Re: [PATCH v10 3/3] net: ax88796c: ASIX AX88796C SPI Ethernet Adapter Driver

On Wed, Jan 20, 2021 at 08:30:14PM +0100, Lukasz Stelmach wrote:

It was <2021-01-15 pią 17:27>, when Jakub Kicinski wrote:

quoted

On Wed, 13 Jan 2021 19:40:28 +0100 Łukasz Stelmach wrote:

quoted

ASIX AX88796[1] is a versatile ethernet adapter chip, that can be
connected to a CPU with a 8/16-bit bus or with an SPI. This driver
supports SPI connection.

The driver has been ported from the vendor kernel for ARTIK5[2]
boards. Several changes were made to adapt it to the current kernel
which include:

+ updated DT configuration,
+ clock configuration moved to DT,
+ new timer, ethtool and gpio APIs,
+ dev_* instead of pr_* and custom printk() wrappers,
+ removed awkward vendor power managemtn.
+ introduced ethtool tunable to control SPI compression

[...]

quoted

quoted

The other ax88796 driver is for NE2000 compatible AX88796L chip. These
chips are not compatible. Hence, two separate drivers are required.

Signed-off-by: Łukasz Stelmach <l.stelmach@samsung.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>

quoted

+static u32 ax88796c_get_priv_flags(struct net_device *ndev)
+{
+	struct ax88796c_device *ax_local = to_ax88796c_device(ndev);
+
+        return ax_local->priv_flags;

stray indent

quoted

+}

Done.

quoted

quoted

+#define MAX(x,y) ((x) > (y) ? (x) : (y))

Please use the standard linux max / max_t macros.

Done.

quoted

quoted

+static struct sk_buff *
+ax88796c_tx_fixup(struct net_device *ndev, struct sk_buff_head *q)
+{
+	struct ax88796c_device *ax_local = to_ax88796c_device(ndev);
+	u8 spi_len = ax_local->ax_spi.comp ? 1 : 4;
+	struct sk_buff *skb;
+	struct tx_pkt_info *info;
+	struct skb_data *entry;
+	u16 pkt_len;
+	u8 padlen, seq_num;
+	u8 need_pages;
+	int headroom;
+	int tailroom;
+
+	if (skb_queue_empty(q))
+		return NULL;
+
+	skb = skb_peek(q);
+	pkt_len = skb->len;
+	need_pages = (pkt_len + TX_OVERHEAD + 127) >> 7;
+	if (ax88796c_check_free_pages(ax_local, need_pages) != 0)
+		return NULL;
+
+	headroom = skb_headroom(skb);
+	tailroom = skb_tailroom(skb);
+	padlen = round_up(pkt_len, 4) - pkt_len;
+	seq_num = ++ax_local->seq_num & 0x1F;
+
+	info = (struct tx_pkt_info *)skb->cb;
+	info->pkt_len = pkt_len;
+
+	if (skb_cloned(skb) ||
+	    (headroom < (TX_OVERHEAD + spi_len)) ||
+	    (tailroom < (padlen + TX_EOP_SIZE))) {
+		size_t h = MAX((TX_OVERHEAD + spi_len) - headroom,0);
+		size_t t = MAX((padlen + TX_EOP_SIZE) - tailroom,0);
+
+		if (pskb_expand_head(skb, h, t, GFP_KERNEL))
+			return NULL;
+	}
+
+	info->seq_num = seq_num;
+	ax88796c_proc_tx_hdr(info, skb->ip_summed);
+
+	/* SOP and SEG header */
+	memcpy(skb_push(skb, TX_OVERHEAD), &info->sop, TX_OVERHEAD);

why use skb->cb to store info? why not declare it on the stack?

