[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CANn89iKXfrfsoYAvbq0Rwx+5FtBgQf0zYU3WAt2v_tpbS59how@mail.gmail.com>
Date: Sat, 16 Feb 2019 10:18:28 -0800
From: Eric Dumazet <edumazet@...gle.com>
To: Kefeng Wang <wangkefeng.wang@...wei.com>
Cc: netdev <netdev@...r.kernel.org>,
Willem de Bruijn <willemb@...gle.com>,
"David S . Miller" <davem@...emloft.net>,
Jay Vosburgh <j.vosburgh@...il.com>,
Veaceslav Falico <vfalico@...il.com>, weiyongjun1@...wei.com
Subject: Re: [RFC PATCH] bonding: use mutex lock in bond_get_stats()
On Fri, Feb 15, 2019 at 9:36 PM Kefeng Wang <wangkefeng.wang@...wei.com> wrote:
>
>
>
> On 2019/2/15 21:57, Eric Dumazet wrote:
> > On Fri, Feb 15, 2019 at 5:37 AM Kefeng Wang <wangkefeng.wang@...wei.com> wrote:
> >>
> >> With CONFIG_DEBUG_SPINLOCK=y, we find following stack,
> >>
> >> BUG: spinlock wrong CPU on CPU#0, ip/16047
> >> lock: 0xffff803f5febc998, .magic: dead4ead, .owner: ip/16047, .owner_cpu: 0
> >> CPU: 1 PID: 16047 Comm: ip Kdump: loaded Tainted: G E 4.19.12.aarch64 #1
> >> Hardware name: Huawei TaiShan 2280 V2/BC82AMDA, BIOS TA BIOS TaiShan 2280 V2 - B900 01/29/2019
> >> Call trace:
> >> dump_backtrace+0x0/0x1c0
> >> show_stack+0x24/0x30
> >> dump_stack+0x90/0xbc
> >> spin_dump+0x84/0xa8
> >> do_raw_spin_unlock+0xf8/0x100
> >> _raw_spin_unlock+0x20/0x30
> >> bond_get_stats+0x110/0x140 [bonding]
> >> rtnl_fill_stats+0x50/0x150
> >> rtnl_fill_ifinfo+0x4d4/0xd18
> >> rtnl_dump_ifinfo+0x200/0x3a8
> >> netlink_dump+0x100/0x2b0
> >> netlink_recvmsg+0x310/0x3e8
> >> sock_recvmsg+0x58/0x68
> >> ___sys_recvmsg+0xd0/0x278
> >> __sys_recvmsg+0x74/0xd0
> >> __arm64_sys_recvmsg+0x2c/0x38
> >> el0_svc_common+0x7c/0x118
> >> el0_svc_handler+0x30/0x40
> >> el0_svc+0x8/0xc
> >>
> >> and then lead to softlockup issue, fix this by using mutex lock instead
> >> of spin lock.
> >>
> >> Signed-off-by: Kefeng Wang <wangkefeng.wang@...wei.com>
> >> ---
> >>
> >> Not sure if this is right fix, please correct me if I'm wrong.
> >>
> >
> > Make sure to also try :
> >
> > cat /proc/net/dev
>
> Yes, no regression with this patch in our test or 'cat /proc/net/dev'
>
You really should have a warning showing up.
Make sure you have : DEBUG_ATOMIC_SLEEP=y (I thought this was
automatically selected with LOCKDEP on, maybe not...)
Please take a look at net/core/net-procfs.c , functions
dev_seq_start(), dev_seq_printf_stats(), dev_seq_show()
We can not use a mutex in a section protected by rcu_read_lock()
Powered by blists - more mailing lists