drm/amdgpu/gfx10: explicitly wait for cp idle after halt/unhalt
authorXiaojie Yuan <xiaojie.yuan@amd.com>
Thu, 14 Nov 2019 08:56:08 +0000 (16:56 +0800)
committerAlex Deucher <alexander.deucher@amd.com>
Fri, 22 Nov 2019 19:54:56 +0000 (14:54 -0500)
50us is not enough to wait for cp ready after gpu reset on some navi asics.

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Suggested-by: Jack Xiao <Jack.Xiao@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c

index b128012..f439dc3 100644 (file)
@@ -2400,7 +2400,7 @@ static int gfx_v10_0_wait_for_rlc_autoload_complete(struct amdgpu_device *adev)
        return 0;
 }
 
-static void gfx_v10_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
+static int gfx_v10_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
 {
        int i;
        u32 tmp = RREG32_SOC15(GC, 0, mmCP_ME_CNTL);
@@ -2413,7 +2413,17 @@ static void gfx_v10_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
                        adev->gfx.gfx_ring[i].sched.ready = false;
        }
        WREG32_SOC15(GC, 0, mmCP_ME_CNTL, tmp);
-       udelay(50);
+
+       for (i = 0; i < adev->usec_timeout; i++) {
+               if (RREG32_SOC15(GC, 0, mmCP_STAT) == 0)
+                       break;
+               udelay(1);
+       }
+
+       if (i >= adev->usec_timeout)
+               DRM_ERROR("failed to %s cp gfx\n", enable ? "unhalt" : "halt");
+
+       return 0;
 }
 
 static int gfx_v10_0_cp_gfx_load_pfp_microcode(struct amdgpu_device *adev)