Done.

quoted

quoted

+	/* Write SPI TXQ header */
+	memcpy(skb_push(skb, spi_len), ax88796c_tx_cmd_buf, spi_len);
+
+	/* Make 32-bit alignment */
+	skb_put(skb, padlen);
+
+	/* EOP header */
+	memcpy(skb_put(skb, TX_EOP_SIZE), &info->eop, TX_EOP_SIZE);
+
+	skb_unlink(skb, q);
+
+	entry = (struct skb_data *)skb->cb;
+	memset(entry, 0, sizeof(*entry));
+	entry->len = pkt_len;
+
+	if (netif_msg_pktdata(ax_local)) {
+		char pfx[IFNAMSIZ + 7];
+
+		snprintf(pfx, sizeof(pfx), "%s:     ", ndev->name);
+
+		netdev_info(ndev, "TX packet len %d, total len %d, seq %d\n",
+			    pkt_len, skb->len, seq_num);
+
+		netdev_info(ndev, "  SPI Header:\n");
+		print_hex_dump(KERN_INFO, pfx, DUMP_PREFIX_OFFSET, 16, 1,
+			       skb->data, 4, 0);
+
+		netdev_info(ndev, "  TX SOP:\n");
+		print_hex_dump(KERN_INFO, pfx, DUMP_PREFIX_OFFSET, 16, 1,
+			       skb->data + 4, TX_OVERHEAD, 0);
+
+		netdev_info(ndev, "  TX packet:\n");
+		print_hex_dump(KERN_INFO, pfx, DUMP_PREFIX_OFFSET, 16, 1,
+			       skb->data + 4 + TX_OVERHEAD,
+			       skb->len - TX_EOP_SIZE - 4 - TX_OVERHEAD, 0);
+
+		netdev_info(ndev, "  TX EOP:\n");
+		print_hex_dump(KERN_INFO, pfx, DUMP_PREFIX_OFFSET, 16, 1,
+			       skb->data + skb->len - 4, 4, 0);
+	}
+
+	return skb;
+}
+
+static int ax88796c_hard_xmit(struct ax88796c_device *ax_local)
+{
+	struct sk_buff *tx_skb;
+	struct skb_data *entry;
+
+	WARN_ON(!mutex_is_locked(&ax_local->spi_lock));
+
+	tx_skb = ax88796c_tx_fixup(ax_local->ndev, &ax_local->tx_wait_q);
+
+	if (!tx_skb)
+		return 0;

tx_dropped++ ?

Done.

quoted

quoted

+	entry = (struct skb_data *)tx_skb->cb;
+
+	AX_WRITE(&ax_local->ax_spi,
+		 (TSNR_TXB_START | TSNR_PKT_CNT(1)), P0_TSNR);
+
+	axspi_write_txq(&ax_local->ax_spi, tx_skb->data, tx_skb->len);
+
+	if (((AX_READ(&ax_local->ax_spi, P0_TSNR) & TXNR_TXB_IDLE) == 0) ||
+	    ((ISR_TXERR & AX_READ(&ax_local->ax_spi, P0_ISR)) != 0)) {
+		/* Ack tx error int */
+		AX_WRITE(&ax_local->ax_spi, ISR_TXERR, P0_ISR);
+
+		ax_local->stats.tx_dropped++;
+
+		netif_err(ax_local, tx_err, ax_local->ndev,
+			  "TX FIFO error, re-initialize the TX bridge\n");

rate limit

Done.

quoted

quoted

+		/* Reinitial tx bridge */
+		AX_WRITE(&ax_local->ax_spi, TXNR_TXB_REINIT |
+			AX_READ(&ax_local->ax_spi, P0_TSNR), P0_TSNR);
+		ax_local->seq_num = 0;
+	} else {
+		ax_local->stats.tx_packets++;
+		ax_local->stats.tx_bytes += entry->len;
+	}
+
+	entry->state = tx_done;
+	dev_kfree_skb(tx_skb);

dev_consume_skb() is better in cases the xmission was correct.
kfree_skb() shows up in packet drop monitor.

