]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/commitdiff
drm/amdgpu: fix coding style in amdgpu_object.c
authorChristian König <christian.koenig@amd.com>
Fri, 12 Aug 2016 14:50:12 +0000 (16:50 +0200)
committerAlex Deucher <alexander.deucher@amd.com>
Tue, 16 Aug 2016 14:44:55 +0000 (10:44 -0400)
Just a few 80 chars problems.

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

index 278017d3dc3fe45e6fc0681211157f099e53bb14..4d084eeb4e09fe2677cefa9c23c423b42a84d4b8 100644 (file)
@@ -44,14 +44,13 @@ void amdgpu_ttm_fini(struct amdgpu_device *adev);
 static u64 amdgpu_get_vis_part_size(struct amdgpu_device *adev,
                                                struct ttm_mem_reg *mem)
 {
-       u64 ret = 0;
-       if (mem->start << PAGE_SHIFT < adev->mc.visible_vram_size) {
-               ret = (u64)((mem->start << PAGE_SHIFT) + mem->size) >
-                          adev->mc.visible_vram_size ?
-                          adev->mc.visible_vram_size - (mem->start << PAGE_SHIFT) :
-                          mem->size;
-       }
-       return ret;
+       if (mem->start << PAGE_SHIFT >= adev->mc.visible_vram_size)
+               return 0;
+
+       return ((mem->start << PAGE_SHIFT) + mem->size) >
+               adev->mc.visible_vram_size ?
+               adev->mc.visible_vram_size - (mem->start << PAGE_SHIFT) :
+               mem->size;
 }
 
 static void amdgpu_update_memory_usage(struct amdgpu_device *adev,
@@ -125,8 +124,9 @@ static void amdgpu_ttm_placement_init(struct amdgpu_device *adev,
                        adev->mc.visible_vram_size < adev->mc.real_vram_size) {
                        placements[c].fpfn =
                                adev->mc.visible_vram_size >> PAGE_SHIFT;
-                       placements[c++].flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED |
-                               TTM_PL_FLAG_VRAM | TTM_PL_FLAG_TOPDOWN;
+                       placements[c++].flags = TTM_PL_FLAG_WC |
+                               TTM_PL_FLAG_UNCACHED | TTM_PL_FLAG_VRAM |
+                               TTM_PL_FLAG_TOPDOWN;
                }
                placements[c].fpfn = 0;
                placements[c++].flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED |
@@ -138,22 +138,24 @@ static void amdgpu_ttm_placement_init(struct amdgpu_device *adev,
        if (domain & AMDGPU_GEM_DOMAIN_GTT) {
                if (flags & AMDGPU_GEM_CREATE_CPU_GTT_USWC) {
                        placements[c].fpfn = 0;
-                       placements[c++].flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_TT |
-                               TTM_PL_FLAG_UNCACHED;
+                       placements[c++].flags = TTM_PL_FLAG_WC |
+                               TTM_PL_FLAG_TT | TTM_PL_FLAG_UNCACHED;
                } else {
                        placements[c].fpfn = 0;
-                       placements[c++].flags = TTM_PL_FLAG_CACHED | TTM_PL_FLAG_TT;
+                       placements[c++].flags = TTM_PL_FLAG_CACHED |
+                               TTM_PL_FLAG_TT;
                }
        }
 
        if (domain & AMDGPU_GEM_DOMAIN_CPU) {
                if (flags & AMDGPU_GEM_CREATE_CPU_GTT_USWC) {
                        placements[c].fpfn = 0;
-                       placements[c++].flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_SYSTEM |
-                               TTM_PL_FLAG_UNCACHED;
+                       placements[c++].flags = TTM_PL_FLAG_WC |
+                               TTM_PL_FLAG_SYSTEM | TTM_PL_FLAG_UNCACHED;
                } else {
                        placements[c].fpfn = 0;
-                       placements[c++].flags = TTM_PL_FLAG_CACHED | TTM_PL_FLAG_SYSTEM;
+                       placements[c++].flags = TTM_PL_FLAG_CACHED |
+                               TTM_PL_FLAG_SYSTEM;
                }
        }
 
