io_uring: switch local task_work to a mpscq

The local (DEFER_TASKRUN) task_work list is an llist, which is LIFO
ordered, and hence __io_run_local_work() has to restore the right
running order with an O(n) llist_reverse_order() pass first. On top of
that, a batch that gets capped by max_events needs the leftover entries
parked on a separate ->retry_llist, as they can't be pushed back to the
shared list.

Switch it to the FIFO mpscq. Adds are wait-free instead of a cmpxchg
retry loop, entries are popped in queue order with no reversal pass,
capping a run simply leaves the remainder on the queue, and
->retry_llist goes away entirely. The consumer cursor, ->work_head,
lives with the rest of the ->uring_lock protected state rather than
next to the queue, so that popping entries doesn't dirty the producer
side cacheline.

For low amounts of task_work, this ends up being a bit more efficient
than the existing scheme. As an example of that, doing multishot
receives for 8 clients has the following task_work overhead:

     1.02%  sock-test  [kernel.kallsyms]  [k] io_req_local_work_add
     0.88%  sock-test  [kernel.kallsyms]  [k] __io_run_local_work_loop
     0.60%  sock-test  [kernel.kallsyms]  [k] llist_reverse_order
     0.14%  sock-test  [kernel.kallsyms]  [k] __io_run_local_work
     2.64% at ~46Gb/sec

and after this change:

     1.08%  sock-test  [kernel.kallsyms]  [k] io_req_local_work_add
     1.03%  sock-test  [kernel.kallsyms]  [k] __io_run_local_work
     2.11% at ~53Gb/sec

which has less overhead even though that test run was faster. For a case
of having 1024 clients on a single ring:

     2.22%  sock-test  [kernel.kallsyms]  [k] llist_reverse_order
     0.84%  sock-test  [kernel.kallsyms]  [k] __io_run_local_work_loop
     0.42%  sock-test  [kernel.kallsyms]  [k] io_req_local_work_add
     0.02%  sock-test  [kernel.kallsyms]  [k] __io_run_local_work
     3.50% at ~24Gb/sec

we start to see the llist reversing taking a considerable amount of
time, and the total add+run task_work overhead is around 3.5%. After
the change:

     0.90%  sock-test  [kernel.kallsyms]  [k] __io_run_local_work
     0.42%  sock-test  [kernel.kallsyms]  [k] io_req_local_work_add
     1.32% at ~26Gb/sec

most of that overhead is gone, and performance is better as well.

Caleb Sander Mateos <csander@purestorage.com> reports that it improves
the performance of a ublk 4kb workload by 4% [1], while testing v1 of
this patchset.

[1] https://lore.kernel.org/io-uring/CADUfDZr-MMYBaP-e+y9+xuRhuiunO2sBTUCmwZyd7AgT8sVtiQ@mail.gmail.com/

Signed-off-by: Jens Axboe <axboe@kernel.dk>
This commit is contained in:
Jens Axboe
2026-06-10 15:19:35 -06:00
parent 50cb44bd0d
commit d46ab2c98a
6 changed files with 87 additions and 89 deletions

View File

@@ -360,6 +360,14 @@ struct io_ring_ctx {
bool poll_multi_queue; bool poll_multi_queue;
struct list_head iopoll_list; struct list_head iopoll_list;
/*
* Consumer cursor for ->work_list, protected by ->uring_lock.
* Deliberately kept away from the producer side of the queue,
* as it's written for every popped entry, and the producer
* cacheline is contended enough as it is.
*/
struct llist_node *work_head;
struct io_file_table file_table; struct io_file_table file_table;
struct io_rsrc_data buf_table; struct io_rsrc_data buf_table;
struct io_alloc_cache node_cache; struct io_alloc_cache node_cache;
@@ -417,8 +425,7 @@ struct io_ring_ctx {
*/ */
struct { struct {
struct io_rings __rcu *rings_rcu; struct io_rings __rcu *rings_rcu;
struct llist_head work_llist; struct mpscq work_list;
struct llist_head retry_llist;
unsigned long check_cq; unsigned long check_cq;
atomic_t cq_wait_nr; atomic_t cq_wait_nr;
atomic_t cq_timeouts; atomic_t cq_timeouts;
@@ -742,8 +749,6 @@ struct io_kiocb {
*/ */
u16 buf_index; u16 buf_index;
unsigned nr_tw;
/* REQ_F_* flags */ /* REQ_F_* flags */
io_req_flags_t flags; io_req_flags_t flags;

View File

@@ -280,7 +280,7 @@ static __cold struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
INIT_LIST_HEAD(&ctx->defer_list); INIT_LIST_HEAD(&ctx->defer_list);
INIT_LIST_HEAD(&ctx->timeout_list); INIT_LIST_HEAD(&ctx->timeout_list);
INIT_LIST_HEAD(&ctx->ltimeout_list); INIT_LIST_HEAD(&ctx->ltimeout_list);
init_llist_head(&ctx->work_llist); mpscq_init(&ctx->work_list, &ctx->work_head);
INIT_LIST_HEAD(&ctx->tctx_list); INIT_LIST_HEAD(&ctx->tctx_list);
mutex_init(&ctx->tctx_lock); mutex_init(&ctx->tctx_lock);
ctx->submit_state.free_list.next = NULL; ctx->submit_state.free_list.next = NULL;

View File

@@ -14,6 +14,7 @@
#include "rw.h" #include "rw.h"
#include "eventfd.h" #include "eventfd.h"
#include "wait.h" #include "wait.h"
#include "mpscq.h"
void io_fallback_req_func(struct work_struct *work) void io_fallback_req_func(struct work_struct *work)
{ {
@@ -170,11 +171,7 @@ static void io_ctx_mark_taskrun(struct io_ring_ctx *ctx)
void io_req_local_work_add(struct io_kiocb *req, unsigned flags) void io_req_local_work_add(struct io_kiocb *req, unsigned flags)
{ {
struct io_ring_ctx *ctx = req->ctx; struct io_ring_ctx *ctx = req->ctx;
unsigned nr_wait, nr_tw, nr_tw_prev; int nr_wait;
struct llist_node *head;
/* See comment above IO_CQ_WAKE_INIT */
BUILD_BUG_ON(IO_CQ_WAKE_FORCE <= IORING_MAX_CQ_ENTRIES);
/* /*
* We don't know how many requests there are in the link and whether * We don't know how many requests there are in the link and whether
@@ -183,56 +180,45 @@ void io_req_local_work_add(struct io_kiocb *req, unsigned flags)
if (req->flags & IO_REQ_LINK_FLAGS) if (req->flags & IO_REQ_LINK_FLAGS)
flags &= ~IOU_F_TWQ_LAZY_WAKE; flags &= ~IOU_F_TWQ_LAZY_WAKE;
guard(rcu)();
head = READ_ONCE(ctx->work_llist.first);
do {
nr_tw_prev = 0;
if (head) {
struct io_kiocb *first_req = container_of(head,
struct io_kiocb,
io_task_work.node);
/*
* Might be executed at any moment, rely on
* SLAB_TYPESAFE_BY_RCU to keep it alive.
*/
nr_tw_prev = READ_ONCE(first_req->nr_tw);
}
/*
* Theoretically, it can overflow, but that's fine as one of
* previous adds should've tried to wake the task.
*/
nr_tw = nr_tw_prev + 1;
if (!(flags & IOU_F_TWQ_LAZY_WAKE))
nr_tw = IO_CQ_WAKE_FORCE;
req->nr_tw = nr_tw;
req->io_task_work.node.next = head;
} while (!try_cmpxchg(&ctx->work_llist.first, &head,
&req->io_task_work.node));
/* /*
* cmpxchg implies a full barrier, which pairs with the barrier * The xchg() in mpscq_push() implies a full barrier, which pairs with
* in set_current_state() on the io_cqring_wait() side. It's used * the barrier in set_current_state() on the io_cqring_wait() side. This
* to ensure that either we see updated ->cq_wait_nr, or waiters * ensures that either we see the updated ->cq_wait_nr, or waiters going
* going to sleep will observe the work added to the list, which * to sleep will observe the work added to the list, which is similar to
* is similar to the wait/wawke task state sync. * the wait/wake task state sync.
*/ */
if (mpscq_push(&ctx->work_list, &req->io_task_work.node)) {
if (!head) {
io_ctx_mark_taskrun(ctx); io_ctx_mark_taskrun(ctx);
if (data_race(ctx->int_flags) & IO_RING_F_HAS_EVFD) if (data_race(ctx->int_flags) & IO_RING_F_HAS_EVFD)
io_eventfd_signal(ctx, false); io_eventfd_signal(ctx, false);
} }
/*
* No one is waiting (IO_CQ_WAKE_INIT), or this cycle's wake up has
* already been issued (zero or negative, see below).
*/
nr_wait = atomic_read(&ctx->cq_wait_nr); nr_wait = atomic_read(&ctx->cq_wait_nr);
/* not enough or no one is waiting */ if (nr_wait <= 0)
if (nr_tw < nr_wait)
return; return;
/* the previous add has already woken it up */ if (flags & IOU_F_TWQ_LAZY_WAKE) {
if (nr_tw_prev >= nr_wait) /*
* ->cq_wait_nr counts down the number of lazy adds, once it
* hits zero we're good to wake the waiter. A producer that
* gets delayed between pushing its entry and getting here
* may count down a later wait cycle. That's OK, it'll be an
* early wake, not a lost one.
*/
if (!atomic_dec_and_test(&ctx->cq_wait_nr))
return;
} else if (atomic_xchg(&ctx->cq_wait_nr, IO_CQ_WAKE_INIT) <= 0) {
/*
* Potentially raced with lazy add, claim the wake. A value
* <= 0 means a lazy add hit zero or another forced add
* claimed IO_CQ_WAKE_INIT. Either way, the wake up for this
* wait cycle has already been done.
*/
return; return;
}
wake_up_state(ctx->submitter_task, TASK_INTERRUPTIBLE); wake_up_state(ctx->submitter_task, TASK_INTERRUPTIBLE);
