Re: [RFC PATCH 6/9] net: dsa: hellcreek: Add debugging mechanisms

On Mon Jun 22 2020, Andrew Lunn wrote:

On Mon, Jun 22, 2020 at 02:32:28PM +0200, Kurt Kanzenbach wrote:

quoted

On Fri Jun 19 2020, Andrew Lunn wrote:

quoted

quoted

quoted

Are trace registers counters?

No. The trace registers provide bits for error conditions and if packets
have been dropped e.g. because of full queues or FCS errors, and so on.

Is there some documentation somewhere? A better understanding of what
they can do might help figure out the correct API.

No, not that I'm aware of.

Actually there are a few more debugging mechanisms and features which
should be exposed somehow. Here's the list:

 * Trace registers for the error conditions. This feature needs to be
   configured for which ports should be traced
 * Memory registers for indicating how many free page and meta pointers
   are available (read-only)
 * Limit registers for configuring:
   * Maximum memory limit per port
   * Reserved memory for critical traffic
   * Background traffic rate
   * Maximum queue depth
 * Re-prioritization of packets based on the ether type (not mac address)
 * Packet logging (-> retrieval of packet time stamps) based on port, traffic class and direction
 * Queue tracking

What API would be useful for these mechanisms?

Hi Kurt

You should take a look at devlink. Many of these fit devlink
resources. Use that where it fits. But you might end up with an out of
tree debugfs patch for your own debugging work. We have something
similar of mv88e6xxx.

I see. Maybe I'll keep the debug stuff out of tree for now and will come
back to it later.

Thanks,
Kurt