HWPOISON: soft offline rework
Subsystems (7)
[email protected]
reviewerv6 cover HWPOISON: soft offline rework
Hi,
This patchset is the latest version of soft offline rework patchset targetted for v5.9.
Since v5, I dropped some patches which tweak refcount handling in madvise_inject_error() to avoid the "unknown refcount page" error. I don't confirm the fix (that didn't reproduce with v5 in my environment), but this change surely call soft_offline_page() after holding refcount, so the error should not happen any more.
Dropped patches
- mm,madvise: call soft_offline_page() without MF_COUNT_INCREASED
- mm,madvise: Refactor madvise_inject_error
- mm,hwpoison: remove MF_COUNT_INCREASED
- mm,hwpoison: remove flag argument from soft offline functions
Thanks,
Naoya Horiguchi
Quoting cover letter of v5:
----
Main focus of this series is to stabilize soft offline. Historically soft offlined pages have suffered from racy conditions because PageHWPoison is used to a little too aggressively, which (directly or indirectly) invades other mm code which cares little about hwpoison. This results in unexpected behavior or kernel panic, which is very far from soft offline's "do not disturb userspace or other kernel component" policy.
Main point of this change set is to contain target page "via buddy allocator", where we first free the target page as we do for normal pages, and remove from buddy only when we confirm that it reaches free list. There is surely race window of page allocation, but that's fine because someone really want that page and the page is still working, so soft offline can happily give up.
v4 from Oscar tries to handle the race around reallocation, but that part seems still work in progress, so I decide to separate it for changes into v5.9. Thank you for your contribution, Oscar.
---
Previous versions:
v1: https://lore.kernel.org/linux-mm/[email protected]/
v2: https://lore.kernel.org/linux-mm/[email protected]/
v3: https://lore.kernel.org/linux-mm/[email protected]/
v4: https://lore.kernel.org/linux-mm/[email protected]/
v5: https://lore.kernel.org/linux-mm/[email protected]/T/#t
---
Summary:
Naoya Horiguchi (5):
mm,hwpoison: cleanup unused PageHuge() check mm, hwpoison: remove recalculating hpage mm,hwpoison-inject: don't pin for hwpoison_filter mm,hwpoison: introduce MF_MSG_UNSPLIT_THP mm,hwpoison: double-check page count in __get_any_page()
Oscar Salvador (7):
mm,hwpoison: Un-export get_hwpoison_page and make it static mm,hwpoison: Kill put_hwpoison_page
mm,hwpoison: Unify THP handling for hard and soft offline mm,hwpoison: Rework soft offline for free pages mm,hwpoison: Rework soft offline for in-use pages mm,hwpoison: Refactor soft_offline_huge_page and __soft_offline_page mm,hwpoison: Return 0 if the page is already poisoned in soft-offline
include/linux/mm.h | 3 +- include/linux/page-flags.h | 6 +- include/ras/ras_event.h | 3 + mm/hwpoison-inject.c | 18 +-- mm/madvise.c | 5 - mm/memory-failure.c | 307 +++++++++++++++++++++------------------------ mm/migrate.c | 11 +- mm/page_alloc.c | 60 +++++++-- 8 files changed, 203 insertions(+), 210 deletions(-)
[email protected]
reviewerPATCH v6 1/12 mm,hwpoison: cleanup unused PageHuge() check
From: Naoya Horiguchi <[email protected]>
Drop the PageHuge check, which is dead code since memory_failure() forks into memory_failure_hugetlb() for hugetlb pages.
memory_failure() and memory_failure_hugetlb() shares some functions like hwpoison_user_mappings() and identify_page_state(), so they should properly handle 4kB page, thp, and hugetlb.
--- mm/memory-failure.c | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-)
mm/memory-failure.c +1 -4
@@ -1382,10 +1382,7 @@ int memory_failure(unsigned long pfn, int flags)* page_remove_rmap() in try_to_unmap_one(). So to determine page status * correctly, we save a copy of the page flags at this time. */ -if (PageHuge (p)) - page_flags = hpage->flags; -else - page_flags = p->flags; + page_flags = p->flags;/* * unpoison always clear PG_hwpoison inside page lock
Signature · 1 line
2.17.1
[email protected]
reviewerPATCH v6 2/12 mm, hwpoison: remove recalculating hpage
In reply to [email protected]From: Naoya Horiguchi <[email protected]>
hpage is never used after try_to_split_thp_page() in memory_failure(), so we don't have to update hpage. So let's not recalculate/use hpage.
--- mm/memory-failure.c | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-)
mm/memory-failure.c +1 -5
@@ -1342,7 +1342,6 @@ int memory_failure(unsigned long pfn, int flags) }unlock_page (p);VM_BUG_ON_PAGE (!page_count (p), p); - hpage =compound_head (p); }/*
@@ -1414,11 +1413,8 @@ int memory_failure(unsigned long pfn, int flags)/* * Now take care of user space mappings. * Abort on fail: __delete_from_page_cache() assumes unmapped page. -* -* When the raw error page is thp tail page, hpage points to the raw -* page after thp split. */ -if (!hwpoison_user_mappings (p, pfn, flags, &hpage)) { +if (!hwpoison_user_mappings (p, pfn, flags, &p)) {action_result (pfn,MF_MSG_UNMAP_FAILED ,MF_IGNORED ); res = -EBUSY ; goto out;
Signature · 1 line
2.17.1
[email protected]
reviewerPATCH v6 3/12 mm,hwpoison-inject: don't pin for hwpoison_filter
In reply to [email protected]From: Naoya Horiguchi <[email protected]>
Another memory error injection interface debugfs:hwpoison/corrupt-pfn also takes bogus refcount for hwpoison_filter(). It's justified because this does a coarse filter, expecting that memory_failure() redoes the check for sure.
--- mm/hwpoison-inject.c | 18 +++++------------- 1 file changed, 5 insertions(+), 13 deletions(-)
mm/hwpoison-inject.c +5 -13
@@ -26,11 +26,6 @@ static int hwpoison_inject(void *data, u64 val) p =pfn_to_page (pfn); hpage =compound_head (p); -/* -* This implies unable to support free buddy pages. -*/ -if (!get_hwpoison_page (p)) -return 0 ;if (!hwpoison_filter_enable) goto inject;
@@ -40,23 +35,20 @@ static int hwpoison_inject(void *data, u64 val)* This implies unable to support non-LRU pages. */ if (!PageLRU (hpage) && !PageHuge (p)) - goto put_out; +return 0 ;/* -* do a racy check with elevated page count, to make sure PG_hwpoison -* will only be set for the targeted owner (or on a free page). +* do a racy check to make sure PG_hwpoison will only be set for +* the targeted owner (or on a free page). * memory_failure() will redo the check reliably inside page lock. */ err =hwpoison_filter (hpage);if (err) - goto put_out; +return 0 ; inject:pr_info ("Injecting memory failure at pfn %#lx\n" , pfn); -return memory_failure (pfn,MF_COUNT_INCREASED ); -put_out: -put_hwpoison_page (p); -return 0 ; +return memory_failure (pfn,0 ); }static int hwpoison_unpoison (void *data,u64 val)
Signature · 1 line
2.17.1
[email protected]
reviewerPATCH v6 4/12 mm,hwpoison: Un-export get_hwpoison_page and make it static
In reply to [email protected]From: Oscar Salvador <[email protected]>
Since get_hwpoison_page is only used in memory-failure code now, let us un-export it and make it private to that code.
