Re: [PATCH v6 0/2] kexec: keep the next kernel off hardware-poisoned pages

On Wed, Aug 12, 2026 at 04:31:50AM -0700, Breno Leitao wrote:
> Memory failures are common enough on large fleets that kexec regularly
> lands the next kernel on a frame the memory failure subsystem has already
> marked bad. Patch 2 teaches the segment placement to avoid those frames.
> 
> Patch 1 is a prerequisite. locate_mem_hole_top_down() walks candidates
> downwards without ever checking that the subtraction stays above zero, so
> the walk can wrap and report success with a destination outside of RAM.
> Patch 2 adds one more downward step to that loop, so the bug is fixed
> first and both patches rely on the same bail-out.

I queued this for v7.4, the patches will appear in the liveupdate tree
after the merge window.
 
> Signed-off-by: Breno Leitao <[email protected]>
> ---
> Changes in v6:
> - Split the pre-existing top-down underflow out into its own patch, and
>   drop the "if (poison < kbuf->memsz)" guard from the hwpoison hunk now
>   that the loop bail-out covers it. (Mike Rapoport, Sashiko)
> - Reword the changelog: the MCE comes from reading the poison back, not
>   from the relocation copy writing it, and the placement paragraphs read
>   better split up. (Mike Rapoport)
> - Move cond_resched() after the struct page dereference, so the scan does
>   not yield between pfn_to_online_page() and is_page_hwpoison().
>   (Pratyush Yadav, Sashiko)
> - Keep cond_resched() per pfn rather than batching it. (Kiryl Shutsemau,
>   Rik van Riel)
> - Link to v5: https://patch.msgid.link/[email protected]
> 
> Changes in v5:
> - Return -EHWPOISON instead of -EADDRNOTAVAIL
> - Leverage is_page_hwpoison() instead of per-page check
> - Link to v4: https://patch.msgid.link/[email protected]
> 
> Changes in v4:
> - Anchor the top-down hole finder on the first poisoned page in the
>   window and the bottom-up one on the last, so each jumps clear of the
>   poison in one step. New range_first_hwpoison(). (Kiryl Shutsemau)
> - Count a poisoned hugetlb folio in full: the flag lives on the folio,
>   not on the subpages, so the per-pfn scan missed poisoned tail pages.
>   (Kiryl Shutsemau)
> - Link to v3: https://patch.msgid.link/[email protected]
> 
> Changes in v3:
> - Return the address of the last poisoned page in the range, or
>   PHYS_ADDR_MAX when it is clean, instead of a bool plus an output
>   parameter. Renamed to range_last_hwpoison(). (Pratyush Yadav)
> - Add cond_resched() to the scan loop, as a segment can span half of
>   memory. (Sashiko)
> - Link to v2: https://patch.msgid.link/[email protected]
> 
> Changes in v2:
> - Change from pfn_to_page() to pfn_to_online_page(). (Miaohe Lin)
> - Return the poisoned address once we find a hit, to avoid the O(n^2)
>   rescan. (Sashiko)
> - Link to v1: https://patch.msgid.link/[email protected]
> 
> To: Andrew Morton <[email protected]>
> To: David Hildenbrand <[email protected]>
> To: Lorenzo Stoakes <[email protected]>
> To: "Liam R. Howlett" <[email protected]>
> To: Vlastimil Babka <[email protected]>
> To: Mike Rapoport <[email protected]>
> To: Suren Baghdasaryan <[email protected]>
> To: Michal Hocko <[email protected]>
> To: Baoquan He <[email protected]>
> To: Pasha Tatashin <[email protected]>
> To: Pratyush Yadav <[email protected]>
> To: Miaohe Lin <[email protected]>
> To: Naoya Horiguchi <[email protected]>
> Cc: [email protected]
> Cc: [email protected]
> Cc: [email protected]
> Cc: [email protected]
> Cc: [email protected]
> 
> ---
> Breno Leitao (2):
>       kexec_file: stop the top-down search before it underflows
>       kexec: keep the next kernel off hardware-poisoned pages
> 
>  include/linux/mm.h  | 14 ++++++++++++++
>  kernel/kexec_core.c | 10 ++++++++++
>  kernel/kexec_file.c | 20 +++++++++++++++++++-
>  mm/memory-failure.c | 40 ++++++++++++++++++++++++++++++++++++++++
>  4 files changed, 83 insertions(+), 1 deletion(-)
> ---
> base-commit: c5e32e86ca02b003f86e095d379b38148999293d
> change-id: 20260727-kexec_posioned-72bb0a4143a0
> 
> Best regards,
> --  
> Breno Leitao <[email protected]>
> 

-- 
Sincerely yours,
Mike.