On Thu, Sep 26, 2024 at 04:21:14PM -0700, Vanshidhar Konda wrote:
> On Thu, Sep 26, 2024 at 12:34:01PM GMT, Beata Michalska wrote:
> > On Tue, Sep 17, 2024 at 05:41:09PM +0530, Sumit Gupta wrote:
> > > Hi Beata,
> > Hi Sumit,
> > >
> > > Thank you for the patches.
> > Thank you for having a look at those.
> > >
> > > On 13/09/24 18:59, Beata Michalska wrote:
> > > > External email: Use caution opening links or attachments
> > > >
> > > >
> > > > With the Frequency Invariance Engine (FIE) being already wired up with
> > > > sched tick and making use of relevant (core counter and constant
> > > > counter) AMU counters, getting the average frequency for a given CPU,
> > > > can be achieved by utilizing the frequency scale factor which reflects
> > > > an average CPU frequency for the last tick period length.
> > > >
> > > > The solution is partially based on APERF/MPERF implementation of
> > > > arch_freq_get_on_cpu.
> > > >
> > > > Suggested-by: Ionela Voinescu <ionela.voinescu@xxxxxxx>
> > > > Signed-off-by: Beata Michalska <beata.michalska@xxxxxxx>
> > > > ---
> > > > arch/arm64/kernel/topology.c | 109 +++++++++++++++++++++++++++++++----
> > > > 1 file changed, 99 insertions(+), 10 deletions(-)
> > > >
> > > > diff --git a/arch/arm64/kernel/topology.c b/arch/arm64/kernel/topology.c
> > > > index cb180684d10d..22e510733336 100644
> > > > --- a/arch/arm64/kernel/topology.c
> > > > +++ b/arch/arm64/kernel/topology.c
> > > > @@ -17,6 +17,7 @@
> > > > #include <linux/cpufreq.h>
> > > > #include <linux/init.h>
> > > > #include <linux/percpu.h>
> > > > +#include <linux/sched/isolation.h>
> > > >
> > > > #include <asm/cpu.h>
> > > > #include <asm/cputype.h>
> > > > @@ -88,18 +89,28 @@ int __init parse_acpi_topology(void)
> > > > * initialized.
> > > > */
> > > > static DEFINE_PER_CPU_READ_MOSTLY(unsigned long, arch_max_freq_scale) = 1UL << (2 * SCHED_CAPACITY_SHIFT);
> > > > -static DEFINE_PER_CPU(u64, arch_const_cycles_prev);
> > > > -static DEFINE_PER_CPU(u64, arch_core_cycles_prev);
> > > > static cpumask_var_t amu_fie_cpus;
> > > >
> > > > +struct amu_cntr_sample {
> > > > + u64 arch_const_cycles_prev;
> > > > + u64 arch_core_cycles_prev;
> > > > + unsigned long last_scale_update;
> > > > +};
> > > > +
> > > > +static DEFINE_PER_CPU_SHARED_ALIGNED(struct amu_cntr_sample, cpu_amu_samples);
> > > > +
> > > > void update_freq_counters_refs(void)
> > > > {
> > > > - this_cpu_write(arch_core_cycles_prev, read_corecnt());
> > > > - this_cpu_write(arch_const_cycles_prev, read_constcnt());
> > > > + struct amu_cntr_sample *amu_sample = this_cpu_ptr(&cpu_amu_samples);
> > > > +
> > > > + amu_sample->arch_core_cycles_prev = read_corecnt();
> > > > + amu_sample->arch_const_cycles_prev = read_constcnt();
> > > > }
> > > >
> > > > static inline bool freq_counters_valid(int cpu)
> > > > {
> > > > + struct amu_cntr_sample *amu_sample = per_cpu_ptr(&cpu_amu_samples, cpu);
> > > > +
> > > > if ((cpu >= nr_cpu_ids) || !cpumask_test_cpu(cpu, cpu_present_mask))
> > > > return false;
> > > >
> > > > @@ -108,8 +119,8 @@ static inline bool freq_counters_valid(int cpu)
> > > > return false;
> > > > }
> > > >
> > > > - if (unlikely(!per_cpu(arch_const_cycles_prev, cpu) ||
> > > > - !per_cpu(arch_core_cycles_prev, cpu))) {
> > > > + if (unlikely(!amu_sample->arch_const_cycles_prev ||
