PFSense on a DEC3840 (Netboard A20)

bpl294

This post is deleted!

bpl294

@stephenw10 here is some of the code changes that he was referring to:

"To make sure the OS can find the serial port, you need to remove some pre production acpi hack, this https://github.com/freebsd/freebsd-src/commit/35af9331 should do the trick.

The 10gbps network card needs a driver, which AMD upstreams to https://github.com/freebsd/freebsd-src/tree/main/sys/dev/axgbe, we do have some additional bug fixes which will likely land later on (you can find them now in our repo, https://github.com/opnsense/src)"

can these changes be applied via a module or does it need to be applied in the kernel itself?

stephenw10

Ok, that is in 2.6 so the first thing to do there is just try a 2.6 snapshot:
https://github.com/pfsense/FreeBSD-src/commit/a7c68340584c942792188ad50593d4ef15cc8982#diff-96de3fc05e938f0fd1d95debb8e797e7c1da4645867d1722e01b1eff85e17186

Steve

bpl294

@stephenw10 the most recent snapshot of 2.6 stalls at the spot below:

stephenw10

So slightly further?

bpl294

@stephenw10 yes:

2.5.2

2.6

bpl294

@stephenw10 support from Decisio metioned the following as well:

"Try to add the following to loader.conf.local hw.uart.console="io:0x3f8,br:115200" "

Would this get added to the loader.conf.local after install on a donor system? or can it be done pre-install?

stephenw10

You could set that at the loader prompt before it boots.

OK> set hw.uart.console="io:0x3f8,br:115200"
OK> boot

You could add it to loader.conf.local on the other system before moving is across.

That was the only line we saw which looked like it could be doing anything so it's certainly worth trying. Those are the default com1 values though.
Having read through the commits it might be setting it back to the default values after the detection quirks that were added for earlier EPYC systems were interfering.

Steve

bpl294

@stephenw10 that did the trick!! Thanks alot man!!

So the loader.conf.local file wouldn't be removed or changed during updates or anything right?

stephenw10

Ah, nice. Interesting that does anything.

Yes, that will be retained across updates.

Steve

bpl294

@stephenw10 One more thing, the cpu is being detected in PFSense as 2 cache groups x2cores. could this cause any performance impact? and is there a way to define what type of cores they are?

stephenw10

That's just how the CPU/BIOS reports it. I wouldn't expect it to make any difference.

More concerning is the fact it shows as running at 1200MHz. Does it ever rise from that?

You might need to enable powerd if you haven't already. Assuming there is a driver to support switching it.

Steve

bpl294

@stephenw10 Yea that was my other concern, it does not. If i disable PowerD, running " sysctl dev.cpu.0.freq " shows 2100. But with PowerD it stays at 1200 and that's with running multiple speed tests at gigabit over wireguard and not.

So i have disabled PowerD for now as the CPU temp is pretty steady at about 115 Fahrenheit.

stephenw10

Hmm, well if it changes when you enable it that shows it can set the frequency.
What profile did you use? Hi-Adaptive is usually the best to use.
It doesn't make a huge amount of difference to power consumption on modern CPUs anyway though.

Steve

bpl294

@stephenw10 I tried it Hi-Adaptive and Maximum, both stayed at 1200.

stephenw10

Hmm, I mean there is a possibility that CPU loading never gets high enough to start ramping up. Or that it's not detecting the loading correctly.

Try running:

sysctl -a | grep freq

So you see cpu frequency levels shown?

If so you can try killing powerd and setting the level manually.

Steve

bpl294

@stephenw10 it returns the following:

Would the command to set it manually be " set dev.cpu.0.freq_levels=2100/1890" ?

[2.6.0-DEVELOPMENT][root@core.sycamore]/root: sysctl -a | grep freq
kern.timecounter.tc.ACPI-fast.frequency: 3579545
kern.timecounter.tc.i8254.frequency: 1193182
kern.timecounter.tc.HPET.frequency: 14318180
kern.timecounter.tc.TSC.frequency: 2096114517
kern.ntp_pll.time_freq: 85052405328768
kern.ntp_pll.pps_freq: 89169788928000
device  cpufreq
kern.eventtimer.et.i8254.frequency: 1193182
kern.eventtimer.et.RTC.frequency: 32768
kern.eventtimer.et.HPET2.frequency: 14318180
kern.eventtimer.et.HPET1.frequency: 14318180
kern.eventtimer.et.HPET.frequency: 14318180
kern.eventtimer.et.LAPIC.frequency: 49907470
kern.acct_chkfreq: 15
net.inet.sctp.sack_freq: 2
debug.cpufreq.verbose: 0
debug.cpufreq.lowest: 0
debug.uart_poll_freq: 50
machdep.tsc_freq: 2096114517
machdep.i8254_freq: 1193182
machdep.acpi_timer_freq: 3579545
dev.cpufreq.0.%parent: cpu0
dev.cpufreq.0.%pnpinfo:
dev.cpufreq.0.%location:
dev.cpufreq.0.%driver: cpufreq
dev.cpufreq.0.%desc:
dev.cpufreq.%parent:
dev.hwpstate.0.freq_settings: 2100/1890 1700/1445 1200/990
dev.cpu.0.freq_levels: 2100/1890 1700/1445 1200/990
dev.cpu.0.freq: 2100

stephenw10

Ah, OK. Looks like it's working.

Kill the powerd process then:

sysctl dev.cpu.0.freq=1700

Or whichever speed you want.
The second number there is meant to be power consumption in mW. But I've always found it to be somewhat random!

Steve

bpl294

@stephenw10 that command did change the freq in the console:

stephenw10

Then maybe your test simply didn't load up the CPU enough.