[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-Id: <20200617104058.14902-1-john.ogness@linutronix.de>
Date: Wed, 17 Jun 2020 12:46:57 +0206
From: John Ogness <john.ogness@...utronix.de>
To: Alexander Viro <viro@...iv.linux.org.uk>
Cc: Sebastian Andrzej Siewior <bigeasy@...utronix.de>,
Peter Zijlstra <peterz@...radead.org>,
Thomas Gleixner <tglx@...utronix.de>,
linux-fsdevel@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: [RFC PATCH 0/1] fs: remove retry loop
Hello,
This patch removes the last retry loop in VFS. It is partially
reverting
commit d3ef3d7351cc ("fs: mnt_want_write speedup")
by re-introducing per-cpu spinlocks for each mount. The patch
includes benchmark results in the diffstat section to show that the
previous optimization work is not undone.
I would have liked to use a percpu_rw_semaphore per mount instead
of the many spinlocks. However, percpu_rw_semaphore can sleep, which
is a problem for sb_prepare_remount_readonly() since it needs to
take the spinlock in @mount_lock in order to iterate @sb->s_mounts.
Perhaps using a mutex to sychronize @sb->s_mounts is an option. I am
not sure. That is why this is an RFC.
I am suggesting this partial revert because it removes the retry
loop and does not show any obvious negative benchmark effects.
John Ogness (1):
fs/namespace.c: use spinlock instead of busy loop
fs/mount.h | 7 +++
fs/namespace.c | 118 +++++++++++++++++++++++++++++++++----------------
2 files changed, 86 insertions(+), 39 deletions(-)
--
2.20.1
Powered by blists - more mailing lists