net: phy: skip LED triggers on PHYs on SFP modules
authorDaniel Golle <daniel@makrotopia.org>
Tue, 12 Dec 2023 00:05:35 +0000 (00:05 +0000)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Mon, 1 Jan 2024 12:42:32 +0000 (12:42 +0000)
commitd98ce1f083cca895982f3e88b8e8c372288e3e23
treed677fa01d6c0311ae1ddfa55be0753c109d7e4c8
parentf0534c0fa81b281bf515a094b854207d8b2b7fe3
net: phy: skip LED triggers on PHYs on SFP modules

[ Upstream commit b1dfc0f76231bbf395c59d20a2070684620d5d0f ]

Calling led_trigger_register() when attaching a PHY located on an SFP
module potentially (and practically) leads into a deadlock.
Fix this by not calling led_trigger_register() for PHYs localted on SFP
modules as such modules actually never got any LEDs.

======================================================
WARNING: possible circular locking dependency detected
6.7.0-rc4-next-20231208+ #0 Tainted: G           O
------------------------------------------------------
kworker/u8:2/43 is trying to acquire lock:
ffffffc08108c4e8 (triggers_list_lock){++++}-{3:3}, at: led_trigger_register+0x4c/0x1a8

but task is already holding lock:
ffffff80c5c6f318 (&sfp->sm_mutex){+.+.}-{3:3}, at: cleanup_module+0x2ba8/0x3120 [sfp]

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #3 (&sfp->sm_mutex){+.+.}-{3:3}:
       __mutex_lock+0x88/0x7a0
       mutex_lock_nested+0x20/0x28
       cleanup_module+0x2ae0/0x3120 [sfp]
       sfp_register_bus+0x5c/0x9c
       sfp_register_socket+0x48/0xd4
       cleanup_module+0x271c/0x3120 [sfp]
       platform_probe+0x64/0xb8
       really_probe+0x17c/0x3c0
       __driver_probe_device+0x78/0x164
       driver_probe_device+0x3c/0xd4
       __driver_attach+0xec/0x1f0
       bus_for_each_dev+0x60/0xa0
       driver_attach+0x20/0x28
       bus_add_driver+0x108/0x208
       driver_register+0x5c/0x118
       __platform_driver_register+0x24/0x2c
       init_module+0x28/0xa7c [sfp]
       do_one_initcall+0x70/0x2ec
       do_init_module+0x54/0x1e4
       load_module+0x1b78/0x1c8c
       __do_sys_init_module+0x1bc/0x2cc
       __arm64_sys_init_module+0x18/0x20
       invoke_syscall.constprop.0+0x4c/0xdc
       do_el0_svc+0x3c/0xbc
       el0_svc+0x34/0x80
       el0t_64_sync_handler+0xf8/0x124
       el0t_64_sync+0x150/0x154

-> #2 (rtnl_mutex){+.+.}-{3:3}:
       __mutex_lock+0x88/0x7a0
       mutex_lock_nested+0x20/0x28
       rtnl_lock+0x18/0x20
       set_device_name+0x30/0x130
       netdev_trig_activate+0x13c/0x1ac
       led_trigger_set+0x118/0x234
       led_trigger_write+0x104/0x17c
       sysfs_kf_bin_write+0x64/0x80
       kernfs_fop_write_iter+0x128/0x1b4
       vfs_write+0x178/0x2a4
       ksys_write+0x58/0xd4
       __arm64_sys_write+0x18/0x20
       invoke_syscall.constprop.0+0x4c/0xdc
       do_el0_svc+0x3c/0xbc
       el0_svc+0x34/0x80
       el0t_64_sync_handler+0xf8/0x124
       el0t_64_sync+0x150/0x154

