Re: [Xen-devel] [PATCH for-4.8] libxc/x86: Report consistent initial APIC value for PV guests

[Boris Ostrovsky <boris.ostrovsky@xxxxxxxxxx>]

On 11/10/2016 10:08 AM, Andrew Cooper wrote:
> On 10/11/16 15:05, Boris Ostrovsky wrote:
>> On 11/10/2016 09:55 AM, Andrew Cooper wrote:
>>> On 10/11/16 14:50, Boris Ostrovsky wrote:
>>>> Currently hypervisor provides PV guest's CPUID(1).EBX[31:24] (initial
>>>> APIC ID) with contents of that field on the processor that launched
>>>> the guest. This results in the guest reporting different initial
>>>> APIC IDs across runs.
>>>>
>>>> We should be consistent in how this value is reported, let's set
>>>> it to 0 (which is also what Linux guests expect).
>>>>
>>>> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@xxxxxxxxxx>
>>> This surely wants to go along with:
>> Probably, although Linux PV always reports APIC ID as zero (whole PV
>> APIC is a mess there as it is tied to topology discovery and we don't do
>> this well, to put it charitably).
> If PV linux always overrides this to 0, why do you need the toolstack
> fix in the first place?

I meant that Linux overrides APICID read from APIC (which, of course, we
don't have) to zero, not CPUID.

>
>>> andrewcoop@andrewcoop:/local/xen.git/xen$ git diff
>>> diff --git a/xen/arch/x86/traps.c b/xen/arch/x86/traps.c
>>> index b51b51b..bdf9339 100644
>>> --- a/xen/arch/x86/traps.c
>>> +++ b/xen/arch/x86/traps.c
>>> @@ -985,6 +985,10 @@ void pv_cpuid(struct cpu_user_regs *regs)
>>>          uint32_t tmp, _ecx, _ebx;
>>>  
>>>      case 0x00000001:
>>> +        /* Fix up VLAPIC details. */
>>> +        b &= 0x00FFFFFFu;
>>> +        b |= (curr->vcpu_id * 2) << 24;
>> Do we also need to multiply by two for PV guests? Or is it just to be
>> consistent with HVM?
> Frankly, until I get CPUID phase 2 sorted, this is all held together
> with good wishes, rather than duck tape.  I am astounded it has held
> together this long.

It does not anymore. We are being yelled at by Linux x86 maintainers to
get our stuff in order.

>
> HVM chooses an even APIC ID to prevent the VM thinking it has hyperthreads.


Right, but it's not needed (I think) by PV. But we might do it that way
to keep the two in sync. Let me test it and see what breaks.

-boris



_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel