On Thu, 2005-10-06 at 12:55, Roland Dreier wrote: > Did we ever figure out how to handle the hotplug issues with the > lifetime of the struct ib_device pointer? Right now this API is > unsafe, because a caller can get a pointer to a device that has > already disappeared.
I think this can be handled as follows: The netdev references would be maintained for the duration each AT call until it completes/times out. If subsequent calls are made based on an ib_device which has been removed, an error could be returned based on the fact that AT maintaining a list of devices and validate the supplied device against its list. ipoib_get_info() would be called only with a valid device and the caller holding a netdev reference for at least the duration of that call. > Also if we do decide to add an API like this, the struct ipoib_info > and ipoib_get_info() declarations should be in <linux/if_infiniband.h> > rather than in the private ipoib.h header. OK. -- Hal _______________________________________________ openib-general mailing list [email protected] http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
