On 14.12.2021 17:26, Oleksandr wrote:
> On 14.12.21 15:37, Jan Beulich wrote:
>> On 03.12.2021 21:33, Oleksandr Tyshchenko wrote:
>>> @@ -2177,14 +2181,22 @@ void *alloc_xenheap_pages(unsigned int order,
>>> unsigned int memflags)
>>>
>>> void free_xenheap_pages(void *v, unsigned int order)
>>> {
>>> + struct page_info *pg;
>>> + unsigned int i;
>>> +
>>> ASSERT(!in_irq());
>>>
>>> if ( v == NULL )
>>> return;
>>>
>>> + pg = virt_to_page(v);
>>> +
>>> memguard_guard_range(v, 1 << (order + PAGE_SHIFT));
>> ... this really want to (logically) move into the new arch hooks.
>> That'll effectively mean to simply drop the Arm stubs afaict (and I
>> notice there's some dead code there on x86, which I guess I'll make
>> a patch to clean up). But first of all this suggests that you want
>> to call the hooks with base page and order, putting the loops there.
>
> I see your point and agree ... However I see the on-list patches that
> remove common memguard_* invocations and x86 bits.
> So I assume, this request is not actual anymore, or I still need to pass
> an order to new arch hooks? Please clarify.
Well, that patch (really just the Arm one) effectively takes care of
part of what I did say above. Irrespective I continue to think that
the hook should take a (page,order) tuple instead of getting invoked
once for every order-0 page. And the hook invocations should be placed
such that they could fulfill the (being removed) memguard function
(iirc that was already the case, at least mostly).
Jan
