Commit 867da60
nfs: avoid i_lock contention in nfs_clear_invalid_mapping
Multi-threaded buffered reads to the same file exposed significant
inode spinlock contention in nfs_clear_invalid_mapping().
Eliminate this spinlock contention by checking flags without locking,
instead using smp_rmb and smp_load_acquire accordingly, but then take
spinlock and double-check these inode flags.
Also refactor nfs_set_cache_invalid() slightly to use
smp_store_release() to pair with nfs_clear_invalid_mapping()'s
smp_load_acquire().
While this fix is beneficial for all multi-threaded buffered reads
issued by an NFS client, this issue was identified in the context of
surprisingly low LOCALIO performance with 4K multi-threaded buffered
read IO. This fix dramatically speeds up LOCALIO performance:
before: read: IOPS=1583k, BW=6182MiB/s (6482MB/s)(121GiB/20002msec)
after: read: IOPS=3046k, BW=11.6GiB/s (12.5GB/s)(232GiB/20001msec)
Fixes: 17dfeb9 ("NFS: Fix races in nfs_revalidate_mapping")
Signed-off-by: Mike Snitzer <[email protected]>
Reviewed-by: Jeff Layton <[email protected]>
Signed-off-by: Anna Schumaker <[email protected]>1 parent bc29408 commit 867da60
1 file changed
+15
-5
lines changed| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
205 | 205 | | |
206 | 206 | | |
207 | 207 | | |
208 | | - | |
| 208 | + | |
| 209 | + | |
| 210 | + | |
209 | 211 | | |
210 | | - | |
211 | | - | |
212 | | - | |
213 | | - | |
| 212 | + | |
| 213 | + | |
| 214 | + | |
| 215 | + | |
| 216 | + | |
214 | 217 | | |
215 | 218 | | |
216 | 219 | | |
| |||
1421 | 1424 | | |
1422 | 1425 | | |
1423 | 1426 | | |
| 1427 | + | |
| 1428 | + | |
| 1429 | + | |
| 1430 | + | |
| 1431 | + | |
| 1432 | + | |
| 1433 | + | |
1424 | 1434 | | |
1425 | 1435 | | |
1426 | 1436 | | |
| |||
0 commit comments