diff options
Diffstat (limited to 'kernel/sched/topology.c')
| -rw-r--r-- | kernel/sched/topology.c | 1107 |
1 files changed, 764 insertions, 343 deletions
diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c index 5d3675c7a76b..c49aea8c1025 100644 --- a/kernel/sched/topology.c +++ b/kernel/sched/topology.c @@ -2,7 +2,8 @@ /* * Scheduler topology setup/handling methods */ -#include "sched.h" + +#include <linux/bsearch.h> DEFINE_MUTEX(sched_domains_mutex); @@ -14,15 +15,15 @@ static cpumask_var_t sched_domains_tmpmask2; static int __init sched_debug_setup(char *str) { - sched_debug_enabled = true; + sched_debug_verbose = true; return 0; } -early_param("sched_debug", sched_debug_setup); +early_param("sched_verbose", sched_debug_setup); static inline bool sched_debug(void) { - return sched_debug_enabled; + return sched_debug_verbose; } #define SD_FLAG(_name, mflags) [__##_name] = { .meta_flags = mflags, .name = #_name }, @@ -74,7 +75,7 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, break; } - if (!cpumask_weight(sched_group_span(group))) { + if (cpumask_empty(sched_group_span(group))) { printk(KERN_CONT "\n"); printk(KERN_ERR "ERROR: empty group\n"); break; @@ -131,7 +132,7 @@ static void sched_domain_debug(struct sched_domain *sd, int cpu) { int level = 0; - if (!sched_debug_enabled) + if (!sched_debug_verbose) return; if (!sd) { @@ -152,7 +153,7 @@ static void sched_domain_debug(struct sched_domain *sd, int cpu) } #else /* !CONFIG_SCHED_DEBUG */ -# define sched_debug_enabled 0 +# define sched_debug_verbose 0 # define sched_domain_debug(sd, cpu) do { } while (0) static inline bool sched_debug(void) { @@ -207,9 +208,72 @@ sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL) DEFINE_STATIC_KEY_FALSE(sched_energy_present); -unsigned int sysctl_sched_energy_aware = 1; -DEFINE_MUTEX(sched_energy_mutex); -bool sched_energy_update; +static unsigned int sysctl_sched_energy_aware = 1; +static DEFINE_MUTEX(sched_energy_mutex); +static bool sched_energy_update; + +static bool sched_is_eas_possible(const struct cpumask *cpu_mask) +{ + bool any_asym_capacity = false; + struct cpufreq_policy *policy; + struct cpufreq_governor *gov; + int i; + + /* EAS is enabled for asymmetric CPU capacity topologies. */ + for_each_cpu(i, cpu_mask) { + if (rcu_access_pointer(per_cpu(sd_asym_cpucapacity, i))) { + any_asym_capacity = true; + break; + } + } + if (!any_asym_capacity) { + if (sched_debug()) { + pr_info("rd %*pbl: Checking EAS, CPUs do not have asymmetric capacities\n", + cpumask_pr_args(cpu_mask)); + } + return false; + } + + /* EAS definitely does *not* handle SMT */ + if (sched_smt_active()) { + if (sched_debug()) { + pr_info("rd %*pbl: Checking EAS, SMT is not supported\n", + cpumask_pr_args(cpu_mask)); + } + return false; + } + + if (!arch_scale_freq_invariant()) { + if (sched_debug()) { + pr_info("rd %*pbl: Checking EAS: frequency-invariant load tracking not yet supported", + cpumask_pr_args(cpu_mask)); + } + return false; + } + + /* Do not attempt EAS if schedutil is not being used. */ + for_each_cpu(i, cpu_mask) { + policy = cpufreq_cpu_get(i); + if (!policy) { + if (sched_debug()) { + pr_info("rd %*pbl: Checking EAS, cpufreq policy not set for CPU: %d", + cpumask_pr_args(cpu_mask), i); + } + return false; + } + gov = policy->governor; + cpufreq_cpu_put(policy); + if (gov != &schedutil_gov) { + if (sched_debug()) { + pr_info("rd %*pbl: Checking EAS, schedutil is mandatory\n", + cpumask_pr_args(cpu_mask)); + } + return false; + } + } + + return true; +} void rebuild_sched_domains_energy(void) { @@ -221,7 +285,7 @@ void rebuild_sched_domains_energy(void) } #ifdef CONFIG_PROC_SYSCTL -int sched_energy_aware_handler(struct ctl_table *table, int write, +static int sched_energy_aware_handler(const struct ctl_table *table, int write, void *buffer, size_t *lenp, loff_t *ppos) { int ret, state; @@ -229,6 +293,15 @@ int sched_energy_aware_handler(struct ctl_table *table, int write, if (write && !capable(CAP_SYS_ADMIN)) return -EPERM; + if (!sched_is_eas_possible(cpu_active_mask)) { + if (write) { + return -EOPNOTSUPP; + } else { + *lenp = 0; + return 0; + } + } + ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); if (!ret && write) { state = static_branch_unlikely(&sched_energy_present); @@ -238,6 +311,26 @@ int sched_energy_aware_handler(struct ctl_table *table, int write, return ret; } + +static const struct ctl_table sched_energy_aware_sysctls[] = { + { + .procname = "sched_energy_aware", + .data = &sysctl_sched_energy_aware, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = sched_energy_aware_handler, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_ONE, + }, +}; + +static int __init sched_energy_aware_sysctl_init(void) +{ + register_sysctl_init("kernel", sched_energy_aware_sysctls); + return 0; +} + +late_initcall(sched_energy_aware_sysctl_init); #endif static void free_pd(struct perf_domain *pd) @@ -326,103 +419,33 @@ static void sched_energy_set(bool has_eas) * 1. an Energy Model (EM) is available; * 2. the SD_ASYM_CPUCAPACITY flag is set in the sched_domain hierarchy. * 3. no SMT is detected. - * 4. the EM complexity is low enough to keep scheduling overheads low; - * 5. schedutil is driving the frequency of all CPUs of the rd; - * 6. frequency invariance support is present; - * - * The complexity of the Energy Model is defined as: - * - * C = nr_pd * (nr_cpus + nr_ps) - * - * with parameters defined as: - * - nr_pd: the number of performance domains - * - nr_cpus: the number of CPUs - * - nr_ps: the sum of the number of performance states of all performance - * domains (for example, on a system with 2 performance domains, - * with 10 performance states each, nr_ps = 2 * 10 = 20). - * - * It is generally not a good idea to use such a model in the wake-up path on - * very complex platforms because of the associated scheduling overheads. The - * arbitrary constraint below prevents that. It makes EAS usable up to 16 CPUs - * with per-CPU DVFS and less than 8 performance states each, for example. + * 4. schedutil is driving the frequency of all CPUs of the rd; + * 5. frequency invariance support is present; */ -#define EM_MAX_COMPLEXITY 2048 - -extern struct cpufreq_governor schedutil_gov; static bool build_perf_domains(const struct cpumask *cpu_map) { - int i, nr_pd = 0, nr_ps = 0, nr_cpus = cpumask_weight(cpu_map); + int i; struct perf_domain *pd = NULL, *tmp; int cpu = cpumask_first(cpu_map); struct root_domain *rd = cpu_rq(cpu)->rd; - struct cpufreq_policy *policy; - struct cpufreq_governor *gov; if (!sysctl_sched_energy_aware) goto free; - /* EAS is enabled for asymmetric CPU capacity topologies. */ - if (!per_cpu(sd_asym_cpucapacity, cpu)) { - if (sched_debug()) { - pr_info("rd %*pbl: CPUs do not have asymmetric capacities\n", - cpumask_pr_args(cpu_map)); - } - goto free; - } - - /* EAS definitely does *not* handle SMT */ - if (sched_smt_active()) { - pr_warn("rd %*pbl: Disabling EAS, SMT is not supported\n", - cpumask_pr_args(cpu_map)); - goto free; - } - - if (!arch_scale_freq_invariant()) { - if (sched_debug()) { - pr_warn("rd %*pbl: Disabling EAS: frequency-invariant load tracking not yet supported", - cpumask_pr_args(cpu_map)); - } + if (!sched_is_eas_possible(cpu_map)) goto free; - } for_each_cpu(i, cpu_map) { /* Skip already covered CPUs. */ if (find_pd(pd, i)) continue; - /* Do not attempt EAS if schedutil is not being used. */ - policy = cpufreq_cpu_get(i); - if (!policy) - goto free; - gov = policy->governor; - cpufreq_cpu_put(policy); - if (gov != &schedutil_gov) { - if (rd->pd) - pr_warn("rd %*pbl: Disabling EAS, schedutil is mandatory\n", - cpumask_pr_args(cpu_map)); - goto free; - } - /* Create the new pd and add it to the local list. */ tmp = pd_init(i); if (!tmp) goto free; tmp->next = pd; pd = tmp; - - /* - * Count performance domains and performance states for the - * complexity check. - */ - nr_pd++; - nr_ps += em_pd_nr_perf_states(pd->em_pd); - } - - /* Bail out if the Energy Model complexity is too high. */ - if (nr_pd * (nr_ps + nr_cpus) > EM_MAX_COMPLEXITY) { - WARN(1, "rd %*pbl: Failed to start EAS, EM complexity is too high\n", - cpumask_pr_args(cpu_map)); - goto free; } perf_domain_debug(cpu_map, pd); @@ -465,9 +488,9 @@ static void free_rootdomain(struct rcu_head *rcu) void rq_attach_root(struct rq *rq, struct root_domain *rd) { struct root_domain *old_rd = NULL; - unsigned long flags; + struct rq_flags rf; - raw_spin_lock_irqsave(&rq->lock, flags); + rq_lock_irqsave(rq, &rf); if (rq->rd) { old_rd = rq->rd; @@ -478,7 +501,7 @@ void rq_attach_root(struct rq *rq, struct root_domain *rd) cpumask_clear_cpu(rq->cpu, old_rd->span); /* - * If we dont want to free the old_rd yet then + * If we don't want to free the old_rd yet then * set old_rd to NULL to skip the freeing later * in this function: */ @@ -493,7 +516,15 @@ void rq_attach_root(struct rq *rq, struct root_domain *rd) if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) set_rq_online(rq); - raw_spin_unlock_irqrestore(&rq->lock, flags); + /* + * Because the rq is not a task, dl_add_task_root_domain() did not + * move the fair server bw to the rd if it already started. + * Add it now. + */ + if (rq->fair_server.dl_server) + __dl_server_attach_root(&rq->fair_server, rq); + + rq_unlock_irqrestore(rq, &rf); if (old_rd) call_rcu(&old_rd->rcu, free_rootdomain); @@ -526,7 +557,7 @@ static int init_rootdomain(struct root_domain *rd) #ifdef HAVE_RT_PUSH_IPI rd->rto_cpu = -1; raw_spin_lock_init(&rd->rto_lock); - init_irq_work(&rd->rto_push_work, rto_push_irq_work_func); + rd->rto_push_work = IRQ_WORK_INIT_HARD(rto_push_irq_work_func); #endif rd->visit_gen = 0; @@ -558,7 +589,7 @@ out: */ struct root_domain def_root_domain; -void init_defrootdomain(void) +void __init init_defrootdomain(void) { init_rootdomain(&def_root_domain); @@ -633,22 +664,25 @@ static void destroy_sched_domains(struct sched_domain *sd) } /* - * Keep a special pointer to the highest sched_domain that has - * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this - * allows us to avoid some pointer chasing select_idle_sibling(). + * Keep a special pointer to the highest sched_domain that has SD_SHARE_LLC set + * (Last Level Cache Domain) for this allows us to avoid some pointer chasing + * select_idle_sibling(). * - * Also keep a unique ID per domain (we use the first CPU number in - * the cpumask of the domain), this allows us to quickly tell if - * two CPUs are in the same cache domain, see cpus_share_cache(). + * Also keep a unique ID per domain (we use the first CPU number in the cpumask + * of the domain), this allows us to quickly tell if two CPUs are in the same + * cache domain, see cpus_share_cache(). */ DEFINE_PER_CPU(struct sched_domain __rcu *, sd_llc); DEFINE_PER_CPU(int, sd_llc_size); DEFINE_PER_CPU(int, sd_llc_id); +DEFINE_PER_CPU(int, sd_share_id); DEFINE_PER_CPU(struct sched_domain_shared __rcu *, sd_llc_shared); DEFINE_PER_CPU(struct sched_domain __rcu *, sd_numa); DEFINE_PER_CPU(struct sched_domain __rcu *, sd_asym_packing); DEFINE_PER_CPU(struct sched_domain __rcu *, sd_asym_cpucapacity); + DEFINE_STATIC_KEY_FALSE(sched_asym_cpucapacity); +DEFINE_STATIC_KEY_FALSE(sched_cluster_active); static void update_top_cache_domain(int cpu) { @@ -657,7 +691,7 @@ static void update_top_cache_domain(int cpu) int id = cpu; int size = 1; - sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); + sd = highest_flag_domain(cpu, SD_SHARE_LLC); if (sd) { id = cpumask_first(sched_domain_span(sd)); size = cpumask_weight(sched_domain_span(sd)); @@ -669,13 +703,24 @@ static void update_top_cache_domain(int cpu) per_cpu(sd_llc_id, cpu) = id; rcu_assign_pointer(per_cpu(sd_llc_shared, cpu), sds); + sd = lowest_flag_domain(cpu, SD_CLUSTER); + if (sd) + id = cpumask_first(sched_domain_span(sd)); + + /* + * This assignment should be placed after the sd_llc_id as + * we want this id equals to cluster id on cluster machines + * but equals to LLC id on non-Cluster machines. + */ + per_cpu(sd_share_id, cpu) = id; + sd = lowest_flag_domain(cpu, SD_NUMA); rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); sd = highest_flag_domain(cpu, SD_ASYM_PACKING); rcu_assign_pointer(per_cpu(sd_asym_packing, cpu), sd); - sd = lowest_flag_domain(cpu, SD_ASYM_CPUCAPACITY); + sd = lowest_flag_domain(cpu, SD_ASYM_CPUCAPACITY_FULL); rcu_assign_pointer(per_cpu(sd_asym_cpucapacity, cpu), sd); } @@ -688,7 +733,6 @@ cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) { struct rq *rq = cpu_rq(cpu); struct sched_domain *tmp; - int numa_distance = 0; /* Remove the sched domains which do not contribute to scheduling. */ for (tmp = sd; tmp; ) { @@ -698,8 +742,12 @@ cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) if (sd_parent_degenerate(tmp, parent)) { tmp->parent = parent->parent; - if (parent->parent) + + if (parent->parent) { parent->parent->child = tmp; + parent->parent->groups->flags = tmp->flags; + } + /* * Transfer SD_PREFER_SIBLING down in case of a * degenerate parent; the spans match for this @@ -716,41 +764,21 @@ cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) tmp = sd; sd = sd->parent; destroy_sched_domain(tmp); - if (sd) - sd->child = NULL; - } + if (sd) { + struct sched_group *sg = sd->groups; - for (tmp = sd; tmp; tmp = tmp->parent) - numa_distance += !!