Messages in this thread | | | Subject | Re: [RFC][PATCH 0/7] improve printk reliability | From | Peter Zijlstra <> | Date | Wed, 21 Dec 2011 12:23:24 +0100 |
| |
On Wed, 2011-12-21 at 11:57 +0100, Peter Zijlstra wrote: > Hi, > > So these are my current efforts to make my own console output more > reliable. I've hit every deadlock described in this series and a few > more. > > With these patches applied I can printk() and WARN() from inside > scheduler context and not have my machine deadlock, even without the > aid of early_printk(). > > While I'll go an remove the sole printk() I found in the scheduler > so that we don't have to rely on the console doing this, it is good > to have a reliable console that can do this. > > As it stands some of the patches, the semaphore one in particular, > are really too ugly to live so I won't push those but wanted to share > anyway. > > Debugging this was 'fun', and I really hope to not have to do it > again :-)
Also, these patches enable lockdep over printk(), allowing lockdep to detect previously undetected deadlocks with console implementations, irrespective of rq->lock.
| |