Commit 692a576b authored by Daniel Vetter's avatar Daniel Vetter
Browse files

drm/i915: don't call shmem_read_mapping unnecessarily 



This speeds up pwrite and pread from ~120 µs ro ~100 µs for
reading/writing 1mb on my snb (if the backing storage pages
are already pinned, of course).

v2: Chris Wilson pointed out a glaring page reference bug - I've
unconditionally dropped the reference. With that fixed (and the
associated reduction of dirt in dmesg) it's now even a notch faster.

v3: Unconditionaly grab a page reference when dropping
dev->struct_mutex to simplify the code-flow.

Tested-by: default avatarChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: default avatarChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: default avatarDaniel Vetter <daniel.vetter@ffwll.ch>
parent 3ae53783

drivers/gpu/drm/i915/i915_gem.c

+30 −12
Original line number Diff line number Diff line
@@ -301,6 +301,7 @@ i915_gem_shmem_pread(struct drm_device *dev, 
	int obj_do_bit17_swizzling, page_do_bit17_swizzling;

	int hit_slowpath = 0;

	int needs_clflush = 0;

	int release_page;



	user_data = (char __user *) (uintptr_t) args->data_ptr;

	remain = args->size;
@@ -335,11 +336,17 @@ i915_gem_shmem_pread(struct drm_device *dev, 
		if ((shmem_page_offset + page_length) > PAGE_SIZE)

			page_length = PAGE_SIZE - shmem_page_offset;



		if (obj->pages) {

			page = obj->pages[offset >> PAGE_SHIFT];

			release_page = 0;

		} else {

			page = shmem_read_mapping_page(mapping, offset >> PAGE_SHIFT);

			if (IS_ERR(page)) {

				ret = PTR_ERR(page);

				goto out;

			}

			release_page = 1;

		}



		page_do_bit17_swizzling = obj_do_bit17_swizzling &&

			(page_to_phys(page) & (1 << 17)) != 0;
@@ -358,7 +365,7 @@ i915_gem_shmem_pread(struct drm_device *dev, 
		}



		hit_slowpath = 1;



		page_cache_get(page);

		mutex_unlock(&dev->struct_mutex);



		vaddr = kmap(page);
@@ -377,8 +384,10 @@ i915_gem_shmem_pread(struct drm_device *dev, 
		kunmap(page);



		mutex_lock(&dev->struct_mutex);

		page_cache_release(page);

next_page:

		mark_page_accessed(page);

		if (release_page)

			page_cache_release(page);



		if (ret) {
@@ -660,6 +669,7 @@ i915_gem_shmem_pwrite(struct drm_device *dev, 
	int shmem_page_offset, page_length, ret = 0;

	int obj_do_bit17_swizzling, page_do_bit17_swizzling;

	int hit_slowpath = 0;

	int release_page;



	user_data = (char __user *) (uintptr_t) args->data_ptr;

	remain = args->size;
@@ -684,11 +694,17 @@ i915_gem_shmem_pwrite(struct drm_device *dev, 
		if ((shmem_page_offset + page_length) > PAGE_SIZE)

			page_length = PAGE_SIZE - shmem_page_offset;



		if (obj->pages) {

			page = obj->pages[offset >> PAGE_SHIFT];

			release_page = 0;

		} else {

			page = shmem_read_mapping_page(mapping, offset >> PAGE_SHIFT);

			if (IS_ERR(page)) {

				ret = PTR_ERR(page);

				goto out;

			}

			release_page = 1;

		}



		page_do_bit17_swizzling = obj_do_bit17_swizzling &&

			(page_to_phys(page) & (1 << 17)) != 0;
@@ -705,7 +721,7 @@ i915_gem_shmem_pwrite(struct drm_device *dev, 
		}



		hit_slowpath = 1;



		page_cache_get(page);

		mutex_unlock(&dev->struct_mutex);



		vaddr = kmap(page);
@@ -720,9 +736,11 @@ i915_gem_shmem_pwrite(struct drm_device *dev, 
		kunmap(page);



		mutex_lock(&dev->struct_mutex);

		page_cache_release(page);

next_page:

		set_page_dirty(page);

		mark_page_accessed(page);

		if (release_page)

			page_cache_release(page);



		if (ret) {

CRA Git | Maintained and supported by SUSTech CRA and CCSE