kernel_optimize_test/kernel/sched/topology.c

1925 lines
48 KiB
C
Raw Normal View History

/*
* Scheduler topology setup/handling methods
*/
#include <linux/sched.h>
#include <linux/mutex.h>
#include "sched.h"
DEFINE_MUTEX(sched_domains_mutex);
/* Protected by sched_domains_mutex: */
cpumask_var_t sched_domains_tmpmask;
sched/topology: Fix overlapping sched_group_capacity When building the overlapping groups we need to attach a consistent sched_group_capacity structure. That is, all 'identical' sched_group's should have the _same_ sched_group_capacity. This can (once again) be demonstrated with a topology like: node 0 1 2 3 0: 10 20 30 20 1: 20 10 20 30 2: 30 20 10 20 3: 20 30 20 10 But we need at least 2 CPUs per node for this to show up, after all, if there is only one CPU per node, our CPU @i is per definition a unique CPU that reaches this domain (aka balance-cpu). Given the above NUMA topo and 2 CPUs per node: [] CPU0 attaching sched-domain(s): [] domain-0: span=0,4 level=DIE [] groups: 0:{ span=0 }, 4:{ span=4 } [] domain-1: span=0-1,3-5,7 level=NUMA [] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 } [] CPU1 attaching sched-domain(s): [] domain-0: span=1,5 level=DIE [] groups: 1:{ span=1 }, 5:{ span=5 } [] domain-1: span=0-2,4-6 level=NUMA [] groups: 1:{ span=1,5 mask=1,5 cap=2048 }, 2:{ span=2,6 mask=2,6 cap=2048 }, 4:{ span=0,4 mask=0,4 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 1:{ span=0-2,4-6 mask=1,5 cap=6144 }, 3:{ span=0,2-4,6-7 mask=3,7 cap=6144 } Observe how CPU0-domain1-group0 and CPU1-domain1-group4 are the 'same' but have a different id (0 vs 4). To fix this, use the group balance CPU to select the SGC. This means we have to compute the full mask for each CPU and require a second temporary mask to store the group mask in (it otherwise lives in the SGC). The fixed topology looks like: [] CPU0 attaching sched-domain(s): [] domain-0: span=0,4 level=DIE [] groups: 0:{ span=0 }, 4:{ span=4 } [] domain-1: span=0-1,3-5,7 level=NUMA [] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 } [] CPU1 attaching sched-domain(s): [] domain-0: span=1,5 level=DIE [] groups: 1:{ span=1 }, 5:{ span=5 } [] domain-1: span=0-2,4-6 level=NUMA [] groups: 1:{ span=1,5 mask=1,5 cap=2048 }, 2:{ span=2,6 mask=2,6 cap=2048 }, 0:{ span=0,4 mask=0,4 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 1:{ span=0-2,4-6 mask=1,5 cap=6144 }, 3:{ span=0,2-4,6-7 mask=3,7 cap=6144 } Debugged-by: Lauro Ramos Venancio <lvenanci@redhat.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-kernel@vger.kernel.org Fixes: e3589f6c81e4 ("sched: Allow for overlapping sched_domain spans") Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-04-25 20:31:11 +08:00
cpumask_var_t sched_domains_tmpmask2;
#ifdef CONFIG_SCHED_DEBUG
static int __init sched_debug_setup(char *str)
{
sched_debug_enabled = true;
return 0;
}
early_param("sched_debug", sched_debug_setup);
static inline bool sched_debug(void)
{
return sched_debug_enabled;
}
static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
struct cpumask *groupmask)
{
struct sched_group *group = sd->groups;
cpumask_clear(groupmask);
printk(KERN_DEBUG "%*s domain-%d: ", level, "", level);
if (!(sd->flags & SD_LOAD_BALANCE)) {
printk("does not load-balance\n");
if (sd->parent)
printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain"
" has parent");
return -1;
}
printk(KERN_CONT "span=%*pbl level=%s\n",
cpumask_pr_args(sched_domain_span(sd)), sd->name);
if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) {
printk(KERN_ERR "ERROR: domain->span does not contain "
"CPU%d\n", cpu);
}
if (!cpumask_test_cpu(cpu, sched_group_span(group))) {
printk(KERN_ERR "ERROR: domain->groups does not contain"
" CPU%d\n", cpu);
}
printk(KERN_DEBUG "%*s groups:", level + 1, "");
do {
if (!group) {
printk("\n");
printk(KERN_ERR "ERROR: group is NULL\n");
break;
}
if (!cpumask_weight(sched_group_span(group))) {
printk(KERN_CONT "\n");
printk(KERN_ERR "ERROR: empty group\n");
break;
}
if (!(sd->flags & SD_OVERLAP) &&
cpumask_intersects(groupmask, sched_group_span(group))) {
printk(KERN_CONT "\n");
printk(KERN_ERR "ERROR: repeated CPUs\n");
break;
}
cpumask_or(groupmask, groupmask, sched_group_span(group));
printk(KERN_CONT " %d:{ span=%*pbl",
group->sgc->id,
cpumask_pr_args(sched_group_span(group)));
if ((sd->flags & SD_OVERLAP) &&
!cpumask_equal(group_balance_mask(group), sched_group_span(group))) {
printk(KERN_CONT " mask=%*pbl",
cpumask_pr_args(group_balance_mask(group)));
}
if (group->sgc->capacity != SCHED_CAPACITY_SCALE)
printk(KERN_CONT " cap=%lu", group->sgc->capacity);
if (group == sd->groups && sd->child &&
!cpumask_equal(sched_domain_span(sd->child),
sched_group_span(group))) {
printk(KERN_ERR "ERROR: domain->groups does not match domain->child\n");
}
printk(KERN_CONT " }");
group = group->next;
if (group != sd->groups)
printk(KERN_CONT ",");
} while (group != sd->groups);
printk(KERN_CONT "\n");
if (!cpumask_equal(sched_domain_span(sd), groupmask))
printk(KERN_ERR "ERROR: groups don't span domain->span\n");
if (sd->parent &&
!cpumask_subset(groupmask, sched_domain_span(sd->parent)))
printk(KERN_ERR "ERROR: parent span is not a superset "
"of domain->span\n");
return 0;
}
static void sched_domain_debug(struct sched_domain *sd, int cpu)
{
int level = 0;
if (!sched_debug_enabled)
return;
if (!sd) {
printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu);
return;
}
printk(KERN_DEBUG "CPU%d attaching sched-domain(s):\n", cpu);
for (;;) {
if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask))
break;
level++;
sd = sd->parent;
if (!sd)
break;
}
}
#else /* !CONFIG_SCHED_DEBUG */
# define sched_debug_enabled 0
# define sched_domain_debug(sd, cpu) do { } while (0)
static inline bool sched_debug(void)
{
return false;
}
#endif /* CONFIG_SCHED_DEBUG */
static int sd_degenerate(struct sched_domain *sd)
{
if (cpumask_weight(sched_domain_span(sd)) == 1)
return 1;
/* Following flags need at least 2 groups */
if (sd->flags & (SD_LOAD_BALANCE |
SD_BALANCE_NEWIDLE |
SD_BALANCE_FORK |
SD_BALANCE_EXEC |
SD_SHARE_CPUCAPACITY |
SD_ASYM_CPUCAPACITY |
SD_SHARE_PKG_RESOURCES |
SD_SHARE_POWERDOMAIN)) {
if (sd->groups != sd->groups->next)
return 0;
}
/* Following flags don't use groups */
if (sd->flags & (SD_WAKE_AFFINE))
return 0;
return 1;
}
static int
sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
{
unsigned long cflags = sd->flags, pflags = parent->flags;
if (sd_degenerate(parent))
return 1;
if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent)))
return 0;
/* Flags needing groups don't count if only 1 group in parent */
if (parent->groups == parent->groups->next) {
pflags &= ~(SD_LOAD_BALANCE |
SD_BALANCE_NEWIDLE |
SD_BALANCE_FORK |
SD_BALANCE_EXEC |
SD_ASYM_CPUCAPACITY |
SD_SHARE_CPUCAPACITY |
SD_SHARE_PKG_RESOURCES |
SD_PREFER_SIBLING |
SD_SHARE_POWERDOMAIN);
if (nr_node_ids == 1)
pflags &= ~SD_SERIALIZE;
}
if (~cflags & pflags)
return 0;
return 1;
}
static void free_rootdomain(struct rcu_head *rcu)
{
struct root_domain *rd = container_of(rcu, struct root_domain, rcu);
cpupri_cleanup(&rd->cpupri);
cpudl_cleanup(&rd->cpudl);
free_cpumask_var(rd->dlo_mask);
free_cpumask_var(rd->rto_mask);
free_cpumask_var(rd->online);
free_cpumask_var(rd->span);
kfree(rd);
}
void rq_attach_root(struct rq *rq, struct root_domain *rd)
{
struct root_domain *old_rd = NULL;
unsigned long flags;
raw_spin_lock_irqsave(&rq->lock, flags);
if (rq->rd) {
old_rd = rq->rd;
if (cpumask_test_cpu(rq->cpu, old_rd->online))
set_rq_offline(rq);
cpumask_clear_cpu(rq->cpu, old_rd->span);
/*
* If we dont want to free the old_rd yet then
* set old_rd to NULL to skip the freeing later
* in this function:
*/
if (!atomic_dec_and_test(&old_rd->refcount))
old_rd = NULL;
}
atomic_inc(&rd->refcount);
rq->rd = rd;
cpumask_set_cpu(rq->cpu, rd->span);
if (cpumask_test_cpu(rq->cpu, cpu_active_mask))
set_rq_online(rq);
raw_spin_unlock_irqrestore(&rq->lock, flags);
if (old_rd)
call_rcu_sched(&old_rd->rcu, free_rootdomain);
}
static int init_rootdomain(struct root_domain *rd)
{
if (!zalloc_cpumask_var(&rd->span, GFP_KERNEL))
goto out;
if (!zalloc_cpumask_var(&rd->online, GFP_KERNEL))
goto free_span;
if (!zalloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL))
goto free_online;
if (!zalloc_cpumask_var(&rd->rto_mask, GFP_KERNEL))
goto free_dlo_mask;
init_dl_bw(&rd->dl_bw);
if (cpudl_init(&rd->cpudl) != 0)
goto free_rto_mask;
if (cpupri_init(&rd->cpupri) != 0)
goto free_cpudl;
return 0;
free_cpudl:
cpudl_cleanup(&rd->cpudl);
free_rto_mask:
free_cpumask_var(rd->rto_mask);
free_dlo_mask:
free_cpumask_var(rd->dlo_mask);
free_online:
free_cpumask_var(rd->online);
free_span:
free_cpumask_var(rd->span);
out:
return -ENOMEM;
}
/*
* By default the system creates a single root-domain with all CPUs as
* members (mimicking the global state we have today).
