From: Kent Overstreet <koverstreet@google.com>
To: linux-kernel@vger.kernel.org, linux-aio@kvack.org,
linux-fsdevel@vger.kernel.org
Cc: Kent Overstreet <koverstreet@google.com>,
zab@redhat.com, bcrl@kvack.org, jmoyer@redhat.com,
axboe@kernel.dk, viro@zeniv.linux.org.uk, tytso@mit.edu
Subject: [PATCH 22/32] aio: percpu reqs_available
Date: Wed, 26 Dec 2012 18:00:01 -0800 [thread overview]
Message-ID: <1356573611-18590-25-git-send-email-koverstreet@google.com> (raw)
In-Reply-To: <1356573611-18590-1-git-send-email-koverstreet@google.com>
See the previous patch for why we want to do this - this basically
implements a per cpu allocator for reqs_available that doesn't actually
allocate anything.
Note that we need to increase the size of the ringbuffer we allocate,
since a single thread won't necessarily be able to use all the
reqs_available slots - some (up to about half) might be on other per cpu
lists, unavailable for the current thread.
We size the ringbuffer based on the nr_events userspace passed to
io_setup(), so this is a slight behaviour change - but nr_events wasn't
being used as a hard limit before, it was being rounded up to the next
page before so this doesn't change the actual semantics.
Signed-off-by: Kent Overstreet <koverstreet@google.com>
---
fs/aio.c | 92 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++-----
1 file changed, 85 insertions(+), 7 deletions(-)
diff --git a/fs/aio.c b/fs/aio.c
index d384eb2..e415b33 100644
--- a/fs/aio.c
+++ b/fs/aio.c
@@ -26,6 +26,7 @@
#include <linux/mm.h>
#include <linux/mman.h>
#include <linux/mmu_context.h>
+#include <linux/percpu.h>
#include <linux/slab.h>
#include <linux/timer.h>
#include <linux/aio.h>
@@ -59,6 +60,10 @@ struct aio_ring {
#define AIO_RING_PAGES 8
+struct kioctx_cpu {
+ unsigned reqs_available;
+};
+
struct kioctx {
atomic_t users;
atomic_t dead;
@@ -67,6 +72,10 @@ struct kioctx {
unsigned long user_id;
struct hlist_node list;
+ struct __percpu kioctx_cpu *cpu;
+
+ unsigned req_batch;
+
unsigned nr;
/* sys_io_setup currently limits this to an unsigned int */
@@ -149,6 +158,9 @@ static int aio_setup_ring(struct kioctx *ctx)
unsigned long size;
int nr_pages;
+ nr_events = max(nr_events, num_possible_cpus() * 4);
+ nr_events *= 2;
+
/* Compensate for the ring buffer's head/tail overlap entry */
nr_events += 2; /* 1 is required, 2 for good luck */
@@ -255,6 +267,8 @@ static int kiocb_cancel(struct kioctx *ctx, struct kiocb *kiocb,
static void free_ioctx_rcu(struct rcu_head *head)
{
struct kioctx *ctx = container_of(head, struct kioctx, rcu_head);
+
+ free_percpu(ctx->cpu);
kmem_cache_free(kioctx_cachep, ctx);
}
@@ -268,7 +282,7 @@ static void free_ioctx(struct kioctx *ctx)
struct aio_ring *ring;
struct io_event res;
struct kiocb *req;
- unsigned head, avail;
+ unsigned cpu, head, avail;
