Re: [dpdk-dev] [PATCH v2 1/6] net/af_xdp: introduce AF XDP PMD driver

[Ye Xiaolong <xiaolong.ye@intel.com>]

Hi, Stephen

On 03/19, Stephen Hemminger wrote:
>Lots of little review comments. This is what I saw in 30 minutes.
>Expect more.

Thanks for taking time to review my patch. They are all valuable inputs.

>
>
>On Tue, 19 Mar 2019 15:12:51 +0800
>Xiaolong Ye <xiaolong.ye@intel.com> wrote:
>
>> +	nb_pkts = nb_pkts < ETH_AF_XDP_RX_BATCH_SIZE ?
>> +		nb_pkts : ETH_AF_XDP_RX_BATCH_SIZE;
>
>Maybe use RTE_MIN() ?

Will do.

>
>> +		mbuf = rte_pktmbuf_alloc(rxq->mb_pool);
>> +		if (mbuf) {
>> +			memcpy(rte_pktmbuf_mtod(mbuf, void*), pkt, len);
>
>Space necessary in "void*"
>Use rte_memcpy.

Will do.

>
>> +static void pull_umem_cq(struct xsk_umem_info *umem, int size)
>> +{
>> +	struct xsk_ring_cons *cq = &umem->cq;
>> +	int i, n;
>> +	uint32_t idx_cq;
>> +	uint64_t addr;
>> +
>> +	n = xsk_ring_cons__peek(cq, size, &idx_cq);
>> +	if (n > 0) {
>> +		for (i = 0; i < n; i++) {
>
>You don't need the if (n > 0) since if n <= 0 the for loop
>would happen 0 times.

Will remove the redundant "if (n > 0)"

>
>> +
>> +static void kick_tx(struct pkt_tx_queue *txq)
>> +{
>> +	struct xsk_umem_info *umem = txq->pair->umem;
>> +	int ret;
>> +
>> +	while (1) {
>> +		ret = sendto(xsk_socket__fd(txq->pair->xsk), NULL, 0,
>> +			     MSG_DONTWAIT, NULL, 0);
>> +
>> +		/* everything is ok */
>> +		if (ret >= 0)
>> +			break;
>
>I would prefer:
>	while ((send(xsk_socket__fd(fd, NULL, 0, MSG_DONTWAIT) < 0) {
>
>Because:
>	- use while() to make looping clearer rather than while(1)
>	- use send() rather than sendto() because you aren't sending with addr
>	- you don't care about return value (ie.ret)
>
>> +
>> +		/* some thing unexpected */
>> +		if (errno != EBUSY && errno != EAGAIN)
>> +			break;
>
>What about EINTR
>
>
>
>> +static void
>> +eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
>> +{
>> +	struct pmd_internals *internals = dev->data->dev_private;
>> +
>> +	dev_info->if_index = internals->if_index;
>> +	dev_info->max_mac_addrs = 1;
>> +	dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
>
>Cast here is unnecessary.

Got it.

>
>> +	dev_info->max_rx_queues = 1;
>> +	dev_info->max_tx_queues = 1;
>> +	dev_info->min_rx_bufsize = 0;
>
>dev_info is already zero, don't need to fill other values.

Got it.

>
>> +
>> +static void
>> +eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
>> +{
>> +	struct pmd_internals *internals = dev->data->dev_private;
>> +
>> +	dev_info->if_index = internals->if_index;
>> +	dev_info->max_mac_addrs = 1;
>> +	dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
>> +	dev_info->max_rx_queues = 1;
>> +	dev_info->max_tx_queues = 1;
>> +	dev_info->min_rx_bufsize = 0;
>> +
>> +	dev_info->default_rxportconf.nb_queues = 1;
>> +	dev_info->default_txportconf.nb_queues = 1;
>> +	dev_info->default_rxportconf.ring_size = ETH_AF_XDP_DFLT_NUM_DESCS;
>> +	dev_info->default_txportconf.ring_size = ETH_AF_XDP_DFLT_NUM_DESCS;
>> +}
>> +
>> +static int
>> +eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
>> +{
>> +	struct pmd_internals *internals = dev->data->dev_private;
>> +	struct xdp_statistics xdp_stats;
>> +	struct pkt_rx_queue *rxq;
>> +	socklen_t optlen;
>> +	int i;
>> +
>> +	optlen = sizeof(struct xdp_statistics);
>
>In theory each call to getsockopt() could change or reduce the value of
>optlen. Best to initialize in loop before each call.