*/
struct root_domain def_root_domain;
void init_defrootdomain(void)
{
init_rootdomain(&def_root_domain);
atomic_set(&def_root_domain.refcount, 1);
}
static struct root_domain *alloc_rootdomain(void)
{
struct root_domain *rd;
rd = kzalloc(sizeof(*rd), GFP_KERNEL);
if (!rd)
return NULL;
if (init_rootdomain(rd) != 0) {
kfree(rd);
return NULL;
}
return rd;
}
static void free_sched_groups(struct sched_group *sg, int free_sgc)
{
struct sched_group *tmp, *first;
if (!sg)
return;
first = sg;
do {
tmp = sg->next;
if (free_sgc && atomic_dec_and_test(&sg->sgc->ref))
kfree(sg->sgc);
sched/topology: Fix memory leak in __sdt_alloc() Found this issue by kmemleak: the 'sg' and 'sgc' pointers from __sdt_alloc() might be leaked as each domain holds many groups' ref, but in destroy_sched_domain(), it only declined the first group ref. Onlining and offlining a CPU can trigger this leak, and cause OOM. Reproducer for my 6 CPUs machine: while true do echo 0 > /sys/devices/system/cpu/cpu5/online; echo 1 > /sys/devices/system/cpu/cpu5/online; done unreferenced object 0xffff88007d772a80 (size 64): comm "cpuhp/5", pid 39, jiffies 4294719962 (age 35.251s) hex dump (first 32 bytes): c0 22 77 7d 00 88 ff ff 02 00 00 00 01 00 00 00 ."w}............ 40 2a 77 7d 00 88 ff ff 00 00 00 00 00 00 00 00 @*w}............ backtrace: [<ffffffff8176525a>] kmemleak_alloc+0x4a/0xa0 [<ffffffff8121efe1>] __kmalloc_node+0xf1/0x280 [<ffffffff810d94a8>] build_sched_domains+0x1e8/0xf20 [<ffffffff810da674>] partition_sched_domains+0x304/0x360 [<ffffffff81139557>] cpuset_update_active_cpus+0x17/0x40 [<ffffffff810bdb2e>] sched_cpu_activate+0xae/0xc0 [<ffffffff810900e0>] cpuhp_invoke_callback+0x90/0x400 [<ffffffff81090597>] cpuhp_up_callbacks+0x37/0xb0 [<ffffffff81090887>] cpuhp_thread_fun+0xd7/0xf0 [<ffffffff810b37e0>] smpboot_thread_fn+0x110/0x160 [<ffffffff810af5d9>] kthread+0x109/0x140 [<ffffffff81770e45>] ret_from_fork+0x25/0x30 [<ffffffffffffffff>] 0xffffffffffffffff unreferenced object 0xffff88007d772a40 (size 64): comm "cpuhp/5", pid 39, jiffies 4294719962 (age 35.251s) hex dump (first 32 bytes): 03 00 00 00 00 00 00 00 00 04 00 00 00 00 00 00 ................ 00 04 00 00 00 00 00 00 4f 3c fc ff 00 00 00 00 ........O<...... backtrace: [<ffffffff8176525a>] kmemleak_alloc+0x4a/0xa0 [<ffffffff8121efe1>] __kmalloc_node+0xf1/0x280 [<ffffffff810da16d>] build_sched_domains+0xead/0xf20 [<ffffffff810da674>] partition_sched_domains+0x304/0x360 [<ffffffff81139557>] cpuset_update_active_cpus+0x17/0x40 [<ffffffff810bdb2e>] sched_cpu_activate+0xae/0xc0 [<ffffffff810900e0>] cpuhp_invoke_callback+0x90/0x400 [<ffffffff81090597>] cpuhp_up_callbacks+0x37/0xb0 [<ffffffff81090887>] cpuhp_thread_fun+0xd7/0xf0 [<ffffffff810b37e0>] smpboot_thread_fn+0x110/0x160 [<ffffffff810af5d9>] kthread+0x109/0x140 [<ffffffff81770e45>] ret_from_fork+0x25/0x30 [<ffffffffffffffff>] 0xffffffffffffffff Reported-by: Chunyu Hu <chuhu@redhat.com> Signed-off-by: Shu Wang <shuwang@redhat.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Acked-by: Chunyu Hu <chuhu@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: liwang@redhat.com Link: http://lkml.kernel.org/r/1502351536-9108-1-git-send-email-shuwang@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-08-10 15:52:16 +08:00
if (atomic_dec_and_test(&sg->ref))
kfree(sg);
sg = tmp;
} while (sg != first);
}
static void destroy_sched_domain(struct sched_domain *sd)
{
/*
* A normal sched domain may have multiple group references, an
* overlapping domain, having private groups, only one. Iterate,
* dropping group/capacity references, freeing where none remain.
*/
sched/topology: Fix memory leak in __sdt_alloc() Found this issue by kmemleak: the 'sg' and 'sgc' pointers from __sdt_alloc() might be leaked as each domain holds many groups' ref, but in destroy_sched_domain(), it only declined the first group ref. Onlining and offlining a CPU can trigger this leak, and cause OOM. Reproducer for my 6 CPUs machine: while true do echo 0 > /sys/devices/system/cpu/cpu5/online; echo 1 > /sys/devices/system/cpu/cpu5/online; done unreferenced object 0xffff88007d772a80 (size 64): comm "cpuhp/5", pid 39, jiffies 4294719962 (age 35.251s) hex dump (first 32 bytes): c0 22 77 7d 00 88 ff ff 02 00 00 00 01 00 00 00 ."w}............ 40 2a 77 7d 00 88 ff ff 00 00 00 00 00 00 00 00 @*w}............ backtrace: [<ffffffff8176525a>] kmemleak_alloc+0x4a/0xa0 [<ffffffff8121efe1>] __kmalloc_node+0xf1/0x280 [<ffffffff810d94a8>] build_sched_domains+0x1e8/0xf20 [<ffffffff810da674>] partition_sched_domains+0x304/0x360 [<ffffffff81139557>] cpuset_update_active_cpus+0x17/0x40 [<ffffffff810bdb2e>] sched_cpu_activate+0xae/0xc0 [<ffffffff810900e0>] cpuhp_invoke_callback+0x90/0x400 [<ffffffff81090597>] cpuhp_up_callbacks+0x37/0xb0 [<ffffffff81090887>] cpuhp_thread_fun+0xd7/0xf0 [<ffffffff810b37e0>] smpboot_thread_fn+0x110/0x160 [<ffffffff810af5d9>] kthread+0x109/0x140 [<ffffffff81770e45>] ret_from_fork+0x25/0x30 [<ffffffffffffffff>] 0xffffffffffffffff unreferenced object 0xffff88007d772a40 (size 64): comm "cpuhp/5", pid 39, jiffies 4294719962 (age 35.251s) hex dump (first 32 bytes): 03 00 00 00 00 00 00 00 00 04 00 00 00 00 00 00 ................ 00 04 00 00 00 00 00 00 4f 3c fc ff 00 00 00 00 ........O<...... backtrace: [<ffffffff8176525a>] kmemleak_alloc+0x4a/0xa0 [<ffffffff8121efe1>] __kmalloc_node+0xf1/0x280 [<ffffffff810da16d>] build_sched_domains+0xead/0xf20 [<ffffffff810da674>] partition_sched_domains+0x304/0x360 [<ffffffff81139557>] cpuset_update_active_cpus+0x17/0x40 [<ffffffff810bdb2e>] sched_cpu_activate+0xae/0xc0 [<ffffffff810900e0>] cpuhp_invoke_callback+0x90/0x400 [<ffffffff81090597>] cpuhp_up_callbacks+0x37/0xb0 [<ffffffff81090887>] cpuhp_thread_fun+0xd7/0xf0 [<ffffffff810b37e0>] smpboot_thread_fn+0x110/0x160 [<ffffffff810af5d9>] kthread+0x109/0x140 [<ffffffff81770e45>] ret_from_fork+0x25/0x30 [<ffffffffffffffff>] 0xffffffffffffffff Reported-by: Chunyu Hu <chuhu@redhat.com> Signed-off-by: Shu Wang <shuwang@redhat.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Acked-by: Chunyu Hu <chuhu@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: liwang@redhat.com Link: http://lkml.kernel.org/r/1502351536-9108-1-git-send-email-shuwang@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-08-10 15:52:16 +08:00
free_sched_groups(sd->groups, 1);
if (sd->shared && atomic_dec_and_test(&sd->shared->ref))
kfree(sd->shared);
kfree(sd);
}
static void destroy_sched_domains_rcu(struct rcu_head *rcu)
{
struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu);
while (sd) {
struct sched_domain *parent = sd->parent;
destroy_sched_domain(sd);
sd = parent;
}
}
static void destroy_sched_domains(struct sched_domain *sd)
{
if (sd)
call_rcu(&sd->rcu, destroy_sched_domains_rcu);
}
/*
* 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().
*
* 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 *, sd_llc);
DEFINE_PER_CPU(int, sd_llc_size);
DEFINE_PER_CPU(int, sd_llc_id);
DEFINE_PER_CPU(struct sched_domain_shared *, sd_llc_shared);
DEFINE_PER_CPU(struct sched_domain *, sd_numa);
DEFINE_PER_CPU(struct sched_domain *, sd_asym);
static void update_top_cache_domain(int cpu)
{
struct sched_domain_shared *sds = NULL;
struct sched_domain *sd;
int id = cpu;
int size = 1;
sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES);
if (sd) {
id = cpumask_first(sched_domain_span(sd));
size = cpumask_weight(sched_domain_span(sd));
sds = sd->shared;
}
rcu_assign_pointer(per_cpu(sd_llc, cpu), sd);
per_cpu(sd_llc_size, cpu) = size;
per_cpu(sd_llc_id, cpu) = id;
rcu_assign_pointer(per_cpu(sd_llc_shared, cpu), sds);
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, cpu), sd);
}
/*
* Attach the domain 'sd' to 'cpu' as its base domain. Callers must
* hold the hotplug lock.
*/
static void
cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu)
{
struct rq *rq = cpu_rq(cpu);
struct sched_domain *tmp;
/* Remove the sched domains which do not contribute to scheduling. */
for (tmp = sd; tmp; ) {
struct sched_domain *parent = tmp->parent;
if (!parent)
break;
if (sd_parent_degenerate(tmp, parent)) {
tmp->parent = parent->parent;
if (parent->parent)
parent->parent->child = tmp;
/*
* Transfer SD_PREFER_SIBLING down in case of a
* degenerate parent; the spans match for this
* so the property transfers.
*/
if (parent->flags & SD_PREFER_SIBLING)
tmp->flags |= SD_PREFER_SIBLING;
destroy_sched_domain(parent);
} else
tmp = tmp->parent;
}
if (sd && sd_degenerate(sd)) {
tmp = sd;
sd = sd->parent;
destroy_sched_domain(tmp);
if (sd)
sd->child = NULL;
}
sched_domain_debug(sd, cpu);
rq_attach_root(rq, rd);
tmp = rq->sd;
rcu_assign_pointer(rq->sd, sd);
dirty_sched_domain_sysctl(cpu);
destroy_sched_domains(tmp);
update_top_cache_domain(cpu);
}
/* Setup the mask of CPUs configured for isolated domains */
static int __init isolated_cpu_setup(char *str)
{
int ret;
alloc_bootmem_cpumask_var(&cpu_isolated_map);
ret = cpulist_parse(str, cpu_isolated_map);
if (ret) {
pr_err("sched: Error, all isolcpus= values must be between 0 and %u\n", nr_cpu_ids);
return 0;
}
return 1;
}
__setup("isolcpus=", isolated_cpu_setup);
struct s_data {
struct sched_domain ** __percpu sd;
struct root_domain *rd;
};
enum s_alloc {
sa_rootdomain,
sa_sd,
sa_sd_storage,
sa_none,
};
/*
* Return the canonical balance CPU for this group, this is the first CPU
* of this group that's also in the balance mask.
*
* The balance mask are all those CPUs that could actually end up at this
* group. See build_balance_mask().
*
* Also see should_we_balance().
*/
int group_balance_cpu(struct sched_group *sg)
{
return cpumask_first(group_balance_mask(sg));
}
/*
* NUMA topology (first read the regular topology blurb below)
*
* Given a node-distance table, for example:
*
* node 0 1 2 3
* 0: 10 20 30 20
* 1: 20 10 20 30
* 2: 30 20 10 20
* 3: 20 30 20 10
*
* which represents a 4 node ring topology like:
*
* 0 ----- 1
* | |
* | |
* | |
* 3 ----- 2
*
* We want to construct domains and groups to represent this. The way we go
* about doing this is to build the domains on 'hops'. For each NUMA level we
* construct the mask of all nodes reachable in @level hops.
*
* For the above NUMA topology that gives 3 levels:
*
* NUMA-2 0-3 0-3 0-3 0-3
* groups: {0-1,3},{1-3} {0-2},{0,2-3} {1-3},{0-1,3} {0,2-3},{0-2}
*
* NUMA-1 0-1,3 0-2 1-3 0,2-3
* groups: {0},{1},{3} {0},{1},{2} {1},{2},{3} {0},{2},{3}
*
* NUMA-0 0 1 2 3
*
*
* As can be seen; things don't nicely line up as with the regular topology.
* When we iterate a domain in child domain chunks some nodes can be
* represented multiple times -- hence the "overlap" naming for this part of
* the topology.
*
* In order to minimize this overlap, we only build enough groups to cover the
* domain. For instance Node-0 NUMA-2 would only get groups: 0-1,3 and 1-3.
