drm/amdgpu: Program ring for vce instance 1 at its register space
authorLeo Liu <leo.liu@amd.com>
Mon, 29 May 2017 17:13:59 +0000 (13:13 -0400)
committerAlex Deucher <alexander.deucher@amd.com>
Wed, 31 May 2017 18:16:37 +0000 (14:16 -0400)
We need program ring buffer on instance 1 register space domain,
when only if instance 1 available, with two instances or instance 0,
and we need only program instance 0 regsiter space domain for ring.

Signed-off-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drivers/gpu/drm/amd/amdgpu/vce_v3_0.c

index fb08193..90332f5 100644 (file)
@@ -77,13 +77,26 @@ static int vce_v3_0_set_clockgating_state(void *handle,
 static uint64_t vce_v3_0_ring_get_rptr(struct amdgpu_ring *ring)
 {
        struct amdgpu_device *adev = ring->adev;
+       u32 v;
+
+       mutex_lock(&adev->grbm_idx_mutex);
+       if (adev->vce.harvest_config == 0 ||
+               adev->vce.harvest_config == AMDGPU_VCE_HARVEST_VCE1)
+               WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(0));
+       else if (adev->vce.harvest_config == AMDGPU_VCE_HARVEST_VCE0)
+               WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(1));
 
        if (ring == &adev->vce.ring[0])
-               return RREG32(mmVCE_RB_RPTR);
+               v = RREG32(mmVCE_RB_RPTR);
        else if (ring == &adev->vce.ring[1])
-               return RREG32(mmVCE_RB_RPTR2);
+               v = RREG32(mmVCE_RB_RPTR2);
        else
-               return RREG32(mmVCE_RB_RPTR3);
+               v = RREG32(mmVCE_RB_RPTR3);
+
+       WREG32(mmGRBM_GFX_INDEX, mmGRBM_GFX_INDEX_DEFAULT);
+       mutex_unlock(&adev->grbm_idx_mutex);
+
+       return v;
 }
 
 /**
@@ -96,13 +109,26 @@ static uint64_t vce_v3_0_ring_get_rptr(struct amdgpu_ring *ring)
 static uint64_t vce_v3_0_ring_get_wptr(struct amdgpu_ring *ring)
 {
        struct amdgpu_device *adev = ring->adev;
+       u32 v;
+
+       mutex_lock(&adev->grbm_idx_mutex);
+       if (adev->vce.harvest_config == 0 ||
+               adev->vce.harvest_config == AMDGPU_VCE_HARVEST_VCE1)
+               WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(0));
+       else if (adev->vce.harvest_config == AMDGPU_VCE_HARVEST_VCE0)
+               WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(1));
 
        if (ring == &adev->vce.ring[0])
-               return RREG32(mmVCE_RB_WPTR);
+               v = RREG32(mmVCE_RB_WPTR);
        else if (ring == &adev->vce.ring[1])
-               return RREG32(mmVCE_RB_WPTR2);
+               v = RREG32(mmVCE_RB_WPTR2);
        else
-               return RREG32(mmVCE_RB_WPTR3);
+               v = RREG32(mmVCE_RB_WPTR3);
+
+       WREG32(mmGRBM_GFX_INDEX, mmGRBM_GFX_INDEX_DEFAULT);
+       mutex_unlock(&adev->grbm_idx_mutex);
+
+       return v;
 }
 
 /**
@@ -116,12 +142,22 @@ static void vce_v3_0_ring_set_wptr(struct amdgpu_ring *ring)
 {
        struct amdgpu_device *adev = ring->adev;
 
