Files
Austin Schuh 02caa59221 io_uring/futex: return a wait's data to the cache when it fails to arm
io_futex_wait() takes its io_futex_data from ctx->futex_cache, and a
completed wait gives it back in io_futex_complete().  A wait that fails
to arm kfree()s it instead.

The usual failure is -EAGAIN from futex_wait_setup(): the futex word no
longer holds the expected value.  That is routine for a waiter that
re-arms with the value it last read, since the word changes whenever
there is something to wake for.  Each such failure takes an object out
of the cache for good, so the cache drains and from then on waits
allocate.

The cache exists so that arming a wait doesn't have to go to the slab
allocator, and that matters most to the waiters that re-arm the most:
event loops, often real-time ones.  A kmalloc() or kfree() that misses
the per-CPU slab takes the node's list_lock, shared with every other
user of that kmalloc size on the system, and may go on to the page
allocator.

On PREEMPT_RT list_lock is a sleeping lock.  A real-time task that
finds it held waits for the holder, which may be any task on any CPU,
to get to run and release it.  Priority inheritance boosts the holder
only up to the waiter's priority, so the waiter still sits behind
whatever outranks that on the holder's CPU, such as threaded interrupt
handlers.  Re-arming a futex wait should not depend on any of that.

Put the data back in the cache on that path too, while the submit lock
still protects it.

In an event loop re-arming one IORING_OP_FUTEX_WAIT per message under
load, 0.39 arms per message failed with -EAGAIN, and io_futex_wait() did
0.39 kmalloc()s and kfree()s per message.  With this change, 100,000
waits that all fail with -EAGAIN do one kmalloc() in total.

Fixes: 194bb58c60 ("io_uring: add support for futex wake and wait")
Assisted-by: LLM
Signed-off-by: Austin Schuh <[email protected]>
Link: https://patch.msgid.link/[email protected]
Signed-off-by: Jens Axboe <[email protected]>
2026-10-20 08:34:27 -06:00

