No devices found

[AbelVM]

It looks like my NPU is not being used at all but still I have no clear idea on how to fix it, as the documentation doesn't cover the case where you only need the SHIM libraries and I would rather not break anything 😅

• AMD Ryzen 9 AI 370
• Ubuntu 25.10
• Kernel 6.18.1 (theoretically, there is no need to install XDNA kernel drivers)
• linux-firmware: 20250901.git993ff19b-0ubuntu1.4
• The libraries at /opt/xilinx/xrt/lib are 2.19 version (installed by OS)
• xrt-smi validate returns [xrt-smi] ERROR: No devices found.

XRT build version: 2.19.0
Build hash: d5835aaa7fcdbcb749f4d837d9a0a605c1a4d312
Build date: 2025-05-01 20:03:07
Git branch: HEAD
PID: 23190
UID: 1000
[Mon Dec 29 07:57:13 2025 GMT]
HOST: aix1pro
EXE: /opt/xilinx/xrt/bin/unwrapped/xrt-smi
[xrt-smi] ERROR: No devices found.    

Might be related to

• New Linux kernel install instructions are missing #833
• Framework 13 AMD Ryzen AI 300; amdxdna_drm_submit_execbuf: Invalid arg bo count 0 #513
• Building the shim without the kernel driver #342

Should I just build latest tagged libraries and install them? Is there any specific step I should be aware of?

Any help is welcome

[AbelVM]

Reloading the XDNA module, makes the xrt-smi error message change:

➜  sudo modprobe -r amdxdna
➜  xrt-smi validate
ERROR: Please specify a device using --device option
 Available devices:
    
➜  xrt-smi examine                      
System Configuration
  OS Name              : Linux
  Release              : 6.18.1-061801-generic
  Machine              : x86_64
  CPU Cores            : 24
  Memory               : 93354 MB
  Distribution         : Ubuntu 25.10
  GLIBC                : 2.42
  Model                : AI Series
  BIOS Vendor          : American Megatrends International, LLC.
  BIOS Version         : 1.02

XRT
  Version              : 2.19.0
  Branch               : HEAD
  Hash                 : d5835aaa7fcdbcb749f4d837d9a0a605c1a4d312
  Hash Date            : 2025-05-01 20:03:07
  amdxdna              : unknown, unknown
  virtio-pci           : unknown, unknown

Device(s) Present
  0 devices found               
``

[AbelVM]

I've compiled and installed the SHIM library as per the readme and now I have my NPU listed

➜  ~ xrt-smi examine           
System Configuration
  OS Name              : Linux
  Release              : 6.18.1-061801-generic
  Machine              : x86_64
  CPU Cores            : 24
  Memory               : 93354 MB
  Distribution         : Ubuntu 25.10
  GLIBC                : 2.42
  Model                : AI Series
  BIOS Vendor          : American Megatrends International, LLC.
  BIOS Version         : 1.02
  Processor            : AMD Ryzen AI 9 HX 370 w/ Radeon 890M

XRT
  Version              : 2.21.0
  Branch               : HEAD
  Hash                 : 92d731b2d6d2660b1ad407bb1bc0341d9a081fc1
  Hash Date            : 2025-12-29 17:19:56
  amdxdna              : 6.18.1-061801-generic
  virtio-pci           : 6.18.1-061801-generic
  NPU Firmware Version : 1.0.0.63

Device(s) Present
|BDF             |Name          |
|----------------|--------------|
|[0000:c6:00.1]  |RyzenAI-npu4  |

But it fails in validation

➜  ~ xrt-smi validate --verbose     
Verbose: Enabling Verbosity
Verbose: Setting power mode to `performance`

