diff options
| author | Linus Torvalds <torvalds@linux-foundation.org> | 2019-03-29 14:43:07 -0700 |
|---|---|---|
| committer | Linus Torvalds <torvalds@linux-foundation.org> | 2019-03-29 14:43:07 -0700 |
| commit | ffb8e45cf33e14d9a565491aec7abe039bebcfce (patch) | |
| tree | 4181a995d8fb1de48ed6af9c4211ac6a27b54b0a /lib | |
| parent | 7376e39ad96583545faefa2e7798bcb6a2a212a7 (diff) | |
| parent | 7bca889ee9297c3e208dee7c41aed7a56a880400 (diff) | |
Merge tag 'for-linus-20190329' of git://git.kernel.dk/linux-block
Pull block fixes from Jens Axboe:
"Small set of fixes that should go into this series. This contains:
- compat signal mask fix for io_uring (Arnd)
- EAGAIN corner case for direct vs buffered writes for io_uring
(Roman)
- NVMe pull request from Christoph with various little fixes
- sbitmap ws_active fix, which caused a perf regression for shared
tags (me)
- sbitmap bit ordering fix (Ming)
- libata on-stack DMA fix (Raymond)"
* tag 'for-linus-20190329' of git://git.kernel.dk/linux-block:
nvmet: fix error flow during ns enable
nvmet: fix building bvec from sg list
nvme-multipath: relax ANA state check
nvme-tcp: fix an endianess miss-annotation
libata: fix using DMA buffers on stack
io_uring: offload write to async worker in case of -EAGAIN
sbitmap: order READ/WRITE freed instance and setting clear bit
blk-mq: fix sbitmap ws_active for shared tags
io_uring: fix big-endian compat signal mask handling
blk-mq: update comment for blk_mq_hctx_has_pending()
blk-mq: use blk_mq_put_driver_tag() to put tag
Diffstat (limited to 'lib')
| -rw-r--r-- | lib/sbitmap.c | 11 |
1 files changed, 11 insertions, 0 deletions
diff --git a/lib/sbitmap.c b/lib/sbitmap.c index 5b382c1244ed..155fe38756ec 100644 --- a/lib/sbitmap.c +++ b/lib/sbitmap.c @@ -591,6 +591,17 @@ EXPORT_SYMBOL_GPL(sbitmap_queue_wake_up); void sbitmap_queue_clear(struct sbitmap_queue *sbq, unsigned int nr, unsigned int cpu) { + /* + * Once the clear bit is set, the bit may be allocated out. + * + * Orders READ/WRITE on the asssociated instance(such as request + * of blk_mq) by this bit for avoiding race with re-allocation, + * and its pair is the memory barrier implied in __sbitmap_get_word. + * + * One invariant is that the clear bit has to be zero when the bit + * is in use. + */ + smp_mb__before_atomic(); sbitmap_deferred_clear_bit(&sbq->sb, nr); /* |