On 11/21/19 4:52 PM, Jan Kiszka wrote: > On 21.11.19 16:31, Philippe Gerum wrote: >> On 11/19/19 6:39 PM, Philippe Gerum wrote: >>> On 11/19/19 5:46 PM, Philippe Gerum via Xenomai wrote: >>>> On 11/18/19 4:13 PM, Lange Norbert via Xenomai wrote: >>>>> Hello, >>>>> >>>>> Here's one of my deadlocks, the output seems interleaved from 2 >>>>> concurrent dumps, >>>>> I ran the crashlog through decode_stacktrace.sh. >>>>> >>>> >>>> Ok, I can reproduce this one, including in a vm. The symptom can be either >>>> a lockup, or recursive faults. I'm on it. >>>> >>> >>> #0b9e81807 seems to have introduced a regression. Something there may be >>> subtly at odds with the core scheduler logic. More later. >> >> Eh, no. #0b9e81807 only exposed a very serious and longstanding issue into >> the scheduler core, which may cause a CPU to pick threads from a remote >> runqueue in rare circumstances. And that bug is definitely mine. I need to >> review more code to make sure no more horror shows alike are waiting for >> prime time. This bug affects all Xenomai series, 3.1, 3.0, 2.x. >> > > Uh... good that we surfaced this now. Curious to see the result! >
Good news is that unlike initially suspected, Xenomai 2.x is _not_ affected by the thread-suspend bug. The synchronous bp feature increased the likeliness of triggering it, as XNDBGSTOP needs to be sent to multiple threads while single-stepping. -- Philippe.
