Re: [PATCH v5 02/10] accel: Introduce 'query-accels' QMP command

[Philippe Mathieu-Daudé]

On 5/5/21 5:01 PM, Markus Armbruster wrote:
> Philippe Mathieu-Daudé <philmd@redhat.com> writes:
> 
>> On 5/5/21 9:07 AM, Markus Armbruster wrote:
>>> Philippe Mathieu-Daudé <philmd@redhat.com> writes:
>>>
>>>> Introduce the 'query-accels' QMP command which returns a list
>>>> of built-in accelerator names.
>>>>
>>>> - Accelerator is a QAPI enum of all existing accelerators,
>>>>
>>>> - AcceleratorInfo is a QAPI structure providing accelerator
>>>>   specific information. Currently the common structure base
>>>>   provides the name of the accelerator, while the specific
>>>>   part is empty, but each accelerator can expand it.
>>>>
>>>> - 'query-accels' QMP command returns a list of @AcceleratorInfo
>>>>
>>>> For example on a KVM-only build we get:
>>>>
>>>>     { "execute": "query-accels" }
>>>>     {
>>>>         "return": [
>>>>             {
>>>>                 "name": "qtest"
>>>>             },
>>>>             {
>>>>                 "name": "kvm"
>>>>             }
>>>>         ]
>>>>     }
>>>>
>>>> Note that we can't make the enum values or union branches conditional
>>>> because of target-specific poisoning of accelerator definitions.
>>>
>>> Let me try to complete this argument;
>>>
>>>   If we did, enum Accelerator could only be used in target-specific
>>>   code.  But we want to use it in generic code, too.
>>>
>>> Which generic code exactly?
>>
>>
>> cpu.c:133:#ifdef CONFIG_TCG
>> hmp-commands-info.hx:271:#if defined(CONFIG_TCG)
>> monitor/misc.c:324:#ifdef CONFIG_TCG
>> softmmu/physmem.c:28:#ifdef CONFIG_TCG
>>
>> and more importantly:
>>
>> include/exec/cpu-all.h:430:#ifdef CONFIG_TCG
>> include/exec/cpu-defs.h:77:#if !defined(CONFIG_USER_ONLY) &&
>> defined(CONFIG_TCG)
>> include/exec/exec-all.h:25:#ifdef CONFIG_TCG
>> include/sysemu/cpu-timers.h:27:#ifdef CONFIG_TCG
>>
>> Thomas might provide more cases, IIRC he audited this recently.
> 
> No need, I'm interested in examples, not a complete list.
> 
> Please amend the commit message:
> 
>     Note that we can't make the enum values or union branches conditional
>     because of target-specific poisoning of accelerator definitions.
>     If we did, enum Accelerator could only be used in target-specific
>     code.  But we want to also use it in generic code, such as ...
> 
> with ... replaced by one or more references to relevant code.
> 
> If we expect readers of the code to later wonder why we're not using
> QAPI conditionals, then we should add a comment to the QAPI schema,
> too.  Not a demand, since I'm not sure what to expect.

OK, will do in v8.