drm/i915: Make the physical object coherent with GTT
Currently objects for which the hardware needs a contiguous physical
address are allocated a shadow backing storage to satisfy the contraint.
This shadow buffer is not wired into the normal obj->pages and so the
physical object is incoherent with accesses via the GPU, GTT and CPU. By
setting up the appropriate scatter-gather table, we can allow userspace
to access the physical object via either a GTT mmaping of or by rendering
into the GEM bo. However, keeping the CPU mmap of the shmemfs backing
storage coherent with the contiguous shadow is not yet possible.
Fortuituously, CPU mmaps of objects requiring physical addresses are not
expected to be coherent anyway.
This allows the physical constraint of the GEM object to be transparent
to userspace and allow it to efficiently render into or update them via
the GTT and GPU.
v2: Fix leak of pci handle spotted by Ville
v3: Remove the now duplicate call to detach_phys_object during free.
v4: Wait for rendering before pwrite. As this patch makes it possible to
render into the phys object, we should make it correct as well!
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 3e0cabe..86cf428 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -208,40 +208,137 @@
return 0;
}
-static void i915_gem_object_detach_phys(struct drm_i915_gem_object *obj)
+static int
+i915_gem_object_get_pages_phys(struct drm_i915_gem_object *obj)
{
- drm_dma_handle_t *phys = obj->phys_handle;
+ struct address_space *mapping = file_inode(obj->base.filp)->i_mapping;
+ char *vaddr = obj->phys_handle->vaddr;
+ struct sg_table *st;
+ struct scatterlist *sg;
+ int i;
- if (!phys)
- return;
+ if (WARN_ON(i915_gem_object_needs_bit17_swizzle(obj)))
+ return -EINVAL;
- if (obj->madv == I915_MADV_WILLNEED) {
+ for (i = 0; i < obj->base.size / PAGE_SIZE; i++) {
+ struct page *page;
+ char *src;
+
+ page = shmem_read_mapping_page(mapping, i);
+ if (IS_ERR(page))
+ return PTR_ERR(page);
+
+ src = kmap_atomic(page);
+ memcpy(vaddr, src, PAGE_SIZE);
+ drm_clflush_virt_range(vaddr, PAGE_SIZE);
+ kunmap_atomic(src);
+
+ page_cache_release(page);
+ vaddr += PAGE_SIZE;
+ }
+
+ i915_gem_chipset_flush(obj->base.dev);
+
+ st = kmalloc(sizeof(*st), GFP_KERNEL);
+ if (st == NULL)
+ return -ENOMEM;
+
+ if (sg_alloc_table(st, 1, GFP_KERNEL)) {
+ kfree(st);
+ return -ENOMEM;
+ }
+
+ sg = st->sgl;
+ sg->offset = 0;
+ sg->length = obj->base.size;
+
+ sg_dma_address(sg) = obj->phys_handle->busaddr;
+ sg_dma_len(sg) = obj->base.size;
+
+ obj->pages = st;
+ obj->has_dma_mapping = true;
+ return 0;
+}
+
+static void
+i915_gem_object_put_pages_phys(struct drm_i915_gem_object *obj)
+{
+ int ret;
+
+ BUG_ON(obj->madv == __I915_MADV_PURGED);
+
+ ret = i915_gem_object_set_to_cpu_domain(obj, true);
+ if (ret) {
+ /* In the event of a disaster, abandon all caches and
+ * hope for the best.
+ */
+ WARN_ON(ret != -EIO);
+ obj->base.read_domains = obj->base.write_domain = I915_GEM_DOMAIN_CPU;
+ }
+
+ if (obj->madv == I915_MADV_DONTNEED)
+ obj->dirty = 0;
+
+ if (obj->dirty) {
struct address_space *mapping = file_inode(obj->base.filp)->i_mapping;
- char *vaddr = phys->vaddr;
+ char *vaddr = obj->phys_handle->vaddr;
int i;
for (i = 0; i < obj->base.size / PAGE_SIZE; i++) {
- struct page *page = shmem_read_mapping_page(mapping, i);
- if (!IS_ERR(page)) {
- char *dst = kmap_atomic(page);
- memcpy(dst, vaddr, PAGE_SIZE);
- drm_clflush_virt_range(dst, PAGE_SIZE);
- kunmap_atomic(dst);
+ struct page *page;
+ char *dst;
- set_page_dirty(page);
+ page = shmem_read_mapping_page(mapping, i);
+ if (IS_ERR(page))
+ continue;
+
+ dst = kmap_atomic(page);
+ drm_clflush_virt_range(vaddr, PAGE_SIZE);
+ memcpy(dst, vaddr, PAGE_SIZE);
+ kunmap_atomic(dst);
+
+ set_page_dirty(page);
+ if (obj->madv == I915_MADV_WILLNEED)
mark_page_accessed(page);
- page_cache_release(page);
- }
+ page_cache_release(page);
vaddr += PAGE_SIZE;
}
- i915_gem_chipset_flush(obj->base.dev);
+ obj->dirty = 0;
}
