diff options
| author | Brennan Shacklett <brennan@genyes.org> | 2013-10-21 12:20:32 -0400 |
|---|---|---|
| committer | Rafael J. Wysocki <rafael.j.wysocki@intel.com> | 2013-10-21 19:15:38 -0400 |
| commit | d253d2a52676cfa3d89b8f0737a08ce7db665207 (patch) | |
| tree | 0764df6b50b68dea4c7d2984b5b9e599ef12f0f8 | |
| parent | 31d141e3a666269a3b6fcccddb0351caf7454240 (diff) | |
intel_pstate: Improve accuracy by not truncating until final result
This patch addresses Bug 60727
(https://bugzilla.kernel.org/show_bug.cgi?id=60727)
which was due to the truncation of intermediate values in the
calculations, which causes the code to consistently underestimate the
current cpu frequency, specifically 100% cpu utilization was truncated
down to the setpoint of 97%. This patch fixes the problem by keeping
the results of all intermediate calculations as fixed point numbers
rather scaling them back and forth between integers and fixed point.
References: https://bugzilla.kernel.org/show_bug.cgi?id=60727
Signed-off-by: Brennan Shacklett <bpshacklett@gmail.com>
Acked-by: Dirk Brandewie <dirk.j.brandewie@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
| -rw-r--r-- | drivers/cpufreq/intel_pstate.c | 33 |
1 files changed, 15 insertions, 18 deletions
diff --git a/drivers/cpufreq/intel_pstate.c b/drivers/cpufreq/intel_pstate.c index badf6206b2b2..8b8677f92700 100644 --- a/drivers/cpufreq/intel_pstate.c +++ b/drivers/cpufreq/intel_pstate.c | |||
| @@ -48,7 +48,7 @@ static inline int32_t div_fp(int32_t x, int32_t y) | |||
| 48 | } | 48 | } |
| 49 | 49 | ||
| 50 | struct sample { | 50 | struct sample { |
| 51 | int core_pct_busy; | 51 | int32_t core_pct_busy; |
| 52 | u64 aperf; | 52 | u64 aperf; |
| 53 | u64 mperf; | 53 | u64 mperf; |
| 54 | int freq; | 54 | int freq; |
| @@ -68,7 +68,7 @@ struct _pid { | |||
| 68 | int32_t i_gain; | 68 | int32_t i_gain; |
| 69 | int32_t d_gain; | 69 | int32_t d_gain; |
| 70 | int deadband; | 70 | int deadband; |
| 71 | int last_err; | 71 | int32_t last_err; |
| 72 | }; | 72 | }; |
| 73 | 73 | ||
| 74 | struct cpudata { | 74 | struct cpudata { |
| @@ -153,16 +153,15 @@ static inline void pid_d_gain_set(struct _pid *pid, int percent) | |||
| 153 | pid->d_gain = div_fp(int_tofp(percent), int_tofp(100)); | 153 | pid->d_gain = div_fp(int_tofp(percent), int_tofp(100)); |
| 154 | } | 154 | } |
| 155 | 155 | ||
| 156 | static signed int pid_calc(struct _pid *pid, int busy) | 156 | static signed int pid_calc(struct _pid *pid, int32_t busy) |
| 157 | { | 157 | { |
| 158 | signed int err, result; | 158 | signed int result; |
| 159 | int32_t pterm, dterm, fp_error; | 159 | int32_t pterm, dterm, fp_error; |
| 160 | int32_t integral_limit; | 160 | int32_t integral_limit; |
| 161 | 161 | ||
| 162 | err = pid->setpoint - busy; | 162 | fp_error = int_tofp(pid->setpoint) - busy; |
| 163 | fp_error = int_tofp(err); | ||
| 164 | 163 | ||
| 165 | if (abs(err) <= pid->deadband) | 164 | if (abs(fp_error) <= int_tofp(pid->deadband)) |
| 166 | return 0; | 165 | return 0; |
