drm/amdgpu: Fix the ring buffer size for queue VM flush
authorPrike Liang <Prike.Liang@amd.com>
Mon, 25 Mar 2024 07:33:34 +0000 (15:33 +0800)
committerAlex Deucher <alexander.deucher@amd.com>
Fri, 26 Apr 2024 21:22:38 +0000 (17:22 -0400)
Here are the corrections needed for the queue ring buffer size
calculation for the following cases:
- Remove the KIQ VM flush ring usage.
- Add the invalidate TLBs packet for gfx10 and gfx11 queue.
- There's no VM flush and PFP sync, so remove the gfx9 real
  ring and compute ring buffer usage.

Signed-off-by: Prike Liang <Prike.Liang@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c

index d5b9242..591e8ed 100644 (file)
@@ -9186,7 +9186,7 @@ static const struct amdgpu_ring_funcs gfx_v10_0_ring_funcs_gfx = {
                7 + /* PIPELINE_SYNC */
                SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
                SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
-               2 + /* VM_FLUSH */
+               4 + /* VM_FLUSH */
                8 + /* FENCE for VM_FLUSH */
                20 + /* GDS switch */
                4 + /* double SWITCH_BUFFER,
@@ -9276,7 +9276,6 @@ static const struct amdgpu_ring_funcs gfx_v10_0_ring_funcs_kiq = {
                7 + /* gfx_v10_0_ring_emit_pipeline_sync */
                SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
                SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
-               2 + /* gfx_v10_0_ring_emit_vm_flush */
                8 + 8 + 8, /* gfx_v10_0_ring_emit_fence_kiq x3 for user fence, vm fence */
        .emit_ib_size = 7, /* gfx_v10_0_ring_emit_ib_compute */
        .emit_ib = gfx_v10_0_ring_emit_ib_compute,
index 5dbfef4..3c01d89 100644 (file)
@@ -6187,7 +6187,7 @@ static const struct amdgpu_ring_funcs gfx_v11_0_ring_funcs_gfx = {
                7 + /* PIPELINE_SYNC */
                SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
                SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
-               2 + /* VM_FLUSH */
+               4 + /* VM_FLUSH */
                8 + /* FENCE for VM_FLUSH */
                20 + /* GDS switch */
                5 + /* COND_EXEC */
@@ -6273,7 +6273,6 @@ static const struct amdgpu_ring_funcs gfx_v11_0_ring_funcs_kiq = {
                7 + /* gfx_v11_0_ring_emit_pipeline_sync */
                SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
                SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
-               2 + /* gfx_v11_0_ring_emit_vm_flush */
                8 + 8 + 8, /* gfx_v11_0_ring_emit_fence_kiq x3 for user fence, vm fence */
        .emit_ib_size = 7, /* gfx_v11_0_ring_emit_ib_compute */
        .emit_ib = gfx_v11_0_ring_emit_ib_compute,
index 71b5559..eff3356 100644 (file)
@@ -6981,7 +6981,6 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_compute = {
                7 + /* gfx_v9_0_ring_emit_pipeline_sync */
                SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
                SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
-               2 + /* gfx_v9_0_ring_emit_vm_flush */
                8 + 8 + 8 + /* gfx_v9_0_ring_emit_fence x3 for user fence, vm fence */
                7 + /* gfx_v9_0_emit_mem_sync */
                5 + /* gfx_v9_0_emit_wave_limit for updating mmSPI_WCL_PIPE_PERCENT_GFX register */
@@ -7019,7 +7018,6 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_kiq = {
                7 + /* gfx_v9_0_ring_emit_pipeline_sync */
                SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
                SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
-               2 + /* gfx_v9_0_ring_emit_vm_flush */
                8 + 8 + 8, /* gfx_v9_0_ring_emit_fence_kiq x3 for user fence, vm fence */
        .emit_ib_size = 7, /* gfx_v9_0_ring_emit_ib_compute */
        .emit_fence = gfx_v9_0_ring_emit_fence_kiq,