spin_lock_irq(&ctx->ctx_lock);
@@ -282,6 +296,13 @@ static void free_ioctx(struct kioctx *ctx)
spin_unlock_irq(&ctx->ctx_lock);
+ for_each_possible_cpu(cpu) {
+ struct kioctx_cpu *kcpu = per_cpu_ptr(ctx->cpu, cpu);
+
+ atomic_add(kcpu->reqs_available, &ctx->reqs_available);
+ kcpu->reqs_available = 0;
+ }
+
ring = kmap_atomic(ctx->ring_pages[0]);
head = ring->head;
kunmap_atomic(ring);
@@ -357,10 +378,16 @@ static struct kioctx *ioctx_alloc(unsigned nr_events)
INIT_LIST_HEAD(&ctx->active_reqs);
- if (aio_setup_ring(ctx) < 0)
+ ctx->cpu = alloc_percpu(struct kioctx_cpu);
+ if (!ctx->cpu)
goto out_freectx;
+ if (aio_setup_ring(ctx) < 0)
+ goto out_freepcpu;
+
atomic_set(&ctx->reqs_available, ctx->nr);
+ ctx->req_batch = ctx->nr / (num_possible_cpus() * 4);
+ BUG_ON(!ctx->req_batch);
/* limit the number of system wide aios */
spin_lock(&aio_nr_lock);
@@ -384,6 +411,8 @@ static struct kioctx *ioctx_alloc(unsigned nr_events)
out_cleanup:
err = -EAGAIN;
aio_free_ring(ctx);
+out_freepcpu:
+ free_percpu(ctx->cpu);
out_freectx:
kmem_cache_free(kioctx_cachep, ctx);
pr_debug("error allocating ioctx %d\n", err);
@@ -482,6 +511,52 @@ void exit_aio(struct mm_struct *mm)
}
}
+static void put_reqs_available(struct kioctx *ctx, unsigned nr)
+{
+ struct kioctx_cpu *kcpu;
+
+ preempt_disable();
+ kcpu = this_cpu_ptr(ctx->cpu);
+
+ kcpu->reqs_available += nr;
+ while (kcpu->reqs_available >= ctx->req_batch * 2) {
+ kcpu->reqs_available -= ctx->req_batch;
+ atomic_add(ctx->req_batch, &ctx->reqs_available);
+ }
+
+ preempt_enable();
+}
+
+static bool get_reqs_available(struct kioctx *ctx)
+{
+ struct kioctx_cpu *kcpu;
+ bool ret = false;
+
+ preempt_disable();
+ kcpu = this_cpu_ptr(ctx->cpu);
+
+ if (!kcpu->reqs_available) {
+ int old, avail = atomic_read(&ctx->reqs_available);
+
+ do {
+ if (avail < ctx->req_batch)
+ goto out;
+
+ old = avail;
+ avail = atomic_cmpxchg(&ctx->reqs_available,
+ avail, avail - ctx->req_batch);
+ } while (avail != old);
+
+ kcpu->reqs_available += ctx->req_batch;
+ }
+
+ ret = true;
+ kcpu->reqs_available--;
+out:
+ preempt_enable();
+ return ret;
+}
+
/* aio_get_req
* Allocate a slot for an aio request. Increments the ki_users count
* of the kioctx so that the kioctx stays around until all requests are
@@ -496,7 +571,7 @@ static inline struct kiocb *aio_get_req(struct kioctx *ctx)
{
struct kiocb *req;
- if (atomic_dec_if_positive(&ctx->reqs_available) <= 0)
+ if (!get_reqs_available(ctx))
return NULL;
req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO);
@@ -505,10 +580,9 @@ static inline struct kiocb *aio_get_req(struct kioctx *ctx)
atomic_set(&req->ki_users, 2);
req->ki_ctx = ctx;
-
return req;
out_put:
- atomic_inc(&ctx->reqs_available);
+ put_reqs_available(ctx, 1);
return NULL;
}
@@ -597,6 +671,10 @@ void aio_complete(struct kiocb *iocb, long res, long res2)
