On Fri, 9 Aug 2024 23:46:26 Taylor R Campbell wrote: > But there's obviously something else going on here that I don't > understand -- and I find that concerning, because it means there's a > _much more interesting_ semantic difference between scsipi_done and > scsipi_done_once than I understand, and that makes me concerned > there's something incoherent about the scsi API that might affect lots > of drivers.
I'm very sorry Taylor. I was confused about the crash point and flow of execution of code. scsipi_done is not re-entrant...Sorry :( I've appended to the bug report a full dmesg, snippets of kernel error messages and a full backtrace on an unpactched kernel. I hope this provides a better understanding. The patch fixes it buy deferring scsipi_run_queue though. Best regards, Nat
