Re: [RFC PATCH v2 09/18] mm/memory-failure: handle folios with PG_rcu_free

[Kiryl Shutsemau <kas@kernel.org>]

On Thu, Nov 05, 2026 at 01:03:41PM +0000, sashiko-bot@kernel.org wrote:
> - [High] mm/memory-failure: unsafe RCU wait in memory_failure() already-poisoned path

In v3 PG_rcu_free stays set until the grace period ends, so the first
case does not happen.  v3 waits there unconditionally anyway: the path
already sleeps, and the test looked at a folio it held no reference to.
The second case is real for shmem; v3 checks PageHWPoison() on each
page in the lockless copy.

> - [High] mm/memory-failure: mf_free_rcu_pages() skips sync while rcu_free_pages() runs

The window between the last put and the deferral is the put_page() race
memory_failure() has without this series.  For the counter, v3 drains
without checking it: the drain waits for the works in flight and starts
no grace period when nothing waits.  The per-cpu half does not happen:
get_hwpoison_page() drains the per-cpu lists before get_any_page().

> - [Medium] mm/memory-failure: order-0 PG_rcu_free page dropped after setting PG_hwpoison

By design: the free drops a poisoned order-0 page, as it does when the
last put races with memory_failure() without this series.  The page is
never reused.
-- 
  Kiryl Shutsemau / Kirill A. Shutemov