Re: [PATCH net-next v2 1/2] hinic: add rss support

[Jakub Kicinski <jakub.kicinski@netronome.com>]

On Tue, 11 Jun 2019 18:12:33 +0000, Xue Chaojing wrote:
> This patch adds rss support for the HINIC driver.
> 
> Signed-off-by: Xue Chaojing <xuechaojing@huawei.com>

> +static int hinic_rss_init(struct hinic_dev *nic_dev)
> +{
> +	u8 default_rss_key[HINIC_RSS_KEY_SIZE] = {
> +			0x6d, 0x5a, 0x56, 0xda, 0x25, 0x5b, 0x0e, 0xc2,
> +			0x41, 0x67, 0x25, 0x3d, 0x43, 0xa3, 0x8f, 0xb0,
> +			0xd0, 0xca, 0x2b, 0xcb, 0xae, 0x7b, 0x30, 0xb4,
> +			0x77, 0xcb, 0x2d, 0xa3, 0x80, 0x30, 0xf2, 0x0c,
> +			0x6a, 0x42, 0xb7, 0x3b, 0xbe, 0xac, 0x01, 0xfa};

netdev_rss_key_fill()

> +	u32 indir_tbl[HINIC_RSS_INDIR_SIZE] = { 0 };
> +	u8 tmpl_idx = nic_dev->rss_tmpl_idx;
> +	int err, i;
> +
> +	for (i = 0; i < HINIC_RSS_INDIR_SIZE; i++)
> +		indir_tbl[i] = (i / HINIC_RSS_INDIR_SIZE) * nic_dev->num_rss +
> +				i % nic_dev->num_rss;
> +
> +	err = hinic_rss_set_template_tbl(nic_dev, tmpl_idx, default_rss_key);
> +	if (err)
> +		return err;
> +
> +	err = hinic_rss_set_indir_tbl(nic_dev, tmpl_idx, indir_tbl);
> +	if (err)
> +		return err;
> +
> +	err = hinic_set_rss_type(nic_dev, tmpl_idx, nic_dev->rss_type);
> +	if (err)
> +		return err;
> +
> +	err = hinic_rss_set_hash_engine(nic_dev, tmpl_idx,
> +					nic_dev->rss_hash_engine);
> +	if (err)
> +		return err;
> +
> +	err = hinic_rss_cfg(nic_dev, 1, tmpl_idx);
> +	if (err)
> +		return err;
> +
> +	return 0;
> +}
> +
> +static void hinic_rss_deinit(struct hinic_dev *nic_dev)
> +{
> +	hinic_rss_cfg(nic_dev, 0, nic_dev->rss_tmpl_idx);
> +}
> +
> +static void hinic_init_rss_parameters(struct hinic_dev *nic_dev)
> +{
> +	nic_dev->rss_hash_engine = HINIC_RSS_HASH_ENGINE_TYPE_XOR;
> +	nic_dev->rss_type.tcp_ipv6_ext = 1;
> +	nic_dev->rss_type.ipv6_ext = 1;
> +	nic_dev->rss_type.tcp_ipv6 = 1;
> +	nic_dev->rss_type.ipv6 = 1;
> +	nic_dev->rss_type.tcp_ipv4 = 1;
> +	nic_dev->rss_type.ipv4 = 1;
> +	nic_dev->rss_type.udp_ipv6 = 1;
> +	nic_dev->rss_type.udp_ipv4 = 1;

Usually UDP is disabled by default because fragmentation leads to
reorders (NICs file all fragmented packets to queue 0 while other
packets are distributed by RSS).

> +}
> +
> +static void hinic_enable_rss(struct hinic_dev *nic_dev)
> +{
> +	struct net_device *netdev = nic_dev->netdev;
> +	struct hinic_hwdev *hwdev = nic_dev->hwdev;
> +	struct hinic_hwif *hwif = hwdev->hwif;
> +	struct pci_dev *pdev = hwif->pdev;
> +	int i, node, err = 0;
> +	u16 num_cpus = 0;
> +
> +	nic_dev->max_qps = hinic_hwdev_max_num_qps(hwdev);
> +	if (nic_dev->max_qps <= 1) {
> +		nic_dev->flags &= ~HINIC_RSS_ENABLE;
> +		nic_dev->rss_limit = nic_dev->max_qps;
> +		nic_dev->num_qps = nic_dev->max_qps;
> +		nic_dev->num_rss = nic_dev->max_qps;
> +
> +		return;
> +	}
> +
> +	err = hinic_rss_template_alloc(nic_dev, &nic_dev->rss_tmpl_idx);
> +	if (err) {
> +		netif_err(nic_dev, drv, netdev,
> +			  "Failed to alloc tmpl_idx for rss, can't enable rss for this function\n");
> +		nic_dev->flags &= ~HINIC_RSS_ENABLE;
> +		nic_dev->max_qps = 1;
> +		nic_dev->rss_limit = nic_dev->max_qps;
> +		nic_dev->num_qps = nic_dev->max_qps;
> +		nic_dev->num_rss = nic_dev->max_qps;
> +
> +		return;
> +	}
> +
> +	nic_dev->flags |= HINIC_RSS_ENABLE;
> +
> +	for (i = 0; i < num_online_cpus(); i++) {
> +		node = cpu_to_node(i);
> +		if (node == dev_to_node(&pdev->dev))
> +			num_cpus++;
> +	}
> +
> +	if (!num_cpus)
> +		num_cpus = num_online_cpus();
> +
> +	nic_dev->num_qps = min_t(u16, nic_dev->max_qps, num_cpus);

We generally use netif_get_num_default_rss_queues() for RX queues
and num_online_cpus() for TX queues but I'm not sure you can do
different counts, so it's probably fine.