Xen status

Xen status

[Jeremy Fitzhardinge]

Well, the PDA stuff took a bit longer to settle down than I would have 
hoped, but it seems fairly sound now.

I just committed the changes needed to make Xen happy with the PDA, 
which means making sure an appropriate GDT is set up early enough.

I also seem to have avoided most of the APIC issues under Xen by simply 
making the cpuid not report the presence of an APIC.

With these in place, things are getting fairly far, but not to 
usermode.  It often gets as far as trying to mount root, but it fails; I 
haven't investigated that yet.  But it also frequently crashes before 
then, depending on when it gets an interrupt.  I'm trying to work out 
what's happening here, but the symptom is eip == 0x00000001 (!).

Chris, what local hacks do you have in your tree?

    j

Re: Xen status

[Zachary Amsden]

Jeremy Fitzhardinge wrote:
> Well, the PDA stuff took a bit longer to settle down than I would have 
> hoped, but it seems fairly sound now.
>
> I just committed the changes needed to make Xen happy with the PDA, 
> which means making sure an appropriate GDT is set up early enough.
>
> I also seem to have avoided most of the APIC issues under Xen by 
> simply making the cpuid not report the presence of an APIC.
>
> With these in place, things are getting fairly far, but not to 
> usermode.  It often gets as far as trying to mount root, but it fails; 
> I haven't investigated that yet.  But it also frequently crashes 
> before then, depending on when it gets an interrupt.  I'm trying to 
> work out what's happening here, but the symptom is eip == 0x00000001 (!).

Bad patches?  I put code in vmi.c to patch IRET and SYSEXIT EIPs to 0x1 
and 0x2, respectively, once upon a time.  Maybe it migrated over to you 
somehow?

Re: Xen status

[Chris Wright]

* Jeremy Fitzhardinge (jeremy@goop.org) wrote:
> Well, the PDA stuff took a bit longer to settle down than I would have 
> hoped, but it seems fairly sound now.
> 
> I just committed the changes needed to make Xen happy with the PDA, 
> which means making sure an appropriate GDT is set up early enough.
> 
> I also seem to have avoided most of the APIC issues under Xen by simply 
> making the cpuid not report the presence of an APIC.
> 
> With these in place, things are getting fairly far, but not to 
> usermode.  It often gets as far as trying to mount root, but it fails; I 
> haven't investigated that yet.  But it also frequently crashes before 
> then, depending on when it gets an interrupt.  I'm trying to work out 
> what's happening here, but the symptom is eip == 0x00000001 (!).

Haven't seen that one.

> Chris, what local hacks do you have in your tree?

Essentially all the local hacks I have are for debugging.  There was
one bit which was just for apic clock setup.  But the symptom w/out
that was a hang, not a crash.  They're also on a remote machine that
has apparently died.  Re-tooling locally ATM.

thanks,
-chris

Re: Xen status

[Jeremy Fitzhardinge]

Chris Wright wrote:
>> With these in place, things are getting fairly far, but not to 
>> usermode.  It often gets as far as trying to mount root, but it fails; I 
>> haven't investigated that yet.  But it also frequently crashes before 
>> then, depending on when it gets an interrupt.  I'm trying to work out 
>> what's happening here, but the symptom is eip == 0x00000001 (!).
>>     
>
> Haven't seen that one.
>   

Yeah, it looks like something is going amiss in interrupt delivery.  
Maybe something to do with interrupt stacks...

I never saw anything strange relating to timer events, but perhaps this 
is a manifestation of that problem.

> Essentially all the local hacks I have are for debugging.  There was
> one bit which was just for apic clock setup.

I think I worked around that by masking off the ACPI bit in CPUID.

>   But the symptom w/out that was a hang, not a crash. 
Yes, I'm past that one.

>  They're also on a remote machine that
> has apparently died.  Re-tooling locally ATM.
>   
That sucks.

    J