diff options
author | Monk Liu <Monk.Liu@amd.com> | 2017-03-24 07:07:52 +0300 |
---|---|---|
committer | Alex Deucher <alexander.deucher@amd.com> | 2017-03-30 06:55:34 +0300 |
commit | 806ba2d4f523f7414103fa0bdd2e3b17a340ac41 (patch) | |
tree | 709309e0dcaf593272f61bd3916cd546c4fa98ba /drivers/gpu/drm | |
parent | 65333e4429c60b8bbbb7a9213cb5612761499acd (diff) | |
download | linux-806ba2d4f523f7414103fa0bdd2e3b17a340ac41.tar.xz |
drm/amdgpu:implement cond_exec for gfx8
when MCBP enabled for gfx8, the cond_exec must also
be implemented, otherwise there will be odds to meet
cross engine (ce and me) deadlock when world switch
happens.
Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Diffstat (limited to 'drivers/gpu/drm')
-rw-r--r-- | drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 30 |
1 files changed, 30 insertions, 0 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c index 423ed68c2e52..0050b09d5883 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c @@ -6781,6 +6781,34 @@ static void gfx_v8_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags) (flags & AMDGPU_VM_DOMAIN) ? AMDGPU_CSA_VADDR : ring->adev->virt.csa_vmid0_addr); } +static unsigned gfx_v8_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring) +{ + unsigned ret; + + amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3)); + amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr)); + amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr)); + amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */ + ret = ring->wptr & ring->buf_mask; + amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */ + return ret; +} + +static void gfx_v8_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned offset) +{ + unsigned cur; + + BUG_ON(offset > ring->buf_mask); + BUG_ON(ring->ring[offset] != 0x55aa55aa); + + cur = (ring->wptr & ring->buf_mask) - 1; + if (likely(cur > offset)) + ring->ring[offset] = cur - offset; + else + ring->ring[offset] = (ring->ring_size >> 2) - offset + cur; +} + + static void gfx_v8_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg) { struct amdgpu_device *adev = ring->adev; @@ -7070,6 +7098,8 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_gfx = { .pad_ib = amdgpu_ring_generic_pad_ib, .emit_switch_buffer = gfx_v8_ring_emit_sb, .emit_cntxcntl = gfx_v8_ring_emit_cntxcntl, + .init_cond_exec = gfx_v8_0_ring_emit_init_cond_exec, + .patch_cond_exec = gfx_v8_0_ring_emit_patch_cond_exec, }; static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = { |