summaryrefslogtreecommitdiffstats
path: root/drivers/gpu/drm/i915/gem/i915_gem_userptr.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/gpu/drm/i915/gem/i915_gem_userptr.c')
-rw-r--r--drivers/gpu/drm/i915/gem/i915_gem_userptr.c91
1 files changed, 46 insertions, 45 deletions
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_userptr.c b/drivers/gpu/drm/i915/gem/i915_gem_userptr.c
index 2caa594322bc..580319b7bf1a 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_userptr.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_userptr.c
@@ -12,11 +12,10 @@
#include <drm/i915_drm.h>
+#include "i915_drv.h"
#include "i915_gem_ioctls.h"
#include "i915_gem_object.h"
#include "i915_scatterlist.h"
-#include "i915_trace.h"
-#include "intel_drv.h"
struct i915_mm_struct {
struct mm_struct *mm;
@@ -93,7 +92,6 @@ userptr_mn_invalidate_range_start(struct mmu_notifier *_mn,
struct i915_mmu_notifier *mn =
container_of(_mn, struct i915_mmu_notifier, mn);
struct interval_tree_node *it;
- struct mutex *unlock = NULL;
unsigned long end;
int ret = 0;
@@ -130,32 +128,14 @@ userptr_mn_invalidate_range_start(struct mmu_notifier *_mn,
}
spin_unlock(&mn->lock);
- if (!unlock) {
- unlock = &mn->mm->i915->drm.struct_mutex;
-
- switch (mutex_trylock_recursive(unlock)) {
- default:
- case MUTEX_TRYLOCK_FAILED:
- if (mutex_lock_killable_nested(unlock, I915_MM_SHRINKER)) {
- i915_gem_object_put(obj);
- return -EINTR;
- }
- /* fall through */
- case MUTEX_TRYLOCK_SUCCESS:
- break;
-
- case MUTEX_TRYLOCK_RECURSIVE:
- unlock = ERR_PTR(-EEXIST);
- break;
- }
- }
-
- ret = i915_gem_object_unbind(obj);
+ ret = i915_gem_object_unbind(obj,
+ I915_GEM_OBJECT_UNBIND_ACTIVE |
+ I915_GEM_OBJECT_UNBIND_BARRIER);
if (ret == 0)
- ret = __i915_gem_object_put_pages(obj, I915_MM_SHRINKER);
+ ret = __i915_gem_object_put_pages(obj);
i915_gem_object_put(obj);
if (ret)
- goto unlock;
+ return ret;
spin_lock(&mn->lock);
@@ -168,10 +148,6 @@ userptr_mn_invalidate_range_start(struct mmu_notifier *_mn,
}
spin_unlock(&mn->lock);
-unlock:
- if (!IS_ERR_OR_NULL(unlock))
- mutex_unlock(unlock);
-
return ret;
}
@@ -427,7 +403,7 @@ struct get_pages_work {
static struct sg_table *
__i915_gem_userptr_alloc_pages(struct drm_i915_gem_object *obj,
- struct page **pvec, int num_pages)
+ struct page **pvec, unsigned long num_pages)
{
unsigned int max_segment = i915_sg_segment_size();
struct sg_table *st;
@@ -473,9 +449,10 @@ __i915_gem_userptr_get_pages_worker(struct work_struct *_work)
{
struct get_pages_work *work = container_of(_work, typeof(*work), work);
struct drm_i915_gem_object *obj = work->obj;
- const int npages = obj->base.size >> PAGE_SHIFT;
+ const unsigned long npages = obj->base.size >> PAGE_SHIFT;
+ unsigned long pinned;
struct page **pvec;
- int pinned, ret;
+ int ret;
ret = -ENOMEM;
pinned = 0;
@@ -484,31 +461,36 @@ __i915_gem_userptr_get_pages_worker(struct work_struct *_work)
if (pvec != NULL) {
struct mm_struct *mm = obj->userptr.mm->mm;
unsigned int flags = 0;
+ int locked = 0;
if (!i915_gem_object_is_readonly(obj))
flags |= FOLL_WRITE;
ret = -EFAULT;
if (mmget_not_zero(mm)) {
- down_read(&mm->mmap_sem);
while (pinned < npages) {
+ if (!locked) {
+ down_read(&mm->mmap_sem);
+ locked = 1;
+ }
ret = get_user_pages_remote
(work->task, mm,
obj->userptr.ptr + pinned * PAGE_SIZE,
npages - pinned,
flags,
- pvec + pinned, NULL, NULL);
+ pvec + pinned, NULL, &locked);
if (ret < 0)
break;
pinned += ret;
}
- up_read(&mm->mmap_sem);
+ if (locked)
+ up_read(&mm->mmap_sem);
mmput(mm);
}
}
- mutex_lock(&obj->mm.lock);
+ mutex_lock_nested(&obj->mm.lock, I915_MM_GET_PAGES);
if (obj->userptr.work == &work->work) {
struct sg_table *pages = ERR_PTR(ret);
@@ -578,7 +560,7 @@ __i915_gem_userptr_get_pages_schedule(struct drm_i915_gem_object *obj)
static int i915_gem_userptr_get_pages(struct drm_i915_gem_object *obj)
{
- const int num_pages = obj->base.size >> PAGE_SHIFT;
+ const unsigned long num_pages = obj->base.size >> PAGE_SHIFT;
struct mm_struct *mm = obj->userptr.mm->mm;
struct page **pvec;
struct sg_table *pages;
@@ -662,8 +644,16 @@ i915_gem_userptr_put_pages(struct drm_i915_gem_object *obj,
__i915_gem_object_release_shmem(obj, pages, true);
i915_gem_gtt_finish_pages(obj, pages);
+ /*
+ * We always mark objects as dirty when they are used by the GPU,
+ * just in case. However, if we set the vma as being read-only we know
+ * that the object will never have been written to.
+ */
+ if (i915_gem_object_is_readonly(obj))
+ obj->mm.dirty = false;
+
for_each_sgt_page(page, sgt_iter, pages) {
- if (obj->mm.dirty)
+ if (obj->mm.dirty && trylock_page(page)) {
/*
* As this may not be anonymous memory (e.g. shmem)
* but exist on a real mapping, we have to lock
@@ -671,8 +661,20 @@ i915_gem_userptr_put_pages(struct drm_i915_gem_object *obj,
* the page reference is not sufficient to
* prevent the inode from being truncated.
* Play safe and take the lock.
+ *
+ * However...!
+ *
+ * The mmu-notifier can be invalidated for a
+ * migrate_page, that is alreadying holding the lock
+ * on the page. Such a try_to_unmap() will result
+ * in us calling put_pages() and so recursively try
+ * to lock the page. We avoid that deadlock with
+ * a trylock_page() and in exchange we risk missing
+ * some page dirtying.
*/
- set_page_dirty_lock(page);
+ set_page_dirty(page);
+ unlock_page(page);
+ }
mark_page_accessed(page);
put_page(page);
@@ -702,6 +704,7 @@ i915_gem_userptr_dmabuf_export(struct drm_i915_gem_object *obj)
static const struct drm_i915_gem_object_ops i915_gem_userptr_ops = {
.flags = I915_GEM_OBJECT_HAS_STRUCT_PAGE |
I915_GEM_OBJECT_IS_SHRINKABLE |
+ I915_GEM_OBJECT_NO_GGTT |
I915_GEM_OBJECT_ASYNC_CANCEL,
.get_pages = i915_gem_userptr_get_pages,
.put_pages = i915_gem_userptr_put_pages,
@@ -749,6 +752,7 @@ i915_gem_userptr_ioctl(struct drm_device *dev,
void *data,
struct drm_file *file)
{
+ static struct lock_class_key lock_class;
struct drm_i915_private *dev_priv = to_i915(dev);
struct drm_i915_gem_userptr *args = data;
struct drm_i915_gem_object *obj;
@@ -776,14 +780,11 @@ i915_gem_userptr_ioctl(struct drm_device *dev,
return -EFAULT;
if (args->flags & I915_USERPTR_READ_ONLY) {
- struct i915_address_space *vm;
-
/*
* On almost all of the older hw, we cannot tell the GPU that
* a page is readonly.
*/
- vm = dev_priv->kernel_context->vm;
- if (!vm || !vm->has_read_only)
+ if (!dev_priv->gt.vm->has_read_only)
return -ENODEV;
}
@@ -792,7 +793,7 @@ i915_gem_userptr_ioctl(struct drm_device *dev,
return -ENOMEM;
drm_gem_private_object_init(dev, &obj->base, args->user_size);
- i915_gem_object_init(obj, &i915_gem_userptr_ops);
+ i915_gem_object_init(obj, &i915_gem_userptr_ops, &lock_class);
obj->read_domains = I915_GEM_DOMAIN_CPU;
obj->write_domain = I915_GEM_DOMAIN_CPU;
i915_gem_object_set_cache_coherency(obj, I915_CACHE_LLC);
OpenPOWER on IntegriCloud