The patch titled Subject: aio: change reqs_active to include unreaped completions has been added to the -mm tree. Its filename is aio-change-reqs_active-to-include-unreaped-completions.patch Before you just go and hit "reply", please: a) Consider who else should be cc'ed b) Prefer to cc a suitable mailing list as well c) Ideally: find the original patch on the mailing list and do a reply-to-all to that, adding suitable additional cc's *** Remember to use Documentation/SubmitChecklist when testing your code *** The -mm tree is included into linux-next and is updated there every 3-4 working days ------------------------------------------------------ From: Kent Overstreet <koverstreet@xxxxxxxxxx> Subject: aio: change reqs_active to include unreaped completions The aio code tries really hard to avoid having to deal with the completion ringbuffer overflowing. To do that, it has to keep track of the number of outstanding kiocbs, and the number of completions currently in the ringbuffer - and it's got to check that every time we allocate a kiocb. Ouch. But - we can improve this quite a bit if we just change reqs_active to mean "number of outstanding requests and unreaped completions" - that means kiocb allocation doesn't have to look at the ringbuffer, which is a fairly significant win. Signed-off-by: Kent Overstreet <koverstreet@xxxxxxxxxx> Cc: Zach Brown <zab@xxxxxxxxxx> Cc: Felipe Balbi <balbi@xxxxxx> Cc: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> Cc: Mark Fasheh <mfasheh@xxxxxxxx> Cc: Joel Becker <jlbec@xxxxxxxxxxxx> Cc: Rusty Russell <rusty@xxxxxxxxxxxxxxx> Cc: Jens Axboe <axboe@xxxxxxxxx> Cc: Asai Thambi S P <asamymuthupa@xxxxxxxxxx> Cc: Selvan Mani <smani@xxxxxxxxxx> Cc: Sam Bradshaw <sbradshaw@xxxxxxxxxx> Cc: Jeff Moyer <jmoyer@xxxxxxxxxx> Cc: Al Viro <viro@xxxxxxxxxxxxxxxxxx> Cc: Benjamin LaHaise <bcrl@xxxxxxxxx> Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> --- fs/aio.c | 38 +++++++++++++++++++++++++------------- 1 file changed, 25 insertions(+), 13 deletions(-) diff -puN fs/aio.c~aio-change-reqs_active-to-include-unreaped-completions fs/aio.c --- a/fs/aio.c~aio-change-reqs_active-to-include-unreaped-completions +++ a/fs/aio.c @@ -71,12 +71,6 @@ struct aio_ring_info { struct page *internal_pages[AIO_RING_PAGES]; }; -static inline unsigned aio_ring_avail(struct aio_ring_info *info, - struct aio_ring *ring) -{ - return (ring->head + info->nr - 1 - ring->tail) % info->nr; -} - struct kioctx { atomic_t users; atomic_t dead; @@ -270,8 +264,11 @@ static void free_ioctx_rcu(struct rcu_he */ static void free_ioctx(struct kioctx *ctx) { + struct aio_ring_info *info = &ctx->ring_info; + struct aio_ring *ring; struct io_event res; struct kiocb *req; + unsigned head, avail; spin_lock_irq(&ctx->ctx_lock); @@ -285,7 +282,21 @@ static void free_ioctx(struct kioctx *ct spin_unlock_irq(&ctx->ctx_lock); - wait_event(ctx->wait, !atomic_read(&ctx->reqs_active)); + ring = kmap_atomic(info->ring_pages[0]); + head = ring->head; + kunmap_atomic(ring); + + while (atomic_read(&ctx->reqs_active) > 0) { + wait_event(ctx->wait, head != info->tail); + + avail = (head < info->tail ? info->tail : info->nr) - head; + + atomic_sub(avail, &ctx->reqs_active); + head += avail; + head %= info->nr; + } + + WARN_ON(atomic_read(&ctx->reqs_active) < 