} }
@@ -273,21 +259,27 @@ void io_req_task_work_add_remote(struct io_kiocb *req, unsigned flags)
void __cold io_move_task_work_from_local(struct io_ring_ctx *ctx) void __cold io_move_task_work_from_local(struct io_ring_ctx *ctx)
{ {
struct llist_node *node; struct llist_node *node, *first = NULL, **tail = &first;
/* /*
* Running the work items may utilize ->retry_llist as a means * The work list consumer side is serialized by ->uring_lock, see
* for capping the number of task_work entries run at the same * __io_run_local_work(). Grab it to guard against racing with normal
* time. But that list can potentially race with moving the work * task_work running, as the task may be exiting.
* from here, if the task is exiting. As any normal task_work
* running holds ->uring_lock already, just guard this slow path
* with ->uring_lock to avoid racing on ->retry_llist.
*/ */
guard(mutex)(&ctx->uring_lock); guard(mutex)(&ctx->uring_lock);
node = llist_del_all(&ctx->work_llist);
__io_fallback_tw(node, false); while (!mpscq_empty(&ctx->work_list)) {
node = llist_del_all(&ctx->retry_llist); node = mpscq_pop(&ctx->work_list, &ctx->work_head);
__io_fallback_tw(node, false); if (!node) {
/* a producer is mid-push, wait for it to link */
cpu_relax();
continue;
}
*tail = node;
tail = &node->next;
}
*tail = NULL;
__io_fallback_tw(first, false);
} }
static bool io_run_local_work_continue(struct io_ring_ctx *ctx, int events, static bool io_run_local_work_continue(struct io_ring_ctx *ctx, int events,
@@ -302,22 +294,23 @@ static bool io_run_local_work_continue(struct io_ring_ctx *ctx, int events,
return false; return false;
} }
static int __io_run_local_work_loop(struct llist_node **node, static int __io_run_local_work_loop(struct io_ring_ctx *ctx,
io_tw_token_t tw, io_tw_token_t tw,
int events) int events)
{ {
int ret = 0; int ret = 0;
while (*node) { while (ret < events) {
struct llist_node *next = (*node)->next; struct llist_node *node = mpscq_pop(&ctx->work_list, &ctx->work_head);
struct io_kiocb *req = container_of(*node, struct io_kiocb, struct io_kiocb *req;
io_task_work.node);
if (!node)
break;
req = container_of(node, struct io_kiocb, io_task_work.node);
INDIRECT_CALL_2(req->io_task_work.func, INDIRECT_CALL_2(req->io_task_work.func,
io_poll_task_func, io_req_rw_complete, io_poll_task_func, io_req_rw_complete,
(struct io_tw_req){req}, tw); (struct io_tw_req){req}, tw);
*node = next; ret++;
if (++ret >= events)
break;
} }
return ret; return ret;
@@ -326,7 +319,6 @@ static int __io_run_local_work_loop(struct llist_node **node,
static int __io_run_local_work(struct io_ring_ctx *ctx, io_tw_token_t tw, static int __io_run_local_work(struct io_ring_ctx *ctx, io_tw_token_t tw,
int min_events, int max_events) int min_events, int max_events)
{ {
struct llist_node *node;
unsigned int loops = 0; unsigned int loops = 0;
int ret = 0; int ret = 0;
@@ -335,24 +327,21 @@ static int __io_run_local_work(struct io_ring_ctx *ctx, io_tw_token_t tw,
if (ctx->flags & IORING_SETUP_TASKRUN_FLAG) if (ctx->flags & IORING_SETUP_TASKRUN_FLAG)
atomic_andnot(IORING_SQ_TASKRUN, &ctx->rings->sq_flags); atomic_andnot(IORING_SQ_TASKRUN, &ctx->rings->sq_flags);
again: again:
/*
* If the last loop made no progress while work is still pending,
* a producer has published a node but hasn't linked it into the
* queue yet (see mpscq_pop()). Give it a chance to finish rather
* than spinning on the queue.
*/
if (unlikely(loops && !ret))
cond_resched();
tw.cancel = io_should_terminate_tw(ctx); tw.cancel = io_should_terminate_tw(ctx);
min_events -= ret; min_events -= ret;
ret = __io_run_local_work_loop(&ctx->retry_llist.first, tw, max_events); ret = __io_run_local_work_loop(ctx, tw, max_events);
if (ctx->retry_llist.first)
goto retry_done;
/*
* llists are in reverse order, flip it back the right way before
* running the pending items.
*/
node = llist_reverse_order(llist_del_all(&ctx->work_llist));
ret += __io_run_local_work_loop(&node, tw, max_events - ret);
ctx->retry_llist.first = node;
loops++; loops++;
if (io_run_local_work_continue(ctx, ret, min_events)) if (io_run_local_work_continue(ctx, ret, min_events))
goto again; goto again;
retry_done:
io_submit_flush_completions(ctx); io_submit_flush_completions(ctx);
if (io_run_local_work_continue(ctx, ret, min_events)) if (io_run_local_work_continue(ctx, ret, min_events))
goto again; goto again;

View File

@@ -6,6 +6,8 @@
#include <linux/percpu-refcount.h> #include <linux/percpu-refcount.h>
#include <linux/io_uring_types.h> #include <linux/io_uring_types.h>
#include "mpscq.h"
#define IO_LOCAL_TW_DEFAULT_MAX 20 #define IO_LOCAL_TW_DEFAULT_MAX 20
/* /*
@@ -89,7 +91,7 @@ static inline int io_run_task_work(void)
static inline bool io_local_work_pending(struct io_ring_ctx *ctx) static inline bool io_local_work_pending(struct io_ring_ctx *ctx)
{ {
return !llist_empty(&ctx->work_llist) || !llist_empty(&ctx->retry_llist); return !mpscq_empty(&ctx->work_list);
} }
static inline bool io_task_work_pending(struct io_ring_ctx *ctx) static inline bool io_task_work_pending(struct io_ring_ctx *ctx)

View File

@@ -98,7 +98,7 @@ static enum hrtimer_restart io_cqring_min_timer_wakeup(struct hrtimer *timer)
if (ctx->flags & IORING_SETUP_DEFER_TASKRUN) { if (ctx->flags & IORING_SETUP_DEFER_TASKRUN) {
atomic_set(&ctx->cq_wait_nr, 1); atomic_set(&ctx->cq_wait_nr, 1);
smp_mb(); smp_mb();
if (!llist_empty(&ctx->work_llist)) if (io_local_work_pending(ctx))
goto out_wake; goto out_wake;
} }

View File

@@ -5,12 +5,14 @@
#include <linux/io_uring_types.h> #include <linux/io_uring_types.h>
/* /*
* No waiters. It's larger than any valid value of the tw counter * ->cq_wait_nr is armed with the number of lazy task_work adds the waiter
* so that tests against ->cq_wait_nr would fail and skip wake_up(). * still needs, and counted down by the add side, with the add reaching zero
* issuing the (single) wake up for this wait cycle. Zero and below means no
* wake up is to be issued: IO_CQ_WAKE_INIT when no task is waiting (also
* what a forced wake up resets it to when claiming one), zero once the
* countdown has fired.
*/ */
#define IO_CQ_WAKE_INIT (-1U) #define IO_CQ_WAKE_INIT (-1)
/* Forced wake up if there is a waiter regardless of ->cq_wait_nr */
#define IO_CQ_WAKE_FORCE (IO_CQ_WAKE_INIT >> 1)
struct ext_arg { struct ext_arg {
size_t argsz; size_t argsz;