]> git.baikalelectronics.ru Git - kernel.git/commitdiff
drm/amdgpu: fix VM PD addr shift
authorChristian König <christian.koenig@amd.com>
Mon, 27 Nov 2017 13:01:51 +0000 (14:01 +0100)
committerAlex Deucher <alexander.deucher@amd.com>
Wed, 6 Dec 2017 17:48:28 +0000 (12:48 -0500)
The block size only affects the leave nodes, everything else is fixed.

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 122379dfc7d8737ebf3bf38187bb1994da4b17a9..f1e541e9b514c61c053636de333eecc255c78750 100644 (file)
@@ -138,6 +138,24 @@ struct amdgpu_prt_cb {
        struct dma_fence_cb cb;
 };
 
+/**
+ * amdgpu_vm_level_shift - return the addr shift for each level
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Returns the number of bits the pfn needs to be right shifted for a level.
+ */
+static unsigned amdgpu_vm_level_shift(struct amdgpu_device *adev,
+                                     unsigned level)
+{
+       if (level != adev->vm_manager.num_level)
+               return 9 * (adev->vm_manager.num_level - level - 1) +
+                       adev->vm_manager.block_size;
+       else
+               /* For the page tables on the leaves */
+               return 0;
+}
+
 /**
  * amdgpu_vm_num_entries - return the number of entries in a PD/PT
  *
@@ -288,8 +306,7 @@ static int amdgpu_vm_alloc_levels(struct amdgpu_device *adev,
                                  uint64_t saddr, uint64_t eaddr,
                                  unsigned level)
 {
-       unsigned shift = (adev->vm_manager.num_level - level) *
-               adev->vm_manager.block_size;
+       unsigned shift = amdgpu_vm_level_shift(adev, level);
        unsigned pt_idx, from, to;
        int r;
        u64 flags;
@@ -1302,18 +1319,19 @@ void amdgpu_vm_get_entry(struct amdgpu_pte_update_params *p, uint64_t addr,
                         struct amdgpu_vm_pt **entry,
                         struct amdgpu_vm_pt **parent)
 {
-       unsigned idx, level = p->adev->vm_manager.num_level;
+       unsigned level = 0;
 
        *parent = NULL;
        *entry = &p->vm->root;
        while ((*entry)->entries) {
-               idx = addr >> (p->adev->vm_manager.block_size * level--);
+               unsigned idx = addr >> amdgpu_vm_level_shift(p->adev, level++);
+
                idx %= amdgpu_bo_size((*entry)->base.bo) / 8;
                *parent = *entry;
                *entry = &(*entry)->entries[idx];
        }
 
-       if (level)
+       if (level != p->adev->vm_manager.num_level)
                *entry = NULL;
 }