io_uring: extend async work merging

commit 6d5d5ac522b20b65167dafe0656b7cad05ec48b3 upstream.

We currently merge async work items if we see a strict sequential hit.
This helps avoid unnecessary workqueue switches when we don't need
them. We can extend this merging to cover cases where it's not a strict
sequential hit, but the IO still fits within the same page. If an
application is doing multiple requests within the same page, we don't
want separate workers waiting on the same page to complete IO. It's much
faster to let the first worker bring in the page, then operate on that
page from the same worker to complete the next request(s).
Reviewed-by: NJeff Moyer <jmoyer@redhat.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>
Signed-off-by: NJoseph Qi <joseph.qi@linux.alibaba.com>
Reviewed-by: NXiaoguang Wang <xiaoguang.wang@linux.alibaba.com>

fs/io_uring.c

@@ -167,7 +167,7 @@ struct async_list {
 	struct list_head	list;
 
 	struct file		*file;
-	off_t			io_end;
+	off_t			io_start;
 	size_t			io_len;
 };
 
@@ -1188,6 +1188,28 @@ static ssize_t io_import_iovec(struct io_ring_ctx *ctx, int rw,
 	return import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter);
 }
 
+static inline bool io_should_merge(struct async_list *al, struct kiocb *kiocb)
+{
+	if (al->file == kiocb->ki_filp) {
+		off_t start, end;
+
+		/*
+		 * Allow merging if we're anywhere in the range of the same
+		 * page. Generally this happens for sub-page reads or writes,
+		 * and it's beneficial to allow the first worker to bring the
+		 * page in and the piggy backed work can then work on the
+		 * cached page.
+		 */
+		start = al->io_start & PAGE_MASK;
+		end = (al->io_start + al->io_len + PAGE_SIZE - 1) & PAGE_MASK;
+		if (kiocb->ki_pos >= start && kiocb->ki_pos <= end)
+			return true;
+	}
+
+	al->file = NULL;
+	return false;
+}
+
 /*
  * Make a note of the last file/offset/direction we punted to async
  * context. We'll use this information to see if we can piggy back a
@@ -1199,9 +1221,8 @@ static void io_async_list_note(int rw, struct io_kiocb *req, size_t len)
 	struct async_list *async_list = &req->ctx->pending_async[rw];
 	struct kiocb *kiocb = &req->rw;
 	struct file *filp = kiocb->ki_filp;
-	off_t io_end = kiocb->ki_pos + len;
 
-	if (filp == async_list->file && kiocb->ki_pos == async_list->io_end) {
+	if (io_should_merge(async_list, kiocb)) {
 		unsigned long max_bytes;
 
 		/* Use 8x RA size as a decent limiter for both reads/writes */
@@ -1214,17 +1235,16 @@ static void io_async_list_note(int rw, struct io_kiocb *req, size_t len)
 			req->flags |= REQ_F_SEQ_PREV;
 			async_list->io_len += len;
 		} else {
-			io_end = 0;
-			async_list->io_len = 0;
+			async_list->file = NULL;
 		}
 	}
 
 	/* New file? Reset state. */
 	if (async_list->file != filp) {
-		async_list->io_len = 0;
+		async_list->io_start = kiocb->ki_pos;
+		async_list->io_len = len;
 		async_list->file = filp;
 	}
-	async_list->io_end = io_end;
 }
 
 static int io_read(struct io_kiocb *req, const struct sqe_submit *s,
@@ -1993,7 +2013,7 @@ static void io_sq_wq_submit_work(struct work_struct *work)
  */
 static bool io_add_to_prev_work(struct async_list *list, struct io_kiocb *req)
 {
-	bool ret = false;
+	bool ret;
 
 	if (!list)
 		return false;