linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Ionela Voinescu <ionela.voinescu@arm.com>
To: catalin.marinas@arm.com, will@kernel.org, mark.rutland@arm.com,
	maz@kernel.org, suzuki.poulose@arm.com, sudeep.holla@arm.com,
	lukasz.luba@arm.com, valentin.schneider@arm.com,
	rjw@rjwysocki.net, ionela.voinescu@arm.com
Cc: peterz@infradead.org, mingo@redhat.com,
	vincent.guittot@linaro.org, viresh.kumar@linaro.org,
	linux-arm-kernel@lists.infradead.org, linux-doc@vger.kernel.org,
	linux-kernel@vger.kernel.org, linux-pm@vger.kernel.org
Subject: [PATCH v3 6/7] arm64: use activity monitors for frequency invariance
Date: Tue, 11 Feb 2020 18:45:41 +0000	[thread overview]
Message-ID: <20200211184542.29585-7-ionela.voinescu@arm.com> (raw)
In-Reply-To: <20200211184542.29585-1-ionela.voinescu@arm.com>

The Frequency Invariance Engine (FIE) is providing a frequency
scaling correction factor that helps achieve more accurate
load-tracking.

So far, for arm and arm64 platforms, this scale factor has been
obtained based on the ratio between the current frequency and the
maximum supported frequency recorded by the cpufreq policy. The
setting of this scale factor is triggered from cpufreq drivers by
calling arch_set_freq_scale. The current frequency used in computation
is the frequency requested by a governor, but it may not be the
frequency that was implemented by the platform.

This correction factor can also be obtained using a core counter and a
constant counter to get information on the performance (frequency based
only) obtained in a period of time. This will more accurately reflect
the actual current frequency of the CPU, compared with the alternative
implementation that reflects the request of a performance level from
the OS.

Therefore, implement arch_scale_freq_tick to use activity monitors, if
present, for the computation of the frequency scale factor.

The use of AMU counters depends on:
 - CONFIG_ARM64_AMU_EXTN - depents on the AMU extension being present
 - (optional) CONFIG_CPU_FREQ - the current frequency obtained using
   counter information is divided by the maximum frequency obtained from
   the cpufreq policy. But the use of cpufreq policy maximum frequency
   is weak and cpu_get_max_freq can be redefined to provide the data
   some other way.

While it is possible to have a combination of CPUs in the system with
and without support for activity monitors, the use of counters for
frequency invariance is only enabled for a CPU if all related CPUs
(CPUs in the same frequency domain) support and have enabled the core
and constant activity monitor counters. In this way, there is a clear
separation between the policies for which arch_set_freq_scale (cpufreq
based FIE) is used, and the policies for which arch_scale_freq_tick
(counter based FIE) is used to set the frequency scale factor. For
this purpose, a late_initcall_sync is registered to trigger validation
work for policies that will enable or disable the use of AMU counters
for frequency invariance. If CONFIG_CPU_FREQ is not defined, the use
of counters is enabled on all CPUs only if all possible CPUs correctly
support the necessary counters.

Signed-off-by: Ionela Voinescu <ionela.voinescu@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Sudeep Holla <sudeep.holla@arm.com>
---
 arch/arm64/include/asm/topology.h |  16 +++
 arch/arm64/kernel/cpufeature.c    |   4 +
 arch/arm64/kernel/topology.c      | 185 ++++++++++++++++++++++++++++++
 drivers/base/arch_topology.c      |   8 ++
 include/linux/topology.h          |   7 ++
 5 files changed, 220 insertions(+)

diff --git a/arch/arm64/include/asm/topology.h b/arch/arm64/include/asm/topology.h
index a4d945db95a2..d910d463cf76 100644
--- a/arch/arm64/include/asm/topology.h
+++ b/arch/arm64/include/asm/topology.h
@@ -16,6 +16,22 @@ int pcibus_to_node(struct pci_bus *bus);
 
