From 384e5699e101b1ee184590a39ee60f10ec0d36b6 Mon Sep 17 00:00:00 2001 From: Viresh Kumar Date: Fri, 8 Jan 2021 16:46:51 +0530 Subject: [PATCH 1/3] arm64: topology: Avoid the have_policy check Every time I have stumbled upon this routine, I get confused with the way 'have_policy' is used and I have to dig in to understand why is it so. Here is an attempt to make it easier to understand, and hopefully it is an improvement. The 'have_policy' check was just an optimization to avoid writing to amu_fie_cpus in case we don't have to, but that optimization itself is creating more confusion than the real work. Lets just do that if all the CPUs support AMUs. It is much cleaner that way. Reviewed-by: Ionela Voinescu Signed-off-by: Viresh Kumar Tested-by: Ionela Voinescu Link: https://lore.kernel.org/r/c125766c4be93461772015ac7c9a6ae45d5756f6.1610104461.git.viresh.kumar@linaro.org Signed-off-by: Will Deacon --- arch/arm64/kernel/topology.c | 20 ++++++-------------- 1 file changed, 6 insertions(+), 14 deletions(-) diff --git a/arch/arm64/kernel/topology.c b/arch/arm64/kernel/topology.c index f6faa697e83e..ebadc73449f9 100644 --- a/arch/arm64/kernel/topology.c +++ b/arch/arm64/kernel/topology.c @@ -199,14 +199,14 @@ static int freq_inv_set_max_ratio(int cpu, u64 max_rate, u64 ref_rate) return 0; } -static inline bool +static inline void 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; + return; } if (cpumask_subset(policy->related_cpus, valid_cpus)) @@ -214,8 +214,6 @@ enable_policy_freq_counters(int cpu, cpumask_var_t valid_cpus) amu_fie_cpus); cpufreq_cpu_put(policy); - - return true; } static DEFINE_STATIC_KEY_FALSE(amu_fie_key); @@ -225,7 +223,6 @@ static int __init init_amu_fie(void) { bool invariance_status = topology_scale_freq_invariant(); cpumask_var_t valid_cpus; - bool have_policy = false; int ret = 0; int cpu; @@ -245,17 +242,12 @@ static int __init init_amu_fie(void) continue; cpumask_set_cpu(cpu, valid_cpus); - have_policy |= enable_policy_freq_counters(cpu, valid_cpus); + enable_policy_freq_counters(cpu, valid_cpus); } - /* - * 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 (!have_policy && cpumask_equal(valid_cpus, cpu_present_mask)) - cpumask_or(amu_fie_cpus, amu_fie_cpus, valid_cpus); + /* Overwrite amu_fie_cpus if all CPUs support AMU */ + if (cpumask_equal(valid_cpus, cpu_present_mask)) + cpumask_copy(amu_fie_cpus, cpu_present_mask); if (!cpumask_empty(amu_fie_cpus)) { pr_info("CPUs[%*pbl]: counters will be used for FIE.", From 47b10b737c0794c2fa584d7c8103b485e274ed51 Mon Sep 17 00:00:00 2001 From: Viresh Kumar Date: Fri, 8 Jan 2021 16:46:52 +0530 Subject: [PATCH 2/3] arm64: topology: Reorder init_amu_fie() a bit This patch does a couple of optimizations in init_amu_fie(), like early exits from paths where we don't need to continue any further, avoid the enable/disable dance, moving the calls to topology_scale_freq_invariant() just when we need them, instead of at the top of the routine, and avoiding calling it for the third time. Reviewed-by: Ionela Voinescu Signed-off-by: Viresh Kumar Tested-by: Ionela Voinescu Link: https://lore.kernel.org/r/a732e71ab9ec28c354eb28dd898c9b47d490863f.1610104461.git.viresh.kumar@linaro.org Signed-off-by: Will Deacon --- arch/arm64/kernel/topology.c | 27 ++++++++++++++------------- 