Message-ID: <f2a3b1f0-0823-4c3a-b0e8-20ac83e855b2@arm.com>
Date: Fri, 15 Dec 2023 14:05:28 +0000
From: Ryan Roberts <ryan.roberts@....com>
To: Alistair Popple <apopple@...dia.com>
Cc: Catalin Marinas <catalin.marinas@....com>, Will Deacon <will@...nel.org>,
 Ard Biesheuvel <ardb@...nel.org>, Marc Zyngier <maz@...nel.org>,
 Oliver Upton <oliver.upton@...ux.dev>, James Morse <james.morse@....com>,
 Suzuki K Poulose <suzuki.poulose@....com>, Zenghui Yu
 <yuzenghui@...wei.com>, Andrey Ryabinin <ryabinin.a.a@...il.com>,
 Alexander Potapenko <glider@...gle.com>,
 Andrey Konovalov <andreyknvl@...il.com>, Dmitry Vyukov <dvyukov@...gle.com>,
 Vincenzo Frascino <vincenzo.frascino@....com>,
 Andrew Morton <akpm@...ux-foundation.org>,
 Anshuman Khandual <anshuman.khandual@....com>,
 Matthew Wilcox <willy@...radead.org>, Yu Zhao <yuzhao@...gle.com>,
 Mark Rutland <mark.rutland@....com>, David Hildenbrand <david@...hat.com>,
 Kefeng Wang <wangkefeng.wang@...wei.com>, John Hubbard
 <jhubbard@...dia.com>, Zi Yan <ziy@...dia.com>,
 Barry Song <21cnbao@...il.com>, Yang Shi <shy828301@...il.com>,
 linux-arm-kernel@...ts.infradead.org, linux-mm@...ck.org,
 linux-kernel@...r.kernel.org
Subject: Re: [PATCH v3 14/15] arm64/mm: Implement ptep_set_wrprotects() to
 optimize fork()

