static void __i915_vma_retire(struct i915_active *ref)
{
struct i915_vma *vma = container_of(ref, typeof(*vma), active);
struct drm_i915_gem_object *obj = vma->obj;
GEM_BUG_ON(!i915_gem_object_is_active(obj));
if (--obj->active_count)
return;
/* Prune the shared fence arrays iff completely idle (inc. external) */
if (reservation_object_trylock(obj->resv)) {
if (reservation_object_test_signaled_rcu(obj->resv, true))
reservation_object_add_excl_fence(obj->resv, NULL);
reservation_object_unlock(obj->resv);
}
/*
* Bump our place on the bound list to keep it roughly in LRU order
* so that we don't steal from recently used but inactive objects
* (unless we are forced to ofc!)
*/
obj_bump_mru(obj);
i915_gem_object_put(obj); /* and drop the active reference */
}
static int gpu_set(struct drm_i915_gem_object *obj,
unsigned long offset,
u32 v)
{
struct drm_i915_private *i915 = to_i915(obj->base.dev);
struct drm_i915_gem_request *rq;
struct i915_vma *vma;
u32 *cs;
int err;
err = i915_gem_object_set_to_gtt_domain(obj, true);
if (err)
return err;
vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0, 0);
if (IS_ERR(vma))
return PTR_ERR(vma);
rq = i915_gem_request_alloc(i915->engine[RCS], i915->kernel_context);
if (IS_ERR(rq)) {
i915_vma_unpin(vma);
return PTR_ERR(rq);
}
cs = intel_ring_begin(rq, 4);
if (IS_ERR(cs)) {
__i915_add_request(rq, false);
i915_vma_unpin(vma);
return PTR_ERR(cs);
}
if (INTEL_GEN(i915) >= 8) {
*cs++ = MI_STORE_DWORD_IMM_GEN4 | 1 << 22;
*cs++ = lower_32_bits(i915_ggtt_offset(vma) + offset);
*cs++ = upper_32_bits(i915_ggtt_offset(vma) + offset);
*cs++ = v;
} else if (INTEL_GEN(i915) >= 4) {
*cs++ = MI_STORE_DWORD_IMM_GEN4 | 1 << 22;
*cs++ = 0;
*cs++ = i915_ggtt_offset(vma) + offset;
*cs++ = v;
} else {
*cs++ = MI_STORE_DWORD_IMM | 1 << 22;
*cs++ = i915_ggtt_offset(vma) + offset;
*cs++ = v;
*cs++ = MI_NOOP;
}
intel_ring_advance(rq, cs);
i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
i915_vma_unpin(vma);
reservation_object_lock(obj->resv, NULL);
reservation_object_add_excl_fence(obj->resv, &rq->fence);
reservation_object_unlock(obj->resv);
__i915_add_request(rq, true);
return 0;
}
static void
i915_vma_retire(struct i915_gem_active *active, struct i915_request *rq)
{
const unsigned int idx = rq->engine->id;
struct i915_vma *vma =
container_of(active, struct i915_vma, last_read[idx]);
struct drm_i915_gem_object *obj = vma->obj;
GEM_BUG_ON(!i915_vma_has_active_engine(vma, idx));
i915_vma_clear_active(vma, idx);
if (i915_vma_is_active(vma))
return;
GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
list_move_tail(&vma->vm_link, &vma->vm->inactive_list);
if (unlikely(i915_vma_is_closed(vma) && !i915_vma_is_pinned(vma)))
WARN_ON(i915_vma_unbind(vma));
GEM_BUG_ON(!i915_gem_object_is_active(obj));
if (--obj->active_count)
return;
/* Prune the shared fence arrays iff completely idle (inc. external) */
if (reservation_object_trylock(obj->resv)) {
if (reservation_object_test_signaled_rcu(obj->resv, true))
reservation_object_add_excl_fence(obj->resv, NULL);
reservation_object_unlock(obj->resv);
}
/* Bump our place on the bound list to keep it roughly in LRU order
* so that we don't steal from recently used but inactive objects
* (unless we are forced to ofc!)
