Merge branches 'sched/domains' and 'sched/clock' into sched/core
authorIngo Molnar <mingo@elte.hu>
Fri, 4 Sep 2009 08:08:43 +0000 (10:08 +0200)
committerIngo Molnar <mingo@elte.hu>
Fri, 4 Sep 2009 08:08:47 +0000 (10:08 +0200)
Merge reason: both topics are ready now, and we want to merge dependent
              changes.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
1  2  3 
kernel/sched.c

diff --cc kernel/sched.c
@@@@ -8478,240 -8476,249 -8378,240 +8576,249 @@@@ static void set_domain_attribute(struc
        }
   }
   
- -/*
- - * Build sched domains for a given set of cpus and attach the sched domains
- - * to the individual cpus
- - */
- -static int __build_sched_domains(const struct cpumask *cpu_map,
- -                              struct sched_domain_attr *attr)
- -{
- -     int i, err = -ENOMEM;
- -     struct root_domain *rd;
- -     cpumask_var_t nodemask, this_sibling_map, this_core_map, send_covered,
- -             tmpmask;
+ +static void __free_domain_allocs(struct s_data *d, enum s_alloc what,
+ +                              const struct cpumask *cpu_map)
+ +{
+ +     switch (what) {
+ +     case sa_sched_groups:
+ +             free_sched_groups(cpu_map, d->tmpmask); /* fall through */
+ +             d->sched_group_nodes = NULL;
+ +     case sa_rootdomain:
+ +             free_rootdomain(d->rd); /* fall through */
+ +     case sa_tmpmask:
+ +             free_cpumask_var(d->tmpmask); /* fall through */
+ +     case sa_send_covered:
+ +             free_cpumask_var(d->send_covered); /* fall through */
+ +     case sa_this_core_map:
+ +             free_cpumask_var(d->this_core_map); /* fall through */
+ +     case sa_this_sibling_map:
+ +             free_cpumask_var(d->this_sibling_map); /* fall through */
+ +     case sa_nodemask:
+ +             free_cpumask_var(d->nodemask); /* fall through */
+ +     case sa_sched_group_nodes:
   #ifdef CONFIG_NUMA
- -     cpumask_var_t domainspan, covered, notcovered;
- -     struct sched_group **sched_group_nodes = NULL;
- -     int sd_allnodes = 0;
- -
- -     if (!alloc_cpumask_var(&domainspan, GFP_KERNEL))
- -             goto out;
- -     if (!alloc_cpumask_var(&covered, GFP_KERNEL))
- -             goto free_domainspan;
- -     if (!alloc_cpumask_var(&notcovered, GFP_KERNEL))
- -             goto free_covered;
  -#endif
  -
  -     if (!alloc_cpumask_var(&nodemask, GFP_KERNEL))
  -             goto free_notcovered;
  -     if (!alloc_cpumask_var(&this_sibling_map, GFP_KERNEL))
  -             goto free_nodemask;
  -     if (!alloc_cpumask_var(&this_core_map, GFP_KERNEL))
  -             goto free_this_sibling_map;
  -     if (!alloc_cpumask_var(&send_covered, GFP_KERNEL))
  -             goto free_this_core_map;
  -     if (!alloc_cpumask_var(&tmpmask, GFP_KERNEL))
  -             goto free_send_covered;
+ +             kfree(d->sched_group_nodes); /* fall through */
+ +     case sa_notcovered:
+ +             free_cpumask_var(d->notcovered); /* fall through */
+ +     case sa_covered:
+ +             free_cpumask_var(d->covered); /* fall through */
+ +     case sa_domainspan:
+ +             free_cpumask_var(d->domainspan); /* fall through */
  +#endif
+ +     case sa_none:
+ +             break;
+ +     }
+ +}
   