*/
if (unlikely(xchg(&iocb->ki_cancel,
KIOCB_CANCELLED) == KIOCB_CANCELLED)) {
+ /*
+ * Can't use the percpu reqs_available here - could race with
+ * free_ioctx()
+ */
atomic_inc(&ctx->reqs_available);
/* Still need the wake_up in case free_ioctx is waiting */
goto put_rq;
@@ -737,7 +815,7 @@ static int aio_read_events_ring(struct kioctx *ctx,
pr_debug("%d h%u t%u\n", ret, head, ctx->tail);
- atomic_add(ret, &ctx->reqs_available);
+ put_reqs_available(ctx, ret);
out:
mutex_unlock(&ctx->ring_lock);
@@ -1156,7 +1234,7 @@ static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
return 0;
out_put_req:
- atomic_inc(&ctx->reqs_available);
+ put_reqs_available(ctx, 1);
aio_put_req(req); /* drop extra ref to req */
aio_put_req(req); /* drop i/o ref to req */
return ret;
--
1.7.12
next prev parent reply other threads:[~2012-12-27 2:00 UTC|newest]
Thread overview: 77+ messages / expand[flat|nested] mbox.gz Atom feed top
2012-12-27 1:59 [PATCH 00/32] AIO performance improvements/cleanups, v3 Kent Overstreet
2012-12-27 1:59 ` [PATCH 01/32] mm: remove old aio use_mm() comment Kent Overstreet
2012-12-27 1:59 ` [PATCH 02/32] aio: remove dead code from aio.h Kent Overstreet
2012-12-27 1:59 ` [PATCH 03/32] gadget: remove only user of aio retry Kent Overstreet
2012-12-27 1:59 ` [PATCH 04/32] aio: remove retry-based AIO Kent Overstreet
2012-12-29 7:36 ` Hillf Danton
2013-01-07 22:12 ` Kent Overstreet
2012-12-29 7:47 ` Hillf Danton
2013-01-07 22:15 ` Kent Overstreet
2012-12-27 1:59 ` [PATCH 05/32] char: add aio_{read,write} to /dev/{null,zero} Kent Overstreet
2012-12-27 1:59 ` [PATCH 06/32] aio: Kill return value of aio_complete() Kent Overstreet
2012-12-27 1:59 ` [PATCH 07/32] aio: kiocb_cancel() Kent Overstreet
2012-12-27 1:59 ` [PATCH 08/32] aio: Move private stuff out of aio.h Kent Overstreet
2012-12-27 1:59 ` [PATCH 09/32] aio: dprintk() -> pr_debug() Kent Overstreet
2012-12-27 1:59 ` [PATCH 10/32] aio: do fget() after aio_get_req() Kent Overstreet
2012-12-27 1:59 ` [PATCH 11/32] aio: Make aio_put_req() lockless Kent Overstreet
2012-12-27 1:59 ` [PATCH 12/32] aio: Refcounting cleanup Kent Overstreet
2012-12-27 1:59 ` [PATCH 13/32] wait: Add wait_event_hrtimeout() Kent Overstreet
2012-12-27 10:37 ` Fubo Chen
2013-01-03 23:08 ` Andrew Morton
2013-01-08 0:09 ` Kent Overstreet
2012-12-27 1:59 ` [PATCH 14/32] aio: Make aio_read_evt() more efficient, convert to hrtimers Kent Overstreet
2013-01-03 23:19 ` Andrew Morton
2013-01-08 0:28 ` Kent Overstreet
2013-01-08 1:00 ` Andrew Morton
2013-01-08 1:28 ` Kent Overstreet
2012-12-27 1:59 ` [PATCH 15/32] aio: Use flush_dcache_page() Kent Overstreet
2012-12-27 1:59 ` [PATCH 16/32] aio: Use cancellation list lazily Kent Overstreet
2012-12-27 1:59 ` [PATCH 17/32] aio: Change reqs_active to include unreaped completions Kent Overstreet