 #include <linux/arch_topology.h>
 
+#ifdef CONFIG_ARM64_AMU_EXTN
+extern unsigned int cpu_get_max_freq(unsigned int cpu);
+/*
+ * Replace default function that signals use of counters
+ * for frequency invariance for given CPUs.
+ */
+bool topology_cpu_freq_counters(struct cpumask *cpus);
+#define arch_cpu_freq_counters topology_cpu_freq_counters
+/*
+ * Replace task scheduler's default counter-based
+ * frequency-invariance scale factor setting.
+ */
+void topology_scale_freq_tick(void);
+#define arch_scale_freq_tick topology_scale_freq_tick
+#endif /* CONFIG_ARM64_AMU_EXTN */
+
 /* Replace task scheduler's default frequency-invariant accounting */
 #define arch_scale_freq_capacity topology_get_freq_scale
 
diff --git a/arch/arm64/kernel/cpufeature.c b/arch/arm64/kernel/cpufeature.c
index 029a473ad273..a4620b269b56 100644
--- a/arch/arm64/kernel/cpufeature.c
+++ b/arch/arm64/kernel/cpufeature.c
@@ -1172,12 +1172,16 @@ bool cpu_has_amu_feat(int cpu)
 	return false;
 }
 
+/* Initialize the use of AMU counters for frequency invariance */
+extern void init_cpu_freq_invariance_counters(void);
+
 static void cpu_amu_enable(struct arm64_cpu_capabilities const *cap)
 {
 	if (has_cpuid_feature(cap, SCOPE_LOCAL_CPU)) {
 		pr_info("detected CPU%d: Activity Monitors Unit (AMU)\n",
 			smp_processor_id());
 		cpumask_set_cpu(smp_processor_id(), amu_cpus);
+		init_cpu_freq_invariance_counters();
 	}
 }
 
diff --git a/arch/arm64/kernel/topology.c b/arch/arm64/kernel/topology.c
index fa9528dfd0ce..3f6e379f07b3 100644
--- a/arch/arm64/kernel/topology.c
+++ b/arch/arm64/kernel/topology.c
@@ -14,6 +14,7 @@
 #include <linux/acpi.h>
 #include <linux/arch_topology.h>
 #include <linux/cacheinfo.h>
+#include <linux/cpufreq.h>
 #include <linux/init.h>
 #include <linux/percpu.h>
 
