Re: [PATCH rdma-next 1/2] RDMA: Pass entire uverbs attr bundle to create cq function

[Leon Romanovsky <leon@kernel.org>]

On Tue, Jun 18, 2024 at 01:13:45PM -0300, Jason Gunthorpe wrote:
> On Tue, Jun 18, 2024 at 07:05:59PM +0300, Leon Romanovsky wrote:
> > On Tue, Jun 18, 2024 at 10:08:54AM -0300, Jason Gunthorpe wrote:
> > > On Tue, Jun 18, 2024 at 08:05:57AM +0300, Leon Romanovsky wrote:
> > > > On Mon, Jun 17, 2024 at 05:10:03PM -0300, Jason Gunthorpe wrote:
> > > > > On Mon, Jun 17, 2024 at 06:49:47PM +0300, Leon Romanovsky wrote:
> > > > > > On Mon, Jun 17, 2024 at 10:44:09AM -0300, Jason Gunthorpe wrote:
> > > > > > > On Sun, Jun 16, 2024 at 07:15:57PM +0300, Leon Romanovsky wrote:
> > > > > > > 
> > > > > > > > @@ -63,6 +63,7 @@ enum uverbs_default_objects {
> > > > > > > >  enum {
> > > > > > > >  	UVERBS_ATTR_UHW_IN = UVERBS_UDATA_DRIVER_DATA_FLAG,
> > > > > > > >  	UVERBS_ATTR_UHW_OUT,
> > > > > > > > +	UVERBS_ATTR_UHW_DRIVER_DATA,
> > > > > > > 
> > > > > > > The start of the driver's attributes is not a "UHW", the UHW is only
> > > > > > > the old structs.
> > > > > > 
> > > > > > I asked from Akiva to keep existing naming convention UVERBS_ATTR_UHW_XXX
> > > > > > to emphasize the namespace and the position of this attribute as
> > > > > > relevant for existing UHW calls.
> > > > > 
> > > > > Well, calling it DRIVER_DATA and UHW is very confusing when it is
> > > > > really the start of the indexing for drivers that use UHW.
> > > > > 
> > > > > A better name is needed
> > > > 
> > > > UVERBS_ATTR_UHW_PRIVATE ????
> > > 
> > > I think it need to have the word "start" in it, because it is the
> > > start of numbers, not an actual number itself.
> > 
> > UVERBS_ATTR_UHW_DRIVER_DATA_START ????
> > What do you suggest instead?
> 
> How about:

ok, let's take this variant.

Thanks

> 
> diff --git a/include/uapi/rdma/ib_user_ioctl_cmds.h b/include/uapi/rdma/ib_user_ioctl_cmds.h
> index dafc7ebe545b8d..e9322f66cd2dec 100644
> --- a/include/uapi/rdma/ib_user_ioctl_cmds.h
> +++ b/include/uapi/rdma/ib_user_ioctl_cmds.h
> @@ -37,9 +37,6 @@
>  #define UVERBS_ID_NS_MASK 0xF000
>  #define UVERBS_ID_NS_SHIFT 12
>  
> -#define UVERBS_UDATA_DRIVER_DATA_NS    1
> -#define UVERBS_UDATA_DRIVER_DATA_FLAG  (1UL << UVERBS_ID_NS_SHIFT)
> -
>  enum uverbs_default_objects {
>         UVERBS_OBJECT_DEVICE, /* No instances of DEVICE are allowed */
>         UVERBS_OBJECT_PD,
> @@ -61,8 +58,10 @@ enum uverbs_default_objects {
>  };
>  
>  enum {
> -       UVERBS_ATTR_UHW_IN = UVERBS_UDATA_DRIVER_DATA_FLAG,
> +       UVERBS_ID_DRIVER_NS = 1U << UVERBS_ID_NS_SHIFT,
> +       UVERBS_ATTR_UHW_IN = UVERBS_ID_DRIVER_NS,
>         UVERBS_ATTR_UHW_OUT,
> +       UVERBS_ID_DRIVER_NS_WITH_UHW,
>  };
>  
>  enum uverbs_methods_device {
> 
> And recommend replacing the open coded UVERBS_ID_DRIVER_NS all over
> the place.
> 
> > > It is also not PRIVATE at all, this is just in the device specific
> > > space number space, not the core space.
> > 
> > Private in the sense of driver specific, like net_priv().
> 
> It is not a private, it is a namespace, that is the naming that was
> used here.
> 
> Jason