* IN THE SOFTWARE.
*/
-#include <stdint.h>
#include <stdlib.h>
#include <unistd.h>
#include <limits.h>
#include <assert.h>
-#include <linux/futex.h>
#include <linux/memfd.h>
-#include <sys/time.h>
#include <sys/mman.h>
-#include <sys/syscall.h>
#include "anv_private.h"
#include "util/hash_table.h"
+#include "util/simple_mtx.h"
#ifdef HAVE_VALGRIND
#define VG_NOACCESS_READ(__ptr) ({ \
#define ANV_MMAP_CLEANUP_INIT ((struct anv_mmap_cleanup){0})
-static inline long
-sys_futex(void *addr1, int op, int val1,
- struct timespec *timeout, void *addr2, int val3)
-{
- return syscall(SYS_futex, addr1, op, val1, timeout, addr2, val3);
-}
-
-static inline int
-futex_wake(uint32_t *addr, int count)
-{
- return sys_futex(addr, FUTEX_WAKE, count, NULL, NULL, 0);
-}
-
-static inline int
-futex_wait(uint32_t *addr, int32_t value)
-{
- return sys_futex(addr, FUTEX_WAIT, value, NULL, NULL, 0);
-}
-
+#ifndef HAVE_MEMFD_CREATE
static inline int
memfd_create(const char *name, unsigned int flags)
{
return syscall(SYS_memfd_create, name, flags);
}
+#endif
static inline uint32_t
ilog2_round_up(uint32_t value)
}
static void
-anv_free_list_push(union anv_free_list *list, void *map, int32_t offset)
+anv_free_list_push(union anv_free_list *list, void *map, int32_t offset,
+ uint32_t size, uint32_t count)
{
union anv_free_list current, old, new;
int32_t *next_ptr = map + offset;
+ /* If we're returning more than one chunk, we need to build a chain to add
+ * to the list. Fortunately, we can do this without any atomics since we
+ * own everything in the chain right now. `offset` is left pointing to the
+ * head of our chain list while `next_ptr` points to the tail.
+ */
+ for (uint32_t i = 1; i < count; i++) {
+ VG_NOACCESS_WRITE(next_ptr, offset + i * size);
+ next_ptr = map + offset + i * size;
+ }
+
old = *list;
do {
current = old;
VkResult
anv_block_pool_init(struct anv_block_pool *pool,
struct anv_device *device,
- uint32_t initial_size)
+ uint64_t start_address,
+ uint32_t initial_size,
+ uint64_t bo_flags)
{
VkResult result;
pool->device = device;
+ pool->bo_flags = bo_flags;
+ pool->start_address = gen_canonical_address(start_address);
+
anv_bo_init(&pool->bo, 0, 0);
pool->fd = memfd_create("block pool", MFD_CLOEXEC);
assert(center_bo_offset >= pool->back_state.end);
assert(size - center_bo_offset >= pool->state.end);
+ /* Assert that we don't go outside the bounds of the memfd */
+ assert(center_bo_offset <= BLOCK_POOL_MEMFD_CENTER);
+ assert(size - center_bo_offset <=
+ BLOCK_POOL_MEMFD_SIZE - BLOCK_POOL_MEMFD_CENTER);
+
cleanup = u_vector_add(&pool->mmap_cleanups);
if (!cleanup)
return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
MAP_SHARED | MAP_POPULATE, pool->fd,
BLOCK_POOL_MEMFD_CENTER - center_bo_offset);
if (map == MAP_FAILED)
- return vk_errorf(VK_ERROR_MEMORY_MAP_FAILED, "mmap failed: %m");
+ return vk_errorf(pool->device->instance, pool->device,
+ VK_ERROR_MEMORY_MAP_FAILED, "mmap failed: %m");
gem_handle = anv_gem_userptr(pool->device, map, size);
if (gem_handle == 0) {
munmap(map, size);
- return vk_errorf(VK_ERROR_TOO_MANY_OBJECTS, "userptr failed: %m");
+ return vk_errorf(pool->device->instance, pool->device,
+ VK_ERROR_TOO_MANY_OBJECTS, "userptr failed: %m");
}
cleanup->map = map;
* hard work for us.