--- include/linux/mm.h | 1 - mm/memory-failure.c | 3 +-- 2 files changed, 1 insertion(+), 3 deletions(-)
include/linux/mm.h +0 -1
@@ -2985,7 +2985,6 @@ extern int memory_failure(unsigned long pfn, int flags);extern void memory_failure_queue (unsigned long pfn,int flags);extern void memory_failure_queue_kick (int cpu);extern int unpoison_memory (unsigned long pfn); -extern int get_hwpoison_page (struct page *page);#define put_hwpoison_page (page) put_page(page)extern int sysctl_memory_failure_early_kill;extern int sysctl_memory_failure_recovery;
mm/memory-failure.c +1 -2
@@ -925,7 +925,7 @@ static int page_action(struct page_state *ps, struct page *p,* Return: return 0 if failed to grab the refcount, otherwise true (some * non-zero value.) */ -int get_hwpoison_page (struct page *page) +static int get_hwpoison_page (struct page *page) {struct page *head =compound_head (page);
@@ -954,7 +954,6 @@ int get_hwpoison_page(struct page *page)return 0 ; } -EXPORT_SYMBOL_GPL (get_hwpoison_page);/* * Do all that is necessary to remove user space mappings. Unmap
Signature · 1 line
2.17.1
[email protected]
reviewerPATCH v6 5/12 mm,hwpoison: Kill put_hwpoison_page
In reply to [email protected]From: Oscar Salvador <[email protected]>
After commit 4e41a30c6d50 ("mm: hwpoison: adjust for new thp refcounting"), put_hwpoison_page got reduced to a put_page.
Let us just use put_page instead.
--- include/linux/mm.h | 1 - mm/memory-failure.c | 30 +++++++++++++++--------------- 2 files changed, 15 insertions(+), 16 deletions(-)
include/linux/mm.h +0 -1
@@ -2985,7 +2985,6 @@ extern int memory_failure(unsigned long pfn, int flags);extern void memory_failure_queue (unsigned long pfn,int flags);extern void memory_failure_queue_kick (int cpu);extern int unpoison_memory (unsigned long pfn); -#define put_hwpoison_page (page) put_page(page)extern int sysctl_memory_failure_early_kill;extern int sysctl_memory_failure_recovery;extern void shake_page (struct page *p,int access);
mm/memory-failure.c +15 -15
@@ -1144,7 +1144,7 @@ static int memory_failure_hugetlb(unsigned long pfn, int flags)pr_err ("Memory failure: %#lx: just unpoisoned\n" , pfn);num_poisoned_pages_dec ();unlock_page (head); -put_hwpoison_page (head); +put_page (head);return 0 ; }
@@ -1336,7 +1336,7 @@ int memory_failure(unsigned long pfn, int flags) pfn);if (TestClearPageHWPoison (p))num_poisoned_pages_dec (); -put_hwpoison_page (p); +put_page (p);return -EBUSY ; }unlock_page (p);
@@ -1389,14 +1389,14 @@ int memory_failure(unsigned long pfn, int flags)pr_err ("Memory failure: %#lx: just unpoisoned\n" , pfn);num_poisoned_pages_dec ();unlock_page (p); -put_hwpoison_page (p); +put_page (p);return 0 ; }if (hwpoison_filter (p)) {if (TestClearPageHWPoison (p))num_poisoned_pages_dec ();unlock_page (p); -put_hwpoison_page (p); +put_page (p);return 0 ; }
@@ -1630,9 +1630,9 @@ int unpoison_memory(unsigned long pfn) }unlock_page (page); -put_hwpoison_page (page); +put_page (page);if (freeit && !(pfn ==my_zero_pfn (0 ) &&page_count (p) ==1 )) -put_hwpoison_page (page); +put_page (page);return 0 ; }
@@ -1683,7 +1683,7 @@ static int get_any_page(struct page *page, unsigned long pfn, int flags)/* * Try to free it. */ -put_hwpoison_page (page); +put_page (page);shake_page (page,1 );/*
@@ -1692,7 +1692,7 @@ static int get_any_page(struct page *page, unsigned long pfn, int flags) ret =__get_any_page (page, pfn,0 );if (ret ==1 && !PageLRU (page)) {/* Drop page reference which is from __get_any_page() */ -put_hwpoison_page (page); +put_page (page);pr_info ("soft_offline: %#lx: unknown non LRU page type %lx (%pGp)\n" , pfn, page->flags, &page->flags);return -EIO ;
@@ -1715,7 +1715,7 @@ static int soft_offline_huge_page(struct page *page, int flags)lock_page (hpage);if (PageHWPoison (hpage)) {unlock_page (hpage); -put_hwpoison_page (hpage); +put_page (hpage);pr_info ("soft offline: %#lx hugepage already poisoned\n" , pfn);return -EBUSY ; }
@@ -1726,7 +1726,7 @@ static int soft_offline_huge_page(struct page *page, int flags)* get_any_page() and isolate_huge_page() takes a refcount each, * so need to drop one here. */ -put_hwpoison_page (hpage); +put_page (hpage);if (!ret) {pr_info ("soft offline: %#lx hugepage failed to isolate\n" , pfn);return -EBUSY ;
@@ -1779,7 +1779,7 @@ static int __soft_offline_page(struct page *page, int flags)wait_on_page_writeback (page);if (PageHWPoison (page)) {unlock_page (page); -put_hwpoison_page (page); +put_page (page);pr_info ("soft offline: %#lx page already poisoned\n" , pfn);return -EBUSY ; }
@@ -1794,7 +1794,7 @@ static int __soft_offline_page(struct page *page, int flags)* would need to fix isolation locking first. */ if (ret ==1 ) { -put_hwpoison_page (page); +put_page (page);pr_info ("soft_offline: %#lx: invalidated\n" , pfn);SetPageHWPoison (page);num_poisoned_pages_inc ();
@@ -1814,7 +1814,7 @@ static int __soft_offline_page(struct page *page, int flags)* Drop page reference which is came from get_any_page() * successful isolate_lru_page() already took another one. */ -put_hwpoison_page (page); +put_page (page);if (!ret) {LIST_HEAD (pagelist);/*
@@ -1858,7 +1858,7 @@ static int soft_offline_in_use_page(struct page *page, int flags)pr_info ("soft offline: %#lx: non anonymous thp\n" ,page_to_pfn (page)); else pr_info("soft offline: %#lx: thp split failed\n", page_to_pfn(page)); -put_hwpoison_page (page); +put_page (page);return -EBUSY ; }unlock_page (page);
@@ -1931,7 +1931,7 @@ int soft_offline_page(unsigned long pfn, int flags)if (PageHWPoison (page)) {pr_info ("soft offline: %#lx page already poisoned\n" , pfn);if (flags &MF_COUNT_INCREASED ) -put_hwpoison_page (page); +put_page (page);return -EBUSY ; }
Signature · 1 line
2.17.1
[email protected]
reviewerPATCH v6 6/12 mm,hwpoison: Unify THP handling for hard and soft offline
In reply to [email protected]From: Oscar Salvador <[email protected]>
Place the THP's page handling in a helper and use it from both hard and soft-offline machinery, so we get rid of some duplicated code.
--- mm/memory-failure.c | 48 +++++++++++++++++++++------------------------ 1 file changed, 22 insertions(+), 26 deletions(-)
mm/memory-failure.c +22 -26
@@ -1103,6 +1103,25 @@ static int identify_page_state(unsigned long pfn, struct page *p,return page_action (ps, p, pfn); } +static int try_to_split_thp_page (struct page *page,const char *msg) +{ +lock_page (page); +if (!PageAnon (page) ||unlikely (split_huge_page (page))) { +unsigned long pfn =page_to_pfn (page); + +unlock_page (page); +if (!PageAnon (page)) +pr_info ("%s: %#lx: non anonymous thp\n" , msg, pfn); +else +pr_info ("%s: %#lx: thp split failed\n" , msg, pfn); +put_page (page); +return -EBUSY ; + } +unlock_page (page); + +return 0 ; +} +static int memory_failure_hugetlb (unsigned long pfn,int flags) {struct page *p =pfn_to_page (pfn);
@@ -1325,21 +1344,8 @@ int memory_failure(unsigned long pfn, int flags) }if (PageTransHuge (hpage)) { -lock_page (p); -if (!PageAnon (p) ||unlikely (split_huge_page (p))) { -unlock_page (p); -if (!PageAnon (p)) -pr_err ("Memory failure: %#lx: non anonymous thp\n" , - pfn); -else -pr_err ("Memory failure: %#lx: thp split failed\n" , - pfn); -if (TestClearPageHWPoison (p)) -num_poisoned_pages_dec (); -put_page (p); +if (try_to_split_thp_page (p,"Memory Failure" ) <0 )return -EBUSY ; - } -unlock_page (p);VM_BUG_ON_PAGE (!page_count (p), p); }
@@ -1850,19 +1856,9 @@ static int soft_offline_in_use_page(struct page *page, int flags)int mt;struct page *hpage =compound_head (page); -if (!PageHuge (page) &&PageTransHuge (hpage)) { -lock_page (page); -if (!PageAnon (page) ||unlikely (split_huge_page (page))) { -unlock_page (page); -if (!PageAnon (page)) -pr_info ("soft offline: %#lx: non anonymous thp\n" ,page_to_pfn (page)); -else -pr_info ("soft offline: %#lx: thp split failed\n" ,page_to_pfn (page)); -put_page (page); +if (!PageHuge (page) &&PageTransHuge (hpage)) +if (try_to_split_thp_page (page,"soft offline" ) <0 )return -EBUSY ; - } -unlock_page (page); - }/* * Setting MIGRATE_ISOLATE here ensures that the page will be linked
Signature · 1 line
2.17.1
[email protected]
reviewerPATCH v6 7/12 mm,hwpoison: Rework soft offline for free pages
In reply to [email protected]From: Oscar Salvador <[email protected]>
When trying to soft-offline a free page, we need to first take it off the buddy allocator.