*/
spin_lock(&rq->i915->mm.obj_lock);
if (obj->bind_count)
list_move_tail(&obj->mm.link, &rq->i915->mm.bound_list);
spin_unlock(&rq->i915->mm.obj_lock);
obj->mm.dirty = true; /* be paranoid */
if (i915_gem_object_has_active_reference(obj)) {
i915_gem_object_clear_active_reference(obj);
i915_gem_object_put(obj);
}
}
/**
* i915_gem_batch_pool_get() - allocate a buffer from the pool
* @pool: the batch buffer pool
* @size: the minimum desired size of the returned buffer
*
* Returns an inactive buffer from @pool with at least @size bytes,
* with the pages pinned. The caller must i915_gem_object_unpin_pages()
* on the returned object.
*
* Note: Callers must hold the struct_mutex
*
* Return: the buffer object or an error pointer
*/
struct drm_i915_gem_object *
i915_gem_batch_pool_get(struct i915_gem_batch_pool *pool,
size_t size)
{
struct drm_i915_gem_object *obj;
struct list_head *list;
int n, ret;
lockdep_assert_held(&pool->engine->i915->drm.struct_mutex);
/* Compute a power-of-two bucket, but throw everything greater than
* 16KiB into the same bucket: i.e. the the buckets hold objects of
* (1 page, 2 pages, 4 pages, 8+ pages).
*/
n = fls(size >> PAGE_SHIFT) - 1;
if (n >= ARRAY_SIZE(pool->cache_list))
n = ARRAY_SIZE(pool->cache_list) - 1;
list = &pool->cache_list[n];
list_for_each_entry(obj, list, batch_pool_link) {
/* The batches are strictly LRU ordered */
if (i915_gem_object_is_active(obj)) {
struct reservation_object *resv = obj->resv;
if (!reservation_object_test_signaled_rcu(resv, true))
break;
i915_retire_requests(pool->engine->i915);
GEM_BUG_ON(i915_gem_object_is_active(obj));
/*
* The object is now idle, clear the array of shared
* fences before we add a new request. Although, we
* remain on the same engine, we may be on a different
* timeline and so may continually grow the array,
* trapping a reference to all the old fences, rather
* than replace the existing fence.
*/
if (rcu_access_pointer(resv->fence)) {
reservation_object_lock(resv, NULL);
reservation_object_add_excl_fence(resv, NULL);
reservation_object_unlock(resv);
}
}
GEM_BUG_ON(!reservation_object_test_signaled_rcu(obj->resv,
true));
if (obj->base.size >= size)
goto found;
}
obj = i915_gem_object_create_internal(pool->engine->i915, size);
if (IS_ERR(obj))
return obj;
found:
ret = i915_gem_object_pin_pages(obj);
if (ret)
return ERR_PTR(ret);
list_move_tail(&obj->batch_pool_link, list);
return obj;
}
list_for_each_entry_continue_reverse(entry, list, head) {
struct ttm_buffer_object *bo = entry->bo;
reservation_object_unlock(bo->resv);
}
bool i915_gem_clflush_object(struct drm_i915_gem_object *obj,
unsigned int flags)
{
struct clflush *clflush;
/*
* Stolen memory is always coherent with the GPU as it is explicitly
* marked as wc by the system, or the system is cache-coherent.
* Similarly, we only access struct pages through the CPU cache, so
* anything not backed by physical memory we consider to be always
* coherent and not need clflushing.
*/
if (!i915_gem_object_has_struct_page(obj)) {
obj->cache_dirty = false;
return false;
}
/* If the GPU is snooping the contents of the CPU cache,
* we do not need to manually clear the CPU cache lines. However,
* the caches are only snooped when the render cache is
* flushed/invalidated. As we always have to emit invalidations
* and flushes when moving into and out of the RENDER domain, correct
* snooping behaviour occurs naturally as the result of our domain
* tracking.
