2013-04-02 15:54:35

by Josh Boyer

[permalink] [raw]
Subject: Stable kernel 3.8.4/3.9-rc3 breaks PNP serial port

Hi All,

We've had a report [1] that the 3.8.4 stable kernel makes a PNP serial
port stop working. After testing 3.8.3, the reporter narrowed it down
to stable commit eec98f82c637 (tty/8250_pnp: serial port detection
regression since v3.7) (upstream commit 77e372a3d82). Reverting that
single commit from 3.8.5 allows the serial port to continue working.

The machine in question is using an AMI UEFI implementation as the
firmware, and on 3.8.3 the serial port has the following in
/sys/bus/pnp/devices/00:0a/resources

state = active
io disabled
irq 4
dma disabled

With 3.8.[45], the same file has:

state = disabled
io disabled
irq 4
dma disabled

In both cases, the io and dma resources are disabled, and an IRQ is
assigned. However, the state remains active for the port on 3.8.3.

The offending commit mentions this is a BIOS bug from InsydeH2O and that
the port is bogus in that case, but we have something similar here with
an AMI UEFI implementation (Version: 0406 Release Date: 06/06/2012)
where the port isn't bogus.

I'm not sure exactly what the solution should be here. Any thoughts?

josh

[1] https://bugzilla.redhat.com/show_bug.cgi?id=928246


2013-04-02 16:23:41

by Greg Kroah-Hartman

[permalink] [raw]
Subject: Re: Stable kernel 3.8.4/3.9-rc3 breaks PNP serial port

On Tue, Apr 02, 2013 at 11:53:44AM -0400, Josh Boyer wrote:
> Hi All,
>
> We've had a report [1] that the 3.8.4 stable kernel makes a PNP serial
> port stop working. After testing 3.8.3, the reporter narrowed it down
> to stable commit eec98f82c637 (tty/8250_pnp: serial port detection
> regression since v3.7) (upstream commit 77e372a3d82). Reverting that
> single commit from 3.8.5 allows the serial port to continue working.
>
> The machine in question is using an AMI UEFI implementation as the
> firmware, and on 3.8.3 the serial port has the following in
> /sys/bus/pnp/devices/00:0a/resources
>
> state = active
> io disabled
> irq 4
> dma disabled
>
> With 3.8.[45], the same file has:
>
> state = disabled
> io disabled
> irq 4
> dma disabled
>
> In both cases, the io and dma resources are disabled, and an IRQ is
> assigned. However, the state remains active for the port on 3.8.3.
>
> The offending commit mentions this is a BIOS bug from InsydeH2O and that
> the port is bogus in that case, but we have something similar here with
> an AMI UEFI implementation (Version: 0406 Release Date: 06/06/2012)
> where the port isn't bogus.
>
> I'm not sure exactly what the solution should be here. Any thoughts?

Sean, should I just revert this patch now, and wait for a better fix
later?

thanks,

greg k-h

2013-04-02 16:34:39

by Sean Young

[permalink] [raw]
Subject: Re: Stable kernel 3.8.4/3.9-rc3 breaks PNP serial port

On Tue, Apr 02, 2013 at 09:23:36AM -0700, Greg Kroah-Hartman wrote:
> On Tue, Apr 02, 2013 at 11:53:44AM -0400, Josh Boyer wrote:
> > Hi All,
> >
> > We've had a report [1] that the 3.8.4 stable kernel makes a PNP serial
> > port stop working. After testing 3.8.3, the reporter narrowed it down
> > to stable commit eec98f82c637 (tty/8250_pnp: serial port detection
> > regression since v3.7) (upstream commit 77e372a3d82). Reverting that
> > single commit from 3.8.5 allows the serial port to continue working.
> >
> > The machine in question is using an AMI UEFI implementation as the
> > firmware, and on 3.8.3 the serial port has the following in
> > /sys/bus/pnp/devices/00:0a/resources
> >
> > state = active
> > io disabled
> > irq 4
> > dma disabled
> >
> > With 3.8.[45], the same file has:
> >
> > state = disabled
> > io disabled
> > irq 4
> > dma disabled
> >
> > In both cases, the io and dma resources are disabled, and an IRQ is
> > assigned. However, the state remains active for the port on 3.8.3.
> >
> > The offending commit mentions this is a BIOS bug from InsydeH2O and that
> > the port is bogus in that case, but we have something similar here with
> > an AMI UEFI implementation (Version: 0406 Release Date: 06/06/2012)
> > where the port isn't bogus.
> >
> > I'm not sure exactly what the solution should be here. Any thoughts?
>
> Sean, should I just revert this patch now, and wait for a better fix
> later?

Looks like although the pnp io is disabled, it does contain a valid port
number, else pnp detection would fail in 3.8.3. We're a little stricter
in 3.8.4, but it should fall back to legacy port detection -- the port
and irq are standard.

I don't understand why the legacy port detection (see SERIAL_PORT_DFNS)
isn't working. Since the fallback isn't working properly maybe this
patch should be reverted for now.

If the pnp information had some virtue we wouldn't have these problems.


Sean

2013-04-02 23:24:38

by Sean Young

[permalink] [raw]
Subject: Re: Stable kernel 3.8.4/3.9-rc3 breaks PNP serial port

On Tue, Apr 02, 2013 at 05:34:35PM +0100, Sean Young wrote:
> On Tue, Apr 02, 2013 at 09:23:36AM -0700, Greg Kroah-Hartman wrote:
> > On Tue, Apr 02, 2013 at 11:53:44AM -0400, Josh Boyer wrote:
> > > Hi All,
> > >
> > > We've had a report [1] that the 3.8.4 stable kernel makes a PNP serial
> > > port stop working. After testing 3.8.3, the reporter narrowed it down
> > > to stable commit eec98f82c637 (tty/8250_pnp: serial port detection
> > > regression since v3.7) (upstream commit 77e372a3d82). Reverting that
> > > single commit from 3.8.5 allows the serial port to continue working.
> > >
> > > The machine in question is using an AMI UEFI implementation as the
> > > firmware, and on 3.8.3 the serial port has the following in
> > > /sys/bus/pnp/devices/00:0a/resources
> > >
> > > state = active
> > > io disabled
> > > irq 4
> > > dma disabled
> > >
> > > With 3.8.[45], the same file has:
> > >
> > > state = disabled
> > > io disabled
> > > irq 4
> > > dma disabled
> > >
> > > In both cases, the io and dma resources are disabled, and an IRQ is
> > > assigned. However, the state remains active for the port on 3.8.3.
> > >
> > > The offending commit mentions this is a BIOS bug from InsydeH2O and that
> > > the port is bogus in that case, but we have something similar here with
> > > an AMI UEFI implementation (Version: 0406 Release Date: 06/06/2012)
> > > where the port isn't bogus.
> > >
> > > I'm not sure exactly what the solution should be here. Any thoughts?
> >
> > Sean, should I just revert this patch now, and wait for a better fix
> > later?
>
> Looks like although the pnp io is disabled, it does contain a valid port
> number, else pnp detection would fail in 3.8.3. We're a little stricter
> in 3.8.4, but it should fall back to legacy port detection -- the port
> and irq are standard.

The device is disabled after pnp probe fails, so the legacy probe fails
as well. I can think of two options:

1) check if both io and irq are disabled, as they are on the bogus bios
for which the patch was written, but not on this board
2) try to detect the port at the pnp io/irq provided, enabled or not, and
fail if no port can be found

I think the second is better but I'm not sure how much code needs to change
for that. I'll try over the next couple of evenings.


Sean