On 23/08/2022 08:21, Jan Beulich wrote:
> On 22.08.2022 23:30, Andrew Cooper wrote:
>> arch_domain_create() can't blindly write into d->arch.hvm union. Move the
>> logic into hvm_domain_initialise(), which involves passing config down.
>>
>> Fixes: 2ce11ce249a3 ("x86/HVM: allow per-domain usage of hardware
>> virtualized APIC")
>> Signed-off-by: Andrew Cooper <[email protected]>
> Reviewed-by: Jan Beulich <[email protected]>
Thanks.
>> This does not fix XenServer's wall of red from testing, but I have at least
>> figured out what's going on. There's a piece of plain RAM in place of a
>> working LAPIC MMIO mapping (accelerated or otherwise), which causes HVMLoader
>> to spin in a tight loop waiting for CPU 1 to come up after failing to send an
>> INIT-SIPI-SIPI.
> Where's that page of plain RAM coming from? And do you meanwhile
> understand why XenServer testing is exposing the issue while osstest
> isn't?
The sink page is allocated and inserted into the P2M based on hardware
capabilities, and the VMCS is (not) configured using the new settings.
APIC-V is still fully disabled in XenServer because there are still
interrupts lost on migration which take out Windows. Jane's work was
the very start of being able to test APIC acceleration rationally, and
ultimately find a fix.
~Andrew
