Re: [PATCH v4 2/5] mm: LARGE_ANON_FOLIO for improved performance

From: Ryan Roberts
Date: Tue Aug 08 2023 - 12:46:37 EST


On 07/08/2023 19:10, Zi Yan wrote:
> On 7 Aug 2023, at 13:45, Ryan Roberts wrote:
>
>> On 05/08/2023 03:50, Yin, Fengwei wrote:
>>>
>>>
>>> On 8/5/2023 5:58 AM, Zi Yan wrote:
>>>> On 4 Aug 2023, at 17:30, David Hildenbrand wrote:
>>>>
>>>>> On 04.08.23 23:26, Yu Zhao wrote:
>>>>>> On Fri, Aug 4, 2023 at 3:13 PM David Hildenbrand <david@xxxxxxxxxx> wrote:
>>>>>>>
>>>>>>> On 04.08.23 23:00, Yu Zhao wrote:
>>>>>>>> On Fri, Aug 4, 2023 at 2:23 PM David Hildenbrand <david@xxxxxxxxxx> wrote:
>>>>>>>>>
>>>>>>>>> On 04.08.23 10:27, Ryan Roberts wrote:
>>>>>>>>>> On 04/08/2023 00:50, Yu Zhao wrote:
>>>>>>>>>>> On Thu, Aug 3, 2023 at 6:43 AM Ryan Roberts <ryan.roberts@xxxxxxx> wrote:
>>>>>>>>>>>>
>>>>>>>>>>>> + Kirill
>>>>>>>>>>>>
>>>>>>>>>>>> On 26/07/2023 10:51, Ryan Roberts wrote:
>>>>>>>>>>>>> Introduce LARGE_ANON_FOLIO feature, which allows anonymous memory to be
>>>>>>>>>>>>> allocated in large folios of a determined order. All pages of the large
>>>>>>>>>>>>> folio are pte-mapped during the same page fault, significantly reducing
>>>>>>>>>>>>> the number of page faults. The number of per-page operations (e.g. ref
>>>>>>>>>>>>> counting, rmap management lru list management) are also significantly
>>>>>>>>>>>>> reduced since those ops now become per-folio.
>>>>>>>>>>>>>
>>>>>>>>>>>>> The new behaviour is hidden behind the new LARGE_ANON_FOLIO Kconfig,
>>>>>>>>>>>>> which defaults to disabled for now; The long term aim is for this to
>>>>>>>>>>>>> defaut to enabled, but there are some risks around internal
>>>>>>>>>>>>> fragmentation that need to be better understood first.
>>>>>>>>>>>>>
>>>>>>>>>>>>> When enabled, the folio order is determined as such: For a vma, process
>>>>>>>>>>>>> or system that has explicitly disabled THP, we continue to allocate
>>>>>>>>>>>>> order-0. THP is most likely disabled to avoid any possible internal
>>>>>>>>>>>>> fragmentation so we honour that request.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Otherwise, the return value of arch_wants_pte_order() is used. For vmas
>>>>>>>>>>>>> that have not explicitly opted-in to use transparent hugepages (e.g.
>>>>>>>>>>>>> where thp=madvise and the vma does not have MADV_HUGEPAGE), then
>>>>>>>>>>>>> arch_wants_pte_order() is limited to 64K (or PAGE_SIZE, whichever is
>>>>>>>>>>>>> bigger). This allows for a performance boost without requiring any
>>>>>>>>>>>>> explicit opt-in from the workload while limitting internal
>>>>>>>>>>>>> fragmentation.
>>>>>>>>>>>>>
>>>>>>>>>>>>> If the preferred order can't be used (e.g. because the folio would
>>>>>>>>>>>>> breach the bounds of the vma, or because ptes in the region are already
>>>>>>>>>>>>> mapped) then we fall back to a suitable lower order; first
>>>>>>>>>>>>> PAGE_ALLOC_COSTLY_ORDER, then order-0.
>>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> ...
>>>>>>>>>>>>
>>>>>>>>>>>>> +#define ANON_FOLIO_MAX_ORDER_UNHINTED \
>>>>>>>>>>>>> + (ilog2(max_t(unsigned long, SZ_64K, PAGE_SIZE)) - PAGE_SHIFT)
>>>>>>>>>>>>> +
>>>>>>>>>>>>> +static int anon_folio_order(struct vm_area_struct *vma)
>>>>>>>>>>>>> +{
>>>>>>>>>>>>> + int order;
>>>>>>>>>>>>> +
>>>>>>>>>>>>> + /*
>>>>>>>>>>>>> + * If THP is explicitly disabled for either the vma, the process or the
>>>>>>>>>>>>> + * system, then this is very likely intended to limit internal
>>>>>>>>>>>>> + * fragmentation; in this case, don't attempt to allocate a large
>>>>>>>>>>>>> + * anonymous folio.
>>>>>>>>>>>>> + *
>>>>>>>>>>>>> + * Else, if the vma is eligible for thp, allocate a large folio of the
>>>>>>>>>>>>> + * size preferred by the arch. Or if the arch requested a very small
>>>>>>>>>>>>> + * size or didn't request a size, then use PAGE_ALLOC_COSTLY_ORDER,
>>>>>>>>>>>>> + * which still meets the arch's requirements but means we still take
>>>>>>>>>>>>> + * advantage of SW optimizations (e.g. fewer page faults).
