Re: [PATCH net-next v8 2/7] net: pse-pd: fire lifecycle events on controller register/unregister

[Carlo Szelinsky <github@szelinsky.de>]

On Mon, 05 Oct 2026 17:33:25 +0000, netdev-bot+sashiko@kernel.org wrote:
> [Severity: Low]
> Once the net fix lands, will this paragraph be out of date? Could the
> message refer to the net fix instead of saying that this commit closes
> the race?

Agreed. In v9 the reordering is its own patch in the PSE core fixes
series, so this commit message no longer talks about it.

> [Severity: Medium]
> This isn't a bug introduced by this patch, but the error paths in
> pse_controller_register() leak memory.
...
> Should the kfifo be sized after that default is set?

The leaks are fixed by the core fixes series. The kfifo size is an
older bug; its fix goes to net.

> [Severity: Medium]
> This is a pre-existing issue, but can this disable_irq() mask an
> interrupt line that the PSE core never owned?

Yes, if the IRQ request failed. That is an older bug; the fix goes to
net: pcdev->irq is cleared again when the request fails.

> [Severity: Medium]
> With disable_irq() now running before the event, can a port be powered on
> from inside the PSE_UNREGISTERED walk?
...
> Is that case worth handling, or is the brief power-on acceptable here?

The brief power-on is acceptable. The same walk switches the port off
again when its PHY drops its handle. A port could only stay on if an
earlier pi_disable() had failed, and I don't think that case is worth
extra code.

> [Severity: High]
> At this point in the series, can ntf_work be queued again after this
> drain?

Yes, with only this patch applied, as the commit message says. The phy
patch at the end of the series closes it, as the review describes.

> [Severity: High]
> This isn't a bug introduced by this patch, but can a shared power domain
> be freed here while another controller still uses it?

Yes. That is an older bug; its fix goes to net: the power domain is no
longer devm memory and is freed only when its last user drops it.

> [Severity: High]
> This is a pre-existing issue, but at this commit, can PHY-held
> pse_control handles reach the freed pcdev->pi and pcdev after unbind?

Yes. The phy patch at the end of the series fixes it, as the review
says.

Carlo