Re: Re: [RFC] hw/nvme: Use irqfd to send interrupts

[Keith Busch]

On Wed, Aug 10, 2022 at 12:48:53AM +0800, 樊金昊 wrote:
>> The driver will create the cq with an allocated vector, but it's not 
>> activated
>> until after the driver wires it up to a handler. I think that's what you're
>> observing with the incomplete MSIx table entry on creation.
> 
> Also, I'm wondering if this is inconsistent with the NVMe spec. In Section 
> 7.6.1
> of the 1.4 spec, it says "After determining the number of I/O Queues, the MSI
> and/or MSI-X registers should be configured;" in Step 8, and CQ creation 
> happens
> in Step 9. Now the driver changes MSI-X registers after CQ creation, is it a
> violation of the spec?

I don't think it's a problem. This is really a more "informative" section of
the spec and doesn't specify any hard requirements. You should be able to rely
on the entry's data being stable after the first queue doorbell ring, though.