>>>>>>>>>>>>> + *
>>>>>>>>>>>>> + * Finally if thp is enabled but the vma isn't eligible, take the
>>>>>>>>>>>>> + * arch-preferred size and limit it to ANON_FOLIO_MAX_ORDER_UNHINTED.
>>>>>>>>>>>>> + * This ensures workloads that have not explicitly opted-in take benefit
>>>>>>>>>>>>> + * while capping the potential for internal fragmentation.
>>>>>>>>>>>>> + */
>>>>>>>>>>>>> +
>>>>>>>>>>>>> + if ((vma->vm_flags & VM_NOHUGEPAGE) ||
>>>>>>>>>>>>> + test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags) ||
>>>>>>>>>>>>> + !hugepage_flags_enabled())
>>>>>>>>>>>>> + order = 0;
>>>>>>>>>>>>> + else {
>>>>>>>>>>>>> + order = max(arch_wants_pte_order(), PAGE_ALLOC_COSTLY_ORDER);
>>>>>>>>>>>>> +
>>>>>>>>>>>>> + if (!hugepage_vma_check(vma, vma->vm_flags, false, true, true))
>>>>>>>>>>>>> + order = min(order, ANON_FOLIO_MAX_ORDER_UNHINTED);
>>>>>>>>>>>>> + }
>>>>>>>>>>>>> +
>>>>>>>>>>>>> + return order;
>>>>>>>>>>>>> +}
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> Hi All,
>>>>>>>>>>>>
>>>>>>>>>>>> I'm writing up the conclusions that we arrived at during discussion in the THP
>>>>>>>>>>>> meeting yesterday, regarding linkage with exiting THP ABIs. It would be great if
>>>>>>>>>>>> I can get explicit "agree" or disagree + rationale from at least David, Yu and
>>>>>>>>>>>> Kirill.
>>>>>>>>>>>>
>>>>>>>>>>>> In summary; I think we are converging on the approach that is already coded, but
>>>>>>>>>>>> I'd like confirmation.
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> The THP situation today
>>>>>>>>>>>> -----------------------
>>>>>>>>>>>>
>>>>>>>>>>>> - At system level: THP can be set to "never", "madvise" or "always"
>>>>>>>>>>>> - At process level: THP can be "never" or "defer to system setting"
>>>>>>>>>>>> - At VMA level: no-hint, MADV_HUGEPAGE, MADV_NOHUGEPAGE
>>>>>>>>>>>>
>>>>>>>>>>>> That gives us this table to describe how a page fault is handled, according to
>>>>>>>>>>>> process state (columns) and vma flags (rows):
>>>>>>>>>>>>
>>>>>>>>>>>> | never | madvise | always
>>>>>>>>>>>> ----------------|-----------|-----------|-----------
>>>>>>>>>>>> no hint | S | S | THP>S
>>>>>>>>>>>> MADV_HUGEPAGE | S | THP>S | THP>S
>>>>>>>>>>>> MADV_NOHUGEPAGE | S | S | S
>>>>>>>>>>>>
>>>>>>>>>>>> Legend:
>>>>>>>>>>>> S allocate single page (PTE-mapped)
>>>>>>>>>>>> LAF allocate lage anon folio (PTE-mapped)
>>>>>>>>>>>> THP allocate THP-sized folio (PMD-mapped)
>>>>>>>>>>>>> fallback (usually because vma size/alignment insufficient for folio)
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> Principles for Large Anon Folios (LAF)
>>>>>>>>>>>> --------------------------------------
>>>>>>>>>>>>
>>>>>>>>>>>> David tells us there are use cases today (e.g. qemu live migration) which use
>>>>>>>>>>>> MADV_NOHUGEPAGE to mean "don't fill any PTEs that are not explicitly faulted"
>>>>>>>>>>>> and these use cases will break (i.e. functionally incorrect) if this request is
>>>>>>>>>>>> not honoured.
>>>>>>>>>>>
>>>>>>>>>>> I don't remember David saying this. I think he was referring to UFFD,
>>>>>>>>>>> not MADV_NOHUGEPAGE, when discussing what we need to absolutely
>>>>>>>>>>> respect.
>>>>>>>>>>
>>>>>>>>>> My understanding was that MADV_NOHUGEPAGE was being applied to regions *before*
>>>>>>>>>> UFFD was being registered, and the app relied on MADV_NOHUGEPAGE to not back any
>>>>>>>>>> unfaulted pages. It's not completely clear to me how not honouring
>>>>>>>>>> MADV_NOHUGEPAGE would break things though. David?
>>>>>>>>>
>>>>>>>>> Sorry, I'm still lagging behind on some threads.
>>>>>>>>>
>>>>>>>>> Imagine the following for VM postcopy live migration:
>>>>>>>>>
>>>>>>>>> (1) Set MADV_NOHUGEPAGE on guest memory and discard all memory (e.g.,
>>>>>>>>> MADV_DONTNEED), to start with a clean slate.
>>>>>>>>> (2) Migrates some pages during precopy from the source and stores them
>>>>>>>>> into guest memory on the destination. Some of the memory locations
>>>>>>>>> will have pages populated.
