From 5eef4e87eb0b2b8482f6a77ebcad067636a867b3 Mon Sep 17 00:00:00 2001
From: Pavel Begunkov <asml.silence@gmail.com>
Date: Fri, 24 Sep 2021 21:59:49 +0100
Subject: [PATCH] io_uring: use single linked list for iopoll
Git-commit: 5eef4e87eb0b2b8482f6a77ebcad067636a867b3
Patch-mainline: v5.16-rc1
References: bsc#1205205
Use single linked lists for keeping iopoll requests, takes less space,
may be faster, but mostly will be of benefit for further patches.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/314033676b100cd485518c3bc55e1b95a0dcd71f.1632516769.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Gabriel Krisman Bertazi <krisman@suse.de>
---
fs/io-wq.h | 3 +++
fs/io_uring.c | 54 +++++++++++++++++++++++++--------------------------
2 files changed, 30 insertions(+), 27 deletions(-)
diff --git a/fs/io-wq.h b/fs/io-wq.h
index c870062105d1..87ba6a733630 100644
--- a/fs/io-wq.h
+++ b/fs/io-wq.h
@@ -32,6 +32,9 @@ struct io_wq_work_list {
#define wq_list_for_each(pos, prv, head) \
for (pos = (head)->first, prv = NULL; pos; prv = pos, pos = (pos)->next)
+#define wq_list_for_each_resume(pos, prv) \
+ for (; pos; prv = pos, pos = (pos)->next)
+
#define wq_list_empty(list) (READ_ONCE((list)->first) == NULL)
#define INIT_WQ_LIST(list) do { \
(list)->first = NULL; \
diff --git a/fs/io_uring.c b/fs/io_uring.c
index 0e46ed6f92ea..fac1d0310d66 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -412,7 +412,7 @@ struct io_ring_ctx {
* For SQPOLL, only the single threaded io_sq_thread() will
* manipulate the list, hence no extra locking is needed there.
*/
- struct list_head iopoll_list;
+ struct io_wq_work_list iopoll_list;
struct hlist_head *cancel_hash;
unsigned cancel_hash_bits;
bool poll_multi_queue;
@@ -1309,7 +1309,7 @@ static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
init_waitqueue_head(&ctx->cq_wait);
spin_lock_init(&ctx->completion_lock);
spin_lock_init(&ctx->timeout_lock);
- INIT_LIST_HEAD(&ctx->iopoll_list);
+ INIT_WQ_LIST(&ctx->iopoll_list);
INIT_LIST_HEAD(&ctx->defer_list);
INIT_LIST_HEAD(&ctx->timeout_list);
INIT_LIST_HEAD(&ctx->ltimeout_list);
@@ -2441,15 +2441,9 @@ static void io_iopoll_complete(struct io_ring_ctx *ctx, struct list_head *done)
io_req_free_batch_finish(ctx, &rb);
}
-/* same as "continue" but starts from the pos, not next to it */
-#define list_for_each_entry_safe_resume(pos, n, head, member) \
- for (n = list_next_entry(pos, member); \
- !list_entry_is_head(pos, head, member); \
- pos = n, n = list_next_entry(n, member))
-
static int io_do_iopoll(struct io_ring_ctx *ctx, bool force_nonspin)
{
- struct io_kiocb *req, *tmp;
+ struct io_wq_work_node *pos, *start, *prev;
unsigned int poll_flags = BLK_POLL_NOSLEEP;
DEFINE_IO_COMP_BATCH(iob);
int nr_events = 0;
@@ -2462,7 +2456,8 @@ static int io_do_iopoll(struct io_ring_ctx *ctx, bool force_nonspin)
if (ctx->poll_multi_queue || force_nonspin)
poll_flags |= BLK_POLL_ONESHOT;
- list_for_each_entry(req, &ctx->iopoll_list, inflight_entry) {
+ wq_list_for_each(pos, start, &ctx->iopoll_list) {
+ struct io_kiocb *req = container_of(pos, struct io_kiocb, comp_list);
struct kiocb *kiocb = &req->rw.kiocb;
int ret;
@@ -2488,14 +2483,20 @@ static int io_do_iopoll(struct io_ring_ctx *ctx, bool force_nonspin)
if (!rq_list_empty(iob.req_list))
iob.complete(&iob);
- list_for_each_entry_safe_resume(req, tmp, &ctx->iopoll_list,
- inflight_entry) {
+ else if (!pos)
+ return 0;
+
+ prev = start;
+ wq_list_for_each_resume(pos, prev) {
+ struct io_kiocb *req = container_of(pos, struct io_kiocb, comp_list);
+
if (!READ_ONCE(req->iopoll_completed))
break;
- list_move_tail(&req->inflight_entry, &done);
+ list_add_tail(&req->inflight_entry, &done);
nr_events++;
}
+ wq_list_cut(&ctx->iopoll_list, prev, start);
if (nr_events)
io_iopoll_complete(ctx, &done);
return nr_events;
@@ -2511,7 +2512,7 @@ static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx)
return;
mutex_lock(&ctx->uring_lock);
- while (!list_empty(&ctx->iopoll_list)) {
+ while (!wq_list_empty(&ctx->iopoll_list)) {
/* let it sleep and repeat later if can't complete a request */
if (io_do_iopoll(ctx, true) == 0)
break;
@@ -2560,7 +2561,7 @@ static int io_iopoll_check(struct io_ring_ctx *ctx, long min)
* forever, while the workqueue is stuck trying to acquire the
* very same mutex.
