summaryrefslogtreecommitdiff
path: root/drivers/gpu/drm/ttm/ttm_bo.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/gpu/drm/ttm/ttm_bo.c')
-rw-r--r--drivers/gpu/drm/ttm/ttm_bo.c159
1 files changed, 100 insertions, 59 deletions
diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 148a322d8f5d..af61fc29e843 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -169,7 +169,7 @@ int ttm_bo_wait_unreserved(struct ttm_buffer_object *bo, bool interruptible)
}
EXPORT_SYMBOL(ttm_bo_wait_unreserved);
-static void ttm_bo_add_to_lru(struct ttm_buffer_object *bo)
+void ttm_bo_add_to_lru(struct ttm_buffer_object *bo)
{
struct ttm_bo_device *bdev = bo->bdev;
struct ttm_mem_type_manager *man;
@@ -191,11 +191,7 @@ static void ttm_bo_add_to_lru(struct ttm_buffer_object *bo)
}
}
-/**
- * Call with the lru_lock held.
- */
-
-static int ttm_bo_del_from_lru(struct ttm_buffer_object *bo)
+int ttm_bo_del_from_lru(struct ttm_buffer_object *bo)
{
int put_count = 0;
@@ -227,9 +223,18 @@ int ttm_bo_reserve_locked(struct ttm_buffer_object *bo,
/**
* Deadlock avoidance for multi-bo reserving.
*/
- if (use_sequence && bo->seq_valid &&
- (sequence - bo->val_seq < (1 << 31))) {
- return -EAGAIN;
+ if (use_sequence && bo->seq_valid) {
+ /**
+ * We've already reserved this one.
+ */
+ if (unlikely(sequence == bo->val_seq))
+ return -EDEADLK;
+ /**
+ * Already reserved by a thread that will not back
+ * off for us. We need to back off.
+ */
+ if (unlikely(sequence - bo->val_seq < (1 << 31)))
+ return -EAGAIN;
}
if (no_wait)
@@ -267,6 +272,13 @@ static void ttm_bo_ref_bug(struct kref *list_kref)
BUG();
}
+void ttm_bo_list_ref_sub(struct ttm_buffer_object *bo, int count,
+ bool never_free)
+{
+ kref_sub(&bo->list_kref, count,
+ (never_free) ? ttm_bo_ref_bug : ttm_bo_release_list);
+}
+
int ttm_bo_reserve(struct ttm_buffer_object *bo,
bool interruptible,
bool no_wait, bool use_sequence, uint32_t sequence)
@@ -282,20 +294,24 @@ int ttm_bo_reserve(struct ttm_buffer_object *bo,
put_count = ttm_bo_del_from_lru(bo);
spin_unlock(&glob->lru_lock);
- while (put_count--)
- kref_put(&bo->list_kref, ttm_bo_ref_bug);
+ ttm_bo_list_ref_sub(bo, put_count, true);
return ret;
}
+void ttm_bo_unreserve_locked(struct ttm_buffer_object *bo)
+{
+ ttm_bo_add_to_lru(bo);
+ atomic_set(&bo->reserved, 0);
+ wake_up_all(&bo->event_queue);
+}
+
void ttm_bo_unreserve(struct ttm_buffer_object *bo)
{
struct ttm_bo_global *glob = bo->glob;
spin_lock(&glob->lru_lock);
- ttm_bo_add_to_lru(bo);
- atomic_set(&bo->reserved, 0);
- wake_up_all(&bo->event_queue);
+ ttm_bo_unreserve_locked(bo);
spin_unlock(&glob->lru_lock);
}
EXPORT_SYMBOL(ttm_bo_unreserve);
@@ -362,8 +378,13 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
int ret = 0;
if (old_is_pci || new_is_pci ||
- ((mem->placement & bo->mem.placement & TTM_PL_MASK_CACHING) == 0))
- ttm_bo_unmap_virtual(bo);
+ ((mem->placement & bo->mem.placement & TTM_PL_MASK_CACHING) == 0)) {
+ ret = ttm_mem_io_lock(old_man, true);
+ if (unlikely(ret != 0))
+ goto out_err;
+ ttm_bo_unmap_virtual_locked(bo);
+ ttm_mem_io_unlock(old_man);
+ }
/*
* Create and bind a ttm if required.