-#ifdef CONFIG_X86
- set_memory_wb((unsigned long)phys->vaddr, phys->size / PAGE_SIZE);
-#endif
- drm_pci_free(obj->base.dev, phys);
- obj->phys_handle = NULL;
+ sg_free_table(obj->pages);
+ kfree(obj->pages);
+
+ obj->has_dma_mapping = false;
+}
+
+static void
+i915_gem_object_release_phys(struct drm_i915_gem_object *obj)
+{
+ drm_pci_free(obj->base.dev, obj->phys_handle);
+}
+
+static const struct drm_i915_gem_object_ops i915_gem_phys_ops = {
+ .get_pages = i915_gem_object_get_pages_phys,
+ .put_pages = i915_gem_object_put_pages_phys,
+ .release = i915_gem_object_release_phys,
+};
+
+static int
+drop_pages(struct drm_i915_gem_object *obj)
+{
+ struct i915_vma *vma, *next;
+ int ret;
+
+ drm_gem_object_reference(&obj->base);
+ list_for_each_entry_safe(vma, next, &obj->vma_list, vma_link)
+ if (i915_vma_unbind(vma))
+ break;
+
+ ret = i915_gem_object_put_pages(obj);
+ drm_gem_object_unreference(&obj->base);
+
+ return ret;
}
int
@@ -249,9 +346,7 @@
int align)
{
drm_dma_handle_t *phys;
- struct address_space *mapping;
- char *vaddr;
- int i;
+ int ret;
if (obj->phys_handle) {
if ((unsigned long)obj->phys_handle->vaddr & (align -1))
@@ -266,41 +361,19 @@
if (obj->base.filp == NULL)
return -EINVAL;
+ ret = drop_pages(obj);
+ if (ret)
+ return ret;
+
/* create a new object */
phys = drm_pci_alloc(obj->base.dev, obj->base.size, align);
if (!phys)
return -ENOMEM;
- vaddr = phys->vaddr;
-#ifdef CONFIG_X86
- set_memory_wc((unsigned long)vaddr, phys->size / PAGE_SIZE);
-#endif
- mapping = file_inode(obj->base.filp)->i_mapping;
- for (i = 0; i < obj->base.size / PAGE_SIZE; i++) {
- struct page *page;
- char *src;
-
- page = shmem_read_mapping_page(mapping, i);
- if (IS_ERR(page)) {
-#ifdef CONFIG_X86
- set_memory_wb((unsigned long)phys->vaddr, phys->size / PAGE_SIZE);
-#endif
- drm_pci_free(obj->base.dev, phys);
- return PTR_ERR(page);
- }
-
- src = kmap_atomic(page);
- memcpy(vaddr, src, PAGE_SIZE);
- kunmap_atomic(src);
-
- mark_page_accessed(page);
- page_cache_release(page);
-
- vaddr += PAGE_SIZE;
- }
-
obj->phys_handle = phys;
- return 0;
+ obj->ops = &i915_gem_phys_ops;
+
+ return i915_gem_object_get_pages(obj);
}
static int
@@ -311,6 +384,14 @@
struct drm_device *dev = obj->base.dev;
void *vaddr = obj->phys_handle->vaddr + args->offset;
char __user *user_data = to_user_ptr(args->data_ptr);
+ int ret;
+
+ /* We manually control the domain here and pretend that it
+ * remains coherent i.e. in the GTT domain, like shmem_pwrite.
+ */
+ ret = i915_gem_object_wait_rendering(obj, false);
+ if (ret)
+ return ret;
if (__copy_from_user_inatomic_nocache(vaddr, user_data, args->size)) {
unsigned long unwritten;
@@ -326,6 +407,7 @@
return -EFAULT;
}
+ drm_clflush_virt_range(vaddr, args->size);
i915_gem_chipset_flush(dev);
return 0;
}
@@ -1046,11 +1128,6 @@
* pread/pwrite currently are reading and writing from the CPU
* perspective, requiring manual detiling by the client.
*/
- if (obj->phys_handle) {
- ret = i915_gem_phys_pwrite(obj, args, file);
- goto out;
- }
-
if (obj->tiling_mode == I915_TILING_NONE &&
obj->base.write_domain != I915_GEM_DOMAIN_CPU &&
cpu_write_needs_clflush(obj)) {
@@ -1060,8 +1137,12 @@
* textures). Fallback to the shmem path in that case. */
}
- if (ret == -EFAULT || ret == -ENOSPC)
- ret = i915_gem_shmem_pwrite(dev, obj, args, file);
+ if (ret == -EFAULT || ret == -ENOSPC) {
+ if (obj->phys_handle)
+ ret = i915_gem_phys_pwrite(obj, args, file);
+ else
+ ret = i915_gem_shmem_pwrite(dev, obj, args, file);
+ }
out:
drm_gem_object_unreference(&obj->base);
@@ -3509,7 +3590,7 @@
* Stolen memory is always coherent with the GPU as it is explicitly
* marked as wc by the system, or the system is cache-coherent.
*/
- if (obj->stolen)
+ if (obj->stolen || obj->phys_handle)
return false;
/* If the GPU is snooping the contents of the CPU cache,
@@ -4471,8 +4552,6 @@
}
}
- i915_gem_object_detach_phys(obj);
-
/* Stolen objects don't hold a ref, but do hold pin count. Fix that up
* before progressing. */
if (obj->stolen)