| Message ID | 20210715150852.1655-1-wangbin224@huawei.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | mm/hwpoison: dissolve error hugepages of file mapping | expand |
On Thu, Jul 15, 2021 at 11:08:52PM +0800, Bin Wang wrote: > We handle error on hugetlb file with hugetlbfs_error_remove_page(), > which removes the error hugepages from page cache. And with the > HWPoison flag in head page, it is safe to free the error hugepages > and dissolve it. How is that safe? If you're going to dissolve the huge page, you need to figure out which subpage needs to have the HWPoison flag set.
Hi Matthew, On Thu, 15 Jul 2021 at 16:22:40 +0100, Matthew Wilcox wrote: > How is that safe? If you're going to dissolve the huge page, you > need to figure out which subpage needs to have the HWPoison flag set. Thanks for your review. The p that we pass to the __page_handle_poison() is the error subpage. And we will move the HWPoison flag to subpage in the dissolve_free_huge_page(): if (PageHWPoison(head) && page != head) { SetPageHWPoison(page); ClearPageHWPoison(head); } Thanks, Bin Wang
diff --git a/mm/memory-failure.c b/mm/memory-failure.c index eefd823deb67..6a3c46b601ef 100644 --- a/mm/memory-failure.c +++ b/mm/memory-failure.c @@ -988,6 +988,13 @@ static int me_huge_page(struct page *p, unsigned long pfn) if (mapping) { res = truncate_error_page(hpage, pfn, mapping); unlock_page(hpage); + if (ret == MF_RECOVERED) { + put_page(hpage); + if (__page_handle_poison(p)) + page_ref_inc(p); + else + ret = MF_FAILED; + } } else { res = MF_FAILED; unlock_page(hpage);
We handle error on hugetlb file with hugetlbfs_error_remove_page(), which removes the error hugepages from page cache. And with the HWPoison flag in head page, it is safe to free the error hugepages and dissolve it. Fixes: 78bb920344b8a ("mm: hwpoison: dissolve in-use hugepage in unrecoverable memory error") Signed-off-by: Bin Wang <wangbin224@huawei.com> --- mm/memory-failure.c | 7 +++++++ 1 file changed, 7 insertions(+)