On Wed, 3 Mar 2021 15:41:35 +0000
"Luck, Tony" <tony.luck@xxxxxxxxx> wrote:
> > For error address with sigbus, i think this is not an issue resulted by the patch i post, before my patch, the issue is already there.
> > I don't find a realizable way to get the correct address for same reason --- we don't know whether the page mapping is there or not when
> > we got to kill_me_maybe(), in some case, we may get it, but there are a lot of parallel issue need to consider, and if failed we have to fallback
> > to the error brach again, remaining current code may be an easy option;
>
> My RFC patch from yesterday removes the uncertainty about whether the page is there or not. After it walks the page
> tables we know that the poison page isn't mapped (note that patch is RFC for a reason ... I'm 90% sure that it should
> do a bit more that just clear the PRESENT bit).
>
> So perhaps memory_failure() has queued a SIGBUS for this task, if so, we take it when we return from kill_me_maybe()
>
> If not, we will return to user mode and re-execute the failing instruction ... but because the page is unmapped we will take a #PF
Got this, I have some error thoughts here.
> The x86 page fault handler will see that the page for this physical address is marked HWPOISON, and it will send the SIGBUS
> (just like it does if the page had been removed by an earlier UCNA/SRAO error).
if your methods works, should it be like this?
1582 pteval = swp_entry_to_pte(make_hwpoison_entry(subpage));
1583 if (PageHuge(page)) {
1584 hugetlb_count_sub(compound_nr(page), mm);
1585 set_huge_swap_pte_at(mm, address,
1586 pvmw.pte, pteval,
1587 vma_mmu_pagesize(vma));
1588 } else {
1589 dec_mm_counter(mm, mm_counter(page));
1590 set_pte_at(mm, address, pvmw.pte, pteval);
1591 }
the page fault check if it's a poison page using is_hwpoison_entry(),
--
Thanks!
Aili Yao
