On 2/7/22 22:09, John Garry wrote: > Hi Damien, > > On 04/02/2022 03:02, Damien Le Moal wrote: >> This is the submission path, not completion. The code is: >> >> (gdb) list *(pm8001_queue_command+0x842) >> 0x3d42 is in pm8001_queue_command (drivers/scsi/pm8001/pm8001_sas.c:491). >> 486 atomic_dec(&pm8001_dev->running_req); >> 487 goto err_out_tag; >> 488 } >> 489 /* TODO: select normal or high priority */ >> 490 spin_lock(&t->task_state_lock); >> 491 t->task_state_flags |= SAS_TASK_AT_INITIATOR; >> 492 spin_unlock(&t->task_state_lock); >> 493 } while (0); >> 494 rc = 0; >> 495 goto out_done; >> >> So the task is already completed when the submission path tries to set >> the state flag ? Debugging... > > JFYI, I am putting together a patch to drop SAS_TASK_AT_INITIATOR > altogether. I just need to ensure that the logic in the isci code is > correct. Great. Less dead code for the pm8001 driver :) I was busy with a nasty libata bug today so I did not continue my investigations. Will try tomorrow. > > Thanks, > John -- Damien Le Moal Western Digital Research