*/
anv_bo_init(&pool->bo, gem_handle, size);
+ if (pool->bo_flags & EXEC_OBJECT_PINNED) {
+ pool->bo.offset = pool->start_address + BLOCK_POOL_MEMFD_CENTER -
+ center_bo_offset;
+ }
+ pool->bo.flags = pool->bo_flags;
pool->bo.map = map;
return VK_SUCCESS;
* the pool and a 4K CPU page.
*/
static uint32_t
-anv_block_pool_grow(struct anv_block_pool *pool, struct anv_block_state *state,
- uint32_t block_size)
+anv_block_pool_grow(struct anv_block_pool *pool, struct anv_block_state *state)
{
VkResult result = VK_SUCCESS;
assert(size > pool->bo.size);
- /* We can't have a block pool bigger than 1GB because we use signed
- * 32-bit offsets in the free list and we don't want overflow. We
- * should never need a block pool bigger than 1GB anyway.
- */
- assert(size <= (1u << 31));
-
/* We compute a new center_bo_offset such that, when we double the size
* of the pool, we maintain the ratio of how much is used by each side.
* This way things should remain more-or-less balanced.
*/
center_bo_offset = ((uint64_t)size * back_used) / total_used;
- /* Align down to a multiple of both the block size and page size */
- uint32_t granularity = MAX2(block_size, PAGE_SIZE);
- assert(util_is_power_of_two(granularity));
- center_bo_offset &= ~(granularity - 1);
+ /* Align down to a multiple of the page size */
+ center_bo_offset &= ~(PAGE_SIZE - 1);
assert(center_bo_offset >= back_used);
/* Make sure we don't shrink the back end of the pool */
- if (center_bo_offset < pool->back_state.end)
- center_bo_offset = pool->back_state.end;
+ if (center_bo_offset < back_required)
+ center_bo_offset = back_required;
/* Make sure that we don't shrink the front end of the pool */
- if (size - center_bo_offset < pool->state.end)
- center_bo_offset = size - pool->state.end;
+ if (size - center_bo_offset < front_required)
+ center_bo_offset = size - front_required;
}
- assert(center_bo_offset % block_size == 0);
assert(center_bo_offset % PAGE_SIZE == 0);
result = anv_block_pool_expand_range(pool, center_bo_offset, size);
- if (pool->device->instance->physicalDevice.has_exec_async)
- pool->bo.flags |= EXEC_OBJECT_ASYNC;
+ pool->bo.flags = pool->bo_flags;
done:
pthread_mutex_unlock(&pool->device->mutex);
{
struct anv_block_state state, old, new;
- assert(util_is_power_of_two(block_size));
-
while (1) {
state.u64 = __sync_fetch_and_add(&pool_state->u64, block_size);
- if (state.next < state.end) {
+ if (state.next + block_size <= state.end) {
assert(pool->map);
return state.next;
- } else if (state.next == state.end) {
+ } else if (state.next <= state.end) {
/* We allocated the first block outside the pool so we have to grow
* the pool. pool_state->next acts a mutex: threads who try to
* allocate now will get block indexes above the current limit and
* hit futex_wait below.
*/
new.next = state.next + block_size;
- new.end = anv_block_pool_grow(pool, pool_state, block_size);
+ do {
+ new.end = anv_block_pool_grow(pool, pool_state);
+ } while (new.end < new.next);
+
old.u64 = __sync_lock_test_and_set(&pool_state->u64, new.u64);
if (old.next != state.next)
futex_wake(&pool_state->end, INT_MAX);
return state.next;
} else {
- futex_wait(&pool_state->end, state.end);
+ futex_wait(&pool_state->end, state.end, NULL);
continue;
}
}
VkResult
anv_state_pool_init(struct anv_state_pool *pool,
struct anv_device *device,
- uint32_t block_size)
+ uint64_t start_address,
+ uint32_t block_size,
+ uint64_t bo_flags)
{
VkResult result = anv_block_pool_init(&pool->block_pool, device,
- block_size * 16);
+ start_address,
+ block_size * 16,
+ bo_flags);
if (result != VK_SUCCESS)
return result;
- assert(util_is_power_of_two(block_size));
+ assert(util_is_power_of_two_or_zero(block_size));
pool->block_size = block_size;
pool->back_alloc_free_list = ANV_FREE_LIST_EMPTY;
for (unsigned i = 0; i < ANV_STATE_BUCKETS; i++) {
struct anv_block_state block, old, new;
uint32_t offset;
+ /* If our state is large, we don't need any sub-allocation from a block.
+ * Instead, we just grab whole (potentially large) blocks.
+ */
+ if (state_size >= block_size)
+ return anv_block_pool_alloc(block_pool, state_size);
+
restart:
block.u64 = __sync_fetch_and_add(&pool->block.u64, state_size);
futex_wake(&pool->block.end, INT_MAX);
return offset;
} else {
- futex_wait(&pool->block.end, block.end);
+ futex_wait(&pool->block.end, block.end, NULL);
goto restart;
}
}
-static struct anv_state
-anv_state_pool_alloc_no_vg(struct anv_state_pool *pool,
- uint32_t size, uint32_t align)
+static uint32_t
+anv_state_pool_get_bucket(uint32_t size)
{
- unsigned size_log2 = ilog2_round_up(size < align ? align : size);
+ unsigned size_log2 = ilog2_round_up(size);
assert(size_log2 <= ANV_MAX_STATE_SIZE_LOG2);
if (size_log2 < ANV_MIN_STATE_SIZE_LOG2)
size_log2 = ANV_MIN_STATE_SIZE_LOG2;
- unsigned bucket = size_log2 - ANV_MIN_STATE_SIZE_LOG2;
+ return size_log2 - ANV_MIN_STATE_SIZE_LOG2;
+}
+
+static uint32_t
+anv_state_pool_get_bucket_size(uint32_t bucket)
+{
+ uint32_t size_log2 = bucket + ANV_MIN_STATE_SIZE_LOG2;
+ return 1 << size_log2;
+}
+
+static struct anv_state
+anv_state_pool_alloc_no_vg(struct anv_state_pool *pool,
+ uint32_t size, uint32_t align)
+{
+ uint32_t bucket = anv_state_pool_get_bucket(MAX2(size, align));
struct anv_state state;
- state.alloc_size = 1 << size_log2;
+ state.alloc_size = anv_state_pool_get_bucket_size(bucket);
/* Try free list first. */
if (anv_free_list_pop(&pool->buckets[bucket].free_list,
goto done;
}
+ /* Try to grab a chunk from some larger bucket and split it up */
+ for (unsigned b = bucket + 1; b < ANV_STATE_BUCKETS; b++) {
+ int32_t chunk_offset;
+ if (anv_free_list_pop(&pool->buckets[b].free_list,
+ &pool->block_pool.map, &chunk_offset)) {
+ unsigned chunk_size = anv_state_pool_get_bucket_size(b);
+
+ /* We've found a chunk that's larger than the requested state size.
+ * There are a couple of options as to what we do with it:
+ *
+ * 1) We could fully split the chunk into state.alloc_size sized
+ * pieces. However, this would mean that allocating a 16B
+ * state could potentially split a 2MB chunk into 512K smaller
+ * chunks. This would lead to unnecessary fragmentation.
+ *
+ * 2) The classic "buddy allocator" method would have us split the
+ * chunk in half and return one half. Then we would split the
+ * remaining half in half and return one half, and repeat as
+ * needed until we get down to the size we want. However, if
+ * you are allocating a bunch of the same size state (which is
+ * the common case), this means that every other allocation has
+ * to go up a level and every fourth goes up two levels, etc.
+ * This is not nearly as efficient as it could be if we did a
+ * little more work up-front.
+ *
+ * 3) Split the difference between (1) and (2) by doing a
+ * two-level split. If it's bigger than some fixed block_size,
+ * we split it into block_size sized chunks and return all but
+ * one of them. Then we split what remains into
+ * state.alloc_size sized chunks and return all but one.
+ *
+ * We choose option (3).
+ */
+ if (chunk_size > pool->block_size &&
+ state.alloc_size < pool->block_size) {
+ assert(chunk_size % pool->block_size == 0);
+ /* We don't want to split giant chunks into tiny chunks. Instead,
+ * break anything bigger than a block into block-sized chunks and
+ * then break it down into bucket-sized chunks from there. Return
+ * all but the first block of the chunk to the block bucket.
+ */
+ const uint32_t block_bucket =
+ anv_state_pool_get_bucket(pool->block_size);
+ anv_free_list_push(&pool->buckets[block_bucket].free_list,
+ pool->block_pool.map,
+ chunk_offset + pool->block_size,
+ pool->block_size,
+ (chunk_size / pool->block_size) - 1);
+ chunk_size = pool->block_size;
+ }
+
+ assert(chunk_size % state.alloc_size == 0);
+ anv_free_list_push(&pool->buckets[bucket].free_list,
+ pool->block_pool.map,
+ chunk_offset + state.alloc_size,
+ state.alloc_size,
+ (chunk_size / state.alloc_size) - 1);
+
+ state.offset = chunk_offset;
+ goto done;
+ }
+ }
+
state.offset = anv_fixed_size_state_pool_alloc_new(&pool->buckets[bucket],
&pool->block_pool,
state.alloc_size,
static void
anv_state_pool_free_no_vg(struct anv_state_pool *pool, struct anv_state state)
{
- assert(util_is_power_of_two(state.alloc_size));
- unsigned size_log2 = ilog2_round_up(state.alloc_size);
- assert(size_log2 >= ANV_MIN_STATE_SIZE_LOG2 &&
- size_log2 <= ANV_MAX_STATE_SIZE_LOG2);
- unsigned bucket = size_log2 - ANV_MIN_STATE_SIZE_LOG2;
+ assert(util_is_power_of_two_or_zero(state.alloc_size));
+ unsigned bucket = anv_state_pool_get_bucket(state.alloc_size);
if (state.offset < 0) {
assert(state.alloc_size == pool->block_size);
anv_free_list_push(&pool->back_alloc_free_list,
- pool->block_pool.map, state.offset);
+ pool->block_pool.map, state.offset,
+ state.alloc_size, 1);
} else {
anv_free_list_push(&pool->buckets[bucket].free_list,
- pool->block_pool.map, state.offset);
+ pool->block_pool.map, state.offset,
+ state.alloc_size, 1);
}
}
assert(alignment <= PAGE_SIZE);
uint32_t offset = align_u32(stream->next, alignment);
- if (offset + size > stream->block_size) {
+ if (offset + size > stream->block.alloc_size) {
+ uint32_t block_size = stream->block_size;
+ if (block_size < size)
+ block_size = round_to_power_of_two(size);
+
stream->block = anv_state_pool_alloc_no_vg(stream->state_pool,
- stream->block_size,
- PAGE_SIZE);
+ block_size, PAGE_SIZE);
struct anv_state_stream_block *sb = stream->block.map;
VG_NOACCESS_WRITE(&sb->block, stream->block);
VG_NOACCESS_WRITE(&sb->next, stream->block_list);
stream->block_list = sb;
- VG_NOACCESS_WRITE(&sb->_vg_ptr, NULL);
+ VG(VG_NOACCESS_WRITE(&sb->_vg_ptr, NULL));
VG(VALGRIND_MAKE_MEM_NOACCESS(stream->block.map, stream->block_size));
stream->next = sizeof(*sb);
offset = align_u32(stream->next, alignment);
- assert(offset + size <= stream->block_size);
+ assert(offset + size <= stream->block.alloc_size);
}
struct anv_state state = stream->block;
};
void
-anv_bo_pool_init(struct anv_bo_pool *pool, struct anv_device *device)
+anv_bo_pool_init(struct anv_bo_pool *pool, struct anv_device *device,
+ uint64_t bo_flags)
{
pool->device = device;
+ pool->bo_flags = bo_flags;
memset(pool->free_list, 0, sizeof(pool->free_list));
VG(VALGRIND_CREATE_MEMPOOL(pool, 0, false));
if (result != VK_SUCCESS)
return result;
+ new_bo.flags = pool->bo_flags;
+
assert(new_bo.size == pow2_size);
new_bo.map = anv_gem_mmap(pool->device, new_bo.gem_handle, 0, pow2_size, 0);
struct bo_pool_bo_link *link = bo.map;
VG_NOACCESS_WRITE(&link->bo, bo);
- assert(util_is_power_of_two(bo.size));
+ assert(util_is_power_of_two_or_zero(bo.size));
const unsigned size_log2 = ilog2_round_up(bo.size);
const unsigned bucket = size_log2 - 12;
assert(bucket < ARRAY_SIZE(pool->free_list));
pthread_mutex_lock(&device->mutex);
__sync_synchronize();
- if (bo->exists)
+ if (bo->exists) {
+ pthread_mutex_unlock(&device->mutex);
return &bo->bo;
+ }
const struct anv_physical_device *physical_device =
&device->instance->physicalDevice;
const struct gen_device_info *devinfo = &physical_device->info;
- /* WaCSScratchSize:hsw
- *
- * Haswell's scratch space address calculation appears to be sparse
- * rather than tightly packed. The Thread ID has bits indicating which
- * subslice, EU within a subslice, and thread within an EU it is.
- * There's a maximum of two slices and two subslices, so these can be
- * stored with a single bit. Even though there are only 10 EUs per
- * subslice, this is stored in 4 bits, so there's an effective maximum
- * value of 16 EUs. Similarly, although there are only 7 threads per EU,
- * this is stored in a 3 bit number, giving an effective maximum value
- * of 8 threads per EU.
- *
- * This means that we need to use 16 * 8 instead of 10 * 7 for the
- * number of threads per subslice.
- */
const unsigned subslices = MAX2(physical_device->subslice_total, 1);
- const unsigned scratch_ids_per_subslice =
- device->info.is_haswell ? 16 * 8 : devinfo->max_cs_threads;
+
+ unsigned scratch_ids_per_subslice;
+ if (devinfo->is_haswell) {
+ /* WaCSScratchSize:hsw
+ *
+ * Haswell's scratch space address calculation appears to be sparse
+ * rather than tightly packed. The Thread ID has bits indicating
+ * which subslice, EU within a subslice, and thread within an EU it
+ * is. There's a maximum of two slices and two subslices, so these
+ * can be stored with a single bit. Even though there are only 10 EUs
+ * per subslice, this is stored in 4 bits, so there's an effective
+ * maximum value of 16 EUs. Similarly, although there are only 7
+ * threads per EU, this is stored in a 3 bit number, giving an
+ * effective maximum value of 8 threads per EU.
+ *
+ * This means that we need to use 16 * 8 instead of 10 * 7 for the
+ * number of threads per subslice.
+ */
+ scratch_ids_per_subslice = 16 * 8;
+ } else if (devinfo->is_cherryview) {
+ /* Cherryview devices have either 6 or 8 EUs per subslice, and each EU
+ * has 7 threads. The 6 EU devices appear to calculate thread IDs as if
+ * it had 8 EUs.
+ */
+ scratch_ids_per_subslice = 8 * 7;
+ } else {
+ scratch_ids_per_subslice = devinfo->max_cs_threads;
+ }
uint32_t max_threads[] = {
[MESA_SHADER_VERTEX] = devinfo->max_vs_threads,
*
* so nothing will ever touch the top page.
*/
- bo->bo.flags &= ~EXEC_OBJECT_SUPPORTS_48B_ADDRESS;
+ assert(!(bo->bo.flags & EXEC_OBJECT_SUPPORTS_48B_ADDRESS));
+
+ if (device->instance->physicalDevice.has_exec_async)
+ bo->bo.flags |= EXEC_OBJECT_ASYNC;
/* Set the exists last because it may be read by other threads */
__sync_synchronize();
if (pthread_mutex_init(&cache->mutex, NULL)) {
_mesa_hash_table_destroy(cache->bo_map, NULL);
- return vk_errorf(VK_ERROR_OUT_OF_HOST_MEMORY,
+ return vk_errorf(NULL, NULL, VK_ERROR_OUT_OF_HOST_MEMORY,
"pthread_mutex_init failed: %m");
}
return bo;
}
-static struct anv_bo *
+UNUSED static struct anv_bo *
anv_bo_cache_lookup(struct anv_bo_cache *cache, uint32_t gem_handle)
{
pthread_mutex_lock(&cache->mutex);
return bo ? &bo->bo : NULL;
}
+#define ANV_BO_CACHE_SUPPORTED_FLAGS \
+ (EXEC_OBJECT_WRITE | \
+ EXEC_OBJECT_ASYNC | \
+ EXEC_OBJECT_SUPPORTS_48B_ADDRESS | \
+ EXEC_OBJECT_PINNED)
+
VkResult
anv_bo_cache_alloc(struct anv_device *device,
struct anv_bo_cache *cache,
- uint64_t size, struct anv_bo **bo_out)
+ uint64_t size, uint64_t bo_flags,
+ struct anv_bo **bo_out)
{
+ assert(bo_flags == (bo_flags & ANV_BO_CACHE_SUPPORTED_FLAGS));
+
struct anv_cached_bo *bo =
vk_alloc(&device->alloc, sizeof(struct anv_cached_bo), 8,
VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
return result;
}
+ bo->bo.flags = bo_flags;
+
+ if (!anv_vma_alloc(device, &bo->bo)) {
+ anv_gem_close(device, bo->bo.gem_handle);
+ vk_free(&device->alloc, bo);
+ return vk_errorf(device->instance, NULL,
+ VK_ERROR_OUT_OF_DEVICE_MEMORY,
+ "failed to allocate virtual address for BO");
+ }
+
assert(bo->bo.gem_handle);
pthread_mutex_lock(&cache->mutex);
VkResult
anv_bo_cache_import(struct anv_device *device,
struct anv_bo_cache *cache,
- int fd, uint64_t size, struct anv_bo **bo_out)
+ int fd, uint64_t bo_flags,
+ struct anv_bo **bo_out)
{
- pthread_mutex_lock(&cache->mutex);
+ assert(bo_flags == (bo_flags & ANV_BO_CACHE_SUPPORTED_FLAGS));
- /* The kernel is going to give us whole pages anyway */
- size = align_u64(size, 4096);
+ pthread_mutex_lock(&cache->mutex);
uint32_t gem_handle = anv_gem_fd_to_handle(device, fd);
if (!gem_handle) {
pthread_mutex_unlock(&cache->mutex);
- return vk_error(VK_ERROR_INVALID_EXTERNAL_HANDLE_KHX);
+ return vk_error(VK_ERROR_INVALID_EXTERNAL_HANDLE_KHR);
}
struct anv_cached_bo *bo = anv_bo_cache_lookup_locked(cache, gem_handle);
if (bo) {
- if (bo->bo.size != size) {
+ /* We have to be careful how we combine flags so that it makes sense.
+ * Really, though, if we get to this case and it actually matters, the
+ * client has imported a BO twice in different ways and they get what
+ * they have coming.
+ */
+ uint64_t new_flags = 0;
+ new_flags |= (bo->bo.flags | bo_flags) & EXEC_OBJECT_WRITE;
+ new_flags |= (bo->bo.flags & bo_flags) & EXEC_OBJECT_ASYNC;
+ new_flags |= (bo->bo.flags & bo_flags) & EXEC_OBJECT_SUPPORTS_48B_ADDRESS;
+ new_flags |= (bo->bo.flags | bo_flags) & EXEC_OBJECT_PINNED;
+
+ /* It's theoretically possible for a BO to get imported such that it's
+ * both pinned and not pinned. The only way this can happen is if it
+ * gets imported as both a semaphore and a memory object and that would
+ * be an application error. Just fail out in that case.
