[v2,2/9] drm/i915: Use memcpy_[from/to]_page() in gem/i915_gem_pyhs.c

Commit Message

Zhao Liu March 29, 2023, 7:32 a.m. UTC

From: Zhao Liu <[e-mail]>

The use of kmap_atomic() is being deprecated in favor of
kmap_local_page()[1],  and this patch converts the call from
kmap_atomic() + memcpy() to memcpy_[from/to]_page(), which use
kmap_local_page() to build local mapping and then do memcpy().

The main difference between atomic and local mappings is that local
mappings doesn't disable page faults or preemption (the preemption is
disabled for !PREEMPT_RT case, otherwise it only disables migration).

With kmap_local_page(), we can avoid the often unwanted side effect of
unnecessary page faults and preemption disables.

In drm/i915/gem/i915_gem_phys.c, the functions
i915_gem_object_get_pages_phys() and i915_gem_object_put_pages_phys()
don't need to disable pagefaults and preemption for mapping because of
2 reasons:

1. The flush operation is safe. In drm/i915/gem/i915_gem_object.c,
i915_gem_object_get_pages_phys() and i915_gem_object_put_pages_phys()
calls drm_clflush_virt_range() to use CLFLUSHOPT or WBINVD to flush.
Since CLFLUSHOPT is global on x86 and WBINVD is called on each cpu in
drm_clflush_virt_range(), the flush operation is global.

2. Any context switch caused by preemption or page faults (page fault
may cause sleep) doesn't affect the validity of local mapping.

Therefore, i915_gem_object_get_pages_phys() and
i915_gem_object_put_pages_phys() are two functions where the uses of
local mappings in place of atomic mappings are correctly suited.

Convert the calls of kmap_atomic() / kunmap_atomic() + memcpy() to
memcpy_from_page() and memcpy_to_page().

[1]: https://lore.kernel.org/all/[e-mail]

v2:
* Used memcpy_from_page() and memcpy_to_page() to replace
  kmap_local_page() + memcpy().
* Dropped hot plug related description since it has nothing to do with
  kmap_local_page().
* Added description of the motivation of using kmap_local_page().

Suggested-by: Dave Hansen <[e-mail]>
Suggested-by: Ira Weiny <[e-mail]>
Suggested-by: Fabio M. De Francesco <[e-mail]>
Signed-off-by: Zhao Liu <[e-mail]>
---
Suggested by credits:
  Dave: Referred to his explanation about cache flush.
  Ira: Referred to his task document, review comments and explanation
       about cache flush.
  Fabio: Referred to his boiler plate commit message and his description
         about why kmap_local_page() should be preferred. Also based on
         his suggestion to use memcpy_[from/to]_page() directly.
---
 drivers/gpu/drm/i915/gem/i915_gem_phys.c | 10 ++--------
 1 file changed, 2 insertions(+), 8 deletions(-)

Comments

Ira Weiny March 30, 2023, 11:01 p.m. UTC | #1

Zhao Liu wrote:
> From: Zhao Liu <[e-mail]>
> 
> The use of kmap_atomic() is being deprecated in favor of
> kmap_local_page()[1],  and this patch converts the call from
> kmap_atomic() + memcpy() to memcpy_[from/to]_page(), which use
> kmap_local_page() to build local mapping and then do memcpy().
> 
> The main difference between atomic and local mappings is that local
> mappings doesn't disable page faults or preemption (the preemption is
> disabled for !PREEMPT_RT case, otherwise it only disables migration).
> 
> With kmap_local_page(), we can avoid the often unwanted side effect of
> unnecessary page faults and preemption disables.
> 
> In drm/i915/gem/i915_gem_phys.c, the functions
> i915_gem_object_get_pages_phys() and i915_gem_object_put_pages_phys()
> don't need to disable pagefaults and preemption for mapping because of
> 2 reasons:
> 
> 1. The flush operation is safe. In drm/i915/gem/i915_gem_object.c,
> i915_gem_object_get_pages_phys() and i915_gem_object_put_pages_phys()
> calls drm_clflush_virt_range() to use CLFLUSHOPT or WBINVD to flush.
> Since CLFLUSHOPT is global on x86 and WBINVD is called on each cpu in
> drm_clflush_virt_range(), the flush operation is global.
> 
> 2. Any context switch caused by preemption or page faults (page fault
> may cause sleep) doesn't affect the validity of local mapping.
> 
> Therefore, i915_gem_object_get_pages_phys() and
> i915_gem_object_put_pages_phys() are two functions where the uses of
> local mappings in place of atomic mappings are correctly suited.
> 
> Convert the calls of kmap_atomic() / kunmap_atomic() + memcpy() to
> memcpy_from_page() and memcpy_to_page().
> 
> [1]: https://lore.kernel.org/all/[e-mail]
> 
> v2:
> * Used memcpy_from_page() and memcpy_to_page() to replace
>   kmap_local_page() + memcpy().
> * Dropped hot plug related description since it has nothing to do with
>   kmap_local_page().
> * Added description of the motivation of using kmap_local_page().
> 
> Suggested-by: Dave Hansen <[e-mail]>
> Suggested-by: Ira Weiny <[e-mail]>

Reviewed-by: Ira Weiny <[e-mail]>

Patch

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_phys.c b/drivers/gpu/drm/i915/gem/i915_gem_phys.c
index 76efe98eaa14..4c6d3f07260a 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_phys.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_phys.c
@@ -64,16 +64,13 @@  static int i915_gem_object_get_pages_phys(struct drm_i915_gem_object *obj)
 	dst = vaddr;
 	for (i = 0; i < obj->base.size / PAGE_SIZE; i++) {
 		struct page *page;
-		void *src;
 
 		page = shmem_read_mapping_page(mapping, i);
 		if (IS_ERR(page))
 			goto err_st;
 
-		src = kmap_atomic(page);
-		memcpy(dst, src, PAGE_SIZE);
+		memcpy_from_page(dst, page, 0, PAGE_SIZE);
 		drm_clflush_virt_range(dst, PAGE_SIZE);
-		kunmap_atomic(src);
 
 		put_page(page);
 		dst += PAGE_SIZE;
@@ -112,16 +109,13 @@  i915_gem_object_put_pages_phys(struct drm_i915_gem_object *obj,
 
 		for (i = 0; i < obj->base.size / PAGE_SIZE; i++) {
 			struct page *page;
-			char *dst;
 
 			page = shmem_read_mapping_page(mapping, i);
 			if (IS_ERR(page))
 				continue;
 
-			dst = kmap_atomic(page);
 			drm_clflush_virt_range(src, PAGE_SIZE);
-			memcpy(dst, src, PAGE_SIZE);
-			kunmap_atomic(dst);
+			memcpy_to_page(page, 0, src, PAGE_SIZE);
 
 			set_page_dirty(page);
 			if (obj->mm.madv == I915_MADV_WILLNEED)