drm/amdgpu: set the executable flag on unused Vega10 PTEs v2
authorChristian König <christian.koenig@amd.com>
Wed, 12 Dec 2018 14:13:10 +0000 (15:13 +0100)
committerAlex Deucher <alexander.deucher@amd.com>
Fri, 25 Jan 2019 21:15:34 +0000 (16:15 -0500)
Otherwise we run into a non-retry fault on access.

It seems to be a hardware bug that the executable bit has
higher priority than the valid bit.

v2: handle clears as well

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c

index ffffe021cccf063d64198e27e8ea5d9052387638..0bc6f553dc081700e58076685aa3314d95948680 100644 (file)
@@ -821,9 +821,16 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev,
                addr += ats_entries * 8;
        }
 
-       if (entries)
+       if (entries) {
+               uint64_t value = 0;
+
+               /* Workaround for fault priority problem on GMC9 */
+               if (level == AMDGPU_VM_PTB && adev->asic_type >= CHIP_VEGA10)
+                       value = AMDGPU_PTE_EXECUTABLE;
+
                amdgpu_vm_set_pte_pde(adev, &job->ibs[0], addr, 0,
-                                     entries, 0, 0);
+                                     entries, 0, value);
+       }
 
        amdgpu_ring_pad_ib(ring, &job->ibs[0]);
 
@@ -1525,20 +1532,27 @@ error:
 }
 
 /**
- * amdgpu_vm_update_huge - figure out parameters for PTE updates
+ * amdgpu_vm_update_flags - figure out flags for PTE updates
  *
  * Make sure to set the right flags for the PTEs at the desired level.
  */
-static void amdgpu_vm_update_huge(struct amdgpu_pte_update_params *params,
-                                 struct amdgpu_bo *bo, unsigned level,
-                                 uint64_t pe, uint64_t addr,
-                                 unsigned count, uint32_t incr,
-                                 uint64_t flags)
+static void amdgpu_vm_update_flags(struct amdgpu_pte_update_params *params,
+                                  struct amdgpu_bo *bo, unsigned level,
+                                  uint64_t pe, uint64_t addr,
+                                  unsigned count, uint32_t incr,
+                                  uint64_t flags)
 
 {
        if (level != AMDGPU_VM_PTB) {
                flags |= AMDGPU_PDE_PTE;
                amdgpu_gmc_get_vm_pde(params->adev, level, &addr, &flags);
+
+       } else if (params->adev->asic_type >= CHIP_VEGA10 &&
+                  !(flags & AMDGPU_PTE_VALID) &&
+                  !(flags & AMDGPU_PTE_PRT)) {
+
+               /* Workaround for fault priority problem on GMC9 */
+               flags |= AMDGPU_PTE_EXECUTABLE;
        }
 
        amdgpu_vm_update_func(params, bo, pe, addr, count, incr, flags);
@@ -1695,9 +1709,9 @@ static int amdgpu_vm_update_ptes(struct amdgpu_pte_update_params *params,
                        uint64_t upd_end = min(entry_end, frag_end);
                        unsigned nptes = (upd_end - frag_start) >> shift;
 
-                       amdgpu_vm_update_huge(params, pt, cursor.level,
-                                             pe_start, dst, nptes, incr,
-                                             flags | AMDGPU_PTE_FRAG(frag));
+                       amdgpu_vm_update_flags(params, pt, cursor.level,
+                                              pe_start, dst, nptes, incr,
+                                              flags | AMDGPU_PTE_FRAG(frag));
 
                        pe_start += nptes * 8;
                        dst += (uint64_t)nptes * AMDGPU_GPU_PAGE_SIZE << shift;