Re: [PATCH qemu] fdmon-io_uring: avoid idle event loop being accounted as IO wait

[Fiona Ebner <f.ebner@proxmox.com>]

Am 24.04.26 um 11:12 AM schrieb Fiona Ebner:
> Am 23.04.26 um 10:37 PM schrieb Thomas Lamprecht:
>> Am 23.04.26 um 17:34 schrieb Fiona Ebner:
>>> Am 22.04.26 um 4:54 PM schrieb Fiona Ebner:
>>>> @@ -317,6 +319,12 @@ struct AioContext {
>>>>  
>>>>      /* Pending callback state for cqe handlers */
>>>>      CqeHandlerSimpleQ cqe_handler_ready_list;
>>>> +
>>>> +    /*
>>>> +     * Number of in-flight requests to be accounted for IO wait.
>>>> +     * Must be accessed using atomics.
>>
>> why though? AioContext is strictly single-threaded nowadays and (enqueue)
>> and process_cqe (dequeue) run on the owning thread. Or is this just
>> defensive protection for potential future changes - as it's cheap I'm
>> fine with it, just wanted to know if I'm overlooking something here.
> 
> No, looking over it again, I think you are right. I was under the
> impression that somehow vCPUs might also add SQEs, but no, that happens
> in the iothread already.

vCPU threads might reach fdmon_io_uring_wait() via aio_poll() (e.g. via
virtio_scsi_dataplane_stop() during Linux guest boot), so atomics are
really necessary.