diff options
| author | Mel Gorman <mgorman@techsingularity.net> | 2016-03-15 17:55:39 -0400 |
|---|---|---|
| committer | Linus Torvalds <torvalds@linux-foundation.org> | 2016-03-15 19:55:16 -0400 |
| commit | ebded02788b5d7c7600f8cff26ae07896d568649 (patch) | |
| tree | 155d79f674e6a3eafbc43bdb6893f0688bcf482d /mm | |
| parent | 32b635298ff4e991d8d8f64dc23782b02eec29c3 (diff) | |
mm: filemap: avoid unnecessary calls to lock_page when waiting for IO to complete during a read
In the generic read paths the kernel looks up a page in the page cache
and if it's up to date, it is used. If not, the page lock is acquired
to wait for IO to complete and then check the page. If multiple
processes are waiting on IO, they all serialise against the lock and
duplicate the checks. This is unnecessary.
The page lock in itself does not give any guarantees to the callers
about the page state as it can be immediately truncated or reclaimed
after the page is unlocked. It's sufficient to wait_on_page_locked and
then continue if the page is up to date on wakeup.
It is possible that a truncated but up-to-date page is returned but the
reference taken during read prevents it disappearing underneath the
caller and the data is still valid if PageUptodate.
The overall impact is small as even if processes serialise on the lock,
the lock section is tiny once the IO is complete. Profiles indicated
that unlock_page and friends are generally a tiny portion of a
read-intensive workload. An artificial test was created that had
instances of dd access a cache-cold file on an ext4 filesystem and
measure how long the read took.
paralleldd
4.4.0 4.4.0
vanilla avoidlock
Amean Elapsd-1 5.28 ( 0.00%) 5.15 ( 2.50%)
Amean Elapsd-4 5.29 ( 0.00%) 5.17 ( 2.12%)
Amean Elapsd-7 5.28 ( 0.00%) 5.18 ( 1.78%)
Amean Elapsd-12 5.20 ( 0.00%) 5.33 ( -2.50%)
Amean Elapsd-21 5.14 ( 0.00%) 5.21 ( -1.41%)
Amean Elapsd-30 5.30 ( 0.00%) 5.12 ( 3.38%)
Amean Elapsd-48 5.78 ( 0.00%) 5.42 ( 6.21%)
Amean Elapsd-79 6.78 ( 0.00%) 6.62 ( 2.46%)
Amean Elapsd-110 9.09 ( 0.00%) 8.99 ( 1.15%)
Amean Elapsd-128 10.60 ( 0.00%) 10.43 ( 1.66%)
The impact is small but intuitively, it makes sense to avoid unnecessary
calls to lock_page.
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Diffstat (limited to 'mm')
| -rw-r--r-- | mm/filemap.c | 49 |
1 files changed, 49 insertions, 0 deletions
diff --git a/mm/filemap.c b/mm/filemap.c index deae0b9ad90b..4a0f5fa79dbd 100644 --- a/mm/filemap.c +++ b/mm/filemap.c | |||
| @@ -1668,6 +1668,15 @@ find_page: | |||
| 1668 | index, last_index - index); | 1668 | index, last_index - index); |
| 1669 | } | 1669 | } |
| 1670 | if (!PageUptodate(page)) { | 1670 | if (!PageUptodate(page)) { |
| 1671 | /* | ||
| 1672 | * See comment in do_read_cache_page on why | ||
| 1673 | * wait_on_page_locked is used to avoid unnecessarily | ||
| 1674 | * serialisations and why it's safe. | ||
| 1675 | */ | ||
| 1676 | wait_on_page_locked_killable(page); | ||
| 1677 | if (PageUptodate(page)) | ||
| 1678 | goto page_ok; | ||
| 1679 | |||
| 1671 | if (inode->i_blkbits == PAGE_CACHE_SHIFT || | 1680 | if (inode->i_blkbits == PAGE_CACHE_SHIFT || |
| 1672 | !mapping->a_ops->is_partially_uptodate) | 1681 | !mapping->a_ops->is_partially_uptodate) |
| 1673 | goto page_not_up_to_date; | 1682 | goto page_not_up_to_date; |
| @@ -2341,12 +2350,52 @@ filler: | |||
| 2341 | if (PageUptodate(page)) | 2350 | if (PageUptodate(page)) |
| 2342 | goto out; | 2351 | goto out; |
| 2343 | 2352 | ||
| 2353 | /* | ||
| 2354 | * Page is not up to date and may be locked due one of the following | ||
| 2355 | * case a: Page is being filled and the page lock is held | ||
| 2356 | * case b: Read/write error clearing the page uptodate status | ||
| 2357 | * case c: Truncation in progress (page locked) | ||
| 2358 | * case d: Reclaim in progress | ||
| 2359 | * | ||
| 2360 | * Case a, the page will be up to date when the page is unlocked. | ||
| 2361 | * There is no need to serialise on the page lock here as the page | ||
| 2362 | * is pinned so the lock gives no additional protection. Even if the | ||
| 2363 | * the page is truncated, the data is still valid if PageUptodate as | ||
| 2364 | * it's a race vs truncate race. | ||
| 2365 | * Case b, the page will not be up to date | ||
| 2366 | * Case c, the page may be truncated but in itself, the data may still | ||
| 2367 | * be valid after IO completes as it's a read vs truncate race. The | ||
| 2368 | * operation must restart if the page is not uptodate on unlock but | ||
| 2369 | * otherwise serialising on page lock to stabilise the mapping gives | ||
| 2370 | * no additional guarantees to the caller as the page lock is | ||
| 2371 | * released before return. | ||
| 2372 | * Case d, similar to truncation. If reclaim holds the page lock, it | ||
| 2373 | * will be a race with remove_mapping that determines if the mapping | ||
| 2374 | * is valid on unlock but otherwise the data is valid and there is | ||
| 2375 | * no need to serialise with page lock. | ||
| 2376 | * | ||
| 2377 | * As the page lock gives no additional guarantee, we optimistically | ||
| 2378 | * wait on the page to be unlocked and check if it's up to date and | ||
| 2379 | * use the page if it is. Otherwise, the page lock is required to | ||
| 2380 | * distinguish between the different cases. The motivation is that we | ||
| 2381 | * avoid spurious serialisations and wakeups when multiple processes | ||
| 2382 | * wait on the same page for IO to complete. | ||
| 2383 | */ | ||
| 2384 | wait_on_page_locked(page); | ||
| 2385 | if (PageUptodate(page)) | ||
| 2386 | goto out; | ||
| 2387 | |||
| 2388 | /* Distinguish between all the cases under the safety of the lock */ | ||
| 2344 | lock_page(page); | 2389 | lock_page(page); |
| 2390 | |||
| 2391 | /* Case c or d, restart the operation */ | ||
| 2345 | if (!page->mapping) { | 2392 | if (!page->mapping) { |
| 2346 | unlock_page(page); | 2393 | unlock_page(page); |
| 2347 | page_cache_release(page); | 2394 | page_cache_release(page); |
| 2348 | goto repeat; | 2395 | goto repeat; |
| 2349 | } | 2396 | } |
| 2397 | |||
| 2398 | /* Someone else locked and filled the page in a very small window */ | ||
| 2350 | if (PageUptodate(page)) { | 2399 | if (PageUptodate(page)) { |
| 2351 | unlock_page(page); | 2400 | unlock_page(page); |
| 2352 | goto out; | 2401 | goto out; |