Re: [PATCH 6/6] mm: Postpone the disabling of kmemleak early logging

[Catalin Marinas]

On 6 May 2014, at 19:15, Sasha Levin <sasha.levin@xxxxxxxxxx> wrote:
> On 05/06/2014 01:05 PM, Catalin Marinas wrote:
>> On Tue, May 06, 2014 at 04:20:27PM +0100, Sasha Levin wrote:
>>> On 05/02/2014 09:41 AM, Catalin Marinas wrote:
>>>> Currently, kmemleak_early_log is disabled at the beginning of the
>>>> kmemleak_init() function, before the full kmemleak tracing is actually
>>>> enabled. In this small window, kmem_cache_create() is called by kmemleak
>>>> which triggers additional memory allocation that are not traced. This
>>>> patch moves the kmemleak_early_log disabling further down and at the
>>>> same time with full kmemleak enabling.
>>>> 
>>>> Signed-off-by: Catalin Marinas <catalin.marinas@xxxxxxx>
>>>> Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
>>> 
>>> This patch makes the kernel die during the boot process:
>>> 
>>> [   24.471801] BUG: unable to handle kernel paging request at ffffffff922f2b93
>>> [   24.472496] IP: [<ffffffff922f2b93>] log_early+0x0/0xcd
>> 
>> Thanks for reporting this. I assume you run with
>> CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF enabled and kmemleak_early_log remains
>> set even though kmemleak is not in use.
>> 
>> Does the patch below fix it?
> 
> Nope, that didn't help as I don't have DEBUG_KMEMLEAK_DEFAULT_OFF enabled.
> 
> For reference:
> 
> $ cat .config | grep KMEMLEAK
> CONFIG_HAVE_DEBUG_KMEMLEAK=y
> CONFIG_DEBUG_KMEMLEAK=y
> CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE=400
> # CONFIG_DEBUG_KMEMLEAK_TEST is not set
> # CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF is not set

I assume your dmesg shows some kmemleak error during boot? I’ll send
another patch tomorrow.

The code around kmemleak_init was changed by commit 8910ae896c8c
(kmemleak: change some global variables to int). It looks like it
wasn’t just a simple conversion but slightly changed the
kmemleak_early_log logic which led to false positives for the kmemleak
cache objects and that’s what my patch was trying to solve.

The failure is caused by kmemleak_alloc() still calling log_early() much
later after the __init section has been freed because kmemleak_early_log
hasn’t been set to 0 (the default off is one path, another is the
kmemleak_error path).

Catalin--
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/