| Message ID | 1560761476-4651-3-git-send-email-n-horiguchi@ah.jp.nec.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | fix return value issue of soft offlining hugepages | expand |
On Mon, Jun 17, 2019 at 05:51:16PM +0900, Naoya Horiguchi wrote: > madvise(MADV_SOFT_OFFLINE) often returns -EBUSY when calling soft offline > for hugepages with overcommitting enabled. That was caused by the suboptimal > code in current soft-offline code. See the following part: > > ret = migrate_pages(&pagelist, new_page, NULL, MPOL_MF_MOVE_ALL, > MIGRATE_SYNC, MR_MEMORY_FAILURE); > if (ret) { > ... > } else { > /* > * We set PG_hwpoison only when the migration source hugepage > * was successfully dissolved, because otherwise hwpoisoned > * hugepage remains on free hugepage list, then userspace will > * find it as SIGBUS by allocation failure. That's not expected > * in soft-offlining. > */ > ret = dissolve_free_huge_page(page); > if (!ret) { > if (set_hwpoison_free_buddy_page(page)) > num_poisoned_pages_inc(); > } > } > return ret; Hi Naoya, just a nit: > > Here dissolve_free_huge_page() returns -EBUSY if the migration source page > was freed into buddy in migrate_pages(), but even in that case we actually > has a chance that set_hwpoison_free_buddy_page() succeeds. So that means > current code gives up offlining too early now. Maybe it is me that I am not really familiar with hugetlb code, but having had a comment pointing out that the releasing of overcommited hugetlb pages into the buddy allocator happens in migrate_pages()->put_page()->free_huge_page() would have been great. > > dissolve_free_huge_page() checks that a given hugepage is suitable for > dissolving, where we should return success for !PageHuge() case because > the given hugepage is considered as already dissolved. > > This change also affects other callers of dissolve_free_huge_page(), > which are cleaned up together. > > Reported-by: Chen, Jerry T <jerry.t.chen@intel.com> > Tested-by: Chen, Jerry T <jerry.t.chen@intel.com> > Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com> > Fixes: 6bc9b56433b76 ("mm: fix race on soft-offlining") > Cc: <stable@vger.kernel.org> # v4.19+ Reviewed-by: Oscar Salvador <osalvador@suse.de>
On 6/17/19 1:51 AM, Naoya Horiguchi wrote: > madvise(MADV_SOFT_OFFLINE) often returns -EBUSY when calling soft offline > for hugepages with overcommitting enabled. That was caused by the suboptimal > code in current soft-offline code. See the following part: > > ret = migrate_pages(&pagelist, new_page, NULL, MPOL_MF_MOVE_ALL, > MIGRATE_SYNC, MR_MEMORY_FAILURE); > if (ret) { > ... > } else { > /* > * We set PG_hwpoison only when the migration source hugepage > * was successfully dissolved, because otherwise hwpoisoned > * hugepage remains on free hugepage list, then userspace will > * find it as SIGBUS by allocation failure. That's not expected > * in soft-offlining. > */ > ret = dissolve_free_huge_page(page); > if (!ret) { > if (set_hwpoison_free_buddy_page(page)) > num_poisoned_pages_inc(); > } > } > return ret; > > Here dissolve_free_huge_page() returns -EBUSY if the migration source page > was freed into buddy in migrate_pages(), but even in that case we actually > has a chance that set_hwpoison_free_buddy_page() succeeds. So that means > current code gives up offlining too early now. > > dissolve_free_huge_page() checks that a given hugepage is suitable for > dissolving, where we should return success for !PageHuge() case because > the given hugepage is considered as already dissolved. > > This change also affects other callers of dissolve_free_huge_page(), > which are cleaned up together. > > Reported-by: Chen, Jerry T <jerry.t.chen@intel.com> > Tested-by: Chen, Jerry T <jerry.t.chen@intel.com> > Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com> Thanks, Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
