AMD AM1 Restarts

[doverosx]

Hello FreeNAS folks,

I have moved from a Skt. 939+DDR configuration and made the jump to the AMD AM1 platform. My board is an ASRock AM1B-ITX paired with an 8GB DDR 1600 RAM stick and an AMD Sempron 3850. I would get the mountroot> shell with my USB3 stick plugged into my USB3 port, however, this was resolved by plugging the USB3 stick into the USB2 port. Moving along, I found that the system would crash, very shortly after start-up.

I would get time to start typing ping and partially through the octets, BAM, restart -> BIOS screen -> boot. Face Palm, repeat. In the past, I have had success simply moving the USB key over from one system to another so long as hardware support was reasonable.

Anyone else experiencing issues with this hardware? I'm running memtest right now and I'll continue disabling BIOS options as they seem relevant; grasping at straws really.

Regards,
Brodey

UPDATE: I have submitted a Bug Report (http://www.freebsd.org/cgi/query-pr.cgi?pr=189267&cat=) on FreeBSD's PR site.

 

[doverosx]

SOLUTION: Disable C6 state.

I'm thinking that the move to FreeBSD 10 will be needed for full hardware compatibility. I'm idling at 42'ish W....7-10W MORE than the old hardware *sigh*. I'm still getting the hwpstate error so common with FreeBSD and AMD CPUs even with the loader.conf option found on many forums. https://forums.freebsd.org/viewtopic.php?t=31451

I s'pose I'll keep this thread as the AM1 info thread.

 

[doverosx]

SOLUTION #2: Disable powerd

Enabling C6 state and CnQ in the BIOS are stable...so long as powerd is disabled.

This is a big time blocker/stopper for those looking at AM1 as a lower power solution (My system still idles at 42W or so in spite of HDD APM and spindown settings....even playing with manual powerd). I'll post up on FreeBSD to see if the devs will back-port support for the hardware from 10 to 9.2.

 

[TheSmoker]

I am thinking that you will save more power if you enable all the power saving feature in bios rather than use powerd which throttles down only one core.
Also you might wanth to lower the clock of the integrated video card to lowest value.

 

[cyberjock]

C6 isn't enabled on FreeNAS. ;)

 

[doverosx]

Really? The way it works (last time I checked) was that the CPUs APIC/HPET tables were presented to the kernel and the kernel would align that to the timer that it picks. In this case, I actually should be seeing HPET but I only see LAPIC/i8254/RTC. Not using the HPET could very well be the cause to the crashing and the higher power usage. Also...being an early adopter is likely working against me now that 9.2 is on the LEGACY tag.

BR here: http://www.freebsd.org/cgi/query-pr.cgi?pr=189267&cat=

 

[cyberjock]

Really? The way it works (last time I checked) was that the CPUs APIC/HPET tables were presented to the kernel and the kernel would align that to the timer that it picks. In this case, I actually should be seeing HPET but I only see LAPIC/i8254/RTC. Not using the HPET could very well be the cause to the crashing and the higher power usage. Also...being an early adopter is likely working against me now that 9.2 is on the LEGACY tag.

BR here: http://www.freebsd.org/cgi/query-pr.cgi?pr=189267&cat=

hehe.. this is where your knowledge exceeds mine. I've tried to find info on how the APCI/HPET stuff works and the "tradeoffs" that take place, but I'm unable to find good documentation on it. I have lots of theories of operation, but I prefer not to start discussing them in detail unless I'm 100% sure. There's already enough idiots talking out their butt that I don't need to contribute to misinformation on the internet. ;)

I do know that even when I booted up DOS 6.22 my system used only 1w more than when I booted up FreeNAS as well as Linux Mint from a livecd. Since I couldn't find an OS that could drop the wattage more I simply concluded that the best we're going to get is what we have and that hardware simply is handling it all automatically(or at least it is for my particular setups).

I really wish AMD was more active with supporting their hardware on open source OSes like FreeBSD. It's quite a shame that they laid off 1/2 their developers last summer. :(

When I asked one of the FreeNAS developers about deeper energy states they had 3 reasons why they only go to C1:

1. The vast majority(80%+) of all of your power savings is already made at C1.
2. The hardware handles itself well enough that it's really not necessary to play with deeper states that may or may not function properly on FreeBSD with given hardware.(you seem to perhaps fall into this category).
3. Below C1 there is a potentially unrestricted performance penalty for going into and out of those states. This may cause problems and may prevent FreeNAS from achieving it's desired function(stable always-on and always available file server).

On one hand I can appreciate the want and desire to use all of the power saving features available. But at the same time I think that because of the risks described in #2 and #3 and the potential small gain makes it not a worthwhile investment for developer resources.