Re: [PATCH v2 00/10] Use generic_file_read_iter() in hugetlbfs

[Matthew Wilcox <willy@infradead.org>]

On Thu, Jul 09, 2026 at 07:47:28PM +0100, Matthew Wilcox (Oracle) wrote:
> This set of patches are separated out from Jane's earlier series.
> https://lore.kernel.org/linux-mm/20260617172534.1740152-1-jane.chu@oracle.com/
> 
> There's enough potentially controversial things in here that I think
> it's worth considering them by themselves.  First, handling hwpoison
> in filemap_read().  I want this for shmem as well as for hugetlbfs (a
> later patchset, maybe).  Second, handling misses in the pagecache as
> holes rather than opportunities to populate the pagecache.
> 
> v1 here: https://lore.kernel.org/linux-mm/20260707181713.208282-1-willy@infradead.org/
> Sashiko had a lot of valuable feedback on that, and that's all addressed
> here except for the things that I don't think are relevant.

Well, Sashiko pointed out a few different things this time.  Will noodle
on them and produce a v3:

https://sashiko.dev/#/patchset/20260709184740.1286561-1-willy%40infradead.org

The only one I think is invalid is:

: Does returning 0 here with an empty batch cause an infinite loop in
: filemap_splice_read()?

No, it doesn't, because we don't call splice_read for hugetlbfs.

The rest all seem like legitimate concerns.  If anyone spots anything
else, let me know.  And really what I want from humans is higher level
feedback like "This is the wrong way to handle hwpoison, it should be
done inside copy_page_to_iter()".