Re: [PATCH net v7] net: dsa: qca8k: preserve port state across MTU changes

Hi Andrew,

On Tue, Oct 27, 2026 at 01:54:16PM +0100, Andrew Lunn wrote:

Is this a real problem which bothers somebody? If it has not been seen
in production, it sounds like it does not bother anybody. So net-next.

It has not been seen in production. Following your suggestion, I will
move it to net-next.

v7 needs another revision anyway. Sashiko's review [1] is right: a
concurrent PORT_STATUS reader such as the PCS poll can corrupt the MTU
snapshot, because the qca8k regmap is created with disable_locking and
shares its work_buf. Clearing the MAC bits unconditionally would not be
enough, since the restore uses the same snapshot and the MAX_FRAME_SIZE
write also goes through work_buf. Stanislaw's patch [2] enables the
regmap lock. With it, the v7 code passes a local KUnit test that forces
this interleaving against the real regmap core.

I will send v8 for net-next once it reopens and [2] is in net-next, and
say in the commit message how the problem was found, as sinfo asked.

[1] https://lore.kernel.org/netdev/179303171241.3134.15019965307136674652@kernel.org/ (local)
[2] https://lore.kernel.org/netdev/20261022192933.384898-2-kuncy7@gmail.com/ (local)

pw-bot: cr