*
* Because:
*
* - the first group of each domain is its child domain; this
* gets us the first 0-1,3
* - the only uncovered node is 2, who's child domain is 1-3.
*
* However, because of the overlap, computing a unique CPU for each group is
* more complicated. Consider for instance the groups of NODE-1 NUMA-2, both
* groups include the CPUs of Node-0, while those CPUs would not in fact ever
* end up at those groups (they would end up in group: 0-1,3).
*
* To correct this we have to introduce the group balance mask. This mask
* will contain those CPUs in the group that can reach this group given the
* (child) domain tree.
*
* With this we can once again compute balance_cpu and sched_group_capacity
* relations.
*
* XXX include words on how balance_cpu is unique and therefore can be
* used for sched_group_capacity links.
*
*
* Another 'interesting' topology is:
*
* node 0 1 2 3
* 0: 10 20 20 30
* 1: 20 10 20 20
* 2: 20 20 10 20
* 3: 30 20 20 10
*
* Which looks a little like:
*
* 0 ----- 1
* | / |
* | / |
* | / |
* 2 ----- 3
*
* This topology is asymmetric, nodes 1,2 are fully connected, but nodes 0,3
* are not.
*
* This leads to a few particularly weird cases where the sched_domain's are
* not of the same number for each cpu. Consider:
*
* NUMA-2 0-3 0-3
* groups: {0-2},{1-3} {1-3},{0-2}
*
* NUMA-1 0-2 0-3 0-3 1-3
*
* NUMA-0 0 1 2 3
*
*/
/*
* Build the balance mask; it contains only those CPUs that can arrive at this
* group and should be considered to continue balancing.
*
* We do this during the group creation pass, therefore the group information
* isn't complete yet, however since each group represents a (child) domain we
* can fully construct this using the sched_domain bits (which are already
* complete).
*/
sched/topology: Fix overlapping sched_group_capacity When building the overlapping groups we need to attach a consistent sched_group_capacity structure. That is, all 'identical' sched_group's should have the _same_ sched_group_capacity. This can (once again) be demonstrated with a topology like: node 0 1 2 3 0: 10 20 30 20 1: 20 10 20 30 2: 30 20 10 20 3: 20 30 20 10 But we need at least 2 CPUs per node for this to show up, after all, if there is only one CPU per node, our CPU @i is per definition a unique CPU that reaches this domain (aka balance-cpu). Given the above NUMA topo and 2 CPUs per node: [] CPU0 attaching sched-domain(s): [] domain-0: span=0,4 level=DIE [] groups: 0:{ span=0 }, 4:{ span=4 } [] domain-1: span=0-1,3-5,7 level=NUMA [] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 } [] CPU1 attaching sched-domain(s): [] domain-0: span=1,5 level=DIE [] groups: 1:{ span=1 }, 5:{ span=5 } [] domain-1: span=0-2,4-6 level=NUMA [] groups: 1:{ span=1,5 mask=1,5 cap=2048 }, 2:{ span=2,6 mask=2,6 cap=2048 }, 4:{ span=0,4 mask=0,4 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 1:{ span=0-2,4-6 mask=1,5 cap=6144 }, 3:{ span=0,2-4,6-7 mask=3,7 cap=6144 } Observe how CPU0-domain1-group0 and CPU1-domain1-group4 are the 'same' but have a different id (0 vs 4). To fix this, use the group balance CPU to select the SGC. This means we have to compute the full mask for each CPU and require a second temporary mask to store the group mask in (it otherwise lives in the SGC). The fixed topology looks like: [] CPU0 attaching sched-domain(s): [] domain-0: span=0,4 level=DIE [] groups: 0:{ span=0 }, 4:{ span=4 } [] domain-1: span=0-1,3-5,7 level=NUMA [] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 } [] CPU1 attaching sched-domain(s): [] domain-0: span=1,5 level=DIE [] groups: 1:{ span=1 }, 5:{ span=5 } [] domain-1: span=0-2,4-6 level=NUMA [] groups: 1:{ span=1,5 mask=1,5 cap=2048 }, 2:{ span=2,6 mask=2,6 cap=2048 }, 0:{ span=0,4 mask=0,4 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 1:{ span=0-2,4-6 mask=1,5 cap=6144 }, 3:{ span=0,2-4,6-7 mask=3,7 cap=6144 } Debugged-by: Lauro Ramos Venancio <lvenanci@redhat.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-kernel@vger.kernel.org Fixes: e3589f6c81e4 ("sched: Allow for overlapping sched_domain spans") Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-04-25 20:31:11 +08:00
static void
build_balance_mask(struct sched_domain *sd, struct sched_group *sg, struct cpumask *mask)
{
const struct cpumask *sg_span = sched_group_span(sg);
struct sd_data *sdd = sd->private;
struct sched_domain *sibling;
int i;
sched/topology: Fix overlapping sched_group_capacity When building the overlapping groups we need to attach a consistent sched_group_capacity structure. That is, all 'identical' sched_group's should have the _same_ sched_group_capacity. This can (once again) be demonstrated with a topology like: node 0 1 2 3 0: 10 20 30 20 1: 20 10 20 30 2: 30 20 10 20 3: 20 30 20 10 But we need at least 2 CPUs per node for this to show up, after all, if there is only one CPU per node, our CPU @i is per definition a unique CPU that reaches this domain (aka balance-cpu). Given the above NUMA topo and 2 CPUs per node: [] CPU0 attaching sched-domain(s): [] domain-0: span=0,4 level=DIE [] groups: 0:{ span=0 }, 4:{ span=4 } [] domain-1: span=0-1,3-5,7 level=NUMA [] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 } [] CPU1 attaching sched-domain(s): [] domain-0: span=1,5 level=DIE [] groups: 1:{ span=1 }, 5:{ span=5 } [] domain-1: span=0-2,4-6 level=NUMA [] groups: 1:{ span=1,5 mask=1,5 cap=2048 }, 2:{ span=2,6 mask=2,6 cap=2048 }, 4:{ span=0,4 mask=0,4 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 1:{ span=0-2,4-6 mask=1,5 cap=6144 }, 3:{ span=0,2-4,6-7 mask=3,7 cap=6144 } Observe how CPU0-domain1-group0 and CPU1-domain1-group4 are the 'same' but have a different id (0 vs 4). To fix this, use the group balance CPU to select the SGC. This means we have to compute the full mask for each CPU and require a second temporary mask to store the group mask in (it otherwise lives in the SGC). The fixed topology looks like: [] CPU0 attaching sched-domain(s): [] domain-0: span=0,4 level=DIE [] groups: 0:{ span=0 }, 4:{ span=4 } [] domain-1: span=0-1,3-5,7 level=NUMA [] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 } [] CPU1 attaching sched-domain(s): [] domain-0: span=1,5 level=DIE [] groups: 1:{ span=1 }, 5:{ span=5 } [] domain-1: span=0-2,4-6 level=NUMA [] groups: 1:{ span=1,5 mask=1,5 cap=2048 }, 2:{ span=2,6 mask=2,6 cap=2048 }, 0:{ span=0,4 mask=0,4 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 1:{ span=0-2,4-6 mask=1,5 cap=6144 }, 3:{ span=0,2-4,6-7 mask=3,7 cap=6144 } Debugged-by: Lauro Ramos Venancio <lvenanci@redhat.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-kernel@vger.kernel.org Fixes: e3589f6c81e4 ("sched: Allow for overlapping sched_domain spans") Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-04-25 20:31:11 +08:00
cpumask_clear(mask);
for_each_cpu(i, sg_span) {
sibling = *per_cpu_ptr(sdd->sd, i);
sched/topology: Fix overlapping sched_group_mask The point of sched_group_mask is to select those CPUs from sched_group_cpus that can actually arrive at this balance domain. The current code gets it wrong, as can be readily demonstrated with a topology like: node 0 1 2 3 0: 10 20 30 20 1: 20 10 20 30 2: 30 20 10 20 3: 20 30 20 10 Where (for example) domain 1 on CPU1 ends up with a mask that includes CPU0: [] CPU1 attaching sched-domain: [] domain 0: span 0-2 level NUMA [] groups: 1 (mask: 1), 2, 0 [] domain 1: span 0-3 level NUMA [] groups: 0-2 (mask: 0-2) (cpu_capacity: 3072), 0,2-3 (cpu_capacity: 3072) This causes sched_balance_cpu() to compute the wrong CPU and consequently should_we_balance() will terminate early resulting in missed load-balance opportunities. The fixed topology looks like: [] CPU1 attaching sched-domain: [] domain 0: span 0-2 level NUMA [] groups: 1 (mask: 1), 2, 0 [] domain 1: span 0-3 level NUMA [] groups: 0-2 (mask: 1) (cpu_capacity: 3072), 0,2-3 (cpu_capacity: 3072) (note: this relies on OVERLAP domains to always have children, this is true because the regular topology domains are still here -- this is before degenerate trimming) Debugged-by: Lauro Ramos Venancio <lvenanci@redhat.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-kernel@vger.kernel.org Cc: stable@vger.kernel.org Fixes: e3589f6c81e4 ("sched: Allow for overlapping sched_domain spans") Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-04-25 20:00:49 +08:00
/*
* Can happen in the asymmetric case, where these siblings are
* unused. The mask will not be empty because those CPUs that
* do have the top domain _should_ span the domain.
*/
if (!sibling->child)
continue;
/* If we would not end up here, we can't continue from here */
if (!cpumask_equal(sg_span, sched_domain_span(sibling->child)))
continue;
sched/topology: Fix overlapping sched_group_capacity When building the overlapping groups we need to attach a consistent sched_group_capacity structure. That is, all 'identical' sched_group's should have the _same_ sched_group_capacity. This can (once again) be demonstrated with a topology like: node 0 1 2 3 0: 10 20 30 20 1: 20 10 20 30 2: 30 20 10 20 3: 20 30 20 10 But we need at least 2 CPUs per node for this to show up, after all, if there is only one CPU per node, our CPU @i is per definition a unique CPU that reaches this domain (aka balance-cpu). Given the above NUMA topo and 2 CPUs per node: [] CPU0 attaching sched-domain(s): [] domain-0: span=0,4 level=DIE [] groups: 0:{ span=0 }, 4:{ span=4 } [] domain-1: span=0-1,3-5,7 level=NUMA [] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 } [] CPU1 attaching sched-domain(s): [] domain-0: span=1,5 level=DIE [] groups: 1:{ span=1 }, 5:{ span=5 } [] domain-1: span=0-2,4-6 level=NUMA [] groups: 1:{ span=1,5 mask=1,5 cap=2048 }, 2:{ span=2,6 mask=2,6 cap=2048 }, 4:{ span=0,4 mask=0,4 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 1:{ span=0-2,4-6 mask=1,5 cap=6144 }, 3:{ span=0,2-4,6-7 mask=3,7 cap=6144 } Observe how CPU0-domain1-group0 and CPU1-domain1-group4 are the 'same' but have a different id (0 vs 4). To fix this, use the group balance CPU to select the SGC. This means we have to compute the full mask for each CPU and require a second temporary mask to store the group mask in (it otherwise lives in the SGC). The fixed topology looks like: [] CPU0 attaching sched-domain(s): [] domain-0: span=0,4 level=DIE [] groups: 0:{ span=0 }, 4:{ span=4 } [] domain-1: span=0-1,3-5,7 level=NUMA [] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 } [] CPU1 attaching sched-domain(s): [] domain-0: span=1,5 level=DIE [] groups: 1:{ span=1 }, 5:{ span=5 } [] domain-1: span=0-2,4-6 level=NUMA [] groups: 1:{ span=1,5 mask=1,5 cap=2048 }, 2:{ span=2,6 mask=2,6 cap=2048 }, 0:{ span=0,4 mask=0,4 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 1:{ span=0-2,4-6 mask=1,5 cap=6144 }, 3:{ span=0,2-4,6-7 mask=3,7 cap=6144 } Debugged-by: Lauro Ramos Venancio <lvenanci@redhat.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-kernel@vger.kernel.org Fixes: e3589f6c81e4 ("sched: Allow for overlapping sched_domain spans") Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-04-25 20:31:11 +08:00
cpumask_set_cpu(i, mask);
}
sched/topology: Fix overlapping sched_group_mask The point of sched_group_mask is to select those CPUs from sched_group_cpus that can actually arrive at this balance domain. The current code gets it wrong, as can be readily demonstrated with a topology like: node 0 1 2 3 0: 10 20 30 20 1: 20 10 20 30 2: 30 20 10 20 3: 20 30 20 10 Where (for example) domain 1 on CPU1 ends up with a mask that includes CPU0: [] CPU1 attaching sched-domain: [] domain 0: span 0-2 level NUMA [] groups: 1 (mask: 1), 2, 0 [] domain 1: span 0-3 level NUMA [] groups: 0-2 (mask: 0-2) (cpu_capacity: 3072), 0,2-3 (cpu_capacity: 3072) This causes sched_balance_cpu() to compute the wrong CPU and consequently should_we_balance() will terminate early resulting in missed load-balance opportunities. The fixed topology looks like: [] CPU1 attaching sched-domain: [] domain 0: span 0-2 level NUMA [] groups: 1 (mask: 1), 2, 0 [] domain 1: span 0-3 level NUMA [] groups: 0-2 (mask: 1) (cpu_capacity: 3072), 0,2-3 (cpu_capacity: 3072) (note: this relies on OVERLAP domains to always have children, this is true because the regular topology domains are still here -- this is before degenerate trimming) Debugged-by: Lauro Ramos Venancio <lvenanci@redhat.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-kernel@vger.kernel.org Cc: stable@vger.kernel.org Fixes: e3589f6c81e4 ("sched: Allow for overlapping sched_domain spans") Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-04-25 20:00:49 +08:00
/* We must not have empty masks here */
sched/topology: Fix overlapping sched_group_capacity When building the overlapping groups we need to attach a consistent sched_group_capacity structure. That is, all 'identical' sched_group's should have the _same_ sched_group_capacity. This can (once again) be demonstrated with a topology like: node 0 1 2 3 0: 10 20 30 20 1: 20 10 20 30 2: 30 20 10 20 3: 20 30 20 10 But we need at least 2 CPUs per node for this to show up, after all, if there is only one CPU per node, our CPU @i is per definition a unique CPU that reaches this domain (aka balance-cpu). Given the above NUMA topo and 2 CPUs per node: [] CPU0 attaching sched-domain(s): [] domain-0: span=0,4 level=DIE [] groups: 0:{ span=0 }, 4:{ span=4 } [] domain-1: span=0-1,3-5,7 level=NUMA [] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 } [] CPU1 attaching sched-domain(s): [] domain-0: span=1,5 level=DIE [] groups: 1:{ span=1 }, 5:{ span=5 } [] domain-1: span=0-2,4-6 level=NUMA [] groups: 1:{ span=1,5 mask=1,5 cap=2048 }, 2:{ span=2,6 mask=2,6 cap=2048 }, 4:{ span=0,4 mask=0,4 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 1:{ span=0-2,4-6 mask=1,5 cap=6144 }, 3:{ span=0,2-4,6-7 mask=3,7 cap=6144 } Observe how CPU0-domain1-group0 and CPU1-domain1-group4 are the 'same' but have a different id (0 vs 4). To fix this, use the group balance CPU to select the SGC. This means we have to compute the full mask for each CPU and require a second temporary mask to store the group mask in (it otherwise lives in the SGC). The fixed topology looks like: [] CPU0 attaching sched-domain(s): [] domain-0: span=0,4 level=DIE [] groups: 0:{ span=0 }, 4:{ span=4 } [] domain-1: span=0-1,3-5,7 level=NUMA [] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 } [] CPU1 attaching sched-domain(s): [] domain-0: span=1,5 level=DIE [] groups: 1:{ span=1 }, 5:{ span=5 } [] domain-1: span=0-2,4-6 level=NUMA [] groups: 1:{ span=1,5 mask=1,5 cap=2048 }, 2:{ span=2,6 mask=2,6 cap=2048 }, 0:{ span=0,4 mask=0,4 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 1:{ span=0-2,4-6 mask=1,5 cap=6144 }, 3:{ span=0,2-4,6-7 mask=3,7 cap=6144 } Debugged-by: Lauro Ramos Venancio <lvenanci@redhat.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-kernel@vger.kernel.org Fixes: e3589f6c81e4 ("sched: Allow for overlapping sched_domain spans") Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-04-25 20:31:11 +08:00
WARN_ON_ONCE(cpumask_empty(mask));
}
/*
* XXX: This creates per-node group entries; since the load-balancer will
* immediately access remote memory to construct this group's load-balance
* statistics having the groups node local is of dubious benefit.
*/
static struct sched_group *
build_group_from_child_sched_domain(struct sched_domain *sd, int cpu)
{
struct sched_group *sg;
struct cpumask *sg_span;
sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(),
GFP_KERNEL, cpu_to_node(cpu));
if (!sg)
return NULL;
sg_span = sched_group_span(sg);
if (sd->child)
cpumask_copy(sg_span, sched_domain_span(sd->child));
else
cpumask_copy(sg_span, sched_domain_span(sd));
sched/topology: Fix memory leak in __sdt_alloc() Found this issue by kmemleak: the 'sg' and 'sgc' pointers from __sdt_alloc() might be leaked as each domain holds many groups' ref, but in destroy_sched_domain(), it only declined the first group ref. Onlining and offlining a CPU can trigger this leak, and cause OOM. Reproducer for my 6 CPUs machine: while true do echo 0 > /sys/devices/system/cpu/cpu5/online; echo 1 > /sys/devices/system/cpu/cpu5/online; done unreferenced object 0xffff88007d772a80 (size 64): comm "cpuhp/5", pid 39, jiffies 4294719962 (age 35.251s) hex dump (first 32 bytes): c0 22 77 7d 00 88 ff ff 02 00 00 00 01 00 00 00 ."w}............ 40 2a 77 7d 00 88 ff ff 00 00 00 00 00 00 00 00 @*w}............ backtrace: [<ffffffff8176525a>] kmemleak_alloc+0x4a/0xa0 [<ffffffff8121efe1>] __kmalloc_node+0xf1/0x280 [<ffffffff810d94a8>] build_sched_domains+0x1e8/0xf20 [<ffffffff810da674>] partition_sched_domains+0x304/0x360 [<ffffffff81139557>] cpuset_update_active_cpus+0x17/0x40 [<ffffffff810bdb2e>] sched_cpu_activate+0xae/0xc0 [<ffffffff810900e0>] cpuhp_invoke_callback+0x90/0x400 [<ffffffff81090597>] cpuhp_up_callbacks+0x37/0xb0 [<ffffffff81090887>] cpuhp_thread_fun+0xd7/0xf0 [<ffffffff810b37e0>] smpboot_thread_fn+0x110/0x160 [<ffffffff810af5d9>] kthread+0x109/0x140 [<ffffffff81770e45>] ret_from_fork+0x25/0x30 [<ffffffffffffffff>] 0xffffffffffffffff unreferenced object 0xffff88007d772a40 (size 64): comm "cpuhp/5", pid 39, jiffies 4294719962 (age 35.251s) hex dump (first 32 bytes): 03 00 00 00 00 00 00 00 00 04 00 00 00 00 00 00 ................ 00 04 00 00 00 00 00 00 4f 3c fc ff 00 00 00 00 ........O<...... backtrace: [<ffffffff8176525a>] kmemleak_alloc+0x4a/0xa0 [<ffffffff8121efe1>] __kmalloc_node+0xf1/0x280 [<ffffffff810da16d>] build_sched_domains+0xead/0xf20 [<ffffffff810da674>] partition_sched_domains+0x304/0x360 [<ffffffff81139557>] cpuset_update_active_cpus+0x17/0x40 [<ffffffff810bdb2e>] sched_cpu_activate+0xae/0xc0 [<ffffffff810900e0>] cpuhp_invoke_callback+0x90/0x400 [<ffffffff81090597>] cpuhp_up_callbacks+0x37/0xb0 [<ffffffff81090887>] cpuhp_thread_fun+0xd7/0xf0 [<ffffffff810b37e0>] smpboot_thread_fn+0x110/0x160 [<ffffffff810af5d9>] kthread+0x109/0x140 [<ffffffff81770e45>] ret_from_fork+0x25/0x30 [<ffffffffffffffff>] 0xffffffffffffffff Reported-by: Chunyu Hu <chuhu@redhat.com> Signed-off-by: Shu Wang <shuwang@redhat.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Acked-by: Chunyu Hu <chuhu@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: liwang@redhat.com Link: http://lkml.kernel.org/r/1502351536-9108-1-git-send-email-shuwang@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-08-10 15:52:16 +08:00
atomic_inc(&sg->ref);
return sg;
}
static void init_overlap_sched_group(struct sched_domain *sd,
sched/topology: Fix overlapping sched_group_capacity When building the overlapping groups we need to attach a consistent sched_group_capacity structure. That is, all 'identical' sched_group's should have the _same_ sched_group_capacity. This can (once again) be demonstrated with a topology like: node 0 1 2 3 0: 10 20 30 20 1: 20 10 20 30 2: 30 20 10 20 3: 20 30 20 10 But we need at least 2 CPUs per node for this to show up, after all, if there is only one CPU per node, our CPU @i is per definition a unique CPU that reaches this domain (aka balance-cpu). Given the above NUMA topo and 2 CPUs per node: [] CPU0 attaching sched-domain(s): [] domain-0: span=0,4 level=DIE [] groups: 0:{ span=0 }, 4:{ span=4 } [] domain-1: span=0-1,3-5,7 level=NUMA [] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 } [] CPU1 attaching sched-domain(s): [] domain-0: span=1,5 level=DIE [] groups: 1:{ span=1 }, 5:{ span=5 } [] domain-1: span=0-2,4-6 level=NUMA [] groups: 1:{ span=1,5 mask=1,5 cap=2048 }, 2:{ span=2,6 mask=2,6 cap=2048 }, 4:{ span=0,4 mask=0,4 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 1:{ span=0-2,4-6 mask=1,5 cap=6144 }, 3:{ span=0,2-4,6-7 mask=3,7 cap=6144 } Observe how CPU0-domain1-group0 and CPU1-domain1-group4 are the 'same' but have a different id (0 vs 4). To fix this, use the group balance CPU to select the SGC. This means we have to compute the full mask for each CPU and require a second temporary mask to store the group mask in (it otherwise lives in the SGC). The fixed topology looks like: [] CPU0 attaching sched-domain(s): [] domain-0: span=0,4 level=DIE [] groups: 0:{ span=0 }, 4:{ span=4 } [] domain-1: span=0-1,3-5,7 level=NUMA [] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 } [] CPU1 attaching sched-domain(s): [] domain-0: span=1,5 level=DIE [] groups: 1:{ span=1 }, 5:{ span=5 } [] domain-1: span=0-2,4-6 level=NUMA [] groups: 1:{ span=1,5 mask=1,5 cap=2048 }, 2:{ span=2,6 mask=2,6 cap=2048 }, 0:{ span=0,4 mask=0,4 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 1:{ span=0-2,4-6 mask=1,5 cap=6144 }, 3:{ span=0,2-4,6-7 mask=3,7 cap=6144 } Debugged-by: Lauro Ramos Venancio <lvenanci@redhat.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-kernel@vger.kernel.org Fixes: e3589f6c81e4 ("sched: Allow for overlapping sched_domain spans") Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-04-25 20:31:11 +08:00
