IRC channel logs

2023-07-27.log

back to list of logs

<damo22>i think we should have gnumach have per cpu timeouts
<damo22>luckyluke: what if we made gnumach have per-cpu timeouts, so the timer_lock isnt so contended?
<damo22>i think currently only cpu0 handles timeouts, but somehow softclock is being called on all cpus
<damo22>in smp with apic, lapic timer expires on all cpus so the timer interrupt happens on all cpus
<damo22>i dont know if softclock is mp safe
<azert>damo22 maybe interesting for you: https://openscholarship.wustl.edu/cgi/viewcontent.cgi?article=1381&context=cse_research
<azert>Shows how they replaced the former softclock in BSD
<azert>with a newer implementation
<azert>Check the comment: save the overhead of a software interrupt, so it now. Could you adopt that approach?
<damo22>im pretty sure a lot of the smp problem is the timeouts