Re: rmmod e1000 hangs (Was Re: 2.6.22-rc2-mm1)

["Kok, Auke" <auke-jan.h.kok@intel.com>]

Herbert Xu wrote:
> On Thu, May 24, 2007 at 08:47:13PM +1000, Herbert Xu wrote:
> > On Thu, May 24, 2007 at 11:36:22AM +0100, Jeremy Fitzhardinge wrote:
> > > I got a hang while rmmodding e1000.  sysrq-t shows:
> > >
> > > rmmod         D 003FFAFC  6616 15923  15911 (NOTLB)
> > >        e9341e44 00000092 82318c15 003ffafc e9341e2c 00000000 e9341e14 823187a1 
> > >        003ffafc 00000000 c0123862 d3dbab80 d3dbad1c c2c08a40 77a67d01 000001ca 
> > >        00000292 e9341e24 c03799cd e9341e54 c0540840 e9341e44 00223389 000000ff 
> > > Call Trace:
> > >  [<c03777b1>] schedule_timeout+0x70/0x8e
> > >  [<c03777e4>] schedule_timeout_uninterruptible+0x15/0x17
> > >  [<c0133d04>] msleep+0x10/0x16
> > >  [<c030d5e0>] dev_close+0x39/0x6b
> > Looks like we're spinning on __LINK_STATE_RX_SCHED.  This means that
> > someone called netif_poll_disable() without re-enabling it again.
> > Perhaps e1000_io_error_detected? Auke?

Should not be, e1000_io_error_detected will call e1000_down which does the 
netif_poll_disable, but e1000_io_resume nicely calls e1000_up again which does 
the netif_poll_enable again, unless io_resume somehow failed

> I think the dual meaning of __LINK_STATE_RX_SCHED is seriously broken.
> In dev_close we are waiting for any outstanding poll to terminate but
> the same bit can either mean an outstanding poll or that poll has
> been disabled.

that seems more likely

> It's a surprise that it has taken so many years for someone to report
> a bug on it.  I'll try to get this fixed up, probably by adding a bit.

I get the feeling that a recent change exposed us to this, our lab has been 
seeing similar OOPS's yesterday out of nothing.


Auke
-
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/