Sorry, I should have been more clear what part of the patch is related to the heisenbug: I suspect the heisenbug to be caused by the register cache being one to small and in the openmoko tree version of wm8753.c the bounds check in the register cache access code is wrong and allows access to one element behind the actual array size. So read/write to the WM8753_ADCTL2 register accesses unalloced and uninitalised memory. Which explains why it showed up only sometimes. - Lars

On Thu, Jul 2, 2009 at 10:51 AM, Timo Juhani Is a prebuilt image+modules available for that revision? It'd like to see how much it's really stable. Last week I started a prototype application to control wpa_supplicant over dbus, and my experiments are going slow because I'm not able to find a really stable kernel. As compile/deploy/test phases are frequent and as my tool enables/disables wifi resource, drives wpa_supplicant, starts/stops udhcpc at every lauch, after a bit every kernel crashes or w...