linux-kernel - Re: [PATCH 2/2] kpageflags: fix wrong KPF_THP on non-pmd-mappable compound pages

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date: Fri, 28 Jun 2024 03:01:43 +0000
From: ran xiaokai <ranxiaokai627@....com>
To: david@...hat.com
Cc: akpm@...ux-foundation.org,
	baohua@...nel.org,
	linux-fsdevel@...r.kernel.org,
	linux-kernel@...r.kernel.org,
	linux-mm@...ck.org,
	peterx@...hat.com,
	ran.xiaokai@....com.cn,
	ranxiaokai627@....com,
	ryan.roberts@....com,
	svetly.todorov@...verge.com,
	vbabka@...e.cz,
	willy@...radead.org,
	ziy@...dia.com
Subject: Re: [PATCH 2/2] kpageflags: fix wrong KPF_THP on non-pmd-mappable compound pages

>On 26.06.24 04:49, ran xiaokai wrote:
>> From: Ran Xiaokai <ran.xiaokai@....com.cn>
>> 
>> KPF_COMPOUND_HEAD and KPF_COMPOUND_TAIL are set on "common" compound
>> pages, which means of any order, but KPF_THP should only be set
>> when the folio is a 2M pmd mappable THP. Since commit 19eaf44954df
>
>"should only be set" -- who says that? :)
>
>The documentation only talks about "Contiguous pages which construct 
>transparent hugepages". Sure, when it was added there were only PMD ones.

Hi, David,
I am working on tools/testing/selftests/mm/split_huge_page_test to support
splitint THP to more orders, not only order-0, and the testcases failed.
It seems the testcodes rely on KPF_THP to indicate only 2M THP.

>> ("mm: thp: support allocation of anonymous multi-size THP"),
>> multiple orders of folios can be allocated and mapped to userspace,
>> so the folio_test_large() check is not sufficient here,
>> replace it with folio_test_pmd_mappable() to fix this.
>> 
>
>A couple of points:
>
>1) If I am not daydreaming, ever since we supported non-PMD THP in the
>    pagecache (much longer than anon mTHP), we already indicate KPF_THP
>    for them here. So this is not anon specific? Or am I getting the
>    PG_lru check all wrong?

Thanks for the clarification.

>2) Anon THP are disabled as default. If some custom tool cannot deal
>    with that "extension" we did with smaller THP, it shall be updated if
>    it really has to special-case PMD-mapped THP, before enabled by the
>    admin.

ok, it seems that it is the testcodes which should be updated.

>
>I think this interface does exactly what we want, as it is right now. 
>Unless there is *good* reason, we should keep it like that.
>
>So I suggest
>
>a) Extend the documentation to just state "THP of any size and any 
>mapping granularity" or sth like that.

yes, it is neccessay to update the documentation to make it more clear.
Since the definition of KPF_THP has been expanded since it was firstly introduced.

>b) Maybe using folio_test_large_rmappable() instead of "(k & (1 <<
>    PG_lru)) || is_anon", so even isolated-from-LRU THPs are indicated
>    properly.

i will investigate on this.

>c) Whoever is interested in getting the folio size, use this flag along
>    with a scan for the KPF_COMPOUND_HEAD.

yes, we can use the combination of KPF_COMPOUND_HEAD and KPF_COMPOUND_TAIL
to figure out the compound order.

>
>I'll note that, scanning documentation, PAGE_IS_HUGE currently only 
>applies to PMD *mapped* THP. It doesn't consider PTE-mapped THP at all 
>(including PMD-ones). Likely, documentation should be updated to state 
>"PMD-mapped THP or any hugetlb page".

i will also investigate on this.

>> Also kpageflags is not only for userspace memory but for all valid pfn
>> pages,including slab pages or drivers used pages, so the PG_lru and
>> is_anon check are unnecessary here.
>
>I'm completely confused about that statements. We do have KPF_OFFLINE, 
>KPF_PGTABLE, KPF_SLAB, ... I'm missing something important.

My statement is wrong, please ignore this.

>> 
>> Signed-off-by: Ran Xiaokai <ran.xiaokai@....com.cn>
>> ---
>>   fs/proc/page.c | 14 ++++----------
>>   1 file changed, 4 insertions(+), 10 deletions(-)
>> 
>> diff --git a/fs/proc/page.c b/fs/proc/page.c
>> index 2fb64bdb64eb..3e7b70449c2f 100644
>> --- a/fs/proc/page.c
>> +++ b/fs/proc/page.c
>> @@ -146,19 +146,13 @@ u64 stable_page_flags(const struct page *page)
>>   		u |= kpf_copy_bit(k, KPF_COMPOUND_HEAD, PG_head);
>>   	else
>>   		u |= 1 << KPF_COMPOUND_TAIL;
>> +
>>   	if (folio_test_hugetlb(folio))
>>   		u |= 1 << KPF_HUGE;
>> -	/*
>> -	 * We need to check PageLRU/PageAnon
>> -	 * to make sure a given page is a thp, not a non-huge compound page.
>> -	 */
>> -	else if (folio_test_large(folio)) {
>> -		if ((k & (1 << PG_lru)) || is_anon)
>> -			u |= 1 << KPF_THP;
>> -		else if (is_huge_zero_folio(folio)) {
>> +	else if (folio_test_pmd_mappable(folio)) {
>
>folio_test_pmd_mappable() would not be future safe. It includes anything 
> >= PMD_ORDER as well.
>
>
>-- 
>Cheers,
>
>David / dhildenb