Looks like a better place for this.
v2: use atomic64_t members instead
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>
struct amdgpu_mman mman;
struct amdgpu_vram_scratch vram_scratch;
struct amdgpu_wb wb;
- atomic64_t vram_usage;
- atomic64_t vram_vis_usage;
atomic64_t num_bytes_moved;
atomic64_t num_evictions;
atomic64_t num_vram_cpu_page_faults;
}
total_vram = adev->mc.real_vram_size - adev->vram_pin_size;
- used_vram = atomic64_read(&adev->vram_usage);
+ used_vram = amdgpu_vram_mgr_usage(&adev->mman.bdev.man[TTM_PL_VRAM]);
free_vram = used_vram >= total_vram ? 0 : total_vram - used_vram;
spin_lock(&adev->mm_stats.lock);
/* Do the same for visible VRAM if half of it is free */
if (adev->mc.visible_vram_size < adev->mc.real_vram_size) {
u64 total_vis_vram = adev->mc.visible_vram_size;
- u64 used_vis_vram = atomic64_read(&adev->vram_vis_usage);
+ u64 used_vis_vram =
+ amdgpu_vram_mgr_vis_usage(&adev->mman.bdev.man[TTM_PL_VRAM]);
if (used_vis_vram < total_vis_vram) {
u64 free_vis_vram = total_vis_vram - used_vis_vram;
ui64 = atomic64_read(&adev->num_vram_cpu_page_faults);
return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0;
case AMDGPU_INFO_VRAM_USAGE:
- ui64 = atomic64_read(&adev->vram_usage);
+ ui64 = amdgpu_vram_mgr_usage(&adev->mman.bdev.man[TTM_PL_VRAM]);
return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0;
case AMDGPU_INFO_VIS_VRAM_USAGE:
- ui64 = atomic64_read(&adev->vram_vis_usage);
+ ui64 = amdgpu_vram_mgr_vis_usage(&adev->mman.bdev.man[TTM_PL_VRAM]);
return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0;
case AMDGPU_INFO_GTT_USAGE:
ui64 = amdgpu_gtt_mgr_usage(&adev->mman.bdev.man[TTM_PL_TT]);
mem.vram.total_heap_size = adev->mc.real_vram_size;
mem.vram.usable_heap_size =
adev->mc.real_vram_size - adev->vram_pin_size;
- mem.vram.heap_usage = atomic64_read(&adev->vram_usage);
+ mem.vram.heap_usage =
+ amdgpu_vram_mgr_usage(&adev->mman.bdev.man[TTM_PL_VRAM]);
mem.vram.max_allocation = mem.vram.usable_heap_size * 3 / 4;
mem.cpu_accessible_vram.total_heap_size =
adev->mc.visible_vram_size -
(adev->vram_pin_size - adev->invisible_pin_size);
mem.cpu_accessible_vram.heap_usage =
- atomic64_read(&adev->vram_vis_usage);
+ amdgpu_vram_mgr_vis_usage(&adev->mman.bdev.man[TTM_PL_VRAM]);
mem.cpu_accessible_vram.max_allocation =
mem.cpu_accessible_vram.usable_heap_size * 3 / 4;
#include "amdgpu.h"
#include "amdgpu_trace.h"
-
-
-static u64 amdgpu_get_vis_part_size(struct amdgpu_device *adev,
- struct ttm_mem_reg *mem)
-{
- 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,
- struct ttm_mem_reg *old_mem,
- struct ttm_mem_reg *new_mem)
-{
- u64 vis_size;
- if (!adev)
- return;
-
- if (new_mem) {
- switch (new_mem->mem_type) {
- case TTM_PL_TT:
- break;
- case TTM_PL_VRAM:
- atomic64_add(new_mem->size, &adev->vram_usage);
- vis_size = amdgpu_get_vis_part_size(adev, new_mem);
- atomic64_add(vis_size, &adev->vram_vis_usage);
- break;
- }
- }
-
- if (old_mem) {
- switch (old_mem->mem_type) {
- case TTM_PL_TT:
- break;
- case TTM_PL_VRAM:
- atomic64_sub(old_mem->size, &adev->vram_usage);
- vis_size = amdgpu_get_vis_part_size(adev, old_mem);
- atomic64_sub(vis_size, &adev->vram_vis_usage);
- break;
- }
- }
-}
-
static void amdgpu_ttm_bo_destroy(struct ttm_buffer_object *tbo)
{
struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
bo = container_of(tbo, struct amdgpu_bo, tbo);
amdgpu_bo_kunmap(bo);
- amdgpu_update_memory_usage(adev, &bo->tbo.mem, NULL);
drm_gem_object_release(&bo->gem_base);
amdgpu_bo_unref(&bo->parent);
return;
/* move_notify is called before move happens */
- amdgpu_update_memory_usage(adev, &bo->mem, new_mem);
-
trace_amdgpu_ttm_bo_move(abo, new_mem->mem_type, old_mem->mem_type);
}
struct ttm_mem_reg *mem);
uint64_t amdgpu_gtt_mgr_usage(struct ttm_mem_type_manager *man);
+uint64_t amdgpu_vram_mgr_usage(struct ttm_mem_type_manager *man);
+uint64_t amdgpu_vram_mgr_vis_usage(struct ttm_mem_type_manager *man);
+
int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset,
uint64_t dst_offset, uint32_t byte_count,
struct reservation_object *resv,
struct amdgpu_vram_mgr {
struct drm_mm mm;
spinlock_t lock;
+ atomic64_t usage;
+ atomic64_t vis_usage;
};
/**
}
/**
+ * amdgpu_vram_mgr_vis_size - Calculate visible node size
+ *
+ * @adev: amdgpu device structure
+ * @node: MM node structure
+ *
+ * Calculate how many bytes of the MM node are inside visible VRAM
+ */
+static u64 amdgpu_vram_mgr_vis_size(struct amdgpu_device *adev,
+ struct drm_mm_node *node)
+{
+ uint64_t start = node->start << PAGE_SHIFT;
+ uint64_t end = (node->size + node->start) << PAGE_SHIFT;
+
+ if (start >= adev->mc.visible_vram_size)
+ return 0;
+
+ return (end > adev->mc.visible_vram_size ?
+ adev->mc.visible_vram_size : end) - start;
+}
+
+/**
* amdgpu_vram_mgr_new - allocate new ranges
*
* @man: TTM memory type manager
const struct ttm_place *place,
struct ttm_mem_reg *mem)
{
+ struct amdgpu_device *adev = amdgpu_ttm_adev(man->bdev);
struct amdgpu_vram_mgr *mgr = man->priv;
struct drm_mm *mm = &mgr->mm;
struct drm_mm_node *nodes;
enum drm_mm_insert_mode mode;
unsigned long lpfn, num_nodes, pages_per_node, pages_left;
+ uint64_t usage = 0, vis_usage = 0;
unsigned i;
int r;
if (unlikely(r))
goto error;
+ usage += nodes[i].size << PAGE_SHIFT;
+ vis_usage += amdgpu_vram_mgr_vis_size(adev, &nodes[i]);
+
/* Calculate a virtual BO start address to easily check if
* everything is CPU accessible.
*/
}
spin_unlock(&mgr->lock);
+ atomic64_add(usage, &mgr->usage);
+ atomic64_add(vis_usage, &mgr->vis_usage);
+
mem->mm_node = nodes;
return 0;
static void amdgpu_vram_mgr_del(struct ttm_mem_type_manager *man,
struct ttm_mem_reg *mem)
{
+ struct amdgpu_device *adev = amdgpu_ttm_adev(man->bdev);
struct amdgpu_vram_mgr *mgr = man->priv;
struct drm_mm_node *nodes = mem->mm_node;
+ uint64_t usage = 0, vis_usage = 0;
unsigned pages = mem->num_pages;
if (!mem->mm_node)
while (pages) {
pages -= nodes->size;
drm_mm_remove_node(nodes);
+ usage += nodes->size << PAGE_SHIFT;
+ vis_usage += amdgpu_vram_mgr_vis_size(adev, nodes);
++nodes;
}
spin_unlock(&mgr->lock);
+ atomic64_sub(usage, &mgr->usage);
+ atomic64_sub(vis_usage, &mgr->vis_usage);
+
kfree(mem->mm_node);
mem->mm_node = NULL;
}
/**
+ * amdgpu_vram_mgr_usage - how many bytes are used in this domain
+ *
+ * @man: TTM memory type manager
+ *
+ * Returns how many bytes are used in this domain.
+ */
+uint64_t amdgpu_vram_mgr_usage(struct ttm_mem_type_manager *man)
+{
+ struct amdgpu_vram_mgr *mgr = man->priv;
+
+ return atomic64_read(&mgr->usage);
+}
+
+/**
+ * amdgpu_vram_mgr_vis_usage - how many bytes are used in the visible part
+ *
+ * @man: TTM memory type manager
+ *
+ * Returns how many bytes are used in the visible part of VRAM
+ */
+uint64_t amdgpu_vram_mgr_vis_usage(struct ttm_mem_type_manager *man)
+{
+ struct amdgpu_vram_mgr *mgr = man->priv;
+
+ return atomic64_read(&mgr->vis_usage);
+}
+
+/**
* amdgpu_vram_mgr_debug - dump VRAM table
*
* @man: TTM memory type manager
static void amdgpu_vram_mgr_debug(struct ttm_mem_type_manager *man,
struct drm_printer *printer)
{
- struct amdgpu_device *adev = amdgpu_ttm_adev(man->bdev);
struct amdgpu_vram_mgr *mgr = man->priv;
spin_lock(&mgr->lock);
spin_unlock(&mgr->lock);
drm_printf(printer, "man size:%llu pages, ram usage:%lluMB, vis usage:%lluMB\n",
- adev->mman.bdev.man[TTM_PL_VRAM].size,
- (u64)atomic64_read(&adev->vram_usage) >> 20,
- (u64)atomic64_read(&adev->vram_vis_usage) >> 20);
+ man->size, amdgpu_vram_mgr_usage(man) >> 20,
+ amdgpu_vram_mgr_vis_usage(man) >> 20);
}
const struct ttm_mem_type_manager_func amdgpu_vram_mgr_func = {
OSDN Git Service
