@@ -988,9 +988,10 @@ static void ctx_flush_and_put(struct io_ring_ctx *ctx, bool *locked)
static void handle_tw_list(struct llist_node *node,
- struct io_ring_ctx **ctx, bool *locked)
+ struct io_ring_ctx **ctx, bool *locked,
+ struct llist_node *fake)
{
- do {
+ while (node && node != fake) {
struct llist_node *next = node->next;
struct io_kiocb *req = container_of(node, struct io_kiocb,
io_task_work.node);
@@ -1006,7 +1007,10 @@ static void handle_tw_list(struct llist_node *node,
}
req->io_task_work.func(req, locked);
node = next;
- } while (node);
+ count++;
+ }
+
+ return count;
}
/**
@@ -1045,11 +1049,15 @@ void tctx_task_work(struct callback_head *cb)
struct io_ring_ctx *ctx = NULL;
struct io_uring_task *tctx = container_of(cb, struct io_uring_task,
task_work);
- struct llist_node *node = llist_del_all(&tctx->task_list);
-
- if (node) {
- handle_tw_list(node, &ctx, &uring_locked);
- cond_resched();
+ struct llist_node fake = {};
+ struct llist_node *node = io_llist_xchg(&tctx->task_list, &fake);
+
+ handle_tw_list(node, &ctx, &uring_locked, &fake);
+ node = io_llist_cmpxchg(&tctx->task_list, &fake, NULL);
+ while (node != &fake) {
+ node = io_llist_xchg(&tctx->task_list, &fake);
+ handle_tw_list(node, &ctx, &uring_locked, &fake);
+ node = io_llist_cmpxchg(&tctx->task_list, &fake, NULL);
}
ctx_flush_and_put(ctx, &uring_locked);
Batching task work up is an important performance optimisation, as task_work_add is expensive. In order to keep the semantics replace the task_list with a fake node while processing the old list, and then do a cmpxchg at the end to see if there is more work. Signed-off-by: Dylan Yudaken <dylany@fb.com> --- io_uring/io_uring.c | 24 ++++++++++++++++-------- 1 file changed, 16 insertions(+), 8 deletions(-)