Re: [PATCH v5 net-next 6/6] net: ethernet: ti: cpsw: add XDP support

[Ivan Khoronzhuk]

On Mon, Jul 01, 2019 at 06:19:01PM +0200, Jesper Dangaard Brouer wrote:
> On Sun, 30 Jun 2019 20:23:48 +0300
> Ivan Khoronzhuk <ivan.khoronzhuk@xxxxxxxxxx> wrote:
>
> > +static int cpsw_ndev_create_xdp_rxq(struct cpsw_priv *priv, int ch)
> > +{
> > +	struct cpsw_common *cpsw = priv->cpsw;
> > +	int ret, new_pool = false;
> > +	struct xdp_rxq_info *rxq;
> > +
> > +	rxq = &priv->xdp_rxq[ch];
> > +
> > +	ret = xdp_rxq_info_reg(rxq, priv->ndev, ch);
> > +	if (ret)
> > +		return ret;
> > +
> > +	if (!cpsw->page_pool[ch]) {
> > +		ret =  cpsw_create_rx_pool(cpsw, ch);
> > +		if (ret)
> > +			goto err_rxq;
> > +
> > +		new_pool = true;
> > +	}
> > +
> > +	ret = xdp_rxq_info_reg_mem_model(rxq, MEM_TYPE_PAGE_POOL,
> > +					 cpsw->page_pool[ch]);
> > +	if (!ret)
> > +		return 0;
> > +
> > +	if (new_pool) {
> > +		page_pool_free(cpsw->page_pool[ch]);
> > +		cpsw->page_pool[ch] = NULL;
> > +	}
> > +
> > +err_rxq:
> > +	xdp_rxq_info_unreg(rxq);
> > +	return ret;
> > +}
>
> Looking at this, and Ilias'es XDP-netsec error handling path, it might
> be a mistake that I removed page_pool_destroy() and instead put the
> responsibility on xdp_rxq_info_unreg().
As for me this is started not from page_pool_free, but rather from calling
unreg_mem_model from rxq_info_unreg. Then, if page_pool_free is hidden
it looks more a while normal to move all chain to be self destroyed.

> As here, we have to detect if page_pool_create() was a success, and then
> if xdp_rxq_info_reg_mem_model() was a failure, explicitly call
> page_pool_free() because the xdp_rxq_info_unreg() call cannot "free"
> the page_pool object given it was not registered.
Yes, it looked a little bit ugly from the beginning, but, frankly,
I have got used to this already.

> Ivan's patch in[1], might be a better approach, which forced all
> drivers to explicitly call page_pool_free(), even-though it just
> dec-refcnt and the real call to page_pool_free() happened via
> xdp_rxq_info_unreg().
>
> To better handle error path, I would re-introduce page_pool_destroy(),
So, you might to do it later as I understand, and not for my special
case but becouse it makes error path to look a little bit more pretty.
I'm perfectly fine with this, and better you add this, for now my
implementation requires only "xdp: allow same allocator usage" patch,
but if you insist I can resend also patch in question afterwards my
series is applied (with modification to cpsw & netsec & mlx5 & page_pool).

What's your choice? I can add to your series patch needed for cpsw to
avoid some misuse.

> as a driver API, that would gracefully handle NULL-pointer case, and
> then call page_pool_free() with the atomic_dec_and_test().  (It should
> hopefully simplify the error handling code a bit)
>
> [1] https://lore.kernel.org/netdev/20190625175948.24771-2-ivan.khoronzhuk@xxxxxxxxxx/
>
>
> > +void cpsw_ndev_destroy_xdp_rxqs(struct cpsw_priv *priv)
> > +{
> > +	struct cpsw_common *cpsw = priv->cpsw;
> > +	struct xdp_rxq_info *rxq;
> > +	int i;
> > +
> > +	for (i = 0; i < cpsw->rx_ch_num; i++) {
> > +		rxq = &priv->xdp_rxq[i];
> > +		if (xdp_rxq_info_is_reg(rxq))
> > +			xdp_rxq_info_unreg(rxq);
> > +	}
> > +}
>
> Are you sure you need to test xdp_rxq_info_is_reg() here?
Yes it's required in my case as it's used in error path where
an rx queue can be even not registered and no need in this warn.

> You should just call xdp_rxq_info_unreg(rxq), if you know that this rxq
> should be registered.  If your assumption failed, you will get a
> WARNing, and discover your driver level bug.  This is one of the ways
> the API is designed to "detect" misuse of the API.  (I found this
> rather useful, when I converted the approx 12 drivers using this
> xdp_rxq_info API).
>
> --
> Best regards,
>  Jesper Dangaard Brouer
>  MSc.CS, Principal Kernel Engineer at Red Hat
>  LinkedIn: http://www.linkedin.com/in/brouer

--
Regards,
Ivan Khoronzhuk