OASIS Mailing List ArchivesView the OASIS mailing list archive below
or browse/search using MarkMail.


Help: OASIS Mailing Lists Help | MarkMail Help

virtio-dev message

[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]

Subject: Re: [virtio] [PATCH v2] initialization: add extra device status handshake

On Wed, Sep 25, 2013 at 10:18:10AM +0200, Paolo Bonzini wrote:
> Il 25/09/2013 09:49, Michael S. Tsirkin ha scritto:
> >> > +
> >> > +8. The DRIVER_OK status bit is set.  This is simply informational, 
> >> > +   as the device may already be active at this point.
> > 
> > So we have a separate issue I think: there's lots
> > of device specific configuration that really
> > would benefit from "OK now go" bit.
> > 
> > For example, we register network device and we
> > really don't want e.g. config interrupts before that is done.
> > 
> > Also qemu relied on DRIVER_OK to kick off vhost net and
> > vhost scsi.
> True, but that is wrong.  The virtio-scsi driver had to be modified to
> support vhost-scsi, because it used the virtqueues before setting
> DRIVER_OK and this didn't work with vhost-scsi.  This was ugly.
> The right thing to do would be to start vhost upon the driver setting

With current proposal, this makes things very hard for the
implementation: we can't support this easily because rings are not yet
setup (not even sized) on FEATURES_OK.
We really want a flag that says "rings are now setup".

> Failure to start vhost (which is fatal for vhost-scsi,
> since there is no userspace fallback unlike vhost-net) could be reported
> by not acknowledging FEATURES_OK.
> Paolo

I think what you are asking for really, is
DEVICE_OK flag that locks both features and rings
and enables device.

It would have to be set from driver quite late in life.


[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]