Message-ID: <878qkohleu.fsf@oracle.com>
Date: Tue, 15 Jul 2025 20:19:53 -0700
From: Ankur Arora <ankur.a.arora@...cle.com>
To: David Hildenbrand <david@...hat.com>
Cc: Ankur Arora <ankur.a.arora@...cle.com>, linux-kernel@...r.kernel.org,
        linux-mm@...ck.org, x86@...nel.org, akpm@...ux-foundation.org,
        bp@...en8.de, dave.hansen@...ux.intel.com, hpa@...or.com,
        mingo@...hat.com, mjguzik@...il.com, luto@...nel.org,
        peterz@...radead.org, acme@...nel.org, namhyung@...nel.org,
        tglx@...utronix.de, willy@...radead.org, raghavendra.kt@....com,
        boris.ostrovsky@...cle.com, konrad.wilk@...cle.com
Subject: Re: [PATCH v5 13/14] mm: memory: support clearing page-extents


David Hildenbrand <david@...hat.com> writes:

> On 10.07.25 02:59, Ankur Arora wrote:
>> folio_zero_user() is constrained to clear in a page-at-a-time
>> fashion because it supports CONFIG_HIGHMEM which means that kernel
>> mappings for pages in a folio are not guaranteed to be contiguous.
>> We don't have this problem when running under configurations with
>> CONFIG_CLEAR_PAGE_EXTENT (implies !CONFIG_HIGHMEM), so zero in
>> longer page-extents.
>> This is expected to be faster because the processor can now optimize
>> the clearing based on the knowledge of the extent.
>> However, clearing in larger chunks can have two other problems:
>>   - cache locality when clearing small folios (< MAX_ORDER_NR_PAGES)
>>     (larger folios don't have any expectation of cache locality).
>>   - preemption latency when clearing large folios.
>> Handle the first by splitting the clearing in three parts: the
>> faulting page and its immediate locality, its left and right
>> regions; the local neighbourhood is cleared last.
>> The second problem is relevant only when running under cooperative
>> preemption models. Limit the worst case preemption latency by clearing
>> in architecture specified ARCH_CLEAR_PAGE_EXTENT units.
>> Signed-off-by: Ankur Arora <ankur.a.arora@...cle.com>
>> ---
>>   mm/memory.c | 86 ++++++++++++++++++++++++++++++++++++++++++++++++++++-
>>   1 file changed, 85 insertions(+), 1 deletion(-)
>> diff --git a/mm/memory.c b/mm/memory.c
>> index b0cda5aab398..c52806270375 100644
>> --- a/mm/memory.c
>> +++ b/mm/memory.c
>> @@ -7034,6 +7034,7 @@ static inline int process_huge_page(
>>   	return 0;
>>   }
>>   +#ifndef CONFIG_CLEAR_PAGE_EXTENT
>>   static void clear_gigantic_page(struct folio *folio, unsigned long addr_hint,
>>   				unsigned int nr_pages)
>>   {
>> @@ -7058,7 +7059,10 @@ static int clear_subpage(unsigned long addr, int idx, void *arg)
>>   /**
>>    * folio_zero_user - Zero a folio which will be mapped to userspace.
>>    * @folio: The folio to zero.
>> - * @addr_hint: The address will be accessed or the base address if uncelar.
>> + * @addr_hint: The address accessed by the user or the base address.
>> + *
>> + * folio_zero_user() uses clear_gigantic_page() or process_huge_page() to
>> + * do page-at-a-time zeroing because it needs to handle CONFIG_HIGHMEM.
>>    */
>>   void folio_zero_user(struct folio *folio, unsigned long addr_hint)
>>   {
>> @@ -7070,6 +7074,86 @@ void folio_zero_user(struct folio *folio, unsigned long addr_hint)
>>   		process_huge_page(addr_hint, nr_pages, clear_subpage, folio);
>>   }
>>   +#else /* CONFIG_CLEAR_PAGE_EXTENT */
>> +
>> +static void clear_pages_resched(void *addr, int npages)
>> +{
>> +	int i, remaining;
>> +
>> +	if (preempt_model_preemptible()) {
>> +		clear_pages(addr, npages);
>> +		goto out;
>> +	}
>> +
>> +	for (i = 0; i < npages/ARCH_CLEAR_PAGE_EXTENT; i++) {
>> +		clear_pages(addr + i * ARCH_CLEAR_PAGE_EXTENT * PAGE_SIZE,
>> +			    ARCH_CLEAR_PAGE_EXTENT);
>> +		cond_resched();
>> +	}
>> +
>> +	remaining = npages % ARCH_CLEAR_PAGE_EXTENT;
>> +
>> +	if (remaining)
>> +		clear_pages(addr + i * ARCH_CLEAR_PAGE_EXTENT * PAGE_SHIFT,
