Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <299889df-db40-e0e2-6bc6-d9eb784ebe89@kernel.dk>
Date:   Wed, 20 Jul 2022 12:13:27 -0600
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux aarch64; rv:91.0) Gecko/20100101
 Thunderbird/91.10.0
Subject: Re: [LKP] Re: [io_uring] 584b0180f0:
 phoronix-test-suite.fio.SequentialWrite.IO_uring.Yes.Yes.1MB.DefaultTestDirectory.mb_s
 -10.2% regression
Content-Language: en-US
From:   Jens Axboe <axboe@kernel.dk>
To:     Yin Fengwei <fengwei.yin@intel.com>,
        kernel test robot <oliver.sang@intel.com>
Cc:     LKML <linux-kernel@vger.kernel.org>, io-uring@vger.kernel.org,
        lkp@lists.01.org, lkp@intel.com
References: <20220527092432.GE11731@xsang-OptiPlex-9020>
 <2085bfef-a91c-8adb-402b-242e8c5d5c55@kernel.dk>
 <0d60aa42-a519-12ad-3c69-72ed12398865@intel.com>
 <26d913ea-7aa0-467d-4caf-a93f8ca5b3ff@kernel.dk>
 <9df150bb-f4fd-7857-aea8-b2c7a06a8791@intel.com>
 <7146c853-0ff8-3c92-c872-ce6615baab40@kernel.dk>
 <81af5cdf-1a13-db2c-7b7b-cfd86f1271e6@intel.com>
 <74d1f308-de03-fd5e-b7f0-0e17980f988e@kernel.dk>
 <2ec953da-78fd-df01-44cf-6f33a5e40864@intel.com>
 <f5d20f6c-5363-231b-b208-b577a59b4ae9@kernel.dk>
In-Reply-To: <f5d20f6c-5363-231b-b208-b577a59b4ae9@kernel.dk>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 7/20/22 11:24 AM, Jens Axboe wrote:
> On 7/19/22 2:58 AM, Yin Fengwei wrote:
>> Hi Jens,
>>
>> On 7/19/2022 10:29 AM, Jens Axboe wrote:
>>> I'll poke at this tomorrow.
>>
>> Just FYI. Another finding (test is based on commit 584b0180f0):
>> If the code block is put to different function, the fio performance result is
>> different:
> 
> I think this turned out to be a little bit of a goose chase. What's
> happening here is that later kernels defer the file assignment, which
> means it isn't set if a request is queued with IOSQE_ASYNC. That in
> turn, for writes, means that we don't hash it on io-wq insertion, and
> then it doesn't get serialized with other writes to that file.
> 
> I'll come up with a patch for this that you can test.

Can you try this? It's against 5.19-rc7.


diff --git a/fs/io_uring.c b/fs/io_uring.c
index a01ea49f3017..34758e95990a 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -2015,6 +2015,64 @@ static inline void io_arm_ltimeout(struct io_kiocb *req)
 		__io_arm_ltimeout(req);
 }
 