-       if (!alloc_cpumask_var(&nodemask, GFP_KERNEL))
-               goto free_notcovered;
-       if (!alloc_cpumask_var(&this_sibling_map, GFP_KERNEL))
-               goto free_nodemask;
-       if (!alloc_cpumask_var(&this_core_map, GFP_KERNEL))
-               goto free_this_sibling_map;
-       if (!alloc_cpumask_var(&send_covered, GFP_KERNEL))
-               goto free_this_core_map;
-       if (!alloc_cpumask_var(&tmpmask, GFP_KERNEL))
-               goto free_send_covered;
-  
+ +static enum s_alloc __visit_domain_allocation_hell(struct s_data *d,
+ +                                                const struct cpumask *cpu_map)
+ +{
   #ifdef CONFIG_NUMA
- -     /*
- -      * Allocate the per-node list of sched groups
- -      */
- -     sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *),
- -                                 GFP_KERNEL);
- -     if (!sched_group_nodes) {
+ +     if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL))
+ +             return sa_none;
+ +     if (!alloc_cpumask_var(&d->covered, GFP_KERNEL))
+ +             return sa_domainspan;
+ +     if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL))
+ +             return sa_covered;
+ +     /* Allocate the per-node list of sched groups */
+ +     d->sched_group_nodes = kcalloc(nr_node_ids,
+ +                                   sizeof(struct sched_group *), GFP_KERNEL);
+ +     if (!d->sched_group_nodes) {
                printk(KERN_WARNING "Can not alloc sched group node list\n");
- -             goto free_tmpmask;
  -     }
  -#endif
  -
  -     rd = alloc_rootdomain();
  -     if (!rd) {
+ +             return sa_notcovered;
  +     }
+ +     sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes;
  +#endif
-  
-       rd = alloc_rootdomain();
-       if (!rd) {
+ +     if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL))
+ +             return sa_sched_group_nodes;
+ +     if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL))
+ +             return sa_nodemask;
+ +     if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL))
+ +             return sa_this_sibling_map;
+ +     if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL))
+ +             return sa_this_core_map;
+ +     if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL))
+ +             return sa_send_covered;
+ +     d->rd = alloc_rootdomain();
+ +     if (!d->rd) {
                printk(KERN_WARNING "Cannot alloc root domain\n");
- -             goto free_sched_groups;
+ +             return sa_tmpmask;
        }
+ +     return sa_rootdomain;
+ +}
   
+ +static struct sched_domain *__build_numa_sched_domains(struct s_data *d,
+ +     const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i)
+ +{
+ +     struct sched_domain *sd = NULL;
   #ifdef CONFIG_NUMA
- -     sched_group_nodes_bycpu[cpumask_first(cpu_map)] = sched_group_nodes;
- -#endif
- -
- -     /*
- -      * Set up domains for cpus specified by the cpu_map.
- -      */
- -     for_each_cpu(i, cpu_map) {
- -             struct sched_domain *sd = NULL, *p;
- -
- -             cpumask_and(nodemask, cpumask_of_node(cpu_to_node(i)), cpu_map);
- -
- -#ifdef CONFIG_NUMA
- -             if (cpumask_weight(cpu_map) >
- -                             SD_NODES_PER_DOMAIN*cpumask_weight(nodemask)) {
- -                     sd = &per_cpu(allnodes_domains, i).sd;
- -                     SD_INIT(sd, ALLNODES);
- -                     set_domain_attribute(sd, attr);
- -                     cpumask_copy(sched_domain_span(sd), cpu_map);
- -                     cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask);
- -                     p = sd;
- -                     sd_allnodes = 1;
- -             } else
- -                     p = NULL;
+ +     struct sched_domain *parent;
   
- -             sd = &per_cpu(node_domains, i).sd;
- -             SD_INIT(sd, NODE);
+ +     d->sd_allnodes = 0;
+ +     if (cpumask_weight(cpu_map) >
+ +         SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) {
+ +             sd = &per_cpu(allnodes_domains, i).sd;
+ +             SD_INIT(sd, ALLNODES);
                set_domain_attribute(sd, attr);
- -             sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd));
- -             sd->parent = p;
- -             if (p)
- -                     p->child = sd;
- -             cpumask_and(sched_domain_span(sd),
- -                         sched_domain_span(sd), cpu_map);
+ +             cpumask_copy(sched_domain_span(sd), cpu_map);
+ +             cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask);
+ +             d->sd_allnodes = 1;
+ +     }
+ +     parent = sd;
+ +
+ +     sd = &per_cpu(node_domains, i).sd;
+ +     SD_INIT(sd, NODE);
+ +     set_domain_attribute(sd, attr);
+ +     sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd));
+ +     sd->parent = parent;
+ +     if (parent)
+ +             parent->child = sd;
+ +     cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map);
   #endif
+ +     return sd;
+ +}
   