This one is OK as it is because

1. dev_kfree_skb() is consume_skb()

    include/linux/skbuff.h:#define dev_kfree_skb(a) consume_skb(a)

2. dev_consume_skb() does not exist. There are dev_consume_skb_irq() and
dev_consume_skb_any(). The former can be used in IRQs, the latter
anywher and de facto calls the former if in IRQ. If not, it calls
dev_kfree_skb() (see above)

3. Last but not least. kfree_skb() and consume_skb() become the same
without CONFIG_TRACEPOINTS (commit be769db2f958).

quoted

quoted

+
+	return 1;
+}

quoted

+static void
+ax88796c_skb_return(struct ax88796c_device *ax_local, struct sk_buff *skb,
+		    struct rx_header *rxhdr)
+{
+	struct net_device *ndev = ax_local->ndev;
+	int status;
+
+	do {
+		if (!(ndev->features & NETIF_F_RXCSUM))
+			break;
+
+		/* checksum error bit is set */
+		if ((rxhdr->flags & RX_HDR3_L3_ERR) ||
+		    (rxhdr->flags & RX_HDR3_L4_ERR))
+			break;
+
+		/* Other types may be indicated by more than one bit. */
+		if ((rxhdr->flags & RX_HDR3_L4_TYPE_TCP) ||
+		    (rxhdr->flags & RX_HDR3_L4_TYPE_UDP))
+			skb->ip_summed = CHECKSUM_UNNECESSARY;
+	} while (0);
+
+	ax_local->stats.rx_packets++;
+	ax_local->stats.rx_bytes += skb->len;
+	skb->dev = ndev;
+
+	skb->protocol = eth_type_trans(skb, ax_local->ndev);
+
+	netif_info(ax_local, rx_status, ndev, "< rx, len %zu, type 0x%x\n",
+		   skb->len + sizeof(struct ethhdr), skb->protocol);
+
+	status = netif_rx_ni(skb);
+	if (status != NET_RX_SUCCESS)
+		netif_info(ax_local, rx_err, ndev,
+			   "netif_rx status %d\n", status);

rate limit

Done.

quoted

quoted

+}
+
+static void
+ax88796c_rx_fixup(struct ax88796c_device *ax_local, struct sk_buff *rx_skb)
+{
+	struct rx_header *rxhdr = (struct rx_header *)rx_skb->data;
+	struct net_device *ndev = ax_local->ndev;
+	u16 len;
+
+	be16_to_cpus(&rxhdr->flags_len);
+	be16_to_cpus(&rxhdr->seq_lenbar);
+	be16_to_cpus(&rxhdr->flags);
+
+	if (((rxhdr->flags_len) & RX_HDR1_PKT_LEN) !=
+			 (~(rxhdr->seq_lenbar) & 0x7FF)) {

Lots of unnecessary parenthesis.

Done.

quoted

quoted

+		netif_err(ax_local, rx_err, ndev, "Header error\n");
+
+		ax_local->stats.rx_frame_errors++;
+		kfree_skb(rx_skb);
+		return;
+	}
+
+	if ((rxhdr->flags_len & RX_HDR1_MII_ERR) ||
+	    (rxhdr->flags_len & RX_HDR1_CRC_ERR)) {
+		netif_err(ax_local, rx_err, ndev, "CRC or MII error\n");
+
+		ax_local->stats.rx_crc_errors++;
+		kfree_skb(rx_skb);
+		return;
+	}
+
+	len = rxhdr->flags_len & RX_HDR1_PKT_LEN;
+	if (netif_msg_pktdata(ax_local)) {
+		char pfx[IFNAMSIZ + 7];
+
+		snprintf(pfx, sizeof(pfx), "%s:     ", ndev->name);
+		netdev_info(ndev, "RX data, total len %d, packet len %d\n",
+			    rx_skb->len, len);
+
+		netdev_info(ndev, "  Dump RX packet header:");
+		print_hex_dump(KERN_INFO, pfx, DUMP_PREFIX_OFFSET, 16, 1,
+			       rx_skb->data, sizeof(*rxhdr), 0);
+
+		netdev_info(ndev, "  Dump RX packet:");
+		print_hex_dump(KERN_INFO, pfx, DUMP_PREFIX_OFFSET, 16, 1,
+			       rx_skb->data + sizeof(*rxhdr), len, 0);
+	}
+
+	skb_pull(rx_skb, sizeof(*rxhdr));
+	pskb_trim(rx_skb, len);
+
+	ax88796c_skb_return(ax_local, rx_skb, rxhdr);
+}
+
+static int ax88796c_receive(struct net_device *ndev)
+{
+	struct ax88796c_device *ax_local = to_ax88796c_device(ndev);
+	struct skb_data *entry;
+	u16 w_count, pkt_len;
+	struct sk_buff *skb;
+	u8 pkt_cnt;
+
+	WARN_ON(!mutex_is_locked(&ax_local->spi_lock));
+
+	/* check rx packet and total word count */
+	AX_WRITE(&ax_local->ax_spi, AX_READ(&ax_local->ax_spi, P0_RTWCR)
+		  | RTWCR_RX_LATCH, P0_RTWCR);
+
+	pkt_cnt = AX_READ(&ax_local->ax_spi, P0_RXBCR2) & RXBCR2_PKT_MASK;
+	if (!pkt_cnt)
+		return 0;
+
+	pkt_len = AX_READ(&ax_local->ax_spi, P0_RCPHR) & 0x7FF;
+
+	w_count = ((pkt_len + 6 + 3) & 0xFFFC) >> 1;

