On 4/10/22 3:08 AM, Zheyu Ma wrote: > Hello, > > I found a bug at the init_one() function in the hfi driver. > When the function xa_alloc_irq() fails, the driver executes the error > handling function sdma_clean(), and this function uses the lock ' > dd->sde_map_lock'. But this lock is initialized after executing the > function xa_alloc_irq(), which causes the following warning: > > [ 23.257762] hfi1 0000:00:05.0: Could not allocate unit ID: error 1 > [ 23.269915] INFO: trying to register non-static key. > [ 23.270318] The code is fine but needs lockdep annotation, or maybe > [ 23.270808] you didn't initialize this object before use? > [ 23.271229] turning off the locking correctness validator. > [ 23.273198] Call Trace: > [ 23.274185] register_lock_class+0x11b/0x880 > [ 23.274525] __lock_acquire+0xf3/0x7930 > [ 23.275769] lock_acquire+0xff/0x2d0 > [ 23.276053] ? sdma_clean+0x42a/0x660 [hfi1] > [ 23.276485] ? lock_release+0x472/0x710 > [ 23.276789] _raw_spin_lock_irq+0x46/0x60 > [ 23.277105] ? sdma_clean+0x42a/0x660 [hfi1] > [ 23.277530] sdma_clean+0x42a/0x660 [hfi1] > [ 23.277945] ? trace_kfree+0x28/0xc0 > [ 23.278232] hfi1_free_devdata+0x3a7/0x420 [hfi1] > [ 23.278688] init_one+0x867/0x11a0 [hfi1] > [ 23.279090] ? _raw_spin_unlock_irqrestore+0x3d/0x60 > [ 23.279482] ? rcu_lock_release+0x20/0x20 [hfi1] > [ 23.279930] pci_device_probe+0x40e/0x8d0 Thanks for the report. Will fix. -Denny
