On Wed, Jan 20, 2021 at 01:48:05PM -0800, Mike Kravetz wrote:

quoted

quoted

This comment addresses both this patch and the next one.

Instead of putting the SetHPageMigratable flag spread over the
allocation paths, would it make more sense to place it in
alloc_huge_page before returning the page?
Then we could opencode SetHPageMigratableIfSupported right there.

and in putback_active_hugepage.

Hi Oscar,

In Muchun's series of hugetlb bug fixes, Michal asked the same question.

https://lore.kernel.org/linux-mm/7e69a55c-d501-6b42-8225-a677f09fb829@oracle.com/ (local)

The 'short answer' is that the this would allow a page to be migrated
after allocation but before the page fault code adds it to the page
cache or page tables.  This actually caused bugs in the past.

Oh, I see. I jumped late into that patchset so I missed some early messages.
Thanks for explaining this again.

-- 
Oscar Salvador
SUSE L3