void amdgpu_ttm_fini(struct amdgpu_device *adev);
static u64 amdgpu_get_vis_part_size(struct amdgpu_device *adev,
- struct ttm_mem_reg * mem)
+ 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):
+ adev->mc.visible_vram_size - (mem->start << PAGE_SHIFT) :
mem->size;
}
return ret;
bo = container_of(tbo, struct amdgpu_bo, tbo);
amdgpu_update_memory_usage(bo->adev, &bo->tbo.mem, NULL);
- amdgpu_mn_unregister(bo);
mutex_lock(&bo->adev->gem.mutex);
list_del_init(&bo->list);
return false;
}
-void amdgpu_ttm_placement_from_domain(struct amdgpu_bo *rbo, u32 domain)
+static void amdgpu_ttm_placement_init(struct amdgpu_device *adev,
+ struct ttm_placement *placement,
+ struct ttm_place *placements,
+ u32 domain, u64 flags)
{
u32 c = 0, i;
- rbo->placement.placement = rbo->placements;
- rbo->placement.busy_placement = rbo->placements;
+
+ placement->placement = placements;
+ placement->busy_placement = placements;
if (domain & AMDGPU_GEM_DOMAIN_VRAM) {
- if (rbo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS &&
- rbo->adev->mc.visible_vram_size < rbo->adev->mc.real_vram_size) {
- rbo->placements[c].fpfn =
- rbo->adev->mc.visible_vram_size >> PAGE_SHIFT;
- rbo->placements[c++].flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED |
- TTM_PL_FLAG_VRAM;
+ if (flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS &&
+ 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;
}
- rbo->placements[c].fpfn = 0;
- rbo->placements[c++].flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED |
- TTM_PL_FLAG_VRAM;
+ placements[c].fpfn = 0;
+ placements[c++].flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED |
+ TTM_PL_FLAG_VRAM;
}
if (domain & AMDGPU_GEM_DOMAIN_GTT) {
- if (rbo->flags & AMDGPU_GEM_CREATE_CPU_GTT_USWC) {
- rbo->placements[c].fpfn = 0;
- rbo->placements[c++].flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_TT |
- TTM_PL_FLAG_UNCACHED;
+ 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;
} else {
- rbo->placements[c].fpfn = 0;
- rbo->placements[c++].flags = TTM_PL_FLAG_CACHED | TTM_PL_FLAG_TT;
+ placements[c].fpfn = 0;
+ placements[c++].flags = TTM_PL_FLAG_CACHED | TTM_PL_FLAG_TT;
}
}
if (domain & AMDGPU_GEM_DOMAIN_CPU) {
- if (rbo->flags & AMDGPU_GEM_CREATE_CPU_GTT_USWC) {
- rbo->placements[c].fpfn = 0;
- rbo->placements[c++].flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_SYSTEM |
- TTM_PL_FLAG_UNCACHED;
+ 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;
} else {
- rbo->placements[c].fpfn = 0;
- rbo->placements[c++].flags = TTM_PL_FLAG_CACHED | TTM_PL_FLAG_SYSTEM;
+ placements[c].fpfn = 0;
+ placements[c++].flags = TTM_PL_FLAG_CACHED | TTM_PL_FLAG_SYSTEM;
}
}
if (domain & AMDGPU_GEM_DOMAIN_GDS) {
- rbo->placements[c++].flags = TTM_PL_FLAG_UNCACHED |
- AMDGPU_PL_FLAG_GDS;
+ placements[c].fpfn = 0;
+ placements[c++].flags = TTM_PL_FLAG_UNCACHED |
+ AMDGPU_PL_FLAG_GDS;
}
if (domain & AMDGPU_GEM_DOMAIN_GWS) {
- rbo->placements[c++].flags = TTM_PL_FLAG_UNCACHED |
- AMDGPU_PL_FLAG_GWS;
+ placements[c].fpfn = 0;
+ placements[c++].flags = TTM_PL_FLAG_UNCACHED |
+ AMDGPU_PL_FLAG_GWS;
}
if (domain & AMDGPU_GEM_DOMAIN_OA) {
- rbo->placements[c++].flags = TTM_PL_FLAG_UNCACHED |
- AMDGPU_PL_FLAG_OA;
+ placements[c].fpfn = 0;
+ placements[c++].flags = TTM_PL_FLAG_UNCACHED |
+ AMDGPU_PL_FLAG_OA;
}
if (!c) {
- rbo->placements[c].fpfn = 0;
