drm/amdgpu: disable GFX ring and disable PQ wptr in hw_fini
authorMonk Liu <Monk.Liu@amd.com>
Mon, 29 Jan 2018 11:24:32 +0000 (19:24 +0800)
committerAlex Deucher <alexander.deucher@amd.com>
Wed, 28 Feb 2018 19:18:08 +0000 (14:18 -0500)
otherwise there will be DMAR reading error comes out from CP since
GFX is still alive and CPC's WPTR_POLL is still enabled, which would
lead to DMAR read error.

fix:
we can hault CPG after hw_fini, but cannot halt CPC becaues KIQ
stil need to be alive to let RLCV invoke, but its WPTR_POLL could
be disabled.

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c

index e9cc03e78bb09d362045d454dfbe368645e0f33d..d73bbb092202c67666eb4f9130ae09d4db9f7deb 100644 (file)
@@ -2961,7 +2961,13 @@ static int gfx_v9_0_hw_fini(void *handle)
                gfx_v9_0_kcq_disable(&adev->gfx.kiq.ring, &adev->gfx.compute_ring[i]);
 
        if (amdgpu_sriov_vf(adev)) {
-               pr_debug("For SRIOV client, shouldn't do anything.\n");
+               gfx_v9_0_cp_gfx_enable(adev, false);
+               /* must disable polling for SRIOV when hw finished, otherwise
+                * CPC engine may still keep fetching WB address which is already
+                * invalid after sw finished and trigger DMAR reading error in
+                * hypervisor side.
+                */
+               WREG32_FIELD15(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0);
                return 0;
        }
        gfx_v9_0_cp_enable(adev, false);