(tmp->flags & SD_NUMA); + /* + * sched groups hold the flags of the child sched + * domain for convenience. Clear such flags since + * the child is being destroyed. + */ + do { + sg->flags = 0; + } while (sg != sd->groups); - /* - * FIXME: Diameter >=3 is misrepresented. - * - * Smallest diameter=3 topology is: - * - * node 0 1 2 3 - * 0: 10 20 30 40 - * 1: 20 10 20 30 - * 2: 30 20 10 20 - * 3: 40 30 20 10 - * - * 0 --- 1 --- 2 --- 3 - * - * NUMA-3 0-3 N/A N/A 0-3 - * groups: {0-2},{1-3} {1-3},{0-2} - * - * NUMA-2 0-2 0-3 0-3 1-3 - * groups: {0-1},{1-3} {0-2},{2-3} {1-3},{0-1} {2-3},{0-2} - * - * NUMA-1 0-1 0-2 1-3 2-3 - * groups: {0},{1} {1},{2},{0} {2},{3},{1} {3},{2} - * - * NUMA-0 0 1 2 3 - * - * The NUMA-2 groups for nodes 0 and 3 are obviously buggered, as the - * group span isn't a subset of the domain span. - */ - WARN_ONCE(numa_distance > 2, "Shortest NUMA path spans too many nodes\n"); + sd->child = NULL; + } + } sched_domain_debug(sd, cpu); @@ -945,10 +973,12 @@ build_group_from_child_sched_domain(struct sched_domain *sd, int cpu) return NULL; sg_span = sched_group_span(sg); - if (sd->child) + if (sd->child) { cpumask_copy(sg_span, sched_domain_span(sd->child)); - else + sg->flags = sd->child->flags; + } else { cpumask_copy(sg_span, sched_domain_span(sd)); + } atomic_inc(&sg->ref); return sg; @@ -963,7 +993,7 @@ static void init_overlap_sched_group(struct sched_domain *sd, int cpu; build_balance_mask(sd, sg, mask); - cpu = cpumask_first_and(sched_group_span(sg), mask); + cpu = cpumask_first(mask); sg->sgc = *per_cpu_ptr(sdd->sgc, cpu); if (atomic_inc_return(&sg->sgc->ref) == 1) @@ -982,6 +1012,31 @@ static void init_overlap_sched_group(struct sched_domain *sd, sg->sgc->max_capacity = SCHED_CAPACITY_SCALE; } +static struct sched_domain * +find_descended_sibling(struct sched_domain *sd, struct sched_domain *sibling) +{ + /* + * The proper descendant would be the one whose child won't span out + * of sd + */ + while (sibling->child && + !cpumask_subset(sched_domain_span(sibling->child), + sched_domain_span(sd))) + sibling = sibling->child; + + /* + * As we are referencing sgc across different topology level, we need + * to go down to skip those sched_domains which don't contribute to + * scheduling because they will be degenerated in cpu_attach_domain + */ + while (sibling->child && + cpumask_equal(sched_domain_span(sibling->child), + sched_domain_span(sibling))) + sibling = sibling->child; + + return sibling; +} + static int build_overlap_sched_groups(struct sched_domain *sd, int cpu) { @@ -1015,6 +1070,41 @@ build_overlap_sched_groups(struct sched_domain *sd, int cpu) if (!cpumask_test_cpu(i, sched_domain_span(sibling))) continue; + /* + * Usually we build sched_group by sibling's child sched_domain + * But for machines whose NUMA diameter are 3 or above, we move + * to build sched_group by sibling's proper descendant's child + * domain because sibling's child sched_domain will span out of + * the sched_domain being built as below. + * + * Smallest diameter=3 topology is: + * + * node 0 1 2 3 + * 0: 10 20 30 40 + * 1: 20 10 20 30 + * 2: 30 20 10 20 + * 3: 40 30 20 10 + * + * 0 --- 1 --- 2 --- 3 + * + * NUMA-3 0-3 N/A N/A 0-3 + * groups: {0-2},{1-3} {1-3},{0-2} + * + * NUMA-2 0-2 0-3 0-3 1-3 + * groups: {0-1},{1-3} {0-2},{2-3} {1-3},{0-1} {2-3},{0-2} + * + * NUMA-1 0-1 0-2 1-3 2-3 + * groups: {0},{1} {1},{2},{0} {2},{3},{1} {3},{2} + * + * NUMA-0 0 1 2 3 + * + * The NUMA-2 groups for nodes 0 and 3 are obviously buggered, as the + * group span isn't a subset of the domain span. + */ + if (sibling->child && + !cpumask_subset(sched_domain_span(sibling->child), span)) + sibling = find_descended_sibling(sd, sibling); + sg = build_group_from_child_sched_domain(sibling, cpu); if (!sg) goto fail; @@ -1022,7 +1112,7 @@ build_overlap_sched_groups(struct sched_domain *sd, int cpu) sg_span = sched_group_span(sg); cpumask_or(covered, covered, sg_span); - init_overlap_sched_group(sd, sg); + init_overlap_sched_group(sibling, sg); if (!first) first = sg; @@ -1050,7 +1140,7 @@ fail: * * - Simultaneous multithreading (SMT) * - Multi-Core Cache (MC) - * - Package (DIE) + * - Package (PKG) * * Where the last one more or less denotes everything up to a NUMA node. * @@ -1072,13 +1162,13 @@ fail: * * CPU 0 1 2 3 4 5 6 7 * - * DIE [ ] + * PKG [ ] * MC [ ] [ ] * SMT [ ] [ ] [ ] [ ] * * - or - * - * DIE 0-7 0-7 0-7 0-7 0-7 0-7 0-7 0-7 + * PKG 0-7 0-7 0-7 0-7 0-7 0-7 0-7 0-7 * MC 0-3 0-3 0-3 0-3 4-7 4-7 4-7 4-7 * SMT 0-1 0-1 2-3 2-3 4-5 4-5 6-7 6-7 * @@ -1094,7 +1184,7 @@ fail: * uniquely identify each group (for a given domain): * * - The first is the balance_cpu (see should_we_balance() and the - * load-balance blub in fair.c); for each group we only want 1 CPU to + * load-balance blurb in fair.c); for each group we only want 1 CPU to * continue balancing at a higher domain. * * - The second is the sched_group_capacity; we want all identical groups @@ -1138,6 +1228,7 @@ static struct sched_group *get_group(int cpu, struct sd_data *sdd) if (child) { cpumask_copy(sched_group_span(sg), sched_domain_span(child)); cpumask_copy(group_balance_mask(sg), sched_group_span(sg)); + sg->flags = child->flags; } else { cpumask_set_cpu(cpu, sched_group_span(sg)); cpumask_set_cpu(cpu, group_balance_mask(sg)); @@ -1206,14 +1297,24 @@ build_sched_groups(struct sched_domain *sd, int cpu) static void init_sched_groups_capacity(int cpu, struct sched_domain *sd) { struct sched_group *sg = sd->groups; + struct cpumask *mask = sched_domains_tmpmask2; WARN_ON(!sg); do { - int cpu, max_cpu = -1; + int cpu, cores = 0, max_cpu = -1; sg->group_weight = cpumask_weight(sched_group_span(sg)); + cpumask_copy(mask, sched_group_span(sg)); + for_each_cpu(cpu, mask) { + cores++; +#ifdef CONFIG_SCHED_SMT + cpumask_andnot(mask, mask, cpu_smt_mask(cpu)); +#endif + } + sg->cores = cores; + if (!