &&
+                  amdgpu_ring_is_reset_type_supported(ring, AMDGPU_RESET_TYPE_PER_QUEUE) &&
+                  ring->funcs->reset) {
                dev_err(adev->dev, "Starting %s ring reset\n",
                        s_job->sched->name);
                r = amdgpu_ring_reset(ring, job->vmid, &job->hw_fence);
index acac646..a5c3f64 100644 (file)
@@ -825,3 +825,34 @@ int amdgpu_ring_reset_helper_end(struct amdgpu_ring *ring,
        drm_sched_wqueue_start(&ring->sched);
        return 0;
 }
+
+bool amdgpu_ring_is_reset_type_supported(struct amdgpu_ring *ring,
+                                        u32 reset_type)
+{
+       switch (ring->funcs->type) {
+       case AMDGPU_RING_TYPE_GFX:
+               if (ring->adev->gfx.gfx_supported_reset & reset_type)
+                       return true;
+               break;
+       case AMDGPU_RING_TYPE_COMPUTE:
+               if (ring->adev->gfx.compute_supported_reset & reset_type)
+                       return true;
+               break;
+       case AMDGPU_RING_TYPE_SDMA:
+               if (ring->adev->sdma.supported_reset & reset_type)
+                       return true;
+               break;
+       case AMDGPU_RING_TYPE_VCN_DEC:
+       case AMDGPU_RING_TYPE_VCN_ENC:
+               if (ring->adev->vcn.supported_reset & reset_type)
+                       return true;
+               break;
+       case AMDGPU_RING_TYPE_VCN_JPEG:
+               if (ring->adev->jpeg.supported_reset & reset_type)
+                       return true;
+               break;
+       default:
+               break;
+       }
+       return false;
+}
index 309e7bb..7670f5d 100644 (file)
@@ -568,4 +568,6 @@ void amdgpu_ring_reset_helper_begin(struct amdgpu_ring *ring,
                                    struct amdgpu_fence *guilty_fence);
 int amdgpu_ring_reset_helper_end(struct amdgpu_ring *ring,
                                 struct amdgpu_fence *guilty_fence);
+bool amdgpu_ring_is_reset_type_supported(struct amdgpu_ring *ring,
+                                        u32 reset_type);
 #endif
index d799bc7..f1f6752 100644 (file)
@@ -1522,9 +1522,6 @@ int amdgpu_vcn_ring_reset(struct amdgpu_ring *ring,
 {
        struct amdgpu_device *adev = ring->adev;
 
-       if (!(adev->vcn.supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        if (adev->vcn.inst[ring->me].using_unified_queue)
                return -EINVAL;
 
index 744f51a..7bd506f 100644 (file)
@@ -9523,9 +9523,6 @@ static int gfx_v10_0_reset_kgq(struct amdgpu_ring *ring,
        u64 addr;
        int r;
 
-       if (!(adev->gfx.gfx_supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues)
                return -EINVAL;
 
@@ -9591,9 +9588,6 @@ static int gfx_v10_0_reset_kcq(struct amdgpu_ring *ring,
        unsigned long flags;
        int i, r;
 
-       if (!(adev->gfx.compute_supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues)
                return -EINVAL;
 
index 465fcd6..c01c241 100644 (file)
@@ -6808,9 +6808,6 @@ static int gfx_v11_0_reset_kgq(struct amdgpu_ring *ring,
        struct amdgpu_device *adev = ring->adev;
        int r;
 
-       if (!(adev->gfx.gfx_supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        amdgpu_ring_reset_helper_begin(ring, timedout_fence);
 
        r = amdgpu_mes_reset_legacy_queue(ring->adev, ring, vmid, false);
@@ -6974,9 +6971,6 @@ static int gfx_v11_0_reset_kcq(struct amdgpu_ring *ring,
        struct amdgpu_device *adev = ring->adev;
        int r = 0;
 
-       if (!(adev->gfx.compute_supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        amdgpu_ring_reset_helper_begin(ring, timedout_fence);
 
        r = amdgpu_mes_reset_legacy_queue(ring->adev, ring, vmid, true);
index 5d32ae4..09bf722 100644 (file)
@@ -5303,9 +5303,6 @@ static int gfx_v12_0_reset_kgq(struct amdgpu_ring *ring,
        struct amdgpu_device *adev = ring->adev;
        int r;
 
