On Tue, May 22, 2018 at 04:05:41PM -0600, Jason Gunthorpe wrote:
> On Tue, May 15, 2018 at 05:09:41PM +0300, Leon Romanovsky wrote:
>
> > +static int uverbs_free_counters(struct ib_uobject *uobject,
> > + enum rdma_remove_reason why)
> > +{
> > + struct ib_counters *counters =
> > + (struct ib_counters *)(uobject->object);
>
> No casts from void *.
>
> > +static int UVERBS_HANDLER(UVERBS_METHOD_COUNTERS_CREATE)(struct ib_device *ib_dev,
> > + struct ib_uverbs_file *file,
> > + struct uverbs_attr_bundle *attrs)
> > +{
> > + const struct uverbs_attr *uattr;
> > + struct ib_counters *counters;
> > + struct ib_uobject *uobj;
> > + int ret;
> > +
> > + if (!ib_dev->create_counters)
> > + return -EOPNOTSUPP;
>
> The methods shouldn't even exist in the parse tree if the function
> pointer is not present - we have systematically made this mistake in
> the new uapi, and it needs more infrastructure to fix, so we can leave
> it here for now.. But noting it to help not forget..
>
> > + uattr = uverbs_attr_get(attrs, UVERBS_ATTR_CREATE_COUNTERS_HANDLE);
> > + uobj = uattr->obj_attr.uobject;
>
> Would prefer to see this patch as part of this series:
>
> https://patchwork.kernel.org/patch/10401247/
>
> And the above to use it, as we are going to forget to do this conversion.
>
> > +DECLARE_UVERBS_NAMED_OBJECT(UVERBS_OBJECT_COUNTERS,
> > + &UVERBS_TYPE_ALLOC_IDR(uverbs_free_counters),
> > + &UVERBS_METHOD(UVERBS_METHOD_COUNTERS_CREATE),
> > + &UVERBS_METHOD(UVERBS_METHOD_COUNTERS_DESTROY));
>
> And this doesn't compile out of sequence with the devx patches, is '0'
> the right ordering level for counters?
It needed DevX patches where we did cleanup in some macros.
>
> Jason
Attachment:
signature.asc
Description: PGP signature