> > > > + !amu_sample->arch_core_cycles_prev)) {
> > > > pr_debug("CPU%d: cycle counters are not enabled.\n", cpu);
> > > > return false;
> > > > }
> > > > @@ -152,17 +163,22 @@ void freq_inv_set_max_ratio(int cpu, u64 max_rate)
> > > >
> > > > static void amu_scale_freq_tick(void)
> > > > {
> > > > + struct amu_cntr_sample *amu_sample = this_cpu_ptr(&cpu_amu_samples);
> > > > u64 prev_core_cnt, prev_const_cnt;
> > > > u64 core_cnt, const_cnt, scale;
> > > >
> > > > - prev_const_cnt = this_cpu_read(arch_const_cycles_prev);
> > > > - prev_core_cnt = this_cpu_read(arch_core_cycles_prev);
> > > > + prev_const_cnt = amu_sample->arch_const_cycles_prev;
> > > > + prev_core_cnt = amu_sample->arch_core_cycles_prev;
> > > >
> > > > update_freq_counters_refs();
> > > >
> > > > - const_cnt = this_cpu_read(arch_const_cycles_prev);
> > > > - core_cnt = this_cpu_read(arch_core_cycles_prev);
> > > > + const_cnt = amu_sample->arch_const_cycles_prev;
> > > > + core_cnt = amu_sample->arch_core_cycles_prev;
> > > >
> > > > + /*
> > > > + * This should not happen unless the AMUs have been reset and the
> > > > + * counter values have not been restored - unlikely
> > > > + */
> > > > if (unlikely(core_cnt <= prev_core_cnt ||
> > > > const_cnt <= prev_const_cnt))
> > > > return;
> > > > @@ -182,6 +198,8 @@ static void amu_scale_freq_tick(void)
> > > >
> > > > scale = min_t(unsigned long, scale, SCHED_CAPACITY_SCALE);
> > > > this_cpu_write(arch_freq_scale, (unsigned long)scale);
> > > > +
> > > > + amu_sample->last_scale_update = jiffies;
> > > > }
> > > >
> > > > static struct scale_freq_data amu_sfd = {
> > > > @@ -189,6 +207,77 @@ static struct scale_freq_data amu_sfd = {
> > > > .set_freq_scale = amu_scale_freq_tick,
> > > > };
> > > >
> > > > +static __always_inline bool amu_fie_cpu_supported(unsigned int cpu)
> > > > +{
> > > > + return cpumask_available(amu_fie_cpus) &&
> > > > + cpumask_test_cpu(cpu, amu_fie_cpus);
> > > > +}
> > > > +
> > > > +#define AMU_SAMPLE_EXP_MS 20
> > > > +
> > > > +int arch_freq_avg_get_on_cpu(int cpu)
> > > > +{
> > > > + struct amu_cntr_sample *amu_sample;
> > > > + unsigned int start_cpu = cpu;
> > > > + unsigned long last_update;
> > > > + unsigned int freq = 0;
> > > > + u64 scale;
> > > > +
> > > > + if (!amu_fie_cpu_supported(cpu) || !arch_scale_freq_ref(cpu))
> > > > + return -EOPNOTSUPP;
> > > > +
> > > > +retry:
> > > > + amu_sample = per_cpu_ptr(&cpu_amu_samples, cpu);
> > > > +
> > > > + last_update = amu_sample->last_scale_update;
> > > > +
> > > > + /*
> > > > + * For those CPUs that are in full dynticks mode, and those that have
> > > 'or those' to match with if condition?
> > Yeah, might be.
> > >
> > > > + * not seen tick for a while, try an alternative source for the counters
> > > > + * (and thus freq scale), if available, for given policy: this boils
> > > > + * down to identifying an active cpu within the same freq domain, if any.
> > > > + */
> > > > + if (!housekeeping_cpu(cpu, HK_TYPE_TICK) ||
> > > > + time_is_before_jiffies(last_update + msecs_to_jiffies(AMU_SAMPLE_EXP_MS))) {
> > > > + struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
> > > > + int ref_cpu = cpu;
> > > > +
> > > > + if (!policy)
> > > > + return 0;
> > > > +
> > >
> > > We can skip the rest of code if policy has a single cpu. AFAIR, one of the
> > > previous versions had similar check.
> > >
> > > if (!policy_is_shared(policy)) {
> > > cpufreq_cpu_put(policy);
> > > goto freq_comput;
> > > }
> > True, we could but then this case is covered by cpumask_next_wrap
> > which for single-cpu policies will render the ref_cpu invalid,
> > so policy_is_shared check seemed unnecessary.
> > >
> > > > + if (!cpumask_intersects(policy->related_cpus,
> > > > + housekeeping_cpumask(HK_TYPE_TICK))) {
> > > > + cpufreq_cpu_put(policy);
> > > > + return -EOPNOTSUPP;
> > > > + }
> > > > +
> > > > +
> > > > + do {
> > > > + ref_cpu = cpumask_next_wrap(ref_cpu, policy->cpus,
> > > > + start_cpu, false);
> > > > +
> > > > + } while (ref_cpu < nr_cpu_ids && idle_cpu(ref_cpu));
> > > > +
> > > > + cpufreq_cpu_put(policy);
> > > > +
> > > > + if (ref_cpu >= nr_cpu_ids)
> > > > + /* No alternative to pull info from */
> > > > + return 0;
> > > > +
> > >
> > > The 'cpuinfo_avg_freq' node gives 'unknown' value for single CPU per policy
> > > as 'ref_cpu' increments to 'nr_cpu_ids'. We can use the same CPU instead of
> > > returning zero if no alternative CPU.
> > >
> > > # cat /sys/devices/system/cpu/cpu2/cpufreq/cpuinfo_avg_freq
> > > <unknown>
> > >
> > > ----
> > > if (ref_cpu >= nr_cpu_ids)
> > > /* Use same CPU if no alternative to pull info from */
> > > goto freq_comput;
> > >
> > > ..
> > > freq_comput:
> > > scale = arch_scale_freq_capacity(cpu);
> > > freq = scale * arch_scale_freq_ref(cpu);
> > > ----
> > >
> > This boils down to the question what that function, and the information it
> > provides, represent really. The 'unknown' here simply says the CPU has been idle
> > for a while and as such the frequency data is a bit stale and it does not
> > represent the average freq the CPU is actually running at anymore, which is
> > the intention here really. Or, that the given CPU is a non-housekeeping one.
> > Either way I believe this is a useful information, instead of providing
> > stale data with no indication on whether the frequency is really the 'current'
> > one or not.
> >
> > If that is somehow undesirable we can discuss this further, though I'd rather
> > avoid exposing an interface where the feedback provided is open to
> > interpretation at all times.
>
> Would it make sense to identify that the frequency reporting is unknown due to
> cpu being idle vs some other issue like being a non-housekeeping CPU? Would
> returning a value of 0 make it easier for tools to represent that the CPU is
> currently idle?
That is an option.
Another one would be to return an error for those cases. This would make it
easier to distinguish between valid frequency &/| idle CPU vs tickless CPU
(EINVAL vs ENOENT) ?
---
BR
Beata
>
> Thanks,
> Vanshidhar
>
> >
> > ---
> > Best Regards
> > Beata
> > > Thank you,
> > > Sumit Gupta
> > >
> > > P.S. Will be on afk for next 2 weeks with no access to email. Please expect
> > > a delay in response.
> > >
> > > > + cpu = ref_cpu;
> > > > + goto retry;
> > > > + }
> > > > + /*
> > > > + * Reversed computation to the one used to determine
> > > > + * the arch_freq_scale value
> > > > + * (see amu_scale_freq_tick for details)
> > > > + */
> > > > + scale = arch_scale_freq_capacity(cpu);
> > > > + freq = scale * arch_scale_freq_ref(cpu);
> > > > + freq >>= SCHED_CAPACITY_SHIFT;
> > > > + return freq;
> > > > +}
> > > > +
> > >
> > > > static void amu_fie_setup(const struct cpumask *cpus)
> > > > {
> > > > int cpu;
> > > > --
> > > > 2.25.1
> > > >