Once we know is out of reach, we can safely flag it as poisoned.
take_page_off_buddy will be used to take a page meant to be poisoned off the buddy allocator.take_page_off_buddy calls break_down_buddy_pages, which splits a higher-order page in case our page belongs to one.
Once the page is under our control, we call page_handle_poison to set it as poisoned and grab a refcount on it.
---
ChangeLog v4 -> v5:
- fix comile error
ChangeLog v2 -> v3:
- use add_to_free_list() instead of add_to_free_area()
- use del_page_from_free_list() instead of del_page_from_free_area()
- add fast return
- move extern definition to header file as warned by checkpatch.pl
--- include/linux/page-flags.h | 1 + mm/memory-failure.c | 18 ++++++---- mm/page_alloc.c | 68 ++++++++++++++++++++++++++++++++++++++ 3 files changed, 81 insertions(+), 6 deletions(-)
include/linux/page-flags.h +1 -0
@@ -423,6 +423,7 @@ PAGEFLAG(HWPoison, hwpoison, PF_ANY) TESTSCFLAG(HWPoison, hwpoison, PF_ANY)#define __PG_HWPOISON (1UL << PG_hwpoison)extern bool set_hwpoison_free_buddy_page (struct page *page); +extern bool take_page_off_buddy (struct page *page);#else PAGEFLAG_FALSE (HWPoison) static inline bool set_hwpoison_free_buddy_page(struct page *page)
mm/memory-failure.c +12 -6
@@ -65,6 +65,13 @@ int sysctl_memory_failure_recovery __read_mostly = 1;atomic_long_t num_poisoned_pages__read_mostly =ATOMIC_LONG_INIT (0 ); +static void page_handle_poison (struct page *page) +{ +SetPageHWPoison (page); +page_ref_inc (page); +num_poisoned_pages_inc (); +} +#if defined(CONFIG_HWPOISON_INJECT ) || defined(CONFIG_HWPOISON_INJECT_MODULE )u32 hwpoison_filter_enable =0 ;
@@ -1879,14 +1886,13 @@ static int soft_offline_in_use_page(struct page *page, int flags)static int soft_offline_free_page (struct page *page) { -int rc =dissolve_free_huge_page (page); +int rc = -EBUSY ; -if (!rc) { -if (set_hwpoison_free_buddy_page (page)) -num_poisoned_pages_inc (); -else - rc = -EBUSY ; +if (!dissolve_free_huge_page (page) &&take_page_off_buddy (page)) { +page_handle_poison (page); + rc =0 ; } +return rc; }
mm/page_alloc.c +68 -0
74 lines not on this page · raw message
Signature · 1 line
2.17.1
[email protected]
reviewerPATCH v6 8/12 mm,hwpoison: Rework soft offline for in-use pages
In reply to [email protected]From: Oscar Salvador <[email protected]>
This patch changes the way we set and handle in-use poisoned pages. Until now, poisoned pages were released to the buddy allocator, trusting that the checks that take place prior to hand the page would act as a safe net and would skip that page.
This has proved to be wrong, as we got some pfn walkers out there, like compaction, that all they care is the page to be PageBuddy and be in a freelist.
Although this might not be the only user, having poisoned pages in the buddy allocator seems a bad idea as we should only have free pages that are ready and meant to be used as such.
Before explaining the taken approach, let us break down the kind of pages we can soft offline.
- Anonymous THP (after the split, they end up being 4K pages)
- Hugetlb
- Order-0 pages (that can be either migrated or invalited)
* Normal pages (order-0 and anon-THP)
- If they are clean and unmapped page cache pages, we invalidate then by means of invalidate_inode_page().
- If they are mapped/dirty, we do the isolate-and-migrate dance.
Either way, do not call put_page directly from those paths. Instead, we keep the page and send it to page_set_poison to perform the right handling.
page_set_poison sets the HWPoison flag and does the last put_page. This call to put_page is mainly to be able to call __page_cache_release, since this function is not exported.
Down the chain, we placed a check for HWPoison page in free_pages_prepare, that just skips any poisoned page, so those pages do not end up in any pcplist/freelist.
After that, we set the refcount on the page to 1 and we increment the poisoned pages counter.
We could do as we do for free pages:
1) wait until the page hits buddy's freelists
2) take it off
3) flag it
The problem is that we could race with an allocation, so by the time we want to take the page off the buddy, the page is already allocated, so we cannot soft-offline it.
This is not fatal of course, but if it is better if we can close the race as does not require a lot of code.
* Hugetlb pages
- We isolate-and-migrate them
After the migration has been successful, we call dissolve_free_huge_page, and we set HWPoison on the page if we succeed.
Hugetlb has a slightly different handling though.
While for non-hugetlb pages we cared about closing the race with an allocation, doing so for hugetlb pages requires quite some additional code (we would need to hook in free_huge_page and some other places). So I decided to not make the code overly complicated and just fail normally if the page we allocated in the meantime.
Because of the way we handle now in-use pages, we no longer need the put-as-isolation-migratetype dance, that was guarding for poisoned pages to end up in pcplists.
--- include/linux/page-flags.h | 5 ----- mm/memory-failure.c | 45 ++++++++++++++------------------------ mm/migrate.c | 11 +++------- mm/page_alloc.c | 28 ------------------------ 4 files changed, 19 insertions(+), 70 deletions(-)
include/linux/page-flags.h +0 -5
@@ -422,14 +422,9 @@ PAGEFLAG_FALSE(Uncached) PAGEFLAG(HWPoison, hwpoison, PF_ANY) TESTSCFLAG(HWPoison, hwpoison, PF_ANY)#define __PG_HWPOISON (1UL << PG_hwpoison) -extern bool set_hwpoison_free_buddy_page (struct page *page);extern bool take_page_off_buddy (struct page *page);#else PAGEFLAG_FALSE(HWPoison) -static inline bool set_hwpoison_free_buddy_page(struct page *page) -{ -return 0 ; -}#define __PG_HWPOISON 0#endif
mm/memory-failure.c +16 -29
94 lines not on this page · raw message
mm/migrate.c +3 -8
@@ -1222,16 +1222,11 @@ static int unmap_and_move(new_page_t get_new_page,* we want to retry. */ if (rc ==MIGRATEPAGE_SUCCESS ) { -put_page (page); -if (reason ==MR_MEMORY_FAILURE ) { +if (reason !=MR_MEMORY_FAILURE )/* -* Set PG_HWPoison on just freed page -* intentionally. Although it's rather weird, -* it's how HWPoison flag works at the moment. +* We release the page in page_handle_poison. */ -if (set_hwpoison_free_buddy_page (page)) -num_poisoned_pages_inc (); - } +put_page (page); }else {if (rc != -EAGAIN ) {if (likely (!__PageMovable (page))) {
mm/page_alloc.c +0 -28
@@ -8843,32 +8843,4 @@ bool take_page_off_buddy(struct page *page)spin_unlock_irqrestore (&zone->lock, flags); return ret; } - -/* -* Set PG_hwpoison flag if a given page is confirmed to be a free page. This -* test is performed under the zone lock to prevent a race against page -* allocation. -*/ -bool set_hwpoison_free_buddy_page (struct page *page) -{ -struct zone *zone =page_zone (page); -unsigned long pfn =page_to_pfn (page); -unsigned long flags; -unsigned int order; -bool hwpoisoned = false; - -spin_lock_irqsave (&zone->lock, flags); -for (order =0 ; order <MAX_ORDER ; order++) { -struct page *page_head = page - (pfn & ((1 << order) -1 )); - -if (PageBuddy (page_head) &&page_order (page_head) >= order) { -if (!TestSetPageHWPoison (page)) - hwpoisoned = true; -break ; - } - } -spin_unlock_irqrestore (&zone->lock, flags); - -return hwpoisoned; -}#endif
Signature · 1 line
2.17.1
Re: [PATCH v6 08/12] mm,hwpoison: Rework soft offline for in-use pages
Quoted 73 lines · On 2020-08-06 20:49, [email protected] wrote:
From: Oscar Salvador <[email protected]>
This patch changes the way we set and handle in-use poisoned pages. Until now, poisoned pages were released to the buddy allocator, trusting that the checks that take place prior to hand the page would act as a safe net and would skip that page.