385 lines
9.7 KiB
C

// SPDX-License-Identifier: GPL-2.0
#include <linux/kernel.h>
#include <linux/errno.h>
#include <linux/fs.h>
#include <linux/file.h>
#include <linux/io_uring.h>
#include <uapi/linux/io_uring.h>
#include "../kernel/futex/futex.h"
#include "io_uring.h"
#include "alloc_cache.h"
#include "futex.h"
struct io_futex {
struct file *file;
void __user *uaddr;
unsigned long futex_val;
unsigned long futex_mask;
u32 futex_flags;
unsigned int futex_nr;
bool futexv_unqueued;
};
struct io_futex_data {
struct futex_q q;
struct io_kiocb *req;
};
struct io_futexv_data {
unsigned long owned;
struct futex_vector futexv[];
};
#define IO_FUTEX_ALLOC_CACHE_MAX 32
bool io_futex_cache_init(struct io_ring_ctx *ctx)
{
return io_alloc_cache_init(&ctx->futex_cache, IO_FUTEX_ALLOC_CACHE_MAX,
sizeof(struct io_futex_data), 0);
}
void io_futex_cache_free(struct io_ring_ctx *ctx)
{
io_alloc_cache_free(&ctx->futex_cache, kfree);
}
static void __io_futex_complete(struct io_tw_req tw_req, io_tw_token_t tw)
{
hlist_del_init(&tw_req.req->hash_node);
io_req_task_complete(tw_req, tw);
}
static void io_futex_complete(struct io_tw_req tw_req, io_tw_token_t tw)
{
struct io_kiocb *req = tw_req.req;
struct io_ring_ctx *ctx = req->ctx;
io_tw_lock(ctx, tw);
io_cache_free(&ctx->futex_cache, req->async_data);
io_req_async_data_clear(req, 0);
__io_futex_complete(tw_req, tw);
}
static void io_futexv_complete(struct io_tw_req tw_req, io_tw_token_t tw)
{
struct io_kiocb *req = tw_req.req;
struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex);
struct io_futexv_data *ifd = req->async_data;
io_tw_lock(req->ctx, tw);
if (!iof->futexv_unqueued) {
int res;
res = futex_unqueue_multiple(ifd->futexv, iof->futex_nr);
if (res != -1)
io_req_set_res(req, res, 0);
}
io_req_async_data_free(req);
__io_futex_complete(tw_req, tw);
}
static bool io_futexv_claim(struct io_futexv_data *ifd)
{
if (test_bit(0, &ifd->owned) || test_and_set_bit_lock(0, &ifd->owned))
return false;
return true;
}
static bool __io_futex_cancel(struct io_kiocb *req)
{
/* futex wake already done or in progress */
if (req->opcode == IORING_OP_FUTEX_WAIT) {
struct io_futex_data *ifd = req->async_data;
if (!futex_unqueue(&ifd->q))
return false;
req->io_task_work.func = io_futex_complete;
} else {
struct io_futexv_data *ifd = req->async_data;
if (!io_futexv_claim(ifd))
return false;
req->io_task_work.func = io_futexv_complete;
}
hlist_del_init(&req->hash_node);
io_req_set_res(req, -ECANCELED, 0);
io_req_task_work_add(req);
return true;
}
int io_futex_cancel(struct io_ring_ctx *ctx, struct io_cancel_data *cd,
unsigned int issue_flags)
{
return io_cancel_remove(ctx, cd, issue_flags, &ctx->futex_list, __io_futex_cancel);
}
bool io_futex_remove_all(struct io_ring_ctx *ctx, struct io_uring_task *tctx,
bool cancel_all)
{
return io_cancel_remove_all(ctx, tctx, &ctx->futex_list, cancel_all, __io_futex_cancel);
}
int io_futex_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
{
struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex);
u32 flags;
if (unlikely(sqe->len || sqe->futex_flags || sqe->buf_index ||
sqe->file_index))
return -EINVAL;
iof->uaddr = u64_to_user_ptr(READ_ONCE(sqe->addr));
iof->futex_val = READ_ONCE(sqe->addr2);
iof->futex_mask = READ_ONCE(sqe->addr3);
flags = READ_ONCE(sqe->fd);
if (flags & ~FUTEX2_VALID_MASK)
return -EINVAL;
iof->futex_flags = futex2_to_flags(flags);
if (!futex_flags_valid(iof->futex_flags))
return -EINVAL;
if (!futex_validate_input(iof->futex_flags, iof->futex_val) ||
!futex_validate_input(iof->futex_flags, iof->futex_mask))
return -EINVAL;
return 0;
}
/*
* A wait on a private futex does not block the submitter. Even a single
* threaded task can go on and have the mm private hash installed by clone()
* or prctl(), which does not pick up waiters from the global hash. Make
* sure the private hash exists before the first waiter is queued.
*/
static int io_futex_prep_private(struct io_kiocb *req)
{
int ret = futex_hash_allocate_default();
if (unlikely(ret))
return ret;
/* inflight tracking only needed for mm private hash */
io_req_track_inflight(req);
return 0;
}
int io_futex_wait_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
{
struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex);
int ret;
ret = io_futex_prep(req, sqe);
if (unlikely(ret))
return ret;
if (!(iof->futex_flags & FLAGS_SHARED))
return io_futex_prep_private(req);
return 0;
}
static void io_futex_wakev_fn(struct wake_q_head *wake_q, struct futex_q *q)
{
struct io_kiocb *req = q->wake_data;
struct io_futexv_data *ifd = req->async_data;
if (!io_futexv_claim(ifd)) {
__futex_wake_mark(q);
return;
}
if (unlikely(!__futex_wake_mark(q)))
return;
io_req_set_res(req, 0, 0);
req->io_task_work.func = io_futexv_complete;