- -             p = sd;
- -             sd = &per_cpu(phys_domains, i).sd;
- -             SD_INIT(sd, CPU);
- -             set_domain_attribute(sd, attr);
- -             cpumask_copy(sched_domain_span(sd), nodemask);
- -             sd->parent = p;
- -             if (p)
- -                     p->child = sd;
- -             cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask);
+ +static struct sched_domain *__build_cpu_sched_domain(struct s_data *d,
+ +     const struct cpumask *cpu_map, struct sched_domain_attr *attr,
+ +     struct sched_domain *parent, int i)
+ +{
+ +     struct sched_domain *sd;
+ +     sd = &per_cpu(phys_domains, i).sd;
+ +     SD_INIT(sd, CPU);
+ +     set_domain_attribute(sd, attr);
+ +     cpumask_copy(sched_domain_span(sd), d->nodemask);
+ +     sd->parent = parent;
+ +     if (parent)
+ +             parent->child = sd;
+ +     cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask);
+ +     return sd;
+ +}
   
+ +static struct sched_domain *__build_mc_sched_domain(struct s_data *d,
+ +     const struct cpumask *cpu_map, struct sched_domain_attr *attr,
+ +     struct sched_domain *parent, int i)
+ +{
+ +     struct sched_domain *sd = parent;
   #ifdef CONFIG_SCHED_MC
- -             p = sd;
- -             sd = &per_cpu(core_domains, i).sd;
- -             SD_INIT(sd, MC);
- -             set_domain_attribute(sd, attr);
- -             cpumask_and(sched_domain_span(sd), cpu_map,
- -                                                cpu_coregroup_mask(i));
- -             sd->parent = p;
- -             p->child = sd;
- -             cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask);
+ +     sd = &per_cpu(core_domains, i).sd;
+ +     SD_INIT(sd, MC);
+ +     set_domain_attribute(sd, attr);
+ +     cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i));
+ +     sd->parent = parent;
+ +     parent->child = sd;
+ +     cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask);
   #endif
+ +     return sd;
+ +}
   
+ +static struct sched_domain *__build_smt_sched_domain(struct s_data *d,
+ +     const struct cpumask *cpu_map, struct sched_domain_attr *attr,
+ +     struct sched_domain *parent, int i)
+ +{
+ +     struct sched_domain *sd = parent;
   #ifdef CONFIG_SCHED_SMT
- -             p = sd;
- -             sd = &per_cpu(cpu_domains, i).sd;
- -             SD_INIT(sd, SIBLING);
- -             set_domain_attribute(sd, attr);
- -             cpumask_and(sched_domain_span(sd),
- -                         topology_thread_cpumask(i), cpu_map);
- -             sd->parent = p;
- -             p->child = sd;
- -             cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask);
+ +     sd = &per_cpu(cpu_domains, i).sd;
+ +     SD_INIT(sd, SIBLING);
+ +     set_domain_attribute(sd, attr);
+ +     cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i));
+ +     sd->parent = parent;
+ +     parent->child = sd;
+ +     cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask);
   #endif
- -     }
+ +     return sd;
+ +}
   