Validate Device           : [0000:c6:00.1]
    Platform              : RyzenAI-npu4
    Power Mode            : Performance
-------------------------------------------------------------------------------
Test 1 [0000:c6:00.1]     : gemm 
    Description           : Measure the TOPS value of GEMM INT8 operations
    Error(s)              : DRM_IOCTL_AMDXDNA_CONFIG_HWCTX IOCTL failed (err=-95):
                            Operation not supported
    Test Status           : [FAILED]
-------------------------------------------------------------------------------
Test 2 [0000:c6:00.1]     : latency 
    Description           : Run end-to-end latency test
    Details               : Average latency: 62.0 us                            
    Test Status           : [PASSED]
-------------------------------------------------------------------------------
Test 3 [0000:c6:00.1]     : throughput 
    Description           : Run end-to-end throughput test
    Error(s)              : runlist failed execution (ERT_CMD_STATE_ABORT)      
    Test Status           : [FAILED]
-------------------------------------------------------------------------------
Validation failed

And now I get some weird logs:

➜  ~ sudo dmesg | grep -Ei "amdxdna|xrt|firmware"                
[   70.237647] systemd[1]: systemd-hibernate-clear.service - Clear Stale Hibernate Storage Info was skipped because of an unmet condition check (ConditionPathExists=/sys/firmware/efi/efivars/HibernateLocation-8cf2644b-4b0b-428f-9387-6d876050dc67).
[   70.451281] amdxdna 0000:c6:00.1: enabling device (0000 -> 0002)
[   70.553251] [drm] Initialized amdxdna_accel_driver 0.1.0 for 0000:c6:00.1 on minor 0
[   70.944511] mt7925e 0000:c2:00.0: WM Firmware Version: ____000000, Build Time: 20250721232943
[   71.568543] amdgpu 0000:c5:00.0: amdgpu: [drm] Optional firmware "amdgpu/isp_4_1_0.bin" was not found
[   71.595342] amdgpu 0000:c5:00.0: amdgpu: [drm] Loading DMUB firmware via PSP: version=0x09002F00
[   71.595514] amdgpu 0000:c5:00.0: amdgpu: [VCN instance 0] Found VCN firmware Version ENC: 1.24 DEC: 9 VEP: 0 Revision: 25
[  133.291712] amdxdna 0000:c6:00.1: xdna_mailbox.131: Message callback ret -22
[  133.291724] amdxdna 0000:c6:00.1: xdna_mailbox.131: Unexpected ret -22, disable irq
[  177.544809] amdxdna 0000:c6:00.1: xdna_mailbox.131: Message callback ret -22
[  177.544823] amdxdna 0000:c6:00.1: xdna_mailbox.131: Unexpected ret -22, disable irq
[  208.451540] amdxdna 0000:c6:00.1: xdna_mailbox.131: Message callback ret -22
[  208.451555] amdxdna 0000:c6:00.1: xdna_mailbox.131: Unexpected ret -22, disable irq
[  517.183232] amdxdna 0000:c6:00.1: xdna_mailbox.131: Message callback ret -22
[  517.183245] amdxdna 0000:c6:00.1: xdna_mailbox.131: Unexpected ret -22, disable irq
[  536.762499] amdxdna 0000:c6:00.1: xdna_mailbox.131: Message callback ret -22
[  536.762521] amdxdna 0000:c6:00.1: xdna_mailbox.131: Unexpected ret -22, disable irq
[ 1078.094206] amdxdna 0000:c6:00.1: xdna_mailbox.131: Message callback ret -22
[ 1078.094220] amdxdna 0000:c6:00.1: xdna_mailbox.131: Unexpected ret -22, disable irq
[ 1090.283501] amdxdna 0000:c6:00.1: xdna_mailbox.131: Message callback ret -22
[ 1090.283515] amdxdna 0000:c6:00.1: xdna_mailbox.131: Unexpected ret -22, disable irq
[ 1985.545056] [drm] Initialized amdxdna_accel_driver 0.1.0 for 0000:c6:00.1 on minor 0
[ 1995.986547] amdxdna 0000:c6:00.1: xdna_mailbox.131: Message callback ret -22
[ 1995.986561] amdxdna 0000:c6:00.1: xdna_mailbox.131: Unexpected ret -22, disable irq
[ 2029.738009] [drm] Initialized amdxdna_accel_driver 0.1.0 for 0000:c6:00.1 on minor 0
[ 2033.485635] amdxdna 0000:c6:00.1: xdna_mailbox.131: Message callback ret -22
[ 2033.485654] amdxdna 0000:c6:00.1: xdna_mailbox.131: Unexpected ret -22, disable irq
[ 2742.166439] amdxdna 0000:c6:00.1: [drm] *ERROR* aie2_get_info: Not supported request parameter 4
[ 2742.166571] amdxdna 0000:c6:00.1: [drm] *ERROR* aie2_get_info: Not supported request parameter 10
[ 2742.171551] amdxdna 0000:c6:00.1: [drm] *ERROR* aie2_get_info: Not supported request parameter 10    

