diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gds.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gds.h index dad2186f4ed5..df8a23554831 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gds.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gds.h @@ -31,7 +31,8 @@ struct amdgpu_gds { uint32_t gds_size; uint32_t gws_size; uint32_t oa_size; - uint32_t gds_compute_max_wave_id; + uint32_t gds_compute_max_wave_id; + uint32_t vgt_gs_max_wave_id; }; struct amdgpu_gds_reg_offset { diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c index ccd5a4968a60..ef222bfb31a1 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c @@ -4197,6 +4197,15 @@ static void gfx_v10_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, unsigned vmid = AMDGPU_JOB_GET_VMID(job); u32 header, control = 0; + /* Prevent a hw deadlock due to a wave ID mismatch between ME and GDS. + * This resets the wave ID counters. (needed by transform feedback) + * TODO: This might only be needed on a VMID switch when we change + * the GDS OA mapping, not sure. + */ + amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); + amdgpu_ring_write(ring, mmVGT_GS_MAX_WAVE_ID); + amdgpu_ring_write(ring, ring->adev->gds.vgt_gs_max_wave_id); + if (ib->flags & AMDGPU_IB_FLAG_CE) header = PACKET3(PACKET3_INDIRECT_BUFFER_CNST, 2); else @@ -4930,7 +4939,7 @@ static const struct amdgpu_ring_funcs gfx_v10_0_ring_funcs_gfx = { 5 + /* HDP_INVL */ 8 + 8 + /* FENCE x2 */ 2, /* SWITCH_BUFFER */ - .emit_ib_size = 4, /* gfx_v10_0_ring_emit_ib_gfx */ + .emit_ib_size = 7, /* gfx_v10_0_ring_emit_ib_gfx */ .emit_ib = gfx_v10_0_ring_emit_ib_gfx, .emit_fence = gfx_v10_0_ring_emit_fence, .emit_pipeline_sync = gfx_v10_0_ring_emit_pipeline_sync, @@ -5078,10 +5087,9 @@ static void gfx_v10_0_set_gds_init(struct amdgpu_device *adev) /* init asic gds info */ switch (adev->asic_type) { case CHIP_NAVI10: - adev->gds.gds_size = 0x10000; - break; default: adev->gds.gds_size = 0x10000; + adev->gds.vgt_gs_max_wave_id = 0x3ff; break; }