Bart Van Assche reported that when the scsi_debug driver was being stress tested with fio, changing the delay paremeter via sysfs caused a cascade of oops-es. The fix presented reads the driver wide delay values (jiffies or nanoseconds) once and remembers in the sdebug_defer object which defer method is used and which method has been initialized. This simplifies handling when command aborts occur. This causes a minor changes in semantic: a SCSI command "in flight" is no longer impacted by changing the delay option after it has been scheduled (i.e. while it is waiting for a work queue or a hr timer to exhaust). Douglas Gilbert (1): scsi_debug: delay fix drivers/scsi/scsi_debug.c | 72 ++++++++++++++++++++++++++++++----------------- 1 file changed, 46 insertions(+), 26 deletions(-) -- 2.14.1