As long as NUMA diameter > 2, building sched_domain by sibling's child domain will definitely create a sched_domain with sched_group which will span out of the sched_domain:
+------+ +------+ +-------+ +------+ | node | 12 |node | 20 | node | 12 |node | | 0 +---------+1 +--------+ 2 +-------+3 | +------+ +------+ +-------+ +------+
domain0 node0 node1 node2 node3
domain1 node0+1 node0+1 node2+3 node2+3 + domain2 node0+1+2 | group: node0+1 | group:node2+3 <-------------------+
when node2 is added into the domain2 of node0, kernel is using the child domain of node2's domain2, which is domain1(node2+3). Node 3 is outside the span of the domain including node0+1+2.
This will make load_balance() run based on the avg_load in the sched_group spanning out of the sched_domain, and it also makes select_task_rq_fair() pick an idle CPU out of the sched_domain.
Real servers which suffer from this problem include Kunpeng920 and 8-node Sun Fire X4600-M2, at least.
Here we move to use the *child* domain of the *child* domain of node2's domain2 to build the sched_group.
+------+ +------+ +-------+ +------+ | node | 12 |node | 20 | node | 12 |node | | 0 +---------+1 +--------+ 2 +-------+3 | +------+ +------+ +-------+ +------+
domain0 node0 node1 +- node2 node3 | domain1 node0+1 node0+1 | node2+3 node2+3 | domain2 node0+1+2 | group: node0+1 | group:node2 <-------------------+
A tricky thing is that we shouldn't use the sgc of the 1st CPU of node2 for the sched_group generated by grandchild, otherwise, when this cpu becomes the balance_cpu of another sched_group of cpus other than node0, our sched_group generated by grandchild will access the same sgc with the sched_group generated by child of another CPU.
So in init_overlap_sched_group(), sgc's capacity be overwritten: build_balance_mask(sd, sg, mask); cpu = cpumask_first_and(sched_group_span(sg), mask);
sg->sgc = *per_cpu_ptr(sdd->sgc, cpu);
And WARN_ON_ONCE(!cpumask_equal(group_balance_mask(sg), mask)) will also be triggered: static void init_overlap_sched_group(struct sched_domain *sd, struct sched_group *sg) { 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)); }
So here move to use the sgc of the 2nd cpu. For the corner case, if NUMA has only one CPU, we will still trigger this WARN_ON_ONCE. But It is really unlikely to be a real case for one NUMA to have one CPU only.
Tested by the below topology: qemu-system-aarch64 -M virt -nographic \ -smp cpus=8 \ -numa node,cpus=0-1,nodeid=0 \ -numa node,cpus=2-3,nodeid=1 \ -numa node,cpus=4-5,nodeid=2 \ -numa node,cpus=6-7,nodeid=3 \ -numa dist,src=0,dst=1,val=12 \ -numa dist,src=0,dst=2,val=20 \ -numa dist,src=0,dst=3,val=22 \ -numa dist,src=1,dst=2,val=22 \ -numa dist,src=2,dst=3,val=12 \ -numa dist,src=1,dst=3,val=24 \ -m 4G -cpu cortex-a57 -kernel arch/arm64/boot/Image
