Re: [PATCH] virtio-pci: alloc only resources actually used.

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

On Thu, Jun 18, 2015 at 10:14:40AM +0200, Gerd Hoffmann wrote:
>   Hi,
> 
> > > +static struct resource *request_capability(struct pci_dev *dev, int off,
> > > +					   const char *name)
> > > +{
> > > +	u8 bar;
> > > +	u32 offset, length;
> > > +
> > > +	pci_read_config_byte(dev, off + offsetof(struct virtio_pci_cap,
> > > +						 bar),
> > > +			     &bar);
> > > +	pci_read_config_dword(dev, off + offsetof(struct virtio_pci_cap, offset),
> > > +			     &offset);
> > > +	pci_read_config_dword(dev, off + offsetof(struct virtio_pci_cap, length),
> > > +			      &length);
> > > +
> > > +	return request_mem_region(pci_resource_start(dev, bar) + offset,
> > > +				  length, name);
> > > +}
> > > +
> > 
> > For device config, this might request too much. The spec says:
> > 	The drivers SHOULD only map part of configuration structure large enough
> > 	for device operation.
> 
> We don't map it here though.  We just reserve what belongs to virtio
> according to the capabilities.
> 
> > I think you should limit this to PAGE_SIZE like we do for map_capability.
> 
> notify is much larger than PAGE_SIZE.

Yes - I mean for the device config.

> > >  	err = -EINVAL;
> > >  	vp_dev->common = map_capability(pci_dev, common,
> > >  					sizeof(struct virtio_pci_common_cfg), 4,
> > 
> > map_capability has a bunch of checks in place to validate the capability
> > structure. With request_capability called earlier, they are now done too
> > late.
> 
> Hmm, lets move the checks to find_capability then?
> 
> cheers,
>   Gerd
>