> > On Sat, Dec 22, 2018 at 11:30:51PM -0800, Stephen Hemminger wrote: > > On Fri, 21 Dec 2018 20:14:20 -0600 > > Honnappa Nagarahalli <honnappa.nagaraha...@arm.com> wrote: > > > > > From: Dharmik Thakkar <dharmik.thak...@arm.com> > > > > > > Add API positive/negative test cases and functional tests. > > > > > > Signed-off-by: Malvika Gupta <malvika.gu...@arm.com> > > > Signed-off-by: Dharmik Thakkar <dharmik.thak...@arm.com> > > > Signed-off-by: Honnappa Nagarahalli <honnappa.nagaraha...@arm.com> > > > Reviewed-by: Gavin Hu <gavin...@arm.com> > > > > Just a thought, could you build stress tests like the kernel RCU tests? > > One worry is that RCU does not play well with blocking threads (and > sometimes preemption). Handling blocking threads is supported right now through register_thread/unregister_thread APIs. If a thread has to make a call to a blocking API, it is expected to unregister itself first. It will be improved further in V3. However, I am not sure what needs to be done for preemption. I would imagine that the threads will be scheduled back at some point (depending on the scheduling policy). If they were using the data structure the updater has to wait.
> > There are similar tests in the userspace RCU library, as well, which can be > found at http://liburcu. I looked at these tests. There is perftest/rperftest(reader only)/uperftest(updater only)/stresstest/benchmark. Currently, we have covered perftest/stresstest/benchmark pretty well (perf numbers need to be added). We will add rperftest and uperftest. > > Thanx, Paul