(sd->flags & SD_ASYM_PACKING)) goto next; @@ -1236,6 +1337,123 @@ next: } /* + * Set of available CPUs grouped by their corresponding capacities + * Each list entry contains a CPU mask reflecting CPUs that share the same + * capacity. + * The lifespan of data is unlimited. + */ +LIST_HEAD(asym_cap_list); + +/* + * Verify whether there is any CPU capacity asymmetry in a given sched domain. + * Provides sd_flags reflecting the asymmetry scope. + */ +static inline int +asym_cpu_capacity_classify(const struct cpumask *sd_span, + const struct cpumask *cpu_map) +{ + struct asym_cap_data *entry; + int count = 0, miss = 0; + + /* + * Count how many unique CPU capacities this domain spans across + * (compare sched_domain CPUs mask with ones representing available + * CPUs capacities). Take into account CPUs that might be offline: + * skip those. + */ + list_for_each_entry(entry, &asym_cap_list, link) { + if (cpumask_intersects(sd_span, cpu_capacity_span(entry))) + ++count; + else if (cpumask_intersects(cpu_map, cpu_capacity_span(entry))) + ++miss; + } + + WARN_ON_ONCE(!count && !list_empty(&asym_cap_list)); + + /* No asymmetry detected */ + if (count < 2) + return 0; + /* Some of the available CPU capacity values have not been detected */ + if (miss) + return SD_ASYM_CPUCAPACITY; + + /* Full asymmetry */ + return SD_ASYM_CPUCAPACITY | SD_ASYM_CPUCAPACITY_FULL; + +} + +static void free_asym_cap_entry(struct rcu_head *head) +{ + struct asym_cap_data *entry = container_of(head, struct asym_cap_data, rcu); + kfree(entry); +} + +static inline void asym_cpu_capacity_update_data(int cpu) +{ + unsigned long capacity = arch_scale_cpu_capacity(cpu); + struct asym_cap_data *insert_entry = NULL; + struct asym_cap_data *entry; + + /* + * Search if capacity already exits. If not, track which the entry + * where we should insert to keep the list ordered descending. + */ + list_for_each_entry(entry, &asym_cap_list, link) { + if (capacity == entry->capacity) + goto done; + else if (!insert_entry && capacity > entry->capacity) + insert_entry = list_prev_entry(entry, link); + } + + entry = kzalloc(sizeof(*entry) + cpumask_size(), GFP_KERNEL); + if (WARN_ONCE(!entry, "Failed to allocate memory for asymmetry data\n")) + return; + entry->capacity = capacity; + + /* If NULL then the new capacity is the smallest, add last. */ + if (!insert_entry) + list_add_tail_rcu(&entry->link, &asym_cap_list); + else + list_add_rcu(&entry->link, &insert_entry->link); +done: + __cpumask_set_cpu(cpu, cpu_capacity_span(entry)); +} + +/* + * Build-up/update list of CPUs grouped by their capacities + * An update requires explicit request to rebuild sched domains + * with state indicating CPU topology changes. + */ +static void asym_cpu_capacity_scan(void) +{ + struct asym_cap_data *entry, *next; + int cpu; + + list_for_each_entry(entry, &asym_cap_list, link) + cpumask_clear(cpu_capacity_span(entry)); + + for_each_cpu_and(cpu, cpu_possible_mask, housekeeping_cpumask(HK_TYPE_DOMAIN)) + asym_cpu_capacity_update_data(cpu); + + list_for_each_entry_safe(entry, next, &asym_cap_list, link) { + if (cpumask_empty(cpu_capacity_span(entry))) { + list_del_rcu(&entry->link); + call_rcu(&entry->rcu, free_asym_cap_entry); + } + } + + /* + * Only one capacity value has been detected i.e. this system is symmetric. + * No need to keep this data around. + */ + if (list_is_singular(&asym_cap_list)) { + entry = list_first_entry(&asym_cap_list, typeof(*entry), link); + list_del_rcu(&entry->link); + call_rcu(&entry->rcu, free_asym_cap_entry); + } +} + +/* * Initializers for schedule domains * Non-inlined to reduce accumulated stack pressure in build_sched_domains() */ @@ -1264,7 +1482,7 @@ static void set_domain_attribute(struct sched_domain *sd, } else request = attr->relax_domain_level; - if (sd->level > request) { + if (sd->level >= request) { /* Turn off idle balance on this domain: */ sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); } @@ -1340,7 +1558,6 @@ static int sched_domains_curr_level; int sched_max_numa_distance; static int *sched_domains_numa_distance; static struct cpumask ***sched_domains_numa_masks; -int __read_mostly node_reclaim_distance = RECLAIM_DISTANCE; #endif /* @@ -1348,11 +1565,12 @@ int __read_mostly node_reclaim_distance = RECLAIM_DISTANCE; * * These flags are purely descriptive of the topology and do not prescribe * behaviour. Behaviour is artificial and mapped in the below sd_init() - * function: + * function. For details, see include/linux/sched/sd_flags.h. * - * SD_SHARE_CPUCAPACITY - describes SMT topologies - * SD_SHARE_PKG_RESOURCES - describes shared caches - * SD_NUMA - describes NUMA topologies + * SD_SHARE_CPUCAPACITY + * SD_SHARE_LLC + * SD_CLUSTER + * SD_NUMA * * Odd one out, which beside describing the topology has a quirk also * prescribes the desired behaviour that goes along with it: @@ -1361,18 +1579,20 @@ int __read_mostly node_reclaim_distance = RECLAIM_DISTANCE; */ #define TOPOLOGY_SD_FLAGS \ (SD_SHARE_CPUCAPACITY | \ - SD_SHARE_PKG_RESOURCES | \ + SD_CLUSTER | \ + SD_SHARE_LLC | \ SD_NUMA | \ SD_ASYM_PACKING) static struct sched_domain * sd_init(struct sched_domain_topology_level *tl, const struct cpumask *cpu_map, - struct sched_domain *child, int dflags, int cpu) + struct sched_domain *child, int cpu) { struct sd_data *sdd = &tl->data; struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); int sd_id, sd_weight, sd_flags = 0; + struct cpumask *sd_span; #ifdef CONFIG_NUMA /* @@ -1389,9 +1609,6 @@ sd_init(struct sched_domain_topology_level *tl, "wrong sd_flags in topology description\n")) sd_flags &= TOPOLOGY_SD_FLAGS; - /* Apply detected topology flags */ - sd_flags |= dflags; - *sd = (struct sched_domain){ .min_interval = sd_weight, .max_interval = 2*sd_weight, @@ -1406,7 +1623,7 @@ sd_init(struct sched_domain_topology_level *tl, | 0*SD_BALANCE_WAKE | 1*SD_WAKE_AFFINE | 0*SD_SHARE_CPUCAPACITY - | 0*SD_SHARE_PKG_RESOURCES + | 0*SD_SHARE_LLC | 0*SD_SERIALIZE | 1*SD_PREFER_SIBLING | 0*SD_NUMA @@ -1416,20 +1633,24 @@ sd_init(struct sched_domain_topology_level *tl, .last_balance = jiffies, .balance_interval = sd_weight, .max_newidle_lb_cost = 0, - .next_decay_max_lb_cost = jiffies, + .last_decay_max_lb_cost = jiffies, .child = child, -#ifdef CONFIG_SCHED_DEBUG .name = tl->name, -#endif }; - cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); - sd_id = cpumask_first(sched_domain_span(sd)); + sd_span = sched_domain_span(sd); + cpumask_and(sd_span, cpu_map, tl->mask(cpu)); + sd_id = cpumask_first(sd_span); + + sd->flags |= asym_cpu_capacity_classify(sd_span, cpu_map); + + WARN_ONCE((sd->flags & (SD_SHARE_CPUCAPACITY | SD_ASYM_CPUCAPACITY)) == + (SD_SHARE_CPUCAPACITY | SD_ASYM_CPUCAPACITY), + "CPU capacity asymmetry not supported on SMT\n"); /* * Convert topological properties into behaviour. */ - /* Don't attempt to spread across CPUs of different capacities. */ if ((sd->flags & SD_ASYM_CPUCAPACITY) && sd->child) sd->child->flags &= ~SD_PREFER_SIBLING; @@ -1437,7 +1658,7 @@ sd_init(struct