thanks for the review!

Make hw.acpi.spmc.intel_dsm_revision sysctl operate directly on Intel DSM set struct.

Make sched_preempt_locked entirely responsible for releasing thread lock and style change.

looks good!

Don't ever use Microsoft DSM UUID for getting constraints, as this isn't supported (checked on Intel acpidump).

Comment about changing sc->constraint_count and more style(9) improvements.

Remove struct acpi_spmc_private and just set DSM set flags directly on softc + style(9) improvements.

See changes/comments in D48387, but will add an interface for PMC suspend and resume which this will use.

Which kind of interrupts do you still see that prevent staying in S0ix? I think you mentioned some IPIs at some point, but which ones? Can we determine their cause?

Remove SPMC callbacks on struct acpi_softc. I will replace them with new pmc_suspend and pmc_resume methods in another revision. This leaves the acpi_spmc_suspend and acpi_spmc_resume (renamed from acpi_spmc_enter/exit) functions unused at the moment. Lmk if this is okay or if you'd rather I add these in a different revision too.

fixed most of the issues highlighted. Will address the allocating-memory-in-probe-function pattern and add an interface for late suspend/early resume (or maybe call this something like pmc_enter and pmc_exit? then again maybe on some platforms the PMC isn't necessarily intended to be called after device suspend) in the future.

Fix issues pointed out by olce@ and apply some suggestions. Main changes:

Remove unused acpi_s2idle_looping

Don't use new IPI_IDLE/IPI_UNIDLE, but use smp_rendezvous_cpus() calling sched_do_idle() instead.

Can totally use smp_rendezvous_cpus() for this instead of adding 2 new IPIs. I thought I had run into a problem using rendezvous when I first implemented this, but I just tried it again and see no problem doing it this way.

Move comment.

Fix more issues pointed out by olce@.

style(9)

style(9)

Tested on my machine (only S5, S4, and S0) and all works

Should we even allow hw.acpi.s4bios to be set if not supported? feels like it could be confusing to users not super familiar with power management that hw.acpi.s4bios is set but the system is not entering S4BIOS.

Accepting assuming no one has objections about changing the size of the sysctl

relevant commit in acpica, for reference: https://github.com/acpica/acpica/commit/79cd933e7b370e8d3fb490bf36ca5d111a12f96a

Reorder includes.

In D54556#1246429, @markj wrote:

That is a possibility, most other fo_stats seem to follow the bzero before setting fields pattern already.

Including vn_statfile(). I think I agree that for now we should patch dma-buf instead, but in the longer term we perhaps should consolidate the zeroing here.

In D54556#1246415, @kib wrote:

stat(2) was micro-optimized to the death. Do not introduce second write to the same memory (not to mention that the next two lines write to some fields third time).

Simplify cpuset initialization.

Remove rev_for_uuid for now and just use revision 0 everywhere.

Finally got around to updating this diff @olce :)

Fix issues pointed out by olce@.

In D54158#1237352, @jhb wrote:

I think we just need some sort of IVAR we can test for and this is just as easy as any other. Can you test this to make sure it works? I have only compile tested it as my only FreeBSD box with TB has the older chipsets that don't work.

looks good, though I wonder if there is a point in actually passing down tb_gen, considering I don't think we'll be supporting non-TB_GEN_USB4 in a while (and I was gonna remove the existing tb_pcib_identifiers after D52861)

ack your review comments, I have a couple changes I need to bring out of a stash commit before I address your comments. My local commit log is getting pretty messy from all the suspend debugging stuff I'm trying ;)

Fix DSM call ordering

thanks for taking a look at this!

Remove previous IDs.