@@ -539,7 +541,8 @@ int amdgpu_bo_pin_restricted(struct amdgpu_bo *bo, u32 domain,
                /* force to pin into visible video ram */
                if ((bo->placements[i].flags & TTM_PL_FLAG_VRAM) &&
                    !(bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS) &&
-                   (!max_offset || max_offset > bo->adev->mc.visible_vram_size)) {
+                   (!max_offset || max_offset >
+                    bo->adev->mc.visible_vram_size)) {
                        if (WARN_ON_ONCE(min_offset >
                                         bo->adev->mc.visible_vram_size))
                                return -EINVAL;
@@ -558,19 +561,23 @@ int amdgpu_bo_pin_restricted(struct amdgpu_bo *bo, u32 domain,
        }
 
        r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
-       if (likely(r == 0)) {
-               bo->pin_count = 1;
-               if (gpu_addr != NULL)
-                       *gpu_addr = amdgpu_bo_gpu_offset(bo);
-               if (domain == AMDGPU_GEM_DOMAIN_VRAM) {
-                       bo->adev->vram_pin_size += amdgpu_bo_size(bo);
-                       if (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS)
-                               bo->adev->invisible_pin_size += amdgpu_bo_size(bo);
-               } else
-                       bo->adev->gart_pin_size += amdgpu_bo_size(bo);
-       } else {
+       if (unlikely(r)) {
                dev_err(bo->adev->dev, "%p pin failed\n", bo);
+               goto error;
+       }
+
+       bo->pin_count = 1;
+       if (gpu_addr != NULL)
+               *gpu_addr = amdgpu_bo_gpu_offset(bo);
+       if (domain == AMDGPU_GEM_DOMAIN_VRAM) {
+               bo->adev->vram_pin_size += amdgpu_bo_size(bo);
+               if (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS)
+                       bo->adev->invisible_pin_size += amdgpu_bo_size(bo);
+       } else {
+               bo->adev->gart_pin_size += amdgpu_bo_size(bo);
        }
+
+error:
        return r;
 }
 
@@ -595,16 +602,20 @@ int amdgpu_bo_unpin(struct amdgpu_bo *bo)
                bo->placements[i].flags &= ~TTM_PL_FLAG_NO_EVICT;
        }
        r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
-       if (likely(r == 0)) {
-               if (bo->tbo.mem.mem_type == TTM_PL_VRAM) {
-                       bo->adev->vram_pin_size -= amdgpu_bo_size(bo);
-                       if (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS)
-                               bo->adev->invisible_pin_size -= amdgpu_bo_size(bo);
-               } else
-                       bo->adev->gart_pin_size -= amdgpu_bo_size(bo);
-       } else {
+       if (unlikely(r)) {
                dev_err(bo->adev->dev, "%p validate failed for unpin\n", bo);
+               goto error;
        }
+
+       if (bo->tbo.mem.mem_type == TTM_PL_VRAM) {
+               bo->adev->vram_pin_size -= amdgpu_bo_size(bo);
+               if (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS)
+                       bo->adev->invisible_pin_size -= amdgpu_bo_size(bo);
+       } else {
+               bo->adev->gart_pin_size -= amdgpu_bo_size(bo);
+       }
+
+error:
        return r;
 }
 
@@ -775,7 +786,8 @@ int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo)
        for (i = 0; i < abo->placement.num_placement; i++) {
                /* Force into visible VRAM */
                if ((abo->placements[i].flags & TTM_PL_FLAG_VRAM) &&
-                   (!abo->placements[i].lpfn || abo->placements[i].lpfn > lpfn))
+                   (!abo->placements[i].lpfn ||
+                    abo->placements[i].lpfn > lpfn))
                        abo->placements[i].lpfn = lpfn;
        }
        r = ttm_bo_validate(bo, &abo->placement, false, false);