- rbo->placements[c++].flags = TTM_PL_MASK_CACHING |
- TTM_PL_FLAG_SYSTEM;
+ placements[c].fpfn = 0;
+ placements[c++].flags = TTM_PL_MASK_CACHING |
+ TTM_PL_FLAG_SYSTEM;
}
- rbo->placement.num_placement = c;
- rbo->placement.num_busy_placement = c;
+ placement->num_placement = c;
+ placement->num_busy_placement = c;
for (i = 0; i < c; i++) {
- if ((rbo->flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED) &&
- (rbo->placements[i].flags & TTM_PL_FLAG_VRAM) &&
- !rbo->placements[i].fpfn)
- rbo->placements[i].lpfn =
- rbo->adev->mc.visible_vram_size >> PAGE_SHIFT;
+ if ((flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED) &&
+ (placements[i].flags & TTM_PL_FLAG_VRAM) &&
+ !placements[i].fpfn)
+ placements[i].lpfn =
+ adev->mc.visible_vram_size >> PAGE_SHIFT;
else
- rbo->placements[i].lpfn = 0;
+ placements[i].lpfn = 0;
}
}
-int amdgpu_bo_create(struct amdgpu_device *adev,
- unsigned long size, int byte_align, bool kernel, u32 domain, u64 flags,
- struct sg_table *sg, struct amdgpu_bo **bo_ptr)
+void amdgpu_ttm_placement_from_domain(struct amdgpu_bo *rbo, u32 domain)
+{
+ amdgpu_ttm_placement_init(rbo->adev, &rbo->placement,
+ rbo->placements, domain, rbo->flags);
+}
+
+static void amdgpu_fill_placement_to_bo(struct amdgpu_bo *bo,
+ struct ttm_placement *placement)
+{
+ BUG_ON(placement->num_placement > (AMDGPU_GEM_DOMAIN_MAX + 1));
+
+ memcpy(bo->placements, placement->placement,
+ placement->num_placement * sizeof(struct ttm_place));
+ bo->placement.num_placement = placement->num_placement;
+ bo->placement.num_busy_placement = placement->num_busy_placement;
+ bo->placement.placement = bo->placements;
+ bo->placement.busy_placement = bo->placements;
+}
+
+int amdgpu_bo_create_restricted(struct amdgpu_device *adev,
+ unsigned long size, int byte_align,
+ bool kernel, u32 domain, u64 flags,
+ struct sg_table *sg,
+ struct ttm_placement *placement,
+ struct reservation_object *resv,
+ struct amdgpu_bo **bo_ptr)
{
struct amdgpu_bo *bo;
enum ttm_bo_type type;
size_t acc_size;
int r;
- /* VI has a hw bug where VM PTEs have to be allocated in groups of 8.
- * do this as a temporary workaround
- */
- if (!(domain & (AMDGPU_GEM_DOMAIN_GDS | AMDGPU_GEM_DOMAIN_GWS | AMDGPU_GEM_DOMAIN_OA))) {
- if (adev->asic_type >= CHIP_TOPAZ) {
- if (byte_align & 0x7fff)
- byte_align = ALIGN(byte_align, 0x8000);
- if (size & 0x7fff)
- size = ALIGN(size, 0x8000);
- }
- }
-
page_align = roundup(byte_align, PAGE_SIZE) >> PAGE_SHIFT;
size = ALIGN(size, PAGE_SIZE);
AMDGPU_GEM_DOMAIN_OA);
bo->flags = flags;
- amdgpu_ttm_placement_from_domain(bo, domain);
+ amdgpu_fill_placement_to_bo(bo, placement);
/* Kernel allocation are uninterruptible */
- down_read(&adev->pm.mclk_lock);
r = ttm_bo_init(&adev->mman.bdev, &bo->tbo, size, type,
&bo->placement, page_align, !kernel, NULL,
- acc_size, sg, NULL, &amdgpu_ttm_bo_destroy);
- up_read(&adev->pm.mclk_lock);
+ acc_size, sg, resv, &amdgpu_ttm_bo_destroy);
if (unlikely(r != 0)) {
return r;
}
return 0;
}
+int amdgpu_bo_create(struct amdgpu_device *adev,
+ unsigned long size, int byte_align,
+ bool kernel, u32 domain, u64 flags,
+ struct sg_table *sg,
+ struct reservation_object *resv,
+ struct amdgpu_bo **bo_ptr)
+{
+ struct ttm_placement placement = {0};
+ struct ttm_place placements[AMDGPU_GEM_DOMAIN_MAX + 1];
+
