efirt: When present, attempt to use EFI runtime services to shutdown

[cem]

Summary

Submitted by: byuu <byuu AT tutanota.com> (previous version)

Test Plan

Unlike byuu's original submission, I left efi_reset_system(void) alone for two
reasons:

• Potential out of tree consumers (maybe Netflix?)
• Laziness: Changing the signature requires updating efirt.9

I would be happy to switch efi_reset_system() to the enum-taking variety if
that seems better.

I made the efi_poweroff knob the default setting on EFIRT-enabled systems, but
do not feel strongly about it. We could make it off by default if you prefer.

We could similarly handle RB_REBOOT / whatever shutdown_final events with
"rt_reset" calls. Again, if you feel strongly about it, I can add that
support. I am considering adding the same functionality to the IPMI
shutdown_final event handler anyway -- we use a different IPMI reset mode than
the one in FreeBSD base tree at ISLN.

[imp]

Netflix doesn't use efi_system_reset() in our version of FreeBSD. Changing the signature is likely fine.

sys/dev/efidev/efirt.c

114

Small quibble.
s/shutdown/power off/
I think that we should use the most specific term here.

465

We (NFLX) don't do anything out of tree with this.

sys/sys/efi.h

44–45

as there's more than 2, maybe having explicit values would be good. Maybe they should have been explicit all the time.

[cem]

Netflix doesn't use efi_system_reset() in our version of FreeBSD. Changing the signature is likely fine.

Ok, works for me.

sys/dev/efidev/efirt.c

114

Will fix

sys/sys/efi.h

44–45

I'm happy to make them explicit if you'd prefer that. They are just 0, 1, 2.

[cem]

• Use explicit terminology in sysctl text.
• Change type of efi_reset_system(9), and document in efirt.9 (which needs cleanup, but that will be left for future work).
• Explicitly define enum efi_reset values.

[cem]

efirt is unloadable; be sure to unregister our eventhandler if the module is unloaded.

[imp]

lgtm now