w/o patch, we get lots of "groups don't span domain->span": [ 0.802139] CPU0 attaching sched-domain(s): [ 0.802193] domain-0: span=0-1 level=MC [ 0.802443] groups: 0:{ span=0 cap=1013 }, 1:{ span=1 cap=979 } [ 0.802693] domain-1: span=0-3 level=NUMA [ 0.802731] groups: 0:{ span=0-1 cap=1992 }, 2:{ span=2-3 cap=1943 } [ 0.802811] domain-2: span=0-5 level=NUMA [ 0.802829] groups: 0:{ span=0-3 cap=3935 }, 4:{ span=4-7 cap=3937 } [ 0.802881] ERROR: groups don't span domain->span [ 0.803058] domain-3: span=0-7 level=NUMA [ 0.803080] groups: 0:{ span=0-5 mask=0-1 cap=5843 }, 6:{ span=4-7 mask=6-7 cap=4077 } [ 0.804055] CPU1 attaching sched-domain(s): [ 0.804072] domain-0: span=0-1 level=MC [ 0.804096] groups: 1:{ span=1 cap=979 }, 0:{ span=0 cap=1013 } [ 0.804152] domain-1: span=0-3 level=NUMA [ 0.804170] groups: 0:{ span=0-1 cap=1992 }, 2:{ span=2-3 cap=1943 } [ 0.804219] domain-2: span=0-5 level=NUMA [ 0.804236] groups: 0:{ span=0-3 cap=3935 }, 4:{ span=4-7 cap=3937 } [ 0.804302] ERROR: groups don't span domain->span [ 0.804520] domain-3: span=0-7 level=NUMA [ 0.804546] groups: 0:{ span=0-5 mask=0-1 cap=5843 }, 6:{ span=4-7 mask=6-7 cap=4077 } [ 0.804677] CPU2 attaching sched-domain(s): [ 0.804687] domain-0: span=2-3 level=MC [ 0.804705] groups: 2:{ span=2 cap=934 }, 3:{ span=3 cap=1009 } [ 0.804754] domain-1: span=0-3 level=NUMA [ 0.804772] groups: 2:{ span=2-3 cap=1943 }, 0:{ span=0-1 cap=1992 } [ 0.804820] domain-2: span=0-5 level=NUMA [ 0.804836] groups: 2:{ span=0-3 mask=2-3 cap=3991 }, 4:{ span=0-1,4-7 mask=4-5 cap=5985 } [ 0.804944] ERROR: groups don't span domain->span [ 0.805108] domain-3: span=0-7 level=NUMA [ 0.805134] groups: 2:{ span=0-5 mask=2-3 cap=5899 }, 6:{ span=0-1,4-7 mask=6-7 cap=6125 } [ 0.805223] CPU3 attaching sched-domain(s): [ 0.805232] domain-0: span=2-3 level=MC [ 0.805249] groups: 3:{ span=3 cap=1009 }, 2:{ span=2 cap=934 } [ 0.805319] domain-1: span=0-3 level=NUMA [ 0.805336] groups: 2:{ span=2-3 cap=1943 }, 0:{ span=0-1 cap=1992 } [ 0.805383] domain-2: span=0-5 level=NUMA [ 0.805399] groups: 2:{ span=0-3 mask=2-3 cap=3991 }, 4:{ span=0-1,4-7 mask=4-5 cap=5985 } [ 0.805458] ERROR: groups don't span domain->span [ 0.805605] domain-3: span=0-7 level=NUMA [ 0.805626] groups: 2:{ span=0-5 mask=2-3 cap=5899 }, 6:{ span=0-1,4-7 mask=6-7 cap=6125 } [ 0.805712] CPU4 attaching sched-domain(s): [ 0.805721] domain-0: span=4-5 level=MC [ 0.805738] groups: 4:{ span=4 cap=984 }, 5:{ span=5 cap=924 } [ 0.805787] domain-1: span=4-7 level=NUMA [ 0.805803] groups: 4:{ span=4-5 cap=1908 }, 6:{ span=6-7 cap=2029 } [ 0.805851] domain-2: span=0-1,4-7 level=NUMA [ 0.805867] groups: 4:{ span=4-7 cap=3937 }, 0:{ span=0-3 cap=3935 } [ 0.805915] ERROR: groups don't span domain->span [ 0.806108] domain-3: span=0-7 level=NUMA [ 0.806130] groups: 4:{ span=0-1,4-7 mask=4-5 cap=5985 }, 2:{ span=0-3 mask=2-3 cap=3991 } [ 0.806214] CPU5 attaching sched-domain(s): [ 0.806222] domain-0: span=4-5 level=MC [ 0.806240] groups: 5:{ span=5 cap=924 }, 4:{ span=4 cap=984 } [ 0.806841] domain-1: span=4-7 level=NUMA [ 0.806866] groups: 4:{ span=4-5 cap=1908 }, 6:{ span=6-7 cap=2029 } [ 0.806934] domain-2: span=0-1,4-7 level=NUMA [ 0.806953] groups: 4:{ span=4-7 cap=3937 }, 0:{ span=0-3 cap=3935 } [ 0.807004] ERROR: groups don't span domain->span [ 0.807312] domain-3: span=0-7 level=NUMA [ 0.807386] groups: 4:{ span=0-1,4-7 mask=4-5 cap=5985 }, 2:{ span=0-3 mask=2-3 cap=3991 } [ 0.807686] CPU6 attaching sched-domain(s): [ 0.807710] domain-0: span=6-7 level=MC [ 0.807750] groups: 6:{ span=6 cap=1017 }, 7:{ span=7 cap=1012 } [ 0.807840] domain-1: span=4-7 level=NUMA [ 0.807870] groups: 6:{ span=6-7 cap=2029 }, 4:{ span=4-5 cap=1908 } [ 0.807952] domain-2: span=0-1,4-7 level=NUMA [ 0.807985] groups: 6:{ span=4-7 mask=6-7 cap=4077 }, 0:{ span=0-5 mask=0-1 cap=5843 } [ 0.808045] ERROR: groups don't span domain->span [ 0.808257] domain-3: span=0-7 level=NUMA [ 0.808571] groups: 6:{ span=0-1,4-7 mask=6-7 cap=6125 }, 2:{ span=0-5 mask=2-3 cap=5899 } [ 0.808848] CPU7 attaching sched-domain(s): [ 0.808860] domain-0: span=6-7 level=MC [ 0.808880] groups: 7:{ span=7 cap=1012 }, 6:{ span=6 cap=1017 } [ 0.808953] domain-1: span=4-7 level=NUMA [ 0.808974] groups: 6:{ span=6-7 cap=2029 }, 4:{ span=4-5 cap=1908 } [ 0.809034] domain-2: span=0-1,4-7 level=NUMA [ 0.809055] groups: 6:{ span=4-7 mask=6-7 cap=4077 }, 0:{ span=0-5 mask=0-1 cap=5843 } [ 0.809128] ERROR: groups don't span domain->span [ 0.810361] domain-3: span=0-7 level=NUMA [ 0.810400] groups: 6:{ span=0-1,4-7 mask=6-7 cap=5961 }, 2:{ span=0-5 mask=2-3 cap=5903 }
w/ patch, we don't get "groups don't span domain->span" any more: [ 0.868907] CPU0 attaching sched-domain(s): [ 0.868962] domain-0: span=0-1 level=MC [ 0.869179] groups: 0:{ span=0 cap=1013 }, 1:{ span=1 cap=983 } [ 0.869405] domain-1: span=0-3 level=NUMA [ 0.869438] groups: 0:{ span=0-1 cap=1996 }, 2:{ span=2-3 cap=2006 } [ 0.869542] domain-2: span=0-5 level=NUMA [ 0.869559] groups: 0:{ span=0-3 cap=4002 }, 5:{ span=4-5 cap=2048 } [ 0.869603] domain-3: span=0-7 level=NUMA [ 0.869618] groups: 0:{ span=0-5 mask=0-1 cap=5980 }, 6:{ span=4-7 mask=6-7 cap=4016 } [ 0.870303] CPU1 attaching sched-domain(s): [ 0.870314] domain-0: span=0-1 level=MC [ 0.870334] groups: 1:{ span=1 cap=983 }, 0:{ span=0 cap=1013 } [ 0.870381] domain-1: span=0-3 level=NUMA [ 0.870396] groups: 0:{ span=0-1 cap=1996 }, 2:{ span=2-3 cap=2006 } [ 0.870440] domain-2: span=0-5 level=NUMA [ 0.870454] groups: 0:{ span=0-3 cap=4002 }, 5:{ span=4-5 cap=2048 } [ 0.870507] domain-3: span=0-7 level=NUMA [ 0.870530] groups: 0:{ span=0-5 mask=0-1 cap=5980 }, 6:{ span=4-7 mask=6-7 cap=4016 } [ 0.870611] CPU2 attaching sched-domain(s): [ 0.870619] domain-0: span=2-3 level=MC [ 0.870634] groups: 2:{ span=2 cap=1007 }, 3:{ span=3 cap=999 } [ 0.870677] domain-1: span=0-3 level=NUMA [ 0.870691] groups: 2:{ span=2-3 cap=2006 }, 0:{ span=0-1 cap=1996 } [ 0.870734] domain-2: span=0-5 level=NUMA [ 0.870748] groups: 2:{ span=0-3 mask=2-3 cap=4054 }, 5:{ span=4-5 cap=2048 } [ 0.870795] domain-3: span=0-7 level=NUMA [ 0.870809] groups: 2:{ span=0-5 mask=2-3 cap=6032 }, 6:{ span=0-1,4-7 mask=6-7 cap=6064 } [ 0.870913] CPU3 attaching sched-domain(s): [ 0.870921] domain-0: span=2-3 level=MC [ 0.870936] groups: 3:{ span=3 cap=999 }, 2:{ span=2 cap=1007 } [ 0.870979] domain-1: span=0-3 level=NUMA [ 0.870993] groups: 2:{ span=2-3 cap=2006 }, 0:{ span=0-1 cap=1996 } [ 0.871035] domain-2: span=0-5 level=NUMA [ 0.871049] groups: 2:{ span=0-3 mask=2-3 cap=4054 }, 5:{ span=4-5 cap=2048 } [ 0.871096] domain-3: span=0-7 level=NUMA [ 0.871110] groups: 2:{ span=0-5 mask=2-3 cap=6032 }, 6:{ span=0-1,4-7 mask=6-7 cap=6064 } [ 0.871177] CPU4 attaching sched-domain(s): [ 0.871185] domain-0: span=4-5 level=MC [ 0.871200] groups: 