+ */
+ if ((bo->bo.flags & EXEC_OBJECT_PINNED) !=
+ (bo_flags & EXEC_OBJECT_PINNED)) {
pthread_mutex_unlock(&cache->mutex);
- return vk_error(VK_ERROR_INVALID_EXTERNAL_HANDLE_KHX);
+ return vk_errorf(device->instance, NULL,
+ VK_ERROR_INVALID_EXTERNAL_HANDLE,
+ "The same BO was imported two different ways");
}
+
+ /* It's also theoretically possible that someone could export a BO from
+ * one heap and import it into another or to import the same BO into two
+ * different heaps. If this happens, we could potentially end up both
+ * allowing and disallowing 48-bit addresses. There's not much we can
+ * do about it if we're pinning so we just throw an error and hope no
+ * app is actually that stupid.
+ */
+ if ((new_flags & EXEC_OBJECT_PINNED) &&
+ (bo->bo.flags & EXEC_OBJECT_SUPPORTS_48B_ADDRESS) !=
+ (bo_flags & EXEC_OBJECT_SUPPORTS_48B_ADDRESS)) {
+ return vk_errorf(device->instance, NULL,
+ VK_ERROR_INVALID_EXTERNAL_HANDLE,
+ "The same BO was imported on two different heaps");
+ }
+
+ bo->bo.flags = new_flags;
+
__sync_fetch_and_add(&bo->refcount, 1);
} else {
- /* For security purposes, we reject BO imports where the size does not
- * match exactly. This prevents a malicious client from passing a
- * buffer to a trusted client, lying about the size, and telling the
- * trusted client to try and texture from an image that goes
- * out-of-bounds. This sort of thing could lead to GPU hangs or worse
- * in the trusted client. The trusted client can protect itself against
- * this sort of attack but only if it can trust the buffer size.
- */
- off_t import_size = lseek(fd, 0, SEEK_END);
- if (import_size == (off_t)-1 || import_size != size) {
+ off_t size = lseek(fd, 0, SEEK_END);
+ if (size == (off_t)-1) {
anv_gem_close(device, gem_handle);
pthread_mutex_unlock(&cache->mutex);
- return vk_error(VK_ERROR_INVALID_EXTERNAL_HANDLE_KHX);
+ return vk_error(VK_ERROR_INVALID_EXTERNAL_HANDLE_KHR);
}
bo = vk_alloc(&device->alloc, sizeof(struct anv_cached_bo), 8,
bo->refcount = 1;
anv_bo_init(&bo->bo, gem_handle, size);
+ bo->bo.flags = bo_flags;
- if (device->instance->physicalDevice.supports_48bit_addresses)
- bo->bo.flags |= EXEC_OBJECT_SUPPORTS_48B_ADDRESS;
-
- if (device->instance->physicalDevice.has_exec_async)
- bo->bo.flags |= EXEC_OBJECT_ASYNC;
+ if (!anv_vma_alloc(device, &bo->bo)) {
+ anv_gem_close(device, bo->bo.gem_handle);
+ pthread_mutex_unlock(&cache->mutex);
+ vk_free(&device->alloc, bo);
+ return vk_errorf(device->instance, NULL,
+ VK_ERROR_OUT_OF_DEVICE_MEMORY,
+ "failed to allocate virtual address for BO");
+ }
_mesa_hash_table_insert(cache->bo_map, (void *)(uintptr_t)gem_handle, bo);
}
pthread_mutex_unlock(&cache->mutex);
-
- /* From the Vulkan spec:
- *
- * "Importing memory from a file descriptor transfers ownership of
- * the file descriptor from the application to the Vulkan
- * implementation. The application must not perform any operations on
- * the file descriptor after a successful import."
- *
- * If the import fails, we leave the file descriptor open.
- */
- close(fd);
-
*bo_out = &bo->bo;
return VK_SUCCESS;
if (bo->bo.map)
anv_gem_munmap(bo->bo.map, bo->bo.size);
+ anv_vma_free(device, &bo->bo);
+
anv_gem_close(device, bo->bo.gem_handle);
/* Don't unlock until we've actually closed the BO. The whole point of