2.4 Hanging during on Apollo Lake

bgarwin

I'm aware of this workaround from the link I've pasted. While I'm OK with this solution, The Apollo Lake support is broken and need a proper fix.

Already reported to FreeBSD
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=222908
Feel free to add more information to the report.

digitalgimpus

Should this tweak be done in /boot/loader.conf.local?

bgarwin

I have it in /boot/loader.conf and it works fine.

digitalgimpus

That gets overwritten on update correct?

bgarwin

I thought it would, but it survived reboot and upgrade.

digitalgimpus

Ok, looks like I've been running ok now for a few days. Despite the installation headache, it seems like pretty solid hardware. Once FreeBSD is updated to better support I'd fully endorse it.

Thanks for the help.

Klaws

I had trouble booting up pfSense 2.4.1 on an Asrock J4205-ITX as well. It seemed to hang atTimecounter "HPET" frequency 19200000 Hz quality 950- but activating verbose mode in the boot menu revealed that it actually further, up to```
msi: Assinging MSI-X IRQ 267 to local APIC 0 vector 52


I (sort of) quickly found out that the board would boot in safe mode. The "safe mode" option which allowed it to boot was disabling SMP (verified by typing "set kern.smp.disabled=1", but that limiting the CPU to a single core was not an option I wanted to take.

What finally did work was adding```
machdep.disable_msix_migration=1
```to /boot/loader.conf.local (had to create this file - this will get read just like loader.conf, but it will not be overwritten by updates/upgrades). Note that some people attempting run various versions of FreeBSD on this board mentioned that it would only boot up when, additionally, the NIC was connected to a network. Not sure about that - I had disabled the internal Realtek NIC, but had a network connected to one of the dual Intel NIC's ports.

Some more details about my hardware:
Asrock J4205-ITX
2 x 8GB RAM
WD Green SSD 120GB
a dual 1GB NIC (Intel 82576-based, manufacturered/sold by 10Gtek®)

The NIC has two 1GB ports and supports PCIe 2.0 x1, so it fits nicely with the board's PCIe 2.0 x1 slot. I disabled the board's Realtek NIC in the BIOS. The BIOS version is 1.40 (tried all four versions, 1.10, 1.20, 1.30 and 1.40, and a BIOS downgrade did not solve the boot issue by itself, so I stuck to 1.40).

My BIOS settings:

Intel SpeedStep Technology: Enabled
CPU C States Support: C6 (tried all each setting, but had no effect on the boot issue, so I left it at default)
Enhanced Halt State(C1E): Enabled
Intel Virtualization Technology: Enabled
VT-d: Disabled
Power Gear: Normal Mode

DRAM Frequency: Auto
DRAM Voltage: Auto
Primary Graphics Adapter: Onboard
Share Memory: Auto (tried to set this to 64MB, but didn't work - perhaps the GPU wanted more, as it was connected to a 4K display via HDMI)
Onboard HD Audio: Disabled
Onboard LAN: Disabled (I decided to go for an Intel NIC right from the start)
PCIE1 Link Speed: Auto
WAN Radio: Disabled
BT Enabled: Disabled
Deep S5: Disabled
Restore on AC/Power Loss: Power On (the box sits behind a UPS…but you never know)
Good Night LED: Disabled

SATA Controller(s): Enabled
SATA Aggressive Link Power Management: Disabled
Hard Disk S.M.A.R.T: Enabled
ASMedia SATA3 Mode: Disabled (the SSD is connected to the SATA3_2 (Intel) port, easier to reach than the SATA3_1 port)

Serial Port 1: Disabled

Suspend to RAM: Auto
ACPI HPET Table: Enabled (this was one of the first I tried to diabled, but had no effect on the boot issue)
All Power-On-options are set to Disabled

Legacy USB Support: Enabled

CPU Fan 1 Setting: Full Speed (there is no CPU fan)
Chassis Fan 1 Setting: Automatic mode (the case does have a nce silent fan, so I decided to plug it in)
Chassis Fan 1 Temp Source: Monitor M/B
Target CPU Temperature: 50 °C/122 °F
Target Fan Speed: Level 9
Case Open Feature: Disabled

Secure Boot: Disabled
Intel(R) Platform Trust Technology: Enabled

Boot Option #1: UEFI OS (SATA 3_2)
Fast Boot: Disabled (this seems to with Windows 10 only anyway)
Boot From Onboard LAN: Disabled
Setup Prompt Timeout: 1
Boot Num-Lock: On
Boot Beep: Disabled (I don't a beeper connected anyway)
Full Screen Logo: Disabled
Boot Failure Guard Message: Enabled
CSM: Disabled

In the pfSense Advanced Setup, Cryptographic & Thermal Hardware, I enabled AES-NI and selected the Intel Core on-die sensors.

And that's what my complete load.conf.local looks like (based on the assumption that with 16GB RAM, I don't need to be conservative):

hw.igb.rxd=4096
hw.igb.txd=4096
net.pf.states_hashsize=2097152
net.pf.source_nodes_hashsize=65536
hw.igb.fc_setting=0
hw.igb.rx_process_limit="-1"
hw.igb.tx_process_limit="-1"
net.inet.tcp.syncache.hashsize="2048"
net.inet.tcp.syncache.bucketlimit="16"
net.inet.tcp.syncache.cachelimit="32768"
machdep.disable_msix_migration=1


The performande settings were mostly inspired by this post: https://forum.pfsense.org/index.php?topic=113496.0

bgarwin

Thank you Klaws, I tried your setting and everything seems to be working better than before.

However, I was hoping it would help with stability problem but it didn't. I still have the occasional lockups, which I can't determine the source of it. Have you encountered any instability problems with j4205, noting i have J3455B-ITX?

Klaws

I had no issues with hangs or instability so far. No reboot or anything for two weeks flat, since my post on October 30th, actually.

Load on my box is probably low - WAN speed is 100MBit/s and I have seen a maximum of 80.000 states or so. Well, squid taxes the CPU a bit, but the box is still not driven to the limits. The HD is an SSD - some WD "Green" 100GB thingy. RAM are two KVR16LS11/8 (giving me 16GB in total), which seems to test nice and stable with memtest x86 and the NIC is a "10Gtek für Intel E1G42ET, Intel 82576 Chip Gigabit Ethernet Konvergierter Netzwerkadapter (NIC), Dual RJ45 Kupfer Ports, PCI Express 2.0 X1" (currently sold out at German Amazon). I have disabled the onboard Realtek NIC in the BIOS and only use the two Intel ports from the add-on board. There are some rumors that Realtek actually has a stable driver for the NIC, but the default driver in most (or all) FreeBSD kernel versions appears to have stability issues, so I didn't start experimenting around and went for the Dual Intel path right away.

latimeria

Hi,
same problem of post #18 here on J3455M-E mobo; I was doing a fresh install because upgrading from 2.4.0 to 2.4.1 through the web console brought to an increase of load average always over 1.
Up to now the only solution was to install 2.3.4 version and wait for a solution on something that seems related to the last kernel.