This has proved to be wrong, as we got some pfn walkers out there, like compaction, that all they care is the page to be PageBuddy and be in a freelist.
Although this might not be the only user, having poisoned pages in the buddy allocator seems a bad idea as we should only have free pages that are ready and meant to be used as such.Before explaining the taken approach, let us break down the kind of pages we can soft offline.
- Anonymous THP (after the split, they end up being 4K pages)
- Hugetlb
- Order-0 pages (that can be either migrated or invalited)* Normal pages (order-0 and anon-THP)
- If they are clean and unmapped page cache pages, we invalidate then by means of
invalidate_inode_page().
- If they are mapped/dirty, we do the isolate-and-migrate dance.Either way, do not call
put_pagedirectly from those paths. Instead, we keep the page and send it topage_set_poisonto perform
the
right handling.
page_set_poisonsets the HWPoison flag and does the lastput_page. This call toput_pageis mainly to be able to call__page_cache_release,
since this function is not exported.Down the chain, we placed a check for HWPoison page in
free_pages_prepare, that just skips any poisoned page, so those pages do not end up in any pcplist/freelist.After that, we set the refcount on the page to 1 and we increment the poisoned pages counter.
We could do as we do for free pages:
1) wait until the page hits buddy's freelists
2) take it off
3) flag itThe problem is that we could race with an allocation, so by the time
we
want to take the page off the buddy, the page is already allocated,
so we
cannot soft-offline it.
This is not fatal of course, but if it is better if we can close the
race
as does not require a lot of code.* Hugetlb pages
- We isolate-and-migrate them
After the migration has been successful, we call
dissolve_free_huge_page,
and we set HWPoison on the page if we succeed. Hugetlb has a slightly different handling though.While for non-hugetlb pages we cared about closing the race with an allocation, doing so for hugetlb pages requires quite some additional code (we would need to hook in
free_huge_pageand some other places). So I decided to not make the code overly complicated and just fail normally if the page we allocated in the meantime.
Quoted 6 lines
Because of the way we handle now in-use pages, we no longer need the put-as-isolation-migratetype dance, that was guarding for poisoned pages to end up in pcplists.
Signed-off-by Oscar Salvador <[email protected]>Signed-off-by Naoya Horiguchi <[email protected]>
Hi Andrew,
I just found out yesterday that the patchset Naoya sent has diverged from mine in some aspects that lead to some bugs [1]. This was due to a misunderstanding so no blame here.
So, patch#8 and patch#9 need a little tweak [2].
I was wondering what do you prefer?
1) I paste the chunks that need to be changed in the offending patches (this and patch#9)
2) I just send them as standalone patches and you applied them on top
I am asking this because although patches had hit -next, we might still have time to change them. If not, do not worry, I will send them as standalone.
[1] https://patchwork.kernel.org/comment/23622881/
[2] https://patchwork.kernel.org/comment/23622985/
I will go ahead and paste the chunks here, in case you lean towards option#1:
Patch#8:
mm/memory-failure.c +0 -0
@@ -67,11 +67,6 @@ atomic_long_t num_poisoned_pages __read_mostly = ATOMIC_LONG_INIT(0); static bool page_handle_poison(struct page *page, bool
hugepage_or_freepage, bool release)
{
- if (release) {
- put_page(page);
- drain_all_pages(page_zone(page));
- }
-
if (hugepage_or_freepage) {
/*
* Doing this check for free pages is also fine sincedissolve_free_huge_page
@@ -89,6 +84,12 @@ static bool page_handle_poison(struct page *page, bool hugepage_or_freepage, boo
}
SetPageHWPoison(page);
+
+ if (release) {
+ put_page(page);
+ drain_all_pages(page_zone(page));
+ }
+
page_ref_inc(page);
num_poisoned_pages_inc();
return true;
mm/page_alloc.c +10 -0
@@ -1173,6 +1173,16 @@ static __always_inline bool free_pages_prepare(struct page *page,trace_mm_page_free (page, order); +if (unlikely (PageHWPoison (page)) && !order) { +/* +* Untie memcg state and reset page's owner +*/ +if (memcg_kmem_enabled () &&PageKmemcg (page)) +__memcg_kmem_uncharge_page (page, order); +reset_page_owner (page, order); +return false; + } +/* * Check tail pages before head page information is cleared to * avoid checking PageCompound for order-0 pag
Patch#9:
mm/memory-failure.c +1 -1
@@ -1835,7 +1835,7 @@ static int __soft_offline_page(struct page *page)if (!ret) {bool release = !huge; -if (!page_handle_poison (page, true, release)) +if (!page_handle_poison (page, huge, release)) ret = -EBUSY ; }else { if (!list_empty(&pagelist)
Thanks ans sorry for the inconvenience.
Andrew Morton
maintainerRe: [PATCH v6 08/12] mm,hwpoison: Rework soft offline for in-use pages
Quoted 6 lines · On Fri, 18 Sep 2020 09:58:22 +0200 [email protected] wrote:
I just found out yesterday that the patchset Naoya sent has diverged from mine in some aspects that lead to some bugs [1]. This was due to a misunderstanding so no blame here.
So, patch#8 and patch#9 need a little tweak [2].I was wondering what do you prefer?
Well. I (and I suspect the rest of the world) have lost track of what's going on here.
So please let's have a full resend of the whole series?
Re: [PATCH v6 08/12] mm,hwpoison: Rework soft offline for in-use pages
Quoted 7 lines · Andrew Morton
Quoted 5 lines · On Fri, 18 Sep 2020 09:58:22 +0200 [email protected] wrote:
I just found out yesterday that the patchset Naoya sent has diverged from mine in some aspects that lead to some bugs [1]. This was due to a misunderstanding so no blame here.
So, patch#8 and patch#9 need a little tweak [2].
Quoted 6 lines
Quoted 1 line
I was wondering what do you prefer?
Well. I (and I suspect the rest of the world) have lost track of what's going on here.
So please let's have a full resend of the whole series?
Sure, I will resend a new version squeezing all changes into it on Monday.
[email protected]
reviewerPATCH v6 9/12 mm,hwpoison: Refactor soft_offline_huge_page and __soft_offline_page
In reply to [email protected]From: Oscar Salvador <[email protected]>
Merging soft_offline_huge_page and __soft_offline_page let us get rid of quite some duplicated code, and makes the code much easier to follow.
Now, __soft_offline_page will handle both normal and hugetlb pages.
Note that move put_page() block to the beginning of page_handle_poison() with drain_all_pages() in order to make sure that the target page is freed and sent into free list to make take_page_off_buddy() work properly.
---
ChangeLog v4 -> v5:
- use "char const *msg_page[]" instead of "const char *msg_page[]"
- move adding drain_all_pages() to 12/16
ChangeLog v2 -> v3:
- use page_is_file_lru() instead of page_is_file_cache(),
- add description about put_page() and drain_all_pages().
