James,
It appears there is a missing class_device_del. The comments for transport_remove_device indicate that transport_remove_classdev should call it (which the attached patch does), but the comment in attribute_container_remove_device:
If you want a * two phase release: remove from visibility and then delete the * device, then you should use this routine with a fn that calls * class_device_del() and then use * attribute_container_device_trigger() to do the final put on the * classdev.
Indicate that maybe transport_destroy_classdev and transport_remove_device are incorrect. However, attribute_container_remove_device will do a list_del(&ic->node); so that later calls to attribute_container_device_trigger would not be able to do the final put on that classdev. I assumed the comments in attribute_container_remove_device are incorrect, so the attached patch made against scsi-rc-fixes-2.6 adds a call to class_device_del to transport_remove_classdev.
I did not fix the comments in attribute_container_remove_device becuase I was not 100% certain what is the correct behavior or usage.
Mike
--- scsi-rc-fixes-2.6.orig/drivers/base/transport_class.c 2005-01-25 13:30:27.000000000 -0800 +++ scsi-rc-fixes-2.6.newtest/drivers/base/transport_class.c 2005-01-26 15:22:52.000000000 -0800 @@ -216,10 +216,16 @@ static int transport_remove_classdev(str struct class_device *classdev) { struct transport_class *tclass = class_to_transport_class(cont->class); + struct class_device_attribute **attrs = cont->attrs; + int i; if (tclass->remove) tclass->remove(dev); + for (i = 0; attrs[i]; i++) + class_device_remove_file(classdev, attrs[i]); + class_device_del(classdev); + return 0; }