@@ -416,11 +437,9 @@ moved:
}
if (bo->mem.mm_node) {
- spin_lock(&bo->lock);
bo->offset = (bo->mem.start << PAGE_SHIFT) +
bdev->man[bo->mem.mem_type].gpu_offset;
bo->cur_placement = bo->mem.placement;
- spin_unlock(&bo->lock);
} else
bo->offset = 0;
@@ -452,7 +471,6 @@ static void ttm_bo_cleanup_memtype_use(struct ttm_buffer_object *bo)
ttm_tt_destroy(bo->ttm);
bo->ttm = NULL;
}
-
ttm_bo_mem_put(bo, &bo->mem);
atomic_set(&bo->reserved, 0);
@@ -474,14 +492,14 @@ static void ttm_bo_cleanup_refs_or_queue(struct ttm_buffer_object *bo)
int put_count;
int ret;
- spin_lock(&bo->lock);
+ spin_lock(&bdev->fence_lock);
(void) ttm_bo_wait(bo, false, false, true);
if (!bo->sync_obj) {
spin_lock(&glob->lru_lock);
/**
- * Lock inversion between bo::reserve and bo::lock here,
+ * Lock inversion between bo:reserve and bdev::fence_lock here,
* but that's OK, since we're only trylocking.
*/
@@ -490,14 +508,13 @@ static void ttm_bo_cleanup_refs_or_queue(struct ttm_buffer_object *bo)
if (unlikely(ret == -EBUSY))
goto queue;
- spin_unlock(&bo->lock);
+ spin_unlock(&bdev->fence_lock);
put_count = ttm_bo_del_from_lru(bo);
spin_unlock(&glob->lru_lock);
ttm_bo_cleanup_memtype_use(bo);
- while (put_count--)
- kref_put(&bo->list_kref, ttm_bo_ref_bug);
+ ttm_bo_list_ref_sub(bo, put_count, true);
return;
} else {
@@ -512,7 +529,7 @@ queue:
kref_get(&bo->list_kref);
list_add_tail(&bo->ddestroy, &bdev->ddestroy);
spin_unlock(&glob->lru_lock);
- spin_unlock(&bo->lock);
+ spin_unlock(&bdev->fence_lock);
if (sync_obj) {
driver->sync_obj_flush(sync_obj, sync_obj_arg);
@@ -537,14 +554,15 @@ static int ttm_bo_cleanup_refs(struct ttm_buffer_object *bo,
bool no_wait_reserve,
bool no_wait_gpu)
{
+ struct ttm_bo_device *bdev = bo->bdev;
struct ttm_bo_global *glob = bo->glob;
int put_count;
int ret = 0;
retry:
- spin_lock(&bo->lock);
+ spin_lock(&bdev->fence_lock);
ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu);
- spin_unlock(&bo->lock);
+ spin_unlock(&bdev->fence_lock);
if (unlikely(ret != 0))
return ret;
@@ -580,8 +598,7 @@ retry:
spin_unlock(&glob->lru_lock);
ttm_bo_cleanup_memtype_use(bo);
- while (put_count--)
- kref_put(&bo->list_kref, ttm_bo_ref_bug);
+ ttm_bo_list_ref_sub(bo, put_count, true);
return 0;
}
@@ -652,6 +669,7 @@ static void ttm_bo_release(struct kref *kref)
struct ttm_buffer_object *bo =
container_of(kref, struct ttm_buffer_object, kref);
struct ttm_bo_device *bdev = bo->bdev;
+ struct ttm_mem_type_manager *man = &bdev->man[bo->mem.mem_type];
if (likely(bo->vm_node != NULL)) {
rb_erase(&bo->vm_rb, &bdev->addr_space_rb);
@@ -659,6 +677,9 @@ static void ttm_bo_release(struct kref *kref)
bo->vm_node = NULL;
}
write_unlock(&bdev->vm_lock);
+ ttm_mem_io_lock(man, false);
+ ttm_mem_io_free_vm(bo);
+ ttm_mem_io_unlock(man);
ttm_bo_cleanup_refs_or_queue(bo);
kref_put(&bo->list_kref, ttm_bo_release_list);
write_lock(&bdev->vm_lock);
@@ -698,9 +719,9 @@ static int ttm_bo_evict(struct ttm_buffer_object *bo, bool interruptible,
struct ttm_placement placement;
int ret = 0;
- spin_lock(&bo->lock);
+ spin_lock(&bdev->fence_lock);
ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu);
- spin_unlock(&bo->lock);
+ spin_unlock(&bdev->fence_lock);
if (unlikely(ret != 0)) {
if (ret != -ERESTARTSYS) {
@@ -715,7 +736,8 @@ static int ttm_bo_evict(struct ttm_buffer_object *bo, bool interruptible,
evict_mem = bo->mem;
evict_mem.mm_node = NULL;
- evict_mem.bus.io_reserved = false;
+ evict_mem.bus.io_reserved_vm = false;
+ evict_mem.bus.io_reserved_count = 0;
placement.fpfn = 0;
placement.lpfn = 0;
@@ -802,8 +824,7 @@ retry:
BUG_ON(ret != 0);
- while (put_count--)
- kref_put(&bo->list_kref, ttm_bo_ref_bug);
+ ttm_bo_list_ref_sub(bo, put_count, true);
ret = ttm_bo_evict(bo, interruptible, no_wait_reserve, no_wait_gpu);
ttm_bo_unreserve(bo);
@@ -1036,6 +1057,7 @@ int ttm_bo_move_buffer(struct ttm_buffer_object *bo,
{
int ret = 0;
struct ttm_mem_reg mem;
+ struct ttm_bo_device *bdev = bo->bdev;
BUG_ON(!atomic_read(&bo->reserved));
@@ -1044,15 +1066,16 @@ int ttm_bo_move_buffer(struct ttm_buffer_object *bo,
* Have the driver move function wait for idle when necessary,
* instead of doing it here.
