[patch] Fix deadlock in pcmcia as found by lockdep

[Arjan van de Ven <arjan@linux.intel.com>]

On Thu, 2006-06-29 at 23:18 -0700, Miles Lane wrote:
> To trigger this, I booted with a U.S. Robotics USR2210 Wifi card
> plugged into my cardbus slot.  I then ran "pccardctl eject" and then
> removed and then reinserted the card.  After looking at the latest
> PCMCIA info, it seems that I may need to add some kernel boot options
> to work around a BIOS or other problem that causes trouble when
> removing a card.
> 
> PM: Removing info for pci:0000:02:00.0
> PCMCIA: socket c1ebc9e0: *** DANGER *** unable to remove socket power


ok this looks like a real bug:

void pcmcia_parse_events(struct pcmcia_socket *s, u_int events)
{
        cs_dbg(s, 4, "parse_events: events %08x\n", events);
        if (s->thread) {
                spin_lock(&s->thread_lock);
                s->thread_events |= events;
                spin_unlock(&s->thread_lock);

                wake_up(&s->thread_wait);
        }
} /* pcmcia_parse_events */


that function gets called from both user context and irq context!

user context:
[<c1181270>] pcmcia_parse_events+0x3e/0x6b
[<c1181945>] pcmcia_register_socket+0x29b/0x2fc
[<c118a8d1>] yenta_probe+0x51b/0x55c
[<c110d537>] pci_device_probe+0x39/0x5b

eg in pcmcia_register_socket:

        ret = kernel_thread(pccardd, socket, CLONE_KERNEL);
        if (ret < 0)
                goto err;

        wait_for_completion(&socket->thread_done);
        if(!socket->thread) {
                printk(KERN_WARNING "PCMCIA: warning: socket thread for
socket %p did not start\n", socket);
                return -EIO;
        }
        pcmcia_parse_events(socket, SS_DETECT);

clearly sleeping/user context


interrupt context:
yenta_interrupt calls pcmcia_parse_events like this:
....
        if (events)
                pcmcia_parse_events(&socket->socket, events);

        return IRQ_HANDLED;
}

and that's the irq handler.

Dominik: this really wants to have _irqsave versions of the spinlock
like this:



the PCMCIA layer calls pcmcia_parse_events both from user context and
IRQ context; the lock thus needs to be irqsave to avoid deadlocks

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>

---
 drivers/pcmcia/cs.c |    5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

Index: linux-2.6.17-mm4/drivers/pcmcia/cs.c
===================================================================
--- linux-2.6.17-mm4.orig/drivers/pcmcia/cs.c
+++ linux-2.6.17-mm4/drivers/pcmcia/cs.c
@@ -697,11 +697,12 @@ static int pccardd(void *__skt)
  */
 void pcmcia_parse_events(struct pcmcia_socket *s, u_int events)
 {
+	unsigned long flags;
 	cs_dbg(s, 4, "parse_events: events %08x\n", events);
 	if (s->thread) {
-		spin_lock(&s->thread_lock);
+		spin_lock_irqsave(&s->thread_lock, flags);
 		s->thread_events |= events;
-		spin_unlock(&s->thread_lock);
+		spin_unlock_irqrestore(&s->thread_lock, flags);
 
 		wake_up(&s->thread_wait);
 	}

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/