struct sched_group *sg)
{
sched/topology: Fix overlapping sched_group_capacity When building the overlapping groups we need to attach a consistent sched_group_capacity structure. That is, all 'identical' sched_group's should have the _same_ sched_group_capacity. This can (once again) be demonstrated with a topology like: node 0 1 2 3 0: 10 20 30 20 1: 20 10 20 30 2: 30 20 10 20 3: 20 30 20 10 But we need at least 2 CPUs per node for this to show up, after all, if there is only one CPU per node, our CPU @i is per definition a unique CPU that reaches this domain (aka balance-cpu). Given the above NUMA topo and 2 CPUs per node: [] CPU0 attaching sched-domain(s): [] domain-0: span=0,4 level=DIE [] groups: 0:{ span=0 }, 4:{ span=4 } [] domain-1: span=0-1,3-5,7 level=NUMA [] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 } [] CPU1 attaching sched-domain(s): [] domain-0: span=1,5 level=DIE [] groups: 1:{ span=1 }, 5:{ span=5 } [] domain-1: span=0-2,4-6 level=NUMA [] groups: 1:{ span=1,5 mask=1,5 cap=2048 }, 2:{ span=2,6 mask=2,6 cap=2048 }, 4:{ span=0,4 mask=0,4 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 1:{ span=0-2,4-6 mask=1,5 cap=6144 }, 3:{ span=0,2-4,6-7 mask=3,7 cap=6144 } Observe how CPU0-domain1-group0 and CPU1-domain1-group4 are the 'same' but have a different id (0 vs 4). To fix this, use the group balance CPU to select the SGC. This means we have to compute the full mask for each CPU and require a second temporary mask to store the group mask in (it otherwise lives in the SGC). The fixed topology looks like: [] CPU0 attaching sched-domain(s): [] domain-0: span=0,4 level=DIE [] groups: 0:{ span=0 }, 4:{ span=4 } [] domain-1: span=0-1,3-5,7 level=NUMA [] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 } [] CPU1 attaching sched-domain(s): [] domain-0: span=1,5 level=DIE [] groups: 1:{ span=1 }, 5:{ span=5 } [] domain-1: span=0-2,4-6 level=NUMA [] groups: 1:{ span=1,5 mask=1,5 cap=2048 }, 2:{ span=2,6 mask=2,6 cap=2048 }, 0:{ span=0,4 mask=0,4 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 1:{ span=0-2,4-6 mask=1,5 cap=6144 }, 3:{ span=0,2-4,6-7 mask=3,7 cap=6144 } Debugged-by: Lauro Ramos Venancio <lvenanci@redhat.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-kernel@vger.kernel.org Fixes: e3589f6c81e4 ("sched: Allow for overlapping sched_domain spans") Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-04-25 20:31:11 +08:00
struct cpumask *mask = sched_domains_tmpmask2;
struct sd_data *sdd = sd->private;
struct cpumask *sg_span;
sched/topology: Fix overlapping sched_group_capacity When building the overlapping groups we need to attach a consistent sched_group_capacity structure. That is, all 'identical' sched_group's should have the _same_ sched_group_capacity. This can (once again) be demonstrated with a topology like: node 0 1 2 3 0: 10 20 30 20 1: 20 10 20 30 2: 30 20 10 20 3: 20 30 20 10 But we need at least 2 CPUs per node for this to show up, after all, if there is only one CPU per node, our CPU @i is per definition a unique CPU that reaches this domain (aka balance-cpu). Given the above NUMA topo and 2 CPUs per node: [] CPU0 attaching sched-domain(s): [] domain-0: span=0,4 level=DIE [] groups: 0:{ span=0 }, 4:{ span=4 } [] domain-1: span=0-1,3-5,7 level=NUMA [] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 } [] CPU1 attaching sched-domain(s): [] domain-0: span=1,5 level=DIE [] groups: 1:{ span=1 }, 5:{ span=5 } [] domain-1: span=0-2,4-6 level=NUMA [] groups: 1:{ span=1,5 mask=1,5 cap=2048 }, 2:{ span=2,6 mask=2,6 cap=2048 }, 4:{ span=0,4 mask=0,4 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 1:{ span=0-2,4-6 mask=1,5 cap=6144 }, 3:{ span=0,2-4,6-7 mask=3,7 cap=6144 } Observe how CPU0-domain1-group0 and CPU1-domain1-group4 are the 'same' but have a different id (0 vs 4). To fix this, use the group balance CPU to select the SGC. This means we have to compute the full mask for each CPU and require a second temporary mask to store the group mask in (it otherwise lives in the SGC). The fixed topology looks like: [] CPU0 attaching sched-domain(s): [] domain-0: span=0,4 level=DIE [] groups: 0:{ span=0 }, 4:{ span=4 } [] domain-1: span=0-1,3-5,7 level=NUMA [] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 } [] CPU1 attaching sched-domain(s): [] domain-0: span=1,5 level=DIE [] groups: 1:{ span=1 }, 5:{ span=5 } [] domain-1: span=0-2,4-6 level=NUMA [] groups: 1:{ span=1,5 mask=1,5 cap=2048 }, 2:{ span=2,6 mask=2,6 cap=2048 }, 0:{ span=0,4 mask=0,4 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 1:{ span=0-2,4-6 mask=1,5 cap=6144 }, 3:{ span=0,2-4,6-7 mask=3,7 cap=6144 } Debugged-by: Lauro Ramos Venancio <lvenanci@redhat.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-kernel@vger.kernel.org Fixes: e3589f6c81e4 ("sched: Allow for overlapping sched_domain spans") Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-04-25 20:31:11 +08:00
int cpu;
build_balance_mask(sd, sg, mask);
cpu = cpumask_first_and(sched_group_span(sg), mask);
sg->sgc = *per_cpu_ptr(sdd->sgc, cpu);
if (atomic_inc_return(&sg->sgc->ref) == 1)
cpumask_copy(group_balance_mask(sg), mask);
else
WARN_ON_ONCE(!cpumask_equal(group_balance_mask(sg), mask));
/*
* Initialize sgc->capacity such that even if we mess up the
* domains and no possible iteration will get us here, we won't
* die on a /0 trap.
*/
sg_span = sched_group_span(sg);
sg->sgc->capacity = SCHED_CAPACITY_SCALE * cpumask_weight(sg_span);
sg->sgc->min_capacity = SCHED_CAPACITY_SCALE;
}
static int
build_overlap_sched_groups(struct sched_domain *sd, int cpu)
{
struct sched_group *first = NULL, *last = NULL, *sg;
const struct cpumask *span = sched_domain_span(sd);
struct cpumask *covered = sched_domains_tmpmask;
struct sd_data *sdd = sd->private;
struct sched_domain *sibling;
int i;
cpumask_clear(covered);
for_each_cpu_wrap(i, span, cpu) {
struct cpumask *sg_span;
if (cpumask_test_cpu(i, covered))
continue;
sibling = *per_cpu_ptr(sdd->sd, i);
/*
* Asymmetric node setups can result in situations where the
* domain tree is of unequal depth, make sure to skip domains
* that already cover the entire range.
*
* In that case build_sched_domains() will have terminated the
* iteration early and our sibling sd spans will be empty.
* Domains should always include the CPU they're built on, so
* check that.
*/
if (!cpumask_test_cpu(i, sched_domain_span(sibling)))
continue;
sg = build_group_from_child_sched_domain(sibling, cpu);
if (!sg)
goto fail;
sg_span = sched_group_span(sg);
cpumask_or(covered, covered, sg_span);
sched/topology: Fix overlapping sched_group_capacity When building the overlapping groups we need to attach a consistent sched_group_capacity structure. That is, all 'identical' sched_group's should have the _same_ sched_group_capacity. This can (once again) be demonstrated with a topology like: node 0 1 2 3 0: 10 20 30 20 1: 20 10 20 30 2: 30 20 10 20 3: 20 30 20 10 But we need at least 2 CPUs per node for this to show up, after all, if there is only one CPU per node, our CPU @i is per definition a unique CPU that reaches this domain (aka balance-cpu). Given the above NUMA topo and 2 CPUs per node: [] CPU0 attaching sched-domain(s): [] domain-0: span=0,4 level=DIE [] groups: 0:{ span=0 }, 4:{ span=4 } [] domain-1: span=0-1,3-5,7 level=NUMA [] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 } [] CPU1 attaching sched-domain(s): [] domain-0: span=1,5 level=DIE [] groups: 1:{ span=1 }, 5:{ span=5 } [] domain-1: span=0-2,4-6 level=NUMA [] groups: 1:{ span=1,5 mask=1,5 cap=2048 }, 2:{ span=2,6 mask=2,6 cap=2048 }, 4:{ span=0,4 mask=0,4 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 1:{ span=0-2,4-6 mask=1,5 cap=6144 }, 3:{ span=0,2-4,6-7 mask=3,7 cap=6144 } Observe how CPU0-domain1-group0 and CPU1-domain1-group4 are the 'same' but have a different id (0 vs 4). To fix this, use the group balance CPU to select the SGC. This means we have to compute the full mask for each CPU and require a second temporary mask to store the group mask in (it otherwise lives in the SGC). The fixed topology looks like: [] CPU0 attaching sched-domain(s): [] domain-0: span=0,4 level=DIE [] groups: 0:{ span=0 }, 4:{ span=4 } [] domain-1: span=0-1,3-5,7 level=NUMA [] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 } [] CPU1 attaching sched-domain(s): [] domain-0: span=1,5 level=DIE [] groups: 1:{ span=1 }, 5:{ span=5 } [] domain-1: span=0-2,4-6 level=NUMA [] groups: 1:{ span=1,5 mask=1,5 cap=2048 }, 2:{ span=2,6 mask=2,6 cap=2048 }, 0:{ span=0,4 mask=0,4 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 1:{ span=0-2,4-6 mask=1,5 cap=6144 }, 3:{ span=0,2-4,6-7 mask=3,7 cap=6144 } Debugged-by: Lauro Ramos Venancio <lvenanci@redhat.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-kernel@vger.kernel.org Fixes: e3589f6c81e4 ("sched: Allow for overlapping sched_domain spans") Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-04-25 20:31:11 +08:00
init_overlap_sched_group(sd, sg);
if (!first)
first = sg;
if (last)
last->next = sg;
last = sg;
last->next = first;
}
sd->groups = first;
return 0;
fail:
free_sched_groups(first, 0);
return -ENOMEM;
}
/*
* Package topology (also see the load-balance blurb in fair.c)
*
* The scheduler builds a tree structure to represent a number of important
* topology features. By default (default_topology[]) these include:
*
* - Simultaneous multithreading (SMT)
* - Multi-Core Cache (MC)
* - Package (DIE)
*
* Where the last one more or less denotes everything up to a NUMA node.
*
* The tree consists of 3 primary data structures:
*
* sched_domain -> sched_group -> sched_group_capacity
* ^ ^ ^ ^
* `-' `-'
*
* The sched_domains are per-cpu and have a two way link (parent & child) and
* denote the ever growing mask of CPUs belonging to that level of topology.
*
* Each sched_domain has a circular (double) linked list of sched_group's, each
* denoting the domains of the level below (or individual CPUs in case of the
* first domain level). The sched_group linked by a sched_domain includes the
* CPU of that sched_domain [*].
*
* Take for instance a 2 threaded, 2 core, 2 cache cluster part:
*
* CPU 0 1 2 3 4 5 6 7
*
* DIE [ ]
* MC [ ] [ ]
* SMT [ ] [ ] [ ] [ ]
*
* - or -
*
* DIE 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
*
* CPU 0 1 2 3 4 5 6 7
*
* One way to think about it is: sched_domain moves you up and down among these
* topology levels, while sched_group moves you sideways through it, at child
* domain granularity.
*
* sched_group_capacity ensures each unique sched_group has shared storage.
*
* There are two related construction problems, both require a CPU that
* 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
* continue balancing at a higher domain.
*
* - The second is the sched_group_capacity; we want all identical groups
* to share a single sched_group_capacity.
*
* Since these topologies are exclusive by construction. That is, its
* impossible for an SMT thread to belong to multiple cores, and cores to
* be part of multiple caches. There is a very clear and unique location
* for each CPU in the hierarchy.
*
* Therefore computing a unique CPU for each group is trivial (the iteration
* mask is redundant and set all 1s; all CPUs in a group will end up at _that_
* group), we can simply pick the first CPU in each group.
*
*
* [*] in other words, the first group of each domain is its child domain.
*/
static struct sched_group *get_group(int cpu, struct sd_data *sdd)
{
struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu);
struct sched_domain *child = sd->child;
struct sched_group *sg;
if (child)
cpu = cpumask_first(sched_domain_span(child));
sg = *per_cpu_ptr(sdd->sg, cpu);
sg->sgc = *per_cpu_ptr(sdd->sgc, cpu);
/* For claim_allocations: */
atomic_inc(&sg->ref);
atomic_inc(&sg->sgc->ref);
if (child) {
cpumask_copy(sched_group_span(sg), sched_domain_span(child));
cpumask_copy(group_balance_mask(sg), sched_group_span(sg));
} else {
cpumask_set_cpu(cpu, sched_group_span(sg));
cpumask_set_cpu(cpu, group_balance_mask(sg));
}
sg->sgc->capacity = SCHED_CAPACITY_SCALE * cpumask_weight(sched_group_span(sg));
sg->sgc->min_capacity = SCHED_CAPACITY_SCALE;
return sg;
}
/*
* build_sched_groups will build a circular linked list of the groups
* covered by the given span, and will set each group's ->cpumask correctly,
* and ->cpu_capacity to 0.