>> +			    remaining);
>> +out:
>> +	cond_resched();
>> +}
>> +
>> +/*
>> + * folio_zero_user - Zero a folio which will be mapped to userspace.
>> + * @folio: The folio to zero.
>> + * @addr_hint: The address accessed by the user or the base address.
>> + *
>> + * Uses architectural support for clear_pages() to zero page extents
>> + * instead of clearing page-at-a-time.
>> + *
>> + * Clearing of small folios (< MAX_ORDER_NR_PAGES) is split in three parts:
>> + * pages in the immediate locality of the faulting page, and its left, right
>> + * regions; the local neighbourhood cleared last in order to keep cache
>> + * lines of the target region hot.
>> + *
>> + * For larger folios we assume that there is no expectation of cache locality
>> + * and just do a straight zero.
>> + */
>> +void folio_zero_user(struct folio *folio, unsigned long addr_hint)
>> +{
>> +	unsigned long base_addr = ALIGN_DOWN(addr_hint, folio_size(folio));
>> +	const long fault_idx = (addr_hint - base_addr) / PAGE_SIZE;
>> +	const struct range pg = DEFINE_RANGE(0, folio_nr_pages(folio) - 1);
>> +	const int width = 2; /* number of pages cleared last on either side */
>> +	struct range r[3];
>> +	int i;
>> +
>> +	if (folio_nr_pages(folio) > MAX_ORDER_NR_PAGES) {
>> +		clear_pages_resched(page_address(folio_page(folio, 0)), folio_nr_pages(folio));
>> +		return;
>> +	}
>> +
>> +	/*
>> +	 * Faulting page and its immediate neighbourhood. Cleared at the end to
>> +	 * ensure it sticks around in the cache.
>> +	 */
>> +	r[2] = DEFINE_RANGE(clamp_t(s64, fault_idx - width, pg.start, pg.end),
>> +			    clamp_t(s64, fault_idx + width, pg.start, pg.end));
>> +
>> +	/* Region to the left of the fault */
>> +	r[1] = DEFINE_RANGE(pg.start,
>> +			    clamp_t(s64, r[2].start-1, pg.start-1, r[2].start));
>> +
>> +	/* Region to the right of the fault: always valid for the common fault_idx=0 case. */
>> +	r[0] = DEFINE_RANGE(clamp_t(s64, r[2].end+1, r[2].end, pg.end+1),
>> +			    pg.end);
>> +
>> +	for (i = 0; i <= 2; i++) {
>> +		int npages = range_len(&r[i]);
>> +
>> +		if (npages > 0)
>> +			clear_pages_resched(page_address(folio_page(folio, r[i].start)), npages);
>> +	}
>> +}
>> +#endif /* CONFIG_CLEAR_PAGE_EXTENT */
>> +
>>   static int copy_user_gigantic_page(struct folio *dst, struct folio *src,
>>   				   unsigned long addr_hint,
>>   				   struct vm_area_struct *vma,
>
> So, folio_zero_user() is only compiled with THP | HUGETLB already.
>
> What we should probably do is scrap the whole new kconfig option and
> do something like this in here:

So, in principle I don't disagree and unifying both of these is cleaner
than introducing a whole new option.

However that still leaves this code having to contort around CONFIG_HIGHMEM
which is probably even less frequently used than THP | HUGETLB.

Maybe we should get rid of ARCH_HAS_CLEAR_PAGES completely and everyone
with !HIGHMEM either use a generic version of clear_pages() which loops
and calls clear_page() or some architectural override.

And, then we can do a similar transformation with copy_pages() (and
copy_user_large_folio()).

At that point, process_huge_page() is used only for !HIGHMEM configs
configs which likely have relatively small caches and so that leaves
it probably over-engineered.

The thing that gives me pause is that non-x86 might perform worse
when they switch away from the left-right zeroing approach in
process_huge_page() to a generic clear_pages().

So, maybe allowing architectures to opt in by having to define
ARCH_HAS_CLEAR_PAGES would allow doing this in a more measured fashion.

(There isn't much justification for the left-right zeroing in the
commit: c79b57e462b "mm: hugetlb: clear target sub-page last when
clearing huge page". Just that it performs better on a benchmark:
anon-w-seq.)


Thanks!

--
ankur

Powered by blists - more mailing lists