On Tue, Dec 18, 2018 at 09:32:16AM -0700, Parav Pandit wrote: > > > @@ -4368,13 +4371,12 @@ static void cm_add_one(struct ib_device > > *ib_device) > > > cm_dev->ib_device = ib_device; > > > cm_dev->ack_delay = ib_device->attrs.local_ca_ack_delay; > > > cm_dev->going_down = 0; > > > - cm_dev->device = device_create(&cm_class, &ib_device->dev, > > > + cm_dev->devnum = ida_alloc(&cm_ida, GFP_KERNEL); > > > + cm_dev->device = device_create(&cm_class, ib_device->dev.parent, > > > MKDEV(0, 0), NULL, > > > > Yikes, this is horrifying, why is it creating a struct device at all like this?? > > > > Probably the CM counter group debug sysfs should just be added directly to > > the actual ib_dev's sysfs?? > > Yes. anchoring to ibdev's sysfs looks right way but that breaks the > ABI/scripts who are accustomed to use it from infiniband_cm. But doesn't this patch rename those paths already? What paths do you think people are using? Jason