+static bool io_bdev_nowait(struct block_device *bdev)
+{
+	return !bdev || blk_queue_nowait(bdev_get_queue(bdev));
+}
+
+/*
+ * If we tracked the file through the SCM inflight mechanism, we could support
+ * any file. For now, just ensure that anything potentially problematic is done
+ * inline.
+ */
+static bool __io_file_supports_nowait(struct file *file, umode_t mode)
+{
+	if (S_ISBLK(mode)) {
+		if (IS_ENABLED(CONFIG_BLOCK) &&
+		    io_bdev_nowait(I_BDEV(file->f_mapping->host)))
+			return true;
+		return false;
+	}
+	if (S_ISSOCK(mode))
+		return true;
+	if (S_ISREG(mode)) {
+		if (IS_ENABLED(CONFIG_BLOCK) &&
+		    io_bdev_nowait(file->f_inode->i_sb->s_bdev) &&
+		    file->f_op != &io_uring_fops)
+			return true;
+		return false;
+	}
+
+	/* any ->read/write should understand O_NONBLOCK */
+	if (file->f_flags & O_NONBLOCK)
+		return true;
+	return file->f_mode & FMODE_NOWAIT;
+}
+
+static inline bool io_file_supports_nowait(struct io_kiocb *req)
+{
+	return req->flags & REQ_F_SUPPORT_NOWAIT;
+}
+
+/*
+ * If we tracked the file through the SCM inflight mechanism, we could support
+ * any file. For now, just ensure that anything potentially problematic is done
+ * inline.
+ */
+static unsigned int io_file_get_flags(struct file *file)
+{
+	umode_t mode = file_inode(file)->i_mode;
+	unsigned int res = 0;
+
+	if (S_ISREG(mode))
+		res |= FFS_ISREG;
+	if (__io_file_supports_nowait(file, mode))
+		res |= FFS_NOWAIT;
+	if (io_file_need_scm(file))
+		res |= FFS_SCM;
+	return res;
+}
+
 static void io_prep_async_work(struct io_kiocb *req)
 {
 	const struct io_op_def *def = &io_op_defs[req->opcode];
@@ -2031,6 +2089,9 @@ static void io_prep_async_work(struct io_kiocb *req)
 	if (req->flags & REQ_F_FORCE_ASYNC)
 		req->work.flags |= IO_WQ_WORK_CONCURRENT;
 
+	if (req->file && !io_req_ffs_set(req))
+		req->flags |= io_file_get_flags(req->file) << REQ_F_SUPPORT_NOWAIT_BIT;
+
 	if (req->flags & REQ_F_ISREG) {
 		if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL))
 			io_wq_hash_work(&req->work, file_inode(req->file));
@@ -3556,64 +3617,6 @@ static void io_iopoll_req_issued(struct io_kiocb *req, unsigned int issue_flags)
 	}
 }
 
-static bool io_bdev_nowait(struct block_device *bdev)
-{
-	return !bdev || blk_queue_nowait(bdev_get_queue(bdev));
-}
-
-/*
- * If we tracked the file through the SCM inflight mechanism, we could support
- * any file. For now, just ensure that anything potentially problematic is done
- * inline.
- */
-static bool __io_file_supports_nowait(struct file *file, umode_t mode)
-{
-	if (S_ISBLK(mode)) {
-		if (IS_ENABLED(CONFIG_BLOCK) &&
-		    io_bdev_nowait(I_BDEV(file->f_mapping->host)))
-			return true;
-		return false;
-	}
-	if (S_ISSOCK(mode))
-		return true;
-	if (S_ISREG(mode)) {
-		if (IS_ENABLED(CONFIG_BLOCK) &&
-		    io_bdev_nowait(file->f_inode->i_sb->s_bdev) &&
-		    file->f_op != &io_uring_fops)
-			return true;
-		return false;
-	}
-
-	/* any ->read/write should understand O_NONBLOCK */
-	if (file->f_flags & O_NONBLOCK)
-		return true;
-	return file->f_mode & FMODE_NOWAIT;
-}
-
-/*
- * If we tracked the file through the SCM inflight mechanism, we could support
- * any file. For now, just ensure that anything potentially problematic is done
- * inline.
- */
-static unsigned int io_file_get_flags(struct file *file)
-{
-	umode_t mode = file_inode(file)->i_mode;
-	unsigned int res = 0;
-
-	if (S_ISREG(mode))
-		res |= FFS_ISREG;
-	if (__io_file_supports_nowait(file, mode))
-		res |= FFS_NOWAIT;
-	if (io_file_need_scm(file))
-		res |= FFS_SCM;
-	return res;
-}
-
-static inline bool io_file_supports_nowait(struct io_kiocb *req)
-{
-	return req->flags & REQ_F_SUPPORT_NOWAIT;
-}
-
 static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 {
 	struct kiocb *kiocb = &req->rw.kiocb;

-- 
Jens Axboe