On 09/10/20 14:47, Juri Lelli wrote: > The following BUG has been reported (slightly edited): > > BUG: using smp_processor_id() in preemptible [00000000] code: handler106/3082 > caller is flow_lookup.isra.15+0x2c/0xf0 [openvswitch] > CPU: 46 PID: 3082 Comm: handler106 Not tainted ... #1 > Hardware name: Dell Inc. PowerEdge R640/06DKY5, BIOS 2.5.4 01/13/2020 > Call Trace: > dump_stack+0x5c/0x80 > check_preemption_disabled+0xc4/0xd0 > flow_lookup.isra.15+0x2c/0xf0 [openvswitch] > ovs_flow_tbl_lookup+0x3b/0x60 [openvswitch] > ovs_flow_cmd_new+0x2d8/0x430 [openvswitch] > ? __switch_to_asm+0x35/0x70 > ? __switch_to_asm+0x41/0x70 > ? __switch_to_asm+0x35/0x70 > genl_family_rcv_msg+0x1d7/0x410 > ? migrate_enable+0x123/0x3a0 > genl_rcv_msg+0x47/0x8c > ? __kmalloc_node_track_caller+0xff/0x2e0 > ? genl_family_rcv_msg+0x410/0x410 > netlink_rcv_skb+0x4c/0x120 > genl_rcv+0x24/0x40 > netlink_unicast+0x197/0x230 > netlink_sendmsg+0x204/0x3d0 > sock_sendmsg+0x4c/0x50 > ___sys_sendmsg+0x29f/0x300 > ? migrate_enable+0x123/0x3a0 > ? ep_send_events_proc+0x8a/0x1f0 > ? ep_scan_ready_list.constprop.23+0x237/0x260 > ? rt_spin_unlock+0x23/0x40 > ? ep_poll+0x1b3/0x390 > ? __fget+0x72/0xa0 > __sys_sendmsg+0x57/0xa0 > do_syscall_64+0x87/0x1a0 > entry_SYSCALL_64_after_hwframe+0x65/0xca > RIP: 0033:0x7f1ed72ccb07 > Code: ... > RSP: 002b:00007f1ecbd9ba80 EFLAGS: 00003293 ORIG_RAX: 000000000000002e > RAX: ffffffffffffffda RBX: 000000000000007b RCX: 00007f1ed72ccb07 > RDX: 0000000000000000 RSI: 00007f1ecbd9bb10 RDI: 000000000000007b > RBP: 00007f1ecbd9bb10 R08: 0000000000000000 R09: 00007f1ecbd9d390 > R10: 0000000019616156 R11: 0000000000003293 R12: 0000000000000000 > R13: 00007f1ecbd9d338 R14: 00007f1ecbd9bfb0 R15: 00007f1ecbd9bb10 > > This happens because openvswitch/flow_table::flow_lookup() accesses > per-cpu data while being preemptible (and migratable). > > Fix it by adding get/put_cpu_light(), so that, even if preempted, the > task executing this code is not migrated (operation is also guarded by > ovs_mutex mutex). > This actually wants also a (sorry for missing it in the first place) Suggested-by: Daniel Bristot de Oliveira <bristot@xxxxxxxxxx> > Signed-off-by: Juri Lelli <juri.lelli@xxxxxxxxxx> > --- > net/openvswitch/flow_table.c | 8 +++++++- > 1 file changed, 7 insertions(+), 1 deletion(-) > > diff --git a/net/openvswitch/flow_table.c b/net/openvswitch/flow_table.c > index e2235849a57e..7df27ef7da09 100644 > --- a/net/openvswitch/flow_table.c > +++ b/net/openvswitch/flow_table.c > @@ -732,11 +732,14 @@ static struct sw_flow *flow_lookup(struct flow_table *tbl, > u32 *n_cache_hit, > u32 *index) > { > - u64 *usage_counters = this_cpu_ptr(ma->masks_usage_cntr); > + u64 *usage_counters; > struct sw_flow *flow; > struct sw_flow_mask *mask; > int i; > > + get_cpu_light(); > + usage_counters = this_cpu_ptr(ma->masks_usage_cntr); > + > if (likely(*index < ma->max)) { > mask = rcu_dereference_ovsl(ma->masks[*index]); > if (mask) { > @@ -746,6 +749,7 @@ static struct sw_flow *flow_lookup(struct flow_table *tbl, > usage_counters[*index]++; > u64_stats_update_end(&ma->syncp); > (*n_cache_hit)++; > + put_cpu_light(); > return flow; > } > } > @@ -766,10 +770,12 @@ static struct sw_flow *flow_lookup(struct flow_table *tbl, > u64_stats_update_begin(&ma->syncp); > usage_counters[*index]++; > u64_stats_update_end(&ma->syncp); > + put_cpu_light(); > return flow; > } > } > > + put_cpu_light(); > return NULL; > } > > -- > 2.26.2 >