Re: [PATCH v2] virtio: Add polling virtio-mmio device reset completion

["Michael S. Tsirkin" <mst@redhat.com>]

On Fri, Sep 27, 2024 at 12:08:34PM +0800, Jason Wang wrote:
> On Thu, Sep 26, 2024 at 2:09 PM Michael S. Tsirkin <mst@redhat.com> wrote:
> >
> > On Thu, Sep 26, 2024 at 11:21:25AM +0800, Jason Wang wrote:
> > > On Thu, Sep 26, 2024 at 10:27 AM Jianguo Sun <quic_jianguos@quicinc.com> wrote:
> > > >
> > > > We write 0 to device_status to initiate reset of a virtio-mmio device.
> > > > The reset operation itself may or may not be completed by the time
> > > > write instruction completes. Add polling device_status to return
> > > > 0 to ensure reset completion before reinitializing the device.
> > > >
> > > > Change-Id: If15d11d090dfd0d4972ad35f49af03e076872413
> > > > Signed-off-by: Jianguo Sun <quic_jianguos@quicinc.com>
> > > > Signed-off-by: Sreenad Menon <quic_sreemeno@quicinc.com>
> > > > Signed-off-by: Anant Goel <quic_anantg@quicinc.com>
> > > > ---
> > > >  drivers/virtio/virtio_mmio.c | 6 ++++++
> > > >  1 file changed, 6 insertions(+)
> > > >
> > > > diff --git a/drivers/virtio/virtio_mmio.c b/drivers/virtio/virtio_mmio.c
> > > > index 90e784e7b721..cf1a4a3dd35d 100644
> > > > --- a/drivers/virtio/virtio_mmio.c
> > > > +++ b/drivers/virtio/virtio_mmio.c
> > > > @@ -70,6 +70,7 @@
> > > >  #include <linux/virtio_config.h>
> > > >  #include <uapi/linux/virtio_mmio.h>
> > > >  #include <linux/virtio_ring.h>
> > > > +#include <linux/delay.h>
> > > >
> > > >
> > > >
> > > > @@ -269,6 +270,11 @@ static void vm_reset(struct virtio_device *vdev)
> > > >
> > > >         /* 0 status means a reset. */
> > > >         writel(0, vm_dev->base + VIRTIO_MMIO_STATUS);
> > > > +       /* After writing 0 to device_status, the driver MUST wait for a read of
> > > > +        * device_status to return 0 before reinitializing the device.
> > > > +        */
> > >
> > > Is this copied from the spec? I can only see a similar description in
> > > the PCI but not MMIO part.
> >
> > Not in spec. We can add something like this in spec, but it
> > can not be a MUST, has to be a SHOULD.
> >
> > > > +       while (readl(vm_dev->base + VIRTIO_MMIO_STATUS))
> > > > +               msleep(1);
> > >
> > > Again, we probably need a feature bit to unbreak existing MMIO devices?
> > >
> > > THanks
> >
> > We can do this unconditionally since legacy devices return 0
> > immediately,
> 
> The problem is that we can't audit all legacy device implementations.
> Current driver still works if a device doesn't reset status to 0.

These would be out of spec though, would they not?
The reset value for the register is 0, and it is R/W.


> > feature negotiation happens after reset, won't be useful
> > here.
> 
> Right, as said in another thread, it could be mmio specific flag or version.
> 
> Thanks
> 
> >
> >
> > > >  }
> > > >
> > > >
> > > > --
> > > > 2.17.1
> > > >
> >