[AbelVM]

On a side comment, it would be great if the documentation pointed that clinfo will crash after this procedure because of unwanted xilinx.icd .

Fix:

cd /etc/OpenCL/vendors/
sudo mv xilinx.icd xilinx.icd.disabled

[maxzhen]

The latest SHIM may require features that is not available yet in 6.18 kernel. @houlz0507 can help to confirm.

[houlz0507]

This is due to the latest shim requires the newer firmware which is not published to linux-firmware yet. Could you please check if the newer one is downloaded to /lib/firmware/amdnpu/17f0_10/npu.dev.sbin? If yes, you may manually change the symbol link to it by: rm /lib/firmware/amdnpu/17f0_10/npu.sbin.zst; ln -s npu.dev.sbin /lib/firmware/amdnpu/17f0_10/npu.dev.sbin. Then, reload the driver and check if the firmware version is newer then "1.0.0.63"

[stsoe]

On a side comment, it would be great if the documentation pointed that clinfo will crash after this procedure because of unwanted xilinx.icd .

Fix:

cd /etc/OpenCL/vendors/
sudo mv xilinx.icd xilinx.icd.disabled

Thank you for noting this, I was not aware of this problem and will look at why it doesn't work.

[AbelVM]

This is due to the latest shim requires the newer firmware which is not published to linux-firmware yet. Could you please check if the newer one is downloaded to /lib/firmware/amdnpu/17f0_10/npu.dev.sbin? If yes, you may manually change the symbol link to it by: rm /lib/firmware/amdnpu/17f0_10/npu.sbin.zst; ln -s npu.dev.sbin /lib/firmware/amdnpu/17f0_10/npu.dev.sbin. Then, reload the driver and check if the firmware version is newer then "1.0.0.63"

Nope, it looks npu.dev.sbin wasn't downloaded

➜   /lib/firmware/amdnpu ls -lash
total 88K
4.0K drwxr-xr-x   5 root root 4.0K Dec 29 17:30 .
 72K drwxr-xr-x 101 root root  68K Dec  2 20:19 ..
4.0K drwxr-xr-x   2 root root 4.0K Dec 29 17:30 1502_00
4.0K drwxr-xr-x   2 root root 4.0K Dec 29 17:30 17f0_10
4.0K drwxr-xr-x   2 root root 4.0K Dec 29 17:30 17f0_11  
➜   /lib/firmware/amdnpu/1502_00 ls -lash
total 80K
4.0K drwxr-xr-x 2 root root 4.0K Dec 29 17:30 .
4.0K drwxr-xr-x 5 root root 4.0K Dec 29 17:30 ..
 72K -rw-r--r-- 1 root root  72K Nov 17 11:46 npu.sbin.1.5.2.380.zst
   0 lrwxrwxrwx 1 root root   22 Nov 17 11:46 npu.sbin.zst -> npu.sbin.1.5.2.380.zst     
➜   /lib/firmware/amdnpu/17f0_10 ls -lash
total 80K
4.0K drwxr-xr-x 2 root root 4.0K Dec 29 17:30 .
4.0K drwxr-xr-x 5 root root 4.0K Dec 29 17:30 ..
 72K -rw-r--r-- 1 root root  72K Nov 17 11:46 npu.sbin.1.0.0.63.zst
   0 lrwxrwxrwx 1 root root   21 Nov 17 11:46 npu.sbin.zst -> npu.sbin.1.0.0.63.zst  
➜   /lib/firmware/amdnpu/17f0_11 ls -lash
total 80K
4.0K drwxr-xr-x 2 root root 4.0K Dec 29 17:30 .
4.0K drwxr-xr-x 5 root root 4.0K Dec 29 17:30 ..
 72K -rw-r--r-- 1 root root  72K Nov 17 11:46 npu.sbin.1.0.0.166.zst
   0 lrwxrwxrwx 1 root root   22 Nov 17 11:46 npu.sbin.zst -> npu.sbin.1.0.0.166.zst   

