summaryrefslogtreecommitdiff
path: root/drivers/gpu/drm/i915/i915_gem_execbuffer.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/gpu/drm/i915/i915_gem_execbuffer.c')
-rw-r--r--drivers/gpu/drm/i915/i915_gem_execbuffer.c130
1 files changed, 89 insertions, 41 deletions
diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index 2c9d9cbaf653..20fef6c50267 100644
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -35,6 +35,9 @@
#define __EXEC_OBJECT_HAS_PIN (1<<31)
#define __EXEC_OBJECT_HAS_FENCE (1<<30)
+#define __EXEC_OBJECT_NEEDS_BIAS (1<<28)
+
+#define BATCH_OFFSET_BIAS (256*1024)
struct eb_vmas {
struct list_head vmas;
@@ -545,7 +548,7 @@ i915_gem_execbuffer_reserve_vma(struct i915_vma *vma,
struct drm_i915_gem_exec_object2 *entry = vma->exec_entry;
bool has_fenced_gpu_access = INTEL_INFO(ring->dev)->gen < 4;
bool need_fence;
- unsigned flags;
+ uint64_t flags;
int ret;
flags = 0;
@@ -559,6 +562,8 @@ i915_gem_execbuffer_reserve_vma(struct i915_vma *vma,
if (entry->flags & EXEC_OBJECT_NEEDS_GTT)
flags |= PIN_GLOBAL;
+ if (entry->flags & __EXEC_OBJECT_NEEDS_BIAS)
+ flags |= BATCH_OFFSET_BIAS | PIN_OFFSET_BIAS;
ret = i915_gem_object_pin(obj, vma->vm, entry->alignment, flags);
if (ret)
@@ -592,6 +597,36 @@ i915_gem_execbuffer_reserve_vma(struct i915_vma *vma,
return 0;
}
+static bool
+eb_vma_misplaced(struct i915_vma *vma, bool has_fenced_gpu_access)
+{
+ struct drm_i915_gem_exec_object2 *entry = vma->exec_entry;
+ struct drm_i915_gem_object *obj = vma->obj;
+ bool need_fence, need_mappable;
+
+ need_fence =
+ has_fenced_gpu_access &&
+ entry->flags & EXEC_OBJECT_NEEDS_FENCE &&
+ obj->tiling_mode != I915_TILING_NONE;
+ need_mappable = need_fence || need_reloc_mappable(vma);
+
+ WARN_ON((need_mappable || need_fence) &&
+ !i915_is_ggtt(vma->vm));
+
+ if (entry->alignment &&
+ vma->node.start & (entry->alignment - 1))
+ return true;
+
+ if (need_mappable && !obj->map_and_fenceable)
+ return true;
+
+ if (entry->flags & __EXEC_OBJECT_NEEDS_BIAS &&
+ vma->node.start < BATCH_OFFSET_BIAS)
+ return true;
+
+ return false;
+}
+
static int
i915_gem_execbuffer_reserve(struct intel_ring_buffer *ring,
struct list_head *vmas,
@@ -653,26 +688,10 @@ i915_gem_execbuffer_reserve(struct intel_ring_buffer *ring,
/* Unbind any ill-fitting objects or pin. */
list_for_each_entry(vma, vmas, exec_list) {
- struct drm_i915_gem_exec_object2 *entry = vma->exec_entry;
- bool need_fence, need_mappable;
-
- obj = vma->obj;
-
if (!drm_mm_node_allocated(&vma->node))
continue;
- need_fence =
- has_fenced_gpu_access &&
- entry->flags & EXEC_OBJECT_NEEDS_FENCE &&
- obj->tiling_mode != I915_TILING_NONE;
- need_mappable = need_fence || need_reloc_mappable(vma);
-
- WARN_ON((need_mappable || need_fence) &&
- !i915_is_ggtt(vma->vm));
-
- if ((entry->alignment &&
- vma->node.start & (entry->alignment - 1)) ||
- (need_mappable && !obj->map_and_fenceable))
+ if (eb_vma_misplaced(vma, has_fenced_gpu_access))
ret = i915_vma_unbind(vma);
else
ret = i915_gem_execbuffer_reserve_vma(vma, ring, need_relocs);
@@ -773,9 +792,9 @@ i915_gem_execbuffer_relocate_slow(struct drm_device *dev,
* relocations were valid.
*/
for (j = 0; j < exec[i].relocation_count; j++) {
- if (copy_to_user(&user_relocs[j].presumed_offset,
- &invalid_offset,
- sizeof(invalid_offset))) {
+ if (__copy_to_user(&user_relocs[j].presumed_offset,
+ &invalid_offset,
+ sizeof(invalid_offset))) {
ret = -EFAULT;
mutex_lock(&dev->struct_mutex);
goto err;
@@ -999,6 +1018,25 @@ i915_reset_gen7_sol_offsets(struct drm_device *dev,
return 0;
}
+static struct drm_i915_gem_object *
+eb_get_batch(struct eb_vmas *eb)
+{
+ struct i915_vma *vma = list_entry(eb->vmas.prev, typeof(*vma), exec_list);
+
+ /*
+ * SNA is doing fancy tricks with compressing batch buffers, which leads
+ * to negative relocation deltas. Usually that works out ok since the
+ * relocate address is still positive, except when the batch is placed
+ * very low in the GTT. Ensure this doesn't happen.
+ *
+ * Note that actual hangs have only been observed on gen7, but for
+ * paranoia do it everywhere.
+ */
+ vma->exec_entry->flags |= __EXEC_OBJECT_NEEDS_BIAS;
+
+ return vma->obj;
+}
+
static int
i915_gem_do_execbuffer(struct drm_device *dev, void *data,
struct drm_file *file,
@@ -1153,7 +1191,7 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data,
goto err;
/* take note of the batch buffer before we might reorder the lists */
- batch_obj = list_entry(eb->vmas.prev, struct i915_vma, exec_list)->obj;
+ batch_obj = eb_get_batch(eb);
/* Move the objects en-masse into the GTT, evicting if necessary. */
need_relocs = (args->flags & I915_EXEC_NO_RELOC) == 0;
@@ -1355,18 +1393,21 @@ i915_gem_execbuffer(struct drm_device *dev, void *data,
ret = i915_gem_do_execbuffer(dev, data, file, &exec2, exec2_list);
if (!ret) {
+ struct drm_i915_gem_exec_object __user *user_exec_list =
+ to_user_ptr(args->buffers_ptr);
+
/* Copy the new buffer offsets back to the user's exec list. */
- for (i = 0; i < args->buffer_count; i++)
- exec_list[i].offset = exec2_list[i].offset;
- /* ... and back out to userspace */
- ret = copy_to_user(to_user_ptr(args->buffers_ptr),
- exec_list,
- sizeof(*exec_list) * args->buffer_count);
- if (ret) {
- ret = -EFAULT;
- DRM_DEBUG("failed to copy %d exec entries "
- "back to user (%d)\n",
- args->buffer_count, ret);
+ for (i = 0; i < args->buffer_count; i++) {
+ ret = __copy_to_user(&user_exec_list[i].offset,
+ &exec2_list[i].offset,
+ sizeof(user_exec_list[i].offset));
+ if (ret) {
+ ret = -EFAULT;
+ DRM_DEBUG("failed to copy %d exec entries "
+ "back to user (%d)\n",
+ args->buffer_count, ret);
+ break;
+ }
}
}
@@ -1412,14 +1453,21 @@ i915_gem_execbuffer2(struct drm_device *dev, void *data,
ret = i915_gem_do_execbuffer(dev, data, file, args, exec2_list);
if (!ret) {
/* Copy the new buffer offsets back to the user's exec list. */
- ret = copy_to_user(to_user_ptr(args->buffers_ptr),
- exec2_list,
- sizeof(*exec2_list) * args->buffer_count);
- if (ret) {
- ret = -EFAULT;
- DRM_DEBUG("failed to copy %d exec entries "
- "back to user (%d)\n",
- args->buffer_count, ret);
+ struct drm_i915_gem_exec_object2 *user_exec_list =
+ to_user_ptr(args->buffers_ptr);
+ int i;
+
+ for (i = 0; i < args->buffer_count; i++) {
+ ret = __copy_to_user(&user_exec_list[i].offset,
+ &exec2_list[i].offset,
+ sizeof(user_exec_list[i].offset));
+ if (ret) {
+ ret = -EFAULT;
+ DRM_DEBUG("failed to copy %d exec entries "
+ "back to user\n",
+ args->buffer_count);
+ break;
+ }
}
}
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-22 15:21:12 +0000