On 14.10.2022 11:28, Henry Wang wrote:
>> -----Original Message-----
>> From: Jan Beulich <jbeul...@suse.com>
>>
>> On 14.10.2022 10:09, Henry Wang wrote:
>>> @@ -1736,6 +1736,17 @@ void p2m_final_teardown(struct domain *d)
>>> if ( !p2m->domain )
>>> return;
>>>
>>> + if ( !page_list_empty(&p2m->pages) )
>>> + p2m_teardown(d, false);
>>> +
>>> + if ( d->arch.paging.p2m_total_pages != 0 )
>>> + {
>>> + spin_lock(&d->arch.paging.lock);
>>> + p2m_set_allocation(d, 0, NULL);
>>> + spin_unlock(&d->arch.paging.lock);
>>> + ASSERT(d->arch.paging.p2m_total_pages == 0);
>>> + }
>>
>> Is it intentional to largely open-code p2m_teardown_allocation() here?
>
> Yes, AFAICT p2m_teardown_allocation() is preemptible and we don't want
> any preemption here.
Well, this can be dealt with by adding a parameter to the function, or
by looping over it until it returns other than -ERESTART. Both would
seem better to me than this duplication of functionality (but I'm not
a maintainer of this code, as you know).
Jan
