From 760a2f284f6d4cd3cd3b1685411bbca21c4ad233 Mon Sep 17 00:00:00 2001 From: Gavin Shan Date: Tue, 27 Jun 2023 20:20:09 +1000 Subject: [PATCH 1/6] numa: Validate cluster and NUMA node boundary if required MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit RH-Author: Gavin Shan RH-MergeRequest: 175: hw/arm: Validate CPU cluster and NUMA node boundary for RHEL machines RH-Bugzilla: 2171363 RH-Acked-by: Cornelia Huck RH-Acked-by: Eric Auger RH-Commit: [1/3] 24580064b9a0076ec4d9a916839d85135ac48cd9 Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=2171363 For some architectures like ARM64, multiple CPUs in one cluster can be associated with different NUMA nodes, which is irregular configuration because we shouldn't have this in baremetal environment. The irregular configuration causes Linux guest to misbehave, as the following warning messages indicate. -smp 6,maxcpus=6,sockets=2,clusters=1,cores=3,threads=1 \ -numa node,nodeid=0,cpus=0-1,memdev=ram0 \ -numa node,nodeid=1,cpus=2-3,memdev=ram1 \ -numa node,nodeid=2,cpus=4-5,memdev=ram2 \ ------------[ cut here ]------------ WARNING: CPU: 0 PID: 1 at kernel/sched/topology.c:2271 build_sched_domains+0x284/0x910 Modules linked in: CPU: 0 PID: 1 Comm: swapper/0 Not tainted 5.14.0-268.el9.aarch64 #1 pstate: 00400005 (nzcv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--) pc : build_sched_domains+0x284/0x910 lr : build_sched_domains+0x184/0x910 sp : ffff80000804bd50 x29: ffff80000804bd50 x28: 0000000000000002 x27: 0000000000000000 x26: ffff800009cf9a80 x25: 0000000000000000 x24: ffff800009cbf840 x23: ffff000080325000 x22: ffff0000005df800 x21: ffff80000a4ce508 x20: 0000000000000000 x19: ffff000080324440 x18: 0000000000000014 x17: 00000000388925c0 x16: 000000005386a066 x15: 000000009c10cc2e x14: 00000000000001c0 x13: 0000000000000001 x12: ffff00007fffb1a0 x11: ffff00007fffb180 x10: ffff80000a4ce508 x9 : 0000000000000041 x8 : ffff80000a4ce500 x7 : ffff80000a4cf920 x6 : 0000000000000001 x5 : 0000000000000001 x4 : 0000000000000007 x3 : 0000000000000002 x2 : 0000000000001000 x1 : ffff80000a4cf928 x0 : 0000000000000001 Call trace: build_sched_domains+0x284/0x910 sched_init_domains+0xac/0xe0 sched_init_smp+0x48/0xc8 kernel_init_freeable+0x140/0x1ac kernel_init+0x28/0x140 ret_from_fork+0x10/0x20 Improve the situation to warn when multiple CPUs in one cluster have been associated with different NUMA nodes. However, one NUMA node is allowed to be associated with different clusters. Signed-off-by: Gavin Shan Acked-by: Philippe Mathieu-Daudé Acked-by: Igor Mammedov Message-Id: <20230509002739.18388-2-gshan@redhat.com> Signed-off-by: Paolo Bonzini (cherry picked from commit a494fdb715832000ee9047a549a35aacfea8175e) Signed-off-by: Gavin Shan --- hw/core/machine.c | 42 ++++++++++++++++++++++++++++++++++++++++++ include/hw/boards.h | 1 + 2 files changed, 43 insertions(+) diff --git a/hw/core/machine.c b/hw/core/machine.c index c28702b690..5abdc8c39b 100644 --- a/hw/core/machine.c +++ b/hw/core/machine.c @@ -1496,6 +1496,45 @@ static void machine_numa_finish_cpu_init(MachineState *machine) g_string_free(s, true); } +static void validate_cpu_cluster_to_numa_boundary(MachineState *ms) +{ + MachineClass *mc = MACHINE_GET_CLASS(ms); + NumaState *state = ms->numa_state; + const CPUArchIdList *possible_cpus = mc->possible_cpu_arch_ids(ms); + const CPUArchId *cpus = possible_cpus->cpus; + int i, j; + + if (state->num_nodes <= 1 || possible_cpus->len <= 1) { + return; + } + + /* + * The Linux scheduling domain can't be parsed when the multiple CPUs + * in one cluster have been associated with different NUMA nodes. However, + * it's fine to associate one NUMA node with CPUs in different clusters. + */ + for (i = 0; i < possible_cpus->len; i++) { + for (j = i + 1; j < possible_cpus->len; j++) { + if (cpus[i].props.has_socket_id && + cpus[i].props.has_cluster_id && + cpus[i].props.has_node_id && + cpus[j].props.has_socket_id && + cpus[j].props.has_cluster_id && + cpus[j].props.has_node_id && + cpus[i].props.socket_id == cpus[j].props.socket_id && + cpus[i].props.cluster_id == cpus[j].props.cluster_id && + cpus[i].props.node_id != cpus[j].props.node_id) { + warn_report("CPU-%d and CPU-%d in socket-%" PRId64 "-cluster-%" PRId64 + " have been associated with node-%" PRId64 " and node-%" PRId64 + " respectively. It can cause OSes like Linux to" + " misbehave", i, j, cpus[i].props.socket_id, + cpus[i].props.cluster_id, cpus[i].props.node_id, + cpus[j].props.node_id); + } + } + } +} + MemoryRegion *machine_consume_memdev(MachineState *machine, HostMemoryBackend *backend) { @@ -1581,6 +1620,9 @@ void machine_run_board_init(MachineState *machine, const char *mem_path, Error * numa_complete_configuration(machine); if (machine->numa_state->num_nodes) { machine_numa_finish_cpu_init(machine); + if (machine_class->cpu_cluster_has_numa_boundary) { + validate_cpu_cluster_to_numa_boundary(machine); + } } } diff --git a/include/hw/boards.h b/include/hw/boards.h index 5f08bd7550..3628671228 100644 --- a/include/hw/boards.h +++ b/include/hw/boards.h @@ -275,6 +275,7 @@ struct MachineClass { bool nvdimm_supported; bool numa_mem_supported; bool auto_enable_numa; + bool cpu_cluster_has_numa_boundary; SMPCompatProps smp_props; const char *default_ram_id; -- 2.39.3