On 2024/12/13 22:00, Matthew Wilcox wrote:
On Fri, Dec 13, 2024 at 03:49:53PM +0800, Kefeng Wang wrote:
On 2024/12/13 12:25, Matthew Wilcox wrote:
On Wed, Dec 11, 2024 at 10:43:46AM +0100, Amir Goldstein wrote:
I think this patch is fine as is.
This patch is complete crap. The test-case is broken. NAK to all of
this.
Hi Matthew, regardless of the test case, the original issue is the
ovl don't respect underlying fs' get_unmapped_area(), the lower fs may
have own rules for vma alignment(own get_unmapped_area callback),
thp_get_unmapped_area() is one case, what's your option/suggestion about
No, filesystems don't "have their own rules" for get_unmapped_area.
get_unmapped_area is for device drivers.
Commit 74d2fad1334d ("thp, dax: add thp_get_unmapped_area for pmd
mappings") to enable PMD mappings as a FSDAX filesystem, and with
commit 1854bc6e2420 ("mm/readahead: Align file mappings for non-DAX")
to enable THPs for mmapped files too, also other filesystem, eg,
tmpfs provide a shmem_get_unmapped_area to decide the mapping address,
see commit c01d5b300774 ("shmem: get_unmapped_area align huge page"),
that is what I think the filesystem have own rules to get the mapping
address, correct me if I misunderstood, thanks.
