diff options
-rw-r--r-- | mm/filemap.c | 20 |
1 files changed, 19 insertions, 1 deletions
diff --git a/mm/filemap.c b/mm/filemap.c index 9b74674e36ad..ee79b5d3439f 100644 --- a/mm/filemap.c +++ b/mm/filemap.c | |||
@@ -139,7 +139,25 @@ static int sync_page(void *word) | |||
139 | page = container_of((page_flags_t *)word, struct page, flags); | 139 | page = container_of((page_flags_t *)word, struct page, flags); |
140 | 140 | ||
141 | /* | 141 | /* |
142 | * FIXME, fercrissake. What is this barrier here for? | 142 | * page_mapping() is being called without PG_locked held. |
143 | * Some knowledge of the state and use of the page is used to | ||
144 | * reduce the requirements down to a memory barrier. | ||
145 | * The danger here is of a stale page_mapping() return value | ||
146 | * indicating a struct address_space different from the one it's | ||
147 | * associated with when it is associated with one. | ||
148 | * After smp_mb(), it's either the correct page_mapping() for | ||
149 | * the page, or an old page_mapping() and the page's own | ||
150 | * page_mapping() has gone NULL. | ||
151 | * The ->sync_page() address_space operation must tolerate | ||
152 | * page_mapping() going NULL. By an amazing coincidence, | ||
153 | * this comes about because none of the users of the page | ||
154 | * in the ->sync_page() methods make essential use of the | ||
155 | * page_mapping(), merely passing the page down to the backing | ||
156 | * device's unplug functions when it's non-NULL, which in turn | ||
157 | * ignore it for all cases but swap, where only page->private is | ||
158 | * of interest. When page_mapping() does go NULL, the entire | ||
159 | * call stack gracefully ignores the page and returns. | ||
160 | * -- wli | ||
143 | */ | 161 | */ |
144 | smp_mb(); | 162 | smp_mb(); |
145 | mapping = page_mapping(page); | 163 | mapping = page_mapping(page); |