Re: Deprecate 32-bit hosts? (was: Re: [PULL 14/14] hw/arm/aspeed: Add Fuji machine type)

[Philippe Mathieu-Daudé]

On 9/15/21 10:37 AM, Daniel P. Berrangé wrote:
> On Wed, Sep 15, 2021 at 09:42:48AM +0200, Thomas Huth wrote:
>> On 14/09/2021 17.22, Richard Henderson wrote:
>>> On 9/14/21 5:26 AM, Peter Maydell wrote:
>>>> (2) RAM blocks should have a length that fits inside a
>>>>      signed 32-bit type on 32-bit hosts (at least I assume this
>>>>      is where the 2047MB limit is coming from; in theory this ought
>>>>      to be improveable but auditing the code for mishandling of
>>>>      RAMblock sizes to ensure we weren't accidentally stuffing
>>>>      their size into a signed 'long' somewhere would be kind
>>>>      of painful)
>>>
>>> Recalling that the win64 abi model is p64, i.e. 'long' is still 32-bit
>>> while pointers are 64-bit, how close do we think we are to this being
>>> fixed already?
>>>
>>>> Even if we did fix (2) we'd need to compromise on (3)
>>>> sometimes still -- if a board has 4GB of RAM that's
>>>> not going to fit in 32 bits regardless. But we would be
>>>> able to let boards with 2GB have 2GB.
>>>
>>> I'm not opposed to deprecating 32-bit hosts...  ;-)
>>
>> I think we should consider this again, indeed. Plain 32-bit CPUs are quite
>> seldom these days, aren't they? And I think we urgently need to decrease the
>> amount of things that we have to test and maintain in our CI and developer
>> branches... So is there still a really really compelling reason to keep
>> 32-bit host support alive?
> 
> I think it probably depends on the architecture to some extent.
> 
> i386 is possibly getting rare enough to consider dropping, though
> IIUC, KVM in the kernel still supports it.  Would feel odd to drop
> it in QEMU if the kernel still thinks it is popular enough to keep
> KVM support.
> 
> armv7 feels like it is relatively common as 64-bit didn't arrive
> in widespread use until relatively recent times compared to x86_64.
> KVM dropped armv7, but then hardware for that was never widespread,
> so armv7 was always TCG dominated
> 
> Other 32-bit arches were/are always rare.

While I could understand there are rare uses of system emulation on
32-bit hosts, I still believe user-emulation is used, but would like
to be proven to the contrary. With that in mind, I'm not sure removing
sysemu on 32-bit hosts is worthful. Maybe we should ask distribution
maintainers first, then eventually poll the community? Or start with
a deprecation warning?

>> Could we maybe also decrease the amount of targets, i.e. merge
>> qemu-system-x86_64 and qemu-system-i386, merge qemu-system-ppc64 and
>> qemu-system-ppc, etc. where it makes sense (i.e. where one of the binaries
>> is a superset of the other)?
> 
> Regards,
> Daniel
>