summaryrefslogtreecommitdiff
path: root/drivers/cpufreq/cpufreq_governor.c
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2014-06-13 00:14:19 +0400
committerLinus Torvalds <torvalds@linux-foundation.org>2014-06-13 00:14:19 +0400
commit19c1940feab777bb037c665a09f495d08a6c4e6c (patch)
tree68d4de31f30c9d69a1f1ea080b8a69ce61e1ead5 /drivers/cpufreq/cpufreq_governor.c
parent7c574cf6aeb75920ba4d3af937bb1b3c42785ac4 (diff)
parentd715a226b0b3dae48865d05e8c36175a8f75a809 (diff)
downloadlinux-19c1940feab777bb037c665a09f495d08a6c4e6c.tar.xz
Merge tag 'pm+acpi-3.16-rc1-2' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm
Pull more ACPI and power management updates from Rafael Wysocki: "These are fixups on top of the previous PM+ACPI pull request, regression fixes (ACPI hotplug, cpufreq ppc-corenet), other bug fixes (ACPI reset, cpufreq), new PM trace points for system suspend profiling and a copyright notice update. Specifics: - I didn't remember correctly that the Hans de Goede's ACPI video patches actually didn't flip the video.use_native_backlight default, although we had discussed that and decided to do that. Since I said we would do that in the previous PM+ACPI pull request, make that change for real now. - ACPI bus check notifications for PCI host bridges don't cause the bus below the host bridge to be checked for changes as they should because of a mistake in the ACPI-based PCI hotplug (ACPIPHP) subsystem that forgets to add hotplug contexts to PCI host bridge ACPI device objects. Create hotplug contexts for PCI host bridges too as appropriate. - Revert recent cpufreq commit related to the big.LITTLE cpufreq driver that breaks arm64 builds. - Fix for a regression in the ppc-corenet cpufreq driver introduced during the 3.15 cycle and causing the driver to use the remainder from do_div instead of the quotient. From Ed Swarthout. - Resets triggered by panic activate a BUG_ON() in vmalloc.c on systems where the ACPI reset register is located in memory address space. Fix from Randy Wright. - Fix for a problem with cpufreq governors that decisions made by them may be suboptimal due to the fact that deferrable timers are used by them for CPU load sampling. From Srivatsa S Bhat. - Fix for a problem with the Tegra cpufreq driver where the CPU frequency is temporarily switched to a "stable" level that is different from both the initial and target frequencies during transitions which causes udelay() to expire earlier than it should sometimes. From Viresh Kumar. - New trace points and rework of some existing trace points for system suspend/resume profiling from Todd Brandt. - Assorted cpufreq fixes and cleanups from Stratos Karafotis and Viresh Kumar. - Copyright notice update for suspend-and-cpuhotplug.txt from Srivatsa S Bhat" * tag 'pm+acpi-3.16-rc1-2' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm: ACPI / hotplug / PCI: Add hotplug contexts to PCI host bridges PM / sleep: trace events for device PM callbacks cpufreq: cpufreq-cpu0: remove dependency on THERMAL and REGULATOR cpufreq: tegra: update comment for clarity cpufreq: intel_pstate: Remove duplicate CPU ID check cpufreq: Mark CPU0 driver with CPUFREQ_NEED_INITIAL_FREQ_CHECK flag PM / Documentation: Update copyright in suspend-and-cpuhotplug.txt cpufreq: governor: remove copy_prev_load from 'struct cpu_dbs_common_info' cpufreq: governor: Be friendly towards latency-sensitive bursty workloads PM / sleep: trace events for suspend/resume cpufreq: ppc-corenet-cpu-freq: do_div use quotient Revert "cpufreq: Enable big.LITTLE cpufreq driver on arm64" cpufreq: Tegra: implement intermediate frequency callbacks cpufreq: add support for intermediate (stable) frequencies ACPI / video: Change the default for video.use_native_backlight to 1 ACPI: Fix bug when ACPI reset register is implemented in system memory
Diffstat (limited to 'drivers/cpufreq/cpufreq_governor.c')
-rw-r--r--drivers/cpufreq/cpufreq_governor.c67
1 files changed, 64 insertions, 3 deletions
diff --git a/drivers/cpufreq/cpufreq_governor.c b/drivers/cpufreq/cpufreq_governor.c
index e1c6433b16e0..1b44496b2d2b 100644
--- a/drivers/cpufreq/cpufreq_governor.c
+++ b/drivers/cpufreq/cpufreq_governor.c
@@ -36,14 +36,29 @@ void dbs_check_cpu(struct dbs_data *dbs_data, int cpu)
struct od_dbs_tuners *od_tuners = dbs_data->tuners;
struct cs_dbs_tuners *cs_tuners = dbs_data->tuners;
struct cpufreq_policy *policy;
+ unsigned int sampling_rate;
unsigned int max_load = 0;
unsigned int ignore_nice;
unsigned int j;
- if (dbs_data->cdata->governor == GOV_ONDEMAND)
+ if (dbs_data->cdata->governor == GOV_ONDEMAND) {
+ struct od_cpu_dbs_info_s *od_dbs_info =
+ dbs_data->cdata->get_cpu_dbs_info_s(cpu);
+
+ /*
+ * Sometimes, the ondemand governor uses an additional
+ * multiplier to give long delays. So apply this multiplier to
+ * the 'sampling_rate', so as to keep the wake-up-from-idle
+ * detection logic a bit conservative.
