Deadlock in fb and tty

Deadlock in fb and tty

[John Tapsell]

Hi,
  I'm consistently and constantly hitting a deadlock.

console_callback in drivers/tty/vt/vt.c does:  console_lock()  and then calls:
  do_blank_screen, which calls:
    vc->vc_sw->con_blank(..)  which can be a pointer to the function:
      fbcon_blank in video/console/fbcon.c.  This is missing a
WARN_CONSOLE_UNLOCKED.  This calls:
        fbcon_generic_blank which does lock_fb_info(info)

So we have a console_lock() followed by a lock_fb_info(info).

Now if while that is running, we have an ioctl call:

do_fb_ioctl in drivers/video/fbmem.c which does:
               if (!lock_fb_info(info))
                        return -ENODEV;
                console_lock();


So it tries to acquire the same locks in the reverse order.  This
deadlocks consistently for me.

(I'm also curious why I'm hitting this continually, when everyone else
seems to be okay.)

I understand that this is really difficult to fix, but if anyone has
even a suggestion on how to hack it to make it work for me, I'd be
very grateful.

Thank you,

John Tapsell

Re: Deadlock in fb and tty

[Peter Hurley]

On 09/11/2013 05:25 AM, John Tapsell wrote:
> Hi,
>    I'm consistently and constantly hitting a deadlock.
>
> console_callback in drivers/tty/vt/vt.c does:  console_lock()  and then calls:
>    do_blank_screen, which calls:
>      vc->vc_sw->con_blank(..)  which can be a pointer to the function:
>        fbcon_blank in video/console/fbcon.c.  This is missing a
> WARN_CONSOLE_UNLOCKED.  This calls:
>          fbcon_generic_blank which does lock_fb_info(info)
>
> So we have a console_lock() followed by a lock_fb_info(info).
>
> Now if while that is running, we have an ioctl call:
>
> do_fb_ioctl in drivers/video/fbmem.c which does:
>                 if (!lock_fb_info(info))
>                          return -ENODEV;
>                  console_lock();
>
>
> So it tries to acquire the same locks in the reverse order.  This
> deadlocks consistently for me.
>
> (I'm also curious why I'm hitting this continually, when everyone else
> seems to be okay.)
>
> I understand that this is really difficult to fix, but if anyone has
> even a suggestion on how to hack it to make it work for me, I'd be
> very grateful.

As a temporary workaround, you can disable timed console blanking on the
kernel command line with 'consoleblank=0'.

Regards,
Peter Hurley

Re: Deadlock in fb and tty

[johnflux]

The following seems to be better:

>From d93c1e9761ff66365d658da7d8d0d33823aa946f Mon Sep 17 00:00:00 2001
From: John Tapsell <johnflux@gmail.com>
Date: Thu, 12 Sep 2013 09:16:12 +0100
Subject: [PATCH] Fix deadlock between fb_info and console.  Do not lock
 fb_info when calling sending the FB_EVENT_CONBLANK

In fbmem.c, the semantics are that we acquire the lock_fb_info first,
and then console_lock.  However when fbcon.c fbcon_generic_blank() is
called, the console lock could already be held.  Locking fb_info can
thus cause a deadlock.

fbmem.c sends the FB_EVENT_BLANK without locking lock_fb_info first, so
this change introduces similar behaviour.
---
 drivers/video/console/fbcon.c | 4 ----
 1 file changed, 4 deletions(-)

diff --git a/drivers/video/console/fbcon.c b/drivers/video/console/fbcon.c
index 6b4fb5c..8546441 100644
--- a/drivers/video/console/fbcon.c
+++ b/drivers/video/console/fbcon.c
@@ -2333,13 +2333,9 @@ static void fbcon_generic_blank(struct vc_data *vc,
struct fb_info *info,
                vc->vc_video_erase_char = oldc;
        }
 
-
-       if (!lock_fb_info(info))
-               return;
        event.info = info;
        event.data = &blank;
        fb_notifier_call_chain(FB_EVENT_CONBLANK, &event);
-       unlock_fb_info(info);
 }
 
 static int fbcon_blank(struct vc_data *vc, int blank, int mode_switch)
-- 
1.8.1.2


How can I get this reviewed/acked please?