+       mutex_lock(&adev->grbm_idx_mutex);
+       if (adev->vce.harvest_config == 0 ||
+               adev->vce.harvest_config == AMDGPU_VCE_HARVEST_VCE1)
+               WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(0));
+       else if (adev->vce.harvest_config == AMDGPU_VCE_HARVEST_VCE0)
+               WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(1));
+
        if (ring == &adev->vce.ring[0])
                WREG32(mmVCE_RB_WPTR, lower_32_bits(ring->wptr));
        else if (ring == &adev->vce.ring[1])
                WREG32(mmVCE_RB_WPTR2, lower_32_bits(ring->wptr));
        else
                WREG32(mmVCE_RB_WPTR3, lower_32_bits(ring->wptr));
+
+       WREG32(mmGRBM_GFX_INDEX, mmGRBM_GFX_INDEX_DEFAULT);
+       mutex_unlock(&adev->grbm_idx_mutex);
 }
 
 static void vce_v3_0_override_vce_clock_gating(struct amdgpu_device *adev, bool override)
@@ -231,33 +267,38 @@ static int vce_v3_0_start(struct amdgpu_device *adev)
        struct amdgpu_ring *ring;
        int idx, r;
 
-       ring = &adev->vce.ring[0];
-       WREG32(mmVCE_RB_RPTR, lower_32_bits(ring->wptr));
-       WREG32(mmVCE_RB_WPTR, lower_32_bits(ring->wptr));
-       WREG32(mmVCE_RB_BASE_LO, ring->gpu_addr);
-       WREG32(mmVCE_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
-       WREG32(mmVCE_RB_SIZE, ring->ring_size / 4);
-
-       ring = &adev->vce.ring[1];
-       WREG32(mmVCE_RB_RPTR2, lower_32_bits(ring->wptr));
-       WREG32(mmVCE_RB_WPTR2, lower_32_bits(ring->wptr));
-       WREG32(mmVCE_RB_BASE_LO2, ring->gpu_addr);
-       WREG32(mmVCE_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
-       WREG32(mmVCE_RB_SIZE2, ring->ring_size / 4);
-
-       ring = &adev->vce.ring[2];
-       WREG32(mmVCE_RB_RPTR3, lower_32_bits(ring->wptr));
-       WREG32(mmVCE_RB_WPTR3, lower_32_bits(ring->wptr));
-       WREG32(mmVCE_RB_BASE_LO3, ring->gpu_addr);
-       WREG32(mmVCE_RB_BASE_HI3, upper_32_bits(ring->gpu_addr));
-       WREG32(mmVCE_RB_SIZE3, ring->ring_size / 4);
-
        mutex_lock(&adev->grbm_idx_mutex);
        for (idx = 0; idx < 2; ++idx) {
                if (adev->vce.harvest_config & (1 << idx))
                        continue;
 
                WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(idx));
+
+               /* Program instance 0 reg space for two instances or instance 0 case
+               program instance 1 reg space for only instance 1 available case */
+               if (idx != 1 || adev->vce.harvest_config == AMDGPU_VCE_HARVEST_VCE0) {
+                       ring = &adev->vce.ring[0];
+                       WREG32(mmVCE_RB_RPTR, lower_32_bits(ring->wptr));
+                       WREG32(mmVCE_RB_WPTR, lower_32_bits(ring->wptr));
+                       WREG32(mmVCE_RB_BASE_LO, ring->gpu_addr);
+                       WREG32(mmVCE_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
+                       WREG32(mmVCE_RB_SIZE, ring->ring_size / 4);
+
+                       ring = &adev->vce.ring[1];
+                       WREG32(mmVCE_RB_RPTR2, lower_32_bits(ring->wptr));
+                       WREG32(mmVCE_RB_WPTR2, lower_32_bits(ring->wptr));
+                       WREG32(mmVCE_RB_BASE_LO2, ring->gpu_addr);
+                       WREG32(mmVCE_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
+                       WREG32(mmVCE_RB_SIZE2, ring->ring_size / 4);
+
+                       ring = &adev->vce.ring[2];
+                       WREG32(mmVCE_RB_RPTR3, lower_32_bits(ring->wptr));
+                       WREG32(mmVCE_RB_WPTR3, lower_32_bits(ring->wptr));
+                       WREG32(mmVCE_RB_BASE_LO3, ring->gpu_addr);
+                       WREG32(mmVCE_RB_BASE_HI3, upper_32_bits(ring->gpu_addr));
+                       WREG32(mmVCE_RB_SIZE3, ring->ring_size / 4);
+               }
+
                vce_v3_0_mc_resume(adev, idx);
                WREG32_FIELD(VCE_STATUS, JOB_BUSY, 1);