*
* Assumes the sched_domain tree is fully constructed
*/
static int
build_sched_groups(struct sched_domain *sd, int cpu)
{
struct sched_group *first = NULL, *last = NULL;
struct sd_data *sdd = sd->private;
const struct cpumask *span = sched_domain_span(sd);
struct cpumask *covered;
int i;
lockdep_assert_held(&sched_domains_mutex);
covered = sched_domains_tmpmask;
cpumask_clear(covered);
for_each_cpu_wrap(i, span, cpu) {
struct sched_group *sg;
if (cpumask_test_cpu(i, covered))
continue;
sg = get_group(i, sdd);
cpumask_or(covered, covered, sched_group_span(sg));
if (!first)
first = sg;
if (last)
last->next = sg;
last = sg;
}
last->next = first;
sd->groups = first;
return 0;
}
/*
* Initialize sched groups cpu_capacity.
*
* cpu_capacity indicates the capacity of sched group, which is used while
* distributing the load between different sched groups in a sched domain.
* Typically cpu_capacity for all the groups in a sched domain will be same
* unless there are asymmetries in the topology. If there are asymmetries,
* group having more cpu_capacity will pickup more load compared to the
* group having less cpu_capacity.
*/
static void init_sched_groups_capacity(int cpu, struct sched_domain *sd)
{
struct sched_group *sg = sd->groups;
WARN_ON(!sg);
do {
int cpu, max_cpu = -1;
sg->group_weight = cpumask_weight(sched_group_span(sg));
if (!(sd->flags & SD_ASYM_PACKING))
goto next;
for_each_cpu(cpu, sched_group_span(sg)) {
if (max_cpu < 0)
max_cpu = cpu;
else if (sched_asym_prefer(cpu, max_cpu))
max_cpu = cpu;
}
sg->asym_prefer_cpu = max_cpu;
next:
sg = sg->next;
} while (sg != sd->groups);
if (cpu != group_balance_cpu(sg))
return;
update_group_capacity(sd, cpu);
}
/*
* Initializers for schedule domains
* Non-inlined to reduce accumulated stack pressure in build_sched_domains()
*/
static int default_relax_domain_level = -1;
int sched_domain_level_max;
static int __init setup_relax_domain_level(char *str)
{
if (kstrtoint(str, 0, &default_relax_domain_level))
pr_warn("Unable to set relax_domain_level\n");
return 1;
}
__setup("relax_domain_level=", setup_relax_domain_level);
static void set_domain_attribute(struct sched_domain *sd,
struct sched_domain_attr *attr)
{
int request;
if (!attr || attr->relax_domain_level < 0) {
if (default_relax_domain_level < 0)
return;
else
request = default_relax_domain_level;
} else
request = attr->relax_domain_level;
if (request < sd->level) {
/* Turn off idle balance on this domain: */
sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
} else {
/* Turn on idle balance on this domain: */
sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
}
}
static void __sdt_free(const struct cpumask *cpu_map);
static int __sdt_alloc(const struct cpumask *cpu_map);
static void __free_domain_allocs(struct s_data *d, enum s_alloc what,
const struct cpumask *cpu_map)
{
switch (what) {
case sa_rootdomain:
if (!atomic_read(&d->rd->refcount))
free_rootdomain(&d->rd->rcu);
/* Fall through */
case sa_sd:
free_percpu(d->sd);
/* Fall through */
case sa_sd_storage:
__sdt_free(cpu_map);
/* Fall through */
case sa_none:
break;
}
}
static enum s_alloc
__visit_domain_allocation_hell(struct s_data *d, const struct cpumask *cpu_map)
{
memset(d, 0, sizeof(*d));
if (__sdt_alloc(cpu_map))
return sa_sd_storage;
d->sd = alloc_percpu(struct sched_domain *);
if (!d->sd)
return sa_sd_storage;
d->rd = alloc_rootdomain();
if (!d->rd)
return sa_sd;
return sa_rootdomain;
}
/*
* NULL the sd_data elements we've used to build the sched_domain and
* sched_group structure so that the subsequent __free_domain_allocs()
* will not free the data we're using.
*/
static void claim_allocations(int cpu, struct sched_domain *sd)
{
struct sd_data *sdd = sd->private;
WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd);
*per_cpu_ptr(sdd->sd, cpu) = NULL;
if (atomic_read(&(*per_cpu_ptr(sdd->sds, cpu))->ref))
*per_cpu_ptr(sdd->sds, cpu) = NULL;
if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref))
*per_cpu_ptr(sdd->sg, cpu) = NULL;
if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref))
*per_cpu_ptr(sdd->sgc, cpu) = NULL;
}
#ifdef CONFIG_NUMA
static int sched_domains_numa_levels;
enum numa_topology_type sched_numa_topology_type;
static int *sched_domains_numa_distance;
int sched_max_numa_distance;
static struct cpumask ***sched_domains_numa_masks;
static int sched_domains_curr_level;
#endif
/*
* SD_flags allowed in topology descriptions.
*
* These flags are purely descriptive of the topology and do not prescribe
* behaviour. Behaviour is artificial and mapped in the below sd_init()
* function:
*
* SD_SHARE_CPUCAPACITY - describes SMT topologies
* SD_SHARE_PKG_RESOURCES - describes shared caches
* SD_NUMA - describes NUMA topologies
* SD_SHARE_POWERDOMAIN - describes shared power domain
* SD_ASYM_CPUCAPACITY - describes mixed capacity topologies
*
* Odd one out, which beside describing the topology has a quirk also
* prescribes the desired behaviour that goes along with it:
*
* SD_ASYM_PACKING - describes SMT quirks
*/
#define TOPOLOGY_SD_FLAGS \
(SD_SHARE_CPUCAPACITY | \
SD_SHARE_PKG_RESOURCES | \
SD_NUMA | \
SD_ASYM_PACKING | \
SD_ASYM_CPUCAPACITY | \
SD_SHARE_POWERDOMAIN)
static struct sched_domain *
sd_init(struct sched_domain_topology_level *tl,
const struct cpumask *cpu_map,
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;
#ifdef CONFIG_NUMA
/*
* Ugly hack to pass state to sd_numa_mask()...
*/
sched_domains_curr_level = tl->numa_level;
#endif
sd_weight = cpumask_weight(tl->mask(cpu));
if (tl->sd_flags)
sd_flags = (*tl->sd_flags)();
if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS,
"wrong sd_flags in topology description\n"))
sd_flags &= ~TOPOLOGY_SD_FLAGS;
*sd = (struct sched_domain){
.min_interval = sd_weight,
.max_interval = 2*sd_weight,
.busy_factor = 32,
.imbalance_pct = 125,
.cache_nice_tries = 0,
.busy_idx = 0,
.idle_idx = 0,
.newidle_idx = 0,
.wake_idx = 0,
.forkexec_idx = 0,
.flags = 1*SD_LOAD_BALANCE
| 1*SD_BALANCE_NEWIDLE
| 1*SD_BALANCE_EXEC
| 1*SD_BALANCE_FORK
| 0*SD_BALANCE_WAKE
| 1*SD_WAKE_AFFINE
| 0*SD_SHARE_CPUCAPACITY
| 0*SD_SHARE_PKG_RESOURCES
| 0*SD_SERIALIZE
| 0*SD_PREFER_SIBLING
| 0*SD_NUMA
| sd_flags
,
.last_balance = jiffies,
.balance_interval = sd_weight,
.smt_gain = 0,
.max_newidle_lb_cost = 0,
.next_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));
/*
* Convert topological properties into behaviour.
*/
if (sd->flags & SD_ASYM_CPUCAPACITY) {
struct sched_domain *t = sd;
for_each_lower_domain(t)
t->flags |= SD_BALANCE_WAKE;
}
if (sd->flags & SD_SHARE_CPUCAPACITY) {
sd->flags |= SD_PREFER_SIBLING;
sd->imbalance_pct = 110;
sd->smt_gain = 1178; /* ~15% */
} else if (sd->flags & SD_SHARE_PKG_RESOURCES) {
sd->flags |= SD_PREFER_SIBLING;
sd->imbalance_pct = 117;
sd->cache_nice_tries = 1;
sd->busy_idx = 2;
#ifdef CONFIG_NUMA
} else if (sd->flags & SD_NUMA) {
sd->cache_nice_tries = 2;
sd->busy_idx = 3;
sd->idle_idx = 2;
sd->flags |= SD_SERIALIZE;
if (sched_domains_numa_distance[tl->numa_level] > RECLAIM_DISTANCE) {
sd->flags &= ~(SD_BALANCE_EXEC |
SD_BALANCE_FORK |
SD_WAKE_AFFINE);
}
#endif
} else {
sd->flags |= SD_PREFER_SIBLING;
sd->cache_nice_tries = 1;
sd->busy_idx = 2;
sd->idle_idx = 1;
}
/*
* For all levels sharing cache; connect a sched_domain_shared
* instance.
*/
if (sd->flags & SD_SHARE_PKG_RESOURCES) {
sd->shared = *per_cpu_ptr(sdd->sds, sd_id);
atomic_inc(&sd->shared->ref);
atomic_set(&sd->shared->nr_busy_cpus, sd_weight);
}
sd->private = sdd;
return sd;
}
/*
* Topology list, bottom-up.
*/
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_MC
{ cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
#endif
{ cpu_cpu_mask, SD_INIT_NAME(DIE) },
{ NULL, },
};
static struct sched_domain_topology_level *sched_domain_topology =
default_topology;
#define for_each_sd_topology(tl) \
for (tl = sched_domain_topology; tl->mask; tl++)
void set_sched_topology(struct sched_domain_topology_level *tl)
{
if (WARN_ON_ONCE(sched_smp_initialized))
return;
sched_domain_topology = tl;
}
#ifdef CONFIG_NUMA
static const struct cpumask *sd_numa_mask(int cpu)
{
return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)];
}
static void sched_numa_warn(const char *str)
{
static int done = false;
int i,j;
if (done)
return;
done = true;
printk(KERN_WARNING "ERROR: %s\n\n", 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));
printk(KERN_CONT "\n");
}
printk(KERN_WARNING "\n");
}
bool find_numa_distance(int distance)
{
int i;
if (distance == node_distance(0, 0))
return true;
for (i = 0; i < sched_domains_numa_levels; i++) {
if (sched_domains_numa_distance[i] == distance)
return true;
}
return false;
}
/*
* A system can have three types of NUMA topology:
* NUMA_DIRECT: all nodes are directly connected, or not a NUMA system
* NUMA_GLUELESS_MESH: some nodes reachable through intermediary nodes
* NUMA_BACKPLANE: nodes can reach other nodes through a backplane
*
* The difference between a glueless mesh topology and a backplane
* topology lies in whether communication between not directly
* connected nodes goes through intermediary nodes (where programs
* could run), or through backplane controllers. This affects
* placement of programs.
*
* The type of topology can be discerned with the following tests:
* - If the maximum distance between any nodes is 1 hop, the system
* is directly connected.
