[BUG] 4.4.x-rt - memcg: refill_stock() use get_cpu_light() has data corruption issue
From: Haiyang HY1 Tan
Date: Tue Nov 21 2017 - 22:23:13 EST
Dear RT experts,
I have a x86 server mainly used as qemu-kvm hypervisor that is installed with linux-RT kernel, the linux kernel and RT patch set are respectively get from:
Linux kernel: https://www.kernel.org/pub/linux/kernel/v4.x/linux-4.4.97.tar.xz
RT patch set: https://www.kernel.org/pub/linux/kernel/projects/rt/4.4/patches-4.4.97-rt110.tar.gz
kernel crash or hung occur occasionally on this server, a typical backtrace is shown blow:
（13:55:23）[167112.371909] BUG: unable to handle kernel NULL pointer dereference at 00000000000003b0
（13:55:23）[167112.371914] IP: [<ffffffff811a6a37>] mem_cgroup_page_lruvec+0x47/0x60
（13:55:23）[167112.371915] PGD 0
（13:55:23）[167112.371916] Oops: 0000 [#1] PREEMPT SMP
（13:55:23）[167112.371938] Modules linked in: xt_mac xt_physdev xt_set ip_set_hash_net ip_set vfio_pci vfio_virqfd ip6table_raw ip6table_mangle iptable_nat nf_nat_ipv4 nf_nat xt_connmark iptable_mangle 8021q garp mrp ebtable_filter ebtables ip6table_filter ip6_tables openvswitch xt_tcpudp xt_multiport xt_conntrack iptable_filter xt_comment xt_CT iptable_raw igb_uio(O) uio intel_rapl iosf_mbi intel_powerclamp coretemp kvm_intel kvm irqbypass crct10dif_pclmul crc32_pclmul ghash_clmulni_intel mxm_wmi mei_me aesni_intel aes_x86_64 glue_helper lrw ablk_helper ipmi_devintf mei lpc_ich mfd_core cryptd sb_edac edac_core ipmi_si ipmi_msghandler acpi_pad shpchp acpi_power_meter wmi tpm_tis vhost_net vhost macvtap macvlan nf_conntrack_ipv6 nf_defrag_ipv6 nf_conntrack_ipv4 nf_defrag_ipv4 ip_tables x_tables raid1 megaraid_sas
（13:55:23）[167112.371942] CPU: 5 PID: 775963 Comm: qemu-kvm Tainted: G O 4.4.70-thinkcloud-nfv #1
（13:55:23）[167112.371943] Hardware name: LENOVO System x3650 M5: -[8871AC1]-/01GR174, BIOS -[TCE124M-2.10]- 06/23/2016
（13:55:23）[167112.371944] task: ffff88022d246a00 ti: ffff88022d3b0000 task.ti: ffff88022d3b0000
（13:55:23）[167112.371946] RIP: 0010:[<ffffffff811a6a37>] [<ffffffff811a6a37>] mem_cgroup_page_lruvec+0x47/0x60
（13:55:23）[167112.371947] RSP: 0018:ffff88022d3b3bc0 EFLAGS: 00010202
（13:55:23）[167112.371947] RAX: 0000000000000340 RBX: 0000000000000001 RCX: 0000000000000000
（13:55:23）[167112.371948] RDX: ffff88114ac02400 RSI: ffff88107fffc000 RDI: 0000000000000006
（13:55:23）[167112.371948] RBP: ffff88022d3b3bc0 R08: 0000000000000001 R09: 0000000000000000
（13:55:23）[167112.371949] R10: ffffea00405de800 R11: 0000000000000000 R12: ffffea0004c14900
（13:55:23）[167112.371949] R13: ffff88103fb50320 R14: ffff88107fffc000 R15: ffff88107fffc000
（13:55:23）[167112.371950] FS: 00007fe24404ac80(0000) GS:ffff88103fb40000(0000) knlGS:0000000000000000
（13:55:23）[167112.371951] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
（13:55:23）[167112.371951] CR2: 00000000000003b0 CR3: 0000000001df1000 CR4: 00000000003426e0
（13:55:23）[167112.371952] Stack:
（13:55:23）[167112.371953] ffff88022d3b3c08 ffffffff81156fed 0000000000000000 ffffffff81157760
（13:55:23）[167112.371954] 0000000000000005 ffff88103fb50520 ffff88022d246a00 00007fe23ed9a000
（13:55:24）[167112.371955] 0000000000000008 ffff88022d3b3c40 ffffffff811581ca 00000000000103a0
（13:55:24）[167112.371955] Call Trace:
（13:55:24）[167112.371964] [<ffffffff81156fed>] pagevec_lru_move_fn+0x8d/0xf0
（13:55:24）[167112.371966] [<ffffffff81157760>] ? __pagevec_lru_add_fn+0x190/0x190
（13:55:24）[167112.371967] [<ffffffff811581ca>] lru_add_drain_cpu+0x8a/0x130
（13:55:24）[167112.371969] [<ffffffff811583ea>] lru_add_drain+0x5a/0x90
（13:55:24）[167112.371971] [<ffffffff81188bdd>] free_pages_and_swap_cache+0x1d/0x90
（13:55:24）[167112.371973] [<ffffffff81173b66>] tlb_flush_mmu_free+0x36/0x60
（13:55:24）[167112.371974] [<ffffffff81175d8d>] unmap_single_vma+0x70d/0x7c0
（13:55:24）[167112.371976] [<ffffffff81176507>] unmap_vmas+0x47/0x90
（13:55:24）[167112.371978] [<ffffffff8117ea88>] exit_mmap+0x98/0x150
（13:55:24）[167112.371982] [<ffffffff8105ea43>] mmput+0x23/0xc0
