Re: [PATCH 3/5] mm: Do not try to write pinned folio during memory cleaning writeback

[John Hubbard <jhubbard@nvidia.com>]

On 2/9/23 17:54, John Hubbard wrote:
> On 2/9/23 04:31, Jan Kara wrote:
>> When a folio is pinned, there is no point in trying to write it during
>> memory cleaning writeback. We cannot reclaim the folio until it is
>> unpinned anyway and we cannot even be sure the folio is really clean.
>> On top of that writeback of such folio may be problematic as the data
>> can change while the writeback is running thus causing checksum or
>> DIF/DIX failures. So just don't bother doing memory cleaning writeback
>> for pinned folios.
>>
>> Signed-off-by: Jan Kara <jack@suse.cz>
>> ---
>>   fs/9p/vfs_addr.c            |  2 +-
>>   fs/afs/file.c               |  2 +-
>>   fs/afs/write.c              |  6 +++---
>>   fs/btrfs/extent_io.c        | 14 +++++++-------
>>   fs/btrfs/free-space-cache.c |  2 +-
>>   fs/btrfs/inode.c            |  2 +-
>>   fs/btrfs/subpage.c          |  2 +-
> 

Oh, and one more fix, below, is required in order to build with my local
test config. Assuming that it is reasonable to deal with pinned pages
here, which I think it is:

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_shmem.c b/drivers/gpu/drm/i915/gem/i915_gem_shmem.c
index 9c759df700ca..c3279fb0edc8 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_shmem.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_shmem.c
@@ -313,7 +313,7 @@ void __shmem_writeback(size_t size, struct address_space *mapping)
  		if (!page)
  			continue;
  
-		if (!page_mapped(page) && clear_page_dirty_for_io(page)) {
+		if (!page_mapped(page) && clear_page_dirty_for_io(&wbc, page)) {
  			int ret;
  
  			SetPageReclaim(page);

thanks,
-- 
John Hubbard
NVIDIA