-       if (!(adev->gfx.gfx_supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        amdgpu_ring_reset_helper_begin(ring, timedout_fence);
 
        r = amdgpu_mes_reset_legacy_queue(ring->adev, ring, vmid, false);
@@ -5421,9 +5418,6 @@ static int gfx_v12_0_reset_kcq(struct amdgpu_ring *ring,
        struct amdgpu_device *adev = ring->adev;
        int r;
 
-       if (!(adev->gfx.compute_supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        amdgpu_ring_reset_helper_begin(ring, timedout_fence);
 
        r = amdgpu_mes_reset_legacy_queue(ring->adev, ring, vmid, true);
index 9c0cdf4..20b30f4 100644 (file)
@@ -7183,9 +7183,6 @@ static int gfx_v9_0_reset_kcq(struct amdgpu_ring *ring,
        unsigned long flags;
        int i, r;
 
-       if (!(adev->gfx.compute_supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues)
                return -EINVAL;
 
index 2763817..51babf5 100644 (file)
@@ -3565,9 +3565,6 @@ static int gfx_v9_4_3_reset_kcq(struct amdgpu_ring *ring,
        unsigned long flags;
        int r;
 
-       if (!(adev->gfx.compute_supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues)
                return -EINVAL;
 
index b93d6af..58239c4 100644 (file)
@@ -773,9 +773,6 @@ static int jpeg_v2_0_ring_reset(struct amdgpu_ring *ring,
 {
        int r;
 
-       if (!(ring->adev->jpeg.supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        amdgpu_ring_reset_helper_begin(ring, timedout_fence);
        r = jpeg_v2_0_stop(ring->adev);
        if (r)
index b6d5ba0..3e2c389 100644 (file)
@@ -650,9 +650,6 @@ static int jpeg_v2_5_ring_reset(struct amdgpu_ring *ring,
                                unsigned int vmid,
                                struct amdgpu_fence *timedout_fence)
 {
-       if (!(ring->adev->jpeg.supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        amdgpu_ring_reset_helper_begin(ring, timedout_fence);
        jpeg_v2_5_stop_inst(ring->adev, ring->me);
        jpeg_v2_5_start_inst(ring->adev, ring->me);
index a229d7e..a44eb26 100644 (file)
@@ -564,9 +564,6 @@ static int jpeg_v3_0_ring_reset(struct amdgpu_ring *ring,
 {
        int r;
 
-       if (!(ring->adev->jpeg.supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        amdgpu_ring_reset_helper_begin(ring, timedout_fence);
        r = jpeg_v3_0_stop(ring->adev);
        if (r)
index f3a9073..da3ee69 100644 (file)
@@ -729,9 +729,6 @@ static int jpeg_v4_0_ring_reset(struct amdgpu_ring *ring,
 {
        int r;
 
-       if (!(ring->adev->jpeg.supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        amdgpu_ring_reset_helper_begin(ring, timedout_fence);
        r = jpeg_v4_0_stop(ring->adev);
        if (r)
index 1892c27..481d1a2 100644 (file)
@@ -774,9 +774,6 @@ static int jpeg_v4_0_5_ring_reset(struct amdgpu_ring *ring,
 {
        int r;
 
-       if (!(ring->adev->jpeg.supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        amdgpu_ring_reset_helper_begin(ring, timedout_fence);
        r = jpeg_v4_0_5_stop(ring->adev);
        if (r)
index 0b4de0c..e0a7190 100644 (file)
@@ -650,9 +650,6 @@ static int jpeg_v5_0_0_ring_reset(struct amdgpu_ring *ring,
 {
        int r;
 
-       if (!(ring->adev->jpeg.supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        amdgpu_ring_reset_helper_begin(ring, timedout_fence);
        r = jpeg_v5_0_0_stop(ring->adev);
        if (r)
index e622db1..54523dc 100644 (file)
@@ -836,9 +836,6 @@ static int jpeg_v5_0_1_ring_reset(struct amdgpu_ring *ring,
                                  unsigned int vmid,
                                  struct amdgpu_fence *timedout_fence)
 {
-       if (amdgpu_sriov_vf(ring->adev))
-               return -EOPNOTSUPP;
-
        amdgpu_ring_reset_helper_begin(ring, timedout_fence);
        jpeg_v5_0_1_core_stall_reset(ring);
        jpeg_v5_0_1_init_jrbc(ring);
index 20fad25..36b1ca7 100644 (file)
@@ -1664,9 +1664,6 @@ static int sdma_v4_4_2_reset_queue(struct amdgpu_ring *ring,
        u32 id = ring->me;
        int r;
 