w_count = round_up(pkt_len + 6, 4) >> 1;

Done.

quoted

quoted

+	skb = netdev_alloc_skb(ndev, (w_count * 2));

parenthesis unnecessary

Done.

quoted

quoted

+	if (!skb) {
+		AX_WRITE(&ax_local->ax_spi, RXBCR1_RXB_DISCARD, P0_RXBCR1);

Increment rx_dropped counter here?

Done.

quoted

quoted

+		return 0;
+	}
+	entry = (struct skb_data *)skb->cb;
+
+	AX_WRITE(&ax_local->ax_spi, RXBCR1_RXB_START | w_count, P0_RXBCR1);
+
+	axspi_read_rxq(&ax_local->ax_spi,
+		       skb_put(skb, w_count * 2), skb->len);
+
+	/* Check if rx bridge is idle */
+	if ((AX_READ(&ax_local->ax_spi, P0_RXBCR2) & RXBCR2_RXB_IDLE) == 0) {
+		netif_err(ax_local, rx_err, ndev,
+			  "Rx Bridge is not idle\n");

rate limit?

Ok.

quoted

quoted

+		AX_WRITE(&ax_local->ax_spi, RXBCR2_RXB_REINIT, P0_RXBCR2);
+
+		entry->state = rx_err;
+	} else {
+		entry->state = rx_done;
+	}
+
+	AX_WRITE(&ax_local->ax_spi, ISR_RXPKT, P0_ISR);
+
+	ax88796c_rx_fixup(ax_local, skb);
+
+	return 1;
+}
+
+static int ax88796c_process_isr(struct ax88796c_device *ax_local)
+{
+	struct net_device *ndev = ax_local->ndev;
+	u8 done = 0;

The logic associated with this variable is "is there more to do" rather
than "done", no?

ax88796c_receive() returns 1, if there may be something more to do. So
yes, let's rename it to todo.

quoted

quoted

+	u16 isr;
+
+	WARN_ON(!mutex_is_locked(&ax_local->spi_lock));
+
+	isr = AX_READ(&ax_local->ax_spi, P0_ISR);
+	AX_WRITE(&ax_local->ax_spi, isr, P0_ISR);
+
+	netif_dbg(ax_local, intr, ndev, "  ISR 0x%04x\n", isr);
+
+	if (isr & ISR_TXERR) {
+		netif_dbg(ax_local, intr, ndev, "  TXERR interrupt\n");
+		AX_WRITE(&ax_local->ax_spi, TXNR_TXB_REINIT, P0_TSNR);
+		ax_local->seq_num = 0x1f;
+	}
+
+	if (isr & ISR_TXPAGES) {
+		netif_dbg(ax_local, intr, ndev, "  TXPAGES interrupt\n");
+		set_bit(EVENT_TX, &ax_local->flags);
+	}
+
+	if (isr & ISR_LINK) {
+		netif_dbg(ax_local, intr, ndev, "  Link change interrupt\n");
+		phy_mac_interrupt(ax_local->ndev->phydev);
+	}
+
+	if (isr & ISR_RXPKT) {
+		netif_dbg(ax_local, intr, ndev, "  RX interrupt\n");
+		done = ax88796c_receive(ax_local->ndev);
+	}
+
+	return done;
+}

