Impossible to set AMD Overdrive sysctls

[valpackett]

Finally got around to trying FreeBSD on my desktop (RX 480). The drm-next kernel does not work with my network card trueos/trueos-core#1463 so I'm glad it's now available as a port :) Everything seems to work pretty well so far. Tried WebGL demos in Firefox, Zelda Wind Waker in Dolphin Emulator (works fine when built from git master; the version from pkg displays a black screen instead of games), and DarkPlaces Quake.

I can't overclock my GPU though:

$ doas sysctl sys.device.drmn0.pp_sclk_od=1
sys.device.drmn0.pp_sclk_od: 0
sysctl: sys.device.drmn0.pp_sclk_od=1: Invalid argument
$ doas sysctl sys.device.drmn0.pp_mclk_od=1
sys.device.drmn0.pp_mclk_od: 0
sysctl: sys.device.drmn0.pp_mclk_od=1: Invalid argument

On Linux, people do something like e.g. echo 5 > pp_sclk_od to set a 5% overclock for the GPU core. FreeBSD sysctls reject numbers as invalid arguments.

[hselasky]

Hi,
Try to set this variable using "kenv" or in /boot/loader.conf before loading the .ko's
--HPS

[valpackett]

No, that didn't help.

BTW, possibly interesting dmesg message:

Can't find requested voltage id in vdd_dep_on_sclk table!
amdgpu: powerplay initialized

Also, apparently pp_table should provide access to the whole powerplay table, but even when asking for the raw value (sysctl -b) it returns just the number 4.

[hselasky]

This might be fixed in 4.11.

[valpackett]

nah, still "Invalid argument"

[valpackett]

Finally got around to debugging this.

Turns out kstrtol returns EINVAL even when it worked fine. lol. Commented out the error check and it works.

https://github.com/freebsd/freebsd/blob/92a4d586c6b7260c4e4469dfd7158839624c9fba/sys/compat/linuxkpi/common/include/linux/kernel.h#L308-L311 — maybe *cp should be 0?

[hselasky]

Hi,

Can you add a printf in the EINVAL case. Print all variables and input. Maybe there is a "\n" there.

--HPS

[valpackett]

I printed the buffer, length, parsed value and return. Didn't realize at first because my print ended with the buffer, but yeah it seems like there is an extra newline after the buffer indeed.

[hselasky]

It looks like linux accepts this.
Can you add to all these functions:
if (*end == '\n')
end++;

[hselasky]

Try this:
https://svnweb.freebsd.org/changeset/base/330944

[valpackett]

It works, thanks!