didn't actually review the diff completly, but maby more content
in the manual page would be useful. anyways, small comment inline.

On Sat, May 04, 2024 at 09:35:35PM +0200, Kirill A. Korinsky wrote:
> @@ -583,7 +591,9 @@ int perfpolicy = PERFPOL_AUTO;
>  #include <sys/sysctl.h>
> 
>  void setperf_auto(void *);
> +void setperf_powersaving(void *);
>  struct timeout setperf_to = TIMEOUT_INITIALIZER(setperf_auto, NULL);
> +struct timeout setperf_to_powersaving = TIMEOUT_INITIALIZER(setperf_powersaving, NULL);
>  extern int hw_power;
> 
>  void
> @@ -653,6 +663,106 @@ faster:
>  	timeout_add_msec(&setperf_to, 100);
>  }
> 
> +void
> +setperf_powersaving(void *v)
> +{
> +	static uint64_t *idleticks, *totalticks;
> +	static int downbeats;
> +	int i, j = 0;
> +	int speedup = 0;
> +	CPU_INFO_ITERATOR cii;
> +	struct cpu_info *ci, *firstoffline = NULL, *lastidle = NULL;
> +	uint64_t idle, total, allidle = 0, alltotal = 0;
> +
> +	if (perfpolicy != PERFPOL_POWERSAVING)
> +		goto recover;
> +
> +	if (cpu_setperf == NULL)
> +		goto recover;
> +
> +	if (hw_power)
> +		goto recover;

does the above check inforce some invariants? if not, could it be
removed?  this could be useful for servers to use less power, even
if they're always running on AC.