On 6/10/24 06:38, Akihiko Odaki wrote:
> On 2024/06/10 4:02, Dmitry Osipenko wrote:
>> On 6/3/24 08:44, Akihiko Odaki wrote:
>>> On 2024/06/03 14:26, Dmitry Osipenko wrote:
>>>> On 6/2/24 08:34, Akihiko Odaki wrote:
>>>>>> +typedef enum {
>>>>>> + RS_START, /* starting state */
>>>>>> + RS_INIT_FAILED, /* failed initialisation */
>>>>>
>>>>> Is the distinction between RS_START and RS_INIT_FAILED really
>>>>> necessary?
>>>>
>>>> The state stays in RS_INIT_FAILED once was failed until virtio-gpu is
>>>> reset, re-initializing virglrenderer isn't allowed in this state.
>>>
>>> Can you elaborate more? Why isn't re-initializing allowed?
>>
>> In practice, if virglrenderer initialization failed once, it will fail
>> second time. Otherwise we will be retrying to init endlessly because
>> guest won't stop sending virgl commands even if they all are timing out.
>> Each initialization failure produces a error msg.
>>
> I see.
>
> A better solution is to add a new function to GraphicHwOps to call back
> after initializating the displays and before starting the guest. You can
> do virgl initialization in such a function, and exit(1) if the
> initialization fails because the guest has not started yet, saving this
> enum. I don't require you to make such a change however; this is not a
> regression of your patches so you have no obligation to fix it.
I'll keep this idea for a follow up patch, thanks! It will take me some
extra time to look through the display code, making sure that callback
is added properly and nothing is missed.
--
Best regards,
Dmitry
