rmustacc sure.

rmustacc but I have identified the problem is that we have already closed the errlog as part of the fmd shutdown, we were attempting to close fltlog, but got error while doing it, and since errlog is already closed, reporting that error did end up with dereferencing NULL pointer and crash

rmustacc so the basic question is, should we just check this pointer to be not NULL, or should fmd.running flag be reset earlier or something like that...

hello. for science(tm) i'm trying to find out real-world 'nice' handling on various systems. here is a test program which compiles on linux and the bsds, i expect it will also compile with minor tweaks on illumos

and here is a test script for it: dpaste.com/C9GB22FJ3

looking for someone(tm) with a *bare metal* illumos install who would be willing to run this

for the test to work it will need to be bound to one cpu, i don't know how to do it on illumos

on linux it would taskset

looks like pbind(8) will do it

pbind -e 3 sh script.sh

or so

cc -O0 -o cpuburner-prio cpuburner-prio.c

let me try it on illumos

make sure the box is mostly idle, or at least nothing messes with the target cpu

also thank you :)

68678: prio 19 ops 14796

68679: prio 0 ops 14821

can you verify on top or similar that they stay on the same cpu for the duration of the test?

oh no, it's being migrated. let me exec using pbind

ok, pbind -e 3 helped (3 is the most idle core)

69320: prio 0 ops 7419

69318: prio 19 ops 7430

ok this adds up, but is also broken :)

what's your uname

joyent_20230309T001400Z

(smartos)

thank you for testing

ChadGPT: the answer can vary dramatically depending on the choice of scheduling class.

TS: 196184: prio 19 ops 9968

196185: prio 0 ops 11399

IA: 196199: prio 0 ops 21377

196198: prio 19 ops 41

FSS: 196210: prio 0 ops 10656

196209: prio 19 ops 10693

RT: + nice -n 20 ./memory_burner 1048576

nice: invalid operation; scheduling class does not support nice

+ nice -n 0 ./memory_burner 1048576

nice: invalid operation; scheduling class does not support nice

:-)

(not sure why the impact is so high for IA vs TS as the documentation makes them seem near-identical)

IA should be just TS plus the ability for the X server to give boosts to processes with focus

it appears that when you put a process into IA it starts boosted.

tsoome: I noticed due to a mistake that libscf has at least one smatch warning that seems worth checking, and I guess clang might notice too. You might want to remove the SMOFF and take a glance

I'm starting to suspect that we use calloc() with the arguments the wrong way around more often than not

tsoome: see msg

i've noticed that

of course a big chunk of the uses of calloc are basically as 'zalloc()' for userland

I don't think any of the calloc implementations in the gate actually care - it's not like they try to align or pad the individual unit size, just multiply it out