acpi: Fix panic on poweroff

[obiwac]

Summary

Previously we were erroneously passing the device softc to
acpi_stype_to_sstate in acpi_wake/run_sleep_prep instead of the ACPI
bus device softc.

Now, we pass the ACPI bus device softc when walking devices to prep for
sleep/wake.

Use the opportunity to pass the stype instead of the ACPI S-state to
acpi_wake_prep_walk.

PR: 289634
Sponsored by: The FreeBSD Foundation

[markj]

Is the proper fix "just" to plumb the acpi softc through acpi_wake_prep_walk()->AcpiWalkNamespace()->acpi_wake_prep()->...? That doesn't seem too painful.

I'm ok with this patch as an intermediate step to fix the panics, but the above should also be straightforward.

[obiwac]

yeah you're absolutely right, I thought acpi_shutdown's dev was a device but its the actual bus of course so this is easy. Will do that in a minute

it's been a really long day, sorry I missed this :D

[obiwac]

Do this correctly.

[bz]

sys/dev/acpica/acpivar.h

137 ↗

(On Diff #162298)

Is there a reason to not keep this file-local in acpi.c?

[obiwac]

sys/dev/acpica/acpivar.h

137 ↗

(On Diff #162298)

I could but I saw that no other structs were being kept file-local so I decided to put this here.

But none of these are entirely local actually so I'll move this to acpi.c.

[obiwac]

Move struct acpi_wake_prep_context out of acpivar.h

[obiwac]

sys/dev/acpica/acpivar.h

137 ↗

(On Diff #162298)

actually struct acpi_interface was, my bad.

[zarychtam_plan-b.pwste.edu.pl]

This patch fixes panic. Please push it when it's ready.

Edit: Unfortunately, suspend is still broken, PC wakes up immediately.

[obiwac]

In D52598#1201370, @zarychtam_plan-b.pwste.edu.pl wrote:

This patch fixes panic. Please push it when it's ready.

Edit: Unfortunately, suspend is still broken, PC wakes up immediately.

Okay, but this is a separate issue I guess. This is going to be hard for me to debug as I do not have an S3 machine. Are you willing to try some patches out for me to fix this issue? Also, can you send me your kernel logs ([e-mail])?

In the meantime I'll just revert the changes which touch the ACPI logic.

[markj]

sys/dev/acpica/acpi.c

3765

Stylistically this looks a bit weird. More typical would be something like

struct acpi_wake_prep_context *ctx = context;

if (AcpiGbl_SystemAwakeAndRunning)
    acpi_wake_sleep_prep(ctx->sc, handle, ctx->stype);
else
    acpi_wake_run_prep(ctx->sc, handle, ctx->stype);

or declare more local variables after, struct acpi_softc *sc = ctx->sc; etc.

[madpilot]

I finally tested this and works fine for me.

I'm sorry I was unable to test it earlier!

[obiwac]

In D52598#1201481, @madpilot wrote:

I finally tested this and works fine for me.

I'm sorry I was unable to test it earlier!

Don't worry about it, I wasn't behind my computer today anyways. Thanks for testing!

[obiwac]

these changes have been rolled into D52705 because the original commit that had these changes had to be reverted anyways.

[obiwac]

sys/dev/acpica/acpi.c

3765

applied this change in D52705