+ +static void build_sched_groups(struct s_data *d, enum sched_domain_level l,
+ +                            const struct cpumask *cpu_map, int cpu)
+ +{
+ +     switch (l) {
   #ifdef CONFIG_SCHED_SMT
- -     /* Set up CPU (sibling) groups */
- -     for_each_cpu(i, cpu_map) {
- -             cpumask_and(this_sibling_map,
- -                         topology_thread_cpumask(i), cpu_map);
- -             if (i != cpumask_first(this_sibling_map))
- -                     continue;
- -
- -             init_sched_build_groups(this_sibling_map, cpu_map,
- -                                     &cpu_to_cpu_group,
- -                                     send_covered, tmpmask);
- -     }
+ +     case SD_LV_SIBLING: /* set up CPU (sibling) groups */
+ +             cpumask_and(d->this_sibling_map, cpu_map,
+ +                         topology_thread_cpumask(cpu));
+ +             if (cpu == cpumask_first(d->this_sibling_map))
+ +                     init_sched_build_groups(d->this_sibling_map, cpu_map,
+ +                                             &cpu_to_cpu_group,
+ +                                             d->send_covered, d->tmpmask);
+ +             break;
   #endif
- -
   #ifdef CONFIG_SCHED_MC
- -     /* Set up multi-core groups */
- -     for_each_cpu(i, cpu_map) {
- -             cpumask_and(this_core_map, cpu_coregroup_mask(i), cpu_map);
- -             if (i != cpumask_first(this_core_map))
- -                     continue;
- -
- -             init_sched_build_groups(this_core_map, cpu_map,
- -                                     &cpu_to_core_group,
- -                                     send_covered, tmpmask);
- -     }
+ +     case SD_LV_MC: /* set up multi-core groups */
+ +             cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu));
+ +             if (cpu == cpumask_first(d->this_core_map))
+ +                     init_sched_build_groups(d->this_core_map, cpu_map,
+ +                                             &cpu_to_core_group,
+ +                                             d->send_covered, d->tmpmask);
+ +             break;
   #endif
- -
- -     /* Set up physical groups */
- -     for (i = 0; i < nr_node_ids; i++) {
- -             cpumask_and(nodemask, cpumask_of_node(i), cpu_map);
- -             if (cpumask_empty(nodemask))
- -                     continue;
- -
- -             init_sched_build_groups(nodemask, cpu_map,
- -                                     &cpu_to_phys_group,
- -                                     send_covered, tmpmask);
- -     }
- -
+ +     case SD_LV_CPU: /* set up physical groups */
+ +             cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map);
+ +             if (!cpumask_empty(d->nodemask))
+ +                     init_sched_build_groups(d->nodemask, cpu_map,
+ +                                             &cpu_to_phys_group,
+ +                                             d->send_covered, d->tmpmask);
+ +             break;
   #ifdef CONFIG_NUMA
- -     /* Set up node groups */
- -     if (sd_allnodes) {
- -             init_sched_build_groups(cpu_map, cpu_map,
- -                                     &cpu_to_allnodes_group,
- -                                     send_covered, tmpmask);
+ +     case SD_LV_ALLNODES:
+ +             init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group,
+ +                                     d->send_covered, d->tmpmask);
+ +             break;
+ +#endif
+ +     default:
+ +             break;
        }
+ +}
   
- -     for (i = 0; i < nr_node_ids; i++) {
- -             /* Set up node groups */
- -             struct sched_group *sg, *prev;
- -             int j;
- -
- -             cpumask_clear(covered);
- -             cpumask_and(nodemask, cpumask_of_node(i), cpu_map);
- -             if (cpumask_empty(nodemask)) {
- -                     sched_group_nodes[i] = NULL;
- -                     continue;
- -             }
+ +/*
+ + * Build sched domains for a given set of cpus and attach the sched domains
+ + * to the individual cpus
+ + */
+ +static int __build_sched_domains(const struct cpumask *cpu_map,
+ +                              struct sched_domain_attr *attr)
+ +{
+ +     enum s_alloc alloc_state = sa_none;
+ +     struct s_data d;
+ +     struct sched_domain *sd;
+ +     int i;
+ +#ifdef CONFIG_NUMA
+ +     d.sd_allnodes = 0;
+ +#endif
   
- -             sched_domain_node_span(i, domainspan);
- -             cpumask_and(domainspan, domainspan, cpu_map);
+ +     alloc_state = __visit_domain_allocation_hell(&d, cpu_map);
+ +     if (alloc_state != sa_rootdomain)
+ +             goto error;
+ +     alloc_state = sa_sched_groups;
   
- -             sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(),
- -                               GFP_KERNEL, i);
- -             if (!sg) {
- -                     printk(KERN_WARNING "Can not alloc domain group for "
- -                             "node %d\n", i);
- -                     goto error;
- -             }
- -             sched_group_nodes[i] = sg;
- -             for_each_cpu(j, nodemask) {
- -                     struct sched_domain *sd;
+ +     /*
+ +      * Set up domains for cpus specified by the cpu_map.
+ +      */
+ +     for_each_cpu(i, cpu_map) {
+ +             cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)),
+ +                         cpu_map);
   
- -                     sd = &per_cpu(node_domains, j).sd;
- -                     sd->groups = sg;
- -             }
- -             sg->__cpu_power = 0;
- -             cpumask_copy(sched_group_cpus(sg), nodemask);
- -             sg->next = sg;
- -             cpumask_or(covered, covered, nodemask);
- -             prev = sg;
+ +             sd = __build_numa_sched_domains(&d, cpu_map, attr, i);
+ +             sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i);
+ +             sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i);
+ +             sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i);
+ +     }
   