- fix coding style warnings by checkpatch.pl
--- mm/memory-failure.c | 183 ++++++++++++++++++++------------------------ 1 file changed, 83 insertions(+), 100 deletions(-)
mm/memory-failure.c +83 -100
@@ -65,15 +65,33 @@ int sysctl_memory_failure_recovery __read_mostly = 1;atomic_long_t num_poisoned_pages__read_mostly =ATOMIC_LONG_INIT (0 ); -static void page_handle_poison (struct page *page,bool release) +static bool page_handle_poison (struct page *page,bool hugepage_or_freepage,bool release) {if (release) {put_page (page);drain_all_pages (page_zone (page)); } + +if (hugepage_or_freepage) { +/* +* Doing this check for free pages is also fine since dissolve_free_huge_page +* returns 0 for non-hugetlb pages as well. +*/ +if (dissolve_free_huge_page (page) || !take_page_off_buddy (page)) +/* +* We could fail to take off the target page from buddy +* for example due to racy page allocaiton, but that's +* acceptable because soft-offlined page is not broken +* and if someone really want to use it, they should +* take it. +*/ +return false; + } +SetPageHWPoison (page);page_ref_inc (page);num_poisoned_pages_inc (); +return true; }#if defined(CONFIG_HWPOISON_INJECT ) || defined(CONFIG_HWPOISON_INJECT_MODULE )
@@ -1718,63 +1736,52 @@ static int get_any_page(struct page *page, unsigned long pfn, int flags)return ret; } -static int soft_offline_huge_page (struct page *page,int flags) +static bool isolate_page (struct page *page,struct list_head *pagelist) { -int ret; -unsigned long pfn =page_to_pfn (page); -struct page *hpage =compound_head (page); -LIST_HEAD (pagelist); +bool isolated = false; +bool lru =PageLRU (page); + +if (PageHuge (page)) { + isolated =isolate_huge_page (page, pagelist); + }else { +if (lru) + isolated = !isolate_lru_page (page); +else + isolated = !isolate_movable_page (page,ISOLATE_UNEVICTABLE ); + +if (isolated) +list_add (&page->lru, pagelist); -/* -* This double-check of PageHWPoison is to avoid the race with -* memory_failure(). See also comment in __soft_offline_page(). -*/ -lock_page (hpage); -if (PageHWPoison (hpage)) { -unlock_page (hpage); -put_page (hpage); -pr_info ("soft offline: %#lx hugepage already poisoned\n" , pfn); -return -EBUSY ; } -unlock_page (hpage); - ret =isolate_huge_page (hpage, &pagelist); +if (isolated && lru) +inc_node_page_state (page,NR_ISOLATED_ANON + +page_is_file_lru (page)); +/* -* get_any_page() and isolate_huge_page() takes a refcount each, -* so need to drop one here. +* If we succeed to isolate the page, we grabbed another refcount on +* the page, so we can safely drop the one we got from get_any_pages(). +* If we failed to isolate the page, it means that we cannot go further +* and we will return an error, so drop the reference we got from +* get_any_pages() as well. */ -put_page (hpage); -if (!ret) { -pr_info ("soft offline: %#lx hugepage failed to isolate\n" , pfn); -return -EBUSY ; - } - - ret =migrate_pages (&pagelist, new_page,NULL ,MPOL_MF_MOVE_ALL , -MIGRATE_SYNC ,MR_MEMORY_FAILURE ); -if (ret) { -pr_info ("soft offline: %#lx: hugepage migration failed %d, type %lx (%pGp)\n" , - pfn, ret, page->flags, &page->flags); -if (!list_empty (&pagelist)) -putback_movable_pages (&pagelist); -if (ret >0 ) - ret = -EIO ; - }else { -/* -* We set PG_hwpoison only when we were able to take the page -* off the buddy. -*/ -if (!dissolve_free_huge_page (page) &&take_page_off_buddy (page)) -page_handle_poison (page, false); -else - ret = -EBUSY ; - } -return ret; +put_page (page); +return isolated; } -static int __soft_offline_page (struct page *page,int flags) +/* +* __soft_offline_page handles hugetlb-pages and non-hugetlb pages. +* If the page is a non-dirty unmapped page-cache page, it simply invalidates. +* If the page is mapped, it migrates the contents over. +*/ +static int __soft_offline_page (struct page *page) { -int ret; +int ret =0 ;unsigned long pfn =page_to_pfn (page); +struct page *hpage =compound_head (page); +char const *msg_page[] = {"page" ,"hugepage" }; +bool huge =PageHuge (page); +LIST_HEAD (pagelist);struct migration_target_control mtc = { .nid =NUMA_NO_NODE , .gfp_mask =GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL,
@@ -1787,98 +1794,74 @@ static int __soft_offline_page(struct page *page, int flags)* so there's no race between soft_offline_page() and memory_failure(). */ lock_page (page); -wait_on_page_writeback (page); +if (!PageHuge (page)) +wait_on_page_writeback (page);if (PageHWPoison (page)) {unlock_page (page);put_page (page);pr_info ("soft offline: %#lx page already poisoned\n" , pfn);return -EBUSY ; } -/* -* Try to invalidate first. This should work for -* non dirty unmapped page cache pages. -*/ - ret =invalidate_inode_page (page); + +if (!PageHuge (page)) +/* +* Try to invalidate first. This should work for +* non dirty unmapped page cache pages. +*/ + ret =invalidate_inode_page (page);unlock_page (page); +/* * RED-PEN would be better to keep it isolated here, but we * would need to fix isolation locking first. */ -if (ret ==1 ) { +if (ret) {pr_info ("soft_offline: %#lx: invalidated\n" , pfn); -page_handle_poison (page, true); +page_handle_poison (page, false, true);return 0 ; } -/* -* Simple invalidation didn't work. -* Try to migrate to a new page instead. migrate.c -* handles a large number of cases for us. -*/ -if (PageLRU (page)) - ret =isolate_lru_page (page); -else - ret =isolate_movable_page (page,ISOLATE_UNEVICTABLE ); -/* -* Drop page reference which is came from get_any_page() -* successful isolate_lru_page() already took another one. -*/ -put_page (page); -if (!ret) { -LIST_HEAD (pagelist); -/* -* After isolated lru page, the PageLRU will be cleared, -* so use !__PageMovable instead for LRU page's mapping -* cannot have PAGE_MAPPING_MOVABLE. -*/ -if (!__PageMovable (page)) -inc_node_page_state (page,NR_ISOLATED_ANON + -page_is_file_lru (page)); -list_add (&page->lru, &pagelist); +if (isolate_page (hpage, &pagelist)) { ret =migrate_pages (&pagelist, alloc_migration_target,NULL , (unsigned long )&mtc,MIGRATE_SYNC ,MR_MEMORY_FAILURE );if (!ret) { -page_handle_poison (page, true); +bool release = !huge; + +if (!page_handle_poison (page, true, release)) + ret = -EBUSY ; }else {if (!list_empty (&pagelist))putback_movable_pages (&pagelist); -pr_info ("soft offline: %#lx: migration failed %d, type %lx (%pGp)\n" , - pfn, ret, page->flags, &page->flags); + +pr_info ("soft offline: %#lx: %s migration failed %d, type %lx (%pGp)\n" , + pfn, msg_page[huge], ret, page->flags, &page->flags);if (ret >0 ) ret = -EIO ; } }else { -pr_info ("soft offline: %#lx: isolation failed: %d, page count %d, type %lx (%pGp)\n" , - pfn, ret,page_count (page), page->flags, &page->flags); +pr_info ("soft offline: %#lx: %s isolation failed: %d, page count %d, type %lx (%pGp)\n" , + pfn, msg_page[huge], ret,page_count (page), page->flags, &page->flags); }return ret; } -static int soft_offline_in_use_page (struct page *page,int flags) +static int soft_offline_in_use_page (struct page *page) { -int ret;struct page *hpage =compound_head (page);if (!PageHuge (page) &&PageTransHuge (hpage))if (try_to_split_thp_page (page,"soft offline" ) <0 )return -EBUSY ; - -if (PageHuge (page)) - ret =soft_offline_huge_page (page, flags); -else - ret =__soft_offline_page (page, flags); -return ret; +return __soft_offline_page (page); }static int soft_offline_free_page (struct page *page) { -int rc = -EBUSY ; +int rc =0 ; -if (!dissolve_free_huge_page (page) &&take_page_off_buddy (page)) { -page_handle_poison (page, false); - rc =0 ; - } +if (!page_handle_poison (page, true, false)) + rc = -EBUSY ;return rc; }
@@ -1929,7 +1912,7 @@ int soft_offline_page(unsigned long pfn, int flags)put_online_mems ();if (ret >0 ) - ret =soft_offline_in_use_page (page, flags); + ret =soft_offline_in_use_page (page);else if (ret ==0 ) ret =soft_offline_free_page (page);
Signature · 1 line
2.17.1
[email protected]
reviewerPATCH v6 10/12 mm,hwpoison: Return 0 if the page is already poisoned in soft-offline
In reply to [email protected]From: Oscar Salvador <[email protected]>
Currently, there is an inconsistency when calling soft-offline from different paths on a page that is already poisoned.
