Re: qemu-kvm broken after ./configure --disable-kvm

[Jan Kiszka <jan.kiszka@web.de>]

[-- Attachment #1: Type: text/plain, Size: 1611 bytes --]

Avi Kivity wrote:
> Jan Kiszka wrote:
>>>>> Building latest git with ./configure --disable-kvm breaks with
>>>>> errors in
>>>>> pcspk.c
>>>>>             
>>>> With latest git, things break much earlier in case your host does not
>>>> provide linux/kvm.h because libkvm-all.h includes it unconditionally.
>>>>
>>>>         
>>> But qemu-kvm carries linux/kvm.h, so this never happens?
>>>     
>>
>> 1. qemu-kvm does not use its own headers when you specify --disble-kvm.
>>    That's the technical reason for this bug.
>>   
> 
> Ah, okay.  This should be fixed (by including the headers) as long as we
> continue to carry kvm.h.
> 
>> 2. Upstream does not, and it's unclear if it ever will (if we push
>>    recent headers into kvm-kmod, I think there is no urgent need
>>    anymore). At least for code to-be-pushed upstream, we must not
>>    rely in this anyway.
> 
> Yes.
> 
> Adding the headers to kvm-kmod.h is the right thing technically, but
> something tells me we'll get a lot of failures by people compiling first
> and installing later, rather than the sequence needed to make things
> work: compile and install kvm-kmod, compile and install qemu[-kvm].  Not
> all of the failures will be visible at compile time.
> 

That could (and probably should - independent of in-tree headers) be
caught by making all KVM_CAPs mandatory, ie. check for the latest and
greatest ones during configure and drop all the #ifdefs from the code.

Whatever the strategy will be, it should be one with the clear goal to
converge over the same approach with upstream.

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 257 bytes --]