+ */
+ sampling_rate = od_tuners->sampling_rate;
+ sampling_rate *= od_dbs_info->rate_mult;
+
ignore_nice = od_tuners->ignore_nice_load;
- else
+ } else {
+ sampling_rate = cs_tuners->sampling_rate;
ignore_nice = cs_tuners->ignore_nice_load;
+ }
policy = cdbs->cur_policy;
@@ -96,7 +111,46 @@ void dbs_check_cpu(struct dbs_data *dbs_data, int cpu)
if (unlikely(!wall_time || wall_time < idle_time))
continue;
- load = 100 * (wall_time - idle_time) / wall_time;
+ /*
+ * If the CPU had gone completely idle, and a task just woke up
+ * on this CPU now, it would be unfair to calculate 'load' the
+ * usual way for this elapsed time-window, because it will show
+ * near-zero load, irrespective of how CPU intensive that task
+ * actually is. This is undesirable for latency-sensitive bursty
+ * workloads.
+ *
+ * To avoid this, we reuse the 'load' from the previous
+ * time-window and give this task a chance to start with a
+ * reasonably high CPU frequency. (However, we shouldn't over-do
+ * this copy, lest we get stuck at a high load (high frequency)
+ * for too long, even when the current system load has actually
+ * dropped down. So we perform the copy only once, upon the
+ * first wake-up from idle.)
+ *
+ * Detecting this situation is easy: the governor's deferrable
+ * timer would not have fired during CPU-idle periods. Hence
+ * an unusually large 'wall_time' (as compared to the sampling
+ * rate) indicates this scenario.
+ *
+ * prev_load can be zero in two cases and we must recalculate it
+ * for both cases:
+ * - during long idle intervals
+ * - explicitly set to zero
+ */
+ if (unlikely(wall_time > (2 * sampling_rate) &&
+ j_cdbs->prev_load)) {
+ load = j_cdbs->prev_load;
+
+ /*
+ * Perform a destructive copy, to ensure that we copy
+ * the previous load only once, upon the first wake-up
+ * from idle.
+ */
+ j_cdbs->prev_load = 0;
+ } else {
+ load = 100 * (wall_time - idle_time) / wall_time;
+ j_cdbs->prev_load = load;
+ }
if (load > max_load)
max_load = load;
@@ -318,11 +372,18 @@ int cpufreq_governor_dbs(struct cpufreq_policy *policy,
for_each_cpu(j, policy->cpus) {
struct cpu_dbs_common_info *j_cdbs =
dbs_data->cdata->get_cpu_cdbs(j);
+ unsigned int prev_load;
j_cdbs->cpu = j;
j_cdbs->cur_policy = policy;
j_cdbs->prev_cpu_idle = get_cpu_idle_time(j,
&j_cdbs->prev_cpu_wall, io_busy);
+
+ prev_load = (unsigned int)
+ (j_cdbs->prev_cpu_wall - j_cdbs->prev_cpu_idle);
+ j_cdbs->prev_load = 100 * prev_load /
+ (unsigned int) j_cdbs->prev_cpu_wall;
+
if (ignore_nice)
j_cdbs->prev_cpu_nice =
kcpustat_cpu(j).cpustat[CPUTIME_NICE];