From aa4ea34da9e7a2bc0fb98f5add3e4e52872b7d45 Mon Sep 17 00:00:00 2001 From: Ethan Zhao Date: Tue, 9 Dec 2014 10:43:19 +0900 Subject: intel_pstate: add kernel parameter to force loading To force loading on Oracle Sun X86 servers, provide one kernel command line parameter intel_pstate = force For those who are aware of the risk of no power capping capabily working and try to get better performance with this driver. Signed-off-by: Ethan Zhao Tested-by: Alexey Kodanev Reviewed-by: Linda Knippers Acked-by: Kristen Carlson Accardi Signed-off-by: Rafael J. Wysocki --- drivers/cpufreq/intel_pstate.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) (limited to 'drivers/cpufreq') diff --git a/drivers/cpufreq/intel_pstate.c b/drivers/cpufreq/intel_pstate.c index 1405b393c93d..0e841eecb743 100644 --- a/drivers/cpufreq/intel_pstate.c +++ b/drivers/cpufreq/intel_pstate.c @@ -948,6 +948,7 @@ static struct cpufreq_driver intel_pstate_driver = { static int __initdata no_load; static int __initdata no_hwp; +static unsigned int force_load; static int intel_pstate_msrs_not_valid(void) { @@ -1094,7 +1095,8 @@ static bool intel_pstate_platform_pwr_mgmt_exists(void) case PSS: return intel_pstate_no_acpi_pss(); case PPC: - return intel_pstate_has_acpi_ppc(); + return intel_pstate_has_acpi_ppc() && + (!force_load); } } @@ -1175,6 +1177,8 @@ static int __init intel_pstate_setup(char *str) no_load = 1; if (!strcmp(str, "no_hwp")) no_hwp = 1; + if (!strcmp(str, "force")) + force_load = 1; return 0; } early_param("intel_pstate", intel_pstate_setup); -- cgit v1.2.3 From e0d4c8f80804000eadc106bc5167b96fc6231d98 Mon Sep 17 00:00:00 2001 From: Kristen Carlson Accardi Date: Wed, 10 Dec 2014 12:39:38 -0800 Subject: intel_pstate: Add a few comments Add a few comments in the code which calculates busyness to clarify parts of the algorithm. Signed-off-by: Kristen Carlson Accardi Signed-off-by: Rafael J. Wysocki --- drivers/cpufreq/intel_pstate.c | 32 +++++++++++++++++++++++++++++++- 1 file changed, 31 insertions(+), 1 deletion(-) (limited to 'drivers/cpufreq') diff --git a/drivers/cpufreq/intel_pstate.c b/drivers/cpufreq/intel_pstate.c index 0e841eecb743..742eefba12c2 100644 --- a/drivers/cpufreq/intel_pstate.c +++ b/drivers/cpufreq/intel_pstate.c @@ -199,7 +199,14 @@ static signed int pid_calc(struct _pid *pid, int32_t busy) pid->integral += fp_error; - /* limit the integral term */ + /* + * We limit the integral here so that it will never + * get higher than 30. This prevents it from becoming + * too large an input over long periods of time and allows + * it to get factored out sooner. + * + * The value of 30 was chosen through experimentation. + */ integral_limit = int_tofp(30); if (pid->integral > integral_limit) pid->integral = integral_limit; @@ -616,6 +623,11 @@ static void intel_pstate_get_min_max(struct cpudata *cpu, int *min, int *max) if (limits.no_turbo || limits.turbo_disabled) max_perf = cpu->pstate.max_pstate; + /* + * performance can be limited by user through sysfs, by cpufreq + * policy, or by cpu specific default values determined through + * experimentation. + */ max_perf_adj = fp_toint(mul_fp(int_tofp(max_perf), limits.max_perf)); *max = clamp_t(int, max_perf_adj, cpu->pstate.min_pstate, cpu->pstate.turbo_pstate); @@ -717,11 +729,29 @@ static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu) u32 duration_us; u32 sample_time; + /* + * core_busy is the ratio of actual performance to max + * max_pstate is the max non turbo pstate available + * current_pstate was the pstate that was requested during + * the last sample period. + * + * We normalize core_busy, which was our actual percent + * performance to what we requested during the last sample + * period. The result will be a percentage of busy at a + * specified pstate. + */ core_busy = cpu->sample.core_pct_busy; max_pstate = int_tofp(cpu->pstate.max_pstate); current_pstate = int_tofp(cpu->pstate.current_pstate); core_busy = mul_fp(core_busy, div_fp(max_pstate, current_pstate)); + /* + * Since we have a deferred timer, it will not fire unless + * we are in C0. So, determine if the actual elapsed time + * is significantly greater (3x) than our sample interval. If it + * is, then we were idle for a long enough period of time + * to adjust our busyness. + */ sample_time = pid_params.sample_rate_ms * USEC_PER_MSEC; duration_us = (u32) ktime_us_delta(cpu->sample.time, cpu->last_sample_time); -- cgit v1.2.3