On 19.02.2026 12:03, Juergen Gross wrote:
> On 19.02.26 11:52, Jan Beulich wrote:
>> On 19.02.2026 00:04, Jason Andryuk wrote:
>>> On 2026-02-18 14:08, Daniel P. Smith wrote:
>>>> --- a/xen/common/domain.c
>>>> +++ b/xen/common/domain.c
>>>> @@ -210,7 +210,7 @@ static void set_domain_state_info(struct
>>>> xen_domctl_get_domain_state *info,
>>>> int get_domain_state(struct xen_domctl_get_domain_state *info, struct
>>>> domain *d,
>>>> domid_t *domid)
>>>> {
>>>> - unsigned int dom;
>>>> + unsigned int dom = -1;
>>>> int rc = -ENOENT;
>>>> struct domain *hdl;
>>>>
>>>> @@ -219,6 +219,10 @@ int get_domain_state(struct
>>>> xen_domctl_get_domain_state *info, struct domain *d,
>>>>
>>>> if ( d )
>>>> {
>>>> + rc = xsm_get_domain_state(XSM_XS_PRIV, d);
>>>> + if ( rc )
>>>> + return rc;
>>>> +
>>>> set_domain_state_info(info, d);
>>>>
>>>> return 0;
>>>> @@ -238,28 +242,39 @@ int get_domain_state(struct
>>>> xen_domctl_get_domain_state *info, struct domain *d,
>>>
>>> Between the two hunks is this:
>>>
>>> hdl = lock_dom_exc_handler();
>>>
>>> /*
>>> * Only domain registered for VIRQ_DOM_EXC event is allowed to query
>>> * domains having changed state.
>>> */
>>> if ( current->domain != hdl )
>>> {
>>> rc = -EACCES;
>>> goto out;
>>> }
>>>
>>> So it is only the domain with VIRQ_DOM_EXC that can enter the while loop:
>>>
>>>>
>>>> while ( dom_state_changed )
>>>> {
>>>> - dom = find_first_bit(dom_state_changed, DOMID_MASK + 1);
>>>> + dom = find_next_bit(dom_state_changed, DOMID_MASK + 1, dom + 1);
>>>> if ( dom >= DOMID_FIRST_RESERVED )
>>>> break;
>>>> +
>>>> + d = rcu_lock_domain_by_id(dom);
>>>> + if ( d && xsm_get_domain_state(XSM_XS_PRIV, d) )
>>>
>>> ... if the VIRQ_DOM_EXC owner is denied for domain d ...
>>>
>>>> + {
>>>> + rcu_unlock_domain(d);
>>>> + d = NULL;
>>>> + continue;
>>>
>>> ... the caller would continue ...
>>>
>>>> + }
>>>> +
>>>> if ( test_and_clear_bit(dom, dom_state_changed) )
>>>
>>> ... and this bit would never be cleared. Should the VIRQ_DOM_EXC owner
>>> always get to clear the bit even if it cannot see the result?
>>
>> I don't think so, no. Whenever a legitimate consumer occurs (the owner of
>> VIRQ_DOM_EXC can change, after all), it'll then consume the bits as needed.
>> More generally, I think we're better off not making the code here depend
>> too much on that special VIRQ_DOM_EXC property.
>
> OTOH a new VIRQ_DOM_EXC owner will result in a complete reset of the bitmap
> anyway (that is: the bits for all existing domains will be set, while all
> others will be cleared).
Yes, while writing my reply I wondered whether I should mention that. To keep
things a little more simple, I didn't. Plus for this aspect the last sentence
of my earlier reply also applies.
Jan