*/
- spin_lock(&bo->lock);
+ spin_lock(&bdev->fence_lock);
ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu);
- spin_unlock(&bo->lock);
+ spin_unlock(&bdev->fence_lock);
if (ret)
return ret;
mem.num_pages = bo->num_pages;
mem.size = mem.num_pages << PAGE_SHIFT;
mem.page_alignment = bo->mem.page_alignment;
- mem.bus.io_reserved = false;
+ mem.bus.io_reserved_vm = false;
+ mem.bus.io_reserved_count = 0;
/*
* Determine where to move the buffer.
*/
@@ -1163,7 +1186,6 @@ int ttm_bo_init(struct ttm_bo_device *bdev,
}
bo->destroy = destroy;
- spin_lock_init(&bo->lock);
kref_init(&bo->kref);
kref_init(&bo->list_kref);
atomic_set(&bo->cpu_writers, 0);
@@ -1172,6 +1194,7 @@ int ttm_bo_init(struct ttm_bo_device *bdev,
INIT_LIST_HEAD(&bo->lru);
INIT_LIST_HEAD(&bo->ddestroy);
INIT_LIST_HEAD(&bo->swap);
+ INIT_LIST_HEAD(&bo->io_reserve_lru);
bo->bdev = bdev;
bo->glob = bdev->glob;
bo->type = type;
@@ -1181,7 +1204,8 @@ int ttm_bo_init(struct ttm_bo_device *bdev,
bo->mem.num_pages = bo->num_pages;
bo->mem.mm_node = NULL;
bo->mem.page_alignment = page_alignment;
- bo->mem.bus.io_reserved = false;
+ bo->mem.bus.io_reserved_vm = false;
+ bo->mem.bus.io_reserved_count = 0;
bo->buffer_start = buffer_start & PAGE_MASK;
bo->priv_flags = 0;
bo->mem.placement = (TTM_PL_FLAG_SYSTEM | TTM_PL_FLAG_CACHED);
@@ -1355,6 +1379,10 @@ int ttm_bo_init_mm(struct ttm_bo_device *bdev, unsigned type,
BUG_ON(type >= TTM_NUM_MEM_TYPES);
man = &bdev->man[type];
BUG_ON(man->has_type);
+ man->io_reserve_fastpath = true;
+ man->use_io_reserve_lru = false;
+ mutex_init(&man->io_reserve_mutex);
+ INIT_LIST_HEAD(&man->io_reserve_lru);
ret = bdev->driver->init_mem_type(bdev, type, man);
if (ret)
@@ -1472,8 +1500,7 @@ int ttm_bo_device_release(struct ttm_bo_device *bdev)
list_del(&bdev->device_list);
mutex_unlock(&glob->device_list_mutex);
- if (!cancel_delayed_work(&bdev->wq))
- flush_scheduled_work();
+ cancel_delayed_work_sync(&bdev->wq);
while (ttm_bo_delayed_delete(bdev, true))
;
@@ -1527,7 +1554,8 @@ int ttm_bo_device_init(struct ttm_bo_device *bdev,
bdev->dev_mapping = NULL;
bdev->glob = glob;
bdev->need_dma32 = need_dma32;
-
+ bdev->val_seq = 0;
+ spin_lock_init(&bdev->fence_lock);
mutex_lock(&glob->device_list_mutex);
list_add_tail(&bdev->device_list, &glob->device_list);
mutex_unlock(&glob->device_list_mutex);
@@ -1561,7 +1589,7 @@ bool ttm_mem_reg_is_pci(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
return true;
}
-void ttm_bo_unmap_virtual(struct ttm_buffer_object *bo)
+void ttm_bo_unmap_virtual_locked(struct ttm_buffer_object *bo)
{
struct ttm_bo_device *bdev = bo->bdev;
loff_t offset = (loff_t) bo->addr_space_offset;
@@ -1570,8 +1598,20 @@ void ttm_bo_unmap_virtual(struct ttm_buffer_object *bo)
if (!bdev->dev_mapping)
return;
unmap_mapping_range(bdev->dev_mapping, offset, holelen, 1);
- ttm_mem_io_free(bdev, &bo->mem);
+ ttm_mem_io_free_vm(bo);
+}
+
+void ttm_bo_unmap_virtual(struct ttm_buffer_object *bo)
+{
+ struct ttm_bo_device *bdev = bo->bdev;
+ struct ttm_mem_type_manager *man = &bdev->man[bo->mem.mem_type];
