Re: [PATCH 8/9] platform/x86: asus-wmi: add apu_mem setting

From: Mario Limonciello
Date: Tue May 28 2024 - 17:36:54 EST


On 5/28/2024 16:34, Luke Jones wrote:


On Wed, 29 May 2024, at 9:16 AM, Mario Limonciello wrote:
For any "new" attributes it's better to put them in that API than code
duplication of the BIOS attributes API as well as a random sysfs file
API that you can never discard.

Do you mean the firmware_attributes API? If so, I'm not opposed to adding all the existing ROG attributes to it also.

If I'm understanding the docs correctly, for example this apu_mem attr would then become:
- /sys/class/firmware-attributes/asus-bios/attributes/apu_mem/type
- /sys/class/firmware-attributes/*/attributes/apu_mem/current_value
- /sys/class/firmware-attributes/*/attributes/apu_mem/default_value
- /sys/class/firmware-attributes/*/attributes/apu_mem/display_name
- /sys/class/firmware-attributes/*/attributes/apu_mem/possible_values
- ..etc

That's absolutely much better than what I've been doing and I wish I'd known about it sooner.

Yup! And even better "fwupdmgr get-bios-setting" and "fwupdmgr
set-bios-setting" [1] will work with bash completion too.

If you want to build a GUI around it, it could use the fwupd dbus
interface for getting those settings and manipulating them.

My existing GUI effort revolves around me adding these attrs to both kernel and a dbus interface. So if I can drop that then all the better

The dbus
interface policy is protected by PK.

What is PK? And I'm guessing this restricts users to requiring permissions to change settings?

So if I go ahead and convert all the new attr to this are there any issues with also converting much of the previous attr? And I'm aware of "don't break userspace" so really I'm a bit unsure how best to manage that (would a new module be better here also? "asus-bios.c" perhaps).

What I don't want is a split between platform and firmware_attributes.


My personal thought is add everything to the new API and then hide the
"old" interface behind some boolean Kconfig option that defaults to
disabled. If someone enables it then when it's "used" (IE read from /
written to) then do something like a pr_warn_once() that it's a
deprecated interface and it will go away. Wait until next LTS kernel,
and then rip it out after that.

I'm happy to do this. On a serious note though, I should make a new module for this right? It would make it much much easier to manage the "conversion" I think, and then it would act as a separation of concerns.

Yeah; I agree with you I think this is a good idea.



Need Hans and/or Ilpo agree to make with that though before you spend
too much effort on it though.

Ack


[1] https://fwupd.github.io/libfwupdplugin/bios-settings.html