（13:55:24）[167112.371984] [<ffffffff81064990>] do_exit+0x240/0xbb0
（13:55:24）[167112.371987] [<ffffffff81424eb7>] ? debug_smp_processor_id+0x17/0x20
（13:55:24）[167112.371989] [<ffffffff810620a6>] ? unpin_current_cpu+0x16/0x70
（13:55:24）[167112.371990] [<ffffffff8106538c>] do_group_exit+0x4c/0xc0
（13:55:24）[167112.371991] [<ffffffff81065414>] SyS_exit_group+0x14/0x20
（13:55:24）[167112.371995] [<ffffffff81a9ab2e>] entry_SYSCALL_64_fastpath+0x12/0x71
（13:55:24）[167112.372007] Code: d2 48 89 c1 48 0f 44 15 50 6b da 00 48 c1 e8 38 48 c1 e9 3a 83 e0 03 48 8d 3c 40 48 8d 04 b8 48 c1 e0 05 48 03 84 ca d0 03 00 00 <48> 3b 70 70 75 02 5d c3 48 89 70 70 5d c3 48 8d 86 10 06 00 00
（13:55:24）[167112.372008] RIP [<ffffffff811a6a37>] mem_cgroup_page_lruvec+0x47/0x60
（13:55:24）[167112.372008] RSP <ffff88022d3b3bc0>
（13:55:24）[167112.372008] CR2: 00000000000003b0
（14:05:24）[167112.777349] ---[ end trace 0000000000000002 ]---
I have review the RT patch set and I think the following patch has a bug. It is shown as blow:
Patch: 0250-memcontrol-Prevent-scheduling-while-atomic-in-cgroup
From 9fa927a8204bbb41c887abff7355c205aa32fb20 Mon Sep 17 00:00:00 2001
From: Mike Galbraith <umgwanakikbuti@xxxxxxxxx>
Date: Sat, 21 Jun 2014 10:09:48 +0200
Subject: [PATCH 250/376] memcontrol: Prevent scheduling while atomic in cgroup
code
mm, memcg: make refill_stock() use get_cpu_light()
Nikita reported the following memcg scheduling while atomic bug:
Call Trace:
[e22d5a90] [c0007ea8] show_stack+0x4c/0x168 (unreliable)
[e22d5ad0] [c0618c04] __schedule_bug+0x94/0xb0
[e22d5ae0] [c060b9ec] __schedule+0x530/0x550
[e22d5bf0] [c060bacc] schedule+0x30/0xbc
[e22d5c00] [c060ca24] rt_spin_lock_slowlock+0x180/0x27c
[e22d5c70] [c00b39dc] res_counter_uncharge_until+0x40/0xc4
[e22d5ca0] [c013ca88] drain_stock.isra.20+0x54/0x98
[e22d5cc0] [c01402ac] __mem_cgroup_try_charge+0x2e8/0xbac
[e22d5d70] [c01410d4] mem_cgroup_charge_common+0x3c/0x70
[e22d5d90] [c0117284] __do_fault+0x38c/0x510
[e22d5df0] [c011a5f4] handle_pte_fault+0x98/0x858
[e22d5e50] [c060ed08] do_page_fault+0x42c/0x6fc
[e22d5f40] [c000f5b4] handle_page_fault+0xc/0x80
What happens:
refill_stock()
get_cpu_var()
drain_stock()
res_counter_uncharge()
res_counter_uncharge_until()
spin_lock() <== boom
Fix it by replacing get/put_cpu_var() with get/put_cpu_light().
Reported-by: Nikita Yushchenko <nyushchenko@xxxxxxxxxxxxx>
Signed-off-by: Mike Galbraith <umgwanakikbuti@xxxxxxxxx>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx>
---
mm/memcontrol.c | 7 +++++--
1 file changed, 5 insertions(+), 2 deletions(-)
diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index e4a020497561..1c619267d9da 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -1925,14 +1925,17 @@ static void drain_local_stock(struct work_struct *dummy)
*/
static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
{
- struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
+ struct memcg_stock_pcp *stock;
+ int cpu = get_cpu_light();
+
+ stock = &per_cpu(memcg_stock, cpu);
if (stock->cached != memcg) { /* reset if necessary */
drain_stock(stock);
stock->cached = memcg;
}
stock->nr_pages += nr_pages;
- put_cpu_var(memcg_stock);
+ put_cpu_light();
}
/*
--
2.13.2
@ Mike Galbraith:
@ Sebastian Andrzej Siewior
This patch replaces “get_cpu_var()” with “get_cpu_light()+per_cpu()”. As we know, the “get_cpu_light()” just disables migrate, but the preemption is still on,
that means a higher priority task-A has chance to interrupt a lower priority task-B on the same cpu which is getting ready to do “drain_stock”, if the task-A
invokes refill_stock() by chance, it will access the same “stock” which task-B is manipulating, so after task-B resume, it will drain the cached memcg that task-A
assigned just now.
I plan to remove the patch: “0250-memcontrol-Prevent-scheduling-while-atomic-in-cgroup” in my RT environment, Whether it will introduce potential issue?
Do you have any suggestions or story behind this patch?
Thanks!
haiyang