Re: ldisc problems with 2.6.32-2.6.37-rc2 (at least)

[Jiri Slaby <jslaby@suse.cz>]

On 11/24/2010 11:26 PM, Sergey Lapin wrote:
> On Wed, Nov 24, 2010 at 11:02:46PM +0100, Jiri Slaby wrote:
>> On 11/24/2010 10:53 PM, Sergey Lapin wrote:
>>> Initially I run it, everything is ok.
>>> Then I run it and get error and backtrace.
>>> And if I run it again, everything is ok again, repeatedly. Probably some resources are properly
>>> set up in error condition, and not in normal condition.
>>>
>>> I use ARM machine for testing and x86, bug is architecture-independent.
>>>
>>> addr2lines:
>>> c01bc110 drivers/tty/tty_ldisc.c:499
>>
>> Got it. Does this fixes the warning?
>> --- a/drivers/tty/tty_ldisc.c
>> +++ b/drivers/tty/tty_ldisc.c
>> @@ -454,6 +454,8 @@ static int tty_ldisc_open(struct tty_struct *tty,
>> struct tty_ldisc *ld)
>>                  /* BTM here locks versus a hangup event */
>>                 WARN_ON(!tty_locked());
>>                 ret = ld->ops->open(tty);
>> +               if (ret)
>> +                       clear_bit(TTY_LDISC_OPEN, &tty->flags);
>>                 return ret;
>>         }
>>         return 0;
>>
>> thanks,
>> -- 
>> js
>> suse labs
> This one fixes this particular backtrace, but now we have another one, which happens less often
> only every 3 times of executing problem and never on ENOMEM case:
> 
> ------------[ cut here ]------------
> WARNING: at drivers/tty/tty_ldisc.c:475 tty_ldisc_close+0x24/0x64()

So this one appeared only after the patch was applied. That means, close
is called even if open fails. Crap. Will look into that later.

> ENOMEM case is one which still wonders me, especially why it happens every second run.

For some reason X25 ldisc fails. Could you verify that it doesn't happen
with some other ldisc?

thanks.
-- 
js
suse labs