lkml.org 
[lkml]   [2017]   [Aug]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
Subject[RFC PATCH 3/4] mm: soft-offline: retry to split and soft-offline the raw error if the original THP offlining fails.
Date
From: Zi Yan <zi.yan@cs.rutgers.edu>

For THP soft-offline support, we first try to migrate a THP without
splitting. If the migration fails, we split the THP and migrate the
raw error page.

migrate_pages() does not split a THP if the migration reason is
MR_MEMORY_FAILURE.

Signed-off-by: Zi Yan <zi.yan@cs.rutgers.edu>
---
mm/memory-failure.c | 77 +++++++++++++++++++++++++++++++++++++----------------
mm/migrate.c | 16 +++++++++++
2 files changed, 70 insertions(+), 23 deletions(-)

diff --git a/mm/memory-failure.c b/mm/memory-failure.c
index 8a9ac6f9e1b0..c05107548d72 100644
--- a/mm/memory-failure.c
+++ b/mm/memory-failure.c
@@ -1598,10 +1598,11 @@ static int soft_offline_huge_page(struct page *page, int flags)
return ret;
}

-static int __soft_offline_page(struct page *page, int flags)
+static int __soft_offline_page(struct page *page, int flags, int *split)
{
int ret;
- unsigned long pfn = page_to_pfn(page);
+ struct page *hpage = compound_head(page);
+ unsigned long pfn = page_to_pfn(hpage);

/*
* Check PageHWPoison again inside page lock because PageHWPoison
@@ -1609,11 +1610,11 @@ static int __soft_offline_page(struct page *page, int flags)
* memory_failure() also double-checks PageHWPoison inside page lock,
* so there's no race between soft_offline_page() and memory_failure().
*/
- lock_page(page);
- wait_on_page_writeback(page);
- if (PageHWPoison(page)) {
- unlock_page(page);
- put_hwpoison_page(page);
+ lock_page(hpage);
+ wait_on_page_writeback(hpage);
+ if (PageHWPoison(hpage)) {
+ unlock_page(hpage);
+ put_hwpoison_page(hpage);
pr_info("soft offline: %#lx page already poisoned\n", pfn);
return -EBUSY;
}
@@ -1621,14 +1622,14 @@ static int __soft_offline_page(struct page *page, int flags)
* Try to invalidate first. This should work for
* non dirty unmapped page cache pages.
*/
- ret = invalidate_inode_page(page);
- unlock_page(page);
+ ret = invalidate_inode_page(hpage);
+ unlock_page(hpage);
/*
* RED-PEN would be better to keep it isolated here, but we
* would need to fix isolation locking first.
*/
if (ret == 1) {
- put_hwpoison_page(page);
+ put_hwpoison_page(hpage);
pr_info("soft_offline: %#lx: invalidated\n", pfn);
SetPageHWPoison(page);
num_poisoned_pages_inc();
@@ -1640,15 +1641,15 @@ static int __soft_offline_page(struct page *page, int flags)
* 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);
+ if (PageLRU(hpage))
+ ret = isolate_lru_page(hpage);
else
- ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
+ ret = isolate_movable_page(hpage, ISOLATE_UNEVICTABLE);
/*
* Drop page reference which is came from get_any_page()
* successful isolate_lru_page() already took another one.
*/
- put_hwpoison_page(page);
+ put_hwpoison_page(hpage);
if (!ret) {
LIST_HEAD(pagelist);
/*
@@ -1657,23 +1658,53 @@ static int __soft_offline_page(struct page *page, int flags)
* cannot have PAGE_MAPPING_MOVABLE.
*/
if (!__PageMovable(page))
- inc_node_page_state(page, NR_ISOLATED_ANON +
- page_is_file_cache(page));
- list_add(&page->lru, &pagelist);
+ mod_node_page_state(page_pgdat(hpage), NR_ISOLATED_ANON +
+ page_is_file_cache(hpage), hpage_nr_pages(hpage));
+retry_subpage:
+ list_add(&hpage->lru, &pagelist);
ret = migrate_pages(&pagelist, new_page, NULL, MPOL_MF_MOVE_ALL,
MIGRATE_SYNC, MR_MEMORY_FAILURE);
if (ret) {
- if (!list_empty(&pagelist))
- putback_movable_pages(&pagelist);
-
+ if (!list_empty(&pagelist)) {
+ if (!PageTransHuge(hpage))
+ putback_movable_pages(&pagelist);
+ else {
+ lock_page(hpage);
+ if (split_huge_page_to_list(hpage, &pagelist)) {
+ unlock_page(hpage);
+ goto failed;
+ }
+ unlock_page(hpage);
+
+ if (split)
+ *split = 1;
+ /*
+ * Pull the raw error page out and put back other subpages.
+ * Then retry the raw error page.
+ */
+ list_del(&page->lru);
+ putback_movable_pages(&pagelist);
+ hpage = page;
+ goto retry_subpage;
+ }
+ }
+failed:
pr_info("soft offline: %#lx: migration failed %d, type %lx (%pGp)\n",
- pfn, ret, page->flags, &page->flags);
+ pfn, ret, hpage->flags, &hpage->flags);
if (ret > 0)
ret = -EIO;
}
+ /*
+ * Set PageHWPoison on the raw error page.
+ *
+ * If the page is a THP, PageHWPoison is set then cleared
+ * in its head page in migrate_pages(). So we need to set the raw error
+ * page here. Otherwise, setting PageHWPoison again is fine.
+ */
+ SetPageHWPoison(page);
} 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);
+ pfn, ret, page_count(hpage), hpage->flags, &hpage->flags);
}
return ret;
}
@@ -1704,7 +1735,7 @@ static int soft_offline_in_use_page(struct page *page, int flags, int *split)
if (PageHuge(page))
ret = soft_offline_huge_page(page, flags);
else
- ret = __soft_offline_page(page, flags);
+ ret = __soft_offline_page(page, flags, split);

return ret;
}
diff --git a/mm/migrate.c b/mm/migrate.c
index f7b69282d216..b44df9cf72fd 100644
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -1118,6 +1118,15 @@ static ICE_noinline int unmap_and_move(new_page_t get_new_page,
}

if (unlikely(PageTransHuge(page) && !PageTransHuge(newpage))) {
+ /*
+ * soft-offline wants to retry the raw error subpage, if the THP
+ * migration fails. So we do not split the THP here and exit directly.
+ */
+ if (reason == MR_MEMORY_FAILURE) {
+ rc = -ENOMEM;
+ goto put_new;
+ }
+
lock_page(page);
rc = split_huge_page(page);
unlock_page(page);
@@ -1164,6 +1173,13 @@ static ICE_noinline int unmap_and_move(new_page_t get_new_page,
*/
if (!test_set_page_hwpoison(page))
num_poisoned_pages_inc();
+
+ /*
+ * Clear PageHWPoison in the head page. The caller
+ * is responsible for setting the raw error page.
+ */
+ if (PageTransHuge(page))
+ ClearPageHWPoison(page);
}
} else {
if (rc != -EAGAIN) {
--
2.13.2
\
 
 \ /
  Last update: 2017-08-15 03:54    [W:0.697 / U:0.012 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site