On 10/18/23 14:48, Bart Van Assche wrote: > > On 10/18/23 12:17, Jason Gunthorpe wrote: >> If siw hangs as well, I definitely comfortable continuing to debug and >> leaving the work queues in-tree for now. > > Regarding the KASAN complaint that I shared about one month ago, can that complaint have any other root cause than the patch "RDMA/rxe: Add > workqueue support for rxe tasks"? That report shows a use-after-free by > rxe code with a pointer to memory that was owned by the rxe driver and > that was freed by the rxe driver. That memory is an skbuff. The rxe > driver manages skbuffs. The SRP driver doesn't even know about these > skbuff objects. See also https://lore.kernel.org/linux-rdma/8ee2869b-3f51-4195-9883-015cd30b4241@xxxxxxx/ > > Thanks, > > Bart. > Bart, I agree with you that that is a rxe issue. But, I haven't been able to reproduce it. However, I am able to generate hangs without the KASAN bug so it seems to me that they are unrelated. In addition to the kernel debugging I have added tracing to ib_srp and ib_srpt which may help delay things. Bob