>>>>>>>>> (3) At some point, decide to enable postcopy: enable userfaultfd on
>>>>>>>>> guest memory.
>>>>>>>>> (4) Discard *selected* pages again that have been dirtied in the
>>>>>>>>> meantime on the source. These are pages that have been migrated
>>>>>>>>> previously.
>>>>>>>>> (5) Start running the VM on the destination.
>>>>>>>>> (6) Anything that's not populated will trigger userfaultfd missing
>>>>>>>>> faults. Then, you can request them from the source and place them.
>>>>>>>>>
>>>>>>>>> Assume you would populate more than required during 2), you can end up
>>>>>>>>> not getting userfaultfd faults during 4) and corrupt your guest state.
>>>>>>>>> It works if during (2) you migrated all guest memory, or if during 4)
>>>>>>>>> you zap everything that still needs migr
>>>>>>>>
>>>>>>>> I see what you mean now. Thanks.
>>>>>>>>
>>>>>>>> Yes, in this case we have to interpret MADV_NOHUGEPAGE as nothing >4KB.
>>
>> I'm glad we have agreement on this.
>>
>> In some threads Yu has been talking about this series in the short term, vs long
>> term roadmap; so to be clear, I interpret this as meaning we must consider that
>> MADV_NOHUGEPAGE means nothing bigger than order-0 both in the context of this
>> series and for the long term - that's behavior that user space depends upon.
>>
>> I think we should also apply the same logic to system/process THP mode =
>> "never", even if the vma does not have MADV_NOHUGEPAGE. If the user has
>> explicitly set "never" on the system or process, that means "nothing bigger than
>> order-0". Shout if you disagree.
>>
>>>>>>>
>>>>>>> Note that it's still even unclear to me why we want to *not* call these
>>>>>>> things THP. It would certainly make everything less confusing if we call
>>>>>>> them THP, but with additional attributes.
>>
>> I think I've stated in the past that I don't have a strong opinion on what we
>> call them. But I do think you make a convincing argument for calling them after
>> THP. Regardless, I'd rather agree on a name up front, before this initial series
>> goes in - it's always better to be consistent across all the commit messages and
>> comments to make things more grepable.
>>
>> The only concrete objection I remember hearing to a name with "THP" in the title
>> was that there are stats (meminfo, vmstats, etc) that count THPs and this
>> becomes confusing if those counters now only mean a subset of THPs. But that
>> feels like a small issue in the scheme of things.
>>
>>>>>>>
>>>>>>> I think that is one of the first things we should figure out because it
>>>>>>> also indirectly tells us what all these toggles mean and how/if we
>>>>>>> should redefine them (and if they even apply).
>>>>>>>
>>>>>>> Currently THP == PMD size
>>>>>>>
>>>>>>> In 2016, Hugh already envisioned PUD/PGD THP (see 49920d28781d ("mm:
>>>>>>> make transparent hugepage size public")) when he explicitly exposed
>>>>>>> "hpage_pmd_size". Not "hpage_size".
>>>>>>>
>>>>>>> For hugetlb on arm64 we already support various sizes that are < PMD
>>>>>>> size and *not* call them differently. It's a huge(tlb) page. Sometimes
>>>>>>> we refer to them as cont-PTE hugetlb pages.
>>>>>>>
>>>>>>>
>>>>>>> So, nowadays we do have "PMD-sized THP", someday we might have
>>>>>>> "PUD-sized THP". Can't we come up with a name to describe sub-PMD THP?
>>
>> I think one subtle difference is that these sub-PMD THPs, likely won't always
>> have a single size.
>>
>>>>>>>
>>>>>>> Is it really of value if we invent a new term for them? Yes, I was not
>>>>>>> enjoying "Flexible THP".
>>
>> How about "variable-order THP"? Or "SW THP" vs "HW THP"?
>
> variable-order THP sounds good to me.
>
> One question I have is that although Ryan is only working on sub-PMD THPs,
> do we want to plan for sub-PUD THPs now? Like are sub-PUD THPs variable-order
> THPs? And leave TODOs and comments like "variable-order THPs can be bigger than
> PMD and smaller than PUD in the future"? Maybe sub-PUD THPs are still too far
> to consider it for now. Just think out loud.