0); aio_free_ring(ctx); @@ -534,7 +545,6 @@ static int kiocb_batch_refill(struct kio unsigned short allocated, to_alloc; long avail; struct kiocb *req, *n; - struct aio_ring *ring; to_alloc = min(batch->count, KIOCB_BATCH_SIZE); for (allocated = 0; allocated < to_alloc; allocated++) { @@ -549,9 +559,8 @@ static int kiocb_batch_refill(struct kio goto out; spin_lock_irq(&ctx->ctx_lock); - ring = kmap_atomic(ctx->ring_info.ring_pages[0]); - avail = aio_ring_avail(&ctx->ring_info, ring) - atomic_read(&ctx->reqs_active); + avail = ctx->ring_info.nr - atomic_read(&ctx->reqs_active); BUG_ON(avail < 0); if (avail < allocated) { /* Trim back the number of requests. */ @@ -566,7 +575,6 @@ static int kiocb_batch_refill(struct kio batch->count -= allocated; atomic_add(allocated, &ctx->reqs_active); - kunmap_atomic(ring); spin_unlock_irq(&ctx->ctx_lock); out: @@ -673,8 +681,11 @@ void aio_complete(struct kiocb *iocb, lo * when the event got cancelled. */ if (unlikely(xchg(&iocb->ki_cancel, - KIOCB_CANCELLED) == KIOCB_CANCELLED)) + KIOCB_CANCELLED) == KIOCB_CANCELLED)) { + atomic_dec(&ctx->reqs_active); + /* Still need the wake_up in case free_ioctx is waiting */ goto put_rq; + } /* * Add a completion event to the ring buffer. Must be done holding @@ -731,7 +742,6 @@ void aio_complete(struct kiocb *iocb, lo put_rq: /* everything turned out well, dispose of the aiocb. */ aio_put_req(iocb); - atomic_dec(&ctx->reqs_active); /* * We have to order our ring_info tail store above and test @@ -812,6 +822,8 @@ static int aio_read_events_ring(struct k flush_dcache_page(info->ring_pages[0]); pr_debug("%d h%u t%u\n", ret, head, info->tail); + + atomic_sub(ret, &ctx->reqs_active); out: mutex_unlock(&info->ring_lock); _ Patches currently in -mm which might be from koverstreet@xxxxxxxxxx are mm-remove-old-aio-use_mm-comment.patch aio-remove-dead-code-from-aioh.patch gadget-remove-only-user-of-aio-retry.patch aio-remove-retry-based-aio.patch char-add-aio_readwrite-to-dev-nullzero.patch aio-kill-return-value-of-aio_complete.patch aio-kiocb_cancel.patch aio-move-private-stuff-out-of-aioh.patch aio-dprintk-pr_debug.patch aio-do-fget-after-aio_get_req.patch aio-make-aio_put_req-lockless.patch aio-refcounting-cleanup.patch wait-add-wait_event_hrtimeout.patch wait-add-wait_event_hrtimeout-fix.patch aio-make-aio_read_evt-more-efficient-convert-to-hrtimers.patch aio-use-flush_dcache_page.patch aio-use-cancellation-list-lazily.patch aio-change-reqs_active-to-include-unreaped-completions.patch aio-kill-batch-allocation.patch aio-kill-struct-aio_ring_info.patch aio-give-shared-kioctx-fields-their-own-cachelines.patch aio-give-shared-kioctx-fields-their-own-cachelines-fix.patch aio-reqs_active-reqs_available.patch aio-percpu-reqs_available.patch generic-dynamic-per-cpu-refcounting.patch aio-percpu-ioctx-refcount.patch aio-use-xchg-instead-of-completion_lock.patch aio-dont-include-aioh-in-schedh.patch aio-kill-ki_key.patch aio-kill-ki_retry.patch block-aio-batch-completion-for-bios-kiocbs.patch virtio-blk-convert-to-batch-completion.patch mtip32xx-convert-to-batch-completion.patch aio-smoosh-struct-kiocb.patch -- To unsubscribe from this list: send the line "unsubscribe mm-commits" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html