PPD doesn't force the SoC to active mode or to load.

As the policy of whether amd-pstate is loaded is controlled by the
kernel having another policy whether power-profiles-daemon isn't
necessary.

Mario Limonciello authored 6 months ago and Marco Trevisan committed 6 months ago

This should fix a possible stack trace of an unset error variable.

Mario Limonciello authored 6 months ago and Marco Trevisan committed 6 months ago

Due a firmware bug on some OEM systems the amd-pstate kernel driver
starts up in active mode but EPP writes don't work.

Admittedly this is a kernel bug that it still advertises support and
it could be handled better by the kernel, but PPD can also validate
writes with a no-op for each CPU at bootup.

Do this to ensure that the CPU driver is only setup if we know it works.

Although the system will save more power at lower frequency, AMD SoCs
have a point when they can operate most efficiently called the lowest
non-linear frequency.

Program the minimum frequency to this value in balanced and performance
modes to improve responsiveness.

Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>

Mario Limonciello authored 10 months ago and Mario Limonciello committed 8 months ago

Apply core performance boost in balanced and performance and
disable it when in power saver.

We can just return directly most of the times

Using GList is sub-optimal and eventually needs just more allocations
and iterations, so use arrays instead which provide a nicer API.

Mario Limonciello authored 1 year ago and Mario Limonciello committed 11 months ago

CI already run with `env["G_DEBUG"] = "fatal-criticals"` and these
errors would cause the daemon to abort.  In the event something
escapes don't show a crash on a regular user's system.

Mario Limonciello authored 1 year ago and Mario Limonciello committed 1 year ago

When the system is on AC and profile is balanced put the EPP value
of balanced_performance.
When the system is on battery and profile is balanced put the EPP value
of balanced_power.

Mario Limonciello authored 1 year ago and Kate Hsuan committed 1 year ago

Make the change specifically when changing modes and only set powersave
for balance and powersaver profiles.

Mario Limonciello authored 1 year ago and Kate Hsuan committed 1 year ago

This mirrors the behavior used by Kernel 6.5 and later where amd-pstate
is configured to default into 'performance' by the kernel.  It's not
as useful in server to be changing it.

Link: https://github.com/torvalds/linux/commit/32f80b9adfdb43f8af248596724f59dde938a190

Mario Limonciello authored 1 year ago and Kate Hsuan committed 1 year ago

This is done by designating all drivers as CPU drivers or platform drivers.
One of each driver can be active at any given time.

When setting an active profile CPU driver is set first as this is less likely to have
failures. If failures occur setting the platform driver then the CPU driver is undone.

If both profiles are degraded then the string for the reason from both drivers will be
comma separated to be compatible with existing API and users.

This is leveraged from !123

Fixes: #107
Fixes: #125
Fixes: #124
Fixes: #122
Fixes: #129
Fixes: #131

Prajna Sariputra authored 1 year ago and Bastien Nocera committed 1 year ago

The AMD P-states driver has added support for active mode in kernel 6.3,
which exposes a similar interface for controlling energy performance
preference as the Intel P-state driver, with the same presets present,
so make use of it.

See:
https://www.kernel.org/doc/html/v6.3/admin-guide/pm/amd-pstate.html

Closes: #108