On 28/06/2023 03:32, Yin Fengwei wrote:
>
>
> On 6/27/23 15:56, Ryan Roberts wrote:
>> On 27/06/2023 06:29, Yu Zhao wrote:
>>> On Mon, Jun 26, 2023 at 8:34 PM Yu Zhao <yuzhao@xxxxxxxxxx> wrote:
>>>>
>>>> On Mon, Jun 26, 2023 at 11:14 AM Ryan Roberts <ryan.roberts@xxxxxxx> wrote:
>>>>>
>>>>> Opportunistically attempt to allocate high-order folios in highmem,
>>>>> optionally zeroed. Retry with lower orders all the way to order-0, until
>>>>> success. Although, of note, order-1 allocations are skipped since a
>>>>> large folio must be at least order-2 to work with the THP machinery. The
>>>>> user must check what they got with folio_order().
>>>>>
>>>>> This will be used to oportunistically allocate large folios for
>>>>> anonymous memory with a sensible fallback under memory pressure.
>>>>>
>>>>> For attempts to allocate non-0 orders, we set __GFP_NORETRY to prevent
>>>>> high latency due to reclaim, instead preferring to just try for a lower
>>>>> order. The same approach is used by the readahead code when allocating
>>>>> large folios.
>>>>>
>>>>> Signed-off-by: Ryan Roberts <ryan.roberts@xxxxxxx>
>>>>> ---
>>>>> mm/memory.c | 33 +++++++++++++++++++++++++++++++++
>>>>> 1 file changed, 33 insertions(+)
>>>>>
>>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>>> index 367bbbb29d91..53896d46e686 100644
>>>>> --- a/mm/memory.c
>>>>> +++ b/mm/memory.c
>>>>> @@ -3001,6 +3001,39 @@ static vm_fault_t fault_dirty_shared_page(struct vm_fault *vmf)
>>>>> return 0;
>>>>> }
>>>>>
>>>>> +static inline struct folio *vma_alloc_movable_folio(struct vm_area_struct *vma,
>>>>> + unsigned long vaddr, int order, bool zeroed)
>>>>> +{
>>>>> + gfp_t gfp = order > 0 ? __GFP_NORETRY | __GFP_NOWARN : 0;
>>>>> +
>>>>> + if (zeroed)
>>>>> + return vma_alloc_zeroed_movable_folio(vma, vaddr, gfp, order);
>>>>> + else
>>>>> + return vma_alloc_folio(GFP_HIGHUSER_MOVABLE | gfp, order, vma,
>>>>> + vaddr, false);
>>>>> +}
>>>>> +
>>>>> +/*
>>>>> + * Opportunistically attempt to allocate high-order folios, retrying with lower
>>>>> + * orders all the way to order-0, until success. order-1 allocations are skipped
>>>>> + * since a folio must be at least order-2 to work with the THP machinery. The
>>>>> + * user must check what they got with folio_order(). vaddr can be any virtual
>>>>> + * address that will be mapped by the allocated folio.
>>>>> + */
>>>>> +static struct folio *try_vma_alloc_movable_folio(struct vm_area_struct *vma,
>>>>> + unsigned long vaddr, int order, bool zeroed)
>>>>> +{
>>>>> + struct folio *folio;
>>>>> +
>>>>> + for (; order > 1; order--) {
>>>>> + folio = vma_alloc_movable_folio(vma, vaddr, order, zeroed);
>>>>> + if (folio)
>>>>> + return folio;
>>>>> + }
>>>>> +
>>>>> + return vma_alloc_movable_folio(vma, vaddr, 0, zeroed);
>>>>> +}
>>>>
>>>> I'd drop this patch. Instead, in do_anonymous_page():
>>>>
>>>> if (IS_ENABLED(CONFIG_ARCH_WANTS_PTE_ORDER))
>>>> folio = vma_alloc_zeroed_movable_folio(vma, addr,
>>>> CONFIG_ARCH_WANTS_PTE_ORDER))
>>>>
>>>> if (!folio)
>>>> folio = vma_alloc_zeroed_movable_folio(vma, addr, 0);
>>>
>>> I meant a runtime function arch_wants_pte_order() (Its default
>>> implementation would return 0.)
>>
>> There are a bunch of things which you are implying here which I'll try to make
>> explicit:
>>
>> I think you are implying that we shouldn't retry allocation with intermediate
>> orders; but only try the order requested by the arch (arch_wants_pte_order())
>> and 0. Correct? For arm64 at least, I would like the VMA's THP hint to be a
>> factor in determining the preferred order (see patches 8 and 9). So I would add
>> a vma parameter to arch_wants_pte_order() to allow for this.
>>
>> For the case where the THP hint is present, then the arch will request 2M (if
>> the page size is 16K or 64K). If that fails to allocate, there is still value in
>> allocating a 64K folio (which is order 2 in the 16K case). Without the retry
>> with intermediate orders logic, we would not get this.
>>
>> We can't just blindly allocate a folio of arch_wants_pte_order() size because it
>> might overlap with existing populated PTEs, or cross the bounds of the VMA (or a
>> number of other things - see calc_anon_folio_order_alloc() in patch 10). Are you
>> implying that if there is any kind of issue like this, then we should go
>> directly to order 0? I can kind of see the argument from a minimizing
>> fragmentation perspective, but for best possible performance I think we are
>> better off "packing the bin" with intermediate orders.
>
> One drawback of the retry is that it could introduce large tail latency (by
> memory zeroing, memory reclaiming or existing populated PTEs). That may not
> be appreciated by some applications. Thanks.
Good point. based on all the discussion, I think the conclusion is:
- ask the arch to for preferred folio order with runtime function
- check the folio will fit (racy) - if does not fit fall back to order-0
- allocate the folio
- take the ptl
- check the folio still fits (not racy) - if does not fit fall back to order-0
So in the worst case the latency will be allocating and zeroing a large folio,
then allocating and zeroing an order-0 folio. Which is obviously better than
iterating through every order from preferred to 0.
I'll work this flow into a v2.
>
>
> Regards
> Yin, Fengwei
>
>>
>> You're also implying that a runtime arch_wants_pte_order() function is better
>> than the Kconfig stuff I did in patch 8. On reflection, I agree with you here. I
>> think you mentioned that AMD supports coalescing 8 pages on some CPUs - so you
>> would probably want runtime logic to determine if you are on an appropriate AMD
>> CPU as part of the decision in that function?
>>
>> The real reason for the existance of try_vma_alloc_movable_folio() is that I'm
>> reusing it on the other fault paths (which are no longer part of this series).
>> But I guess that's not a good reason to keep this until we get to those patches.
