Fixed
Created: Jun 18, 2025
Updated: Oct 14, 2025
Resolved Date: Oct 13, 2025
Found In Version: 10.21.20.1
Fix Version: 10.21.20.14
Severity: Standard
Applicable for: Wind River Linux LTS 21
Component/s: Kernel
In the Linux kernel, the following vulnerability has been resolved:EOL][EOL]driver core: fix potential deadlock in __driver_attach[EOL][EOL]In __driver_attach function, There are also AA deadlock problem,[EOL]like the commit b232b02bf3c2 ("driver core: fix deadlock in[EOL]__device_attach").[EOL][EOL]stack like commit b232b02bf3c2 ("driver core: fix deadlock in[EOL]__device_attach").[EOL]list below:[EOL] In __driver_attach function, The lock holding logic is as follows:[EOL] ...[EOL] __driver_attach[EOL] if (driver_allows_async_probing(drv))[EOL] device_lock(dev) // get lock dev[EOL] async_schedule_dev(__driver_attach_async_helper, dev); // func[EOL] async_schedule_node[EOL] async_schedule_node_domain(func)[EOL] entry = kzalloc(sizeof(struct async_entry), GFP_ATOMIC);[EOL] /* when fail or work limit, sync to execute func, but[EOL] __driver_attach_async_helper will get lock dev as[EOL] will, which will lead to A-A deadlock. */[EOL] if (!entry (| atomic_read(&entry_count) > MAX_WORK) {[EOL) func;EOL] else[EOL] queue_work_node(node, system_unbound_wq, &entry->work)[EOL] device_unlock(dev)[EOL][EOL] As above show, when it is allowed to do async probes, because of[EOL] out of memory or work limit, async work is not be allowed, to do[EOL] sync execute instead. it will lead to A-A deadlock because of[EOL] __driver_attach_async_helper getting lock dev.[EOL][EOL]Reproduce:[EOL]and it can be reproduce by make the condition[EOL](if (!entry (| atomic_read(&entry_count) > MAX_WORK)) untenable, like[EOL)below:[EOL][EOL][ 370.785650] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables[EOL]this message.[EOL][ 370.787154] task:swapper/0 state:D stack: 0 pid: 1 ppid:[EOL]0 flags:0x00004000[EOL][ 370.788865] Call Trace:[EOL][ 370.789374] <TASK>[EOL][ 370.789841] __schedule+0x482/0x1050[EOL][ 370.790613] schedule+0x92/0x1a0[EOL][ 370.791290] schedule_preempt_disabled+0x2c/0x50[EOL][ 370.792256] __mutex_lock.isra.0+0x757/0xec0[EOL][ 370.793158] __mutex_lock_slowpath+0x1f/0x30[EOL][ 370.794079] mutex_lock+0x50/0x60[EOL][ 370.794795] __device_driver_lock+0x2f/0x70[EOL][ 370.795677] ? driver_probe_device+0xd0/0xd0[EOL][ 370.796576] __driver_attach_async_helper+0x1d/0xd0[EOL][ 370.797318] ? driver_probe_device+0xd0/0xd0[EOL][ 370.797957] async_schedule_node_domain+0xa5/0xc0[EOL][ 370.798652] async_schedule_node+0x19/0x30[EOL][ 370.799243] __driver_attach+0x246/0x290[EOL][ 370.799828] ? driver_allows_async_probing+0xa0/0xa0[EOL][ 370.800548] bus_for_each_dev+0x9d/0x130[EOL][ 370.801132] driver_attach+0x22/0x30[EOL][ 370.801666] bus_add_driver+0x290/0x340[EOL][ 370.802246] driver_register+0x88/0x140[EOL][ 370.802817] ? virtio_scsi_init+0x116/0x116[EOL][ 370.803425] scsi_register_driver+0x1a/0x30[EOL][ 370.804057] init_sd+0x184/0x226[EOL][ 370.804533] do_one_initcall+0x71/0x3a0[EOL][ 370.805107] kernel_init_freeable+0x39a/0x43a[EOL][ 370.805759] ? rest_init+0x150/0x150[EOL][ 370.806283] kernel_init+0x26/0x230[EOL][ 370.806799] ret_from_fork+0x1f/0x30[EOL][EOL]To fix the deadlock, move the async_schedule_dev outside device_lock,[EOL]as we can see, in async_schedule_node_domain, the parameter of[EOL]queue_work_node is system_unbound_wq, so it can accept concurrent[EOL]operations. which will also not change the code logic, and will[EOL]not lead to deadlock.