public inbox for linux-kernel@vger.kernel.org
 help / color / mirror / Atom feed
* Help with 2.6.10 concurrency issue
@ 2006-02-09 21:08 martin rogers
  0 siblings, 0 replies; 2+ messages in thread
From: martin rogers @ 2006-02-09 21:08 UTC (permalink / raw)
  To: linux-kernel

All,
I need help with a concurrency issue on 2.6.10.
CONFIG_PREEMPT not set, but my code must run on both uni and SMP systems.


I have a function, that runs from a tastlet:

void readlist(unsigned long arg)
{    
  int flags;
  my_type *entry, *n;

  if (list_empty(&mylist.list)) return;
  list_for_each_entry_safe(entry, n, &mylist.list, list)
  {
        spin_lock_irqsave(&mylock, flags); // protect against intr
        list_del(&entry->list);
        spin_unlock_irqrestore(&mylock, flags);
        INIT_LIST_HEAD(&entry->list);
        do_stuff(entry);
  }
}


And the func that puts things on the list:

void writeList(my_type *record)
{
  spin_lock(&mylock);
  list_add_tail(&record->list, &mylist.list);
  spin_unlock(&mylock);
  tasklet_schedule(&mytasklet.tlet);
}


Problem is, the function writeList can be called from a H/W intr,
and a workqueue (and that intr could of course happen while either
the workqueue or the tasklet is running, right?).

If I use spin_lock_irqsave in writeList, it protects against the intr
but not the tasklet.  If I use spin_lock_bh, I don't get protection
from the intr I think; plus, I get :

Badness in local_bh_enable at kernel/softirq.c:142

when the intr runs (what does this mean?).

So, how can I protect my data (my list) from both intrs (calling
writeList) and the tasklet (calling readList) while the workqueue is
inside of writeList?  Combination of spin_lock_irqsave and local_bh_disable
inside writeList ?

Thanks to all,
Martin Rogers
Wind River


^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: Help with 2.6.10 concurrency issue
       [not found] <5Erdv-5iS-39@gated-at.bofh.it>
@ 2006-02-10  0:48 ` Robert Hancock
  0 siblings, 0 replies; 2+ messages in thread
From: Robert Hancock @ 2006-02-10  0:48 UTC (permalink / raw)
  To: linux-kernel

martin rogers wrote:
> Problem is, the function writeList can be called from a H/W intr,
> and a workqueue (and that intr could of course happen while either
> the workqueue or the tasklet is running, right?).
> 
> If I use spin_lock_irqsave in writeList, it protects against the intr
> but not the tasklet.

Yes, it protects against the tasklet as well. Disabling interrupts also 
implicity disables BH execution (tasklets).

   If I use spin_lock_bh, I don't get protection
> from the intr I think; plus, I get :
> 
> Badness in local_bh_enable at kernel/softirq.c:142
> 
> when the intr runs (what does this mean?).

It means you're enabling BHs when interrupts are disabled, which doesn't 
make any sense. You can't use spin_lock_bh/spin_unlock_bh when 
interrupts are disabled.

-- 
Robert Hancock      Saskatoon, SK, Canada
To email, remove "nospam" from hancockr@nospamshaw.ca
Home Page: http://www.roberthancock.com/


^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2006-02-10  0:47 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
     [not found] <5Erdv-5iS-39@gated-at.bofh.it>
2006-02-10  0:48 ` Help with 2.6.10 concurrency issue Robert Hancock
2006-02-09 21:08 martin rogers

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox