> -----Original Message----- > From: Jason Gunthorpe > Sent: Thursday, December 20, 2018 11:05 AM > To: Parav Pandit <parav@xxxxxxxxxxxx> > Cc: Leon Romanovsky <leon@xxxxxxxxxx>; Doug Ledford > <dledford@xxxxxxxxxx>; Leon Romanovsky <leonro@xxxxxxxxxxxx>; RDMA > mailing list <linux-rdma@xxxxxxxxxxxxxxx>; Ira Weiny <ira.weiny@xxxxxxxxx>; > Jack Morgenstein <jackm@xxxxxxxxxxxxxxxxxx> > Subject: Re: [PATCH rdma-next 4/6] IB/umad: Refactor code to use > cdev_device_add() > > On Wed, Dec 19, 2018 at 11:41:20PM -0700, Parav Pandit wrote: > > > I looked at this during v0 version. Couldn't do it because both issm > > and umad devices have their ibdev and port files and they are part of > > single structure. So container_of() from the device cannot teach back > > to umad_device as offsets for two devices are different. > > Creating another structure and storing umad_device structure pointer > > is not worth overhead. > > umad and issm don't share code, do they? I would expect to do the > appropriate container_of for each struct device? > They share two trival sysfs files. creating one generic structure and putting their devices in such new structure, and reaching back to ib_device was not worth the code. I think get/set_drvdata() is fine, given that this is get/set on device created by the ib_umad module.
