Re: mmotm 2010-06-03-16-36 lots of suspected kmemleak

From: Dave Young
Date: Wed Jun 09 2010 - 05:19:14 EST


On Wed, Jun 9, 2010 at 10:37 AM, Dave Young <hidave.darkstar@xxxxxxxxx> wrote:
> On Tue, Jun 8, 2010 at 12:43 AM, Catalin Marinas
> <catalin.marinas@xxxxxxx> wrote:
>> On Mon, 2010-06-07 at 11:00 +0100, Dave Young wrote:
>>> On Mon, Jun 7, 2010 at 5:19 PM, Catalin Marinas <catalin.marinas@xxxxxxx> wrote:
>>> > On Mon, 2010-06-07 at 06:20 +0100, Dave Young wrote:
>>> >> On Fri, Jun 4, 2010 at 9:55 PM, Dave Young <hidave.darkstar@xxxxxxxxx> wrote:
>>> >> > On Fri, Jun 4, 2010 at 6:50 PM, Catalin Marinas <catalin.marinas@xxxxxxx> wrote:
>>> >> >> Dave Young <hidave.darkstar@xxxxxxxxx> wrote:
>>> >> >>> With mmotm 2010-06-03-16-36, I gots tuns of kmemleaks
>>> >> >>
>>> >> >> Do you have CONFIG_NO_BOOTMEM enabled? I posted a patch for this but
>>> >> >> hasn't been reviewed yet (I'll probably need to repost, so if it fixes
>>> >> >> the problem for you a Tested-by would be nice):
>>> >> >>
>>> >> >> http://lkml.org/lkml/2010/5/4/175
>>> >> >
>>> >> >
>>> >> > I'd like to test, but I can not access the test pc during weekend. So
>>> >> > I will test it next monday.
>>> >>
>>> >> Bad news, the patch does not fix this issue.
>>> >
>>> > Thanks for trying. Could you please just disable CONFIG_NO_BOOTMEM and
>>> > post the kmemleak reported leaks again?
>>>
>>> Still too many suspected leaks, results similar with
>>> (CONFIG_NO_BOOTMEM = y && apply your patch), looks like a little
>>> different from original ones? I just copy some of them here:
>>>
>>> unreferenced object 0xde3c7420 (size 44):
>>> Â comm "bash", pid 1631, jiffies 4294897023 (age 223.573s)
>>> Â hex dump (first 32 bytes):
>>> Â Â 05 05 00 00 ad 4e ad de ff ff ff ff ff ff ff ff Â.....N..........
>>> Â Â 98 42 d9 c1 00 00 00 00 50 fe 63 c1 10 32 8f dd Â.B......P.c..2..
>>> Â backtrace:
>>> Â Â [<c1498ad2>] kmemleak_alloc+0x4a/0x83
>>> Â Â [<c10c1ace>] kmem_cache_alloc+0xde/0x12a
>>> Â Â [<c10b421b>] anon_vma_fork+0x31/0x88
>>> Â Â [<c102c71d>] dup_mm+0x1d3/0x38f
>>> Â Â [<c102d20d>] copy_process+0x8ce/0xf39
>>> Â Â [<c102d990>] do_fork+0x118/0x295
>>> Â Â [<c1007fe0>] sys_clone+0x1f/0x24
>>> Â Â [<c10029b1>] ptregs_clone+0x15/0x24
>>> Â Â [<ffffffff>] 0xffffffff
>>
>> I'll try to test the mmotm kernel as well. I don't get any kmemleak
>> reports with the 2.6.35-rc1 kernel.
>
> Manually bisected mm patches, the memleak caused by following patch:
>
> mm-extend-ksm-refcounts-to-the-anon_vma-root.patch

Add following debug code:

void drop_anon_vma(struct anon_vma *anon_vma)
{
+ int a, b;
+ a = anonvma_external_refcount(anon_vma);
+ b = anonvma_external_refcount(anon_vma->root);
+ if (!a || !b) {
+ printk("drop_anon_vma: ref %d ", a);
+ printk("root ref %d\n", b);
+ }

result in below debug output:

[ 52.948614] drop_anon_vma: ref 0 root ref 0
[ 52.949770] Pid: 1403, comm: ps Not tainted 2.6.35-rc1-mm1 #29
[ 52.951386] Call Trace:
[ 52.952062] [<c14b1128>] ? printk+0x20/0x24
[ 52.953210] [<c10b409c>] drop_anon_vma+0x37/0xb3
[ 52.954503] [<c10b418c>] unlink_anon_vmas+0x74/0xc4
[ 52.955854] [<c10aeaa0>] free_pgtables+0x45/0x95
[ 52.957142] [<c10b00fd>] exit_mmap+0xab/0xfe
[ 52.958325] [<c102fafa>] ? exit_mm+0xdd/0xec
[ 52.959497] [<c102c25d>] mmput+0x49/0xcf
[ 52.960605] [<c102fb01>] exit_mm+0xe4/0xec
[ 52.961750] [<c103137c>] do_exit+0x1b4/0x64b
[ 52.962921] [<c1031875>] do_group_exit+0x62/0x85
[ 52.964212] [<c10318ab>] sys_exit_group+0x13/0x17
[ 52.965523] [<c14b344d>] syscall_call+0x7/0xb

So I guess the refcount break, either drop-without-get or over-drop

>
> cc Rik van Riel
>
>>
>> Can you send me your .config file? Do you have CONFIG_HUGETLBFS enabled?
>>
>> Thanks.
>>
>> --
>> Catalin
>>
>>
>
>
>
> --
> Regards
> dave
>



--
Regards
dave
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/