On 2020-11-09 15:37:03 [+0100], Daniel Wagner wrote: > > this looks odd. So rt1 has 415, rt2 has 399 and rt3 has 420 so lets say > > it is the same. And then rt4 should reduce it to 340. The only part that > > could have some influence is the are the highmem/kmap patches. But for > > ARM64 these are still a nop and in both cases kmap_atomic() disables > > migrate & page-fault. > > > > Are you sure those numbers always reproducible and not something that > > goes wrong and sometimes it is captured at 300us and sometimes 400us. > > These test run only very short with hackbench as worlkload (5 minutes). > Though I running these tests now for more than year with v4.4-rt and > some times the newer -rt releases and I've never seen the latency > numbers above 200us unless something was broken. Given that 5 minutes is > not really long, I'll let those test run for longer to see if I get the > same results when they run for one hour. oki. > > I've been staring at the code of signaltest on Friday and I might need > > to stare longer to figure out what it does. > > I hear you. Anyway, I gave the current head a run with lazy preemption > disabled as you asked for. … > 5.10.0-rc2-rt4 vs 5.10.0-rc2-rt4(lazy preemption disabled) > > 0_cyclicdeadline t2-max-latency pass/pass 274.00/ 61.00 349.18% So the value went from 274us to 61us after disabling lazy-preempt? > cyclicdeadline seems heavily affected by the change. Sebastian
