Michael Kelley <[email protected]> writes:

> From: Vitaly Kuznetsov <[email protected]> Sent: Wednesday, November 18, 
> 2026 6:02 AM
>> 
>> Mukesh R <[email protected]> writes:
>> 
>> > On 11/17/26 09:41, Michael Kelley wrote:
>> 
>> ...
>> 
>> > Confirming, that it is same behavior all the way back to 2016.
>> >
>> >> Sounds good. FWIW, old TLFS version 2.0A, which is specifically
>> >> titled to cover Windows Server 2008 R2, has the paragraph
>> >> about hypercall continuation. So that functionality was evidently
>> >> present very early on. It is still curious that Vitaly found it
>> >> necessary to include the looping behavior in the Linux guest
>> >> code.
>> >
>> > Maybe Vitaly was running into a genuine bug, not sure.
>> >
>> 
>> I'm not sure either :-(
>> 
>> WS2012 was definitely around back then, not sure about 2008. It may also
>> make sense to add at least a debug print in case we find ourselves
>> running on something that old. AFAICS, even the TLFS version 2.0A
>> mentions PV TLB flush hypercalls. If looping is still required for these
>> old versions, we may introduce some very hard-to-find bugs if we drop
>> it.
>> 
>
> Vitaly -- I'll point out that support for running Linux guests
> on any Hyper-V older than WS2016/Win10 has been removed
> from the upstream code tree as of 7.4. 

Missed that, thanks for pointing out!

> The removal is
> impactful enough that 7.4 and later kernels won't boot in a
> guest on these older Hyper-V versions. Hopefully whatever bug
> you encountered is no longer present in WS2016 and later. But
> even if the looping is removed, the code could still check and
> warn on a hypercall completion with no error but not all reps
> completed. Then at least we'd know there is a problem.

Yes, WARN[_ON_ONCE]() should be sufficient.

-- 
Vitaly


Reply via email to