On Wed, Jan 29, 2014 at 04:38:58PM +1030, Rusty Russell wrote:
> "Michael S. Tsirkin" <> writes:
> > Two comments received from a co-worker:
> >
> > 3.1 "The device MUST NOT consume buffers before DRIVER_OK ..." - I
> > personally
> > would like to see also something like: "device must not generate
> > interrupts if
> > DRIVER_OK is not set” to better handle power management transitions.
>
> That addition seems to make sense, but I don't understand the reference
> to power management transitions?
Well - ATM we do things like suspend by device reset and re-initialization,
right? I'm guessing during e.g. resume it's important not to get
interrupts too early. I'll check.
Hmm actually we have a chapter about initialization but not cleanup.
Might be a good addition.
Also, I note that QEMU implemented master abort bit by
clearing DRIVER_OK - AFAIK useful to stop device in a
standard way if driver malfunctions.
Also with QEMU you can clear DRIVER_OK and this stops the device. I
wonder if someone uses this.
> > 4.1.3.1.1.
> > Must clarify that all devices MUST support all cfg_type values 1-5.
> > This way BIOS can limit type 5 exclusively and drivers don't
> > need to add support for type 5.
>
> The word "support" here is unclear to me. Are you saying that a device
> MUST offer a VIRTIO_PCI_CAP_PCI_CFG value? Something like:
>
> Drivers MUST offer at least one capability of each type.
Exactly. Except
> Thanks,
> Rusty.