-       if (!(adev->sdma.supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        amdgpu_amdkfd_suspend(adev, true);
        r = amdgpu_sdma_reset_engine(adev, id, false);
        amdgpu_amdkfd_resume(adev, true);
index d8c1960..7dc67a2 100644 (file)
@@ -1547,9 +1547,6 @@ static int sdma_v5_0_reset_queue(struct amdgpu_ring *ring,
        struct amdgpu_device *adev = ring->adev;
        int r;
 
-       if (!(adev->sdma.supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        if (ring->me >= adev->sdma.num_instances) {
                dev_err(adev->dev, "sdma instance not found\n");
                return -EINVAL;
index 5d56029..3bd44c2 100644 (file)
@@ -1461,9 +1461,6 @@ static int sdma_v5_2_reset_queue(struct amdgpu_ring *ring,
        struct amdgpu_device *adev = ring->adev;
        int r;
 
-       if (!(adev->sdma.supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        if (ring->me >= adev->sdma.num_instances) {
                dev_err(adev->dev, "sdma instance not found\n");
                return -EINVAL;
index 5d31f31..e6d8edd 100644 (file)
@@ -1578,9 +1578,6 @@ static int sdma_v6_0_reset_queue(struct amdgpu_ring *ring,
        struct amdgpu_device *adev = ring->adev;
        int r;
 
-       if (!(adev->sdma.supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        if (ring->me >= adev->sdma.num_instances) {
                dev_err(adev->dev, "sdma instance not found\n");
                return -EINVAL;
index 13abc0c..b8b06d4 100644 (file)
@@ -809,9 +809,6 @@ static int sdma_v7_0_reset_queue(struct amdgpu_ring *ring,
        struct amdgpu_device *adev = ring->adev;
        int r;
 
-       if (!(adev->sdma.supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        if (ring->me >= adev->sdma.num_instances) {
                dev_err(adev->dev, "sdma instance not found\n");
                return -EINVAL;
index f642a06..1924e07 100644 (file)
@@ -1976,9 +1976,6 @@ static int vcn_v4_0_ring_reset(struct amdgpu_ring *ring,
        struct amdgpu_vcn_inst *vinst = &adev->vcn.inst[ring->me];
        int r;
 
-       if (!(adev->vcn.supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        amdgpu_ring_reset_helper_begin(ring, timedout_fence);
        r = vcn_v4_0_stop(vinst);
        if (r)
index 3bda19b..2a3663b 100644 (file)
@@ -1603,12 +1603,6 @@ static int vcn_v4_0_3_ring_reset(struct amdgpu_ring *ring,
        struct amdgpu_device *adev = ring->adev;
        struct amdgpu_vcn_inst *vinst = &adev->vcn.inst[ring->me];
 
-       if (amdgpu_sriov_vf(ring->adev))
-               return -EOPNOTSUPP;
-
-       if (!(adev->vcn.supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        amdgpu_ring_reset_helper_begin(ring, timedout_fence);
 
        vcn_inst = GET_INST(VCN, ring->me);
index 039989a..caf2d95 100644 (file)
@@ -1474,9 +1474,6 @@ static int vcn_v4_0_5_ring_reset(struct amdgpu_ring *ring,
        struct amdgpu_vcn_inst *vinst = &adev->vcn.inst[ring->me];
        int r;
 
-       if (!(adev->vcn.supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        amdgpu_ring_reset_helper_begin(ring, timedout_fence);
        r = vcn_v4_0_5_stop(vinst);
        if (r)
index ec0268b..07a6e95 100644 (file)
@@ -1201,9 +1201,6 @@ static int vcn_v5_0_0_ring_reset(struct amdgpu_ring *ring,
        struct amdgpu_vcn_inst *vinst = &adev->vcn.inst[ring->me];
        int r;
 
-       if (!(adev->vcn.supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
-               return -EOPNOTSUPP;
-
        amdgpu_ring_reset_helper_begin(ring, timedout_fence);
        r = vcn_v5_0_0_stop(vinst);
        if (r)