Re: Need help in bug in isolate_migratepages_range

[Vlastimil Babka <vbabka@suse.cz>]

On 02/03/2014 05:20 PM, Michal Hocko wrote:
> On Mon 03-02-14 14:29:22, Holger Kiehl wrote:
>> I have attached it. Please, tell me if you do not get the attachment.
>
> I hoped it would help me to get a closer compiled code to yours but I am
> probably using too different gcc.
> Anyway I've tried to check whether I can hook on something and it seems
> that this is a race with thp merge/split or something like that.
>
> [...]
>>>>    Jan 31 13:07:43 asterix kernel: BUG: unable to handle kernel NULL pointer dereference at 000000000000001c
>>>>    Jan 31 13:07:43 asterix kernel: IP: [<ffffffff810af0ac>] isolate_migratepages_range+0x32d/0x653
>>>>    Jan 31 13:07:43 asterix kernel: PGD 7d3074067 PUD 7d3073067 PMD 0
>>>>    Jan 31 13:07:43 asterix kernel: Oops: 0000 [#1] SMP
>>>>    Jan 31 13:07:43 asterix kernel: Modules linked in: drbd lru_cache coretemp ipmi_devintf bonding nf_conntrack_ftp binfmt_misc usbhid i2c_i801 sg ehci_pci i2c_core ehci_hcd uhci_hcd i5000_edac i5k_amb ipmi_si ipmi_msghandler usbcore usb_common [last unloaded: microcode]
>>>>    Jan 31 13:07:43 asterix kernel: CPU: 5 PID: 14164 Comm: java Not tainted 3.12.9 #1
>>>>    Jan 31 13:07:43 asterix kernel: Hardware name: FUJITSU SIEMENS PRIMERGY RX300 S4             /D2519, BIOS 4.06  Rev. 1.04.2519             07/30/2008
>>>>    Jan 31 13:07:43 asterix kernel: task: ffff8807d30b08c0 ti: ffff8807d30b2000 task.ti: ffff8807d30b2000
>>>>    Jan 31 13:07:43 asterix kernel: RIP: 0010:[<ffffffff810af0ac>]  [<ffffffff810af0ac>] isolate_migratepages_range+0x32d/0x653
>>>>    Jan 31 13:07:43 asterix kernel: RSP: 0000:ffff8807d30b3928  EFLAGS: 00010286
>>>>    Jan 31 13:07:43 asterix kernel: RAX: 0000000000000000 RBX: 000000000020ec09 RCX: 0000000000000002
>>>>    Jan 31 13:07:43 asterix kernel: RDX: 2c00000000008000 RSI: 0000000000000004 RDI: 000000000000006c
>>>>    Jan 31 13:07:43 asterix kernel: RBP: ffff8807d30b39f8 R08: ffff88083fbde390 R09: 0000000000000001
>>>>    Jan 31 13:07:43 asterix kernel: R10: 0000000000000000 R11: ffffea000733a000 R12: ffff8807d30b3a58
>>>>    Jan 31 13:07:43 asterix kernel: R13: ffffea000733a1f8 R14: 0000000000000000 R15: ffff88083ffe1d80
>>>>    Jan 31 13:07:43 asterix kernel: FS:  00007f9d9e72f910(0000) GS:ffff88083fd40000(0000) knlGS:0000000000000000
>>>>    Jan 31 13:07:43 asterix kernel: CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
>>>>    Jan 31 13:07:43 asterix kernel: CR2: 000000000000001c CR3: 00000007d3070000 CR4: 00000000000407e0
>>>>    Jan 31 13:07:43 asterix kernel: Stack:
>>>>    Jan 31 13:07:43 asterix kernel: 0000000000000009 ffff88083ffe16c0 ffffea00002e6af0 ffff8807d30b3998
>>>>    Jan 31 13:07:43 asterix kernel: ffff8807d30b2010 00ff8807d30b08c0 ffff8807d30b08c0 000000000020f000
>>>>    Jan 31 13:07:43 asterix kernel: 0000000000000000 000000000000083b 000000000000000a ffff8807d30b3a68
>>>>    Jan 31 13:07:43 asterix kernel: Call Trace:
>>>>    Jan 31 13:07:43 asterix kernel: [<ffffffff810a161f>] ? lru_add_drain_cpu+0x25/0x97
>>>>    Jan 31 13:07:43 asterix kernel: [<ffffffff810af687>] compact_zone+0x2b5/0x319
