Re: [PATCH 17 of 20] ipath - infiniband verbs support

[Roland Dreier <rdreier@cisco.com>]

 > +	/*
 > +	 * We don't need to register a MAD agent, we just need to create
 > +	 * a linker dependency on ib_mad so the module is loaded before
 > +	 * this module is initialized.  The call to ib_register_device()
 > +	 * above will then cause ib_mad to create QP 0 & 1.
 > +	 */
 > +	(void) ib_register_mad_agent(dev, 1, (enum ib_qp_type) 2,
 > +				     NULL, 0, NULL, NULL, NULL);

This looks shady to me.  Can this be solved in userspace by just
making sure that modprobe loads ib_mad before this module?

As it stands you're leaking a mad agent at the very least, not to
mention the hard-coded 2 in there.

 > +	number_of_devices = ipath_layer_get_num_of_dev();
 > +	i = number_of_devices * sizeof(struct ipath_ibdev *);
 > +	ipath_devices = kmalloc(i, GFP_ATOMIC);
 > +	if (ipath_devices == NULL)
 > +		return -ENOMEM;
 > +
 > +	for (i = 0; i < number_of_devices; i++) {
 > +		struct ipath_devdata *dd;
 > +		int ret = ipath_verbs_register(i, ipath_ib_piobufavail,
 > +					       ipath_ib_rcv, ipath_ib_timer,
 > +					       &dd);

What happens if a device is hot plugged or unplugged after you call
ipath_layer_get_num_of_dev() but before you call ipath_verbs_register()?

For that matter, what happens if a device is hot plugged after this
module loads?

 - R.