On Tue, 23 Sep 2014 23:25:12 -0500 Brassow Jonathan <jbrassow@xxxxxxxxxx> wrote: > > Sorry, still not there yet. No?? I was *sure* I had it that time! > > I'm sorry I haven't had more time to spend on this. I'll try to get some help (perhaps from Heinz) and see if we can pitch-in instead of making you do all the work. Could you include /proc/PID/stack for the two md kernel threads? They are unlikely to differ from what we have seen before but as they are the strongest pointer to where the problem is, I like to be able to see them :-) Also I'd like to see the disassembly of raise_barrier() just to be certain which of the two 'wait's it is waiting in. One thing that might be useful is to change the various wait_event_lock_irq() calls to wait_event_lock_irq_cmd() with an extra argument 'dotrace(conf)' where dotrace() is a new function that first calls schedule_timeout(60*HZ); and then if that returns 0, prints out all fields that might be of interest. Also get it to print which 'wait' it was called from (e.g. pass a string to dotrace). Thanks, NeilBrown
Attachment:
signature.asc
Description: PGP signature