@@ -120,4 +121,188 @@ int __init parse_acpi_topology(void)
 }
 #endif
 
+#ifdef CONFIG_ARM64_AMU_EXTN
 
+#undef pr_fmt
+#define pr_fmt(fmt) "AMU: " fmt
+
+static DEFINE_PER_CPU_READ_MOSTLY(unsigned long, arch_max_freq_scale);
+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;
+
+/* Obtain max frequency (in KHz) as reported by hardware */
+__weak unsigned int cpu_get_max_freq(unsigned int cpu)
+{
+	return 0;
+}
+
+#ifdef CONFIG_CPU_FREQ
+/* Replace max frequency getter with cpufreq based function */
+#define cpu_get_max_freq cpufreq_get_hw_max_freq
+#endif
+
+/* Initialize counter reference per-cpu variables for the current CPU */
+void init_cpu_freq_invariance_counters(void)
+{
+	this_cpu_write(arch_core_cycles_prev,
+		       read_sysreg_s(SYS_AMEVCNTR0_CORE_EL0));
+	this_cpu_write(arch_const_cycles_prev,
+		       read_sysreg_s(SYS_AMEVCNTR0_CONST_EL0));
+}
+
+static int validate_cpu_freq_invariance_counters(int cpu)
+{
+	u64 max_freq_hz, ratio;
+
+	if (!cpu_has_amu_feat(cpu)) {
+		pr_debug("CPU%d: counters are not supported.\n", cpu);
+		return -EINVAL;
+	}
+
+	if (unlikely(!per_cpu(arch_const_cycles_prev, cpu) ||
+		     !per_cpu(arch_core_cycles_prev, cpu))) {
+		pr_debug("CPU%d: cycle counters are not enabled.\n", cpu);
+		return -EINVAL;
+	}
+
+	/* Convert maximum frequency from KHz to Hz and validate */
+	max_freq_hz = cpu_get_max_freq(cpu) * 1000;
+	if (unlikely(!max_freq_hz)) {
+		pr_debug("CPU%d: invalid maximum frequency.\n", cpu);
+		return -EINVAL;
+	}
+
+	/*
+	 * Pre-compute the fixed ratio between the frequency of the constant
+	 * counter and the maximum frequency of the CPU.
+	 *
+	 *			      const_freq
+	 * arch_max_freq_scale =   ---------------- * SCHED_CAPACITY_SCALE²
+	 *			   cpuinfo_max_freq
+	 *
+	 * We use a factor of 2 * SCHED_CAPACITY_SHIFT -> SCHED_CAPACITY_SCALE²
+	 * in order to ensure a good resolution for arch_max_freq_scale for
+	 * very low arch timer frequencies (up to the KHz range which should be
+	 * unlikely).
+	 */
+	ratio = (u64)arch_timer_get_rate() << (2 * SCHED_CAPACITY_SHIFT);
+	ratio = div64_u64(ratio, max_freq_hz);
+	if (!ratio) {
+		pr_err("System timer frequency too low.\n");
+		return -EINVAL;
+	}
+
+	per_cpu(arch_max_freq_scale, cpu) = (unsigned long)ratio;
+
+	return 0;
+}
+
+static inline int
+enable_policy_freq_counters(int cpu, cpumask_var_t valid_cpus)
+{
+	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
+
+	if (!policy) {
+		pr_debug("CPU%d: No cpufreq policy found.\n", cpu);
+		return false;
+	}
+
+	if (cpumask_subset(policy->related_cpus, valid_cpus)) {
+		cpumask_or(amu_fie_cpus, policy->related_cpus,
+			   amu_fie_cpus);
+		pr_info("CPUs[%*pbl]: counters will be used for FIE.",
+			cpumask_pr_args(amu_fie_cpus));
+	}
+
+	cpufreq_cpu_put(policy);
+
+	return true;
+}
+
+static int __init init_amu_fie(void)
+{
+	cpumask_var_t valid_cpus;
