[Qemu-devel] Linux 3.4, VersatilePB and PCI

[Qemu-devel] Linux 3.4, VersatilePB and PCI

[Marc Marí]

Hi

In short:
I have a Linux 3.4 kernel on top of a VersatilePB machine that
fails to mount root file system with message:

sym53c8xx 0000:00:0c.0: BAR 0: can't reserve [io 0x44000000-0x440000ff]

after booting with:

qemu-system-arm -M versatilepb -serial telnet::4444,server -kernel
vmlinuz-3.4.0 -initrd initrd.img-3.4.0 -hda hda.img -hdb hdb.img
-append "console=ttyAMA0 root=/dev/sda1"

In long:
I have a VersatilePB machine with Debian and Linux 3.2 on top. I
downloaded Linux 3.4 and, inside the machine (no cross-compilation, to
avoid any little issues), I copied the actual 3.2 config to the new
kernel, run menuconfig, just changed SCSI ACARD support (because it
does not compile in 3.4) and kernel compression to GZIP, and run
make-kpkg --initrd kernel_image (Debian-specific tools, to avoid any
little issues). When it finished, I installed the resulting .deb, and
in the host I copied the /boot/vmlinuz-3.4.0 and the /boot/initrd-3.4.0
from the guest, and tried to boot again. Initrd will fail to mount root
file system with the message above.

But if I do exactly the same process with a Linux 3.2.0 it does work.

Am I doing something wrong with the kernels? Or is this a bug either in
QEMU or Linux? I'd like a bit of help in, at least, knowing if the
error is anything related with QEMU.

Thanks
Marc

Re: [Qemu-devel] Linux 3.4, VersatilePB and PCI

[Peter Maydell]

On 13 January 2015 at 09:05, Marc Marí <marc.mari.barcelo@gmail.com> wrote:
> Hi
>
> In short:
> I have a Linux 3.4 kernel on top of a VersatilePB machine that
> fails to mount root file system with message:
>
> sym53c8xx 0000:00:0c.0: BAR 0: can't reserve [io 0x44000000-0x440000ff]
>
> after booting with:
>
> qemu-system-arm -M versatilepb -serial telnet::4444,server -kernel
> vmlinuz-3.4.0 -initrd initrd.img-3.4.0 -hda hda.img -hdb hdb.img
> -append "console=ttyAMA0 root=/dev/sda1"

3.4 is pretty ancient; have you tried a newer kernel?

-- PMM

Re: [Qemu-devel] Linux 3.4, VersatilePB and PCI

[Marc Marí]

El Tue, 13 Jan 2015 10:06:00 +0000
Peter Maydell <peter.maydell@linaro.org> escribió:
> On 13 January 2015 at 09:05, Marc Marí <marc.mari.barcelo@gmail.com>
> wrote:
> > Hi
> >
> > In short:
> > I have a Linux 3.4 kernel on top of a VersatilePB machine that
> > fails to mount root file system with message:
> >
> > sym53c8xx 0000:00:0c.0: BAR 0: can't reserve [io
> > 0x44000000-0x440000ff]
> >
> > after booting with:
> >
> > qemu-system-arm -M versatilepb -serial telnet::4444,server -kernel
> > vmlinuz-3.4.0 -initrd initrd.img-3.4.0 -hda hda.img -hdb hdb.img
> > -append "console=ttyAMA0 root=/dev/sda1"
> 
> 3.4 is pretty ancient; have you tried a newer kernel?
> 
> -- PMM

I was fearing that answer :). No, I can't. I want to compare the
behaviour of a "normal" Linux 3.4 with another modified Linux 3.4, and
the second one is fixed.

And is not that ancient, it only has 2.5 years :)

Thanks
Marc

Re: [Qemu-devel] Linux 3.4, VersatilePB and PCI

[Peter Maydell]

On 13 January 2015 at 10:35, Marc Marí <marc.mari.barcelo@gmail.com> wrote:
> I was fearing that answer :). No, I can't. I want to compare the
> behaviour of a "normal" Linux 3.4 with another modified Linux 3.4, and
> the second one is fixed.
>
> And is not that ancient, it only has 2.5 years :)

In that case you'll need to debug the issue like you would
any other "kernel misbehaving" issue -- figure out what the
kernel is doing to cause it to print that message, identify
whether that's because the kernel itself is buggy or because
QEMU's hardware emulation is buggy, and go from there...

There were a number of bugfixes applied to the versatile
PCI controller code at some point after 3.4, so I wouldn't
be very surprised if the answer turns out to be "this is
just broken in 3.4".

In particular, commit 9b0f7e39923 which isn't in 3.4 AFAIK
basically says "we broke I/O access", which is probably what
you're running into. I expect you'll need other fixes too.

So you can try to find and cherry-pick the necessary fixes, which
could be painful, or you can move to a newer kernel which already
has them.

-- PMM

Re: [Qemu-devel] Linux 3.4, VersatilePB and PCI

[Marc Marí]

El Tue, 13 Jan 2015 10:46:20 +0000
Peter Maydell <peter.maydell@linaro.org> escribió:
> On 13 January 2015 at 10:35, Marc Marí <marc.mari.barcelo@gmail.com>
> wrote:
> > I was fearing that answer :). No, I can't. I want to compare the
> > behaviour of a "normal" Linux 3.4 with another modified Linux 3.4,
> > and the second one is fixed.
> >
> > And is not that ancient, it only has 2.5 years :)
> 
> In that case you'll need to debug the issue like you would
> any other "kernel misbehaving" issue -- figure out what the
> kernel is doing to cause it to print that message, identify
> whether that's because the kernel itself is buggy or because
> QEMU's hardware emulation is buggy, and go from there...
> 
> There were a number of bugfixes applied to the versatile
> PCI controller code at some point after 3.4, so I wouldn't
> be very surprised if the answer turns out to be "this is
> just broken in 3.4".
> 
> In particular, commit 9b0f7e39923 which isn't in 3.4 AFAIK
> basically says "we broke I/O access", which is probably what
> you're running into. I expect you'll need other fixes too.
> 
> So you can try to find and cherry-pick the necessary fixes, which
> could be painful, or you can move to a newer kernel which already
> has them.

Thank you very much. I didn't know where to start looking, so I was
lost. This helps me find my way again.

Thanks
Marc