On Wed, 31 Jul 2019 15:50:26 -0600, David Ahern wrote:

On 7/30/19 12:08 AM, Jiri Pirko wrote:

quoted

Mon, Jul 29, 2019 at 10:17:25PM CEST, dsahern@gmail.com wrote:  

quoted

On 7/27/19 3:44 AM, Jiri Pirko wrote:  

quoted

From: Jiri Pirko <redacted>

Devlink from the beginning counts with network namespaces, but the
instances has been fixed to init_net. The first patch allows user
to move existing devlink instances into namespaces:
 

so you intend for an asic, for example, to have multiple devlink
instances where each instance governs a set of related ports (e.g.,
ports that share a set of hardware resources) and those instances can be
managed from distinct network namespaces?  

No, no multiple devlink instances for asic intended.  

So it should be allowed for an asic to have resources split across
network namespaces. e.g., something like this:

   namespace 1 |  namespace 2  | ... | namespace N
               |               |     |
  { ports 1 }  |   { ports 2 } | ... | { ports N }
               |               |     |
   devlink 1   |    devlink 2  | ... |  devlink N
  =================================================
                   driver

Can you elaborate further? Ports for most purposes are represented by
netdevices. Devlink port instances expose global topological view of
the ports which is primarily relevant if you can see the entire ASIC.
I think the global configuration and global view of resources is still
the most relevant need, so in your diagram you must account for some
"all-seeing" instance, e.g.:

   namespace 1 |  namespace 2  | ... | namespace N
               |               |     |
  { ports 1 }  |   { ports 2 } | ... | { ports N }
               |               |     |
subdevlink 1   | subdevlink 2  | ... |  subdevlink N
         \______      |              _______/
                 master ASIC devlink
  =================================================
                   driver

No?