[houlz0507]

Could you download:
https://gitlab.com/kernel-firmware/drm-firmware/-/raw/amd-ipu-staging/amdnpu/17f0_10/1.6_npu.sbin.1.0.20.31
then create the symbol link to it?

[AbelVM]

Could you download: https://gitlab.com/kernel-firmware/drm-firmware/-/raw/amd-ipu-staging/amdnpu/17f0_10/1.6_npu.sbin.1.0.20.31 then create the symbol link to it?

Yep, I'll test that tomorrow indeed. But it looks like there's a newer version available, isn't it?

xdna-driver/tools/info.json

Line 26 in ff5edc9

"url": "https://gitlab.com/kernel-firmware/drm-firmware/-/raw/amd-ipu-staging/amdnpu/17f0_10/1.7_npu.sbin.1.1.0.59",

In the meantime, I've realized that no firmware at all was downloaded during the build

xdna-driver/build/build.sh

Line 293 in 17b6eec

DOWNLOAD_BINS_DIR=./amdxdna_bins

xdna-driver/build/build.sh

Line 80 in 17b6eec

local firmware_dir=${DOWNLOAD_BINS_DIR}/firmware

➜  amdxdna_bins git:(main) ls -lash
total 16K
4.0K drwxrwxr-x 4 abel abel 4.0K Dec 29 17:36 .
4.0K drwxrwxr-x 4 abel abel 4.0K Dec 29 17:51 ..
4.0K drwxrwxr-x 3 abel abel 4.0K Dec 29 17:36 download_raw
4.0K drwxrwxr-x 2 abel abel 4.0K Dec 29 17:52 vtd_archives  

Because I only built the SHIM library!!

xdna-driver/build/build.sh

Line 184 in 17b6eec

if [[ $skip_kmod == 0 ]]; then

So, it's somehow broken, because if you build and install an updated SHIM, it doesn't get the needed updated firmwares and everything fails later because of the lack of verions sync

[houlz0507]

Thanks for trying all these.
And I just noticed that it might not work for the latest 1.1 firmware due to the pending kernel patch. https://lore.kernel.org/dri-devel/20251219014356.2234241-2-lizhi.hou@amd.com/
So could you please manually download the one I posted previously for now.

[maxzhen]

It is by design not to include fw when you don't build driver. The fw and driver are sort of closely coupled.

[AbelVM]

