Re: Probable Bug (or configuration error) in kmemleak

[Sankar P]

On Fri, Jun 18, 2010 at 6:55 PM, Sankar P <sankar.curiosity@xxxxxxxxx> wrote:
> On Fri, Jun 18, 2010 at 6:30 PM, Catalin Marinas
> <catalin.marinas@xxxxxxx> wrote:
>> On Fri, 2010-06-18 at 09:11 +0100, Sankar P wrote:
>>> On Thu, Jun 17, 2010 at 11:06 PM, Luis R. Rodriguez
>>> <lrodriguez@xxxxxxxxxxx> wrote:
>>> > On Thu, Jun 17, 2010 at 02:21:56AM -0700, Sankar P wrote:
>>> >> Hi,
>>> >>
>>> >> I wanted to detect memory leaks in one of my kernel modules. So I
>>> >> built Linus' tree  with the following config options enabled (on top
>>> >> of make defconfig)
>>> >>
>>> >> CONFIG_DEBUG_KMEMLEAK=y
>>> >> CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE=400
>>> >> CONFIG_DEBUG_KMEMLEAK_TEST=y
>>> >>
>>> >> If I boot with this kernel, debugfs is automatically mounted. But I do
>>> >> not have the file:
>>> >>
>>> >> /sys/kernel/debug/kmemleak
>>> >>
>>> >> created at all. There are other files like kprobes in the mounted
>>> >> /sys/kernel/debug directory btw. So I am not able to detect any of the
>>> >> memory leaks. Is there anything I am doing wrong or missing (or) is
>>> >> this a bug in kmemleak ?
>>> >>
>>> >> Please let me know your suggestions to fix this and get memory leaks
>>> >> reporting working. Thanks.
>>> >>
>>> >> The full .config file is also attached with this mail. Sorry for the
>>> >> attachment, I did not want to paste 5k lines in the mail. Sorry if it
>>> >> is wrong.
>>> >
>>> >
>>> > This is odd.. Do you see this message on your kernel ring buffer?
>>> >
>>> > Failed to create the debugfs kmemleak file
>>> >
>>>
>>> I dont see such an error in the dmesg output. But I got another
>>> interesting error:
>>>
>>> [    0.000000] kmemleak: Early log buffer exceeded, please increase
>>> DEBUG_KMEMLEAK_EARLY_LOG_SIZE
>>> [    0.000000] kmemleak: Kernel memory leak detector disabled
>>
>> You would need to increase DEBUG_KMEMLEAK_EARLY_LOG_SIZE. The default of
>> 400 seems ok for me but it may not work with some other kernel
>> configurations (that's a static array for logging memory allocations
>> before the kmemleak is fully initialised and can start tracking them).
>>
>>> But after that also, I see some other lines like:
>>>
>>> [    0.511641] kmemleak: vmalloc(64) = f7857000
>>> [    0.511645] kmemleak: vmalloc(64) = f785a000
>>
>> This is because you compiler the test module into the kernel
>> (DEBUG_KMEMLEAK_TEST). It's not kmemleak printing this but it's testing
>> module (which leaks memory on purpose).
>>
>>> The variable  DEBUG_KMEMLEAK_EARLY_LOG_SIZE was set to 400 by default.
>>> I changed it to 4000 and then 40000 (may be should try < 32567 ?) but
>>> still I get the same error message and the file
>>> /sys/kernel/debug/kmem* is never created at all.
>>
>> This shouldn't usually happen with values greater than 2000. From your
>> kernel log, the version seems to be 2.6.32. Do you have the same
>> problems with 2.6.35-rc3?
>>
>> Your .config seems to refer to the 2.6.35-rc3 kernel - are you checking
>> the right image?
>>
>
> Ah sorry. I am testing by ssh into a remote machine. After installing
> my kernel (after increasing the DEBUG_KMEMLEAK_EARLY_LOG_SIZE), Before
> rebooting, I forgot to change the default-kernel this time. I will
> check once again and will let you know. Thank you a lot for the
> "dmesg" pointer.
>
>

Thanks for your "dmesg" pointer, I was able to solve the problem. I
found that whenever I ran "defconfig", the value of 400 for
DEBUG_KMEMLEAK_EARLY_LOG_SIZE seem to be inadequate always. So I made
a trivial patch incrementing the value to 1000. This worked fine for
my debugging. So, take the patch I will send next, if you prefer.


-- 
Sankar P
http://psankar.blogspot.com
--
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/