2012-12-27 1:59 ` [PATCH 18/32] aio: Kill batch allocation Kent Overstreet
2012-12-27 1:59 ` [PATCH 19/32] aio: Kill struct aio_ring_info Kent Overstreet
2012-12-27 1:59 ` [PATCH 20/32] aio: Give shared kioctx fields their own cachelines Kent Overstreet
2013-01-03 23:25 ` Andrew Morton
2013-01-07 23:48 ` Kent Overstreet
2012-12-27 2:00 ` [PATCH 21/32] aio: reqs_active -> reqs_available Kent Overstreet
2012-12-27 2:00 ` Kent Overstreet [this message]
2012-12-27 2:00 ` [PATCH 23/32] Generic dynamic per cpu refcounting Kent Overstreet
2013-01-03 22:48 ` Andrew Morton
2013-01-07 23:47 ` Kent Overstreet
2013-01-08 1:03 ` [PATCH] percpu-refcount: Sparse fixes Kent Overstreet
2013-01-25 0:51 ` [PATCH 23/32] Generic dynamic per cpu refcounting Tejun Heo
2013-01-25 1:13 ` Kent Overstreet
2013-01-25 2:03 ` Tejun Heo
2013-01-25 2:09 ` Tejun Heo
2013-01-28 17:48 ` Kent Overstreet
2013-01-28 18:18 ` Tejun Heo
2013-01-25 6:15 ` Rusty Russell
2013-01-28 17:53 ` Kent Overstreet
2013-01-28 17:59 ` Tejun Heo
2013-01-28 18:32 ` Kent Overstreet
2013-01-28 18:57 ` Christoph Lameter
2013-02-08 14:44 ` Tejun Heo
2013-02-08 14:49 ` Jens Axboe
2013-02-08 17:50 ` Andrew Morton
2013-02-08 21:27 ` Kent Overstreet
2013-02-11 14:21 ` Jeff Moyer
2013-02-08 21:17 ` Kent Overstreet
2012-12-27 2:00 ` [PATCH 24/32] aio: Percpu ioctx refcount Kent Overstreet
2012-12-27 2:00 ` [PATCH 25/32] aio: use xchg() instead of completion_lock Kent Overstreet
2013-01-03 23:34 ` Andrew Morton
2013-01-07 23:21 ` Kent Overstreet
2013-01-07 23:35 ` Andrew Morton
2013-01-08 0:01 ` Kent Overstreet
2012-12-27 2:00 ` [PATCH 26/32] aio: Don't include aio.h in sched.h Kent Overstreet
2012-12-27 2:00 ` [PATCH 27/32] aio: Kill ki_key Kent Overstreet
2012-12-27 2:00 ` [PATCH 28/32] aio: Kill ki_retry Kent Overstreet
2012-12-27 2:00 ` [PATCH 29/32] block, aio: Batch completion for bios/kiocbs Kent Overstreet
2013-01-04 9:22 ` Jens Axboe
2013-01-07 23:34 ` Kent Overstreet
2013-01-08 15:33 ` Jeff Moyer
2013-01-08 16:06 ` Kent Overstreet
2013-01-08 16:15 ` Jeff Moyer
2013-01-08 16:48 ` Kent Overstreet
2012-12-27 2:00 ` [PATCH 30/32] virtio-blk: Convert to batch completion Kent Overstreet
2012-12-27 2:00 ` [PATCH 31/32] mtip32xx: " Kent Overstreet
2012-12-27 2:00 ` [PATCH 32/32] aio: Smoosh struct kiocb Kent Overstreet
2013-01-04 9:22 ` [PATCH 00/32] AIO performance improvements/cleanups, v3 Jens Axboe
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1356573611-18590-25-git-send-email-koverstreet@google.com \
--to=koverstreet@google.com \
--cc=axboe@kernel.dk \
--cc=bcrl@kvack.org \
--cc=jmoyer@redhat.com \
--cc=linux-aio@kvack.org \
--cc=linux-fsdevel@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=tytso@mit.edu \
--cc=viro@zeniv.linux.org.uk \
--cc=zab@redhat.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).