Downloaded and linked 1.6_npu.sbin.1.0.20.31

➜  ~ cd  /lib/firmware/amdnpu/17f0_10/     
➜  17f0_10 ls -lash
total 540K
4.0K drwxr-xr-x 2 root root 4.0K Jan  6 11:12 .
4.0K drwxr-xr-x 5 root root 4.0K Dec 29 17:30 ..
460K -rw-rw-r-- 1 abel abel 458K Jan  6 11:11 1.6_npu.sbin.1.0.20.31
 72K -rw-r--r-- 1 root root  72K Dec  8 13:47 npu.sbin.1.0.0.63.zst
   0 lrwxrwxrwx 1 root root   21 Dec  8 13:47 npu.sbin.zst -> npu.sbin.1.0.0.63.zst
➜  17f0_10 sudo rm npu.sbin.zst
➜  17f0_10 sudo ln -s 1.6_npu.sbin.1.0.20.31 npu.dev.sbin
➜  17f0_10 ls -lash
total 540K
4.0K drwxr-xr-x 2 root root 4.0K Jan  6 11:31 .
4.0K drwxr-xr-x 5 root root 4.0K Dec 29 17:30 ..
460K -rw-rw-r-- 1 abel abel 458K Jan  6 11:11 1.6_npu.sbin.1.0.20.31
   0 lrwxrwxrwx 1 root root   22 Jan  6 11:31 npu.dev.sbin -> 1.6_npu.sbin.1.0.20.31
 72K -rw-r--r-- 1 root root  72K Dec  8 13:47 npu.sbin.1.0.0.63.zst 

Reload and... no device found

➜  17f0_10 sudo modprobe -r amdxdna
➜  17f0_10 sudo modprobe amdxdna
➜  17f0_10 source /opt/xilinx/xrt/setup.sh
Autocomplete not enabled for XRT tools
XILINX_XRT        : /opt/xilinx/xrt
PATH              : /opt/xilinx/xrt/bin:/opt/xilinx/xrt/bin:/opt/amdgpu-pro/bin:/opt/amdgpu/bin:/home/abel/bin:/home/abel/.local/bin:/usr/local/bin:/home/abel/.local/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/snap/bin:/snap/bin:/home/abel/.lmstudio/bin:/home/abel/.lmstudio/bin
LD_LIBRARY_PATH   : /opt/xilinx/xrt/lib:/opt/xilinx/xrt/lib
PYTHONPATH        : /opt/xilinx/xrt/python:/opt/xilinx/xrt/python     
➜ 17f0_10  xrt-smi examine                  
System Configuration
  OS Name              : Linux
  Release              : 6.18.1-061801-generic
  Machine              : x86_64
  CPU Cores            : 24
  Memory               : 93354 MB
  Distribution         : Ubuntu 25.10
  GLIBC                : 2.42
  Model                : AI Series
  BIOS Vendor          : American Megatrends International, LLC.
  BIOS Version         : 1.02
  Processor            : AMD Ryzen AI 9 HX 370 w/ Radeon 890M

XRT
  Version              : 2.21.0
  Branch               : HEAD
  Hash                 : 92d731b2d6d2660b1ad407bb1bc0341d9a081fc1
  Hash Date            : 2025-12-29 17:19:56
  amdxdna              : 6.18.1-061801-generic
  virtio-pci           : 6.18.1-061801-generic

Device(s) Present
  0 devices found  

It looks like the symlink with dev tag isn't used:

➜ 17f0_10 sudo dmesg | grep -Ei "amdxdna|xrt|firmware"    
[sudo: authenticate] Password: 
[   69.608720] systemd[1]: systemd-hibernate-clear.service - Clear Stale Hibernate Storage Info was skipped because of an unmet condition check (ConditionPathExists=/sys/firmware/efi/efivars/HibernateLocation-8cf2644b-4b0b-428f-9387-6d876050dc67).
[   69.843667] amdxdna 0000:c6:00.1: Direct firmware load for amdnpu/17f0_10/npu.sbin failed with error -2
[   69.843679] amdxdna 0000:c6:00.1: [drm] *ERROR* aie2_init: failed to request_firmware amdnpu/17f0_10/npu.sbin, ret -2
[   69.843686] amdxdna 0000:c6:00.1: [drm] *ERROR* amdxdna_probe: Hardware init failed, ret -2
[   69.843697] amdxdna 0000:c6:00.1: probe with driver amdxdna failed with error -2
[   70.377213] mt7925e 0000:c2:00.0: WM Firmware Version: ____000000, Build Time: 20251015213023
[   71.014958] amdgpu 0000:c5:00.0: amdgpu: [drm] Optional firmware "amdgpu/isp_4_1_0.bin" was not found
[   71.042510] amdgpu 0000:c5:00.0: amdgpu: [drm] Loading DMUB firmware via PSP: version=0x09002F00
[   71.042733] amdgpu 0000:c5:00.0: amdgpu: [VCN instance 0] Found VCN firmware Version ENC: 1.24 DEC: 9 VEP: 0 Revision: 25

