Hi, I filed https://gitlab.freedesktop.org/drm/amd/-/issues/5679 a couple of days ago and wanted to flag it here since it may be interesting to whoever owns the PSP init path on gfx1151.
Short version: on my ASUS ROG Flow Z13 GZ302EA (Ryzen AI MAX+ 395 / Radeon 8060S), amdgpu has never initialized. The PSP itself raises a memory fault while setting up its Trusted Memory Region, and hw_init of the psp block fails. The exact sequence, identical on every attempt: amdgpu 0000:65:00.0: [mmhub] page fault ... Faulty UTCL2 client ID: MP0 (0x6) amdgpu 0000:65:00.0: amdgpu: psp gfx command ID_LOAD_TOC(0x20) failed and response status is (0x0) amdgpu 0000:65:00.0: amdgpu: PSP tmr init failed! amdgpu 0000:65:00.0: amdgpu: hw_init of IP block <psp> failed -22 On top of that, the cleanup path then hits a vcn_v4_0_5_sw_fini Oops with IRQs disabled, which can hard-hang the machine. nomodeset is the only stable state. What makes it interesting: Windows 11 + Adrenalin works perfectly on the same unit, and at least one other GZ302EA on the same BIOS (.311) reportedly boots Linux fine — so this smells like per-unit PSP/firmware state rather than a generic driver bug. I reproduced it across 5 kernels (6.18 LTS through 7.2 — newer ones actually fail earlier), a BIOS re-flash via EZ Flash, live USB, and every relevant module parameter I could find. Full 837-line dmesg of the manual modprobe sequence is attached to the issue. Two questions I couldn't answer from the outside: 1. Is there any way to dump or reset the PSP NVRAM/state on Strix Halo from the OS or BIOS? (The firmware setup variables include an ASUS "AsusClearPspNvram" knob I have not touched yet — is that path known/safe?) 2. Is there an expected TMR base/size on this platform that the BIOS might be misprogramming on some units? The machine is set up for safe remote testing (SSH from a second laptop, amdgpu blacklisted at boot), so I'm happy to try patches or dump anything useful. Thanks, Abdulaziz AlMutairi
