Piling onto this thread, there's actually another use in the SCSI target stack that I think wants something similar to this. In drivers/scsi/qla2xxx/tcm_qla2xxx.c there's some code (for example tcm_qla2xxx_tpg_store_enable()) which is called from a configfs callback, and defers a call to configfs_depend_item() (indirectly via tcm_qla2xxx_depend_tpg() -> target_depend_item()) to a workqueue because "configfs callbacks can't call configfs_depend_item()." The code here is trying to express the fact that enabling a target on an NPIV device (virtual FC port) depends on keeping the base physical port around. So I think we need case 1) (depend on an item in the same subsystem without relocking). But of course deferring to a workqueue doesn't solve anything because schedule_work() and wait_for_completion() on a completion completed by the work item does not avoid any lock dependencies — it's pretty much equivalent to a direct function call. So something like this patch would be welcome for solving that problem too. - R. -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
