[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <alpine.DEB.2.00.1002091858180.31159@chino.kir.corp.google.com>
Date: Tue, 9 Feb 2010 19:05:48 -0800 (PST)
From: David Rientjes <rientjes@...gle.com>
To: Américo Wang <xiyou.wangcong@...il.com>
cc: Neil Brown <neilb@...e.de>,
"Eric W. Biederman" <ebiederm@...ssion.com>,
Tejun Heo <tj@...nel.org>, Greg Kroah-Hartman <gregkh@...e.de>,
Hugh Dickins <hugh.dickins@...cali.co.uk>,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH] sysfs: differentiate between locking links and
non-links
On Wed, 10 Feb 2010, Américo Wang wrote:
> > What happens for hard links such as writing to
> > /sys/devices/block/xxx/queue/scheduler to change an I/O scheduler which
> > requires sd->dep_map and sd->parent->dep_map in sysfs_get_active_two() to
> > pin both? The call to kobject_del() invokes the destruction that also
> > requires sd->dep_map in sysfs_deactivate() because of the s_active lockdep
> > annotation.
> >
>
> This is not related with Neil's case at all.
>
> The I/O scheduler switch case should be a bogus, I am working on
> it. We have more similar cases of cpu hotplug. Trust me, I am working
> on a fix to all of them, this is not as easy as you may think about.
>
You should be able to reuse Neil's sysfs_dirent_init_lockdep(sd, type) to
seperate the lock classes for the sd getting pinned in
sysfs_get_active_two() from sysfs_deactivate(), although using subclasses
would probably be optimal since there is a clear parent -> child relationship.
Powered by blists - more mailing lists