1 file changed, 14 insertions(+), 13 deletions(-) diff --git a/arch/arm64/kernel/topology.c b/arch/arm64/kernel/topology.c index ebadc73449f9..57267d694495 100644 --- a/arch/arm64/kernel/topology.c +++ b/arch/arm64/kernel/topology.c @@ -221,8 +221,8 @@ static DEFINE_STATIC_KEY_FALSE(amu_fie_key); static int __init init_amu_fie(void) { - bool invariance_status = topology_scale_freq_invariant(); cpumask_var_t valid_cpus; + bool invariant; int ret = 0; int cpu; @@ -249,18 +249,19 @@ static int __init init_amu_fie(void) if (cpumask_equal(valid_cpus, cpu_present_mask)) cpumask_copy(amu_fie_cpus, cpu_present_mask); - if (!cpumask_empty(amu_fie_cpus)) { - pr_info("CPUs[%*pbl]: counters will be used for FIE.", - cpumask_pr_args(amu_fie_cpus)); - static_branch_enable(&amu_fie_key); - } + if (cpumask_empty(amu_fie_cpus)) + goto free_valid_mask; - /* - * If the system is not fully invariant after AMU init, disable - * partial use of counters for frequency invariance. - */ - if (!topology_scale_freq_invariant()) - static_branch_disable(&amu_fie_key); + invariant = topology_scale_freq_invariant(); + + /* We aren't fully invariant yet */ + if (!invariant && !cpumask_equal(amu_fie_cpus, cpu_present_mask)) + goto free_valid_mask; + + static_branch_enable(&amu_fie_key); + + pr_info("CPUs[%*pbl]: counters will be used for FIE.", + cpumask_pr_args(amu_fie_cpus)); /* * Task scheduler behavior depends on frequency invariance support, @@ -268,7 +269,7 @@ static int __init init_amu_fie(void) * a result of counter initialisation and use, retrigger the build of * scheduling domains to ensure the information is propagated properly. */ - if (invariance_status != topology_scale_freq_invariant()) + if (!invariant) rebuild_sched_domains_energy(); free_valid_mask: From a5f1b187cd24f7da35eb7a48fc50839c554d52a2 Mon Sep 17 00:00:00 2001 From: Viresh Kumar Date: Fri, 8 Jan 2021 16:46:53 +0530 Subject: [PATCH 3/3] arm64: topology: Make AMUs work with modular cpufreq drivers The AMU counters won't get used today if the cpufreq driver is built as a module as the amu core requires everything to be ready by late init. Fix that properly by registering for cpufreq policy notifier. Note that the amu core don't have any cpufreq dependency after the first time CPUFREQ_CREATE_POLICY notifier is called for all the CPUs. And so we don't need to do anything on the CPUFREQ_REMOVE_POLICY notifier. And for the same reason we check if the CPUs are already parsed in the beginning of amu_fie_setup() and skip if that is true. Alternatively we can shoot a work from there to unregister the notifier instead, but that seemed too much instead of this simple check. While at it, convert the print message to pr_debug instead of pr_info. Signed-off-by: Viresh Kumar Reviewed-by: Ionela Voinescu Tested-by: Ionela Voinescu Link: https://lore.kernel.org/r/89c1921334443e133c9c8791b4693607d65ed9f5.1610104461.git.viresh.kumar@linaro.org Signed-off-by: Will Deacon --- arch/arm64/kernel/topology.c | 92 +++++++++++++++++++----------------- 1 file changed, 48 insertions(+), 44 deletions(-) diff --git a/arch/arm64/kernel/topology.c b/arch/arm64/kernel/topology.c index 57267d694495..e08a4126453a 100644 --- a/arch/arm64/kernel/topology.c +++ b/arch/arm64/kernel/topology.c @@ -199,69 +199,38 @@ static int freq_inv_set_max_ratio(int cpu, u64 max_rate, u64 ref_rate) return 0; } -static inline void -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; - } - - if (cpumask_subset(policy->related_cpus, valid_cpus)) - cpumask_or(amu_fie_cpus, policy->related_cpus, - amu_fie_cpus); - - cpufreq_cpu_put(policy); -} - static DEFINE_STATIC_KEY_FALSE(amu_fie_key); #define amu_freq_invariant() static_branch_unlikely(&amu_fie_key) -static int __init init_amu_fie(void) +static void amu_fie_setup(const struct cpumask *cpus) { - cpumask_var_t valid_cpus; bool invariant; - int ret = 0; int cpu; - if (!zalloc_cpumask_var(&valid_cpus, GFP_KERNEL)) - return -ENOMEM; + /* We are already set since the last insmod of cpufreq driver */ + if (unlikely(cpumask_subset(cpus, amu_fie_cpus))) + return; - if (!zalloc_cpumask_var(&amu_fie_cpus, GFP_KERNEL)) { - ret = -ENOMEM; - goto free_valid_mask; - } - - for_each_present_cpu(cpu) { + for_each_cpu(cpu, cpus) { if (!freq_counters_valid(cpu) || freq_inv_set_max_ratio(cpu, cpufreq_get_hw_max_freq(cpu) * 1000, arch_timer_get_rate())) - continue; - - cpumask_set_cpu(cpu, valid_cpus); - enable_policy_freq_counters(cpu, valid_cpus); + return; } - /* Overwrite amu_fie_cpus if all CPUs support AMU */ - if (cpumask_equal(valid_cpus, cpu_present_mask)) - cpumask_copy(amu_fie_cpus, cpu_present_mask); - - if (cpumask_empty(amu_fie_cpus)) - goto free_valid_mask; + cpumask_or(amu_fie_cpus, amu_fie_cpus, cpus); invariant = topology_scale_freq_invariant(); /* We aren't fully invariant yet */ if (!invariant && !cpumask_equal(amu_fie_cpus, cpu_present_mask)) - goto free_valid_mask; + return; static_branch_enable(&amu_fie_key); - pr_info("CPUs[%*pbl]: counters will be used for FIE.", - cpumask_pr_args(amu_fie_cpus)); + pr_debug("CPUs[%*pbl]: counters will be used for FIE.", + cpumask_pr_args(cpus)); /* * Task scheduler behavior depends on frequency invariance support, @@ -271,13 +240,48 @@ static int __init init_amu_fie(void) */ if (!invariant) rebuild_sched_domains_energy(); +} -free_valid_mask: - free_cpumask_var(valid_cpus); +static int init_amu_fie_callback(struct notifier_block *nb, unsigned long val, + void *data) +{ + struct cpufreq_policy *policy = data; + + if (val == CPUFREQ_CREATE_POLICY) + amu_fie_setup(policy->related_cpus); + + /* + * We don't need to handle CPUFREQ_REMOVE_POLICY event as the AMU + * counters don't have any dependency on cpufreq driver once we have + * initialized AMU support and enabled invariance. The AMU counters will + * keep on working just fine in the absence of the cpufreq driver, and + * for the CPUs for which there are no counters available, the last set + * value of freq_scale will remain valid as that is the frequency those + * CPUs are running at. + */ + + return 0; +} + +static struct notifier_block init_amu_fie_notifier = { + .notifier_call = init_amu_fie_callback, +}; + +static int __init init_amu_fie(void) +{ + int ret; + + if (!zalloc_cpumask_var(&amu_fie_cpus, GFP_KERNEL)) + return -ENOMEM; + + ret = cpufreq_register_notifier(&init_amu_fie_notifier, + CPUFREQ_POLICY_NOTIFIER); + if (ret) + free_cpumask_var(amu_fie_cpus); return ret; } -late_initcall_sync(init_amu_fie); +core_initcall(init_amu_fie); bool arch_freq_counters_available(const struct cpumask *cpus) {