+	bool have_policy = false;
+	int cpu;
+
+	if (!zalloc_cpumask_var(&valid_cpus, GFP_KERNEL) ||
+	    !zalloc_cpumask_var(&amu_fie_cpus, GFP_KERNEL))
+		return -ENOMEM;
+
+	for_each_possible_cpu(cpu) {
+		if (validate_cpu_freq_invariance_counters(cpu))
+			continue;
+		cpumask_set_cpu(cpu, valid_cpus);
+		have_policy = enable_policy_freq_counters(cpu, valid_cpus) ||
+			      have_policy;
+	}
+
+	if (!have_policy) {
+		/*
+		 * If we are not restricted by cpufreq policies, we only enable
+		 * the use of the AMU feature for FIE if all CPUs support AMU.
+		 * Otherwise, enable_policy_freq_counters has already enabled
+		 * policy cpus.
+		 */
+		if (cpumask_equal(valid_cpus, cpu_possible_mask)) {
+			cpumask_or(amu_fie_cpus, amu_fie_cpus, valid_cpus);
+			pr_info("CPUs[%*pbl]: counters will be used for FIE.",
+				cpumask_pr_args(amu_fie_cpus));
+		}
+	}
+
+	free_cpumask_var(valid_cpus);
+
+	return 0;
+}
+late_initcall_sync(init_amu_fie);
+
+bool topology_cpu_freq_counters(struct cpumask *cpus)
+{
+	return cpumask_available(amu_fie_cpus) &&
+	       cpumask_subset(cpus, amu_fie_cpus);
+}
+
+void topology_scale_freq_tick(void)
+{
+	u64 prev_core_cnt, prev_const_cnt;
+	u64 core_cnt, const_cnt, scale;
+	int cpu = smp_processor_id();
+
+	if (!cpumask_available(amu_fie_cpus) ||
+	    !cpumask_test_cpu(cpu, amu_fie_cpus))
+		return;
+
+	const_cnt = read_sysreg_s(SYS_AMEVCNTR0_CONST_EL0);
+	core_cnt = read_sysreg_s(SYS_AMEVCNTR0_CORE_EL0);
+	prev_const_cnt = this_cpu_read(arch_const_cycles_prev);
+	prev_core_cnt = this_cpu_read(arch_core_cycles_prev);
+
+	if (unlikely(core_cnt <= prev_core_cnt ||
+		     const_cnt <= prev_const_cnt))
+		goto store_and_exit;
+
+	/*
+	 *	    /\core    arch_max_freq_scale
+	 * scale =  ------- * --------------------
+	 *	    /\const   SCHED_CAPACITY_SCALE
+	 *
+	 * We shift by SCHED_CAPACITY_SHIFT (divide by SCHED_CAPACITY_SCALE)
+	 * in order to compensate for the SCHED_CAPACITY_SCALE² factor in
+	 * arch_max_freq_scale (used to ensure its resolution) while keeping
+	 * the scale value in the 0-SCHED_CAPACITY_SCALE capacity range.
+	 */
+	scale = core_cnt - prev_core_cnt;
+	scale *= this_cpu_read(arch_max_freq_scale);
+	scale = div64_u64(scale >> SCHED_CAPACITY_SHIFT,
+			  const_cnt - prev_const_cnt);
+
+	scale = min_t(unsigned long, scale, SCHED_CAPACITY_SCALE);
+	this_cpu_write(freq_scale, (unsigned long)scale);
+
+store_and_exit:
+	this_cpu_write(arch_core_cycles_prev, core_cnt);
+	this_cpu_write(arch_const_cycles_prev, const_cnt);
+}
+#endif /* CONFIG_ARM64_AMU_EXTN */
diff --git a/drivers/base/arch_topology.c b/drivers/base/arch_topology.c
index 1eb81f113786..1ab2b7503d63 100644
--- a/drivers/base/arch_topology.c
+++ b/drivers/base/arch_topology.c
@@ -29,6 +29,14 @@ void arch_set_freq_scale(struct cpumask *cpus, unsigned long cur_freq,
 	unsigned long scale;
 	int i;
 
