Rock 5B plus - NVME detected only sometimes!

[Chris C]

Hello,

 

I've recently bought a rock 5B plus. I have a brand new crucial P310 nvme 1TB with the goal to boot from it.

 

As the title says, sometimes the nvme drive is properly detected, sometimes not. (Everytime the controller is visible in lspci, but only sometimes shows in lsblk).

When the drive is properly detected, I can write to it and it works fine, without issues.

I've noticed that it is more likely to be detected on a cold boot (ie. if I turn off the power and come back the next day, the first time it will probably show up).

 

I have the official radxa 65w PD power adapter (although I've also tried a lot more during the process with the same results).

 

I tried with the official radxa debian image, and with armbian, and the results are exactly the same on both.

 

Btw, I would like to mention that armbian noble gnome doesn't boot even from the sd card (blue led but no display output, I let it for over 10 minutes), only the minimal/iot works.

 

I don't know what else to try at this point, I can post any info / logs you may want.

 

Thanks!

[royk]

You could try to add one of these in armbianEnv.txt
extraargs=pci=pcie_gen2

extraargs=nvme_core.io_timeout=4294967295

extraargs=nvme_core.default_ps_max_latency_us=0

[Chris C]

Thanks for the suggestion, I appreciate it.

 

Unfortunately it didn't do anything. I've tried the args one at a time, combined, still nothing.

 

One thing I noticed, I can consistently get the nvme to properly show up, during the first boot of radxa's image. (ie. flashing the radxa image to an SD, booting for the first time to the desktop, the nvme will be detected and work properly). Rebooting after that is a hit or miss and generally only cold boots sometimes work. I also tried the args you suggested in the radxa official image and still the same.

 

Btw for everything I mention, the armbian SPI image was installed.

My only assumption is that some irrelevant delay related to the first boot, gives time for the nvme to initialize? On subsequent boots I get "pcie link fail" during boot, and again, lspci works, lsblk doesn't.

 

Also I would like to use armbian ubuntu desktop which I can't at the moment, since the noble gnome image doesn't boot at all.

[royk]

@Chris C In the past there was a commit to fix sata issues with the Rock 5a, but this gave problems for nvme with Orange Pi. After that I also read some problems with nvme and the Rock 5B although I don't know if that was related.
https://forum.armbian.com/topic/46601-2484-latest-apt-upgrade-breaks-multiple-things-including-the-gnome-desktop/?do=findComment&comment=204829

 

 

This is a bit older kernel with RT patch where I reverted that commit. If that works you could compile a "normal" kernel with that commit reverted.
https://drive.google.com/drive/folders/1r76sUsfG_F8pq0pkzzPgEyzqdRG6fGcz?usp=drive_link

[Chris C]

Thanks @royk,

 

Can you please elaborate a bit on how to try out the kernel you sent?

Isn't it for orange pi 5? How do I build an image out of it for the rock 5b+?

[going]

23.08.2025 в 12:05, Chris C сказал:

I can post any info / logs you may want.

UART console log?

 

In fact, only the u-boot version and the kernel version are interested. On my devices with rk3588 (s) chips, old versions of u-boot were originally installed and they did not allow loading Armbian images. I used the instructions from the manufacturer's official website and recorded the bootloader provided by Armbian on the SPI chip using special software.

 

In other words, Armbian delivers his images from u-boot v2025.XX but my orangepi5plus was loaded using u-boot v2019 recorded on an SPI chip. And that may be the problem. After overwriting the new version, my device loads any images and PCI (NVME) devices register correctly. I checked several of these from different manufacturers.

Hope this information helps with your problem.

[royk]

@Chris C In case going's suggestion doesn't help, the kernel should be compatible with the Rock 5b. I don't think that you need to install the dtb if you're already running Armbian with vendor kernel 6.1.x. So just install with "sudo apt install ./linux-image*.deb and the headers if you need them to compile drivers that you need for testing. To be sure if the install went correct you should look if the symlinks are all to the correct kernel. "ls -l /boot" otherwise "sudo ln -sf correct-vmlinuz Image" and so on
But adding a log is always helpful indeed. Send it after the nvme stopped working.

Edited 1 hour ago by royk

[going]

32 минуты назад, Chris C сказал:

How do I build an image out of it for the rock 5b+?

./compile.sh

And choose this:

[eselarm]

7 hours ago, Chris C said:

On subsequent boots I get "pcie link fail" during boot, and again, lspci works, lsblk doesn't.

I have seen this many times on my ROCK3A and had to do with how the various 'PCI-E I/O' is operated. Some are multi-PHY SERDES, so can act also as SATA or USB3 or Mii. It very much matters which U-Boot and also overlays, which are kernel dependent. I now got 2-lane PCI-E for a Samsung 970 EVO 500MB  (M-Key slot) and SATA3 (E-Key slot) working simultaneously. Is a Radxa U-Boot in SPI and vendor 6.1.115 kernel, else it won't work.

 

For the RK3588 on my ROCK5B it works more flexible w.r.t. bootloader and kernel, 3 sets work, so also latest mainline generic distro Linux.

 

The ROCK5Bplus has 2x M-2 M-key, so that is different from a ROCK5B. I am not sure how lanes are routed, but did once check a similar issue RK35xx board and there the bifurcation seemed to be the problem. There are options I think for lspci where you can see how many lanes and that should also be in kernel log. Might need loglevel=7 to see it.

 

It might also be that the Crucial P310 just won't work reliably with the ROCK5Bplus, we have seen something this similar before for an other NVM-E + RK3588 SBC combinations (I think it was OPi5+ and same SSD as I have). Other NVM-E brand worked without issues.

 

Edited 55 minutes ago by eselarm