axelera M2 and radxa 5B+

HI there ​@mardistal , great to see you here!

Hmm… I think that’s the same host that ​@Bohdanken was using a few weeks ago. Although I think Ubuntu worked there. It needed to be run natively but official support for Ubuntu 24.04 has been added since then.  Are you running this in a container, or natively?

I will ask around, but I’m not sure if Armbian is currently supported...

I am running natively

I did also ran it with the same host and system that Bohdanken used

One quick thought - I don’t suppose it could be a power issue at the M.2 slot? I don’t believe the Radxa 5B+ has been officially tested, so I’m not sure what kind of power delivery it has at the M.2 port…

Actually, thinking along similar lines, do we know if the host system itself requires a bit of a beefier PSU (like the Raspberry Pi does, for example). 65W or something.

All I’m thinking is this kind of instability could potentially be the result of insufficient power...

I am using a fix PSU 88W

Hi ​@mardistal ,

Welcome to our Community!

It looks like your system (Radxa ROCK 5B Plus) has the following m.2 connectors:

• 2x M.2 M Key Connector each with PCIe 3.0 2-lane

Since the m.2 connectors are gen 3.0, that means they are likely specced to ~8W of peak power. Metis has higher peaks by default, but you can tune that. Note that the 2 lanes also may mean degraded performance due to PCIe datatransfer bottlenecks. This bottleneck highly depends on your application.

Did you already find the MVM power limiters? You can play with it with the environment variable

AXELERA_CONFIGURE_BOARD=,30

where 30 means 30% power to the mvm unit. This limits the peak power that Metis uses. For OrangePI, I generally use 20 or 30.

You may need to do a full system shutdown before trying.

Alternatively, you could look into our pcie troubleshooting guide, which has some nice settings that you could try

https://support.axelera.ai/hc/en-us/articles/26261515755538-PCIe-Troubleshooting-Guide

Since Armbian is not on our officially supported list, you could try the following to see if the installation works

1. Check Driver Installation

First, verify the metis driver is properly installed:

# Check if driver is loaded
lsmod | grep metis

# Check driver version, should return a number like 1.5.1 or similar
cat /sys/class/metis/version

# If not loaded, try loading it manually
sudo modprobe metis

# Check if device node exists
ls -la /dev/metis*

2. Try Basic Recovery Commands

# Refresh PCIe and reload firmware
axdevice --refresh -v
axdevice --reload-firmware

# Try PCIe rescan
axdevice --pcie-rescan

Hi ​@mardistal ,

I see your host is Rockchip based, correct?

Can you share the output of 

sudo dmesg

with us as an attached file?

I think I know what it can be, as Rockchip hosts need to get the device tree modified to assign enough memory to communicate with the Metis card.

I will share further info if that is indeed the case after checking your sudo dmesg output.

Best,

Victor

when I installed the voyager SDK I had to also install linux-headers to compile the drivers because

the installation just give me the drivers source but no binary ,I also made  changes in the device tree may be I didn’t do it properly,but after this change I could see pcie BARs enumeration

after executing axdevice -v command the system just freeze ,even after using AXELERA_CONFIGURE_BOARD=,30

I also try to monitor dmsg via ssh but the system freeze too early  

even  pcie_asmp=off ,iommu=soft ,pci=noaer didn’t change any thing

Hi ​@mardistal ,

Thank you for sharing the logs. It looks like to device tree must have been modified correctly.

I see in the logs that you are using Armbian, which isn’t officially supported with our SDK unless you use Docker.

Is it possible that you image the host with the Joshua Riek Ubuntu (22.04 or 24.04), then you modify the device tree and share sudo dmesg and we take it from there?

Both Ubuntu 22.04 and Ubuntu 24.04 are officially supported for our SDK, so it’s better to use one of those OSs to debug any issues.

Best,

Victor

it’s sound to me that I’ve already try this configuration,but anyway I restart this one once again

it doesnt work with  Joshua Riek Ubuntu  either 2204 or 2404

something more under armbian

lspci -vv

0001:11:00.0 Processing accelerators: Axelera AI Metis AIPU (rev 02)

    Subsystem: Axelera AI Metis AIPU (rev 02)
    Control: I/O- Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx+
    Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
    Latency: 0
    Interrupt: pin A routed to IRQ 188
    Region 0: Memory at f5010000 (64-bit, non-prefetchable) [size=4K]
    Region 2: Memory at f6000000 (32-bit, non-prefetchable) [size=32M]
    Expansion ROM at f5000000 [virtual] [disabled] [size=64K]
    Capabilities: <access denied>
    Kernel driver in use: axl
    Kernel modules: metis

and  lsmod

Module                  Size  Used by
rfcomm                 61440  4
squashfs               53248  0
bnep                   24576  2
pwm_fan                20480  0
rk805_pwrkey           16384  0
joydev                 24576  0
rtw89_8852be           16384  0
metis                  86016  110475390
rtw89_8852b           282624  1 rtw89_8852be
rtk_btusb              69632  0
rtw89_pci              45056  1 rtw89_8852be
rtw89_core            249856  2 rtw89_8852b,rtw89_pci
btusb                  53248  0
btmtk                  16384  1 btusb
panthor               118784  3
btrtl                  24576  1 btusb
drm_gpuvm              24576  1 panthor
drm_exec               16384  2 drm_gpuvm,panthor
drm_shmem_helper       20480  1 panthor
btbcm                  24576  1 btusb
btintel                36864  1 btusb
gpu_sched              40960  1 panthor
input_leds             16384  0
lte_em05               16384  0
uio_pdrv_genirq        16384  0
uio                    20480  1 uio_pdrv_genirq
zram                   28672  3
zsmalloc               20480  1 zram
binfmt_misc            20480  1
sch_fq_codel           20480  2
fuse                  118784  3
dm_mod                110592  0
nfnetlink              20480  1
ip_tables              28672  0
ipv6                  458752  102
rockchip_cpuinfo       16384  0
r8169                  77824  0
 

NOTICE THIS:

metis                  86016  110475390

hello

Modify the device tree in joshua riek 24.04 distribution just remove the node tree at lspci and the 22.04 is not a solution either concerning stability.

in all the cases it’s seem to me that those distributions are not maintain or well maintain any more

then I have no interest waste more time trying to make it work on this ones.

I am not using my board just to make the axelera work on it,the armbian distribution that I am using is more stable and features proof

I will try investigate on differents boards and distribution that suites my needs,and if nothing work I will just drop that solution in the mean time and come back later for sure ,because your device deserve it.

regards

Hello ​@mardistal ,

Sorry for the delay in replying (Christams holidays). Thank you for providing the info and context. 

As native install on armbian is not officially supported yet, can you use Docker?

See here the guide on how to install on Docker:

https://support.axelera.ai/hc/en-us/articles/25953148201362-Install-Voyager-SDK-in-a-Docker-Container

Let us know how it goes and if it works inside Docker. FYI ​@Spanner 

Best,

Victor