4:{ span=4 cap=977 }, 5:{ span=5 cap=1001 } [ 0.871243] domain-1: span=4-7 level=NUMA [ 0.871257] groups: 4:{ span=4-5 cap=1978 }, 6:{ span=6-7 cap=1968 } [ 0.871300] domain-2: span=0-1,4-7 level=NUMA [ 0.871314] groups: 4:{ span=4-7 cap=3946 }, 1:{ span=0-1 cap=2048 } [ 0.871356] domain-3: span=0-7 level=NUMA [ 0.871370] groups: 4:{ span=0-1,4-7 mask=4-5 cap=5994 }, 2:{ span=0-3 mask=2-3 cap=4054 } [ 0.871436] CPU5 attaching sched-domain(s): [ 0.871443] domain-0: span=4-5 level=MC [ 0.871457] groups: 5:{ span=5 cap=1001 }, 4:{ span=4 cap=977 } [ 0.871512] domain-1: span=4-7 level=NUMA [ 0.871893] groups: 4:{ span=4-5 cap=1978 }, 6:{ span=6-7 cap=1968 } [ 0.871949] domain-2: span=0-1,4-7 level=NUMA [ 0.871966] groups: 4:{ span=4-7 cap=3946 }, 1:{ span=0-1 cap=2048 } [ 0.872010] domain-3: span=0-7 level=NUMA [ 0.872025] groups: 4:{ span=0-1,4-7 mask=4-5 cap=5994 }, 2:{ span=0-3 mask=2-3 cap=4054 } [ 0.872115] CPU6 attaching sched-domain(s): [ 0.872123] domain-0: span=6-7 level=MC [ 0.872139] groups: 6:{ span=6 cap=993 }, 7:{ span=7 cap=975 } [ 0.872186] domain-1: span=4-7 level=NUMA [ 0.872202] groups: 6:{ span=6-7 cap=1968 }, 4:{ span=4-5 cap=1978 } [ 0.872246] domain-2: span=0-1,4-7 level=NUMA [ 0.872260] groups: 6:{ span=4-7 mask=6-7 cap=4016 }, 1:{ span=0-1 cap=2048 } [ 0.872309] domain-3: span=0-7 level=NUMA [ 0.872323] groups: 6:{ span=0-1,4-7 mask=6-7 cap=6064 }, 2:{ span=0-5 mask=2-3 cap=6032 } [ 0.872392] CPU7 attaching sched-domain(s): [ 0.872399] domain-0: span=6-7 level=MC [ 0.872414] groups: 7:{ span=7 cap=975 }, 6:{ span=6 cap=993 } [ 0.872458] domain-1: span=4-7 level=NUMA [ 0.872472] groups: 6:{ span=6-7 cap=1968 }, 4:{ span=4-5 cap=1978 } [ 0.872662] domain-2: span=0-1,4-7 level=NUMA [ 0.872685] groups: 6:{ span=4-7 mask=6-7 cap=4016 }, 1:{ span=0-1 cap=2048 } [ 0.872737] domain-3: span=0-7 level=NUMA [ 0.872752] groups: 6:{ span=0-1,4-7 mask=6-7 cap=6064 }, 2:{ span=0-5 mask=2-3 cap=6032 }
Reported-by: Valentin Schneider valentin.schneider@arm.com Tested-by: Meelis Roos mroos@linux.ee Signed-off-by: Barry Song song.bao.hua@hisilicon.com --- Differences with RFC v2 * added tested-by Meelis Roos for the fixed "8-node Sun Fire X4600-M2" * removed the hacking code in balance_mask and should_we_balance() * removed the redundant "from_grandchild" field from sched_group
The patch is based on 5.11-rc6;
kernel/sched/topology.c | 83 ++++++++++++++++++++++++++--------------- 1 file changed, 52 insertions(+), 31 deletions(-)
diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c index 5d3675c7a76b..100feb2fd8a0 100644 --- a/kernel/sched/topology.c +++ b/kernel/sched/topology.c @@ -723,35 +723,6 @@ cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) for (tmp = sd; tmp; tmp = tmp->parent) numa_distance += !!(tmp->flags & SD_NUMA);
- /* - * FIXME: Diameter >=3 is misrepresented. - * - * Smallest diameter=3 topology is: - * - * node 0 1 2 3 - * 0: 10 20 30 40 - * 1: 20 10 20 30 - * 2: 30 20 10 20 - * 3: 40 30 20 10 - * - * 0 --- 1 --- 2 --- 3 - * - * NUMA-3 0-3 N/A N/A 0-3 - * groups: {0-2},{1-3} {1-3},{0-2} - * - * NUMA-2 0-2 0-3 0-3 1-3 - * groups: {0-1},{1-3} {0-2},{2-3} {1-3},{0-1} {2-3},{0-2} - * - * NUMA-1 0-1 0-2 1-3 2-3 - * groups: {0},{1} {1},{2},{0} {2},{3},{1} {3},{2} - * - * NUMA-0 0 1 2 3 - * - * The NUMA-2 groups for nodes 0 and 3 are obviously buggered, as the - * group span isn't a subset of the domain span. - */ - WARN_ONCE(numa_distance > 2, "Shortest NUMA path spans too many nodes\n"); - sched_domain_debug(sd, cpu);
rq_attach_root(rq, rd); @@ -916,6 +887,11 @@ build_balance_mask(struct sched_domain *sd, struct sched_group *sg, struct cpuma if (!sibling->child) continue;
+ while (sibling->child && + !cpumask_subset(sched_domain_span(sibling->child), + sched_domain_span(sd))) + sibling = sibling->child; + /* If we would not end up here, we can't continue from here */ if (!cpumask_equal(sg_span, sched_domain_span(sibling->child))) continue; @@ -955,7 +931,8 @@ build_group_from_child_sched_domain(struct sched_domain *sd, int cpu) }
static void init_overlap_sched_group(struct sched_domain *sd, - struct sched_group *sg) + struct sched_group *sg, + int from_grandchild) { struct cpumask *mask = sched_domains_tmpmask2; struct sd_data *sdd = sd->private; @@ -964,6 +941,12 @@ static void init_overlap_sched_group(struct sched_domain *sd,
build_balance_mask(sd, sg, mask); cpu = cpumask_first_and(sched_group_span(sg), mask); + /* + * for the group generated by grandchild, use the sgc of 2nd cpu + * because the 1st cpu might be used by another sched_group + */ + if (from_grandchild && cpumask_weight(mask) > 1) + cpu = cpumask_next_and(cpu, sched_group_span(sg), mask);
sg->sgc = *per_cpu_ptr(sdd->sgc, cpu); if (atomic_inc_return(&sg->sgc->ref) == 1) @@ -996,6 +979,7 @@ build_overlap_sched_groups(struct sched_domain *sd, int cpu)
for_each_cpu_wrap(i, span, cpu) { struct cpumask *sg_span; + int from_grandchild = 0;
if (cpumask_test_cpu(i, covered)) continue; @@ -1015,6 +999,43 @@ build_overlap_sched_groups(struct sched_domain *sd, int cpu) if (!cpumask_test_cpu(i, sched_domain_span(sibling))) continue;
+ /* + * for NUMA diameter >= 3, building sched_domain by sibling's + * child's child domain to prevent sched_group from spanning + * out of sched_domain + * if we don't do this, Diameter >=3 is misrepresented: + * + * Smallest diameter=3 topology is: + * + * node 0 1 2 3 + * 0: 10 20 30 40 + * 1: 20 10 20 30 + * 2: 30 20 10 20 + * 3: 40 30 20 10 + * + * 0 --- 1 --- 2 --- 3 + * + * NUMA-3 0-3 N/A N/A 0-3 + * groups: {0-2},{1-3} {1-3},{0-2} + * + * NUMA-2 0-2 0-3 0-3 1-3 + * groups: {0-1},{1-3} {0-2},{2-3} {1-3},{0-1} {2-3},{0-2} + * + * NUMA-1 0-1 0-2 1-3 2-3 + * groups: {0},{1} {1},{2},{0} {2},{3},{1} {3},{2} + * + * NUMA-0 0 1 2 3 + * + * The NUMA-2 groups for nodes 0 and 3 are obviously buggered, as the + * group span isn't a subset of the domain span. + */ + while (sibling->child && + !cpumask_subset(sched_domain_span(sibling->child), + span)) { + sibling = sibling->child; + from_grandchild = 1; + } + sg = build_group_from_child_sched_domain(sibling, cpu); if (!sg) goto fail; @@ -1022,7 +1043,7 @@ build_overlap_sched_groups(struct sched_domain *sd, int cpu) sg_span = sched_group_span(sg); cpumask_or(covered, covered, sg_span);
- init_overlap_sched_group(sd, sg); + init_overlap_sched_group(sd, sg, from_grandchild);
if (!first) first = sg;