-> #1 (&led_cdev->trigger_lock){++++}-{3:3}:
       down_write+0x4c/0x13c
       led_trigger_write+0xf8/0x17c
       sysfs_kf_bin_write+0x64/0x80
       kernfs_fop_write_iter+0x128/0x1b4
       vfs_write+0x178/0x2a4
       ksys_write+0x58/0xd4
       __arm64_sys_write+0x18/0x20
       invoke_syscall.constprop.0+0x4c/0xdc
       do_el0_svc+0x3c/0xbc
       el0_svc+0x34/0x80
       el0t_64_sync_handler+0xf8/0x124
       el0t_64_sync+0x150/0x154

-> #0 (triggers_list_lock){++++}-{3:3}:
       __lock_acquire+0x12a0/0x2014
       lock_acquire+0x100/0x2ac
       down_write+0x4c/0x13c
       led_trigger_register+0x4c/0x1a8
       phy_led_triggers_register+0x9c/0x214
       phy_attach_direct+0x154/0x36c
       phylink_attach_phy+0x30/0x60
       phylink_sfp_connect_phy+0x140/0x510
       sfp_add_phy+0x34/0x50
       init_module+0x15c/0xa7c [sfp]
       cleanup_module+0x1d94/0x3120 [sfp]
       cleanup_module+0x2bb4/0x3120 [sfp]
       process_one_work+0x1f8/0x4ec
       worker_thread+0x1e8/0x3d8
       kthread+0x104/0x110
       ret_from_fork+0x10/0x20

other info that might help us debug this:

Chain exists of:
  triggers_list_lock --> rtnl_mutex --> &sfp->sm_mutex

 Possible unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  lock(&sfp->sm_mutex);
                               lock(rtnl_mutex);
                               lock(&sfp->sm_mutex);
  lock(triggers_list_lock);

 *** DEADLOCK ***

4 locks held by kworker/u8:2/43:
 #0: ffffff80c000f938 ((wq_completion)events_power_efficient){+.+.}-{0:0}, at: process_one_work+0x150/0x4ec
 #1: ffffffc08214bde8 ((work_completion)(&(&sfp->timeout)->work)){+.+.}-{0:0}, at: process_one_work+0x150/0x4ec
 #2: ffffffc0810902f8 (rtnl_mutex){+.+.}-{3:3}, at: rtnl_lock+0x18/0x20
 #3: ffffff80c5c6f318 (&sfp->sm_mutex){+.+.}-{3:3}, at: cleanup_module+0x2ba8/0x3120 [sfp]

stack backtrace:
CPU: 0 PID: 43 Comm: kworker/u8:2 Tainted: G           O       6.7.0-rc4-next-20231208+ #0
Hardware name: Bananapi BPI-R4 (DT)
Workqueue: events_power_efficient cleanup_module [sfp]
Call trace:
 dump_backtrace+0xa8/0x10c
 show_stack+0x14/0x1c
 dump_stack_lvl+0x5c/0xa0
 dump_stack+0x14/0x1c
 print_circular_bug+0x328/0x430
 check_noncircular+0x124/0x134
 __lock_acquire+0x12a0/0x2014
 lock_acquire+0x100/0x2ac
 down_write+0x4c/0x13c
 led_trigger_register+0x4c/0x1a8
 phy_led_triggers_register+0x9c/0x214
 phy_attach_direct+0x154/0x36c
 phylink_attach_phy+0x30/0x60
 phylink_sfp_connect_phy+0x140/0x510
 sfp_add_phy+0x34/0x50
 init_module+0x15c/0xa7c [sfp]
 cleanup_module+0x1d94/0x3120 [sfp]
 cleanup_module+0x2bb4/0x3120 [sfp]
 process_one_work+0x1f8/0x4ec
 worker_thread+0x1e8/0x3d8
 kthread+0x104/0x110
 ret_from_fork+0x10/0x20

Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Fixes: 01e5b728e9e4 ("net: phy: Add a binding for PHY LEDs")
Link: https://lore.kernel.org/r/102a9dce38bdf00215735d04cd4704458273ad9c.1702339354.git.daniel@makrotopia.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
drivers/net/phy/phy_device.c