PCENGINES APU[1-7] Coreboot SeaBIOS Open Source Firmware

[Gary7]

My sources of optimization settings included calomel.org, teklager.se, and some notes from OPNsense forum.
I tried to consolidate the settings from the various sources, determine what was in agreement between sources, and implement best guess what might work for me.
Some settings work in /etc/sysctl.conf.local and some settings don't work in sysctl.conf.local. Those settings need to be added in the tunable page.
I did my configuration months ago. It's time that I review all the configuration settings to see if anything needs to be updated.
I did a lot of work for 200 Mbps Internet speed. My APU2D4 is basically "coasting" for my usage.

[Gary7]

I'm trying to determine why I don't see any decrease in performance.
When I run a speed test at the same time I'm executing top at the console, I consistently get a speed of 236 Mbps and top shows approximately 75% idle.

Just another thought. With BIOS v4.11.0.5, add these lines to /boot/loader.conf.local:
# Intel igb(4) kernel driver, preload
if_igb_load="YES"

# agree with Intel license terms
legal.intel_ipw.license_ack=1
legal.intel_iwi.license_ack=1
legal.intel_igb.license_ack=1

-----------
In the GUI, in the tunable page, add these variables:
dev.igb.0.eee_disabled      = 1
dev.igb.1.eee_disabled      = 1
dev.igb.2.eee_disabled      = 1

I suspect, but can't prove, that accepting the Intel license terms enables some additional features of the Intel NIC.

Then, reboot and see if it makes any difference at all.

[miroco]

Unless you have these wireless adaptors (ipw, iwi) I see little use in agreeing to those license terms, and as far as I can tell there are no license terms to agree on regarding the igb driver in the Tunables section and consequently not in a /boot/loader.conf.local - file. These are my findings.

ipw -- Intel PRO/Wireless 2100 IEEE 802.11 driver
iwi -- Intel PRO/Wireless 2200BG/2225BG/2915ABG IEEE 802.11 driver
igb -- Intel(R) PRO/1000 PCI Express Gigabit Ethernet adapter driver

https://www.freebsd.org/cgi/man.cgi?ipw
https://www.freebsd.org/cgi/man.cgi?iwi
https://www.freebsd.org/cgi/man.cgi?igb

The Teklager "OPNsense performance optimization for gigabit speed" guide seems to have been based on OPNsense 19.1 or 19.7 judging from the copyright information at the bottom of the top screen shot (OPNsense (c) 2014-2019 Deciso B.V.)

https://teklager.se/en/knowledge-base/opnsense-performance-optimization/

The settings referring to "hardware checksum offload, hardware segmentation offload and large receive offload" are all found under the Interface - Settings section.

That leaves the "tx" and rx" receiving queues settings.
hw.igb.rx_process_limit="-1"
hw.igb.tx_process_limit="-1"

Enabling EEE seems to come with a penalty.
WARNING: enabling EEE will significantly delay DHCP leases and the network interface will flip a few times on boot. https://en.wikipedia.org/wiki/Energy-Efficient_Ethernet

https://calomel.org/freebsd_network_tuning.html
https://www.thomas-krenn.com/de/wiki/OPNsense_igb_EEE_Funktion_deaktivieren

I think the BIOS is the joker in the game, especially since the release of v4.11.0.5.

[Gary7]

Yes. I agree that the iwi and ipw are for wireless adapters. Having those legal acknowledgments in the loader.conf.local file doesn't hurt, but doesn't help either.

I did a quick test by commenting out the 3 legal acknowledgement lines and rebooted.
I ran some Internet speed tests and I couldn't tell any difference. I was still getting almost exactly the same speed as before, 236 Mbps. That idea was a dead end. I thought that I would see some (small) difference in network speed even though I'm well under Gbit speed.

You do want to disable Energy Efficient networking, dev.igb.0.eee_disabled = 1

Didn't OPNsense 19.1 or 19.7 run on top of FreeBSD 11.2 (hardened) ?
OPNsense 20.1 uses FreeBSD 11.2 (hardened)
I'm looking for network (igb) settings that apply to FreeBSD 11.2.

