linux-kernel - Re: [BUG] Crash on x86_32 for: mm: page_alloc: avoid merging non-fallbackable pageblocks with others

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: <B7789993-75EC-4F74-B4E6-AF1CC2CBD9D4@nvidia.com>
Date:   Wed, 30 Mar 2022 16:29:28 -0400
From:   Zi Yan <ziy@...dia.com>
To:     Steven Rostedt <rostedt@...dmis.org>,
        Linus Torvalds <torvalds@...ux-foundation.org>
Cc:     LKML <linux-kernel@...r.kernel.org>,
        Mel Gorman <mgorman@...hsingularity.net>,
        David Hildenbrand <david@...hat.com>,
        Vlastimil Babka <vbabka@...e.cz>,
        Mike Rapoport <rppt@...ux.ibm.com>,
        Oscar Salvador <osalvador@...e.de>,
        Andrew Morton <akpm@...ux-foundation.org>,
        Linux-MM <linux-mm@...ck.org>
Subject: Re: [BUG] Crash on x86_32 for: mm: page_alloc: avoid merging non-fallbackable pageblocks with others

On 30 Mar 2022, at 16:05, Linus Torvalds wrote:

> On Wed, Mar 30, 2022 at 12:42 PM Steven Rostedt <rostedt@...dmis.org> wrote:
>>
>> I started testing new patches and it crashed when doing the x86-32 test on
>> boot up.
>>
>> Initializing HighMem for node 0 (000375fe:0021ee00)
>> BUG: kernel NULL pointer dereference, address: 00000878
>> #PF: supervisor read access in kernel mode
>> #PF: error_code(0x0000) - not-present page
>> *pdpt = 0000000000000000 *pde = f0000000f000eef3
>> Oops: 0000 [#1] PREEMPT SMP PTI
>> CPU: 0 PID: 0 Comm: swapper Not tainted 5.17.0-test+ #469
>> Hardware name: MSI MS-7823/CSM-H87M-G43 (MS-7823), BIOS V1.6 02/22/2014
>> EIP: get_pfnblock_flags_mask+0x2c/0x36
>> Code: 6d ea ff 55 89 e5 56 89 ce 53 8b 18 89 d8 c1 eb 1e e8 f7 fb ff ff 69 db c0 02 00 00 89 c1 89 c2 c1 ea 05 8b 83 7c d7 79 c1 5b <8b> 04 90 d3 e8 21 f0 5e 5d c3 55 89 e5 57 56 89 d6 53 89 c3 64 a1
>
> The whole function is in that Code: thing, and it decodes to:
>
>    0: 55                    push   %ebp
>    1: 89 e5                mov    %esp,%ebp
>    3: 56                    push   %esi
>    4: 89 ce                mov    %ecx,%esi
>    6: 53                    push   %ebx
>    7: 8b 18                mov    (%eax),%ebx
>    9: 89 d8                mov    %ebx,%eax
>    b: c1 eb 1e              shr    $0x1e,%ebx
>    e: e8 f7 fb ff ff        call   0xfffffc0a
>   13: 69 db c0 02 00 00    imul   $0x2c0,%ebx,%ebx
>   19: 89 c1                mov    %eax,%ecx
>   1b: 89 c2                mov    %eax,%edx
>   1d: c1 ea 05              shr    $0x5,%edx
>   20: 8b 83 7c d7 79 c1    mov    -0x3e862884(%ebx),%eax
>   26: 5b                    pop    %ebx
>   27:* 8b 04 90              mov    (%eax,%edx,4),%eax <-- trapping instruction
>   2a: d3 e8                shr    %cl,%eax
>   2c: 21 f0                and    %esi,%eax
>   2e: 5e                    pop    %esi
>   2f: 5d                    pop    %ebp
>   30: c3                    ret
>
> with '%eax' being NULL, and %edx being 0x21e.
>
> (The call seems to be to 'pfn_to_bitidx().isra.0' if my compiler does
> similar code generation, so it's out-of-lined part of pfn_to_bitidx()
> despite being marked inline)
>
> So that oops is that
>
>         word = bitmap[word_bitidx];
>
> line, with 'bitmap' being NULL (and %edx contains 'word_bitidx').
>
> Looking around, your 'config-bad' doesn't even have
> CONFIG_MEMORY_ISOLATION enabled, and so I suspect the culprit is this
> part of the change:
>
> -               if (unlikely(has_isolate_pageblock(zone))) {
>
> which used to always be false for that config, and now the code is
> suddenly enabled.
>
> Alternatively, that code just can't deal with highmem properly.
>
> But I didn't really analyze things, I'm mainly doing pattern matching here.
>
> Zi Yan - and all the people who ack'ed and reviewed this - please take
> a deeper look..
>

In the original code, it will jump back to continue_merging and still tries
to find the buddy. The crash means the found buddy is not valid, since its
pageblock migratetype is NULL. That seems to suggest the physical memory
range is not aligned to MAX_ORDER_NR_PAGES, which should not be the case.
But if (!page_is_buddy(page, buddy, order)) prevents further buddy merging.
I must be missing something.


Hi Steven,

Can you try the patch below to see if it fixes the crash? Thanks.

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index bdc8f60ae462..83a90e2973b7 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -1108,6 +1108,8 @@ static inline void __free_one_page(struct page *page,

                buddy_pfn = __find_buddy_pfn(pfn, order);
                buddy = page + (buddy_pfn - pfn);
+               if (!page_is_buddy(page, buddy, order))
+                       goto done_merging;
                buddy_mt = get_pageblock_migratetype(buddy);

                if (migratetype != buddy_mt


--
Best Regards,
Yan, Zi

Download attachment "signature.asc" of type "application/pgp-signature" (855 bytes)