So, let's try symlinking npu.sbin instead

➜  ~  cd  /lib/firmware/amdnpu/17f0_10/ 
➜  17f0_10 sudo rm npu.dev.sbin
➜  17f0_10 sudo ln -s 1.6_npu.sbin.1.0.20.31 npu.sbin  
➜  17f0_10 sudo modprobe -r amdxdna
➜  17f0_10 sudo modprobe amdxdna
➜  17f0_10 source /opt/xilinx/xrt/setup.sh    
Autocomplete not enabled for XRT tools
XILINX_XRT        : /opt/xilinx/xrt
PATH              : /opt/xilinx/xrt/bin:/opt/xilinx/xrt/bin:/opt/amdgpu-pro/bin:/opt/amdgpu/bin:/home/abel/bin:/home/abel/.local/bin:/usr/local/bin:/home/abel/.local/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/snap/bin:/snap/bin:/home/abel/.lmstudio/bin:/home/abel/.lmstudio/bin
LD_LIBRARY_PATH   : /opt/xilinx/xrt/lib:/opt/xilinx/xrt/lib
PYTHONPATH        : /opt/xilinx/xrt/python:/opt/xilinx/xrt/python
➜  17f0_10 xrt-smi examine  
System Configuration
  OS Name              : Linux
  Release              : 6.18.1-061801-generic
  Machine              : x86_64
  CPU Cores            : 24
  Memory               : 93354 MB
  Distribution         : Ubuntu 25.10
  GLIBC                : 2.42
  Model                : AI Series
  BIOS Vendor          : American Megatrends International, LLC.
  BIOS Version         : 1.02
  Processor            : AMD Ryzen AI 9 HX 370 w/ Radeon 890M

XRT
  Version              : 2.21.0
  Branch               : HEAD
  Hash                 : 92d731b2d6d2660b1ad407bb1bc0341d9a081fc1
  Hash Date            : 2025-12-29 17:19:56
  amdxdna              : 6.18.1-061801-generic
  virtio-pci           : 6.18.1-061801-generic
  NPU Firmware Version : 1.0.20.31

Device(s) Present
|BDF             |Name          |
|----------------|--------------|
|[0000:c6:00.1]  |RyzenAI-npu4  |

npu4 is back and using the 1.0.20.31 firmware! 🎉 But still... it fails the validation 😞

➜  17f0_10 xrt-smi validate --verbose
Verbose: Enabling Verbosity
WARNING: User doesn't have admin permissions to set performance mode. Running validate in Default mode
Validate Device           : [0000:c6:00.1]
    Platform              : RyzenAI-npu4
    Power Mode            : Default
-------------------------------------------------------------------------------
Test 1 [0000:c6:00.1]     : gemm 
    Description           : Measure the TOPS value of GEMM INT8 operations
    Error(s)              : DRM_IOCTL_AMDXDNA_CONFIG_HWCTX IOCTL failed (err=-95):
                            Operation not supported
    Test Status           : [FAILED]
-------------------------------------------------------------------------------
Test 2 [0000:c6:00.1]     : latency 
    Description           : Run end-to-end latency test
    Details               : Average latency: 49.0 us                            
    Test Status           : [PASSED]
-------------------------------------------------------------------------------
Test 3 [0000:c6:00.1]     : throughput 
    Description           : Run end-to-end throughput test
    Details               : Average throughput: 74564.0 op/s                    
    Test Status           : [PASSED]
-------------------------------------------------------------------------------
Validation failed 

[houlz0507]

Are you able to compile and try the latest linux kernel? The required driver changes for latest SHIM were upstream just few weeks ago.