+
+ ttm_mem_io_lock(man, false);
+ ttm_bo_unmap_virtual_locked(bo);
+ ttm_mem_io_unlock(man);
}
+
+
EXPORT_SYMBOL(ttm_bo_unmap_virtual);
static void ttm_bo_vm_insert_rb(struct ttm_buffer_object *bo)
@@ -1651,6 +1691,7 @@ int ttm_bo_wait(struct ttm_buffer_object *bo,
bool lazy, bool interruptible, bool no_wait)
{
struct ttm_bo_driver *driver = bo->bdev->driver;
+ struct ttm_bo_device *bdev = bo->bdev;
void *sync_obj;
void *sync_obj_arg;
int ret = 0;
@@ -1664,9 +1705,9 @@ int ttm_bo_wait(struct ttm_buffer_object *bo,
void *tmp_obj = bo->sync_obj;
bo->sync_obj = NULL;
clear_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags);
- spin_unlock(&bo->lock);
+ spin_unlock(&bdev->fence_lock);
driver->sync_obj_unref(&tmp_obj);
- spin_lock(&bo->lock);
+ spin_lock(&bdev->fence_lock);
continue;
}
@@ -1675,29 +1716,29 @@ int ttm_bo_wait(struct ttm_buffer_object *bo,
sync_obj = driver->sync_obj_ref(bo->sync_obj);
sync_obj_arg = bo->sync_obj_arg;
- spin_unlock(&bo->lock);
+ spin_unlock(&bdev->fence_lock);
ret = driver->sync_obj_wait(sync_obj, sync_obj_arg,
lazy, interruptible);
if (unlikely(ret != 0)) {
driver->sync_obj_unref(&sync_obj);
- spin_lock(&bo->lock);
+ spin_lock(&bdev->fence_lock);
return ret;
}
- spin_lock(&bo->lock);
+ spin_lock(&bdev->fence_lock);
if (likely(bo->sync_obj == sync_obj &&
bo->sync_obj_arg == sync_obj_arg)) {
void *tmp_obj = bo->sync_obj;
bo->sync_obj = NULL;
clear_bit(TTM_BO_PRIV_FLAG_MOVING,
&bo->priv_flags);
- spin_unlock(&bo->lock);
+ spin_unlock(&bdev->fence_lock);
driver->sync_obj_unref(&sync_obj);
driver->sync_obj_unref(&tmp_obj);
- spin_lock(&bo->lock);
+ spin_lock(&bdev->fence_lock);
} else {
- spin_unlock(&bo->lock);
+ spin_unlock(&bdev->fence_lock);
driver->sync_obj_unref(&sync_obj);
- spin_lock(&bo->lock);
+ spin_lock(&bdev->fence_lock);
}
}
return 0;
@@ -1706,6 +1747,7 @@ EXPORT_SYMBOL(ttm_bo_wait);
int ttm_bo_synccpu_write_grab(struct ttm_buffer_object *bo, bool no_wait)
{
+ struct ttm_bo_device *bdev = bo->bdev;
int ret = 0;
/*
@@ -1715,9 +1757,9 @@ int ttm_bo_synccpu_write_grab(struct ttm_buffer_object *bo, bool no_wait)
ret = ttm_bo_reserve(bo, true, no_wait, false, 0);
if (unlikely(ret != 0))
return ret;
- spin_lock(&bo->lock);
+ spin_lock(&bdev->fence_lock);
ret = ttm_bo_wait(bo, false, true, no_wait);
- spin_unlock(&bo->lock);
+ spin_unlock(&bdev->fence_lock);
if (likely(ret == 0))
atomic_inc(&bo->cpu_writers);
ttm_bo_unreserve(bo);
@@ -1783,16 +1825,15 @@ static int ttm_bo_swapout(struct ttm_mem_shrink *shrink)
put_count = ttm_bo_del_from_lru(bo);
spin_unlock(&glob->lru_lock);
- while (put_count--)
- kref_put(&bo->list_kref, ttm_bo_ref_bug);
+ ttm_bo_list_ref_sub(bo, put_count, true);
/**
* Wait for GPU, then move to system cached.
*/
- spin_lock(&bo->lock);
+ spin_lock(&bo->bdev->fence_lock);
ret = ttm_bo_wait(bo, false, false, false);
- spin_unlock(&bo->lock);
+ spin_unlock(&bo->bdev->fence_lock);
if (unlikely(ret != 0))
goto out;
generated by cgit v1.2.3-58-ga151 (git 2.39.0) at 2024-12-23 04:33:18 +0000