* - If for two nodes A and B, located N > 1 hops away from each other,
* 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)
{
int a, b, c, n;
n = sched_max_numa_distance;
if (sched_domains_numa_levels <= 1) {
sched_numa_topology_type = NUMA_DIRECT;
return;
}
for_each_online_node(a) {
for_each_online_node(b) {
/* 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) {
if (node_distance(a, c) < n &&
node_distance(b, c) < n) {
sched_numa_topology_type =
NUMA_GLUELESS_MESH;
return;
}
}
sched_numa_topology_type = NUMA_BACKPLANE;
return;
}
}
}
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, GFP_KERNEL);
if (!sched_domains_numa_distance)
return;
sched/topology: Introduce NUMA identity node sched domain On AMD Family17h-based (EPYC) system, a logical NUMA node can contain upto 8 cores (16 threads) with the following topology. ---------------------------- C0 | T0 T1 | || | T0 T1 | C4 --------| || |-------- C1 | T0 T1 | L3 || L3 | T0 T1 | C5 --------| || |-------- C2 | T0 T1 | #0 || #1 | T0 T1 | C6 --------| || |-------- C3 | T0 T1 | || | T0 T1 | C7 ---------------------------- Here, there are 2 last-level (L3) caches per logical NUMA node. A socket can contain upto 4 NUMA nodes, and a system can support upto 2 sockets. With full system configuration, current scheduler creates 4 sched domains: domain0 SMT (span a core) domain1 MC (span a last-level-cache) domain2 NUMA (span a socket: 4 nodes) domain3 NUMA (span a system: 8 nodes) Note that there is no domain to represent cpus spaning a logical NUMA node. With this hierarchy of sched domains, the scheduler does not balance properly in the following cases: Case1: When running 8 tasks, a properly balanced system should schedule a task per logical NUMA node. This is not the case for the current scheduler. Case2: In some cases, threads are scheduled on the same cpu, while other cpus are idle. This results in run-to-run inconsistency. For example: taskset -c 0-7 sysbench --num-threads=8 --test=cpu \ --cpu-max-prime=100000 run Total execution time ranges from 25.1s to 33.5s depending on threads placement, where 25.1s is when all 8 threads are balanced properly on 8 cpus. Introducing NUMA identity node sched domain, which is based on how SRAT/SLIT table define a logical NUMA node. This results in the following hierarchy of sched domains on the same system described above. domain0 SMT (span a core) domain1 MC (span a last-level-cache) domain2 NODE (span a logical NUMA node) domain3 NUMA (span a socket: 4 nodes) domain4 NUMA (span a system: 8 nodes) This fixes the improper load balancing cases mentioned above. Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: bp@suse.de Link: http://lkml.kernel.org/r/1504768805-46716-1-git-send-email-suravee.suthikulpanit@amd.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-09-07 15:20:05 +08:00
/* Includes NUMA identity node at level 0. */
sched_domains_numa_distance[level++] = curr_distance;
sched_domains_numa_levels = level;
/*
* O(nr_nodes^2) deduplicating 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;
/*
* 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");
if (sched_debug() && i && !find_numa_distance(distance))
sched_numa_warn("Node-0 not representative");
}
if (next_distance != curr_distance) {
sched_domains_numa_distance[level++] = next_distance;
sched_domains_numa_levels = level;
curr_distance = next_distance;
} else break;
}
/*
* In case of sched_debug() we verify the above assumption.
*/
if (!sched_debug())
break;
}
if (!level)
return;
/*
sched/topology: Introduce NUMA identity node sched domain On AMD Family17h-based (EPYC) system, a logical NUMA node can contain upto 8 cores (16 threads) with the following topology. ---------------------------- C0 | T0 T1 | || | T0 T1 | C4 --------| || |-------- C1 | T0 T1 | L3 || L3 | T0 T1 | C5 --------| || |-------- C2 | T0 T1 | #0 || #1 | T0 T1 | C6 --------| || |-------- C3 | T0 T1 | || | T0 T1 | C7 ---------------------------- Here, there are 2 last-level (L3) caches per logical NUMA node. A socket can contain upto 4 NUMA nodes, and a system can support upto 2 sockets. With full system configuration, current scheduler creates 4 sched domains: domain0 SMT (span a core) domain1 MC (span a last-level-cache) domain2 NUMA (span a socket: 4 nodes) domain3 NUMA (span a system: 8 nodes) Note that there is no domain to represent cpus spaning a logical NUMA node. With this hierarchy of sched domains, the scheduler does not balance properly in the following cases: Case1: When running 8 tasks, a properly balanced system should schedule a task per logical NUMA node. This is not the case for the current scheduler. Case2: In some cases, threads are scheduled on the same cpu, while other cpus are idle. This results in run-to-run inconsistency. For example: taskset -c 0-7 sysbench --num-threads=8 --test=cpu \ --cpu-max-prime=100000 run Total execution time ranges from 25.1s to 33.5s depending on threads placement, where 25.1s is when all 8 threads are balanced properly on 8 cpus. Introducing NUMA identity node sched domain, which is based on how SRAT/SLIT table define a logical NUMA node. This results in the following hierarchy of sched domains on the same system described above. domain0 SMT (span a core) domain1 MC (span a last-level-cache) domain2 NODE (span a logical NUMA node) domain3 NUMA (span a socket: 4 nodes) domain4 NUMA (span a system: 8 nodes) This fixes the improper load balancing cases mentioned above. Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: bp@suse.de Link: http://lkml.kernel.org/r/1504768805-46716-1-git-send-email-suravee.suthikulpanit@amd.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-09-07 15:20:05 +08:00
* 'level' contains the number of unique distances
*
* The sched_domains_numa_distance[] array includes the actual distance
* numbers.
*/
/*
* 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
* 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.
*/
sched_domains_numa_levels = 0;
sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL);
if (!sched_domains_numa_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])
return;
for (j = 0; j < nr_node_ids; j++) {
struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL);
if (!mask)
return;
sched_domains_numa_masks[i][j] = mask;
for_each_node(k) {
if (node_distance(j, k) > sched_domains_numa_distance[i])
continue;
cpumask_or(mask, mask, cpumask_of_node(k));
}
}
}
/* Compute default topology size */
for (i = 0; sched_domain_topology[i].mask; i++);
tl = kzalloc((i + level + 1) *
sizeof(struct sched_domain_topology_level), GFP_KERNEL);
if (!tl)
return;
/*
* Copy the default topology bits..
*/
for (i = 0; sched_domain_topology[i].mask; i++)
tl[i] = sched_domain_topology[i];
sched/topology: Introduce NUMA identity node sched domain On AMD Family17h-based (EPYC) system, a logical NUMA node can contain upto 8 cores (16 threads) with the following topology. ---------------------------- C0 | T0 T1 | || | T0 T1 | C4 --------| || |-------- C1 | T0 T1 | L3 || L3 | T0 T1 | C5 --------| || |-------- C2 | T0 T1 | #0 || #1 | T0 T1 | C6 --------| || |-------- C3 | T0 T1 | || | T0 T1 | C7 ---------------------------- Here, there are 2 last-level (L3) caches per logical NUMA node. A socket can contain upto 4 NUMA nodes, and a system can support upto 2 sockets. With full system configuration, current scheduler creates 4 sched domains: domain0 SMT (span a core) domain1 MC (span a last-level-cache) domain2 NUMA (span a socket: 4 nodes) domain3 NUMA (span a system: 8 nodes) Note that there is no domain to represent cpus spaning a logical NUMA node. With this hierarchy of sched domains, the scheduler does not balance properly in the following cases: Case1: When running 8 tasks, a properly balanced system should schedule a task per logical NUMA node. This is not the case for the current scheduler. Case2: In some cases, threads are scheduled on the same cpu, while other cpus are idle. This results in run-to-run inconsistency. For example: taskset -c 0-7 sysbench --num-threads=8 --test=cpu \ --cpu-max-prime=100000 run Total execution time ranges from 25.1s to 33.5s depending on threads placement, where 25.1s is when all 8 threads are balanced properly on 8 cpus. Introducing NUMA identity node sched domain, which is based on how SRAT/SLIT table define a logical NUMA node. This results in the following hierarchy of sched domains on the same system described above. domain0 SMT (span a core) domain1 MC (span a last-level-cache) domain2 NODE (span a logical NUMA node) domain3 NUMA (span a socket: 4 nodes) domain4 NUMA (span a system: 8 nodes) This fixes the improper load balancing cases mentioned above. Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: bp@suse.de Link: http://lkml.kernel.org/r/1504768805-46716-1-git-send-email-suravee.suthikulpanit@amd.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-09-07 15:20:05 +08:00
/*
* Add the NUMA identity distance, aka single NODE.
*/
tl[i++] = (struct sched_domain_topology_level){
.mask = sd_numa_mask,
.numa_level = 0,
SD_INIT_NAME(NODE)
};
/*
* .. and append 'j' levels of NUMA goodness.
*/
sched/topology: Introduce NUMA identity node sched domain On AMD Family17h-based (EPYC) system, a logical NUMA node can contain upto 8 cores (16 threads) with the following topology. ---------------------------- C0 | T0 T1 | || | T0 T1 | C4 --------| || |-------- C1 | T0 T1 | L3 || L3 | T0 T1 | C5 --------| || |-------- C2 | T0 T1 | #0 || #1 | T0 T1 | C6 --------| || |-------- C3 | T0 T1 | || | T0 T1 | C7 ---------------------------- Here, there are 2 last-level (L3) caches per logical NUMA node. A socket can contain upto 4 NUMA nodes, and a system can support upto 2 sockets. With full system configuration, current scheduler creates 4 sched domains: domain0 SMT (span a core) domain1 MC (span a last-level-cache) domain2 NUMA (span a socket: 4 nodes) domain3 NUMA (span a system: 8 nodes) Note that there is no domain to represent cpus spaning a logical NUMA node. With this hierarchy of sched domains, the scheduler does not balance properly in the following cases: Case1: When running 8 tasks, a properly balanced system should schedule a task per logical NUMA node. This is not the case for the current scheduler. Case2: In some cases, threads are scheduled on the same cpu, while other cpus are idle. This results in run-to-run inconsistency. For example: taskset -c 0-7 sysbench --num-threads=8 --test=cpu \ --cpu-max-prime=100000 run Total execution time ranges from 25.1s to 33.5s depending on threads placement, where 25.1s is when all 8 threads are balanced properly on 8 cpus. Introducing NUMA identity node sched domain, which is based on how SRAT/SLIT table define a logical NUMA node. This results in the following hierarchy of sched domains on the same system described above. domain0 SMT (span a core) domain1 MC (span a last-level-cache) domain2 NODE (span a logical NUMA node) domain3 NUMA (span a socket: 4 nodes) domain4 NUMA (span a system: 8 nodes) This fixes the improper load balancing cases mentioned above. Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: bp@suse.de Link: http://lkml.kernel.org/r/1504768805-46716-1-git-send-email-suravee.suthikulpanit@amd.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-09-07 15:20:05 +08:00
for (j = 1; j < level; i++, j++) {
tl[i] = (struct sched_domain_topology_level){
.mask = sd_numa_mask,
.sd_flags = cpu_numa_flags,
.flags = SDTL_OVERLAP,
.numa_level = j,
SD_INIT_NAME(NUMA)
};
}
sched_domain_topology = tl;
sched_domains_numa_levels = level;
sched_max_numa_distance = sched_domains_numa_distance[level - 1];
init_numa_topology_type();
}
void sched_domains_numa_masks_set(unsigned int cpu)
{
int node = cpu_to_node(cpu);
int i, j;
for (i = 0; i < sched_domains_numa_levels; i++) {