1) madvise:
madvise_inject_error skips any poisoned page and continues the loop.
If that was the only page to madvise, it returns 0.
2) /sys/devices/system/memory/:
When calling soft_offline_page_store()->soft_offline_page(), we return -EBUSY in case the page is already poisoned. This is inconsistent with a) the above example and b) memory_failure, where we return 0 if the page was poisoned.
Fix this by dropping the PageHWPoison() check in madvise_inject_error, and let soft_offline_page return 0 if it finds the page already poisoned.
Please, note that this represents a user-api change, since now the return error when calling soft_offline_page_store()->soft_offline_page() will be different.
--- mm/madvise.c | 5 ----- mm/memory-failure.c | 4 ++-- 2 files changed, 2 insertions(+), 7 deletions(-)
mm/madvise.c +0 -5
@@ -910,11 +910,6 @@ static int madvise_inject_error(int behavior,*/ size =page_size (compound_head (page)); -if (PageHWPoison (page)) { -put_page (page); -continue ; - } -if (behavior ==MADV_SOFT_OFFLINE ) {pr_info ("Soft offlining pfn %#lx at process virtual address %#lx\n" , pfn, start);
mm/memory-failure.c +2 -2
@@ -1800,7 +1800,7 @@ static int __soft_offline_page(struct page *page)unlock_page (page);put_page (page);pr_info ("soft offline: %#lx page already poisoned\n" , pfn); -return -EBUSY ; +return 0 ; }if (!PageHuge (page))
@@ -1904,7 +1904,7 @@ int soft_offline_page(unsigned long pfn, int flags)pr_info ("soft offline: %#lx page already poisoned\n" , pfn);if (flags &MF_COUNT_INCREASED )put_page (page); -return -EBUSY ; +return 0 ; }get_online_mems ();
Signature · 1 line
2.17.1
[email protected]
reviewerPATCH v6 11/12 mm,hwpoison: introduce MF_MSG_UNSPLIT_THP
In reply to [email protected]From: Naoya Horiguchi <[email protected]>
memory_failure() is supposed to call action_result() when it handles a memory error event, but there's one missing case. So let's add it.
I find that include/ras/ras_event.h has some other MF_MSG_* undefined, so this patch also adds them.
--- include/linux/mm.h | 1 + include/ras/ras_event.h | 3 +++ mm/memory-failure.c | 5 ++++- 3 files changed, 8 insertions(+), 1 deletion(-)
include/linux/mm.h +1 -0
@@ -3024,6 +3024,7 @@ enum mf_action_page_type {MF_MSG_BUDDY ,MF_MSG_BUDDY_2ND ,MF_MSG_DAX , +MF_MSG_UNSPLIT_THP , MF_MSG_UNKNOWN, };
include/ras/ras_event.h +3 -0
@@ -361,6 +361,7 @@ TRACE_EVENT(aer_event,EM (MF_MSG_POISONED_HUGE ,"huge page already hardware poisoned" ) \ EM ( MF_MSG_HUGE, "huge page" ) \ EM ( MF_MSG_FREE_HUGE, "free huge page" ) \ + EM ( MF_MSG_NON_PMD_HUGE, "non-pmd-sized huge page" ) \ EM ( MF_MSG_UNMAP_FAILED, "unmapping failed page" ) \ EM ( MF_MSG_DIRTY_SWAPCACHE, "dirty swapcache page" ) \ EM ( MF_MSG_CLEAN_SWAPCACHE, "clean swapcache page" ) \
@@ -373,6 +374,8 @@ TRACE_EVENT(aer_event,EM (MF_MSG_TRUNCATED_LRU ,"already truncated LRU page" ) \ EM ( MF_MSG_BUDDY, "free buddy page" ) \ EM ( MF_MSG_BUDDY_2ND, "free buddy page (2nd try)" ) \ + EM ( MF_MSG_DAX, "dax page" ) \ + EM ( MF_MSG_UNSPLIT_THP, "unsplit thp" ) \ EMe ( MF_MSG_UNKNOWN, "unknown page" )/*
mm/memory-failure.c +4 -1
@@ -583,6 +583,7 @@ static const char * const action_page_types[] = { [MF_MSG_BUDDY] = "free buddy page", [MF_MSG_BUDDY_2ND] = "free buddy page (2nd try)", [MF_MSG_DAX] = "dax page", + [MF_MSG_UNSPLIT_THP] = "unsplit thp", [MF_MSG_UNKNOWN] = "unknown page", };
@@ -1373,8 +1374,10 @@ int memory_failure(unsigned long pfn, int flags) }if (PageTransHuge (hpage)) { -if (try_to_split_thp_page (p,"Memory Failure" ) <0 ) +if (try_to_split_thp_page (p,"Memory Failure" ) <0 ) { +action_result (pfn,MF_MSG_UNSPLIT_THP ,MF_IGNORED );return -EBUSY ; + }VM_BUG_ON_PAGE (!page_count (p), p); }
Signature · 1 line
2.17.1
[email protected]
reviewerPATCH v6 12/12 mm,hwpoison: double-check page count in __get_any_page()
In reply to [email protected]From: Naoya Horiguchi <[email protected]>
Soft offlining could fail with EIO due to the race condition with hugepage migration. This issuse became visible due to the change by previous patch that makes soft offline handler take page refcount by its own. We have no way to directly pin zero refcount page, and the page considered as a zero refcount page could be allocated just after the first check.
This patch adds the second check to find the race and gives us chance to handle it more reliably.
--- mm/memory-failure.c | 6 ++++++ 1 file changed, 6 insertions(+)
mm/memory-failure.c +6 -0
@@ -1700,6 +1700,9 @@ static int __get_any_page(struct page *p, unsigned long pfn, int flags) }else if (is_free_buddy_page (p)) {pr_info ("%s: %#lx free buddy page\n" , __func__, pfn); ret =0 ; + }else if (page_count (p)) { +/* raced with allocation */ + ret = -EBUSY ; }else {pr_info ("%s: %#lx: unknown zero refcount page type %lx\n" , __func__, pfn, p->flags);
@@ -1716,6 +1719,9 @@ static int get_any_page(struct page *page, unsigned long pfn, int flags) {int ret =__get_any_page (page, pfn, flags); +if (ret == -EBUSY ) + ret =__get_any_page (page, pfn, flags); +if (ret ==1 && !PageHuge (page) && !PageLRU (page) && !__PageMovable (page)) {/*
Signature · 1 line
2.17.1
Re: [PATCH v6 12/12] mm,hwpoison: double-check page count in __get_any_page()
Quoted 9 lines · On Thu, Aug 06, 2020 at 06:49:23PM +0000, [email protected] wrote:
From: Naoya Horiguchi <[email protected]>
Soft offlining could fail with EIO due to the race condition with hugepage migration. This issuse became visible due to the change by previous patch that makes soft offline handler take page refcount by its own. We have no way to directly pin zero refcount page, and the page considered as a zero refcount page could be allocated just after the first check.
Quoted 5 lines
This patch adds the second check to find the race and gives us chance to handle it more reliably.
Reported-by Qian Cai <[email protected]>Signed-off-by Naoya Horiguchi <[email protected]>
Signature · 2 lines
Oscar Salvador SUSE L3
Quoted 10 lines · On Thu, Aug 06, 2020 at 06:49:11PM +0000, [email protected] wrote:
Hi,
This patchset is the latest version of soft offline rework patchset targetted for v5.9.