--
View this message in context: http://linux-kernel.2935.n7.nabble.com/Deadlock-in-fb-and-tty-tp717929p718779.html
Sent from the Linux Kernel mailing list archive at Nabble.com.

Re: Deadlock in fb and tty

[Peter Hurley]

On 09/12/2013 09:22 AM, johnflux wrote:
> The following seems to be better:
>
>  From d93c1e9761ff66365d658da7d8d0d33823aa946f Mon Sep 17 00:00:00 2001
> From: John Tapsell<johnflux@gmail.com>
> Date: Thu, 12 Sep 2013 09:16:12 +0100
> Subject: [PATCH] Fix deadlock between fb_info and console.  Do not lock
>   fb_info when calling sending the FB_EVENT_CONBLANK
>
> In fbmem.c, the semantics are that we acquire the lock_fb_info first,
> and then console_lock.  However when fbcon.c fbcon_generic_blank() is
> called, the console lock could already be held.  Locking fb_info can
> thus cause a deadlock.
>
> fbmem.c sends the FB_EVENT_BLANK without locking lock_fb_info first, so
> this change introduces similar behaviour.
> ---
>   drivers/video/console/fbcon.c | 4 ----
>   1 file changed, 4 deletions(-)
>
> diff --git a/drivers/video/console/fbcon.c b/drivers/video/console/fbcon.c
> index 6b4fb5c..8546441 100644
> --- a/drivers/video/console/fbcon.c
> +++ b/drivers/video/console/fbcon.c
> @@ -2333,13 +2333,9 @@ static void fbcon_generic_blank(struct vc_data *vc,
> struct fb_info *info,
>                  vc->vc_video_erase_char = oldc;
>          }
>
> -
> -       if (!lock_fb_info(info))
> -               return;
>          event.info = info;
>          event.data = &blank;
>          fb_notifier_call_chain(FB_EVENT_CONBLANK, &event);
> -       unlock_fb_info(info);
>   }
>
>   static int fbcon_blank(struct vc_data *vc, int blank, int mode_switch)
> -- 1.8.1.2

> How can I get this reviewed/acked please?

The get_maintainer.pl script can help to discover to whom to
address patches. For example,

peter@thor:~/src/kernels/next$ ./scripts/get_maintainer.pl -f drivers/video/console/fbcon.c
Jean-Christophe Plagniol-Villard <plagnioj@jcrosoft.com> (maintainer:FRAMEBUFFER LAYER)
Tomi Valkeinen <tomi.valkeinen@ti.com> (maintainer:FRAMEBUFFER LAYER)
Dave Airlie <airlied@redhat.com> (commit_signer:5/11=45%)
Greg Kroah-Hartman <gregkh@linuxfoundation.org> (commit_signer:4/11=36%)
Wang YanQing <udknight@gmail.com> (commit_signer:4/11=36%)
Andrew Morton <akpm@linux-foundation.org> (commit_signer:3/11=27%)
Kamal Mostafa <kamal@whence.com> (commit_signer:1/11=9%)
linux-fbdev@vger.kernel.org (open list:FRAMEBUFFER LAYER)
linux-kernel@vger.kernel.org (open list)

In this case, you'll want to send this patch addressed to the two maintainers
and cc linux-fbdev and linux-kernel. You can cc me if you want.

This is covered in Documentation/SubmittingPatches, which also covers the
required subject line format <hint: too long, needs subsystem identifier>.

Also, if you have the lockdep report of the deadlock, it's customary to
include it in the commit message.

Regards,
Peter Hurley

PS - Don't put stuff after the '--' tag at the end of the patch. Many
mailers treat that as don't care. I missed your question first time around
because Thunderbird grays that text :(