diff --git v5.2-rc4/mm/hugetlb.c v5.2-rc4_patched/mm/hugetlb.c index ac843d3..ede7e7f 100644 --- v5.2-rc4/mm/hugetlb.c +++ v5.2-rc4_patched/mm/hugetlb.c @@ -1510,16 +1510,29 @@ static int free_pool_huge_page(struct hstate *h, nodemask_t *nodes_allowed, /* * Dissolve a given free hugepage into free buddy pages. This function does - * nothing for in-use (including surplus) hugepages. Returns -EBUSY if the - * dissolution fails because a give page is not a free hugepage, or because - * free hugepages are fully reserved. + * nothing for in-use hugepages and non-hugepages. + * This function returns values like below: + * + * -EBUSY: failed to dissolved free hugepages or the hugepage is in-use + * (allocated or reserved.) + * 0: successfully dissolved free hugepages or the page is not a + * hugepage (considered as already dissolved) */ int dissolve_free_huge_page(struct page *page) { int rc = -EBUSY; + /* Not to disrupt normal path by vainly holding hugetlb_lock */ + if (!PageHuge(page)) + return 0; + spin_lock(&hugetlb_lock); - if (PageHuge(page) && !page_count(page)) { + if (!PageHuge(page)) { + rc = 0; + goto out; + } + + if (!page_count(page)) { struct page *head = compound_head(page); struct hstate *h = page_hstate(head); int nid = page_to_nid(head); @@ -1564,11 +1577,9 @@ int dissolve_free_huge_pages(unsigned long start_pfn, unsigned long end_pfn) for (pfn = start_pfn; pfn < end_pfn; pfn += 1 << minimum_order) { page = pfn_to_page(pfn); - if (PageHuge(page) && !page_count(page)) { - rc = dissolve_free_huge_page(page); - if (rc) - break; - } + rc = dissolve_free_huge_page(page); + if (rc) + break; } return rc; diff --git v5.2-rc4/mm/memory-failure.c v5.2-rc4_patched/mm/memory-failure.c index 8ee7b16..d9cc660 100644 --- v5.2-rc4/mm/memory-failure.c +++ v5.2-rc4_patched/mm/memory-failure.c @@ -1856,11 +1856,8 @@ static int soft_offline_in_use_page(struct page *page, int flags) static int soft_offline_free_page(struct page *page) { - int rc = 0; - struct page *head = compound_head(page); + int rc = dissolve_free_huge_page(page); - if (PageHuge(head)) - rc = dissolve_free_huge_page(page); if (!rc) { if (set_hwpoison_free_buddy_page(page)) num_poisoned_pages_inc();
madvise(MADV_SOFT_OFFLINE) often returns -EBUSY when calling soft offline for hugepages with overcommitting enabled. That was caused by the suboptimal code in current soft-offline code. See the following part: ret = migrate_pages(&pagelist, new_page, NULL, MPOL_MF_MOVE_ALL, MIGRATE_SYNC, MR_MEMORY_FAILURE); if (ret) { ... } else { /* * We set PG_hwpoison only when the migration source hugepage * was successfully dissolved, because otherwise hwpoisoned * hugepage remains on free hugepage list, then userspace will * find it as SIGBUS by allocation failure. That's not expected * in soft-offlining. */ ret = dissolve_free_huge_page(page); if (!ret) { if (set_hwpoison_free_buddy_page(page)) num_poisoned_pages_inc(); } } return ret; Here dissolve_free_huge_page() returns -EBUSY if the migration source page was freed into buddy in migrate_pages(), but even in that case we actually has a chance that set_hwpoison_free_buddy_page() succeeds. So that means current code gives up offlining too early now. dissolve_free_huge_page() checks that a given hugepage is suitable for dissolving, where we should return success for !PageHuge() case because the given hugepage is considered as already dissolved. This change also affects other callers of dissolve_free_huge_page(), which are cleaned up together. Reported-by: Chen, Jerry T <jerry.t.chen@intel.com> Tested-by: Chen, Jerry T <jerry.t.chen@intel.com> Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com> Fixes: 6bc9b56433b76 ("mm: fix race on soft-offlining") Cc: <stable@vger.kernel.org> # v4.19+ --- ChangeLog v2->v3: - add PageHuge check in dissolve_free_huge_page() outside hugetlb_lock - update comment on dissolve_free_huge_page() about return value --- mm/hugetlb.c | 29 ++++++++++++++++++++--------- mm/memory-failure.c | 5 +---- 2 files changed, 21 insertions(+), 13 deletions(-)