On 27.02.2025 15:36, Alejandro Vallejo wrote:
> On Wed Feb 26, 2025 at 2:05 PM GMT, Jan Beulich wrote:
>> On 24.02.2025 15:49, Alejandro Vallejo wrote:
>>> Open question to whoever reviews this...
>>>
>>> On Mon Feb 24, 2025 at 1:27 PM GMT, Alejandro Vallejo wrote:
>>>> spin_lock(&heap_lock);
>>>> - /* adjust domain outstanding pages; may not go negative */
>>>> - dom_before = d->outstanding_pages;
>>>> - dom_after = dom_before - pages;
>>>> - BUG_ON(dom_before < 0);
>>>> - dom_claimed = dom_after < 0 ? 0 : dom_after;
>>>> - d->outstanding_pages = dom_claimed;
>>>> - /* flag accounting bug if system outstanding_claims would go negative
>>>> */
>>>> - sys_before = outstanding_claims;
>>>> - sys_after = sys_before - (dom_before - dom_claimed);
>>>> - BUG_ON(sys_after < 0);
>>>> - outstanding_claims = sys_after;
>>>> + BUG_ON(outstanding_claims < d->outstanding_pages);
>>>> + if ( pages > 0 && d->outstanding_pages < pages )
>>>> + {
>>>> + /* `pages` exceeds the domain's outstanding count. Zero it out. */
>>>> + outstanding_claims -= d->outstanding_pages;
>>>> + d->outstanding_pages = 0;
>>>
>>> While this matches the previous behaviour, do we _really_ want it? It's
>>> weird,
>>> quirky, and it hard to extend to NUMA-aware claims (which is something in
>>> midway through).
>>>
>>> Wouldn't it make sense to fail the allocation (earlier) if the claim has run
>>> out? Do we even expect this to ever happen this late in the allocation call
>>> chain?
>>
>> This goes back to what a "claim" means. Even without any claim, a domain may
>> allocate memory. So a claim having run out doesn't imply allocation has to
>> fail.
>
> Hmmm... but that violates the purpose of the claim infra as far as I
> understand
> it. If a domain may overallocate by (e.g) ballooning in memory it can distort
> the
> ability of another domain to start up, even if it succeeded in its own claim.
Why would that be? As long as we hold back enough memory to cover the claim, it
shouldn't matter what kind of allocation we want to process. I'd say that a PV
guest starting ballooned ought to be able to deflate its balloon as far as
there was a claim established for it up front.
> We might also break the invariant that total claims are strictly >=
> total_avail_pages.
Same here - I don't see why this would happen as long as all accounting is
working correctly.
> I'm somewhat puzzled at the "why" of having separate concepts for max_mem and
> claims. I guess it simply grew the way it did. Reinstating sanity would
> probably involve making max_mem effectively the claim, but that's a ton of
> work I really would rather not do for now.
To me the two are different (beyond claim being global while max-mem is per-
domain). max-mem is a hard boundary (beyond which allocations _will_ fail),
whereas a claim is a softer one, beyond which allocations _may_ fail.
Jan
