IRC channel logs
2023-03-05.log
back to list of logs
<damo22>ok i figured it out, PIT timer is redirected to IOAPIC pin 2 <damo22>so we can use the method youpi1 suggested <damo22>sent in a patch for this already <damo22>but smp is still slow with cpus > 1 <damo22>it must be related to the fine locking <damo22>something must be locked for too long and causes things to stall? <damo22>cpus spinning unnecessarily in loops <damo22>i wonder if there is a way to count the number of spins in each lock and print some stats <damo22>maybe one lock is dominating the rest <damo22>or the number of times the lock is taken <damo22>SUCCESS FAIL MASKED STACK TIME LOCK/CALLER <damo22>10790 total locks, 0 empty buckets <damo22>1251873 50/0 290938/23 248429/0 0/0 0xc10a4100(c10a4100) <damo22>DiffieHellman: it works but its currently so slow that the system is unusable <damo22>there is definitely something wrong with it <damo22>i think i fixed MACH_LOCK_MON lock monitoring <damo22>but i dont quite understand what the output means <damo22>youpi1: do you know what the lock_info_print lip() function is supposed to show? i hooked it up <damo22>looks like there are just too many successful locks <damo22>with -smp 1 it boots fast to this position: <damo22>SUCCESS FAIL MASKED STACK TIME LOCK/CALLER <damo22>15126 total locks, 0 empty buckets <damo22>2287490 0/0 441021/19 3898186/1 0/0 0xc10a4100(c10a4100) <damo22>with -smp 2 it boots slow to the same position: <damo22>SUCCESS FAIL MASKED STACK TIME LOCK/CALLER <damo22>15384 total locks, 0 empty buckets <damo22>2376675 480/0 471399/19 3981849/1 0/0 0xc10a4100(c10a4100) <damo22>it looks like roughly the same number of locks are being taken, but smp 2 has 480 failed locks and smp 1 has zero <damo22>SUCCESS FAIL MASKED STACK TIME LOCK/CALLER <damo22>15555 total locks, 0 empty buckets <damo22>2390319 941/0 480270/20 7974790/3 0/0 0xc10a4100(c10a4100) <damo22>SUCCESS FAIL MASKED STACK TIME LOCK/CALLER <damo22>18391 total locks, 0 empty buckets <damo22>2401894 1296/0 484905/20 4344562/1 0/0 0xc10a4100(c10a4100) <youpi1>damo22: I don't know anything about the lock mon, but the numbers you show look alright <youpi1>DiffieHellman: that's not really surprising