On Wed, Feb 18, 2026 at 04:08:17PM +0000, Russell King (Oracle) wrote:

On Wed, Feb 18, 2026 at 06:05:51PM +0200, Vladimir Oltean wrote:

quoted

There are multiple levels at which this could be avoided:
- add an "if (dp->pl)" in sja1105_static_config_reload()
- make the phylink replay helpers NULL-tolerant
- mark ports as DSA_PORT_TYPE_UNUSED after dsa_port_phylink_destroy()
  has run, such that subsequent dsa_switch_for_each_available_port()
  iterations skip them
- disconnect the entire switch at once from switchdev and
  NETDEV_CHANGEUPPER events while unbinding, not just port by port,
  likely using a "ds->unbinding = true" mechanism or similar

however options 3 and 4 are quite heavy and might have side effects.
Although 2 allows to keep the driver simpler, the phylink API it not
NULL-tolerant in general and is not responsible for the NULL pointer
(this is something done by dsa_port_phylink_destroy()). So I went
with 1.

^^^^^^^^

quoted

v1->v2: select option 1 instead of 2

Given this, I think the paragraph above needs updating.

Sorry, I don't understand what needs updating.