Since v5, I dropped some patches which tweak refcount handling in
madvise_inject_error()to avoid the "unknown refcount page" error. I don't confirm the fix (that didn't reproduce with v5 in my environment), but this change surely callsoft_offline_page()after holding refcount, so the error should not happen any more.
With this patchset, arm64 is still suffering from premature 512M-size hugepages allocation failures.
# git clone https://gitlab.com/cailca/linux-mm # cd linux-mm; make # ./random 1 - start: migrate_huge_offline - use NUMA nodes 0,1. - mmap and free 2147483648 bytes hugepages on node 0 - mmap and free 2147483648 bytes hugepages on node 1 madvise: Cannot allocate memory
[ 292.456538][ T3685] soft offline: 0x8a000: hugepage isolation failed: 0, page count 2, type 7ffff80001000e (referenced|uptodate|dirty|head) [ 292.469113][ T3685] Soft offlining pfn 0x8c000 at process virtual address 0xffff60000000 [ 292.983855][ T3685] Soft offlining pfn 0x88000 at process virtual address 0xffff40000000 [ 293.271369][ T3685] Soft offlining pfn 0x8a000 at process virtual address 0xffff60000000 [ 293.834030][ T3685] Soft offlining pfn 0xa000 at process virtual address 0xffff40000000 [ 293.851378][ T3685] soft offline: 0xa000: hugepage migration failed -12, type 7ffff80001000e (referenced|uptodate|dirty|head)
The fresh-booted system still had 40G+ memory free before running the test.
Reverting the following commits allowed the test to run succesfully over and over again.
"mm, hwpoison: remove recalculating hpage" "mm,hwpoison-inject: don't pin for hwpoison_filter" "mm,hwpoison: Un-export get_hwpoison_page and make it static" "mm,hwpoison: kill put_hwpoison_page"
"mm,hwpoison: unify THP handling for hard and soft offline" "mm,hwpoison: rework soft offline for free pages" "mm,hwpoison: rework soft offline for in-use pages" "mm,hwpoison: refactor soft_offline_huge_page and __soft_offline_page"
i.e., it is not enough to only revert,
mm,hwpoison: double-check page count in __get_any_page() mm,hwpoison: introduce MF_MSG_UNSPLIT_THP
mm,hwpoison: return 0 if the page is already poisoned in soft-offline
Quoted 64 lines
Dropped patches
- mm,madvise: callsoft_offline_page()withoutMF_COUNT_INCREASED
- mm,madvise: Refactormadvise_inject_error
- mm,hwpoison: removeMF_COUNT_INCREASED
- mm,hwpoison: remove flag argument from soft offline functionsThanks,
Naoya HoriguchiQuoting cover letter of v5:
----
Main focus of this series is to stabilize soft offline. Historically soft offlined pages have suffered from racy conditions because PageHWPoison is used to a little too aggressively, which (directly or indirectly) invades other mm code which cares little about hwpoison. This results in unexpected behavior or kernel panic, which is very far from soft offline's "do not disturb userspace or other kernel component" policy.Main point of this change set is to contain target page "via buddy allocator", where we first free the target page as we do for normal pages, and remove from buddy only when we confirm that it reaches free list. There is surely race window of page allocation, but that's fine because someone really want that page and the page is still working, so soft offline can happily give up.
v4 from Oscar tries to handle the race around reallocation, but that part seems still work in progress, so I decide to separate it for changes into v5.9. Thank you for your contribution, Oscar.
---
Previous versions:
v1: https://lore.kernel.org/linux-mm/[email protected]/
v2: https://lore.kernel.org/linux-mm/[email protected]/
v3: https://lore.kernel.org/linux-mm/[email protected]/
v4: https://lore.kernel.org/linux-mm/[email protected]/
v5: https://lore.kernel.org/linux-mm/[email protected]/T/#t
---
Summary:Naoya Horiguchi (5):
mm,hwpoison: cleanup unusedPageHuge()check mm, hwpoison: remove recalculating hpage mm,hwpoison-inject: don't pin forhwpoison_filtermm,hwpoison: introduceMF_MSG_UNSPLIT_THPmm,hwpoison: double-check page count in__get_any_page()Oscar Salvador (7):
mm,hwpoison: Un-exportget_hwpoison_pageand make it static mm,hwpoison: Killput_hwpoison_page
mm,hwpoison: Unify THP handling for hard and soft offline mm,hwpoison: Rework soft offline for free pages mm,hwpoison: Rework soft offline for in-use pages mm,hwpoison: Refactorsoft_offline_huge_pageand__soft_offline_pagemm,hwpoison: Return 0 if the page is already poisoned in soft-offlineinclude/linux/mm.h | 3 +- include/linux/page-flags.h | 6 +- include/ras/ras_event.h | 3 + mm/hwpoison-inject.c | 18 +-- mm/madvise.c | 5 - mm/memory-failure.c | 307 +++++++++++++++++++++------------------------ mm/migrate.c | 11 +- mm/page_alloc.c | 60 +++++++-- 8 files changed, 203 insertions(+), 210 deletions(-)
Quoted 24 lines · Qian Cai
Quoted 10 lines · On Thu, Aug 06, 2020 at 06:49:11PM +0000, [email protected] wrote:
Hi,
This patchset is the latest version of soft offline rework patchset targetted for v5.9.
Since v5, I dropped some patches which tweak refcount handling in
madvise_inject_error()to avoid the "unknown refcount page" error. I don't confirm the fix (that didn't reproduce with v5 in my environment), but this change surely callsoft_offline_page()after holding refcount, so the error should not happen any more.With this patchset, arm64 is still suffering from premature 512M-size hugepages allocation failures.
# git clone https://gitlab.com/cailca/linux-mm # cd linux-mm; make # ./random 1 - start: migrate_huge_offline - use NUMA nodes 0,1. - mmap and free 2147483648 bytes hugepages on node 0 - mmap and free 2147483648 bytes hugepages on node 1 madvise: Cannot allocate memory
Quoted 8 lines
[ 292.456538][ T3685] soft offline: 0x8a000: hugepage isolation failed: 0, page count 2, type 7ffff80001000e (referenced|uptodate|dirty|head) [ 292.469113][ T3685] Soft offlining pfn 0x8c000 at process virtual address 0xffff60000000 [ 292.983855][ T3685] Soft offlining pfn 0x88000 at process virtual address 0xffff40000000 [ 293.271369][ T3685] Soft offlining pfn 0x8a000 at process virtual address 0xffff60000000 [ 293.834030][ T3685] Soft offlining pfn 0xa000 at process virtual address 0xffff40000000 [ 293.851378][ T3685] soft offline: 0xa000: hugepage migration failed -12, type 7ffff80001000e (referenced|uptodate|dirty|head)The fresh-booted system still had 40G+ memory free before running the test.
As I commented over v5, this failure is expected and it doesn't mean kernel issue. Once we successfully soft offline a hugepage, the memory range covering the hugepage will never participate in hugepage because one of the subpages is removed from buddy. So if you iterate soft offlining hugepages, all memory range are "holed" finally, and no hugepage will be available in the system.
Please fix your test program to properly determine nubmer of loop (NR_LOOP) so that you can assume that you can always allocate hugepage during testing. For example, if you can use 40G memory and hugepage size is 512MB, NR_LOOP should not be larger than 80.
Quoted 11 lines
Reverting the following commits allowed the test to run succesfully over and over again.
"mm, hwpoison: remove recalculating hpage" "mm,hwpoison-inject: don't pin for
hwpoison_filter" "mm,hwpoison: Un-exportget_hwpoison_pageand make it static" "mm,hwpoison: killput_hwpoison_page"
"mm,hwpoison: unify THP handling for hard and soft offline" "mm,hwpoison: rework soft offline for free pages" "mm,hwpoison: rework soft offline for in-use pages" "mm,hwpoison: refactorsoft_offline_huge_pageand__soft_offline_page"
I'm still not sure why the test succeeded by reverting these because current mainline kernel provides similar mechanism to prevent reuse of soft offlined page. So this success seems to me something suspicious.
