Re: [PATCH v1 2/2] mm: remove redundant smp_wmb()

[Qi Zheng <zhengqi.arch@bytedance.com>]

On 2021/8/31 PM6:20, Vlastimil Babka wrote:
> On 8/28/21 06:23, Qi Zheng wrote:
>> The smp_wmb() which is in the __pte_alloc() is used to
>> ensure all ptes setup is visible before the pte is made
>> visible to other CPUs by being put into page tables. We
>> only need this when the pte is actually populated, so
>> move it to pte_install(). __pte_alloc_kernel(),
> 
> It's named pmd_install()?

Yes, I will update it in the next version.

> 
>> __p4d_alloc(), __pud_alloc() and __pmd_alloc() are similar
>> to this case.
>>
>> We can also defer smp_wmb() to the place where the pmd entry
>> is really populated by preallocated pte. There are two kinds
>> of user of preallocated pte, one is filemap & finish_fault(),
>> another is THP. The former does not need another smp_wmb()
>> because the smp_wmb() has been done by pte_install().
> 
> Same here.
> 
>> Fortunately, the latter also does not need another smp_wmb()
>> because there is already a smp_wmb() before populating the
>> new pte when the THP uses a preallocated pte to split a huge
>> pmd.
>>
>> Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
>> Reviewed-by: Muchun Song <songmuchun@bytedance.com>
>> ---
>>   mm/memory.c         | 47 ++++++++++++++++++++---------------------------
>>   mm/sparse-vmemmap.c |  2 +-
>>   2 files changed, 21 insertions(+), 28 deletions(-)
>>
>> diff --git a/mm/memory.c b/mm/memory.c
>> index ef7b1762e996..9c7534187454 100644
>> --- a/mm/memory.c
>> +++ b/mm/memory.c
>> @@ -439,6 +439,20 @@ void pmd_install(struct mm_struct *mm, pmd_t *pmd, pgtable_t *pte)
>>   
>>   	if (likely(pmd_none(*pmd))) {	/* Has another populated it ? */
>>   		mm_inc_nr_ptes(mm);
>> +		/*
>> +		 * Ensure all pte setup (eg. pte page lock and page clearing) are
>> +		 * visible before the pte is made visible to other CPUs by being
>> +		 * put into page tables.
>> +		 *
>> +		 * The other side of the story is the pointer chasing in the page
>> +		 * table walking code (when walking the page table without locking;
>> +		 * ie. most of the time). Fortunately, these data accesses consist
>> +		 * of a chain of data-dependent loads, meaning most CPUs (alpha
>> +		 * being the notable exception) will already guarantee loads are
>> +		 * seen in-order. See the alpha page table accessors for the
>> +		 * smp_rmb() barriers in page table walking code.
>> +		 */
>> +		smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
> 
> So, could it? :)
> 

Yes, it could, but we don't have smp_wmb__after_spin_lock() now.