I'm not personally planning to do any work here. Such a thing would need similar
but separate implementation IMHO, since you would be working at the PMD level
not the PTE level.

Then there is the question of what the benefit would be. My working assumption
would be that you will not be getting any further HW benefits until its big
enough for a PUD. And the SW costs of allocating such a large contig block would
very likely outweigh the benefits of a few less page faults; you're surely
better off allocating multiple PMD-sized THPs?

>
>
>>
>>>>>>>
>>>>>>>
>>>>>>> Once we figured that out, we should figure out if MADV_HUGEPAGE meant
>>>>>>> "only PMD-sized THP" or anything else?
>>>>>>>
>>>>>>> Also, we can then figure out if MADV_NOHUGEPAGE meant "only PMD-sized
>>>>>>> THP" or anything else?
>>
>> Based on the existing user space expectation that MADV_NOHUGEPAGE means "nothing
>> bigger than order-0" I'm not sure how we could ever decide MADV_NOHUGEPAGE means
>> anything different? This feels set in stone to me.
>>
>>>>>>>
>>>>>>>
>>>>>>> The simplest approach to me would be "they imply any THP, and once we
>>>>>>> need more tunables we might add some", similar to what Kirill also raised.
>>
>> Agreed.
>>
>>>>>>>
>>>>>>>
>>>>>>> Again, it's all unclear to me at this point and I'm happy to hear
>>>>>>> opinions, because I really don't know.
>>>>>>
>>>>>> I agree these points require more discussion. But I don't think we
>>>>>> need to conclude them now, unless they cause correctness issues like
>>>>>> ignoring MADV_NOHUGEPAGE would. My concern is that if we decide to go
>>>>>> with "they imply any THP" and *expose this to userspace now*, we might
>>>>>> regret later.
>>>>>
>>>>> If we don't think they are THP, probably MADV_NOHUGEPAGE should not apply and we should be ready to find other ways to deal with the mess we eventually create. If we want to go down that path, sure.
>>>>>
>>>>> If they are THP, to me there is not really a question if MADV_NOHUGEPAGE applies to them or not. Unless we want to build a confusing piece of software ;)
>>>>
>>>> I think it is good to call them THP, since they are transparent huge (>order-0) pages.
>>>> But the concern is that before we have a reasonable management policy for order>0 &&
>>>> order<9 THPs, mixing them with existing order-9 THP might give user unexpected
>>>> performance outcome. Unless we are sure they will always performance improvement,
>>>> we might repeat the old THP path, namely users begin to disable THP by default
>>>> to avoid unexpected performance hiccup. That is the reason Yu wants to separate
>>>> LAF from THP at the moment.
>>
>> (for the purposes of this; LAF="sub-PMD THP", THP="PMD-size THP", we treat them
>> both as forms of THP)...
>>
>> How about this for a strawman:
>>
>> When introducing LAF we can either use an opt-in or an opt-out model. The opt-in
>> model would require a new ABI from day 1 (something I think there is concensus
>> that we do not want to do) and would prevent apps from automatically getting
>> benefit. So I don't like that model.
>>
>> If going with the opt-out model, we already have an opt-out mechanism
>> (thp="never" and MADV_NOHUGEPAGE) that we can piggyback. But that mechanism
>> doesn't give us all the control we would like for benchmarking/characterizing
>> the interactions between LAF/THP for different workloads. Ideally we need a way
>> to enable THP while keeping LAF disabled and enable LAF while keeping THP disabled.
>>
>> Can we do this with debugfs? I think controls in there can come and go without
>> too much concern about back-compat?
>
> Is debugfs always available on all distros? For system without debugfs, user is
> going to lose control of LAF. IMHO, the two knobs below can live in
> /sys/kernel/mm/transparent_hugepage/ and could be in sync with "enabled" once
> we think LAF is well studied and goes along well with existing PMD THPs,
> namely when setting "always", "madvise", or "never" to "enabled", "laf_enabled"
> is set to the same value.

