[ltp] Memory leak with vmalloc?

Hamish Marson linux-thinkpad@linux-thinkpad.org
Thu, 20 Apr 2006 20:46:53 +0100


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Dave Hansen wrote:

> On Sun, 2006-03-26 at 21:53 +0100, Hamish Marson wrote:
>
>> Is anyone else finding the same thing? It doesn't sem to be
>> associated with susped/resume itself... Dumping /proc/meminfo
>> before & after shows no loss during the suspend/resume cycle...
>> At least not when I try by hand..
>
>
> Module code itself sits in vmalloc'd areas, and the e1000 driver
> also does some vmallocs of its own. I'd venture a guess that one
> of those is responsible, first, but no guarantees. Putting some
> printks around the vmallocs in those two places might be helpful.
>
> I can also write you a debugging patch to print out what all of the
> callers of each vmalloc area were. Would take an hour or two to
> put together...
>

Hi.

Thanks... I think I've found a way to stop it from happening... I
compiled 2.6.16.1 and changed the memory split to 2G/2G (With high mem
set to 4GB) And set 'Allocate 3rd level pagetables from highmem'.

VmallocTotal is now (With that kernel) at >500MB. Used doesn't seem to
grow. I haven't tried 2.6.16.1 with the same .config as 2.6.15.2 which
I know does have those problems... (I leaft 2.6.15.2 at default mem
split & 3rd level pagetables although it did have 4GB highmem support.


I'll try the patch you sent as well if you think it's still worth it,
but apart from ipw2100 not working in 2.6.16.1 it's going well (Hmm...
Maybe that's a problem with the highmem changes I did. I didn't think
of that one). I won't have time for a few days though. I have a TKD
grading this weekend to get through first.

> -- Dave
>

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFER+Ut/3QXwQQkZYwRApRZAKDQafq9+GIVrjW9+4xzdRDue/diQgCePL4o
29/Zcror6tSHAvoty6uJUDQ=
=7NW6
-----END PGP SIGNATURE-----