*/
- if (list_empty(&ctx->iopoll_list)) {
+ if (wq_list_empty(&ctx->iopoll_list)) {
u32 tail = ctx->cached_cq_tail;
mutex_unlock(&ctx->uring_lock);
@@ -2569,7 +2570,7 @@ static int io_iopoll_check(struct io_ring_ctx *ctx, long min)
/* some requests don't go through iopoll_list */
if (tail != ctx->cached_cq_tail ||
- list_empty(&ctx->iopoll_list))
+ wq_list_empty(&ctx->iopoll_list))
break;
}
ret = io_do_iopoll(ctx, !min);
@@ -2729,14 +2730,13 @@ static void io_iopoll_req_issued(struct io_kiocb *req)
* how we do polling eventually, not spinning if we're on potentially
* different devices.
*/
- if (list_empty(&ctx->iopoll_list)) {
+ if (wq_list_empty(&ctx->iopoll_list)) {
ctx->poll_multi_queue = false;
} else if (!ctx->poll_multi_queue) {
struct io_kiocb *list_req;
- list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb,
- inflight_entry);
-
+ list_req = container_of(ctx->iopoll_list.first, struct io_kiocb,
+ comp_list);
if (list_req->file != req->file)
ctx->poll_multi_queue = true;
}
@@ -2746,9 +2746,9 @@ static void io_iopoll_req_issued(struct io_kiocb *req)
* it to the front so we find it first.
*/
if (READ_ONCE(req->iopoll_completed))
- list_add(&req->inflight_entry, &ctx->iopoll_list);
+ wq_list_add_head(&req->comp_list, &ctx->iopoll_list);
else
- list_add_tail(&req->inflight_entry, &ctx->iopoll_list);
+ wq_list_add_tail(&req->comp_list, &ctx->iopoll_list);
if (unlikely(in_async)) {
/*
@@ -7322,14 +7322,14 @@ static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries)
if (cap_entries && to_submit > IORING_SQPOLL_CAP_ENTRIES_VALUE)
to_submit = IORING_SQPOLL_CAP_ENTRIES_VALUE;
- if (!list_empty(&ctx->iopoll_list) || to_submit) {
+ if (!wq_list_empty(&ctx->iopoll_list) || to_submit) {
const struct cred *creds = NULL;
if (ctx->sq_creds != current_cred())
creds = override_creds(ctx->sq_creds);
mutex_lock(&ctx->uring_lock);
- if (!list_empty(&ctx->iopoll_list))
+ if (!wq_list_empty(&ctx->iopoll_list))
io_do_iopoll(ctx, true);
/*
@@ -7407,7 +7407,7 @@ static int io_sq_thread(void *data)
list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
int ret = __io_sq_thread(ctx, cap_entries);
- if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list)))
+ if (!sqt_spin && (ret > 0 || !wq_list_empty(&ctx->iopoll_list)))
sqt_spin = true;
}
if (io_run_task_work())
@@ -7428,7 +7428,7 @@ static int io_sq_thread(void *data)
io_ring_set_wakeup_flag(ctx);
if ((ctx->flags & IORING_SETUP_IOPOLL) &&
- !list_empty_careful(&ctx->iopoll_list)) {
+ !wq_list_empty(&ctx->iopoll_list)) {
needs_sched = false;
break;
}
@@ -9583,7 +9583,7 @@ static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
/* SQPOLL thread does its own polling */
if ((!(ctx->flags & IORING_SETUP_SQPOLL) && cancel_all) ||
(ctx->sq_data && ctx->sq_data->thread == current)) {
- while (!list_empty_careful(&ctx->iopoll_list)) {
+ while (!wq_list_empty(&ctx->iopoll_list)) {
io_iopoll_try_reap_events(ctx);
ret = true;
}
--
2.35.3