On Fri, Dec 21, 2018 at 04:19:22PM +0200, Leon Romanovsky wrote: > From: Leon Romanovsky <leonro@xxxxxxxxxxxx> > > Changelog > v1->v2: > - Dropped horizontal white spaces > - Use c99 flex array syntax for ports > - Changed code for dev_release() to do ib_umad_dev_put() > - Use struct_size() > - Changed the patch order to use umad_class.name that reduced one patch > v0->v1: > - Address Jason's comment to reuse class name during char device creation time > - Changed subject from RDMA/umad to IB/umad to be consistent with rest patches > - Removed extra space after comment in patch-2 > > >From Parav, > > Refactor ib_umad class and device registration code for following items. > 1. To make use of more static definitions and utilize class_groups and > dev_groups to register class and device specific attributes This is > similar to commits b53b1c08a23e and c5c4d92e70f3 for ib_uverbs module. > 2. Use kref instead of kobject as its lighter for ib_umad_device > reference counting. > 3. Avoid accessing undefined dev->name during ioctl() operation for > device > under destruction. > 4. Refactor for minor cosmetics for alignments. > 5. Return more appropriate error code instead of -1. > > Thanks > > Parav Pandit (5): > IB/umad: Simplify and avoid dynamic allocation of class > IB/umad: Avoid destroying device while it is accessed > IB/umad: Refactor code to use cdev_device_add() > IB/umad: Use class_groups and let core create class file > IB/umad: Start using dev_groups of class Applied to for-next, thanks Jason