To investigate more, I want to have additional info about the page states of the relevant pages after soft offlining. Could you collect it by the following steps?
- modify random.c not to run hotplug_memory() in migrate_huge_hotplug_memory(),
- compile it and run "./random 1" once,
- to collect page state with hwpoisoned pages, run "./page-types -Nlr -b hwpoison", where page-types is available under tools/vm in kernel source tree.
- choose a few pfns of soft offlined pages from kernel message
"Soft offlining pfn ...", and run "./page-types -Nlr -a <pfn>".
Thanks,
Naoya Horiguchi
Quoted 7 lines
i.e., it is not enough to only revert,
mm,hwpoison: double-check page count in
__get_any_page()mm,hwpoison: introduceMF_MSG_UNSPLIT_THP
mm,hwpoison: return 0 if the page is already poisoned in soft-offline
Quoted 7 lines · HORIGUCHI NAOYA(堀口 直也)
I'm still not sure why the test succeeded by reverting these because current mainline kernel provides similar mechanism to prevent reuse of soft offlined page. So this success seems to me something suspicious.
Even if we call munmap() on the range, it still can’t be be reused? If so, how to recover those memory then?
Quoted 4 lines
To investigate more, I want to have additional info about the page states of the relevant pages after soft offlining. Could you collect it by the following steps?
Do you want to collect those from the failed or succeed kernel?
Quoted 9 lines · Qian Cai
Quoted 7 lines · HORIGUCHI NAOYA(堀口 直也)
I'm still not sure why the test succeeded by reverting these because current mainline kernel provides similar mechanism to prevent reuse of soft offlined page. So this success seems to me something suspicious.
Even if we call
munmap()on the range, it still can’t be be reused? If so, how to recover those memory then?
No, it can't, because soft offline isolates the physical page. so even after calling munmap(), the side effect remains on the page. In your random.c, memory online/offline resets the status of hwpoison. So you can reallocate hugepages in another run of the program.
Quoted 7 lines
Quoted 5 lines
To investigate more, I want to have additional info about the page states of the relevant pages after soft offlining. Could you collect it by the following steps?
Do you want to collect those from the failed or succeed kernel?
I'd like to check on the succeeded kernel.
Sorry for the lack of information.
Thanks,
Naoya Horiguchi
Quoted 8 lines · HORIGUCHI NAOYA(堀口 直也)
I'm still not sure why the test succeeded by reverting these because current mainline kernel provides similar mechanism to prevent reuse of soft offlined page. So this success seems to me something suspicious.
To investigate more, I want to have additional info about the page states of the relevant pages after soft offlining. Could you collect it by the following steps?
Quoted 6 lines
- modify
random.cnot to runhotplug_memory()inmigrate_huge_hotplug_memory(),
- compile it and run "./random 1" once,
- to collect page state with hwpoisoned pages, run "./page-types -Nlr -b hwpoison", where page-types is available undertools/vmin kernel source tree.
- choose a few pfns of soft offlined pages from kernel message"Soft offlining pfn ...", and run "./page-types -Nlr -a <pfn>".
# ./page-types -Nlr -b hwpoison offset len flags 99a000 1 __________B________X_______________________ 99c000 1 __________B________X_______________________ 99e000 1 __________B________X_______________________ 9a0000 1 __________B________X_______________________ ba6000 1 __________B________X_______________________ baa000 1 __________B________X_______________________
Every single one of pfns was like this,
# ./page-types -Nlr -a 0x99a000 offset len flags 99a000 1 __________B________X_______________________
# ./page-types -Nlr -a 0x99e000 offset len flags 99e000 1 __________B________X_______________________
# ./page-types -Nlr -a 0x99c000 offset len flags 99c000 1 __________B________X_______________________
Naoya Horiguchi
reviewerQuoted 16 lines · Qian Cai
Quoted 8 lines · HORIGUCHI NAOYA(堀口 直也)
I'm still not sure why the test succeeded by reverting these because current mainline kernel provides similar mechanism to prevent reuse of soft offlined page. So this success seems to me something suspicious.
To investigate more, I want to have additional info about the page states of the relevant pages after soft offlining. Could you collect it by the following steps?
Quoted 6 lines
- modify
random.cnot to runhotplug_memory()inmigrate_huge_hotplug_memory(),
- compile it and run "./random 1" once,
- to collect page state with hwpoisoned pages, run "./page-types -Nlr -b hwpoison", where page-types is available undertools/vmin kernel source tree.
- choose a few pfns of soft offlined pages from kernel message"Soft offlining pfn ...", and run "./page-types -Nlr -a <pfn>".
Quoted 8 lines
# ./page-types -Nlr -b hwpoison offset len flags 99a000 1 __________B________X_______________________ 99c000 1 __________B________X_______________________ 99e000 1 __________B________X_______________________ 9a0000 1 __________B________X_______________________ ba6000 1 __________B________X_______________________ baa000 1 __________B________X_______________________
Thank you. It only shows 6 lines of records, which is unexpected to me because random.c iterates soft offline 2 hugepages with madvise() 1000 times. Somehow (maybe in arch specific way?) other hwpoisoned pages might be cleared? If they really are, the success of this test is a fake, and this patchset can be considered as a fix.
Quoted 14 lines
Every single one of pfns was like this,
# ./page-types -Nlr -a 0x99a000 offset len flags 99a000 1 __________B________X_______________________# ./page-types -Nlr -a 0x99e000 offset len flags 99e000 1 __________B________X_______________________# ./page-types -Nlr -a 0x99c000 offset len flags 99c000 1 __________B________X_______________________
Quoted 26 lines · Naoya Horiguchi
Quoted 16 lines · Qian Cai
Quoted 8 lines · HORIGUCHI NAOYA(堀口 直也)
I'm still not sure why the test succeeded by reverting these because current mainline kernel provides similar mechanism to prevent reuse of soft offlined page. So this success seems to me something suspicious.
To investigate more, I want to have additional info about the page states of the relevant pages after soft offlining. Could you collect it by the following steps?
Quoted 6 lines
- modify
random.cnot to runhotplug_memory()inmigrate_huge_hotplug_memory(),
- compile it and run "./random 1" once,
- to collect page state with hwpoisoned pages, run "./page-types -Nlr -b hwpoison", where page-types is available undertools/vmin kernel source tree.
- choose a few pfns of soft offlined pages from kernel message"Soft offlining pfn ...", and run "./page-types -Nlr -a <pfn>".Quoted 8 lines
# ./page-types -Nlr -b hwpoison offset len flags 99a000 1 __________B________X_______________________ 99c000 1 __________B________X_______________________ 99e000 1 __________B________X_______________________ 9a0000 1 __________B________X_______________________ ba6000 1 __________B________X_______________________ baa000 1 __________B________X_______________________
Quoted 5 lines
Thank you. It only shows 6 lines of records, which is unexpected to me because
random.citerates soft offline 2 hugepages withmadvise()1000 times. Somehow (maybe in arch specific way?) other hwpoisoned pages might be cleared? If they really are, the success of this test is a fake, and this patchset can be considered as a fix.
The test was designed to catch a previous bug (the latest patchset fixed that) where kernel will be enterting into an endless loop.
https://lore.kernel.org/lkml/[email protected]/
However, I don't understand why mmap() does not return ENOMEM in the first place where overcommit_memory == 0 instead of munmap() or/and madvise() returning ENOMEM. I suppose that is the price to pay with heuristic, and I can't easily confirm if it is related to this patchset or not.
addr = mmap(NULL, length, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB, -1, 0);
if (addr == MAP_FAILED) {
if (i == 0 || errno != ENOMEM) {
perror("mmap");
return 1;
}
usleep(1000);
continue;
}
memset(addr, 0, length);
code = madvise(addr, length, MADV_SOFT_OFFLINE);
if(safe_munmap(addr, length))
return 1;
/* madvise() could return >= 0 on success. */
if (code < 0 && errno != EBUSY) {
perror("madvise");
return 1;
}Otherwise, our test will keep running and ignore ENOMEM correctly. I did also confirm that this patchset has a higher success rate of soft-offlining ("page-types" shows 400+ lines) which changes the existing assumption (looks like in a good way in this case).