quoted

+static void ax88796c_work(struct work_struct *work)
+{
+	struct ax88796c_device *ax_local =
+			container_of(work, struct ax88796c_device, ax_work);
+
+	mutex_lock(&ax_local->spi_lock);
+
+	if (test_bit(EVENT_SET_MULTI, &ax_local->flags)) {
+		ax88796c_set_hw_multicast(ax_local->ndev);
+		clear_bit(EVENT_SET_MULTI, &ax_local->flags);
+	}
+
+	if (test_bit(EVENT_INTR, &ax_local->flags)) {
+		AX_WRITE(&ax_local->ax_spi, IMR_MASKALL, P0_IMR);
+
+		while (1) {
+			if (!ax88796c_process_isr(ax_local))
+				break;

while (ax88796c_process_isr(ax_local))
	/* nothing */;
?

Ok.

quoted

quoted

+		}
+
+		clear_bit(EVENT_INTR, &ax_local->flags);
+
+		AX_WRITE(&ax_local->ax_spi, IMR_DEFAULT, P0_IMR);
+
+		enable_irq(ax_local->ndev->irq);
+	}
+
+	if (test_bit(EVENT_TX, &ax_local->flags)) {
+		while (skb_queue_len(&ax_local->tx_wait_q)) {
+			if (!ax88796c_hard_xmit(ax_local))
+				break;
+		}
+
+		clear_bit(EVENT_TX, &ax_local->flags);
+
+		if (netif_queue_stopped(ax_local->ndev) &&
+		    (skb_queue_len(&ax_local->tx_wait_q) < TX_QUEUE_LOW_WATER))
+			netif_wake_queue(ax_local->ndev);
+	}
+
+	mutex_unlock(&ax_local->spi_lock);
+}

quoted

+static void ax88796c_set_csums(struct ax88796c_device *ax_local)
+{
+	struct net_device *ndev = ax_local->ndev;
+
+	WARN_ON(!mutex_is_locked(&ax_local->spi_lock));

lockdep_assert_held() in all those cases

Done.

quoted

quoted

+static void ax88796c_free_skb_queue(struct sk_buff_head *q)
+{
+	struct sk_buff *skb;
+
+	while (q->qlen) {
+		skb = skb_dequeue(q);
+		kfree_skb(skb);
+	}

__skb_queue_purge()

Done.

quoted

quoted

+}
+
+static int
+ax88796c_close(struct net_device *ndev)
+{
+	struct ax88796c_device *ax_local = to_ax88796c_device(ndev);
+
+	netif_stop_queue(ndev);
+	phy_stop(ndev->phydev);
+
+	mutex_lock(&ax_local->spi_lock);
+
+	/* Disable MAC interrupts */
+	AX_WRITE(&ax_local->ax_spi, IMR_MASKALL, P0_IMR);
+	ax88796c_free_skb_queue(&ax_local->tx_wait_q);
+	ax88796c_soft_reset(ax_local);
+
+	mutex_unlock(&ax_local->spi_lock);
+
+	free_irq(ndev->irq, ndev);
+
+	return 0;
+}

quoted

+struct ax88796c_device {
+	struct spi_device	*spi;
+	struct net_device	*ndev;
+	struct net_device_stats	stats;

You need to use 64 bit stats, like struct rtnl_link_stats64.
On a 32bit system at 100Mbps ulong can wrap in minutes.

Let me see. At first glance

git grep -l ndo_get_stats\\\> drivers/net/ethernet/  | xargs grep -li SPEED_100\\\>

quite a number of Fast Ethernet drivers use net_device_stats. Let me
calculate.

- bytes
  100Mbps is ~10MiB/s
  sending 4GiB at 10MiB/s takes 27 minutes

- packets
  minimum frame size is 84 bytes (840 bits on the wire) on 100Mbps means
  119048 pps at this speed it takse 10 hours to transmit 2^32 packets

Anyway, I switched to rtnl_link_stats64. Tell me, is it OK to just
memcpy() in .ndo_get_stats64?

You should uses the helpers in include/linux/u64_stats_sync.h to make
sure the synchronization between readers and writers is correct. A
memcpy() within those primitives should be O.K.

	 Andrew