Got it.

>
>> +	for (i = 0; i < dev->data->nb_rx_queues; i++) {
>> +		rxq = &internals->rx_queues[i];
>> +		stats->q_ipackets[i] = internals->rx_queues[i].rx_pkts;
>> +		stats->q_ibytes[i] = internals->rx_queues[i].rx_bytes;
>> +
>> +		stats->q_opackets[i] = internals->tx_queues[i].tx_pkts;
>> +		stats->q_obytes[i] = internals->tx_queues[i].tx_bytes;
>> +
>> +		stats->ipackets += stats->q_ipackets[i];
>> +		stats->ibytes += stats->q_ibytes[i];
>> +		stats->imissed += internals->rx_queues[i].rx_dropped;
>> +		getsockopt(xsk_socket__fd(rxq->xsk), SOL_XDP, XDP_STATISTICS,
>> +				&xdp_stats, &optlen);
>
>You need to check return value of getsockopt() otherwise coverity will complain.

Got it.

>
>> +static void
>> +eth_stats_reset(struct rte_eth_dev *dev)
>> +{
>> +	struct pmd_internals *internals = dev->data->dev_private;
>> +	int i;
>> +
>> +	for (i = 0; i < ETH_AF_XDP_MAX_QUEUE_PAIRS; i++) {
>> +		internals->rx_queues[i].rx_pkts = 0;
>> +		internals->rx_queues[i].rx_bytes = 0;
>> +		internals->rx_queues[i].rx_dropped = 0;
>> +
>> +		internals->tx_queues[i].tx_pkts = 0;
>> +		internals->tx_queues[i].err_pkts = 0;
>> +		internals->tx_queues[i].tx_bytes = 0;
>
>Put all the statistics together and use memset?

Sounds good, will do.

>
>> +static struct xsk_umem_info *xdp_umem_configure(void)
>> +{
>> +	struct xsk_umem_info *umem;
>> +	struct xsk_umem_config usr_config = {
>> +		.fill_size = ETH_AF_XDP_DFLT_NUM_DESCS,
>> +		.comp_size = ETH_AF_XDP_DFLT_NUM_DESCS,
>> +		.frame_size = ETH_AF_XDP_FRAME_SIZE,
>> +		.frame_headroom = ETH_AF_XDP_DATA_HEADROOM };
>> +	void *bufs = NULL;
>> +	char ring_name[0x100];
>
>0x100 is unconvential here. Instead use RTE_RING_NAMESIZE
>but variable is unnecessary, see below

Got it.

>
>> +	int ret;
>> +	uint64_t i;
>> +
>> +	umem = calloc(1, sizeof(*umem));
>
>Why not use rte_zmalloc_node to:
>   1. work with primary/secondary
>   2. guarantee memory on correct numa node?

Will do.

>
>> +	if (!umem) {
>> +		RTE_LOG(ERR, AF_XDP, "Failed to allocate umem info");
>> +		return NULL;
>> +	}
>> +
>> +	snprintf(ring_name, 0x100, "af_xdp_ring");
>
>If ring is always the same, why copy it. Just use string literal

Got it.

>
>> +/** parse name argument */
>> +static int
>> +parse_name_arg(const char *key __rte_unused,
>> +	       const char *value, void *extra_args)
>> +{
>> +	char *name = extra_args;
>> +
>> +	if (strlen(value) > IFNAMSIZ) {
>
>Why not:
>	if (strnlen(value, IFNAMSIZ) >= IFNAMSIZ) {

Will do.

>
>> +
>> +static int
>> +init_internals(struct rte_vdev_device *dev,
>> +	       const char *if_name,
>> +	       int queue_idx,
>> +	       struct rte_eth_dev **eth_dev)
>
>If you changed the function to return the new eth_dev (and return NULL on error)
>then you wouldn't have to pass eth_dev by reference.
>
>static struct rte_eth_dev *
>allocate_ethdev(struct rte_vdev_device *dev, const char *if_name, uint16_t queue_idx)

Sounds better, will do.

>{ 
>
>> +{
>> +	const char *name = rte_vdev_device_name(dev);
>> +	const unsigned int numa_node = dev->device.numa_node;
>> +	struct pmd_internals *internals = NULL;
>
>Useless initialization, first thing you do is allocate this.

Got it.

Thanks,
Xiaolong

>
>
>> +	int ret;
>> +	int i;
>> +
>> +	internals = rte_zmalloc_socket(name, sizeof(*internals), 0, numa_node);
>