On 15/12/2023 04:32, Alistair Popple wrote:
> 
> Ryan Roberts <ryan.roberts@....com> writes:
> 
>> On 08/12/2023 01:37, Alistair Popple wrote:
>>>
>>> Ryan Roberts <ryan.roberts@....com> writes:
>>>
>>>> With the core-mm changes in place to batch-copy ptes during fork, we can
>>>> take advantage of this in arm64 to greatly reduce the number of tlbis we
>>>> have to issue, and recover the lost fork performance incured when adding
>>>> support for transparent contiguous ptes.
>>>>
>>>> If we are write-protecting a whole contig range, we can apply the
>>>> write-protection to the whole range and know that it won't change
>>>> whether the range should have the contiguous bit set or not. For ranges
>>>> smaller than the contig range, we will still have to unfold, apply the
>>>> write-protection, then fold if the change now means the range is
>>>> foldable.
>>>>
>>>> This optimization is possible thanks to the tightening of the Arm ARM in
>>>> respect to the definition and behaviour when 'Misprogramming the
>>>> Contiguous bit'. See section D21194 at
>>>> https://developer.arm.com/documentation/102105/latest/
>>>>
>>>> Performance tested with the following test written for the will-it-scale
>>>> framework:
>>>>
>>>> -------
>>>>
>>>> char *testcase_description = "fork and exit";
>>>>
>>>> void testcase(unsigned long long *iterations, unsigned long nr)
>>>> {
>>>> 	int pid;
>>>> 	char *mem;
>>>>
>>>> 	mem = malloc(SZ_128M);
>>>> 	assert(mem);
>>>> 	memset(mem, 1, SZ_128M);
>>>>
>>>> 	while (1) {
>>>> 		pid = fork();
>>>> 		assert(pid >= 0);
>>>>
>>>> 		if (!pid)
>>>> 			exit(0);
>>>>
>>>> 		waitpid(pid, NULL, 0);
>>>>
>>>> 		(*iterations)++;
>>>> 	}
>>>> }
>>>>
>>>> -------
>>>>
>>>> I see huge performance regression when PTE_CONT support was added, then
>>>> the regression is mostly fixed with the addition of this change. The
>>>> following shows regression relative to before PTE_CONT was enabled
>>>> (bigger negative value is bigger regression):
>>>>
>>>> |   cpus |   before opt |   after opt |
>>>> |-------:|-------------:|------------:|
>>>> |      1 |       -10.4% |       -5.2% |
>>>> |      8 |       -15.4% |       -3.5% |
>>>> |     16 |       -38.7% |       -3.7% |
>>>> |     24 |       -57.0% |       -4.4% |
>>>> |     32 |       -65.8% |       -5.4% |
>>>>
>>>> Signed-off-by: Ryan Roberts <ryan.roberts@....com>
>>>> ---
>>>>  arch/arm64/include/asm/pgtable.h | 30 ++++++++++++++++++++---
>>>>  arch/arm64/mm/contpte.c          | 42 ++++++++++++++++++++++++++++++++
>>>>  2 files changed, 69 insertions(+), 3 deletions(-)
>>>>
>>>> diff --git a/arch/arm64/include/asm/pgtable.h b/arch/arm64/include/asm/pgtable.h
>>>> index 15bc9cf1eef4..9bd2f57a9e11 100644
>>>> --- a/arch/arm64/include/asm/pgtable.h
>>>> +++ b/arch/arm64/include/asm/pgtable.h
>>>> @@ -984,6 +984,16 @@ static inline void __ptep_set_wrprotect(struct mm_struct *mm,
>>>>  	} while (pte_val(pte) != pte_val(old_pte));
>>>>  }
>>>>  
>>>> +static inline void __ptep_set_wrprotects(struct mm_struct *mm,
>>>> +					unsigned long address, pte_t *ptep,
>>>> +					unsigned int nr)
>>>> +{
>>>> +	unsigned int i;
>>>> +
>>>> +	for (i = 0; i < nr; i++, address += PAGE_SIZE, ptep++)
>>>> +		__ptep_set_wrprotect(mm, address, ptep);
>>>> +}
>>>> +
>>>>  #ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>>>  #define __HAVE_ARCH_PMDP_SET_WRPROTECT
>>>>  static inline void pmdp_set_wrprotect(struct mm_struct *mm,
>>>> @@ -1139,6 +1149,8 @@ extern int contpte_ptep_test_and_clear_young(struct vm_area_struct *vma,
>>>>  				unsigned long addr, pte_t *ptep);
>>>>  extern int contpte_ptep_clear_flush_young(struct vm_area_struct *vma,
>>>>  				unsigned long addr, pte_t *ptep);
>>>> +extern void contpte_set_wrprotects(struct mm_struct *mm, unsigned long addr,
>>>> +				pte_t *ptep, unsigned int nr);
>>>>  extern int contpte_ptep_set_access_flags(struct vm_area_struct *vma,
>>>>  				unsigned long addr, pte_t *ptep,
>>>>  				pte_t entry, int dirty);
>>>> @@ -1290,13 +1302,25 @@ static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
>>>>  	return contpte_ptep_clear_flush_young(vma, addr, ptep);
>>>>  }
>>>>  
>>>> +#define ptep_set_wrprotects ptep_set_wrprotects
>>>> +static inline void ptep_set_wrprotects(struct mm_struct *mm, unsigned long addr,
>>>> +				pte_t *ptep, unsigned int nr)
>>>> +{
>>>> +	if (!contpte_is_enabled(mm))
>>>> +		__ptep_set_wrprotects(mm, addr, ptep, nr);
>>>> +	else if (nr == 1) {
>>>
>>> Why do we need the special case here? Couldn't we just call
>>> contpte_set_wrprotects() with nr == 1?
>>
>> My intention is for this to be a fast path for ptep_set_wrprotect(). I'm having
>> to work hard to prevent regressing the order-0 folios case.
> 
> This ends up calling three functions anyway so I'm curious - does
> removing the one function call really make that much of difference?

Yes; big time. All the functions in the fast path are inlined. The version
regresses a fork() microbenchmark that David gave me by ~30%. I've had to work
quite hard to reduce that to 2%, even from this starting point. There is so
little in the inner loop that even the __ptep_get(ptep) (which is a READ_ONCE())
makes a measurable difference.

Anyway, I'll be posting v4 with these optimizations and all the supporting
benchmark data on Monday.

