On 8/6/26 3:36 AM, Jan Beulich wrote:
On 06.08.2026 03:10, Daniel P. Smith wrote:
On 7/28/26 9:23 AM, Jan Beulich wrote:
--- a/xen/xsm/flask/hooks.c
+++ b/xen/xsm/flask/hooks.c
@@ -1022,14 +1022,9 @@ static char *cf_check flask_show_irq_sid
#ifdef CONFIG_HAS_PIRQ
-static int cf_check flask_map_domain_pirq(struct domain *d)
+static int cf_check flask_map_domain_pirq(struct domain *d, bool access)
{
- return current_has_perm(d, SECCLASS_RESOURCE, RESOURCE__ADD);
-}
-
-static int cf_check flask_unmap_domain_pirq(struct domain *d)
-{
- return current_has_perm(d, SECCLASS_RESOURCE, RESOURCE__REMOVE);
+ return current_has_perm(d, SECCLASS_RESOURCE, resource_to_perm(access));
}
#endif /* CONFIG_HAS_PIRQ */
I am not opposed to collapsing the calls as long as the semantic is not
lost, which I feel the reuse of the xsm_map_domain_pirq does looses it
much less provides an opportunity for confusion. Something like
xsm_domain_pirq(..., access) makes more semantic sense to me, as it
would read, grant domain pirq access T/F.
In fact I was as well wondering about naming (and a possible name change)
here. I decided against it because of the other hooks (touched by patch
19), which all follow a similar model of their names really more expressing
the "positive" form than the "negative" one. Further, entirely losing "map"
from the name also doesn't look quite right, as physdev_{,un}map_pirq() is
where they're called from. To follow the naming of some of the other hook,
maybe xsm_domain_pirq_mapping(..., bool map)? Possibly even with "domain"
dropped from the name, fully fitting xsm_io{mem,port}_mapping()? Which
would then further raise the question whether patch 19 should maybe rename
the parameters of those two hooks to "map" at the same time.
Sound like you and I were running through the same argument with
ourselves, except you went right and I went left.
For me, I find the false case of xsm_map_domain_pirq(..., access) reads
a bit weird, "Can the target map domain pirq access=false?" I do like
your mapping suggestion along with the parameter rename to map. It
changes the reading to "Is target allowed map(true)/unmap(false) for the
domain pirq mapping?" This could be carried through to the other cases.
v/r,
dps