*/
if (!(flags & I915_CLFLUSH_FORCE) &&
obj->cache_coherent & I915_BO_CACHE_COHERENT_FOR_READ)
return false;
trace_i915_gem_object_clflush(obj);
clflush = NULL;
if (!(flags & I915_CLFLUSH_SYNC))
clflush = kmalloc(sizeof(*clflush), GFP_KERNEL);
if (clflush) {
GEM_BUG_ON(!obj->cache_dirty);
dma_fence_init(&clflush->dma,
&i915_clflush_ops,
&clflush_lock,
to_i915(obj->base.dev)->mm.unordered_timeline,
0);
i915_sw_fence_init(&clflush->wait, i915_clflush_notify);
clflush->obj = i915_gem_object_get(obj);
INIT_WORK(&clflush->work, i915_clflush_work);
dma_fence_get(&clflush->dma);
i915_sw_fence_await_reservation(&clflush->wait,
obj->resv, NULL,
true, I915_FENCE_TIMEOUT,
I915_FENCE_GFP);
reservation_object_lock(obj->resv, NULL);
reservation_object_add_excl_fence(obj->resv, &clflush->dma);
reservation_object_unlock(obj->resv);
i915_sw_fence_commit(&clflush->wait);
} else if (obj->mm.pages) {
__i915_do_clflush(obj);
} else {
GEM_BUG_ON(obj->base.write_domain != I915_GEM_DOMAIN_CPU);
}
obj->cache_dirty = false;
return true;
}
/*
* vgem_fence_attach_ioctl (DRM_IOCTL_VGEM_FENCE_ATTACH):
*
* Create and attach a fence to the vGEM handle. This fence is then exposed
* via the dma-buf reservation object and visible to consumers of the exported
* dma-buf. If the flags contain VGEM_FENCE_WRITE, the fence indicates the
* vGEM buffer is being written to by the client and is exposed as an exclusive
* fence, otherwise the fence indicates the client is current reading from the
* buffer and all future writes should wait for the client to signal its
* completion. Note that if a conflicting fence is already on the dma-buf (i.e.
* an exclusive fence when adding a read, or any fence when adding a write),
* -EBUSY is reported. Serialisation between operations should be handled
* by waiting upon the dma-buf.
*
* This returns the handle for the new fence that must be signaled within 10
* seconds (or otherwise it will automatically expire). See
* vgem_fence_signal_ioctl (DRM_IOCTL_VGEM_FENCE_SIGNAL).
*
* If the vGEM handle does not exist, vgem_fence_attach_ioctl returns -ENOENT.
*/
int vgem_fence_attach_ioctl(struct drm_device *dev,
void *data,
struct drm_file *file)
{
struct drm_vgem_fence_attach *arg = data;
struct vgem_file *vfile = file->driver_priv;
struct reservation_object *resv;
struct drm_gem_object *obj;
struct dma_fence *fence;
int ret;
if (arg->flags & ~VGEM_FENCE_WRITE)
return -EINVAL;
if (arg->pad)
return -EINVAL;
obj = drm_gem_object_lookup(file, arg->handle);
if (!obj)
return -ENOENT;
ret = attach_dmabuf(dev, obj);
if (ret)
goto err;
fence = vgem_fence_create(vfile, arg->flags);
if (!fence) {
ret = -ENOMEM;
goto err;
}
/* Check for a conflicting fence */
resv = obj->dma_buf->resv;
if (!reservation_object_test_signaled_rcu(resv,
arg->flags & VGEM_FENCE_WRITE)) {
ret = -EBUSY;
goto err_fence;
}
/* Expose the fence via the dma-buf */
ret = 0;
reservation_object_lock(resv, NULL);
if (arg->flags & VGEM_FENCE_WRITE)
reservation_object_add_excl_fence(resv, fence);
else if ((ret = reservation_object_reserve_shared(resv)) == 0)
reservation_object_add_shared_fence(resv, fence);
reservation_object_unlock(resv);
/* Record the fence in our idr for later signaling */
if (ret == 0) {
mutex_lock(&vfile->fence_mutex);
ret = idr_alloc(&vfile->fence_idr, fence, 1, 0, GFP_KERNEL);
mutex_unlock(&vfile->fence_mutex);
if (ret > 0) {
arg->out_fence = ret;
ret = 0;
}
}
err_fence:
if (ret) {
dma_fence_signal(fence);
dma_fence_put(fence);
}
err:
drm_gem_object_put_unlocked(obj);
return ret;
}