+ memset(&placements, 0,
+ (AMDGPU_GEM_DOMAIN_MAX + 1) * sizeof(struct ttm_place));
+
+ amdgpu_ttm_placement_init(adev, &placement,
+ placements, domain, flags);
+
+ return amdgpu_bo_create_restricted(adev, size, byte_align, kernel,
+ domain, flags, sg, &placement,
+ resv, bo_ptr);
+}
+
int amdgpu_bo_kmap(struct amdgpu_bo *bo, void **ptr)
{
bool is_iomem;
*bo = NULL;
}
-int amdgpu_bo_pin_restricted(struct amdgpu_bo *bo, u32 domain, u64 max_offset,
+int amdgpu_bo_pin_restricted(struct amdgpu_bo *bo, u32 domain,
+ u64 min_offset, u64 max_offset,
u64 *gpu_addr)
{
int r, i;
+ unsigned fpfn, lpfn;
if (amdgpu_ttm_tt_has_userptr(bo->tbo.ttm))
return -EPERM;
+ if (WARN_ON_ONCE(min_offset > max_offset))
+ return -EINVAL;
+
if (bo->pin_count) {
bo->pin_count++;
if (gpu_addr)
if (max_offset != 0) {
u64 domain_start;
-
if (domain == AMDGPU_GEM_DOMAIN_VRAM)
domain_start = bo->adev->mc.vram_start;
else
for (i = 0; i < bo->placement.num_placement; i++) {
/* 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))
- bo->placements[i].lpfn =
- bo->adev->mc.visible_vram_size >> PAGE_SHIFT;
- else
- bo->placements[i].lpfn = max_offset >> PAGE_SHIFT;
-
+ !(bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS) &&
+ (!max_offset || max_offset > bo->adev->mc.visible_vram_size)) {
+ if (WARN_ON_ONCE(min_offset >
+ bo->adev->mc.visible_vram_size))
+ return -EINVAL;
+ fpfn = min_offset >> PAGE_SHIFT;
+ lpfn = bo->adev->mc.visible_vram_size >> PAGE_SHIFT;
+ } else {
+ fpfn = min_offset >> PAGE_SHIFT;
+ lpfn = max_offset >> PAGE_SHIFT;
+ }
+ if (fpfn > bo->placements[i].fpfn)
+ bo->placements[i].fpfn = fpfn;
+ if (lpfn && lpfn < bo->placements[i].lpfn)
+ bo->placements[i].lpfn = lpfn;
bo->placements[i].flags |= TTM_PL_FLAG_NO_EVICT;
}
int amdgpu_bo_pin(struct amdgpu_bo *bo, u32 domain, u64 *gpu_addr)
{
- return amdgpu_bo_pin_restricted(bo, domain, 0, gpu_addr);
+ return amdgpu_bo_pin_restricted(bo, domain, 0, 0, gpu_addr);
}
int amdgpu_bo_unpin(struct amdgpu_bo *bo)
int amdgpu_bo_evict_vram(struct amdgpu_device *adev)
{
/* late 2.6.33 fix IGP hibernate - we need pm ops to do this correct */
- if (0 && (adev->flags & AMDGPU_IS_APU)) {
+ if (0 && (adev->flags & AMD_IS_APU)) {
/* Useless to evict on IGP chips */
return 0;
}
}
dev_err(adev->dev, "Userspace still has active objects !\n");
list_for_each_entry_safe(bo, n, &adev->gem.objects, list) {
- mutex_lock(&adev->ddev->struct_mutex);
dev_err(adev->dev, "%p %p %lu %lu force free\n",
&bo->gem_base, bo, (unsigned long)bo->gem_base.size,
*((unsigned long *)&bo->gem_base.refcount));
list_del_init(&bo->list);
mutex_unlock(&bo->adev->gem.mutex);
/* this should unref the ttm bo */
- drm_gem_object_unreference(&bo->gem_base);
- mutex_unlock(&adev->ddev->struct_mutex);
+ drm_gem_object_unreference_unlocked(&bo->gem_base);
}
}
* @shared: true if fence should be added shared
*
*/
-void amdgpu_bo_fence(struct amdgpu_bo *bo, struct amdgpu_fence *fence,
+void amdgpu_bo_fence(struct amdgpu_bo *bo, struct fence *fence,
bool shared)
{
struct reservation_object *resv = bo->tbo.resv;
if (shared)
- reservation_object_add_shared_fence(resv, &fence->base);
+ reservation_object_add_shared_fence(resv, fence);
else
- reservation_object_add_excl_fence(resv, &fence->base);
+ reservation_object_add_excl_fence(resv, fence);
}
projects / deliverable / linux.git / blobdiff