for (j = 0; j < nr_node_ids; j++) {
if (node_distance(j, node) <= sched_domains_numa_distance[i])
cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]);
}
}
}
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]);
}
}
#endif /* CONFIG_NUMA */
static int __sdt_alloc(const struct cpumask *cpu_map)
{
struct sched_domain_topology_level *tl;
int j;
for_each_sd_topology(tl) {
struct sd_data *sdd = &tl->data;
sdd->sd = alloc_percpu(struct sched_domain *);
if (!sdd->sd)
return -ENOMEM;
sdd->sds = alloc_percpu(struct sched_domain_shared *);
if (!sdd->sds)
return -ENOMEM;
sdd->sg = alloc_percpu(struct sched_group *);
if (!sdd->sg)
return -ENOMEM;
sdd->sgc = alloc_percpu(struct sched_group_capacity *);
if (!sdd->sgc)
return -ENOMEM;
for_each_cpu(j, cpu_map) {
struct sched_domain *sd;
struct sched_domain_shared *sds;
struct sched_group *sg;
struct sched_group_capacity *sgc;
sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(),
GFP_KERNEL, cpu_to_node(j));
if (!sd)
return -ENOMEM;
*per_cpu_ptr(sdd->sd, j) = sd;
sds = kzalloc_node(sizeof(struct sched_domain_shared),
GFP_KERNEL, cpu_to_node(j));
if (!sds)
return -ENOMEM;
*per_cpu_ptr(sdd->sds, j) = sds;
sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(),
GFP_KERNEL, cpu_to_node(j));
if (!sg)
return -ENOMEM;
sg->next = sg;
*per_cpu_ptr(sdd->sg, j) = sg;
sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(),
GFP_KERNEL, cpu_to_node(j));
if (!sgc)
return -ENOMEM;
#ifdef CONFIG_SCHED_DEBUG
sgc->id = j;
#endif
*per_cpu_ptr(sdd->sgc, j) = sgc;
}
}
return 0;
}
static void __sdt_free(const struct cpumask *cpu_map)
{
struct sched_domain_topology_level *tl;
int j;
for_each_sd_topology(tl) {
struct sd_data *sdd = &tl->data;
for_each_cpu(j, cpu_map) {
struct sched_domain *sd;
if (sdd->sd) {
sd = *per_cpu_ptr(sdd->sd, j);
if (sd && (sd->flags & SD_OVERLAP))
free_sched_groups(sd->groups, 0);
kfree(*per_cpu_ptr(sdd->sd, j));
}
if (sdd->sds)
kfree(*per_cpu_ptr(sdd->sds, j));
if (sdd->sg)
kfree(*per_cpu_ptr(sdd->sg, j));
if (sdd->sgc)
kfree(*per_cpu_ptr(sdd->sgc, j));
}
free_percpu(sdd->sd);
sdd->sd = NULL;
free_percpu(sdd->sds);
sdd->sds = NULL;
free_percpu(sdd->sg);
sdd->sg = NULL;
free_percpu(sdd->sgc);
sdd->sgc = NULL;
}
}
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 cpu)
{
struct sched_domain *sd = sd_init(tl, cpu_map, child, cpu);
if (child) {
sd->level = child->level + 1;
sched_domain_level_max = max(sched_domain_level_max, sd->level);
child->parent = sd;
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),
sched_domain_span(child));
}
}
set_domain_attribute(sd, attr);
return sd;
}
/*
* 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;
struct sched_domain *sd;
struct s_data d;
struct rq *rq = NULL;
int i, ret = -ENOMEM;
alloc_state = __visit_domain_allocation_hell(&d, cpu_map);
if (alloc_state != sa_rootdomain)
goto error;
/* Set up domains for CPUs specified by the cpu_map: */
for_each_cpu(i, cpu_map) {
struct sched_domain_topology_level *tl;
sd = NULL;
for_each_sd_topology(tl) {
sd = build_sched_domain(tl, cpu_map, attr, sd, i);
if (tl == sched_domain_topology)
*per_cpu_ptr(d.sd, i) = sd;
if (tl->flags & SDTL_OVERLAP)
sd->flags |= SD_OVERLAP;
if (cpumask_equal(cpu_map, sched_domain_span(sd)))
break;
}
}
/* Build the groups for the domains */
for_each_cpu(i, cpu_map) {
for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) {
sd->span_weight = cpumask_weight(sched_domain_span(sd));
if (sd->flags & SD_OVERLAP) {
if (build_overlap_sched_groups(sd, i))
goto error;
} else {
if (build_sched_groups(sd, i))
goto error;
}
}
}
/* Calculate CPU capacity for physical packages and nodes */
for (i = nr_cpumask_bits-1; i >= 0; i--) {
if (!cpumask_test_cpu(i, cpu_map))
continue;
for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) {
claim_allocations(i, sd);
init_sched_groups_capacity(i, sd);
}
}
/* Attach the domains */
rcu_read_lock();
for_each_cpu(i, cpu_map) {
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);
}
rcu_read_unlock();
if (rq && sched_debug_enabled) {
pr_info("span: %*pbl (max cpu_capacity = %lu)\n",
cpumask_pr_args(cpu_map), rq->rd->max_cpu_capacity);
}
ret = 0;
error:
__free_domain_allocs(&d, alloc_state, cpu_map);
return ret;
}
/* Current sched domains: */
static cpumask_var_t *doms_cur;
/* Number of sched domains in 'doms_cur': */
static int ndoms_cur;
/* Attribues of custom domains in 'doms_cur' */
static struct sched_domain_attr *dattr_cur;
/*
* Special case: If a kmalloc() of a doms_cur partition (array of
* cpumask) fails, then fallback to a single sched domain,
* as determined by the single cpumask fallback_doms.
*/
static cpumask_var_t fallback_doms;
/*
* arch_update_cpu_topology lets virtualized architectures update the
* CPU core maps. It is supposed to return 1 if the topology changed
* or 0 if it stayed the same.
*/
int __weak arch_update_cpu_topology(void)
{
return 0;
}
cpumask_var_t *alloc_sched_domains(unsigned int ndoms)
{
int i;
cpumask_var_t *doms;
doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL);
if (!doms)
return NULL;
for (i = 0; i < ndoms; i++) {
if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) {
free_sched_domains(doms, i);
return NULL;
}
}
return doms;
}
void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms)
{
unsigned int i;
for (i = 0; i < ndoms; i++)
free_cpumask_var(doms[i]);
kfree(doms);
}
/*
* Set up scheduler domains and groups. Callers must hold the hotplug lock.
* 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 err;
zalloc_cpumask_var(&sched_domains_tmpmask, GFP_KERNEL);
sched/topology: Fix overlapping sched_group_capacity When building the overlapping groups we need to attach a consistent sched_group_capacity structure. That is, all 'identical' sched_group's should have the _same_ sched_group_capacity. This can (once again) be demonstrated with a topology like: node 0 1 2 3 0: 10 20 30 20 1: 20 10 20 30 2: 30 20 10 20 3: 20 30 20 10 But we need at least 2 CPUs per node for this to show up, after all, if there is only one CPU per node, our CPU @i is per definition a unique CPU that reaches this domain (aka balance-cpu). Given the above NUMA topo and 2 CPUs per node: [] CPU0 attaching sched-domain(s): [] domain-0: span=0,4 level=DIE [] groups: 0:{ span=0 }, 4:{ span=4 } [] domain-1: span=0-1,3-5,7 level=NUMA [] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 } [] CPU1 attaching sched-domain(s): [] domain-0: span=1,5 level=DIE [] groups: 1:{ span=1 }, 5:{ span=5 } [] domain-1: span=0-2,4-6 level=NUMA [] groups: 1:{ span=1,5 mask=1,5 cap=2048 }, 2:{ span=2,6 mask=2,6 cap=2048 }, 4:{ span=0,4 mask=0,4 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 1:{ span=0-2,4-6 mask=1,5 cap=6144 }, 3:{ span=0,2-4,6-7 mask=3,7 cap=6144 } Observe how CPU0-domain1-group0 and CPU1-domain1-group4 are the 'same' but have a different id (0 vs 4). To fix this, use the group balance CPU to select the SGC. This means we have to compute the full mask for each CPU and require a second temporary mask to store the group mask in (it otherwise lives in the SGC). The fixed topology looks like: [] CPU0 attaching sched-domain(s): [] domain-0: span=0,4 level=DIE [] groups: 0:{ span=0 }, 4:{ span=4 } [] domain-1: span=0-1,3-5,7 level=NUMA [] groups: 0:{ span=0,4 mask=0,4 cap=2048 }, 1:{ span=1,5 mask=1,5 cap=2048 }, 3:{ span=3,7 mask=3,7 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 0:{ span=0-1,3-5,7 mask=0,4 cap=6144 }, 2:{ span=1-3,5-7 mask=2,6 cap=6144 } [] CPU1 attaching sched-domain(s): [] domain-0: span=1,5 level=DIE [] groups: 1:{ span=1 }, 5:{ span=5 } [] domain-1: span=0-2,4-6 level=NUMA [] groups: 1:{ span=1,5 mask=1,5 cap=2048 }, 2:{ span=2,6 mask=2,6 cap=2048 }, 0:{ span=0,4 mask=0,4 cap=2048 } [] domain-2: span=0-7 level=NUMA [] groups: 1:{ span=0-2,4-6 mask=1,5 cap=6144 }, 3:{ span=0,2-4,6-7 mask=3,7 cap=6144 } Debugged-by: Lauro Ramos Venancio <lvenanci@redhat.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-kernel@vger.kernel.org Fixes: e3589f6c81e4 ("sched: Allow for overlapping sched_domain spans") Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-04-25 20:31:11 +08:00
zalloc_cpumask_var(&sched_domains_tmpmask2, GFP_KERNEL);
zalloc_cpumask_var(&fallback_doms, GFP_KERNEL);
arch_update_cpu_topology();
ndoms_cur = 1;
doms_cur = alloc_sched_domains(ndoms_cur);
if (!doms_cur)
doms_cur = &fallback_doms;
cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map);
err = build_sched_domains(doms_cur[0], NULL);
register_sched_domain_sysctl();
return err;
}
/*
* Detach sched domains from a group of CPUs specified in cpu_map
* These CPUs will now be attached to the NULL domain
*/
static void detach_destroy_domains(const struct cpumask *cpu_map)
{
int i;
rcu_read_lock();
for_each_cpu(i, cpu_map)
cpu_attach_domain(NULL, &def_root_domain, i);
rcu_read_unlock();
}
/* handle null as "default" */
static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur,
struct sched_domain_attr *new, int idx_new)
{
struct sched_domain_attr tmp;
/* Fast path: */
if (!new && !cur)
return 1;
tmp = SD_ATTR_INIT;
return !memcmp(cur ? (cur + idx_cur) : &tmp,
new ? (new + idx_new) : &tmp,
sizeof(struct sched_domain_attr));
}
/*
* Partition sched domains as specified by the 'ndoms_new'
* cpumasks in the array doms_new[] of cpumasks. This compares
* doms_new[] to the current sched domain partitioning, doms_cur[].
* It destroys each deleted domain and builds each new domain.
*
* 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'.
* The masks don't intersect (don't overlap.) We should setup one
* sched domain for each mask. CPUs not in any of the cpumasks will
* not be load balanced. If the same cpumask appears both in the
* current 'doms_cur' domains and in the new 'doms_new', we can leave
* it as it is.
*
* The passed in 'doms_new' should be allocated using
* alloc_sched_domains. This routine takes ownership of it and will
* free_sched_domains it when done with it. If the caller failed the
* alloc call, then it can pass in doms_new == NULL && ndoms_new == 1,
* and partition_sched_domains() will fallback to the single partition
* 'fallback_doms', it also forces the domains to be rebuilt.
*
* If doms_new == NULL it will be replaced with cpu_online_mask.
* ndoms_new == 0 is a special case for destroying existing domains,
* and it will not create the default domain.
*
* Call with hotplug lock held
*/
void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[],
struct sched_domain_attr *dattr_new)
{
int i, j, n;
int new_topology;
mutex_lock(&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();
if (!doms_new) {
WARN_ON_ONCE(dattr_new);
n = 0;
doms_new = alloc_sched_domains(1);
if (doms_new) {
n = 1;
cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map);
}
} else {
n = ndoms_new;
}
/* Destroy deleted domains: */
for (i = 0; i < ndoms_cur; i++) {
for (j = 0; j < n && !new_topology; j++) {
if (cpumask_equal(doms_cur[i], doms_new[j])
&& dattrs_equal(dattr_cur, i, dattr_new, j))
goto match1;
}
/* No match - a current sched domain not in new doms_new[] */
detach_destroy_domains(doms_cur[i]);
match1:
;
}
n = ndoms_cur;
if (!doms_new) {
n = 0;
doms_new = &fallback_doms;
cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map);
}
/* Build new domains: */
for (i = 0; i < ndoms_new; i++) {
for (j = 0; j < n && !new_topology; j++) {
if (cpumask_equal(doms_new[i], doms_cur[j])
&& dattrs_equal(dattr_new, i, dattr_cur, j))
goto match2;
}
/* No match - add a new doms_new */
build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL);
match2:
;
}
/* Remember the new sched domains: */
if (doms_cur != &fallback_doms)
free_sched_domains(doms_cur, ndoms_cur);
kfree(dattr_cur);
doms_cur = doms_new;
dattr_cur = dattr_new;
ndoms_cur = ndoms_new;
register_sched_domain_sysctl();
mutex_unlock(&sched_domains_mutex);
}