Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.33 as permitted sender) client-ip=23.128.96.33;
Message-ID: <tencent_4967A8AA14EACCA0BB4999F40066023FDC07@qq.com>
From:   wenyang.linux@foxmail.com
To:     Alexander Viro <viro@zeniv.linux.org.uk>,
        Jens Axboe <axboe@kernel.dk>,
        Christian Brauner <brauner@kernel.org>
Cc:     Wen Yang <wenyang.linux@foxmail.com>,
        Christoph Hellwig <hch@lst.de>, Dylan Yudaken <dylany@fb.com>,
        David Woodhouse <dwmw@amazon.co.uk>,
        Matthew Wilcox <willy@infradead.org>,
        Eric Biggers <ebiggers@google.com>,
        linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: [PATCH V2] eventfd: move 'eventfd-count' printing out of spinlock
Date:   Mon,  2 Oct 2023 23:13:02 +0800
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

From: Wen Yang <wenyang.linux@foxmail.com>

When printing eventfd->count, interrupts will be disabled and a spinlock
will be obtained, competing with eventfd_write().

In a single CPU container, one process frequently calls eventfd_write()
and another process frequently prints eventfd->count, as follows:

Tasks: 372 total,   3 running, 369 sleeping,   0 stopped,   0 zombie
%Cpu(s): 1.8 us, 4.5 sy, 0.0 ni, 92.5 id, 0.0 wa, 0.0 hi, 1.2 si, 0.0 st
MiB Mem :  13595.9 total,   4627.0 free,   2536.6 used,   6432.2 buff/cache
MiB Swap:   2048.0 total,   2048.0 free,      0.0 used.  10727.2 avail Mem

    PID USER  PR  NI    VIRT  RES  SHR S  %CPU  %MEM   TIME+ COMMAND    P
1893151 test  20   0    2364  896  896 R  49.2   0.0  6:03.49 a.out     3
1893080 test  20   0  244748 5376 4224 R  48.8   0.0  6:52.34 test_zmq  3

 # ./perf stat -e  'probe:eventfd_write' -I 10000
 #           time             counts unit events
    10.004950650            396,394      probe:eventfd_write
    20.015871354            396,403      probe:eventfd_write
    30.026916173            397,624      probe:eventfd_write
    40.037815761            396,363      probe:eventfd_write
    50.048964887            396,168      probe:eventfd_write
    60.059711824            396,656      probe:eventfd_write
    70.065021841            397,286      probe:eventfd_write
    80.072792225            397,476      probe:eventfd_write
    90.083883162            394,014      probe:eventfd_write

By moving the "eventfd-count" print out of the spinlock and merging
multiple seq_printf() into one, it could improve a bit.
After applying this patch, the competition has been slightly alleviated,
and eventfd_writ() has more opportunities to be executed, as follows:

 # ./perf stat -e  'probe:eventfd_write' -I 10000
 #           time             counts unit events
    10.010058911            397,550      probe:eventfd_write
    20.021065452            397,771      probe:eventfd_write
    30.032054749            397,509      probe:eventfd_write
    40.034853809            397,605      probe:eventfd_write
    50.045754449            396,926      probe:eventfd_write
    60.056373938            396,867      probe:eventfd_write
    70.067280837            397,542      probe:eventfd_write
    80.078232498            397,011      probe:eventfd_write
    90.089215771            397,311      probe:eventfd_write

Signed-off-by: Wen Yang <wenyang.linux@foxmail.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Christian Brauner <brauner@kernel.org>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dylan Yudaken <dylany@fb.com>
Cc: David Woodhouse <dwmw@amazon.co.uk>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Eric Biggers <ebiggers@google.com>
Cc: linux-fsdevel@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
---
 fs/eventfd.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)

diff --git a/fs/eventfd.c b/fs/eventfd.c
index 8aa36cd37351..3b893f21923d 100644
--- a/fs/eventfd.c
+++ b/fs/eventfd.c
@@ -295,13 +295,18 @@ static ssize_t eventfd_write(struct file *file, const char __user *buf, size_t c
 static void eventfd_show_fdinfo(struct seq_file *m, struct file *f)
 {
 	struct eventfd_ctx *ctx = f->private_data;
+	unsigned long long cnt;
 
 	spin_lock_irq(&ctx->wqh.lock);
-	seq_printf(m, "eventfd-count: %16llx\n",
-		   (unsigned long long)ctx->count);
+	cnt = ctx->count;
 	spin_unlock_irq(&ctx->wqh.lock);
-	seq_printf(m, "eventfd-id: %d\n", ctx->id);
-	seq_printf(m, "eventfd-semaphore: %d\n",
+
+	seq_printf(m,
+		   "eventfd-count: %16llx\n"
+		   "eventfd-id: %d\n"
+		   "eventfd-semaphore: %d\n",
+		   cnt,
+		   ctx->id,
 		   !!(ctx->flags & EFD_SEMAPHORE));
 }
 #endif
-- 
2.25.1