| 167 | 166 | ||
| 168 | pterm = mul_fp(pid->p_gain, fp_error); | 167 | pterm = mul_fp(pid->p_gain, fp_error); |
| @@ -176,8 +175,8 @@ static signed int pid_calc(struct _pid *pid, int busy) | |||
| 176 | if (pid->integral < -integral_limit) | 175 | if (pid->integral < -integral_limit) |
| 177 | pid->integral = -integral_limit; | 176 | pid->integral = -integral_limit; |
| 178 | 177 | ||
| 179 | dterm = mul_fp(pid->d_gain, (err - pid->last_err)); | 178 | dterm = mul_fp(pid->d_gain, fp_error - pid->last_err); |
| 180 | pid->last_err = err; | 179 | pid->last_err = fp_error; |
| 181 | 180 | ||
| 182 | result = pterm + mul_fp(pid->integral, pid->i_gain) + dterm; | 181 | result = pterm + mul_fp(pid->integral, pid->i_gain) + dterm; |
| 183 | 182 | ||
| @@ -436,8 +435,9 @@ static inline void intel_pstate_calc_busy(struct cpudata *cpu, | |||
| 436 | struct sample *sample) | 435 | struct sample *sample) |
| 437 | { | 436 | { |
| 438 | u64 core_pct; | 437 | u64 core_pct; |
| 439 | core_pct = div64_u64(sample->aperf * 100, sample->mperf); | 438 | core_pct = div64_u64(int_tofp(sample->aperf * 100), |
| 440 | sample->freq = cpu->pstate.max_pstate * core_pct * 1000; | 439 | sample->mperf); |
| 440 | sample->freq = fp_toint(cpu->pstate.max_pstate * core_pct * 1000); | ||
| 441 | 441 | ||
| 442 | sample->core_pct_busy = core_pct; | 442 | sample->core_pct_busy = core_pct; |
| 443 | } | 443 | } |
| @@ -469,22 +469,19 @@ static inline void intel_pstate_set_sample_time(struct cpudata *cpu) | |||
| 469 | mod_timer_pinned(&cpu->timer, jiffies + delay); | 469 | mod_timer_pinned(&cpu->timer, jiffies + delay); |
| 470 | } | 470 | } |
| 471 | 471 | ||
| 472 | static inline int intel_pstate_get_scaled_busy(struct cpudata *cpu) | 472 | static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu) |
| 473 | { | 473 | { |
| 474 | int32_t busy_scaled; | ||
| 475 | int32_t core_busy, max_pstate, current_pstate; | 474 | int32_t core_busy, max_pstate, current_pstate; |
| 476 | 475 | ||
| 477 | core_busy = int_tofp(cpu->samples[cpu->sample_ptr].core_pct_busy); | 476 | core_busy = cpu->samples[cpu->sample_ptr].core_pct_busy; |
| 478 | max_pstate = int_tofp(cpu->pstate.max_pstate); | 477 | max_pstate = int_tofp(cpu->pstate.max_pstate); |
| 479 | current_pstate = int_tofp(cpu->pstate.current_pstate); | 478 | current_pstate = int_tofp(cpu->pstate.current_pstate); |
| 480 | busy_scaled = mul_fp(core_busy, div_fp(max_pstate, current_pstate)); | 479 | return mul_fp(core_busy, div_fp(max_pstate, current_pstate)); |
| 481 | |||
| 482 | return fp_toint(busy_scaled); | ||
| 483 | } | 480 | } |
| 484 | 481 | ||
| 485 | static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu) | 482 | static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu) |
| 486 | { | 483 | { |
| 487 | int busy_scaled; | 484 | int32_t busy_scaled; |
| 488 | struct _pid *pid; | 485 | struct _pid *pid; |
| 489 | signed int ctl = 0; | 486 | signed int ctl = 0; |
| 490 | int steps; | 487 | int steps; |
