On 08.06.22 03:31, HORIGUCHI NAOYA(堀口 直也) wrote: > On Tue, Jun 07, 2022 at 03:04:15PM +0200, David Hildenbrand wrote: >> On 02.06.22 07:06, Naoya Horiguchi wrote: >>> From: Naoya Horiguchi <naoya.horiguchi@xxxxxxx> >>> >>> When handling memory error on a hugetlb page, the error handler tries to >>> dissolve and turn it into 4kB pages. If it's successfully dissolved, >>> PageHWPoison flag is moved to the raw error page, so that's all right. >>> However, dissolve sometimes fails, then the error page is left as >>> hwpoisoned hugepage. It's useful if we can retry to dissolve it to save >>> healthy pages, but that's not possible now because the information about >>> where the raw error page is lost. >>> >>> Use the private field of a tail page to keep that information. The code >>> path of shrinking hugepage pool used this info to try delayed dissolve. >>> This only keeps one hwpoison page for now, which might be OK because it's >>> simple and multiple hwpoison pages in a hugepage can be rare. But it can >>> be extended in the future. >>> >>> >> >> But what would happen now if you have multiple successive MCE events on >> such a page now? > > The 2nd and later events are ignored due to "already hwpoisoned hugepage", > this might not be good when the hwpoisoned hugepage is freed/dissolved later. > So a temporal workaround is to remember "hugepage has multiple hwpoison pages" > and disable free/dissolve for such hugepages. Right. We might want to indicate exactly one vs. multiple, and then in __update_and_free_page(), move the hwpoison flag to one subpage, eventually exposing other corrupted subpages to the system. -- Thanks, David / dhildenb
