On Tue, Jun 30, 2020 at 09:06:30AM -0300, Jason Gunthorpe wrote:
> On Tue, Jun 30, 2020 at 02:52:24PM +0300, Leon Romanovsky wrote:
> > On Tue, Jun 30, 2020 at 08:37:29AM -0300, Jason Gunthorpe wrote:
> > > On Tue, Jun 30, 2020 at 10:21:37AM +0300, Leon Romanovsky wrote:
> > > > On Mon, Jun 29, 2020 at 12:39:07PM -0300, Jason Gunthorpe wrote:
> > > > > On Wed, Jun 24, 2020 at 01:54:22PM +0300, Leon Romanovsky wrote:
> > > > > > @@ -4018,8 +4028,7 @@ const struct uapi_definition uverbs_def_write_intf[] = {
> > > > > > IB_USER_VERBS_EX_CMD_DESTROY_RWQ_IND_TBL,
> > > > > > ib_uverbs_ex_destroy_rwq_ind_table,
> > > > > > UAPI_DEF_WRITE_I(
> > > > > > - struct ib_uverbs_ex_destroy_rwq_ind_table),
> > > > > > - UAPI_DEF_METHOD_NEEDS_FN(destroy_rwq_ind_table))),
> > > > > > + struct ib_uverbs_ex_destroy_rwq_ind_table))),
> > > > >
> > > > > Removing these is kind of troublesome.. This misses the one for ioctl:
> > > > >
> > > > > UAPI_DEF_CHAIN_OBJ_TREE_NAMED(
> > > > > UVERBS_OBJECT_RWQ_IND_TBL,
> > > > > UAPI_DEF_OBJ_NEEDS_FN(destroy_rwq_ind_table)),
> > > >
> > > > I will remove, but it seems that we have some gap here, I would expect
> > > > any sort of compilation error for mlx4.
> > >
> > > Why would there be a compilation error?
> >
> > I would expect BUILD_BUG_ON_ZERO() is thrown if ibdev_fn == NULL
>
> ??
>
> > > And it should not be removed, it needs to be reworked to point to some
> > > other function I suppose.
> >
> > Why?
>
> The destroy function should not be registered at all if rwq_ind is not
> supported by the driver - this is the methodlogy.
And here comes mlx4 case that needs create() call but doesn't need
destroy(), because it will be empty after this refactoring.
Thanks
>
> Jason
