Re: [RFC v1 00/38] arm cleanup experiment for kvm-only build

[Claudio Fontana <cfontana@suse.de>]

On 3/1/21 5:23 PM, Alex Bennée wrote:
> 
> Claudio Fontana <cfontana@suse.de> writes:
> 
>> On 2/23/21 10:18 AM, Philippe Mathieu-Daudé wrote:
>>> On 2/22/21 8:00 PM, Alex Bennée wrote:
>>>>
>>>> Claudio Fontana <cfontana@suse.de> writes:
>>>>
>>>>> Hi all,
>>>>>
>>>>> this is an experiment, a cleanup based on and requiring the series
>>>>> "i386 cleanup PART 2":
>>>>>
>>>>> https://lists.gnu.org/archive/html/qemu-devel/2021-02/msg05935.html
>>>>>
>>>>> The goal is to split the code between TCG-only and non-TCG code,
>>>>> fixing the KVM-only build (configure --disable-tcg),
>>>>>
>>>>> and laying the ground for further cleanups and the use of the
>>>>> new accel objects in the hierarchy to specialize the cpu
>>>>> according to the accelerator.
>>>>>
>>>>> This is known to be an early state, with probably a lot of work
>>>>> still needed.
>>>>
>>>> Well early work is looking pretty good:
>>>>
>>>>   18:59:22 [alex@idun:~/l/q/b/default] review/arm-cleanup-rfc1|… ± ls -lh qemu-system-aarch64
>>>>   -rwxr-xr-x 1 alex alex 107M Feb 22 18:08 qemu-system-aarch64*
>>>>   18:59:29 [alex@idun:~/l/q/b/default] review/arm-cleanup-rfc1|… ± ls -lh ../disable.tcg/qemu-system-aarch64
>>>>   -rwxr-xr-x 1 alex alex 76M Feb 22 17:47 ../disable.tcg/qemu-system-aarch64*
>>>
>>> :~)
>>>
>>>>
>>>> and I've tested the KVM side works well enough with a basic image.
>>>
>>>
>>
>> I am working on the next version, one thing I noticed among others as I get close to the v2,
>> is the fact that tests/ for arm require tcg in many cases.
> 
> I think in a lot of cases they are historical because developers
> generally weren't running on native hardware. That said off the top of
> my head:
> 
>   tests/tcg - linux-user, so implies TCG
>   tests/tcg/system - use semihosting (at least for arm/aarch64) - which implies TCG
>   tests/acceptance/[replay_kernel/reverse_debugging/tcg_plugins] - all need TCG features
> 
> I don't think there is any reason the others can't run with KVM - and
> probably should on real hardware.


One thing I noticed is that tests try to run qemu-system-aarch64 with accel "qtest" and machine "virt",
and the thing tries to create a cortex-a15 cpu model for some unknown reason.

Digging and sweating..

Ciao,

Claudio

> 
>>
>> So there is even more cleanup needed to discern which are actually tcg-only, and how to tweak the others into working also with only kvm available..
>>
>> Ciao,
>>
>> Claudio
> 
>