Re: [BUG] s2idle: unrecoverable sleep on ThinkPad P16s Gen 4 AMD, (Strix Point) when more than 16 logical CPUs are online

[Fourhundred Thecat <400thecat@ik.me>]

On 2026-09-23 23:10, Mario Limonciello wrote:
> On 9/23/26 15:48, Fourhundred Thecat wrote:
>> the bug reproduces on 7.3-rc4.
>>
>> Built 7.3-rc4 from the torvalds snapshot, migrated my 6.18.51 config 
>> with make olddefconfig, IOMMU enabled, all 24 CPUs online, no IOMMU 
>> boot parameters. Same result: the machine suspends and never wakes, 
>> forced power off required.
>>
>> Both attempts recorded by the same script, neither produced a resume 
>> block:
>>
>>    === 2026-09-23T22:15:02 BEFORE
>>    cmdline:     (6.18.51, no iommu parameters)
>>    cpus online: 0-23
>>    iommu:       ivhd0
>>    ff_rt_clk:   0
>>    S0ix Entry Time: 0 / Exit Time: 0 / Residency Time: 0
>>    suspending with +30s alarm
>>
>>    === 2026-09-23T22:39:06 BEFORE
>>    cmdline:     (7.3.0-rc4, no iommu parameters)
>>    cpus online: 0-23
>>    iommu:       ivhd0
>>    ff_rt_clk:   0
>>    S0ix Entry Time: 0 / Exit Time: 0 / Residency Time: 0
>>    suspending with +30s alarm
>>
>> Both used a programmed RTC wakealarm at +30s and were left untouched 
>> for several minutes.
>>
>> Relevant details of the 7.3-rc4 build, so this is not a configuration 
>> difference:
>>
>>    CONFIG_AMD_IOMMU=y, CONFIG_IOMMU_PT=y, CONFIG_IOMMU_PT_AMDV1=y
>>    CONFIG_NR_CPUS=24, CONFIG_MODULES not set
>>    CONFIG_DEBUG_FS=y, CONFIG_PM_DEBUG=y, CONFIG_DYNAMIC_DEBUG=y
>>    CONFIG_AMD_MP2_STB=y, CONFIG_X86_MSR=y, CONFIG_DMIID=y
>>    CONFIG_LOG_BUF_SHIFT=20, CONFIG_RANDSTRUCT_NONE=y
>>    kernel is not tainted, builds with zero warnings
>>
>> Only 243 config lines differ between my 6.18.51 and 7.3-rc4 configs. 
>> Both commits you pointed me at are present in 7.3-rc4: rtc-cmos uses 
>> platform_get_irq_optional(), and iommu/amd carries the "No ACPI device 
>> matched UID" check. As on 6.18, that FW_BUG message does not appear 
>> here either.
>>
>> Two things worth noting about the 7.3 build specifically. The new 
>> IOMMU page table layer is in use (CONFIG_IOMMU_PT / IOMMU_PT_AMDV1 
>> replacing CONFIG_IOMMU_IO_PGTABLE), so that rework does not change the 
>> outcome. And AMD_PMF and DRM_ACCEL_AMDXDNA are not compiled into 
>> either kernel - DRM_ACCEL and AMD_SFH_HID are both off in my config - 
>> so neither driver is involved in any of these results at all, which is 
>> a stronger statement than the initcall_blacklist tests I reported 
>> earlier.
>>
>> Where that leaves things. With the IOMMU enabled and all 24 CPUs, on 
>> both 6.18.51 and 7.3-rc4:
>>
>>    lid switch, internal keyboard, power button, USB mouse with wakeup
>>    armed, and a programmed ACPI RTC alarm        all fail to wake
>>
>>    intremap=off, iommu=pt, amd_iommu_intr=legacy  all still hang
>>    pm_test freezer / devices / platform           all pass
>>    PC6 and CC6 enabled per amd-s2idle             yes
>>
>>    amd_iommu=off                                  wakes normally,
>>                                                   Last S0i3 Status 
>> Success,
>>                                                   13.3s residency,
>>                                                   pm_wakeup_irq 9
>>
>> Since it fails identically on 6.18 and 7.3 there is no working kernel 
>> version to bisect against. The only variable that changes the outcome 
>> is whether the IOMMU is enabled, and secondarily whether more than 16 
>> CPUs are brought up at boot.
> 
> To me this still feels like a BIOS bug.  I wouldn't discount the 
> possibility that think-lmi didn't apply a setting properly or something 
> like that.
> 
> Can you please try to manually go into your BIOS and reset BIOS default 
> settings?  Does it reproduce with IOMMU left on?

I did another test:

I bootet latest Debian live CD image before touching the BIOS, and it 
reproduces.

Debian live, kernel 7.1.12+deb14-amd64 #1 SMP PREEMPT_DYNAMIC Debian 
7.1.12-1 (2026-08-28). Stock distro kernel with modules, systemd and 
udev, every device driver bound, distro defaults throughout, nothing of 
mine involved except the BIOS settings. echo mem > /sys/power/state 
suspends and never wakes. Forced power off required, same as always.

So my custom kernel configuration is not the cause.

That now leaves three kernels reproducing it:

   6.18.51    my build, minimal config
   7.3-rc4    my build, migrated config
   7.1.12     stock Debian live image, distro config

I will do the BIOS defaults reset next, as you asked.