@@ -1359,6 +1359,7 @@ struct pqi_ctrl_info {
struct mutex scan_mutex;
struct mutex lun_reset_mutex;
+ struct lock_class_key lun_reset_key;
bool controller_online;
bool block_requests;
bool scan_blocked;
@@ -8789,9 +8789,11 @@ static struct pqi_ctrl_info *pqi_alloc_ctrl_info(int numa_node)
return NULL;
mutex_init(&ctrl_info->scan_mutex);
- mutex_init(&ctrl_info->lun_reset_mutex);
mutex_init(&ctrl_info->ofa_mutex);
+ lockdep_register_key(&ctrl_info->lun_reset_key);
+ mutex_init_with_key(&ctrl_info->lun_reset_mutex, &ctrl_info->lun_reset_key);
+
INIT_LIST_HEAD(&ctrl_info->scsi_device_list);
spin_lock_init(&ctrl_info->scsi_device_list_lock);
@@ -8830,6 +8832,10 @@ static struct pqi_ctrl_info *pqi_alloc_ctrl_info(int numa_node)
static inline void pqi_free_ctrl_info(struct pqi_ctrl_info *ctrl_info)
{
+ mutex_destroy(&ctrl_info->scan_mutex);
+ mutex_destroy(&ctrl_info->ofa_mutex);
+ mutex_destroy(&ctrl_info->lun_reset_mutex);
+ lockdep_unregister_key(&ctrl_info->lun_reset_key);
kfree(ctrl_info);
}
A lockdep recursive locking splat happens when shutting down a debug kernel on some systems. ============================================ WARNING: possible recursive locking detected -------------------------------------------- reboot/15103 is trying to acquire lock: ffff8881435af8c8 (&ctrl_info->lun_reset_mutex){+.+.}-{3:3}, at: pqi_shutdown+0x112/0x4b0 [smartpqi] but task is already holding lock: ffff8888929278c8 (&ctrl_info->lun_reset_mutex){+.+.}-{3:3}, at: pqi_shutdown+0x112/0x4b0 [smartpqi] other info that might help us debug this: Possible unsafe locking scenario: CPU0 ---- lock(&ctrl_info->lun_reset_mutex); lock(&ctrl_info->lun_reset_mutex); *** DEADLOCK *** May be due to missing lock nesting notation 4 locks held by reboot/15103: #0: ffffffff8db76f40 (system_transition_mutex){+.+.}-{3:3}, at: __do_sys_reboot+0x12b/0x2f0 #1: ffff8888929278c8 (&ctrl_info->lun_reset_mutex){+.+.}-{3:3}, at: pqi_shutdown+0x112/0x4b0 [smartpqi] #2: ffff88810a8921a8 (&dev->mutex){....}-{3:3}, at: device_shutdown+0x1de/0x540 #3: ffff88810a9141a8 (&dev->mutex){....}-{3:3}, at: device_shutdown+0x1ec/0x540 Call Trace: dump_stack_lvl+0x6f/0xb0 print_deadlock_bug.cold+0xbd/0xca validate_chain+0x37b/0x570 __lock_acquire+0x55b/0xac0 lock_acquire.part.0+0xf5/0x2b0 mutex_lock_nested+0x4b/0x190 pqi_shutdown+0x112/0x4b0 [smartpqi] pci_device_shutdown+0x76/0x110 device_shutdown+0x2ea/0x540 kernel_restart+0x64/0xa0 __do_sys_reboot+0x1d8/0x2f0 do_syscall_64+0x92/0x180 The fact that there are two dev->mutex'es acquired in device_shutdown() means both a parent and a child devices are being worked on and likely that the lun_reset_mutex'es of these two devices are being acquired here too. However, the way lun_reset_mutex is initialized in pqi_alloc_ctrl_info() will casue all the lun_reset_mutex'es to be treated as in the same class leading to this false positive warning. Fix that by initializing each instance of lun_reset_mutex with its own unique key so that they will be treated as different by lockdep. Also call mutex_destroy() and lockdep_unregister_key() in pqi_free_ctrl_info() before ctrl_info is freed. With this patch applied, the lockdep splat no longer happens. Signed-off-by: Waiman Long <longman@redhat.com> --- drivers/scsi/smartpqi/smartpqi.h | 1 + drivers/scsi/smartpqi/smartpqi_init.c | 8 +++++++- 2 files changed, 8 insertions(+), 1 deletion(-)