amdxdna: firmware <-> driver compatibility

amdxdna: firmware <-> driver compatibility

[Helmut Grohne]

Hello,

I recently attempted updating my linux-firmware. This included updating
amdnpu/17f0_10/npu.sbin.1.0.0.63 to amdnpu/17f0_10/npu.sbin.1.1.2.64.
Once loading the new firmware, I see the following messages on a Debian
6.19.6+deb14+1-amd64 kernel:

amdxdna 0000:66:00.1: enabling device (0000 -> 0002)
amdxdna 0000:66:00.1: [drm] *ERROR* aie2_check_protocol: Incompatible firmware protocol major 7 minor 2
amdxdna 0000:66:00.1: [drm] *ERROR* aie2_hw_start: firmware is not alive
amdxdna 0000:66:00.1: [drm] *ERROR* aie2_smu_exec: smu cmd 4 failed, 0xff
amdxdna 0000:66:00.1: [drm] *ERROR* aie2_smu_fini: Power off failed, ret -22
amdxdna 0000:66:00.1: [drm] *ERROR* aie2_init: start npu failed, ret -22
amdxdna 0000:66:00.1: [drm] *ERROR* amdxdna_probe: Hardware init failed, ret -22
amdxdna 0000:66:00.1: probe with driver amdxdna failed with error -22

This indicates that the updated firmware is incompatible with a 6.19
series kernel. A web search suggests that a 7.0+ kernel requires a npu
firmware >= 1.1. This is an ABI break. When the interface changes in
incompatible ways, the firmware filename should change to avoid such
confusion. The firmware upgrade should be supplying both firmware
versions as long as 6.19 remains supported.

Helmut

Re: amdxdna: firmware <-> driver compatibility

[Lizhi Hou]

Hi Helmut,


Thanks for reporting this. Does your linux-firmware have this change?

https://gitlab.com/kernel-firmware/linux-firmware/-/commit/58cf579b98c0c4878deeda1acb5db70e699875b5

Under the same firmware directory, there should be npu_7.sbin which 
links to npu.sbin.1.1.2.64. This is for the latest driver.

And npu.sbin still links to npu.sbin.1.0.0.63 which is for previous drivers.


Could you check if the symbol links are created as expected under 
amdnpu/17f0_10/?


Thanks,

Lizhi

On 3/20/26 09:26, Helmut Grohne wrote:
> Hello,
>
> I recently attempted updating my linux-firmware. This included updating
> amdnpu/17f0_10/npu.sbin.1.0.0.63 to amdnpu/17f0_10/npu.sbin.1.1.2.64.
> Once loading the new firmware, I see the following messages on a Debian
> 6.19.6+deb14+1-amd64 kernel:
>
> amdxdna 0000:66:00.1: enabling device (0000 -> 0002)
> amdxdna 0000:66:00.1: [drm] *ERROR* aie2_check_protocol: Incompatible firmware protocol major 7 minor 2
> amdxdna 0000:66:00.1: [drm] *ERROR* aie2_hw_start: firmware is not alive
> amdxdna 0000:66:00.1: [drm] *ERROR* aie2_smu_exec: smu cmd 4 failed, 0xff
> amdxdna 0000:66:00.1: [drm] *ERROR* aie2_smu_fini: Power off failed, ret -22
> amdxdna 0000:66:00.1: [drm] *ERROR* aie2_init: start npu failed, ret -22
> amdxdna 0000:66:00.1: [drm] *ERROR* amdxdna_probe: Hardware init failed, ret -22
> amdxdna 0000:66:00.1: probe with driver amdxdna failed with error -22
>
> This indicates that the updated firmware is incompatible with a 6.19
> series kernel. A web search suggests that a 7.0+ kernel requires a npu
> firmware >= 1.1. This is an ABI break. When the interface changes in
> incompatible ways, the firmware filename should change to avoid such
> confusion. The firmware upgrade should be supplying both firmware
> versions as long as 6.19 remains supported.
>
> Helmut
>

Re: amdxdna: firmware <-> driver compatibility

[Helmut Grohne]

Hi Lizhi,

On Mon, Mar 23, 2026 at 09:25:01AM -0700, Lizhi Hou wrote:
> Thanks for reporting this. Does your linux-firmware have this change?
> 
> https://gitlab.com/kernel-firmware/linux-firmware/-/commit/58cf579b98c0c4878deeda1acb5db70e699875b5

No. My report is a duplicate of
https://gitlab.freedesktop.org/drm/amd/-/issues/5009, which is the one
resolved via the commit you point at.

I'll take it from here and ensure that Debian includes that commit.

Helmut