Re: [PATCH] virtio: Define and use per-architecture "pfn shift" constants

[Hollis Blanchard <hollisb@us.ibm.com>]

On Thu, 2008-11-06 at 10:49 +0000, Mark McLoughlin wrote:
> On Wed, 2008-11-05 at 22:49 -0600, Hollis Blanchard wrote:
> 
> > -       info->queue = kzalloc(PAGE_ALIGN(vring_size(num,PAGE_SIZE)), GFP_KERNEL);
> > +       vring_bytes = PAGE_ALIGN(vring_size(num, VRING_PAGE_SIZE));
> > +       info->queue = kzalloc(vring_bytes, GFP_KERNEL);
> 
> You're still aligning the size to PAGE_SIZE rather than VRING_PAGE_SIZE?

The original code was calling kzalloc with a multiple of PAGE_SIZE, so I
kept that behavior here...

> But actually, why do we align the size anyway?

I assume it's so that the last page in the ring (containing the "used"
fields) could be safely mapped into another guest's address space,
without fear of exposing other data.

I don't know how valuable that is, but that's not really my concern so I
preserved the behavior.

> Also might make sense for vring_init() and vring_size() not to take a
> pagesize argument and hard-code them to use VRING_PAGE_SIZE.

I think that's a good idea. Anthony mentioned earlier the code was done
this way so it could be copied to userspace, where PAGE_SIZE is
unavailable, but that isn't an issue if we switch to VRING_PAGE_SIZE.

-- 
Hollis Blanchard
IBM Linux Technology Center