> 
> Either way I think a comment justifying the special case (ie. that this
> is simply a fast path for nr == 1) would be good.

I've added a comment here in v4.

> 
> Thanks.
> 
>>>
>>>> +		contpte_try_unfold(mm, addr, ptep, __ptep_get(ptep));
>>>> +		__ptep_set_wrprotects(mm, addr, ptep, 1);
>>>> +		contpte_try_fold(mm, addr, ptep, __ptep_get(ptep));
>>>> +	} else
>>>> +		contpte_set_wrprotects(mm, addr, ptep, nr);
>>>> +}
>>>> +
>>>>  #define __HAVE_ARCH_PTEP_SET_WRPROTECT
>>>>  static inline void ptep_set_wrprotect(struct mm_struct *mm,
>>>>  				unsigned long addr, pte_t *ptep)
>>>>  {
>>>> -	contpte_try_unfold(mm, addr, ptep, __ptep_get(ptep));
>>>> -	__ptep_set_wrprotect(mm, addr, ptep);
>>>> -	contpte_try_fold(mm, addr, ptep, __ptep_get(ptep));
>>>> +	ptep_set_wrprotects(mm, addr, ptep, 1);
>>>>  }
>>>>  
>>>>  #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
>>>> diff --git a/arch/arm64/mm/contpte.c b/arch/arm64/mm/contpte.c
>>>> index e079ec61d7d1..2a57df16bf58 100644
>>>> --- a/arch/arm64/mm/contpte.c
>>>> +++ b/arch/arm64/mm/contpte.c
>>>> @@ -303,6 +303,48 @@ int contpte_ptep_clear_flush_young(struct vm_area_struct *vma,
>>>>  }
>>>>  EXPORT_SYMBOL(contpte_ptep_clear_flush_young);
>>>>  
>>>> +void contpte_set_wrprotects(struct mm_struct *mm, unsigned long addr,
>>>> +					pte_t *ptep, unsigned int nr)
>>>> +{
>>>> +	unsigned long next;
>>>> +	unsigned long end = addr + (nr << PAGE_SHIFT);
>>>> +
>>>> +	do {
>>>> +		next = pte_cont_addr_end(addr, end);
>>>> +		nr = (next - addr) >> PAGE_SHIFT;
>>>> +
>>>> +		/*
>>>> +		 * If wrprotecting an entire contig range, we can avoid
>>>> +		 * unfolding. Just set wrprotect and wait for the later
>>>> +		 * mmu_gather flush to invalidate the tlb. Until the flush, the
>>>> +		 * page may or may not be wrprotected. After the flush, it is
>>>> +		 * guarranteed wrprotected. If its a partial range though, we
>>>> +		 * must unfold, because we can't have a case where CONT_PTE is
>>>> +		 * set but wrprotect applies to a subset of the PTEs; this would
>>>> +		 * cause it to continue to be unpredictable after the flush.
>>>> +		 */
>>>> +		if (nr != CONT_PTES)
>>>> +			contpte_try_unfold(mm, addr, ptep, __ptep_get(ptep));
>>>> +
>>>> +		__ptep_set_wrprotects(mm, addr, ptep, nr);
>>>> +
>>>> +		addr = next;
>>>> +		ptep += nr;
>>>> +
>>>> +		/*
>>>> +		 * If applying to a partial contig range, the change could have
>>>> +		 * made the range foldable. Use the last pte in the range we
>>>> +		 * just set for comparison, since contpte_try_fold() only
>>>> +		 * triggers when acting on the last pte in the contig range.
>>>> +		 */
>>>> +		if (nr != CONT_PTES)
>>>> +			contpte_try_fold(mm, addr - PAGE_SIZE, ptep - 1,
>>>> +					 __ptep_get(ptep - 1));
>>>> +
>>>> +	} while (addr != end);
>>>> +}
>>>> +EXPORT_SYMBOL(contpte_set_wrprotects);
>>>> +
>>>>  int contpte_ptep_set_access_flags(struct vm_area_struct *vma,
>>>>  					unsigned long addr, pte_t *ptep,
>>>>  					pte_t entry, int dirty)
>>>
> 

Powered by blists - more mailing lists