>>>>    Jan 31 13:07:43 asterix kernel: [<ffffffff810da586>] ? put_super+0x20/0x2c
>>>>    Jan 31 13:07:43 asterix kernel: [<ffffffff810afa4d>] compact_zone_order+0xad/0xc4
>>>>    Jan 31 13:07:43 asterix kernel: [<ffffffff810afaf5>] try_to_compact_pages+0x91/0xe8
>>>>    Jan 31 13:07:43 asterix kernel: [<ffffffff8109b92d>] ? page_alloc_cpu_notify+0x3e/0x3e
>>>>    Jan 31 13:07:43 asterix kernel: [<ffffffff8109da34>] __alloc_pages_direct_compact+0xae/0x195
>>>>    Jan 31 13:07:43 asterix kernel: [<ffffffff8109e45d>] __alloc_pages_nodemask+0x772/0x7b5
>>>>    Jan 31 13:07:43 asterix kernel: [<ffffffff810c85a3>] alloc_pages_vma+0xd6/0x101
>>>>    Jan 31 13:07:43 asterix kernel: [<ffffffff810d47e3>] do_huge_pmd_anonymous_page+0x199/0x2ee
>>>>    Jan 31 13:07:43 asterix kernel: [<ffffffff810b3884>] handle_mm_fault+0x1b7/0xceb
>>>>    Jan 31 13:07:43 asterix kernel: [<ffffffff8105dedc>] ? __dequeue_entity+0x2e/0x33
>>>>    Jan 31 13:07:43 asterix kernel: [<ffffffff8102d8c3>] __do_page_fault+0x3bd/0x3e4
>>>>    Jan 31 13:07:43 asterix kernel: [<ffffffff810bbe1a>] ? mprotect_fixup+0x1c9/0x1fb
>>>>    Jan 31 13:07:43 asterix kernel: [<ffffffff810aa0f0>] ? vm_mmap_pgoff+0x6d/0x8f
>>>>    Jan 31 13:07:43 asterix kernel: [<ffffffff810795f5>] ? SyS_futex+0x103/0x13d
>>>>    Jan 31 13:07:43 asterix kernel: [<ffffffff8102d8f3>] do_page_fault+0x9/0xb
>>>>    Jan 31 13:07:43 asterix kernel: [<ffffffff813d3672>] page_fault+0x22/0x30
>>>>    Jan 31 13:07:43 asterix kernel: Code: 00 41 f7 45 00 ff ff ff 01 0f 85 43 02 00 00 41 8b 45 18 85 c0 0f 89 37 02 00 00 49 8b 55 00 4c 89 e8 66 85 d2 79 04 49 8b 45 30 <8b> 40 1c 83 f8 01 0f 85 1b 02 00 00 49 8b 55 08 30 c0 48 85 d2
>>>>    Jan 31 13:07:43 asterix kernel: RIP  [<ffffffff810af0ac>] isolate_migratepages_range+0x32d/0x653
>>>>    Jan 31 13:07:43 asterix kernel: RSP <ffff8807d30b3928>
>>>>    Jan 31 13:07:43 asterix kernel: CR2: 000000000000001c
>>>>    Jan 31 13:07:43 asterix kernel: ---[ end trace fba75c5b0b9175ea ]---
>
> This seems to match:
>     17027:       49 8b 17                mov    (%r15),%rdx	# page->flags
>     1702a:       4c 89 f8                mov    %r15,%rax
>     1702d:       80 e6 80                and    $0x80,%dh	# PageTail test
>     17030:       74 04                   je     17036 <isolate_migratepages_range+0x2bf>
>     17032:       49 8b 47 30             mov    0x30(%r15),%rax	# page = page->first_page
>     17036:       8b 40 1c                mov    0x1c(%rax),%eax	<<< page->_count
>     17039:       ff c8                   dec    %eax
>
> Which seems to be inlined compound_head. DH is 0x80 so this is a tail
> page. This would suggest that tail page doesn't have firs_pages set up
> properly and it contains NULL.

It seems to come from balloon_page_movable() and its test 
page_count(page) == 1.

> But maybe I've just matched the code incorrectly. Could you try to
> disassemble your vmlinux a send the generated code, please?
>
> Something like
> objdump -d vmlinux > vmlinux.dis
> and cut out isolate_migratepages_range function. Or simply upload your
> vmlinux.dis somewhere so that we can download it.
>

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>