Re: [net-next v14 04/12] net: mtip: The L2 switch driver for imx287

[Paolo Abeni <pabeni@redhat.com>]

On 7/1/25 1:49 PM, Lukasz Majewski wrote:
> Changes for v14:
> - Increase the maximal received frame size to 1536 (for VLAN)
> - Use spin_{un}lock_irq{save|restore} when altering dynamic table of the
>   switch and mtip_adjust_link() as both cannot be done when switch IRQ is
>   potentially enabled

Why?

 (the previous one alters entries in switching table
>   the latter one may reset the whole IP block)

What really matters is the scope (process/atomic, bh, hardirq) of the
relevant callers (the functions that do acquire the given locks).


> +/* dynamicms MAC address table learn and migration */
> +static void
> +mtip_atable_dynamicms_learn_migration(struct switch_enet_private *fep,
> +				      int curr_time, unsigned char *mac,
> +				      u8 *rx_port)
> +{
> +	u8 port = MTIP_PORT_FORWARDING_INIT;
> +	struct mtip_port_info *port_info;
> +	u32 rx_mac_lo = 0, rx_mac_hi = 0;
> +	unsigned long flags;
> +	int index;
> +
> +	spin_lock_irqsave(&fep->learn_lock, flags);

If the _irqsave() part is needed (and I don't see why??!) than all the
other `learn_lock` users should also use such variant, unless already in
hardirq scope.

[...]
> +static void mtip_adjust_link(struct net_device *dev)
> +{
> +	struct mtip_ndev_priv *priv = netdev_priv(dev);
> +	struct switch_enet_private *fep = priv->fep;
> +	struct phy_device *phy_dev;
> +	int status_change = 0, idx;
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&fep->hw_lock, flags);

Same here.

/P