NVM Express SSD support on Jetson TX1

Hi All

I’ve been testing an NVM Express SSD connected to my Jetson TX1 board with some success and some issues. I wanted to report findings here so others can benefit and also ask about a couple of issues I see. Here are the steps I undertook:

  1. Stock Jetson TX1 install with 3.10.67-g3a5c467 kernel.

  2. On a remote host I cross-compiled the stock TX1 kernel after running menuconfig to turn on NVM Express support (as a module). I had to make a couple of changes as noted in other threads to get the code to compile but it went pretty smooth.

  3. scp’ed the nvme.ko from host to Jetson board.

  4. sudo insmod nvme.ko

  5. Installed the nvme-cli tool from source (GitHub - linux-nvme/nvme-cli: NVMe management command line interface.). There is a Ubuntu PPA for this but it does not support arm or aarch64 yet. I am working with Canonical to get that fixed.

  6. sudo nvme list recognizes the attached device (a PMC NVM Express NVRAM card). Note this is a DRAM backed SSD which is why capacity seems small.

batesste@tegra-ubuntu:~$ sudo nvme list
Node Model Version Namepace Usage Format FW Rev


/dev/nvme0n1 MTR_MLC_TS_4GB 1.0 1 0.00 B / 4.26 GB 512 B + 0 B 2.05.36

  1. Installed a EXT4 filesystem on the SSD and issue some IO using both dd and fio. All seems to work OK but performance seems a bit low (ping me if you want performance numbers).

  2. I noticed messages like this in dmesg:

[ 7760.015863] process_cq did something
[ 7760.017047] mc-err: [mcerr] (0) csw_afiw: EMEM address decode error
[ 7760.023479] mc-err: [mcerr] status = 0x20010031; addr = 0x00000000
[ 7760.030281] mc-err: [mcerr] secure: no, access-type: write, SMMU fault: none

Does anyone have any idea what this is?

Thanks

Stephen

Hi sbates,
From the dmesg, it’s MC errors related with SMMU driver, the IO is not able to be the max performance, but still can work as what you’re seeing now,
We are currently investigating the issue. Once any progress, we will update to you.

Thanks

Hi kaycc

Thanks for this update. I missed it when you posted it. Just to add to my original post, I have now tested 4 different NVMe SSDs from different vendors and I see the same messages on each one. Let me know when you have an update in terms of a new kernel I can try or even just a replacement SMMU module if you have it.

Cheers

Stephen