From: Lukas Wunner <[email protected]> The removal of the igc module leads to a deadlock:
|[Mon Apr 8 17:38:55 2024] __mutex_lock.constprop.0+0x3e5/0x7a0 |[Mon Apr 8 17:38:55 2024] ? preempt_count_add+0x85/0xd0 |[Mon Apr 8 17:38:55 2024] __mutex_lock_slowpath+0x13/0x20 |[Mon Apr 8 17:38:55 2024] mutex_lock+0x3b/0x50 |[Mon Apr 8 17:38:55 2024] rtnl_lock+0x19/0x20 |[Mon Apr 8 17:38:55 2024] unregister_netdevice_notifier+0x2a/0xc0 |[Mon Apr 8 17:38:55 2024] netdev_trig_deactivate+0x25/0x70 |[Mon Apr 8 17:38:55 2024] led_trigger_set+0xe2/0x2d0 |[Mon Apr 8 17:38:55 2024] led_classdev_unregister+0x4f/0x100 |[Mon Apr 8 17:38:55 2024] devm_led_classdev_release+0x15/0x20 |[Mon Apr 8 17:38:55 2024] release_nodes+0x47/0xc0 |[Mon Apr 8 17:38:55 2024] devres_release_all+0x9f/0xe0 |[Mon Apr 8 17:38:55 2024] device_del+0x272/0x3c0 |[Mon Apr 8 17:38:55 2024] netdev_unregister_kobject+0x8c/0xa0 |[Mon Apr 8 17:38:55 2024] unregister_netdevice_many_notify+0x530/0x7c0 |[Mon Apr 8 17:38:55 2024] unregister_netdevice_queue+0xad/0xf0 |[Mon Apr 8 17:38:55 2024] unregister_netdev+0x21/0x30 |[Mon Apr 8 17:38:55 2024] igc_remove+0xfb/0x1f0 [igc] |[Mon Apr 8 17:38:55 2024] pci_device_remove+0x42/0xb0 |[Mon Apr 8 17:38:55 2024] device_remove+0x43/0x70 unregister_netdev() acquires the RNTL lock and releases the LEDs bound to that netdevice. However, netdev_trig_deactivate() and later unregister_netdevice_notifier() try to acquire the RTNL lock again. Avoid this situation by not using the device-managed LED class functions. Link: https://lore.kernel.org/r/CAEhC_B=ksywxcg_+aqqxurgegkq+4mqnsv8ebhokbc3-obj...@mail.gmail.com/ Link: https://lore.kernel.org/r/ZhRD3cOtz5i-61PB@mail-itl/ Reported-by: Roman Lozko <[email protected]> Reported-by: "Marek Marczykowski-Górecki" <[email protected]> Fixes: ea578703b03d ("igc: Add support for LEDs on i225/i226") Signed-off-by: Lukas Wunner <[email protected]> [Kurt: Wrote commit message and tested on i225] Signed-off-by: Kurt Kanzenbach <[email protected]> --- Changes in v2: - Add Lukas' SoB (Lukas) - Add Reported-by (Lukas) - Add links to discussions - Take care of error path (Lukas) - Remove forward declaration (Lukas) - Link to v1: https://lore.kernel.org/r/[email protected] --- drivers/net/ethernet/intel/igc/igc.h | 2 ++ drivers/net/ethernet/intel/igc/igc_leds.c | 38 ++++++++++++++++++++++++------- drivers/net/ethernet/intel/igc/igc_main.c | 3 +++ 3 files changed, 35 insertions(+), 8 deletions(-) diff --git a/drivers/net/ethernet/intel/igc/igc.h b/drivers/net/ethernet/intel/igc/igc.h index 90316dc58630..6bc56c7c181e 100644 --- a/drivers/net/ethernet/intel/igc/igc.h +++ b/drivers/net/ethernet/intel/igc/igc.h @@ -298,6 +298,7 @@ struct igc_adapter { /* LEDs */ struct mutex led_mutex; + struct igc_led_classdev *leds; }; void igc_up(struct igc_adapter *adapter); @@ -723,6 +724,7 @@ void igc_ptp_read(struct igc_adapter *adapter, struct timespec64 *ts); void igc_ptp_tx_tstamp_event(struct igc_adapter *adapter); int igc_led_setup(struct igc_adapter *adapter); +void igc_led_free(struct igc_adapter *adapter); #define igc_rx_pg_size(_ring) (PAGE_SIZE << igc_rx_pg_order(_ring)) diff --git a/drivers/net/ethernet/intel/igc/igc_leds.c b/drivers/net/ethernet/intel/igc/igc_leds.c index bf240c5daf86..3929b25b6ae6 100644 --- a/drivers/net/ethernet/intel/igc/igc_leds.c +++ b/drivers/net/ethernet/intel/igc/igc_leds.c @@ -236,8 +236,8 @@ static void igc_led_get_name(struct igc_adapter *adapter, int index, char *buf, pci_dev_id(adapter->pdev), index); } -static void igc_setup_ldev(struct igc_led_classdev *ldev, - struct net_device *netdev, int index) +static int igc_setup_ldev(struct igc_led_classdev *ldev, + struct net_device *netdev, int index) { struct igc_adapter *adapter = netdev_priv(netdev); struct led_classdev *led_cdev = &ldev->led; @@ -257,24 +257,46 @@ static void igc_setup_ldev(struct igc_led_classdev *ldev, led_cdev->hw_control_get = igc_led_hw_control_get; led_cdev->hw_control_get_device = igc_led_hw_control_get_device; - devm_led_classdev_register(&netdev->dev, led_cdev); + return led_classdev_register(&netdev->dev, led_cdev); } int igc_led_setup(struct igc_adapter *adapter) { struct net_device *netdev = adapter->netdev; - struct device *dev = &netdev->dev; struct igc_led_classdev *leds; - int i; + int i, err; mutex_init(&adapter->led_mutex); - leds = devm_kcalloc(dev, IGC_NUM_LEDS, sizeof(*leds), GFP_KERNEL); + leds = kcalloc(IGC_NUM_LEDS, sizeof(*leds), GFP_KERNEL); if (!leds) return -ENOMEM; - for (i = 0; i < IGC_NUM_LEDS; i++) - igc_setup_ldev(leds + i, netdev, i); + for (i = 0; i < IGC_NUM_LEDS; i++) { + err = igc_setup_ldev(leds + i, netdev, i); + if (err) + goto err; + } + + adapter->leds = leds; return 0; + +err: + for (i--; i >= 0; i--) + led_classdev_unregister(&((leds + i)->led)); + + kfree(leds); + return err; +} + +void igc_led_free(struct igc_adapter *adapter) +{ + struct igc_led_classdev *leds = adapter->leds; + int i; + + for (i = 0; i < IGC_NUM_LEDS; i++) + led_classdev_unregister(&((leds + i)->led)); + + kfree(leds); } diff --git a/drivers/net/ethernet/intel/igc/igc_main.c b/drivers/net/ethernet/intel/igc/igc_main.c index 35ad40a803cb..4d975d620a8e 100644 --- a/drivers/net/ethernet/intel/igc/igc_main.c +++ b/drivers/net/ethernet/intel/igc/igc_main.c @@ -7021,6 +7021,9 @@ static void igc_remove(struct pci_dev *pdev) cancel_work_sync(&adapter->watchdog_task); hrtimer_cancel(&adapter->hrtimer); + if (IS_ENABLED(CONFIG_IGC_LEDS)) + igc_led_free(adapter); + /* Release control of h/w to f/w. If f/w is AMT enabled, this * would have already happened in close and is redundant. */ --- base-commit: 7efd0a74039fb6b584be2cb91c1d0ef0bd796ee1 change-id: 20240411-igc_led_deadlock-7abd85954f5e Best regards, -- Kurt Kanzenbach <[email protected]>
