Re: [PATCH 4/4] x86/xen: Drop CONFIG_XEN_PVHVM_SMP

[Jan Beulich <jbeulich@suse.com>]

On 05.08.2026 11:44, Jürgen Groß wrote:
> On 05.08.26 11:04, Jan Beulich wrote:
>> On 05.08.2026 10:55, Juergen Gross wrote:
>>> On 05.08.26 10:44, Jan Beulich wrote:
>>>> On 05.08.2026 10:21, Juergen Gross wrote:
>>>>> --- a/arch/x86/xen/Makefile
>>>>> +++ b/arch/x86/xen/Makefile
>>>>> @@ -37,8 +37,8 @@ obj-$(CONFIG_XEN_PVH)		+= enlighten_pvh.o
>>>>>    obj-$(CONFIG_EVENT_TRACING)	+= trace.o
>>>>>    
>>>>>    obj-$(CONFIG_SMP)		+= smp.o
>>>>> +obj-$(CONFIG_SMP)		+= smp_hvm.o
>>>>>    obj-$(CONFIG_XEN_PV_SMP)  	+= smp_pv.o
>>>>> -obj-$(CONFIG_XEN_PVHVM_SMP)  	+= smp_hvm.o
>>>>
>>>> Similar issue here - in a PV-only config smp_hvm.o doesn't want / shouldn't
>>>> need building.
>>>
>>> Note that I didn't change any functionality.
>>>
>>> I agree that it seems a little bit strange, but in the end I believe
>>> the current status is okay-ish. PV-only hasn't been something in upstream
>>> Linux since Xen support was added, as PV was always meant to be an
>>> alternative to bare metal support via paravirt patching. It might have
>>> been possible to build a kernel not really functional on bare metal, but
>>> this was more like the ability to build a x86 kernel not working on any
>>> existing machine.
>>>
>>> IMO the Xen kernel config options should allow for adding Xen-specific
>>> features, but minimum Xen support should always have basic HVM support,
>>> which includes the Xen specific performance enhancements.
>>
>> I fear I don't understand this. If I want a kernel just to run as PV Dom0,
>> why would it need to carry anything HVM-ish? That is (or should be)
>> entirely unrelated to being able to also run this same kernel on baremetal
>> then.
> 
> The fact is that Xen PV-mode was always a feature not really liked especially
> by x86 maintainers (this is the polite way to phrase it).
> 
> Changing something outside of xen-specific parts of the kernel in favor of PV is
> nearly always a fight and I'm pretty sure I only get changes in by playing nice
> (not changing more than absolutely necessary and cleaning up constantly). I will
> certainly not try to push for a "PV-only" kernel while the hope of the x86
> maintainers is more "PV will go away some time in future".
> 
> The capability to configure the kernel without HVM stuff but with Xen support
> is the way it has been since many years now. Nobody has missed a PV-only kernel,
> so I conclude there is no need for that.

Just FTR - I did, even if maybe I never said so explicitly. But as it's just
me ...

Jan

> The closest we will get here is a PVH
> only kernel, and this is possible by not enabling CONFIG_XEN_PVHVM_GUEST.
> 
> 
> Juergen