On Mon, Jun 15, 2009 at 03:41:50PM -0700, Andrew Morton wrote:
> On Mon, 15 Jun 2009 17:17:35 -0500
> "Mike Miller (OS Dev)" <mikem@xxxxxxxxxxxxxxxxxxxxxxx> wrote:
>
> > On Mon, Jun 15, 2009 at 02:25:50PM -0700, Andrew Morton wrote:
> > > On Mon, 15 Jun 2009 15:39:55 -0500
> > > "Mike Miller (OS Dev)" <mikem@xxxxxxxxxxxxxxxxxxxxxxx> wrote:
> > >
> > > > I'm getting reports that the driver hangs in the scan thread during rmmod. I
> > > > call kthread_stop in cciss_remove_one. Do I also neesd to call complete()?
> > >
> > > Yep.
> > >
> > > static int scan_thread(void *data)
> > > {
> > > ctlr_info_t *h = data;
> > > int rc;
> > > DECLARE_COMPLETION_ONSTACK(wait);
> > > h->rescan_wait = &wait;
> > >
> > > for (;;) {
> > > rc = wait_for_completion_interruptible(&wait);
> > > if (kthread_should_stop())
> > > break;
> > > if (!rc)
> > > rebuild_lun_table(h, 0);
> > > }
> > > return 0;
> > > }
> > >
> > > Two things will cause that wait_for_completion_interruptible() to
> > > return: a complete() and a signal_pending(). But this is a kernel
> > > thread, and kernel threads start out with all signals blocked.
> > >
> > > > During my testing everything seemed to working OK.
> > >
> > > That's odd.
> >
> > Yeah, I guess so. Here's a patch to fix that problem. I add a call to
> > complete to cciss_remove_one to stop the thread for rmmod. Please consider
> > this for inclusion. Or should I submit separately from this mail?
>
> Well it's a bit crappy - someone need to cook up a suitable title and
> write a suitable changelog. I'd prefer that it be you ;)
>
> <writes a title and changelog. Sigh>
>
> > Signed-off-by: Mike Miller <mike.miller@xxxxxx>
> >
> > diff --git a/drivers/block/cciss.c b/drivers/block/cciss.c
> > index 4d4d5e0..e51a0b2 100644
> > --- a/drivers/block/cciss.c
> > +++ b/drivers/block/cciss.c
> > @@ -3935,6 +3935,7 @@ static void __devexit cciss_remove_one(struct pci_dev *pdev)
> > return;
> > }
> >
> > + complete(hba[i]->rescan_wait);
> > kthread_stop(hba[i]->cciss_scan_thread);
> >
> > remove_proc_entry(hba[i]->devname, proc_cciss);
>
> Has this been confirmed to fix things?
Apparently, not. I'm still having issues. I'm sure this worked before my
patch was finally accepted. That's why I made the thread interruptible. Now
I see an article on LWN saying that kthread_stop doesn't use signals. It
says call kthread_should_stop periodically, which I do. How else to stop a
thread???
http://lwn.net/Articles/118935/
>
> This is needed in 2.6.30.x, yes?
Yes, this is needed in 2.6.30.x. I'm working it. Will submit a proper patch
with changelog when complete.
Sorry for the inconvenience.
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
