cpu_hotplug: unmap cpu2node when the cpu is hotremoved

When a cpu is hotpluged, we call acpi_map_cpu2node() in _acpi_map_lsapic() to store the cpu's node. But we don't clear the cpu's node in acpi_unmap_lsapic() when this cpu is hotremove. If the node is also hotremoved, We will get the following messages:

[ 1646.771485] kernel BUG at include/linux/gfp.h:329!
[ 1646.828729] invalid opcode: 0000 [#1] SMP
[ 1646.877872] Modules linked in: ebtable_nat ebtables ipt_MASQUERADE iptable_nat nf_nat xt_CHECKSUM iptable_mangle bridge stp llc sunrpc ipt_REJECT nf_conntrack_ipv4 nf_defrag_ipv4 iptable_filter ip_tables ip6t_REJECT nf_conntrack_ipv6 nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables binfmt_misc dm_mirror dm_region_hash dm_log dm_mod vhost_net macvtap macvlan tun uinput iTCO_wdt iTCO_vendor_support coretemp kvm_intel kvm crc32c_intel microcode pcspkr i2c_i801 i2c_core lpc_ich mfd_core ioatdma e1000e i7core_edac edac_core sg acpi_memhotplug igb dca sd_mod crc_t10dif megaraid_sas mptsas mptscsih mptbase scsi_transport_sas scsi_mod
[ 1647.588773] Pid: 3126, comm: init Not tainted 3.6.0-rc3-tangchen-hostbridge+ #13 FUJITSU-SV PRIMEQUEST 1800E/SB
[ 1647.711545] RIP: 0010:[<ffffffff811bc3fd>]  [<ffffffff811bc3fd>] allocate_slab+0x28d/0x300
[ 1647.810492] RSP: 0018:ffff88078a049cf8  EFLAGS: 00010246
[ 1647.874028] RAX: 0000000000000000 RBX: 0000000000000001 RCX: 0000000000000000
[ 1647.959339] RDX: 0000000000000001 RSI: 0000000000000001 RDI: 0000000000000246
[ 1648.044659] RBP: ffff88078a049d38 R08: 00000000000040d0 R09: 0000000000000001
[ 1648.129953] R10: 0000000000000000 R11: 0000000000000b5f R12: 00000000000052d0
[ 1648.215259] R13: ffff8807c1417300 R14: 0000000000030038 R15: 0000000000000003
[ 1648.300572] FS:  00007fa9b1b44700(0000) GS:ffff8807c3800000(0000) knlGS:0000000000000000
[ 1648.397272] CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
[ 1648.465985] CR2: 00007fa9b09acca0 CR3: 000000078b855000 CR4: 00000000000007e0
[ 1648.551265] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 1648.636565] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
[ 1648.721838] Process init (pid: 3126, threadinfo ffff88078a048000, task ffff8807bb6f2650)
[ 1648.818534] Stack:
[ 1648.842548]  ffff8807c39d7fa0 ffffffff000040d0 00000000000000bb 00000000000080d0
[ 1648.931469]  ffff8807c1417300 ffff8807c39d7fa0 ffff8807c1417300 0000000000000001
[ 1649.020410]  ffff88078a049d88 ffffffff811bc4a0 ffff8807c1410c80 0000000000000000
[ 1649.109464] Call Trace:
[ 1649.138713]  [<ffffffff811bc4a0>] new_slab+0x30/0x1b0
[ 1649.199075]  [<ffffffff811bc978>] __slab_alloc+0x358/0x4c0
[ 1649.264683]  [<ffffffff810b71c0>] ? alloc_fair_sched_group+0xd0/0x1b0
[ 1649.341695]  [<ffffffff811be7d4>] kmem_cache_alloc_node_trace+0xb4/0x1e0
[ 1649.421824]  [<ffffffff8109d188>] ? hrtimer_init+0x48/0x100
[ 1649.488414]  [<ffffffff810b71c0>] ? alloc_fair_sched_group+0xd0/0x1b0
[ 1649.565402]  [<ffffffff810b71c0>] alloc_fair_sched_group+0xd0/0x1b0
[ 1649.640297]  [<ffffffff810a8bce>] sched_create_group+0x3e/0x110
[ 1649.711040]  [<ffffffff810bdbcd>] sched_autogroup_create_attach+0x4d/0x180
[ 1649.793260]  [<ffffffff81089614>] sys_setsid+0xd4/0xf0
[ 1649.854694]  [<ffffffff8167a029>] system_call_fastpath+0x16/0x1b
[ 1649.926483] Code: 89 c4 e9 73 fe ff ff 31 c0 89 de 48 c7 c7 45 de 9e 81 44 89 45 c8 e8 22 05 4b 00 85 db 44 8b 45 c8 0f 89 4f ff ff ff 0f 0b eb fe <0f> 0b 90 eb fd 0f 0b eb fe 89 de 48 c7 c7 45 de 9e 81 31 c0 44
[ 1650.161454] RIP  [<ffffffff811bc3fd>] allocate_slab+0x28d/0x300
[ 1650.232348]  RSP <ffff88078a049cf8>
[ 1650.274029] ---[ end trace adf84c90f3fea3e5 ]---

The reason is that: the cpu's node is not NUMA_NO_NODE, we will call alloc_pages_exact_node() to alloc memory on the node, but the node is offlined.

---
 arch/x86/kernel/acpi/boot.c |    5 +++++
 1 files changed, 5 insertions(+), 0 deletions(-)

index b2297e5..c27c0c6 100644
@@ -691,6 +691,11 @@ EXPORT_SYMBOL(acpi_map_lsapic);
 
 int acpi_unmap_lsapic(int cpu)
 {
+#ifdef CONFIG_ACPI_NUMA
+	set_apicid_to_node(per_cpu(x86_cpu_to_apicid, cpu), NUMA_NO_NODE);
+	numa_clear_node(cpu);
+#endif
+
 	per_cpu(x86_cpu_to_apicid, cpu) = -1;
 	set_cpu_present(cpu, false);
 	num_processors--;

1.7.1

If we don't debug per_cpu maps, the cpu's node is stored in per_cpu variable numa_node. If node is NUMA_NO_NODE, it means the caller want to clear the cpu's node. So we should also call set_cpu_numa_node() in this case.

---
 arch/x86/mm/numa.c |    3 +--
 1 files changed, 1 insertions(+), 2 deletions(-)

index 2d125be..21d02f0 100644
@@ -97,8 +97,7 @@ void __cpuinit numa_set_node(int cpu, int node)
 #endif
 	per_cpu(x86_cpu_to_node_map, cpu) = node;
 
-	if (node != NUMA_NO_NODE)
-		set_cpu_numa_node(cpu, node);
+	set_cpu_numa_node(cpu, node);
 }
 
 void __cpuinit numa_clear_node(int cpu)

1.7.1

The changelog is missing important information.

What is the runtime effect of the patch? In other words, please fully describe the runtime effects of the bug which the patch fixed.

Please always provide this information. It will help others decide which kernel version(s) should be patched, and will help the maintainers of other kernel trees (especially vendor trees) to work out whether they should backport the fix into their kernels.

At 09/05/2012 07:11 AM, Andrew Morton Wrote:

Sorry for later reply. I found this bug when I try to fix a bug by patch 1/2 (The bug is descriptioned in patch 1/2).

Thanks
Wen Congyang