Re: Kmemleak feature

Re: Kmemleak feature

[Pekka Enberg]

On Tue, 8 May 2012, Anbu Manikandan wrote:
> I am very much interested in learning about kmemleak feature in linux
> 3.0-RT kernel.
> I have already read the kmemleak document (Documentation/kmemleak.txt)
> provided in the kernel source.
> 
> Furthermore, I would like to understand the internal functionality of
> kmemleak source code.
> Could you please share some documents, links or presentations to
> understand the kmemleak internal functionality?

Catalin is the main author of kmemleak. I'm not aware of such 
presentations.

> One more thing, I have tried kmemleak test module in an ARM Target
> board(CortexA9 SMP Dual core Board) with 3.0-RT kernel and found that
> memory leak entries are not properly shown in
> sys/kernel/debug/kmemleak file.
> Instead of getting 25 memory leak entries in
> /sys/kernel/debug/kmemleak file, I have got only 5 memory leaks
> entries.
> May be the other memory leaks are called as false negatives as per the
> kmemleak document.
> Please suggest some way to overcome this problem.

I'm CC'ing LKML.

			Pekka

Re: Kmemleak feature

[Catalin Marinas]

On Wed, May 09, 2012 at 07:27:28AM +0100, Pekka Enberg wrote:
> On Tue, 8 May 2012, Anbu Manikandan wrote:
> > I am very much interested in learning about kmemleak feature in linux
> > 3.0-RT kernel.
> > I have already read the kmemleak document (Documentation/kmemleak.txt)
> > provided in the kernel source.
> > 
> > Furthermore, I would like to understand the internal functionality of
> > kmemleak source code.
> > Could you please share some documents, links or presentations to
> > understand the kmemleak internal functionality?
> 
> Catalin is the main author of kmemleak. I'm not aware of such 
> presentations.

There was one last year at LinuxCon in Prague:

https://events.linuxfoundation.org/images/stories/pdf/lceu11_marinas.pdf

-- 
Catalin

Re: Kmemleak feature

[Catalin Marinas]

On Thu, May 10, 2012 at 05:42:35PM +0100, Anbu Manikandan wrote:
> As suggested by you, I have executed Kmemleak on 2.6.35-kernel and 3.0-RT
> kernel.
> I have noticed that Kmemleak behavior is little different between 2.6.35 and
> 3.0-RT kernel for vmalloc allocation memory leak. There is no entry for vmalloc
> allocation memory leak in 3.0-RT kernel.
...
> I tried modifying min_count value for vmalloc allocation from "3" to "4" as
> below in 3.0-RT kernel:

I can reproduce the missing vmalloc report on a non-rt 3.4-rc3 kernel as
well, so something must have changed in the vmalloc code in the past
months. A suspecting commit is f5252e00 but I need to investigate
whether a third reference to the vmalloc'ed pointer is stored in the
generic code somewhere before we decide to bump the min_count to 4.

-- 
Catalin