On 6/4/2019 7:47 PM, Benjamin Herrenschmidt wrote: >> Should we replace these device lock with try lock loop with some sleep >> statements. This could solve the immediate deadlock issues until >> someone implements granular locking in pci. > That won't necessarily solve this AB->BA problem. I think the issue > here is that sriov shouldn't device_lock before doing something that > can take the pci_bus_sem. > I agree. I found out about lock ordering issue after reading the email one more time.
