net: enetc: avoid deadlock in enetc_tx_onestep_tstamp()
authorVladimir Oltean <vladimir.oltean@nxp.com>
Thu, 12 Jan 2023 10:54:40 +0000 (12:54 +0200)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 1 Feb 2023 07:27:08 +0000 (08:27 +0100)
commit8232e5a84d25a84a5cbda0f241a00793fb6eb608
treedeeb15dd57d50dfd53d5f70d8ac50d62cd84727b
parent95347e41cac62ca5f378a0533fcf564a4c610c8d
net: enetc: avoid deadlock in enetc_tx_onestep_tstamp()

[ Upstream commit 3c463721a73bdb57a913e0d3124677a3758886fc ]

This lockdep splat says it better than I could:

================================
WARNING: inconsistent lock state
6.2.0-rc2-07010-ga9b9500ffaac-dirty #967 Not tainted
--------------------------------
inconsistent {IN-SOFTIRQ-W} -> {SOFTIRQ-ON-W} usage.
kworker/1:3/179 [HC0[0]:SC0[0]:HE1:SE1] takes:
ffff3ec4036ce098 (_xmit_ETHER#2){+.?.}-{3:3}, at: netif_freeze_queues+0x5c/0xc0
{IN-SOFTIRQ-W} state was registered at:
  _raw_spin_lock+0x5c/0xc0
  sch_direct_xmit+0x148/0x37c
  __dev_queue_xmit+0x528/0x111c
  ip6_finish_output2+0x5ec/0xb7c
  ip6_finish_output+0x240/0x3f0
  ip6_output+0x78/0x360
  ndisc_send_skb+0x33c/0x85c
  ndisc_send_rs+0x54/0x12c
  addrconf_rs_timer+0x154/0x260
  call_timer_fn+0xb8/0x3a0
  __run_timers.part.0+0x214/0x26c
  run_timer_softirq+0x3c/0x74
  __do_softirq+0x14c/0x5d8
  ____do_softirq+0x10/0x20
  call_on_irq_stack+0x2c/0x5c
  do_softirq_own_stack+0x1c/0x30
  __irq_exit_rcu+0x168/0x1a0
  irq_exit_rcu+0x10/0x40
  el1_interrupt+0x38/0x64
irq event stamp: 7825
hardirqs last  enabled at (7825): [<ffffdf1f7200cae4>] exit_to_kernel_mode+0x34/0x130
hardirqs last disabled at (7823): [<ffffdf1f708105f0>] __do_softirq+0x550/0x5d8
softirqs last  enabled at (7824): [<ffffdf1f7081050c>] __do_softirq+0x46c/0x5d8
softirqs last disabled at (7811): [<ffffdf1f708166e0>] ____do_softirq+0x10/0x20

other info that might help us debug this:
 Possible unsafe locking scenario:

       CPU0
       ----
  lock(_xmit_ETHER#2);
  <Interrupt>
    lock(_xmit_ETHER#2);

 *** DEADLOCK ***

3 locks held by kworker/1:3/179:
 #0: ffff3ec400004748 ((wq_completion)events){+.+.}-{0:0}, at: process_one_work+0x1f4/0x6c0
 #1: ffff80000a0bbdc8 ((work_completion)(&priv->tx_onestep_tstamp)){+.+.}-{0:0}, at: process_one_work+0x1f4/0x6c0
 #2: ffff3ec4036cd438 (&dev->tx_global_lock){+.+.}-{3:3}, at: netif_tx_lock+0x1c/0x34

Workqueue: events enetc_tx_onestep_tstamp
Call trace:
 print_usage_bug.part.0+0x208/0x22c
 mark_lock+0x7f0/0x8b0
 __lock_acquire+0x7c4/0x1ce0
 lock_acquire.part.0+0xe0/0x220
 lock_acquire+0x68/0x84
 _raw_spin_lock+0x5c/0xc0
 netif_freeze_queues+0x5c/0xc0
 netif_tx_lock+0x24/0x34
 enetc_tx_onestep_tstamp+0x20/0x100
 process_one_work+0x28c/0x6c0
 worker_thread+0x74/0x450
 kthread+0x118/0x11c

but I'll say it anyway: the enetc_tx_onestep_tstamp() work item runs in
process context, therefore with softirqs enabled (i.o.w., it can be
interrupted by a softirq). If we hold the netif_tx_lock() when there is
an interrupt, and the NET_TX softirq then gets scheduled, this will take
the netif_tx_lock() a second time and deadlock the kernel.

To solve this, use netif_tx_lock_bh(), which blocks softirqs from
running.

Fixes: 7294380c5211 ("enetc: support PTP Sync packet one-step timestamping")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Link: https://lore.kernel.org/r/20230112105440.1786799-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
drivers/net/ethernet/freescale/enetc/enetc.c