Re: Fw: 2.5.61 oops running SDET

["Martin J. Bligh" <mbligh@aracnet.com>]

>> > diff -urpN -X /home/fletch/.diff.exclude virgin/fs/proc/array.c sdet3/fs/proc/array.c
>> > --- virgin/fs/proc/array.c	Sat Feb 15 16:11:45 2003
>> > +++ sdet3/fs/proc/array.c	Sun Feb 16 11:44:24 2003
>> > @@ -252,8 +252,11 @@ int proc_pid_status(struct task_struct *
>> > 		buffer = task_mem(mm, buffer);
>> > 		mmput(mm);
>> > 	}
>> > -	buffer = task_sig(task, buffer);
>> > +	task_lock(task);
>> > +	if (task->sighand)
>> > +		buffer = task_sig(task, buffer);
>> > 	buffer = task_cap(task, buffer);
>> > +	task_unlock(task);
>> > # if defined(CONFIG_ARCH_S390)
>> > 	buffer = task_show_regs(task, buffer);
>> > # endif
>> >  
>> > 
>> I think it's needed for 2.4, too.
> 
> It's not wrong, but it shouldn't help. Simply because "task_lock()"  
> isn't even relevant to "task->sighand" as it stands now. It will be 
> cleared without holding the task lock, as far as I can see.
> 
> So I suspect it fixes things for Martin only because it changes timings 
> enough not to hit the race.

Ah, fair enough ... it's probably the if, rather than the task_lock.

So what does protect sighand? tasklist_lock? It doesn't seem to
as people do things like:

spin_unlock_irq(&current->sighand->siglock);

all the time ... so is it just protected by good faith and the direction
of the wind?

M.