+	/*
+	 * If the use of counters for FIE is enabled, just return as we don't
+	 * want to update the scale factor with information from CPUFREQ.
+	 * Instead the scale factor will be updated from arch_scale_freq_tick.
+	 */
+	if (arch_cpu_freq_counters(cpus))
+		return;
+
 	scale = (cur_freq << SCHED_CAPACITY_SHIFT) / max_freq;
 
 	for_each_cpu(i, cpus)
diff --git a/include/linux/topology.h b/include/linux/topology.h
index eb2fe6edd73c..397aad6ae163 100644
--- a/include/linux/topology.h
+++ b/include/linux/topology.h
@@ -227,5 +227,12 @@ static inline const struct cpumask *cpu_cpu_mask(int cpu)
 	return cpumask_of_node(cpu_to_node(cpu));
 }
 
+#ifndef arch_cpu_freq_counters
+static __always_inline
+bool arch_cpu_freq_counters(struct cpumask *cpus)
+{
+	return false;
+}
+#endif
 
 #endif /* _LINUX_TOPOLOGY_H */
-- 
2.17.1


  parent reply	other threads:[~2020-02-11 18:46 UTC|newest]

Thread overview: 42+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-02-11 18:45 [PATCH v3 0/7] arm64: ARMv8.4 Activity Monitors support Ionela Voinescu
2020-02-11 18:45 ` [PATCH v3 1/7] arm64: add support for the AMU extension v1 Ionela Voinescu
2020-02-12 11:30   ` Suzuki Kuruppassery Poulose
2020-02-12 14:54     ` Valentin Schneider
2020-02-12 16:10     ` Ionela Voinescu
2020-02-12 16:20       ` Suzuki Kuruppassery Poulose
2020-02-12 18:20         ` Ionela Voinescu
2020-02-12 19:24           ` Suzuki K Poulose
2020-02-12 20:19         ` Ionela Voinescu
2020-02-12 16:24       ` Vladimir Murzin
2020-02-12 18:27         ` Ionela Voinescu
2020-02-11 18:45 ` [PATCH v3 2/7] arm64: trap to EL1 accesses to AMU counters from EL0 Ionela Voinescu
2020-02-12 11:44   ` Suzuki Kuruppassery Poulose
2020-02-12 15:36   ` Valentin Schneider
2020-02-11 18:45 ` [PATCH v3 3/7] arm64/kvm: disable access to AMU registers from kvm guests Ionela Voinescu
2020-02-12 15:36   ` Valentin Schneider
2020-02-12 16:33   ` Suzuki Kuruppassery Poulose
2020-02-11 18:45 ` [PATCH v3 4/7] Documentation: arm64: document support for the AMU extension Ionela Voinescu
2020-02-12 15:36   ` Valentin Schneider
2020-02-11 18:45 ` [PATCH v3 5/7] cpufreq: add function to get the hardware max frequency Ionela Voinescu
2020-02-12  4:14   ` Viresh Kumar
2020-02-13 11:59   ` Valentin Schneider
2020-02-13 12:59     ` Ionela Voinescu
2020-02-13 15:22       ` Valentin Schneider
2020-02-11 18:45 ` Ionela Voinescu [this message]
2020-02-12 18:59   ` [PATCH v3 6/7] arm64: use activity monitors for frequency invariance Lukasz Luba
2020-02-13  9:47     ` Ionela Voinescu
2020-02-17 16:59   ` Valentin Schneider
2020-02-23 18:49     ` Ionela Voinescu
2020-02-11 18:45 ` [PATCH v3 7/7] clocksource/drivers/arm_arch_timer: validate arch_timer_rate Ionela Voinescu
2020-02-12  9:30   ` Valentin Schneider
2020-02-12 10:32     ` Ionela Voinescu
2020-02-12 10:01   ` Lukasz Luba
2020-02-12 10:12     ` Marc Zyngier
2020-02-12 10:54       ` Ionela Voinescu
2020-02-12 10:55       ` Lukasz Luba
2020-02-12 11:10         ` Marc Zyngier
2020-02-12 11:43           ` Lukasz Luba
2020-02-12 11:12         ` Valentin Schneider
2020-02-14  0:35   ` Thomas Gleixner
2020-02-14 15:45     ` Ionela Voinescu
2020-02-14 15:57       ` Ionela Voinescu

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20200211184542.29585-7-ionela.voinescu@arm.com \
    --to=ionela.voinescu@arm.com \
    --cc=catalin.marinas@arm.com \
    --cc=linux-arm-kernel@lists.infradead.org \
    --cc=linux-doc@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-pm@vger.kernel.org \
    --cc=lukasz.luba@arm.com \
    --cc=mark.rutland@arm.com \
    --cc=maz@kernel.org \
    --cc=mingo@redhat.com \
    --cc=peterz@infradead.org \
    --cc=rjw@rjwysocki.net \
    --cc=sudeep.holla@arm.com \
    --cc=suzuki.poulose@arm.com \
    --cc=valentin.schneider@arm.com \
    --cc=vincent.guittot@linaro.org \
    --cc=viresh.kumar@linaro.org \
    --cc=will@kernel.org \
    --subject='Re: [PATCH v3 6/7] arm64: use activity monitors for frequency invariance' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).