> -----Original Message-----
> From: Jason Gunthorpe <jgg@xxxxxxxx>
> Sent: Saturday, January 12, 2019 11:35 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>
> Subject: Re: [PATCH rdma-next 3/7] RDMA/core: Introduce helper functions
> for cache cleanup, update
>
> On Sat, Jan 12, 2019 at 05:20:19PM +0000, Parav Pandit wrote:
> >
> >
> > > From: Jason Gunthorpe <jgg@xxxxxxxx>
> > > Sent: Saturday, January 12, 2019 11:14 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>
> > > Subject: Re: [PATCH rdma-next 3/7] RDMA/core: Introduce helper
> > > functions for cache cleanup, update
> > >
> > > On Sat, Jan 12, 2019 at 04:39:25AM +0000, Parav Pandit wrote:
> > > > > > +
> > > > > > +void ib_device_cache_update(struct ib_device *device); void
> > > > > > +ib_gid_table_cleanup_one(struct ib_device *device);
> > > > >
> > > > > This is super goofy, why do we have a 'cleanup' followed by 'update'
> > > > > API design?
> > > > Cleanup removes all entries from the gid table.
> > >
> > > The work 'cleanup' should be a final act, there shouldn't be another
> > > touch to the datastructure until it is re-inited. So either
> > > 'cleanup' or 'update' are the wrong words for what these functions are
> doing.
> > >
> > Update is wrong word. A wrapper such as reinit() for a update() would
> help?
>
> Sure, but if we have reinit do we need the two steps to be exposed?
>
reinit() looks like below.
Is that what you are looking for?
gid_cache_reinit() {
ib_gid_table_cleanup_one(device);
rdma_roce_rescan_device(device);
from_start_port to end_port {
config_non_roce_gid_cache(device);
}
> Jason
