We are coming from https://gitlab.freedesktop.org/drm/amd/-/issues/1824. Mario Limonciello, however, thinks it is a kernel bug in the end. The problem is as follows. On the Lenovo P14s Gen2 AMD, there are three configurations available: * "Windows 10" mode selected in BIOS, "s2idle" selected in /sys/power/mem_sleep * "Linux" mode selected in BIOS, "s2idle" selected in /sys/power/mem_sleep * "Linux" mode selected in BIOS, "deep" selected in /sys/power/mem_sleep The first and last combination works as expected. However, the second combination ("deep" in BIOS, "s2idle" selected in Linux), causes weird resume crashes. In the original thread, Mario thinks that the sheer offering of "s2idle" when the BIOS is set to "deep" is a kernel bug - it shouldn't be offering "s2idle" on x86 when the firmware is set to "deep".
We decided not to stop exposing it but instead have some messaging if a user tries to suspend in a bad combination that there will be higher power consumption. https://github.com/torvalds/linux/blob/v5.18-rc4/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c#L1062