Hi Dmitry,
On vr, 2016-02-05 at 17:06 +0100, Paul Bolle wrote:
> On vr, 2016-02-05 at 14:28 +0100, Dmitry Vyukov wrote:
> > I wonder why you don't see the leak I am seeing...
>
> So do I, for a few days now.
0) I finally managed to reliably trigger this leak on an i686, single
core machine (yet another ThinkPad).
1) Note that on that machine the leak was noticeable under the kmalloc
-512 line (struct ser_cardstate is 456 bytes on that machine). I'm
_guessing_ the kmalloc-2048 line, which I stared at for quite some time,
is only relevant here for x86_64 and when there's a bit of
instrumentation, or whatever, added to the slab objects (as they are in
your VM?).
2) More important was that this i686 machine ran a tree that actually
included the offending commit:
25cad69f21f5 ("base/platform: Fix platform drivers with no probe
callback").
See, after staring at the gigaset code for way too long I decided to
just use brute force. Ie, I bisected this issue.
2) Anyhow, thanks again for the report. Now on to the next question: how
on earth does that commit make ser_gigaset leak struct ser_cardstate?
To be continued,
Paul Bolle
