P-state switching on Lenovo X121e / AMD E-450

[Grigori Goronzy]

Hi,

I have a Lenovo X121e notebook, and the model variant in question uses
the E-450 APU. I have various problems with the GPU part of it. The most
severe seems to be that switching between power states does not work
correctly.

First off, the PowerPlay table looks rather strange. Here's what the
Linux kernel (3.5.0rc5) tells me about it:

> [    2.447536] [drm:radeon_pm_print_states], 6 Power State(s)
> [    2.447540] [drm:radeon_pm_print_states], State 0: Default
> [    2.447544] [drm:radeon_pm_print_states],    1 Clock Mode(s)
> [    2.447547] [drm:radeon_pm_print_states],            0 e: 275000
> [    2.447551] [drm:radeon_pm_print_states], State 1: Default
> [    2.447595] [drm:radeon_pm_print_states],    1 Clock Mode(s)
> [    2.447598] [drm:radeon_pm_print_states],            0 e: 507700
> [    2.447601] [drm:radeon_pm_print_states], State 2: Battery
> [    2.447604] [drm:radeon_pm_print_states],    1 Clock Mode(s)
> [    2.447607] [drm:radeon_pm_print_states],            0 e: 275000
> [    2.447611] [drm:radeon_pm_print_states], State 3: Performance
> [    2.447614] [drm:radeon_pm_print_states],    2 Clock Mode(s)
> [    2.447616] [drm:radeon_pm_print_states],            0 e: 275000     No display only
> [    2.447620] [drm:radeon_pm_print_states],            1 e: 507700
> [    2.447623] [drm:radeon_pm_print_states], State 4: Default
> [    2.447626] [drm:radeon_pm_print_states],    Default
> [    2.447629] [drm:radeon_pm_print_states], 
> [    2.447631]  2 Clock Mode(s)
> [    2.447633] [drm:radeon_pm_print_states],            0 e: 200000     No display only
> [    2.447636] [drm:radeon_pm_print_states],            1 e: 200000
> [    2.447639] [drm:radeon_pm_print_states], State 5: Default
> [    2.447642] [drm:radeon_pm_print_states],    1 Clock Mode(s)
> [    2.447645] [drm:radeon_pm_print_states],            0 e: 173690

I don't understand why there are so many different "default" states, and
why there are so many states altogether. I wonder, does the driver
actually parse the table correctly?

Nonetheless, the frequencies the power management code picks from that
for profile-based power management make some sense. low/mid use 275 MHz,
high uses 507 MHz, and default is 200 MHz. However, the default clock
reported by another of the AtomBIOS tables (I forgot its name) is 200
MHz, and that is exactly where the trouble is: radeon_set_power_state in
radeon_pm.c [1] restricts the maximum clock to the default clock! So
essentially, I am limited to just 200 MHz. I removed the checks and can
now use up to 507 MHz without any problems.

IMHO, this limitation makes little sense, and it is fine to have a
default clock that is lower than the maximum clock. The driver should
limit the maximum clock to the maximum encountered in the PowerPlay tables.

Best regards,
Grigori

[1]
http://lxr.free-electrons.com/source/drivers/gpu/drm/radeon/radeon_pm.c#L194