Re: Re: NX related build failure in latest pv_ops dom0 hackery tree

[Jeremy Fitzhardinge <jeremy@goop.org>]

Christophe Saout wrote:
> Ok, you are right.  The hackery kernel sees a different issue.  The AHCI
> setup actually works and hard disks are detected, but then it crashes in
> my initrd:
>
> Write protecting the kernel read-only data: 5868k
> init[1] general protection ip:ffffe42f sp:ffc5cfa4 error:0
>
> (the init process is a bash in this case)
>
> Note that this probably has to do with the fact that my initrd is 32 bit
> and hence runs in 32 bit compat mode (I was just to lazy to make me a
> new one when I initially bought my first 64 machine).
>
> The funny thing is that I also see this when running the kernel under
> qemu but just assumed a problem in qemu and never investigated.  qemu is
> emulating an AMD CPU I think, so maybe it's something related to
> sysenter vs. syscall or something. (?)  Because 32 bit compat mode under
> XEN works on my Intel notebook.
>   

Yes, likely.  There's an outstanding bug in 32-bit compat on AMD 
systems, but I haven't got around to setting up a system to repro+debug 
on.  The workaround is vdso32=0 on the kernel command line.

> So I guess if I redo the initramfs as 64 bit the hackery kernel would
> boot.  Maybe someone else who has the kernel running on an AMD machine
> can test 32 bit compat mode to confirm this observation?
>
> Ok, back to the AHCI issue, test push2/xen/dom0/master is ongoing.
>   

Thanks,
    J