On Fri, Aug 03, 2018 at 08:44:24PM +0000, Ruhl, Michael J wrote:
> >diff --git a/drivers/infiniband/core/uverbs_std_types.c
> >b/drivers/infiniband/core/uverbs_std_types.c
> >index 3aa7c7deac749a..7f22b820a21ba0 100644
> >+++ b/drivers/infiniband/core/uverbs_std_types.c
> >@@ -316,4 +316,3 @@ const struct uverbs_object_tree_def
> >*uverbs_default_get_objects(void)
> > {
> > return &uverbs_default_objects;
> > }
> >-EXPORT_SYMBOL_GPL(uverbs_default_get_objects);
>
> That is tremendously cleaner than the original path.
>
> This comment in uverbs_alloc_spec_tree() that seems to imply that device
> can create a parse tree that will disable this interface:
>
> /*
> * Devices which don't want to support ib_uverbs, should just allocate
> * an empty parsing tree. Every user-space command won't hit any valid
> * entry in the parsing tree and thus will fail.
> */
Yes, that was something that was supported, but I have removed in this
series.
Drivers creating a struct ib_device *MUST* support verbs, period.
The only exception is usnic, and that error will never be repeated.
That said, the general infrastructure in this series is still general,
and it would be easy to create a a 'struct uverbs_api' that does not
include the verbs methods - eg for rdma-cm or something. Some gentle
rework would be needed but it is quite doable.
This is the approach to take if this interface is desired without
verbs (and run it on a different cdev than /dev/uverbs)
> It seems that this patch would break this ability. Is this correct, or should
> the comment be updated/removed?
Yes, it should be removed. A later patch in this series drops that
comment already..
> If you would like it:
>
> Reviewed-by: Michael J. Ruhl <michael.j.ruhl@xxxxxxxxx>
Thanks!
Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