I've optimized several network setting including the ones that you've mentioned.
In my case, I don't see the slowdown in v11.4.0.5. Maybe, I'm not running high enough Internet speed.

Someday, I'm sure that we will all figure this out.

[katamadone [CH]]

https://pcengines.github.io/firmware/2020/03/31/PC-Engines-Firmware-v4-11-0-6.html

There are issues adressed... Cpu related and so on

Gesendet von meinem POCO F1 mit Tapatalk

[Gary7]

With BIOS version 4.11.0.6, PCIe power management is disabled by default.

I haven't seen any performance difference since my Internet speed is a little over 200Mbps.
Has anybody tried enabling PCIe power management in the BIOS and determine how much network performance is affected at gigabit speed?

After upgrading v4.11.0.6, I kept PCIe power management at default (disabled) for a day or so. Then, I enabled PCIe power management. So far, I'm seeing about 3 degrees lower CPU temperature. Since this is my home firewall, the load on the hardware is quite low.

[k0ns0l3]

it runs with new bios as well as good 









With friendly greetings k0ns0l3

[mik]

And bandwith overall?

[k0ns0l3]

I didn't notice anything  negative, everything ok   

With friendly greetings k0ns0l3

[hushcoden]

it runs with new bios as well as good 

Where did you find the widget "Hardware information" ?

Thanks.

[katamadone [CH]]

If I'm right, thats os-dmidecode

Gesendet von meinem POCO F1 mit Tapatalk

[hushcoden]

If I'm right, thats os-dmidecode

Indeed, thank you.

[Steve28]

I have an APU2C4 that I have been using for a few years with pfSense.  I recently moved to OPNSense, and in the process of building the new system, I wanted to make sure to optimize performance I stumbled upon this thread as well as the https://teklager.se/en/knowledge-base/opnsense-performance-optimization/ link.  I have the latest mainline BIOS installed:

PC Engines apu2
coreboot build 20202604
BIOS version v4.11.0.6

The TekLager site suggests setting both net.inet.tcp.tso and net.inet.udp.checksum to 1, but they were already set this way on my fresh OPNSense 20.1 install.  Presumably this is reflected with the checkboxes in Interfaces->Settings with both Disable hardware checksum offload and Disable hardware TCP segmentation offload being unchecked.

Teklager also suggest setting the following:

hw.igb.rx_process_limit="-1"
hw.igb.tx_process_limit="-1"
legal.intel_igb.license_ack="1"

However when I do that, I see the following during bootup:

Code: [Select]

sysctl: oid 'hw.igb.rx_process_limit' is a read only tunable
sysctl: Tunable values are set in /boot/loader.conf
sysctl: oid 'hw.igb.tx_process_limit' is a read only tunable
sysctl: Tunable values are set in /boot/loader.conf
SO here come the questions:

• Does the above text mean that these are not applied?
• ...and is there some way to GET them to apply?
• In Interfaces->Settings there is an option for Disable hardware large receive offload that is currently CHECKED.  Should I leave it that way?
• The TekLager site also mentions the CPU boost available to 1.4 Ghz in the firmware.  I found on a separate site that I need to set hint.p4tcc.0.disabled=1, hint.acpi_throttle.0.disabled=1, and hint.acpi_perf.0.disabled=1.  Any reason I should NOT do that?

Thanks!

[hushcoden]

By reading this article https://github.com/pcengines/apu2-documentation/blob/master/docs/apu_CPU_boost.md I understood that (but I might be wrong):

1) since coreboot v4.9.0.2 the CPU is capable of running up to 1.4GHz as long as this option has been enabled in the BIOS and system modify/increase the CPU frequency automatically based on actual workload...

2) the three 'tunetables' you mentioned are simply to allow the system to visualise the true frquency of the CPU otherwise (maybe a bug) it will still show 1GHz...

[Steve28]

If I am understand you correctly, that means by installing 4.11.0.6, I am automatically getting the boost frequency without taking any other action?

Also, any insight into the other questions?