Re: [PATCH RFC 11/20] ctlv: Prepare runtime introspection

[Jason Gunthorpe]

On Tue, Nov 03, 2026 at 10:29:45AM +0100, Jiri Pirko wrote:
> Tue, Nov 03, 2026 at 02:05:37AM +0100, jgg@xxxxxxxx wrote:
> >On Fri, Oct 30, 2026 at 10:53:11AM +0100, Jiri Pirko wrote:
> >> >So.. how about could we drop the sets, use a direct description of the
> >> >layout and do something else for the ID numbering? Maybe the numbering
> >> >can be solved directly by only listing the names in:
> >> >
> >> >numbering:
> >> >  - device-get.reply
> >> >  - device-get.reply.caps
> >> >  - device-get.caps.evidence
> >> >  -[..]
> >> >
> >> >?
> >> 
> >> I'm still committed to implicit numbering. For "inline records" I
> >> resolved it by having it encoded as a tree with 4 levels:
> >> 
> >>  63         48 47 46 45        32 31  28 27  24 23  20 19  16 15               0
> >> +-------------+-----+------------+------+------+------+------+------------------+
> >> |   family    |kind |   index    |  L1  |  L2  |  L3  |  L4  |      field       |
> >> +-------------+-----+------------+------+------+------+------+------------------+
> >>               |<------------------ record ------------------>|
> >
> >Fascinating idea. Sort of like an oid
> >
> >I guarentee we will run out of bits someday :) I saw recently we ran
> >out of bits in RDMA :\ 3 bits seems like quite a small number..
> 
> Well, this is just for inlines. I think if user comes to the point this
> will not be enough, he can use named record, as he probably should with
> more than 8 inline records anyway :) Also, we can introduce "mapping"
> mechanism to allow inline conversion to named record without breaking
> userspace. I just didn't bother to implement now as it is a theoretical
> problem now as I see it.

Oh OK, thats interesting.

Jason