Re: [PATCH] hw/nvme: Use ioeventfd to handle doorbell updates

[Jinhao Fan]

> That looks correct since we don't need the ioevent is an optional 
> optimization.
> 
> I would just suggest making this easier to read. For example, in
> nvme_init_sq_ioeventfd(), instead of assigning within a conditional:
> 
>    if ((ret = event_notifier_init(&cq->notifier, 0))) {
> 
> Do each part separately:
> 
>    ret = event_notifier_init(&cq->notifier, 0);
>    if (ret) {

Thanks for the comment! Will change in the next version.

> 
>> I’ve also been wondering whether using irqfd for sending interrupts can
>> bring some benefits. I’m not familiar with how QEMU emulates interrupts.
>> What do you think of irqfd’s?
> 
> Not sure about this mechanism, I'll need to look into it.

Since irqfd is internally a counter, I guess it may be able to “coalesce”
interrupts so that performance can be improved. I will try it tomorrow.