__io_req_task_work_add(req, IOU_F_TWQ_IN_WAKE);
}
int io_futexv_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
{
struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex);
struct io_futexv_data *ifd;
unsigned int i;
int ret;
/* No flags or mask supported for waitv */
if (unlikely(sqe->fd || sqe->buf_index || sqe->file_index ||
sqe->addr2 || sqe->futex_flags || sqe->addr3))
return -EINVAL;
iof->uaddr = u64_to_user_ptr(READ_ONCE(sqe->addr));
iof->futex_nr = READ_ONCE(sqe->len);
if (!iof->futex_nr || iof->futex_nr > FUTEX_WAITV_MAX)
return -EINVAL;
ifd = kzalloc_flex(struct io_futexv_data, futexv, iof->futex_nr,
GFP_KERNEL_ACCOUNT);
if (!ifd)
return -ENOMEM;
ret = futex_parse_waitv(ifd->futexv, iof->uaddr, iof->futex_nr,
io_futex_wakev_fn, req);
if (ret) {
kfree(ifd);
return ret;
}
for (i = 0; i < iof->futex_nr; i++) {
if (!(ifd->futexv[i].w.flags & FLAGS_SHARED)) {
ret = io_futex_prep_private(req);
if (ret) {
kfree(ifd);
return ret;
}
break;
}
}
iof->futexv_unqueued = 0;
req->flags |= REQ_F_ASYNC_DATA;
req->async_data = ifd;
return 0;
}
static void io_futex_wake_fn(struct wake_q_head *wake_q, struct futex_q *q)
{
struct io_futex_data *ifd = container_of(q, struct io_futex_data, q);
struct io_kiocb *req = ifd->req;
if (unlikely(!__futex_wake_mark(q)))
return;
io_req_set_res(req, 0, 0);
req->io_task_work.func = io_futex_complete;
__io_req_task_work_add(req, IOU_F_TWQ_IN_WAKE);
}
int io_futexv_wait(struct io_kiocb *req, unsigned int issue_flags)
{
struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex);
struct io_futexv_data *ifd = req->async_data;
struct io_ring_ctx *ctx = req->ctx;
int ret, woken = -1;
io_ring_submit_lock(ctx, issue_flags);
ret = futex_wait_multiple_setup(ifd->futexv, iof->futex_nr, &woken);
/*
* Error case, ret is < 0. Mark the request as failed.
*/
if (unlikely(ret < 0)) {
io_ring_submit_unlock(ctx, issue_flags);
req_set_fail(req);
io_req_set_res(req, ret, 0);
io_req_async_data_free(req);
return IOU_COMPLETE;
}
/*
* 0 return means that we successfully setup the waiters, and that
* nobody triggered a wakeup while we were doing so. If the wakeup
* happened post setup, the task_work will be run post this issue and
* under the submission lock. 1 means We got woken while setting up,
* let that side do the completion. Note that
* futex_wait_multiple_setup() will have unqueued all the futexes in
* this case. Mark us as having done that already, since this is
* different from normal wakeup.
*/
if (!ret) {
/*
* If futex_wait_multiple_setup() returns 0 for a
* successful setup, then the task state will not be
* runnable. This is fine for the sync syscall, as
* it'll be blocking unless we already got one of the
* futexes woken, but it obviously won't work for an
* async invocation. Mark us runnable again.
*/
__set_current_state(TASK_RUNNING);
hlist_add_head(&req->hash_node, &ctx->futex_list);
if (io_wq_cancel_seen(req, issue_flags))
__io_futex_cancel(req);
} else {
iof->futexv_unqueued = 1;
if (woken != -1)
io_req_set_res(req, woken, 0);
}
io_ring_submit_unlock(ctx, issue_flags);
return IOU_ISSUE_SKIP_COMPLETE;
}
int io_futex_wait(struct io_kiocb *req, unsigned int issue_flags)
{
struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex);
struct io_ring_ctx *ctx = req->ctx;
struct io_futex_data *ifd = NULL;
int ret;
if (!iof->futex_mask) {
ret = -EINVAL;
goto done;
}
io_ring_submit_lock(ctx, issue_flags);
ifd = io_cache_alloc(&ctx->futex_cache, GFP_NOWAIT);
if (!ifd) {
ret = -ENOMEM;
goto done_unlock;
}
req->flags |= REQ_F_ASYNC_DATA;
req->async_data = ifd;
ifd->q = futex_q_init;
ifd->q.bitset = iof->futex_mask;
ifd->q.wake = io_futex_wake_fn;
ifd->req = req;
ret = futex_wait_setup(iof->uaddr, iof->futex_val, iof->futex_flags,
&ifd->q, NULL, NULL);
if (!ret) {
hlist_add_head(&req->hash_node, &ctx->futex_list);
if (io_wq_cancel_seen(req, issue_flags))
__io_futex_cancel(req);
io_ring_submit_unlock(ctx, issue_flags);
return IOU_ISSUE_SKIP_COMPLETE;
}
done_unlock:
if (ifd) {
/* As in io_futex_complete(), or failed arms would drain it. */
io_cache_free(&ctx->futex_cache, ifd);
io_req_async_data_clear(req, 0);
}
io_ring_submit_unlock(ctx, issue_flags);
done:
if (ret < 0)
req_set_fail(req);
io_req_set_res(req, ret, 0);
return IOU_COMPLETE;
}
int io_futex_wake(struct io_kiocb *req, unsigned int issue_flags)
{
struct io_futex *iof = io_kiocb_to_cmd(req, struct io_futex);
int ret;
/*
* Strict flags - ensure that waking 0 futexes yields a 0 result.
* See commit 43adf8449510 ("futex: FLAGS_STRICT") for details.
*/
ret = futex_wake(iof->uaddr, FLAGS_STRICT | iof->futex_flags, NULL,
iof->futex_val, iof->futex_mask);
if (ret < 0)
req_set_fail(req);
io_req_set_res(req, ret, 0);
return IOU_COMPLETE;
}