On Fri, 2006-05-19 at 17:31 +0900, Isaku Yamahata wrote:
> page = mfn_to_page(mfn);
> ret = get_page(page, page_get_owner(page));
> BUG_ON(ret == 0);
> - assign_domain_page_replace(d, gpaddr, mfn, flags);
> -
> +
> + arflags = _PAGE_AR_RWX;
> + if (flags & GNTMAP_readonly) {
> + arflags = _PAGE_AR_R;
> + }
> + assign_domain_page_replace(d, gpaddr, mfn, arflags);
> return GNTST_okay;
> }
I think we're still parsing flags too high up in the stack. Couldn't
we pass flags "as is" to assign_domain_page_replace(), then do
arflags = (flags & GNTMAP_readonly) ? _PAGE_AR_R : _PAGE_AR_RWX;
in that end function? It also seems more logical to call
__assign_domain_page() as:
__assign_domain_page(d, j, io_ranges[i].type, GNTMAP_readonly /* or 0 for RWX
*/)
than to pass in explicit page flags. Doing these, we could get rid of
the BUG_ON checks. For example:
__assign_domain_page(struct domain *d,
unsigned long mpaddr, unsigned long physaddr,
unsigned long flags)
{
pte_t *pte;
unsigned long arflags = (flags & GNTMAP_readonly) ? _PAGE_AR_R :
_PAGE_AR_RWX;
Thoughts? I think there are only a couple minor places in patches 2 and
3 that would need to be updated to reflect this. Thanks,
Alex
--
Alex Williamson HP Linux & Open Source Lab
_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel
|
Home