[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CA+55aFysA72FVTF6k=HnqYjCTqXQO+jcMLeR6MpM=eJm-H+EBQ@mail.gmail.com>
Date: Sat, 2 Jun 2012 12:59:03 -0700
From: Linus Torvalds <torvalds@...ux-foundation.org>
To: Eric Dumazet <eric.dumazet@...il.com>
Cc: Alan Cox <alan@...rguk.ukuu.org.uk>,
Alan Cox <alan@...ux.intel.com>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
Jens Axboe <jaxboe@...ionio.com>
Subject: Re: [PATCH] tty: add lockdep annotations
On Sat, Jun 2, 2012 at 11:38 AM, Linus Torvalds
<torvalds@...ux-foundation.org> wrote:
>
> I'm pretty sure this is the bug. And there's no way we can make
> 'tty_mutex' protect every tty_kref_put(). So I think we have two
> options:
>
> - revert all the tty locking changes
>
> - make a new global lock that protects just driver->ops->lookup(),
> driver->ttys[idx], and driver->ops->remove()
>
> Hmm?
Ok, here's the second option. THIS PATCH IS TOTALLY UNTESTED!
Basic concepts:
- the tty driver lookup is protected with "tty_lookup_mutex",
similarly to how the "current task mutex" is protected by the signal
lock.
- At lookup, we now always increment the kref (which in the case of
tty_open_current_tty() means that we do *not* drop the kref), and in
the case of a tty driver lookup, we use "atomic_inc_not_zero()" to
make sure that we get a tty that is not on its way out.
I *think* this makes sense. But it has has had absolutely zero
testing. It compiles in my config, and the locking is at least
sensible and well-localized, but maybe I'm missing something.
Comments?
Linus
Download attachment "tty-lookup-locking.patch" of type "application/octet-stream" (4054 bytes)
Powered by blists - more mailing lists