Re: [PATCH v2 2/3] mm: rmap: Move the cache flushing to the correct place for hugetlb PMD sharing

[Baolin Wang <baolin.wang@linux.alibaba.com>]

On 4/28/2022 1:55 PM, Muchun Song wrote:
> On Wed, Apr 27, 2022 at 06:52:06PM +0800, Baolin Wang wrote:
>> The cache level flush will always be first when changing an existing
>> virtual–>physical mapping to a new value, since this allows us to
>> properly handle systems whose caches are strict and require a
>> virtual–>physical translation to exist for a virtual address. So we
>> should move the cache flushing before huge_pmd_unshare().
>>
> 
> Right.
> 
>> As Muchun pointed out[1], now the architectures whose supporting hugetlb
>> PMD sharing have no cache flush issues in practice. But I think we
>> should still follow the cache/TLB flushing rules when changing a valid
>> virtual address mapping in case of potential issues in future.
> 
> Right. One point i need to clarify. I do not object this change but
> want you to clarify this (not an issue in practice) in commit log
> to let others know they do not need to bp this.
> 
>>
>> [1] https://lore.kernel.org/all/YmT%2F%2FhuUbFX+KHcy@FVFYT0MHHV2J.usts.net/
>> Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
>> ---
>>   mm/rmap.c | 40 ++++++++++++++++++++++------------------
>>   1 file changed, 22 insertions(+), 18 deletions(-)
>>
>> diff --git a/mm/rmap.c b/mm/rmap.c
>> index 61e63db..4f0d115 100644
>> --- a/mm/rmap.c
>> +++ b/mm/rmap.c
>> @@ -1535,15 +1535,16 @@ static bool try_to_unmap_one(struct folio *folio, struct vm_area_struct *vma,
>>   			 * do this outside rmap routines.
>>   			 */
>>   			VM_BUG_ON(!(flags & TTU_RMAP_LOCKED));
>> +			/*
>> +			 * huge_pmd_unshare may unmap an entire PMD page.
>> +			 * There is no way of knowing exactly which PMDs may
>> +			 * be cached for this mm, so we must flush them all.
>> +			 * start/end were already adjusted above to cover this
>> +			 * range.
>> +			 */
>> +			flush_cache_range(vma, range.start, range.end);
>> +
> 
> flush_cache_range() is always called even if we do not need to flush.

Right, this is intended. In the original code, if it is not a shared 
PMD, we will use flush_cache_page() to do cache flushing. However the 
flush_cache_page() can not cover the whole size of a hugetlb page on 
some architectures, which is fixed by patch 3.

> How about introducing a new helper like hugetlb_pmd_shared() which
> returns true for shared PMD? Then:
> 
> 	if (hugetlb_pmd_shared(mm, vma, pvmw.pte)) {
> 		flush_cache_range(vma, range.start, range.end);
> 		huge_pmd_unshare(mm, vma, &address, pvmw.pte);
> 		flush_tlb_range(vma, range.start, range.end);
> 	}
> 
> The code could be a little simpler. Right?

IMHO after patch 3, the code will be changed as below, so seems no need 
to separate the validation of the shared PMDs from huge_pmd_unshare() 
into a new function.

if (folio_test_hugetlb(folio)) {
	flush_cache_range(vma, range.start, range.end);

	if (!folio_test_anon(folio)) {
		VM_BUG_ON(!(flags & TTU_RMAP_LOCKED));

		if (hugetlb_pmd_shared(mm, vma, pvmw.pte)) {
			huge_pmd_unshare(mm, vma, &address, pvmw.pte));
			flush_tlb_range(vma, range.start, range.end);
			......
			break;
		}
	}
} else {
	......
}