linux-kernel - Re: [PATCH v1 2/2] mm: don't install PMD mappings when THPs are disabled by the hw/process/vma

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

Message-ID: <b6ba0313-6a3f-4bfc-9237-547355cd7b00@redhat.com>
Date: Fri, 11 Oct 2024 13:40:33 +0200
From: David Hildenbrand <david@...hat.com>
To: Ryan Roberts <ryan.roberts@....com>, linux-kernel@...r.kernel.org
Cc: linux-mm@...ck.org, kvm@...r.kernel.org,
 Andrew Morton <akpm@...ux-foundation.org>, Hugh Dickins <hughd@...gle.com>,
 Thomas Huth <thuth@...hat.com>, "Matthew Wilcox (Oracle)"
 <willy@...radead.org>, Christian Borntraeger <borntraeger@...ux.ibm.com>,
 Janosch Frank <frankja@...ux.ibm.com>,
 Claudio Imbrenda <imbrenda@...ux.ibm.com>,
 Kefeng Wang <wangkefeng.wang@...wei.com>, Leo Fu <bfu@...hat.com>
Subject: Re: [PATCH v1 2/2] mm: don't install PMD mappings when THPs are
 disabled by the hw/process/vma

On 11.10.24 13:36, Ryan Roberts wrote:
> On 11/10/2024 12:33, David Hildenbrand wrote:
>> On 11.10.24 13:29, Ryan Roberts wrote:
>>> On 11/10/2024 11:24, David Hildenbrand wrote:
>>>> We (or rather, readahead logic :) ) might be allocating a THP in the
>>>> pagecache and then try mapping it into a process that explicitly disabled
>>>> THP: we might end up installing PMD mappings.
>>>>
>>>> This is a problem for s390x KVM, which explicitly remaps all PMD-mapped
>>>> THPs to be PTE-mapped in s390_enable_sie()->thp_split_mm(), before
>>>> starting the VM.
>>>>
>>>> For example, starting a VM backed on a file system with large folios
>>>> supported makes the VM crash when the VM tries accessing such a mapping
>>>> using KVM.
>>>>
>>>> Is it also a problem when the HW disabled THP using
>>>> TRANSPARENT_HUGEPAGE_UNSUPPORTED? At least on x86 this would be the case
>>>> without X86_FEATURE_PSE.
>>>>
>>>> In the future, we might be able to do better on s390x and only disallow
>>>> PMD mappings -- what s390x and likely TRANSPARENT_HUGEPAGE_UNSUPPORTED
>>>> really wants. For now, fix it by essentially performing the same check as
>>>> would be done in __thp_vma_allowable_orders() or in shmem code, where this
>>>> works as expected, and disallow PMD mappings, making us fallback to PTE
>>>> mappings.
>>>>
>>>> Reported-by: Leo Fu <bfu@...hat.com>
>>>> Fixes: 793917d997df ("mm/readahead: Add large folio readahead")
>>>
>>> Will this patch be difficult to backport given it depends on the previous patch
>>> and that doesn't have a Fixes tag?
>>
>> "difficult" -- not really. Andrew might want to tag patch #1  with "Fixes:" as
>> well, but I can also send simple stable backports that avoid patch #1.
>>
>> (Thinking again, I assume we want to Cc:stable)
>>
>>>
>>>> Cc: Thomas Huth <thuth@...hat.com>
>>>> Cc: Matthew Wilcox (Oracle) <willy@...radead.org>
>>>> Cc: Ryan Roberts <ryan.roberts@....com>
>>>> Cc: Christian Borntraeger <borntraeger@...ux.ibm.com>
>>>> Cc: Janosch Frank <frankja@...ux.ibm.com>
>>>> Cc: Claudio Imbrenda <imbrenda@...ux.ibm.com>
>>>> Signed-off-by: David Hildenbrand <david@...hat.com>
>>>> ---
>>>>    mm/memory.c | 9 +++++++++
>>>>    1 file changed, 9 insertions(+)
>>>>
>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>> index 2366578015ad..a2e501489517 100644
>>>> --- a/mm/memory.c
>>>> +++ b/mm/memory.c
>>>> @@ -4925,6 +4925,15 @@ vm_fault_t do_set_pmd(struct vm_fault *vmf, struct
>>>> page *page)
>>>>        pmd_t entry;
>>>>        vm_fault_t ret = VM_FAULT_FALLBACK;
>>>>    +    /*
>>>> +     * It is too late to allocate a small folio, we already have a large
>>>> +     * folio in the pagecache: especially s390 KVM cannot tolerate any
>>>> +     * PMD mappings, but PTE-mapped THP are fine. So let's simply refuse any
>>>> +     * PMD mappings if THPs are disabled.
>>>> +     */
>>>> +    if (thp_disabled_by_hw() || vma_thp_disabled(vma, vma->vm_flags))
>>>> +        return ret;
>>>
>>> Why not just call thp_vma_allowable_orders()?
>>
>> Why call thp_vma_allowable_orders() that does a lot more work that doesn't
>> really apply here? :)
> 
> Yeah fair enough, I was just thinking it makes the code simpler to keep all the
> checks in one place. But no strong opinion.
> 
> Either way:
> 
> Reviewed-by: Ryan Roberts <ryan.roberts@....com>

Thanks!

Also, I decided to not use "thp_vma_allowable_orders" because we are 
past the allocation phase (as indicated in the comment) and can really 
just change the way how we map the folio (PMD vs. PTE), not really 
*what* folio to use.

Ideally, in the future we have a different way of just saying "no PMD 
mappings please", decoupling the mapping from the allocation granularity.

-- 
Cheers,

David / dhildenb