On Monday 02 February 2009, Ian Campbell wrote:
> On Sun, 2009-02-01 at 17:21 +0100, Bartlomiej Zolnierkiewicz wrote:
> > On Friday 30 January 2009, Ian Campbell wrote:
> > > On Tue, 2009-01-27 at 17:41 +0100, Bartlomiej Zolnierkiewicz wrote:
> > > > Hi,
> > > >
> > > > On Friday 23 January 2009, Ian Campbell wrote:
> > > > > On reboot the loop in device_shutdown gets confused by these devices
> > > > > and goes into an infinite loop. I think this is because they don't
> > > > > have dev->bus or dev->driver which is because hwif_register_devices
> > > > > doesn't get called if !hwif->present.
> > > > >
> > > > > Signed-off-by: Ian Campbell <ian.campbell@xxxxxxxxxx>
> > > > > Cc: Bartlomiej Zolnierkiewicz <bzolnier@xxxxxxxxx>
> > > > > ---
> > > > > drivers/ide/ide-probe.c | 10 ++++++----
> > > > > 1 files changed, 6 insertions(+), 4 deletions(-)
> > > > >
> > > > > diff --git a/drivers/ide/ide-probe.c b/drivers/ide/ide-probe.c
> > > > > index 312127e..120f65b 100644
> > > > > --- a/drivers/ide/ide-probe.c
> > > > > +++ b/drivers/ide/ide-probe.c
> > > > > @@ -1520,6 +1520,8 @@ int ide_host_register(struct ide_host *host, const struct ide_port_info *d,
> > > > > if (hwif_init(hwif) == 0) {
> > > > > printk(KERN_INFO "%s: failed to initialize IDE "
> > > > > "interface\n", hwif->name);
> > > > > + device_unregister(&hwif->gendev);
> > > > > + device_destroy(ide_port_class, MKDEV(0, 0));
> > > > > hwif->present = 0;
> > > >
> > > > I've been wondering about this a bit and I think that we should just nuke
> > > > port from struct ide_host in such situation so I added ide_disable_port()
> > > > helper for this (+ fixed ide_register_port() handling while at it).
> > > >
> > > > Could you please re-base your fix on top of attached patch?
> > >
> > > Sure.
> > >
> > > Subject: IDE: Disable and unregister devices if initialization fails.
> > >
> > > On reboot the loop in device_shutdown gets confused by these partially
> > > initialized devices and goes into an infinite loop. Therefore disable
> > > and unregister these devices.
> > >
> > > Signed-off-by: Ian Campbell <ian.campbell@xxxxxxxxxx>
> > >
> > > ---
> > > drivers/ide/ide-probe.c | 10 ++++++----
> > > 1 file changed, 6 insertions(+), 4 deletions(-)
> > >
> > > diff -r 12d1a7848ddf drivers/ide/ide-probe.c
> > > --- a/drivers/ide/ide-probe.c Fri Jan 30 16:37:15 2009 +0000
> > > +++ b/drivers/ide/ide-probe.c Fri Jan 30 17:00:08 2009 +0000
> > > @@ -1549,6 +1549,8 @@
> > > if (hwif_init(hwif) == 0) {
> > > printk(KERN_INFO "%s: failed to initialize IDE "
> > > "interface\n", hwif->name);
> > > + ide_disable_port(hwif);
> > > + device_unregister(&hwif->gendev);
> > > hwif->present = 0;
> > > continue;
> >
> > Hmmm, ide_disable_port() may free hwif so it needs to happen last...
>
> Oops, new version below.
>
> > Also did you have a chance to retest this? [ I couldn't reproduce
> > the issue so was hoping that you could test combined fixes. ]
>
> Yep, I can easily repro. I've actually fixed the underlying issue (a Xen
> IRQ problem) causing the probe to fail but can easily revert that fix to
> test these patches. If you wanted to repro yourself you could probably
> make the request_irq() in init_irq() artificially fail.
>
> Ian.
Ok, thanks for explaining this (I get "The Big Picture" now).
> IDE: Disable and unregister devices if initialization fails.
>
> On reboot the loop in device_shutdown gets confused by these partially
> initialized devices and goes into an infinite loop. Therefore disable
> and unregister these devices.
>
> Signed-off-by: Ian Campbell <ian.campbell@xxxxxxxxxx>
> ---
> drivers/ide/ide-probe.c | 10 ++++++----
> 1 file changed, 6 insertions(+), 4 deletions(-)
>
>
> diff -r 9d5441419978 -r f596f17b02c3 drivers/ide/ide-probe.c
> --- a/drivers/ide/ide-probe.c Fri Jan 30 17:39:53 2009 +0000
> +++ b/drivers/ide/ide-probe.c Mon Feb 02 10:02:32 2009 +0000
> @@ -1549,6 +1549,8 @@
> if (hwif_init(hwif) == 0) {
> printk(KERN_INFO "%s: failed to initialize IDE "
> "interface\n", hwif->name);
> + device_unregister(&hwif->gendev);
> + ide_disable_port(hwif);
> hwif->present = 0;
I applied the patch (removing leftover hwif->present while at it)
and sucessfully tested it locally (by making hwif_init() to fail).
Thanks,
Bart
--
To unsubscribe from this list: send the line "unsubscribe linux-ide" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
