Re: AMD PMF: CCP PSP fails to reinitialize after hibernation causing TEE errors

[Mario Limonciello <superm1@kernel.org>]

On 10/22/25 3:04 PM, Lars Francke wrote:
> Hi,
> 
> while I've been active in the open-source ecosystem for years, this is
> my first report to the kernel. Please be gentle if I made mistakes.
> 
> I'm debugging hibernation & sleep issues with my HP ZBook Ultra G1a.
> It has a AMD RYZEN AI MAX+ PRO 395 w/ Radeon 8060S.
> 
> After resuming from hibernation I get this log line once a second:
>    amd-pmf AMDI0105:00: TEE enact cmd failed. err: ffff000e, ret:0
> 
> Right after hibernation and before the first of those errors I get:
>    ccp 0000:c3:00.2: tee: command 0x5 timed out, disabling PSP
> 
> On boot it looks good:
>    ccp 0000:c3:00.2: psp enabled
> 
> I had to google what TEE, CCP and PSP mean so it's partial guesswork
> on my end that these are related. It looks like PSP hardware(?)
> doesn't come out of hibernation properly and is disabled which means
> all further calls to it fail and I get the error message.
> It looks scary but seems harmless?
> 
> Commit 11e298f3548a6fe5e6ad78f811abfba15e6ebbc1 from 2024 has
> more or less exactly my error message but it doesn't seem
> to be fixed for this case.
> https://lore.kernel.org/all/20240216064112.962582-2-Shyam-sundar.S-k@amd.com/
> (I picked my recipients partially from that mail & maintainers.pl output)
> 
> My Environment
> - Kernel: 6.17.4-arch2-1
> - Hardware: HP ZBook Ultra G1a with AMD Ryzen AI Max+ Pro 395
> - Distribution: Arch Linux
> 
> I can (try to) test patches and provide additional debug output if needed.
> I have not tried any of the 6.18 RCs yet but I'll try to figure out
> how I can install them next. Looking at the git history I can't
> see any changes between 6.17.4 and 6.18 that seem relevant though.
> 
> Thanks for your help!
> 
> Cheers,
> Lars
> 

Thanks for the report!

Right now the PMF driver runs the same code during suspend/resume and 
hibernate/resume (there are common callbacks).

However different things happen to the platform during suspend/resume 
and hibernate/resume.

In suspend/resume the TA that was loaded into the TEE environment 
continues to be loaded.

In hibernate/resume the system will lose all context and the TA is no 
longer loaded.

So I feel the correct solution here is to add a dedicated restore() 
callback which reloads the TA before going down the common path.