Re: [RFC net-next 9/9] net: dsa: mv88e6xxx: MST Offloading

["Russell King (Oracle)" <linux@armlinux.org.uk>]

Hi,

On Wed, Feb 16, 2022 at 02:29:34PM +0100, Tobias Waldekranz wrote:
> +static int mv88e6xxx_sid_new(struct mv88e6xxx_chip *chip, u8 *sid)
> +{
> +	DECLARE_BITMAP(busy, MV88E6XXX_N_SID) = { 0 };
> +	struct mv88e6xxx_mst *mst;
> +
> +	set_bit(0, busy);
> +
> +	list_for_each_entry(mst, &chip->msts, node) {
> +		set_bit(mst->stu.sid, busy);
> +	}

Do you need these set_bit() operations to be atomic? Would __set_bit()
produce better code?

> +
> +	*sid = find_first_zero_bit(busy, MV88E6XXX_N_SID);
> +
> +	return (*sid >= mv88e6xxx_max_sid(chip)) ? -ENOSPC : 0;

Hmm. Let's hope that mv88e6xxx_max_sid() never returns a value larger
than MV88E6XXX_N_SID.

> +}
> +
...
> +static int mv88e6xxx_sid_get(struct mv88e6xxx_chip *chip, struct net_device *br,
> +			     u16 mstid, u8 *sid)
> +{
> +	struct mv88e6xxx_mst *mst;
> +	int err;
> +
> +	if (!br)
> +		return 0;
> +
> +	if (!mv88e6xxx_has_stu(chip))
> +		return -EOPNOTSUPP;
> +
> +	list_for_each_entry(mst, &chip->msts, node) {
> +		if (mst->br == br && mst->mstid == mstid) {
> +			refcount_inc(&mst->refcnt);
> +			*sid = mst->stu.sid;
> +			return 0;
> +		}
> +	}
> +
> +	err = mv88e6xxx_sid_new(chip, sid);
> +	if (err)
> +		return err;
> +
> +	mst = kzalloc(sizeof(*mst), GFP_KERNEL);
> +	if (!mst)
> +		return -ENOMEM;
> +
> +	INIT_LIST_HEAD(&mst->node);

There is no need to initialise the node if you're then going to be
adding it to the list.

> +	refcount_set(&mst->refcnt, 1);
> +	mst->br = br;
> +	mst->mstid = mstid;
> +	mst->stu.valid = true;
> +	mst->stu.sid = *sid;
> +	list_add_tail(&mst->node, &chip->msts);
> +	return mv88e6xxx_stu_loadpurge(chip, &mst->stu);

I haven't checked what the locking is here - I hope it's not possible
for two of these to run concurrently.

Thanks.

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 40Mbps down 10Mbps up. Decent connectivity at last!