sched_domain_topology_level *tl, if (sd->flags & SD_SHARE_CPUCAPACITY) { sd->imbalance_pct = 110; - } else if (sd->flags & SD_SHARE_PKG_RESOURCES) { + } else if (sd->flags & SD_SHARE_LLC) { sd->imbalance_pct = 117; sd->cache_nice_tries = 1; @@ -1462,7 +1683,7 @@ sd_init(struct sched_domain_topology_level *tl, * For all levels sharing cache; connect a sched_domain_shared * instance. */ - if (sd->flags & SD_SHARE_PKG_RESOURCES) { + if (sd->flags & SD_SHARE_LLC) { sd->shared = *per_cpu_ptr(sdd->sds, sd_id); atomic_inc(&sd->shared->ref); atomic_set(&sd->shared->nr_busy_cpus, sd_weight); @@ -1480,25 +1701,32 @@ static struct sched_domain_topology_level default_topology[] = { #ifdef CONFIG_SCHED_SMT { cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) }, #endif + +#ifdef CONFIG_SCHED_CLUSTER + { cpu_clustergroup_mask, cpu_cluster_flags, SD_INIT_NAME(CLS) }, +#endif + #ifdef CONFIG_SCHED_MC { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) }, #endif - { cpu_cpu_mask, SD_INIT_NAME(DIE) }, + { cpu_cpu_mask, SD_INIT_NAME(PKG) }, { NULL, }, }; static struct sched_domain_topology_level *sched_domain_topology = default_topology; +static struct sched_domain_topology_level *sched_domain_topology_saved; #define for_each_sd_topology(tl) \ for (tl = sched_domain_topology; tl->mask; tl++) -void set_sched_topology(struct sched_domain_topology_level *tl) +void __init set_sched_topology(struct sched_domain_topology_level *tl) { if (WARN_ON_ONCE(sched_smp_initialized)) return; sched_domain_topology = tl; + sched_domain_topology_saved = NULL; } #ifdef CONFIG_NUMA @@ -1522,8 +1750,12 @@ static void sched_numa_warn(const char *str) for (i = 0; i < nr_node_ids; i++) { printk(KERN_WARNING " "); - for (j = 0; j < nr_node_ids; j++) - printk(KERN_CONT "%02d ", node_distance(i,j)); + for (j = 0; j < nr_node_ids; j++) { + if (!node_state(i, N_CPU) || !node_state(j, N_CPU)) + printk(KERN_CONT "(%02d) ", node_distance(i,j)); + else + printk(KERN_CONT " %02d ", node_distance(i,j)); + } printk(KERN_CONT "\n"); } printk(KERN_WARNING "\n"); @@ -1531,19 +1763,34 @@ static void sched_numa_warn(const char *str) bool find_numa_distance(int distance) { - int i; + bool found = false; + int i, *distances; if (distance == node_distance(0, 0)) return true; + rcu_read_lock(); + distances = rcu_dereference(sched_domains_numa_distance); + if (!distances) + goto unlock; for (i = 0; i < sched_domains_numa_levels; i++) { - if (sched_domains_numa_distance[i] == distance) - return true; + if (distances[i] == distance) { + found = true; + break; + } } +unlock: + rcu_read_unlock(); - return false; + return found; } +#define for_each_cpu_node_but(n, nbut) \ + for_each_node_state(n, N_CPU) \ + if (n == nbut) \ + continue; \ + else + /* * A system can have three types of NUMA topology: * NUMA_DIRECT: all nodes are directly connected, or not a NUMA system @@ -1563,7 +1810,7 @@ bool find_numa_distance(int distance) * there is an intermediary node C, which is < N hops away from both * nodes A and B, the system is a glueless mesh. */ -static void init_numa_topology_type(void) +static void init_numa_topology_type(int offline_node) { int a, b, c, n; @@ -1574,14 +1821,14 @@ static void init_numa_topology_type(void) return; } - for_each_online_node(a) { - for_each_online_node(b) { + for_each_cpu_node_but(a, offline_node) { + for_each_cpu_node_but(b, offline_node) { /* Find two nodes furthest removed from each other. */ if (node_distance(a, b) < n) continue; /* Is there an intermediary node between a and b? */ - for_each_online_node(c) { + for_each_cpu_node_but(c, offline_node) { if (node_distance(a, c) < n && node_distance(b, c) < n) { sched_numa_topology_type = @@ -1594,68 +1841,67 @@ static void init_numa_topology_type(void) return; } } + + pr_err("Failed to find a NUMA topology type, defaulting to DIRECT\n"); + sched_numa_topology_type = NUMA_DIRECT; } -void sched_init_numa(void) -{ - int next_distance, curr_distance = node_distance(0, 0); - struct sched_domain_topology_level *tl; - int level = 0; - int i, j, k; - sched_domains_numa_distance = kzalloc(sizeof(int) * (nr_node_ids + 1), GFP_KERNEL); - if (!sched_domains_numa_distance) - return; +#define NR_DISTANCE_VALUES (1 << DISTANCE_BITS) - /* Includes NUMA identity node at level 0. */ - sched_domains_numa_distance[level++] = curr_distance; - sched_domains_numa_levels = level; +void sched_init_numa(int offline_node) +{ + struct sched_domain_topology_level *tl; + unsigned long *distance_map; + int nr_levels = 0; + int i, j; + int *distances; + struct cpumask ***masks; /* - * O(nr_nodes^2) deduplicating selection sort -- in order to find the + * O(nr_nodes^2) de-duplicating selection sort -- in order to find the * unique distances in the node_distance() table. - * - * Assumes node_distance(0,j) includes all distances in - * node_distance(i,j) in order to avoid cubic time. */ - next_distance = curr_distance; - for (i = 0; i < nr_node_ids; i++) { - for (j = 0; j < nr_node_ids; j++) { - for (k = 0; k < nr_node_ids; k++) { - int distance = node_distance(i, k); - - if (distance > curr_distance && - (distance < next_distance || - next_distance == curr_distance)) - next_distance = distance; + distance_map = bitmap_alloc(NR_DISTANCE_VALUES, GFP_KERNEL); + if (!distance_map) + return; - /* - * While not a strong assumption it would be nice to know - * about cases where if node A is connected to B, B is not - * equally connected to A. - */ - if (sched_debug() && node_distance(k, i) != distance) - sched_numa_warn("Node-distance not symmetric"); + bitmap_zero(distance_map, NR_DISTANCE_VALUES); + for_each_cpu_node_but(i, offline_node) { + for_each_cpu_node_but(j, offline_node) { + int distance = node_distance(i, j); - if (sched_debug() && i && !find_numa_distance(distance)) - sched_numa_warn("Node-0 not representative"); + if (distance < LOCAL_DISTANCE || distance >= NR_DISTANCE_VALUES) { + sched_numa_warn("Invalid distance value range"); + bitmap_free(distance_map); + return; } - if (next_distance != curr_distance) { - sched_domains_numa_distance[level++] = next_distance; - sched_domains_numa_levels = level; - curr_distance = next_distance; - } else break; + + bitmap_set(distance_map, distance, 1); } + } + /* + * We can now figure out how many unique distance values there are and + * allocate memory accordingly. + */ + nr_levels = bitmap_weight(distance_map, NR_DISTANCE_VALUES); - /* - * In case of sched_debug() we verify the above assumption. - */ - if (!sched_debug()) - break; + distances = kcalloc(nr_levels, sizeof(int), GFP_KERNEL); + if (!distances) { + bitmap_free(distance_map); + return; } + for (i = 0, j = 0; i < nr_levels; i++, j++) { + j = find_next_bit(distance_map, NR_DISTANCE_VALUES, j); + distances[i] = j; + } + rcu_assign_pointer(sched_domains_numa_distance, distances); + + bitmap_free(distance_map); + /* - * 'level' contains the number of unique distances + * 'nr_levels' contains the number of unique distances * * The sched_domains_numa_distance[] array includes the actual distance * numbers. @@ -1664,36 +1910,40 @@ void sched_init_numa(void) /* * Here, we should temporarily reset sched_domains_numa_levels to 0. * If it fails to allocate memory for array sched_domains_numa_masks[][], - * the array will contain less then 'level' members. This could be + * the array will contain less then 'nr_levels' members. This could be * dangerous when we use it to iterate array sched_domains_numa_masks[][] * in other functions. * - * We reset it to 'level' at the end of this function. + * We reset it to 'nr_levels' at the end of this function. */ sched_domains_numa_levels = 0; - sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); - if (!sched_domains_numa_masks) + masks = kzalloc(sizeof(void *) * nr_levels, GFP_KERNEL); + if (!masks) return; /* * Now for each level, construct a mask per node which contains all * CPUs of nodes that are that many hops away from us. */ - for (i = 0; i < level; i++) { - sched_domains_numa_masks[i] = - kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); - if (!sched_domains_numa_masks[i]) + for (i = 0; i < nr_levels; i++) { + masks[i] = kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); + if (!masks[i]) return; - for (j = 0; j < nr_node_ids; j++) { + for_each_cpu_node_but(j, offline_node) { struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); + int k; + if (!mask) return; - sched_domains_numa_masks[i][j] = mask; + masks[i][j] = mask; + + for_each_cpu_node_but(k, offline_node) { + if (sched_debug() && (node_distance(j, k) != node_distance(k, j))) + sched_numa_warn("Node-distance not symmetric"); - for_each_node(k) { if (node_distance(j, k) > sched_domains_numa_distance[i]) continue; @@ -1701,11 +1951,12 @@ void sched_init_numa(void) } } } + rcu_assign_pointer(sched_domains_numa_masks, masks); /* Compute default topology size */ for (i = 0; sched_domain_topology[i].mask; i++); - tl = kzalloc((i + level + 1) * + tl = kzalloc((i + nr_levels + 1) * sizeof(struct sched_domain_topology_level), GFP_KERNEL); if (!tl) return; @@ -1728,7 +1979,7 @@ void sched_init_numa(void) /* * .. and append 'j' levels of NUMA goodness. */ - for (j = 1; j < level; i++, j++) { + for (j = 1; j < nr_levels; i++, j++) { tl[i] = (struct sched_domain_topology_level){ .mask = sd_numa_mask, .sd_flags = cpu_numa_flags, @@ -1738,12 +1989,67 @@ void sched_init_numa(void) }; } + sched_domain_topology_saved = sched_domain_topology; sched_domain_topology = tl; - sched_domains_numa_levels = level; - sched_max_numa_distance = sched_domains_numa_distance[level - 1]; + sched_domains_numa_levels = nr_levels; + WRITE_ONCE(sched_max_numa_distance, sched_domains_numa_distance[nr_levels - 1]); + + init_numa_topology_type(offline_node); +} + + +static void sched_reset_numa(void) +{ + int nr_levels, *distances; + struct cpumask ***masks; + + nr_levels = sched_domains_numa_levels; + sched_domains_numa_levels = 0; + sched_max_numa_distance = 0; + sched_numa_topology_type = NUMA_DIRECT; + distances = sched_domains_numa_distance; + rcu_assign_pointer(sched_domains_numa_distance, NULL); + masks = sched_domains_numa_masks; + rcu_assign_pointer(sched_domains_numa_masks, NULL); + if (distances || masks) { + int i, j; + + synchronize_rcu(); + kfree(distances); + for (i = 0; i < nr_levels && masks; i++) { + if (!masks[i]) + continue; + for_each_node(j) + kfree(masks[i][j]); + kfree(masks[i]); + } + kfree(masks); + } + if (sched_domain_topology_saved) { + kfree(sched_domain_topology); + sched_domain_topology = sched_domain_topology_saved; + sched_domain_topology_saved = NULL; + } +} + +/* + * Call with hotplug lock held + */ +void sched_update_numa(int cpu, bool online) +{ + int node; + + node = cpu_to_node(cpu); + /* + * Scheduler NUMA topology is updated when the first CPU of a + * node is onlined or the last CPU of a node is offlined. + */ + if (cpumask_weight(cpumask_of_node(node)) != 1) + return; - init_numa_topology_type(); + sched_reset_numa(); + sched_init_numa(online ? NUMA_NO_NODE : node); } void sched_domains_numa_masks_set(unsigned int cpu) @@ -1753,6 +2059,10 @@ void sched_domains_numa_masks_set(unsigned int cpu) for (i = 0; i < sched_domains_numa_levels; i++) { for (j = 0; j < nr_node_ids; j++) { + if (!node_state(j, N_CPU)) + continue; + + /* Set ourselves in the remote node's masks */ if (node_distance(j, node) <= sched_domains_numa_distance[i]) cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); } @@ -1764,8 +2074,10 @@ void sched_domains_numa_masks_clear(unsigned int cpu) int i, j; for (i = 0; i < sched_domains_numa_levels; i++) { - for (j = 0; j < nr_node_ids; j++) - cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); + for (j = 0; j < nr_node_ids; j++) { + if (sched_domains_numa_masks[i][j]) + cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); + } } } @@ -1779,16 +2091,130 @@ void sched_domains_numa_masks_clear(unsigned int cpu) */ int sched_numa_find_closest(const struct cpumask *cpus, int cpu) { - int i, j = cpu_to_node(cpu); + int i, j = cpu_to_node(cpu), found = nr_cpu_ids; + struct cpumask ***masks; + rcu_read_lock(); + masks = rcu_dereference(sched_domains_numa_masks); + if (!masks) + goto unlock; for (i = 0; i < sched_domains_numa_levels; i++) { - cpu = cpumask_any_and(cpus, sched_domains_numa_masks[i][j]); - if (cpu < nr_cpu_ids) - return cpu; + if (!masks[i][j]) + break; + cpu = cpumask_any_and(cpus, masks[i][j]); + if (cpu < nr_cpu_ids) { + found = cpu; + break; + } } - return nr_cpu_ids; +unlock: + rcu_read_unlock(); + + return found; } +struct __cmp_key { + const struct cpumask *cpus; + struct cpumask ***masks; + int node; + int cpu; + int w; +}; + +static int hop_cmp(const void *a, const void *b) +{ + struct cpumask **prev_hop, **cur_hop = *(struct cpumask ***)b; + struct __cmp_key *k = (struct __cmp_key *)a; + + if (cpumask_weight_and(k->cpus, cur_hop[k->node]) <= k->cpu) + return 1; + + if (b == k->masks) { + k->w = 0; + return 0; + } + + prev_hop = *((struct cpumask ***)b - 1); + k->w = cpumask_weight_and(k->cpus, prev_hop[k->node]); + if (k->w <= k->cpu) + return 0; + + return -1; +} + +/** + * sched_numa_find_nth_cpu() - given the NUMA topology, find the Nth closest CPU + * from @cpus to @cpu, taking into account distance + * from a given @node. + * @cpus: cpumask to find a cpu from + * @cpu: CPU to start searching + * @node: NUMA node to order CPUs by distance + * + * Return: cpu, or nr_cpu_ids when nothing found. + */ +int sched_numa_find_nth_cpu(const struct cpumask *cpus, int cpu, int node) +{ + struct __cmp_key k = { .cpus = cpus, .cpu = cpu }; + struct cpumask ***hop_masks; + int hop, ret = nr_cpu_ids; + + if (node == NUMA_NO_NODE) + return cpumask_nth_and(cpu, cpus, cpu_online_mask); + + rcu_read_lock(); + + /* CPU-less node entries are uninitialized in sched_domains_numa_masks */ + node = numa_nearest_node(node, N_CPU); + k.node = node; + + k.masks = rcu_dereference(sched_domains_numa_masks); + if (!k.masks) + goto unlock; + + hop_masks = bsearch(&k, k.masks, sched_domains_numa_levels, sizeof(k.masks[0]), hop_cmp); + hop = hop_masks - k.masks; + + ret = hop ? + cpumask_nth_and_andnot(cpu - k.w, cpus, k.masks[hop][node], k.masks[hop-1][node]) : + cpumask_nth_and(cpu, cpus, k.masks[0][node]); +unlock: + rcu_read_unlock(); + return ret; +} +EXPORT_SYMBOL_GPL(sched_numa_find_nth_cpu); + +/** + * sched_numa_hop_mask() - Get the cpumask of CPUs at most @hops hops away from + * @node + * @node: The node to count hops from. + * @hops: Include CPUs up to that many hops away. 0 means local node. + * + * Return: On success, a pointer to a cpumask of CPUs at most @hops away from + * @node, an error value otherwise. + * + * Requires rcu_lock to be held. Returned cpumask is only valid within that + * read-side section, copy it if required beyond that. + * + * Note that not all hops are equal in distance; see sched_init_numa() for how + * distances and masks are handled. + * Also note that this is a reflection of sched_domains_numa_masks, which may change + * during the lifetime of the system (offline nodes are taken out of the masks). + */ +const struct cpumask *sched_numa_hop_mask(unsigned int node, unsigned int hops) +{ + struct cpumask ***masks; + + if (node >= nr_node_ids || hops >= sched_domains_numa_levels) + return ERR_PTR(-EINVAL); + + masks = rcu_dereference(sched_domains_numa_masks); + if (!masks) + return ERR_PTR(-EBUSY); + + return masks[hops][node]; +} +EXPORT_SYMBOL_GPL(sched_numa_hop_mask); + #endif /* CONFIG_NUMA */ static int __sdt_alloc(const struct cpumask *cpu_map) @@ -1898,9 +2324,9 @@ static void __sdt_free(const struct cpumask *cpu_map) static struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, const struct cpumask *cpu_map, struct sched_domain_attr *attr, - struct sched_domain *child, int dflags, int cpu) + struct sched_domain *child, int cpu) { - struct sched_domain *sd = sd_init(tl, cpu_map, child, dflags, cpu); + struct sched_domain *sd = sd_init(tl, cpu_map, child, cpu); if (child) { sd->level = child->level + 1; @@ -1910,10 +2336,8 @@ static struct sched_domain *build_sched_domain(struct sched_domain_topology_leve if (!cpumask_subset(sched_domain_span(child), sched_domain_span(sd))) { pr_err("BUG: arch topology borken\n"); -#ifdef CONFIG_SCHED_DEBUG pr_err(" the %s domain not a subset of the %s domain\n", child->name, sd->name); -#endif /* Fixup, ensure @sd has at least @child CPUs. */ cpumask_or(sched_domain_span(sd), sched_domain_span(sd), @@ -1933,7 +2357,7 @@ static struct sched_domain *build_sched_domain(struct sched_domain_topology_leve static bool topology_span_sane(struct sched_domain_topology_level *tl, const struct cpumask *cpu_map, int cpu) { - int i; + int i = cpu + 1; /* NUMA levels are allowed to overlap */ if (tl->flags & SDTL_OVERLAP) @@ -1945,9 +2369,7 @@ static bool topology_span_sane(struct sched_domain_topology_level *tl, * breaking the sched_group lists - i.e. a later get_group() pass * breaks the linking done for an earlier span. */ - for_each_cpu(i, cpu_map) { - if (i == cpu) - continue; + for_each_cpu_from(i, cpu_map) { /* * We should 'and' all those masks with 'cpu_map' to exactly * match the topology we're about to build, but that can only @@ -1963,65 +2385,6 @@ static bool topology_span_sane(struct sched_domain_topology_level *tl, } /* - * Find the sched_domain_topology_level where all CPU capacities are visible - * for all CPUs. - */ -static struct sched_domain_topology_level -*asym_cpu_capacity_level(const struct cpumask *cpu_map) -{ - int i, j, asym_level = 0; - bool asym = false; - struct sched_domain_topology_level *tl, *asym_tl = NULL; - unsigned long cap; - - /* Is there any asymmetry? */ - cap = arch_scale_cpu_capacity(cpumask_first(cpu_map)); - - for_each_cpu(i, cpu_map) { - if (arch_scale_cpu_capacity(i) != cap) { - asym = true; - break; - } - } - - if (!asym) - return NULL; - - /* - * Examine topology from all CPU's point of views to detect the lowest - * sched_domain_topology_level where a highest capacity CPU is visible - * to everyone. - */ - for_each_cpu(i, cpu_map) { - unsigned long max_capacity = arch_scale_cpu_capacity(i); - int tl_id = 0; - - for_each_sd_topology(tl) { - if (tl_id < asym_level) - goto next_level; - - for_each_cpu_and(j, tl->mask(i), cpu_map) { - unsigned long capacity; - - capacity = arch_scale_cpu_capacity(j); - - if (capacity <= max_capacity) - continue; - - max_capacity = capacity; - asym_level = tl_id; - asym_tl = tl; - } -next_level: - tl_id++; - } - } - - return asym_tl; -} - - -/* * Build sched domains for a given set of CPUs and attach the sched domains * to the individual CPUs */ @@ -2033,8 +2396,8 @@ build_sched_domains(const struct cpumask *cpu_map, struct sched_domain_attr *att struct s_data d; struct rq *rq = NULL; int i, ret = -ENOMEM; - struct sched_domain_topology_level *tl_asym; bool has_asym = false; + bool has_cluster = false; if (WARN_ON(cpumask_empty(cpu_map))) goto error; @@ -2043,24 +2406,19 @@ build_sched_domains(const struct cpumask *cpu_map, struct sched_domain_attr *att if (alloc_state != sa_rootdomain) goto error; - tl_asym = asym_cpu_capacity_level(cpu_map); - /* Set up domains for CPUs specified by the cpu_map: */ for_each_cpu(i, cpu_map) { struct sched_domain_topology_level *tl; - int dflags = 0; sd = NULL; for_each_sd_topology(tl) { - if (tl == tl_asym) { - dflags |= SD_ASYM_CPUCAPACITY; - has_asym = true; - } if (WARN_ON(!topology_span_sane(tl, cpu_map, i))) goto error; - sd = build_sched_domain(tl, cpu_map, attr, sd, dflags, i); + sd = build_sched_domain(tl, cpu_map, attr, sd, i); + + has_asym |= sd->flags & SD_ASYM_CPUCAPACITY; if (tl == sched_domain_topology) *per_cpu_ptr(d.sd, i) = sd; @@ -2085,6 +2443,64 @@ build_sched_domains(const struct cpumask *cpu_map, struct sched_domain_attr *att } } + /* + * Calculate an allowed NUMA imbalance such that LLCs do not get + * imbalanced. + */ + for_each_cpu(i, cpu_map) { + unsigned int imb = 0; + unsigned int imb_span = 1; + + for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { + struct sched_domain *child = sd->child; + + if (!