On 21/01/2011 20:40, Mathieu Desnoyers wrote:
I'm setting the compilation off now. I will give it a whirl some time Saturday or Sunday.* Richard Mortimer (richm@oldelvet.org.uk) wrote: [...]P.S. I saw your followup mail so hopefully this matches what you have found!Thanks for the info! At first glance, it does not seem to contradict my findings. When you find time, can you have a try at v3 I just posted ?
Before I did that I created kernel/sched.s and have verified that the two error locations are accesses to it_func_ptr as you suspected.
ldx [%g1+%lo(__tracepoint_sched_wakeup)+28], %l0 !, it_func_ptr and ldx [%g3], %l2 !, it_func_ptrwith v3 of you patches applied it uses +32 to get the offset of it_func_ptr so that looks good.
Make sure to start tracing in your tests, as I think the unaligned access only happens when accessing the struct tracepoint fields below the "int state" field.
Will do. Regards Richard