Re: [virtio-dev] Re: [PATCH v1] content: support SR-IOV

[Tiwei Bie <tiwei.bie@intel.com>]

On Tue, May 22, 2018 at 04:55:05AM +0300, Michael S. Tsirkin wrote:
> On Tue, May 22, 2018 at 09:28:22AM +0800, Tiwei Bie wrote:
> > On Tue, May 22, 2018 at 01:32:29AM +0300, Michael S. Tsirkin wrote:
> > > On Mon, May 21, 2018 at 06:52:06PM +0800, Tiwei Bie wrote:
[...]
> > > >  If VIRTIO_F_IN_ORDER has been negotiated, a device MUST use
> > > >  buffers in the same order in which they have been available.
> > > >  
> > > > +A device SHOULD offer VIRTIO_F_SR_IOV if it presents a PCI
> > > > +SR-IOV capability structure.  A device MAY fail to operate
> > > > +further if VIRTIO_F_SR_IOV is not accepted.
> > > 
> > > Why is the last sentence here? What are you trying to accomplish?  It
> > > seems that if we allow drivers not to accept the bit, we should require
> > > the devices to function without it or at least not go out of our way to
> > > say they do not have to.
> > 
> > The last sentence "A device MAY fail to operate further
> > if VIRTIO_F_SR_IOV is not accepted" means if the driver
> > doesn't accept this feature, the device may fail to work,
> > e.g. setting VIRTIO_CONFIG_S_FEATURES_OK may fail. And
> > in this case (i.e. the device fails to work), it will be
> > necessary for driver to accept this feature in order to
> > work with this device.
> > 
> > So it's not to allow drivers not to accept the feature,
> > instead, it just tells device's behaviour (i.e. device
> > may fail to work) if this feature isn't accepted. And
> > it leaves the choice to device's implementation.
> 
> The choice is always there. However the preferred response
> to features is for drivers only use what they actually
> need, and devices to try to accomodate subsets of features.
> 
> Is there a reason for this feature to be different?
> 
> 
> > It's similar to the description for the IOMMU_PLATFORM
> > feature:
> > 
> > """
> > A device SHOULD offer VIRTIO_F_IOMMU_PLATFORM if it is behind an IOMMU that
> > translates bus addresses from the device into physical addresses in memory.
> > A device MAY fail to operate further if VIRTIO_F_IOMMU_PLATFORM is not
> > accepted.
> > """
> 
> Well IOMMU platform is very special in that it's a security feature.
> We really can't support legacy drivers with these devices.
> Most features aren't like this.
> 
> Something like this would be appropriate for the barrier
> feature.
> 
> But this specific feature I don't see why can't driver just use
> the PF.

Okay. I'll remove the last sentence.

Best regards,
Tiwei Bie