Linus reported a ~50% performance regression on single-threaded workloads on his AMD Ryzen system, and bisected it to: 9c0b4bb7f630 ("sched/cpufreq: Rework schedutil governor performance estimation") When frequency invariance is not enabled, get_capacity_ref_freq(policy) is supposed to return the current frequency and the performance margin applied by map_util_perf(), enabling the utilization to go above the maximum compute capacity and to select a higher frequency than the current one. After the changes in 9c0b4bb7f630, the performance margin was applied earlier in the path to take into account utilization clampings and we couldn't get a utilization higher than the maximum compute capacity, and the CPU remained 'stuck' at lower frequencies. To fix this, we must use a frequency above the current frequency to get a chance to select a higher OPP when the current one becomes fully used. Apply the same margin and return a frequency 25% higher than the current one in order to switch to the next OPP before we fully use the CPU at the current one. [ mingo: Clarified the changelog. ]
Unified · Split
| @@ -133,7 +133,11 @@ unsigned long get_capacity_ref_freq(struct cpufreq_policy *policy) · 132 unmodified lines | ||
| 133 | 133 | if (arch_scale_freq_invariant()) |
| 134 | 134 | return policy->cpuinfo.max_freq; |
| 135 | 135 | |
| 136 | − return policy->cur; | |
| 136 | + /* | |
| 137 | + * Apply a 25% margin so that we select a higher frequency than | |
| 138 | + * the current one before the CPU is fully busy: | |
| 139 | + */ | |
| 140 | + return policy->cur + (policy->cur >> 2); | |
| 137 | 141 | } |
| 138 | 142 | |
| 139 | 143 | /** |