[PATCH] rtnetlink: add rtnl_lock debug log

[Rocco.Yue]

> > We often encounter system hangs caused by certain processes
> > holding rtnl_lock for a long time. Even if there is a lock
> > detection mechanism in Linux, it is a bit troublesome and
> > affects the system performance. We hope to add a lightweight
> > debugging mechanism for detecting rtnl_lock.
> >
> > Up to now, we have discovered and solved some potential bugs
> > through such debug information of this lightweight rtnl_lock,
> > which is helpful for us.
> >
> > When you say Y for RTNL_LOCK_DEBUG, then the kernel will detect
> > if any function hold rtnl_lock too long and some key information
> > will be printed to help identify the issue point.
> >
> > i.e: from the following logs, we can clear know that the pid=5546
> 
> clearly
> 
> > RfxSender_4 process hold rtnl_lock for a long time, causing the
> 
> holds
> 
> > system hang. And we can also speculate that the delay operation
> 
> to hang
> 

Thanks for your review, Andy.

I Will fix them.

> > may be performed in devinet_ioctl(), resulting in rtnl_lock was
> > not released in time.
> >
> > <6>[  141.151364] ----------- rtnl_print_btrace start -----------
> 
> Can you, please, shrink this to the point?
> 

Will shrink these points.

> > <6>[  141.152079] RfxSender_4[5546][R] hold rtnl_lock more than 2 sec,
> > start time: 139129481562
> > <4>[  141.153114]  rtnl_lock+0x88/0xfc
> > <4>[  141.153523]  devinet_ioctl+0x190/0x1268
> > <4>[  141.154007]  inet_ioctl+0x108/0x1f4
> > <4>[  141.154449]  sock_do_ioctl+0x88/0x200
> > <4>[  141.154911]  sock_ioctl+0x4b0/0x884
> > <4>[  141.155367]  do_vfs_ioctl+0x6b0/0xcc4
> > <4>[  141.155830]  __arm64_sys_ioctl+0xc0/0xec
> > <4>[  141.156326]  el0_svc_common+0x130/0x2c0
> > <4>[  141.156810]  el0_svc_handler+0xd0/0xe0
> > <4>[  141.157283]  el0_svc+0x8/0xc
> > <4>[  141.157646] Call trace:
> > <4>[  141.157956]  dump_backtrace+0x0/0x240
> > <4>[  141.158418]  show_stack+0x18/0x24
> > <4>[  141.158836]  rtnl_print_btrace+0x138/0x1cc
> > <4>[  141.159362]  call_timer_fn+0x120/0x47c
> > <4>[  141.159834]  expire_timers+0x28c/0x420
> > <4>[  141.160306]  __run_timers+0x3d0/0x494
> > <4>[  141.160768]  run_timer_softirq+0x24/0x48
> > <4>[  141.161262]  __do_softirq+0x26c/0x968
> > <4>[  141.161725]  irq_exit+0x1f8/0x2b4
> > <4>[  141.162145]  __handle_domain_irq+0xdc/0x15c
> > <4>[  141.162672]  gic_handle_irq+0xe4/0x188
> > <4>[  141.163144]  el1_irq+0x104/0x200
> > <4>[  141.163559]  __const_udelay+0x118/0x1b0
> > <4>[  141.164044]  devinet_ioctl+0x1a0/0x1268
> > <4>[  141.164527]  inet_ioctl+0x108/0x1f4
> > <4>[  141.164968]  sock_do_ioctl+0x88/0x200
> > <4>[  141.165428]  sock_ioctl+0x4b0/0x884
> > <4>[  141.165868]  do_vfs_ioctl+0x6b0/0xcc4
> > <4>[  141.166330]  __arm64_sys_ioctl+0xc0/0xec
> > <4>[  141.166825]  el0_svc_common+0x130/0x2c0
> > <4>[  141.167308]  el0_svc_handler+0xd0/0xe0
> > <4>[  141.167786]  el0_svc+0x8/0xc
> > <6>[  141.168153] ------------ rtnl_print_btrace end -----------
> >
> > <6>[  147.321389] rtnl_lock is held by [5546] from
> > [139129481562] to [147321378812]
> 
> 
> ...
> 
> > +static struct rtnl_debug_btrace_t rtnl_instance = {
> > +       .task           = NULL,
> > +       .pid            = 0,
> > +       .start_time     = 0,
> > +       .end_time       = 0,
> > +       .nr_entries     = 0,
> 
> static assumes all 0:s, what's the point?
> 

will fix it in the patch v2.

> > +};
> 
> ...
> 
> > +static void rtnl_print_btrace(struct timer_list *unused)
> > +{
> > +       pr_info("----------- %s start -----------\n", __func__);
> > +       pr_info("%s[%d][%c] hold rtnl_lock more than 2 sec, start time: %llu\n",
> > +               rtnl_instance.task->comm,
> > +               rtnl_instance.pid,
> > +               task_state_to_char(rtnl_instance.task),
> > +               rtnl_instance.start_time);
> > +       stack_trace_print(rtnl_instance.addrs, rtnl_instance.nr_entries, 0);
> 
> > +       show_stack(rtnl_instance.task, NULL, KERN_DEBUG);
> 
> Unaligned debug level.
> 

KERN_INFO is expected, I will fix it in patch v2.

> > +       pr_info("------------ %s end -----------\n", __func__);
> 
> Looking into tons of these, perhaps you need to define pr_fmt(). I
> haven't checked if it's already defined, though.
> 

In the files related rtnetlink, pr_fmt() is not defined. I am not sure
if I can define one since there are other error logs in the rtnetlink.c.

In addition, there is "rtnl_lock" field in the log, I usually use it to
retrieve whether rtnl_lock is held for a long time.

If adding pr_fmt() is better, it is ok to me :-)

> > +}
> 
> ...
> 
> > +       if (rtnl_instance.end_time - rtnl_instance.start_time > 2000000000ULL) {
> 
> Perhaps you should use one of the defined constants from time64.h ?
> 

Will have a separate patch for this.

> > +               pr_info("rtnl_lock is held by [%d] from [%llu] to [%llu]\n",
> > +                       rtnl_instance.pid,
> > +                       rtnl_instance.start_time,
> > +                       rtnl_instance.end_time);
> > +       }
> 

Best Regard
Rocco