- -             for (j = 0; j < nr_node_ids; j++) {
- -                     int n = (i + j) % nr_node_ids;
+ +     for_each_cpu(i, cpu_map) {
+ +             build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i);
+ +             build_sched_groups(&d, SD_LV_MC, cpu_map, i);
+ +     }
   
- -                     cpumask_complement(notcovered, covered);
- -                     cpumask_and(tmpmask, notcovered, cpu_map);
- -                     cpumask_and(tmpmask, tmpmask, domainspan);
- -                     if (cpumask_empty(tmpmask))
- -                             break;
+ +     /* Set up physical groups */
+ +     for (i = 0; i < nr_node_ids; i++)
+ +             build_sched_groups(&d, SD_LV_CPU, cpu_map, i);
   
- -                     cpumask_and(tmpmask, tmpmask, cpumask_of_node(n));
- -                     if (cpumask_empty(tmpmask))
- -                             continue;
+ +#ifdef CONFIG_NUMA
+ +     /* Set up node groups */
+ +     if (d.sd_allnodes)
+ +             build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0);
   
- -                     sg = kmalloc_node(sizeof(struct sched_group) +
- -                                       cpumask_size(),
- -                                       GFP_KERNEL, i);
- -                     if (!sg) {
- -                             printk(KERN_WARNING
- -                             "Can not alloc domain group for node %d\n", j);
- -                             goto error;
- -                     }
- -                     sg->__cpu_power = 0;
- -                     cpumask_copy(sched_group_cpus(sg), tmpmask);
- -                     sg->next = prev->next;
- -                     cpumask_or(covered, covered, tmpmask);
- -                     prev->next = sg;
- -                     prev = sg;
- -             }
- -     }
+ +     for (i = 0; i < nr_node_ids; i++)
+ +             if (build_numa_sched_groups(&d, cpu_map, i))
+ +                     goto error;
   #endif
   
        /* Calculate CPU power for physical packages and nodes */
   #else
                sd = &per_cpu(phys_domains, i).sd;
   #endif
- -             cpu_attach_domain(sd, rd, i);
-       }
-  
-       err = 0;
-  
-  free_tmpmask:
-       free_cpumask_var(tmpmask);
-  free_send_covered:
-       free_cpumask_var(send_covered);
-  free_this_core_map:
-       free_cpumask_var(this_core_map);
-  free_this_sibling_map:
-       free_cpumask_var(this_sibling_map);
-  free_nodemask:
-       free_cpumask_var(nodemask);
-  free_notcovered:
-  #ifdef CONFIG_NUMA
-       free_cpumask_var(notcovered);
-  free_covered:
-       free_cpumask_var(covered);
-  free_domainspan:
-       free_cpumask_var(domainspan);
-  out:
-  #endif
-       return err;
+ +             cpu_attach_domain(sd, d.rd, i);
+       }
   
  -     err = 0;
  -
  -free_tmpmask:
  -     free_cpumask_var(tmpmask);
  -free_send_covered:
  -     free_cpumask_var(send_covered);
  -free_this_core_map:
  -     free_cpumask_var(this_core_map);
  -free_this_sibling_map:
  -     free_cpumask_var(this_sibling_map);
  -free_nodemask:
  -     free_cpumask_var(nodemask);
  -free_notcovered:
  -#ifdef CONFIG_NUMA
  -     free_cpumask_var(notcovered);
  -free_covered:
  -     free_cpumask_var(covered);
  -free_domainspan:
  -     free_cpumask_var(domainspan);
  -out:
  -#endif
  -     return err;
  -
- -free_sched_groups:
- -#ifdef CONFIG_NUMA
- -     kfree(sched_group_nodes);
- -#endif
- -     goto free_tmpmask;
+ +     d.sched_group_nodes = NULL; /* don't free this we still need it */
+ +     __free_domain_allocs(&d, sa_tmpmask, cpu_map);
+ +     return 0;
   
- -#ifdef CONFIG_NUMA
   error:
- -     free_sched_groups(cpu_map, tmpmask);
- -     free_rootdomain(rd);
- -     goto free_tmpmask;
- -#endif
+ +     __free_domain_allocs(&d, alloc_state, cpu_map);
+ +     return -ENOMEM;
   }
   
   static int build_sched_domains(const struct cpumask *cpu_map)