Re: [Xen-devel] [PATCH v2 0/2] Remove PAGE_SIZE from public headers

[Stefano Stabellini <sstabellini@xxxxxxxxxx>]

On Thu, 20 Oct 2016, Andrew Cooper wrote:
> On 20/10/16 20:09, Stefano Stabellini wrote:
> > On Thu, 20 Oct 2016, Juergen Gross wrote:
> >> On 19/10/16 21:21, stefano@xxxxxxxxxxx wrote:
> >>> Reference to PAGE_SIZE slipped in to two public header files. QEMU build 
> >>> on
> >>> ARM64 is broken by this. PAGE_SIZE should not be used because it could
> >>> be undefined or it could be defined differently on different operating
> >>> systems.
> >>>
> >>> Stefano Stabellini (2):
> >>>       usbif.h: replace PAGE_SIZE with USBIF_RING_SIZE
> >>>       vscsiif.h: replace PAGE_SIZE with VSCSIIF_PAGE_SIZE
> >>>
> >>>  xen/include/public/io/usbif.h   | 5 +++--
> >>>  xen/include/public/io/vscsiif.h | 3 ++-
> >>>  2 files changed, 5 insertions(+), 3 deletions(-)
> >>>
> >> In principle I'm okay with these changes. But wouldn't it be
> >> better to add
> >>
> >> #define XEN_RING_PAGE_SIZE 4096
> >>
> >> to xen/include/public/io/ring.h instead of having each interface
> >> file its own definition?
> > I would prefer to avoid having to spell out the page size for all the io
> > interfaces, because the same interfaces should work fine with other page
> > granularities, if the grant table hypercalls supported them.
> >
> > However in reality frontend and backend would have to agree on the page
> > granularity to use beforehand. Some sort of consensus protocol is
> > required. When that protocol is introduced, we could get rid of
> > #define XEN_RING_PAGE_SIZE 4096.
> >
> > What do people think about this?
> 
> Not all current rings are 4k in size, and not all future rings should be
> implicitly restricted to 4k.
> 
> -1 to this introducing XEN_RING_PAGE_SIZE.  It will be more confusing
> than helpful.

Indeed, you are right. Since you are at it, feel like acking the current
patches? :-)

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel