In 2.6.20, I can boot an arm kernel up to a shell prompt including a virtual
scsi hard drive. In 2.6.21-rc1, this stopped working.
I tried "git bisect" and found out there's a range of about 5000 commits
between the two where arm doesn't compile. At the start of this range, the
controller worked. At the end, it didn't anymore.
How can YOU reproduce this problem? I'm glad you asked:
The miniconfig I'm using is attached. You'll need an arm compiler to build
it, of course. (If you haven't got one, the cross compiler I made is
at "http://landley.net/code/firmware/downloads/cross-compiler/". Download
the armv4l version for the appropriate host, extract the tarball and add
the "bin" directory under that to your path. The x86 version should work on
Ubuntu 6.06 or newer, the x86-64 version was built on 7.04.)
Configure with:
make ARCH=arm allnoconfig KCONFIG_ALLCONFIG=miniconfig-linux
make ARCH=arm CROSS_COMPILE=armv4l-
And then run qemu on it:
qemu-system-arm -M versatilepb -nographic -no-reboot \
-hda /dev/null -kernel zImage-armv4l -append \
'rw panic=1 root=/dev/sdaconsole=ttyAMA0'
The failing system will loop resetting the scsi controller with lots of
timeouts, and takes several minutes to get the the panic where it can't mount
root. The working system will panic due to root being /dev/null fairly
quickly, without pages of error messages and a very long wait first.
Rob
--
"One of my most productive days was throwing away 1000 lines of code."
- Ken Thompson.
Rob Landley wrote:
> In 2.6.20, I can boot an arm kernel up to a shell prompt including a virtual
> scsi hard drive. In 2.6.21-rc1, this stopped working.
>
It's broken with x86 too. I looked into it for a bit and it appears
that the Linux driver is using more LSI commands than it was before.
The odd thing is that it's probing some space that is marked reserved in
the LSI manual. Unfortunately, I couldn't understand the SCSI system
well enough to understand what changed in Linux.
Regards,
Anthony Liguori
> I tried "git bisect" and found out there's a range of about 5000 commits
> between the two where arm doesn't compile. At the start of this range, the
> controller worked. At the end, it didn't anymore.
>
> How can YOU reproduce this problem? I'm glad you asked:
>
> The miniconfig I'm using is attached. You'll need an arm compiler to build
> it, of course. (If you haven't got one, the cross compiler I made is
> at "http://landley.net/code/firmware/downloads/cross-compiler/". Download
> the armv4l version for the appropriate host, extract the tarball and add
> the "bin" directory under that to your path. The x86 version should work on
> Ubuntu 6.06 or newer, the x86-64 version was built on 7.04.)
>
> Configure with:
>
> make ARCH=arm allnoconfig KCONFIG_ALLCONFIG=miniconfig-linux
> make ARCH=arm CROSS_COMPILE=armv4l-
>
> And then run qemu on it:
>
> qemu-system-arm -M versatilepb -nographic -no-reboot \
> -hda /dev/null -kernel zImage-armv4l -append \
> 'rw panic=1 root=/dev/sdaconsole=ttyAMA0'
>
> The failing system will loop resetting the scsi controller with lots of
> timeouts, and takes several minutes to get the the panic where it can't mount
> root. The working system will panic due to root being /dev/null fairly
> quickly, without pages of error messages and a very long wait first.
>
> Rob