[PATCH 6/6] io_uring: batch completion in prior_task_list

[Hao Xu <haoxu@linux.alibaba.com>]

In previous patches, we have already gathered some tw with
io_req_task_complete() as callback in prior_task_list, let's complete
them in batch while we cannot grab uring lock. In this way, we batch
the req_complete_post path.

Signed-off-by: Hao Xu <haoxu@linux.alibaba.com>
---
 fs/io_uring.c | 58 ++++++++++++++++++++++++++++++++++++++++++++-------
 1 file changed, 51 insertions(+), 7 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index e9c67f19d585..b474340c5ea5 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -2223,6 +2223,43 @@ static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req)
 	return io_put_kbuf(req, req->kbuf);
 }
 
+static inline void ctx_commit_and_unlock(struct io_ring_ctx *ctx)
+{
+	io_commit_cqring(ctx);
+	spin_unlock(&ctx->completion_lock);
+	io_cqring_ev_posted(ctx);
+}
+
+static void handle_prior_tw_list(struct io_wq_work_node *node, struct io_ring_ctx **ctx,
+				 bool *locked)
+{
+	do {
+		struct io_wq_work_node *next = node->next;
+		struct io_kiocb *req = container_of(node, struct io_kiocb,
+						    io_task_work.node);
+
+		if (req->ctx != *ctx) {
+			if (unlikely(!*locked) && *ctx)
+				ctx_commit_and_unlock(*ctx);
+			ctx_flush_and_put(*ctx, locked);
+			*ctx = req->ctx;
+			/* if not contended, grab and improve batching */
+			*locked = mutex_trylock(&(*ctx)->uring_lock);
+			percpu_ref_get(&(*ctx)->refs);
+			if (unlikely(*locked))
+				spin_lock(&(*ctx)->completion_lock);
+		}
+		if (likely(*locked))
+			req->io_task_work.func(req, locked);
+		else
+			__io_req_complete_post(req, req->result, io_put_rw_kbuf(req));
+		node = next;
+	} while (node);
+
+	if (unlikely(!*locked) && *ctx)
+		ctx_commit_and_unlock(*ctx);
+}
+
 static void handle_tw_list(struct io_wq_work_node *node, struct io_ring_ctx **ctx, bool *locked)
 {
 	do {
@@ -2250,21 +2287,28 @@ static void tctx_task_work(struct callback_head *cb)
 						  task_work);
 
 	while (1) {
-		struct io_wq_work_node *node;
+		struct io_wq_work_node *node1, *node2;
 
-		if (!tctx->prior_task_list.first &&
-		    !tctx->task_list.first && locked)
+		if (!tctx->task_list.first &&
+		    !tctx->prior_task_list.first && locked)
 			io_submit_flush_completions(ctx);
 
 		spin_lock_irq(&tctx->task_lock);
-		node= wq_list_merge(&tctx->prior_task_list, &tctx->task_list);
-		if (!node)
+		node1 = tctx->prior_task_list.first;
+		node2 = tctx->task_list.first;
+		INIT_WQ_LIST(&tctx->task_list);
+		INIT_WQ_LIST(&tctx->prior_task_list);
+		if (!node2 && !node1)
 			tctx->task_running = false;
 		spin_unlock_irq(&tctx->task_lock);
-		if (!node)
+		if (!node2 && !node1)
 			break;
 
-		handle_tw_list(node, &ctx, &locked);
+		if (node1)
+			handle_prior_tw_list(node1, &ctx, &locked);
+
+		if (node2)
+			handle_tw_list(node2, &ctx, &locked);
 		cond_resched();
 	}
 
-- 
2.24.4