linux-kernel - Re: [PATCH] KVM: SEV: Pin SEV guest memory out of CMA area

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]

Message-ID: <abfd37b9-335d-4af9-b593-7dcf4db98a26@126.com>
Date: Fri, 10 Jan 2025 17:40:23 +0800
From: Ge Yang <yangge1116@....com>
To: David Hildenbrand <david@...hat.com>, pbonzini@...hat.com
Cc: kvm@...r.kernel.org, linux-kernel@...r.kernel.org, michael.roth@....com,
 21cnbao@...il.com, baolin.wang@...ux.alibaba.com, liuzixing@...on.cn
Subject: Re: [PATCH] KVM: SEV: Pin SEV guest memory out of CMA area



在 2025/1/10 16:58, David Hildenbrand 写道:
> On 10.01.25 09:48, yangge1116@....com wrote:
>> From: yangge <yangge1116@....com>
> 
> I would suggest the title to be something like
> 
> "KVM: SEV: fix wrong pinning of pages"
> 
> Then you can describe that without FOLL_LONGTERM, the pages will not get 
> migrated out of MIGRATE_CMA/ZONE_MOVABLE, violating these mechanisms to 
> avoid fragmentation with unmovable pages, for example making CMA 
> allocations fail.
> 
> (CMA allocations failing is only one symptom of the missed usage of 
> FOLL_LONGTERM)
> 
>>
>> When pin_user_pages_fast() pins SEV guest memory without the
>> FOLL_LONGTERM flag, the pinned pages may inadvertently end up in the
>> CMA (Contiguous Memory Allocator) area. This can subsequently cause
>> cma_alloc() to fail in allocating these pages, due to the fact that
>> the pinned pages are not migratable.
>>
>> To address the aforementioned problem, we propose adding the
>> FOLL_LONGTERM flag to the pin_user_pages_fast() function. By doing
>> so, we ensure that the pages allocated will not occupy space within
>> the CMA area, thereby preventing potential allocation failures.
>>
>> Signed-off-by: yangge <yangge1116@....com>
>> ---
>>   arch/x86/kvm/svm/sev.c | 5 ++++-
>>   1 file changed, 4 insertions(+), 1 deletion(-)
>>
>> diff --git a/arch/x86/kvm/svm/sev.c b/arch/x86/kvm/svm/sev.c
>> index 943bd07..35d0714 100644
>> --- a/arch/x86/kvm/svm/sev.c
>> +++ b/arch/x86/kvm/svm/sev.c
>> @@ -630,6 +630,7 @@ static struct page **sev_pin_memory(struct kvm 
>> *kvm, unsigned long uaddr,
>>       unsigned long locked, lock_limit;
>>       struct page **pages;
>>       unsigned long first, last;
>> +    unsigned int flags = 0;
> 
> Why not set
> 
> unsigned int flags = FOLL_LONGTERM;
> 
>>       int ret;
>>       lockdep_assert_held(&kvm->lock);
>> @@ -662,8 +663,10 @@ static struct page **sev_pin_memory(struct kvm 
>> *kvm, unsigned long uaddr,
>>       if (!pages)
>>           return ERR_PTR(-ENOMEM);
>> +    flags = write ? FOLL_WRITE : 0;
> 
> 
> and here do
> 
> flags |= write ? FOLL_WRITE : 0;
> 
>> +
>>       /* Pin the user virtual address. */
>> -    npinned = pin_user_pages_fast(uaddr, npages, write ? FOLL_WRITE : 
>> 0, pages);
>> +    npinned = pin_user_pages_fast(uaddr, npages, flags | 
>> FOLL_LONGTERM, pages);
>>       if (npinned != npages) {
>>           pr_err("SEV: Failure locking %lu pages.\n", npages);
>>           ret = -ENOMEM;
> 
> 
Ok, thanks.