I really don't want to add any sysfs knobs until we properly understand what we
need/want. I couldn't tell you what the availability of debugfs is like across
all distros though; certainly its available on Ubuntu and userdebug builds of
Android.

Yu suggested another policy [1], which would allow us to disable THP while
keeping LAF enabled (but not the other way around) without having to add any
knobs, so perhaps that's the way to go. It does assume its safe to use LAF when
thp=never though. (Copying here for completeness):

| prctl/fw | sysfs | sysfs | sysfs
| disable | never | madvise | always
----------------|-----------|-----------|-----------|-----------
no hint | S | LAF>S | LAF>S | THP>LAF>S
MADV_HUGEPAGE | S | LAF>S | THP>LAF>S | THP>LAF>S
MADV_NOHUGEPAGE | S | S | S | S

Where "prctl/fw disable" trumps the sysfs setting.


[1] https://lore.kernel.org/linux-mm/20469f02-d62d-d925-3536-d6a1f1099fda@xxxxxxx/

>
>>
>> Perhaps 2 controls:
>>
>> laf_enable=0|1
>> enable/disable LAF independently of THP
>> default=1
>>
>> laf_max_order=N
>> applies to both LAF and THP
>> when max_order < PMD-order, THP acts like thp="never"
>> puts a ceiling on folio order allocated by LAF
>> default=PMD-order
>
> I think it is better to keep it independent of PMD THP. Just make
> laf_max_order can never be bigger than PMD-order. Later, when we understand
> the performance impact of mixing LAF with PMD THP, we can lift this limit
> to allow laf_max_order to be any possible page order.
>
>>
>> This gives:
>>
>>
>> laf_enable=1, laf_max_order=PMD-order (LAF+THP):
>>
>> | never | madvise | always
>> ----------------|-----------|-----------|-----------
>> no hint | S | LAF>S | THP>LAF>S
>> MADV_HUGEPAGE | S | THP>LAF>S | THP>LAF>S
>> MADV_NOHUGEPAGE | S | S | S
>>
>>
>> laf_enable=0, laf_max_order=PMD-order (THP only):
>>
>> | never | madvise | always
>> ----------------|-----------|-----------|-----------
>> no hint | S | S | THP>S
>> MADV_HUGEPAGE | S | THP>S | THP>S
>> MADV_NOHUGEPAGE | S | S | S
>>
>>
>> laf_enable=1, laf_max_order=(PMD-order - 1) (LAF only):
>>
>> | never | madvise | always
>> ----------------|-----------|-----------|-----------
>> no hint | S | LAF>S | LAF>S
>> MADV_HUGEPAGE | S | LAF>S | LAF>S
>> MADV_NOHUGEPAGE | S | S | S
>>
>>
>> This would allow us to get something into the kernel that would allow people to
>> more broadly characterize different workloads under THP, LAF, THP+LAF, which
>> would give us a better understanding of if/how we want to design ABIs for the
>> long term.
>>
>>
>>>>
>>>> Maybe call it THP (experimental) for now and merge it to THP when we have a stable
>>>> policy. For knobs, we might add "any-order" to the existing "never", "madvise"
>>>> and another interface to specify max hinted order (enforcing <9) for "any-order".
>>>> Later, we can allow users to specify any max hinted order, including 9. Just an
>>>> idea.
>>> I suspect that all the config knobs (enable/disable mixing mode, define "any-order"
>>> or "specific-order") will be exist long term. Because there are always new workloads
>>> need be tuned against these configs.
>>>
>>>
>>> Regards
>>> Yin, Fengwei
>>>
>>>>
>>>>
>>>> --
>>>> Best Regards,
>>>> Yan, Zi
>
>
> --
> Best Regards,
> Yan, Zi