(sd->flags & SD_SHARE_LLC) && child && + (child->flags & SD_SHARE_LLC)) { + struct sched_domain __rcu *top_p; + unsigned int nr_llcs; + + /* + * For a single LLC per node, allow an + * imbalance up to 12.5% of the node. This is + * arbitrary cutoff based two factors -- SMT and + * memory channels. For SMT-2, the intent is to + * avoid premature sharing of HT resources but + * SMT-4 or SMT-8 *may* benefit from a different + * cutoff. For memory channels, this is a very + * rough estimate of how many channels may be + * active and is based on recent CPUs with + * many cores. + * + * For multiple LLCs, allow an imbalance + * until multiple tasks would share an LLC + * on one node while LLCs on another node + * remain idle. This assumes that there are + * enough logical CPUs per LLC to avoid SMT + * factors and that there is a correlation + * between LLCs and memory channels. + */ + nr_llcs = sd->span_weight / child->span_weight; + if (nr_llcs == 1) + imb = sd->span_weight >> 3; + else + imb = nr_llcs; + imb = max(1U, imb); + sd->imb_numa_nr = imb; + + /* Set span based on the first NUMA domain. */ + top_p = sd->parent; + while (top_p && !(top_p->flags & SD_NUMA)) { + top_p = top_p->parent; + } + imb_span = top_p ? top_p->span_weight : sd->span_weight; + } else { + int factor = max(1U, (sd->span_weight / imb_span)); + + sd->imb_numa_nr = imb * factor; + } + } + } + /* Calculate CPU capacity for physical packages and nodes */ for (i = nr_cpumask_bits-1; i >= 0; i--) { if (!cpumask_test_cpu(i, cpu_map)) @@ -2102,21 +2518,21 @@ build_sched_domains(const struct cpumask *cpu_map, struct sched_domain_attr *att rq = cpu_rq(i); sd = *per_cpu_ptr(d.sd, i); - /* Use READ_ONCE()/WRITE_ONCE() to avoid load/store tearing: */ - if (rq->cpu_capacity_orig > READ_ONCE(d.rd->max_cpu_capacity)) - WRITE_ONCE(d.rd->max_cpu_capacity, rq->cpu_capacity_orig); - cpu_attach_domain(sd, d.rd, i); + + if (lowest_flag_domain(i, SD_CLUSTER)) + has_cluster = true; } rcu_read_unlock(); if (has_asym) static_branch_inc_cpuslocked(&sched_asym_cpucapacity); - if (rq && sched_debug_enabled) { - pr_info("root domain span: %*pbl (max cpu_capacity = %lu)\n", - cpumask_pr_args(cpu_map), rq->rd->max_cpu_capacity); - } + if (has_cluster) + static_branch_inc_cpuslocked(&sched_cluster_active); + + if (rq && sched_debug_verbose) + pr_info("root domain span: %*pbl\n", cpumask_pr_args(cpu_map)); ret = 0; error: @@ -2131,7 +2547,7 @@ static cpumask_var_t *doms_cur; /* Number of sched domains in 'doms_cur': */ static int ndoms_cur; -/* Attribues of custom domains in 'doms_cur' */ +/* Attributes of custom domains in 'doms_cur' */ static struct sched_domain_attr *dattr_cur; /* @@ -2180,7 +2596,7 @@ void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) * Set up scheduler domains and groups. For now this just excludes isolated * CPUs, but could be used to exclude other special cases in the future. */ -int sched_init_domains(const struct cpumask *cpu_map) +int __init sched_init_domains(const struct cpumask *cpu_map) { int err; @@ -2189,13 +2605,13 @@ int sched_init_domains(const struct cpumask *cpu_map) zalloc_cpumask_var(&fallback_doms, GFP_KERNEL); arch_update_cpu_topology(); + asym_cpu_capacity_scan(); ndoms_cur = 1; doms_cur = alloc_sched_domains(ndoms_cur); if (!doms_cur) doms_cur = &fallback_doms; - cpumask_and(doms_cur[0], cpu_map, housekeeping_cpumask(HK_FLAG_DOMAIN)); + cpumask_and(doms_cur[0], cpu_map, housekeeping_cpumask(HK_TYPE_DOMAIN)); err = build_sched_domains(doms_cur[0], NULL); - register_sched_domain_sysctl(); return err; } @@ -2212,6 +2628,9 @@ static void detach_destroy_domains(const struct cpumask *cpu_map) if (rcu_access_pointer(per_cpu(sd_asym_cpucapacity, cpu))) static_branch_dec_cpuslocked(&sched_asym_cpucapacity); + if (static_branch_unlikely(&sched_cluster_active)) + static_branch_dec_cpuslocked(&sched_cluster_active); + rcu_read_lock(); for_each_cpu(i, cpu_map) cpu_attach_domain(NULL, &def_root_domain, i); @@ -2270,11 +2689,11 @@ void partition_sched_domains_locked(int ndoms_new, cpumask_var_t doms_new[], lockdep_assert_held(&sched_domains_mutex); - /* Always unregister in case we don't destroy any domains: */ - unregister_sched_domain_sysctl(); - /* Let the architecture update CPU core mappings: */ new_topology = arch_update_cpu_topology(); + /* Trigger rebuilding CPU capacity asymmetry data */ + if (new_topology) + asym_cpu_capacity_scan(); if (!doms_new) { WARN_ON_ONCE(dattr_new); @@ -2283,7 +2702,7 @@ void partition_sched_domains_locked(int ndoms_new, cpumask_var_t doms_new[], if (doms_new) { n = 1; cpumask_and(doms_new[0], cpu_active_mask, - housekeeping_cpumask(HK_FLAG_DOMAIN)); + housekeeping_cpumask(HK_TYPE_DOMAIN)); } } else { n = ndoms_new; @@ -2298,9 +2717,11 @@ void partition_sched_domains_locked(int ndoms_new, cpumask_var_t doms_new[], /* * This domain won't be destroyed and as such - * its dl_bw->total_bw needs to be cleared. It - * will be recomputed in function - * update_tasks_root_domain(). + * its dl_bw->total_bw needs to be cleared. + * Tasks contribution will be then recomputed + * in function dl_update_tasks_root_domain(), + * dl_servers contribution in function + * dl_restore_server_root_domain(). */ rd = cpu_rq(cpumask_any(doms_cur[i]))->rd; dl_clear_root_domain(rd); @@ -2318,7 +2739,7 @@ match1: n = 0; doms_new = &fallback_doms; cpumask_and(doms_new[0], cpu_active_mask, - housekeeping_cpumask(HK_FLAG_DOMAIN)); + housekeeping_cpumask(HK_TYPE_DOMAIN)); } /* Build new domains: */ @@ -2335,7 +2756,7 @@ match2: } #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL) - /* Build perf. domains: */ + /* Build perf domains: */ for (i = 0; i < ndoms_new; i++) { for (j = 0; j < n && !sched_energy_update; j++) { if (cpumask_equal(doms_new[i], doms_cur[j]) && @@ -2344,7 +2765,7 @@ match2: goto match3; } } - /* No match - add perf. domains for a new rd */ + /* No match - add perf domains for a new rd */ has_eas |= build_perf_domains(doms_new[i]); match3: ; @@ -2361,7 +2782,7 @@ match3: dattr_cur = dattr_new; ndoms_cur = ndoms_new; - register_sched_domain_sysctl(); + update_sched_domain_debugfs(); } /* |