mvpp2: Ensure driver works with netmap buffers in pools

The previous change for populating the buffer pools with netmap buffers
was incomplete in several ways:
* The native driver resets the buffer offset when the mvpp2_open is
  called because it initialises the rx queues again. (e.g. when the
  device is made admin up)
* Changing the mtu potentially switches between per-cpu and shared
  buffer pools which will interfere with the netmap use of queues.
* Changing the mtu can alter which pools the port's queues may use which
  would interfere with the netmap pool use.

To correct these:
* Hook the rxq_offset_set call from the rxq init and set it according to
  netmap on state.
* Hook the mtu change function to avoid the various troublesome
  interactions with the native driver, refuse changes that would not fit
  in the netmap buffer size.
* Ensure the port rxq's are disabled before emptying them during netmap
  start and stop.

LINUX/final-patches/vanilla--mvpp2--40406--99999

@@ -1,7 +1,8 @@
-diff -u a/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c b/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
---- a/mvpp2/mvpp2_main.c	2023-05-24 16:36:55.000000000 +0000
-+++ b/mvpp2/mvpp2_main.c	2023-08-28 01:50:47.764324982 +0000
-@@ -43,6 +43,10 @@
+diff --git a/mvpp2/mvpp2_main.c b/mvpp2/mvpp2_main.c
+index d3f644cad57d..386664b767d7 100644
+--- a/mvpp2/mvpp2_main.c
++++ b/mvpp2/mvpp2_main.c
+@@ -44,6 +44,10 @@
  #include "mvpp2_prs.h"
  #include "mvpp2_cls.h"
 
@@ -12,7 +13,7 @@ diff -u a/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c b/drivers/net/ethernet
  enum mvpp2_bm_pool_log_num {
  	MVPP2_BM_SHORT,
  	MVPP2_BM_LONG,
-@@ -688,7 +692,7 @@
+@@ -688,7 +692,7 @@ static int mvpp2_bm_init(struct device *dev, struct mvpp2 *priv)
  	}
 
  	/* Allocate and initialize BM pools */
@@ -21,19 +22,31 @@ diff -u a/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c b/drivers/net/ethernet
  				      sizeof(*priv->bm_pools), GFP_KERNEL);
  	if (!priv->bm_pools)
  		return -ENOMEM;
-@@ -1435,7 +1439,11 @@ static void mvpp2_interrupts_unmask(void *arg)
-
-	val = MVPP2_CAUSE_MISC_SUM_MASK |
-		MVPP2_CAUSE_RXQ_OCCUP_DESC_ALL_MASK(port->priv->hw_version);
+@@ -1461,7 +1465,11 @@ static void mvpp2_interrupts_unmask(void *arg)
+ 
+ 	val = MVPP2_CAUSE_MISC_SUM_MASK |
+ 		MVPP2_CAUSE_RXQ_OCCUP_DESC_ALL_MASK(port->priv->hw_version);
 +#ifdef CONFIG_NETMAP
 +	if (port->has_tx_irqs && !nm_netmap_on(NA(port->dev)))
 +#else
-	if (port->has_tx_irqs)
+ 	if (port->has_tx_irqs)
++#endif
+ 		val |= MVPP2_CAUSE_TXQ_OCCUP_DESC_ALL_MASK;
+
+ 	mvpp2_thread_write(port->priv, thread,
+@@ -2976,7 +2984,11 @@ static int mvpp2_rxq_init(struct mvpp2_port *port,
+ 	put_cpu();
+
+ 	/* Set Offset */
++#ifdef CONFIG_NETMAP
++	mvpp2_netmap_rxq_offset_set(port, rxq->id);
++#else
+ 	mvpp2_rxq_offset_set(port, rxq->id, MVPP2_SKB_HEADROOM);
 +#endif
-		val |= MVPP2_CAUSE_TXQ_OCCUP_DESC_ALL_MASK;
-
-	mvpp2_thread_write(port->priv, thread,
-@@ -3340,6 +3348,11 @@ static irqreturn_t mvpp2_isr(int irq, void *dev_id)
+ 
+ 	/* Set coalescing pkts and time */
+ 	mvpp2_rx_pkts_coal_set(port, rxq);
+@@ -3366,6 +3378,11 @@ static irqreturn_t mvpp2_isr(int irq, void *dev_id)
  {
  	struct mvpp2_queue_vector *qv = dev_id;
 
@@ -45,7 +58,7 @@ diff -u a/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c b/drivers/net/ethernet
  	mvpp2_qvec_interrupt_disable(qv);
 
  	napi_schedule(&qv->napi);
-@@ -4353,6 +4366,12 @@ static netdev_tx_t mvpp2_tx(struct sk_buff *skb, struct net_device *dev)
+@@ -4382,6 +4399,12 @@ static netdev_tx_t mvpp2_tx(struct sk_buff *skb, struct net_device *dev)
  	}
  	frags = skb_shinfo(skb)->nr_frags + 1;
 
@@ -58,7 +71,19 @@ diff -u a/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c b/drivers/net/ethernet
  	/* Check number of available descriptors */
  	if (mvpp2_aggr_desc_num_check(port, aggr_txq, frags) ||
  	    mvpp2_txq_reserved_desc_num_proc(port, txq, txq_pcpu, frags)) {
-@@ -7004,6 +7023,9 @@ static void mvpp2_port_remove(struct mvpp2_port *port)
+@@ -5831,7 +5854,11 @@ static const struct net_device_ops mvpp2_netdev_ops = {
+ 	.ndo_start_xmit		= mvpp2_tx,
+ 	.ndo_set_rx_mode	= mvpp2_set_rx_mode,
+ 	.ndo_set_mac_address	= mvpp2_set_mac_address,
++#ifdef CONFIG_NETMAP
++	.ndo_change_mtu		= mvpp2_netmap_change_mtu,
++#else
+ 	.ndo_change_mtu		= mvpp2_change_mtu,
++#endif
+ 	.ndo_get_stats64	= mvpp2_get_stats64,
+ 	.ndo_eth_ioctl		= mvpp2_ioctl,
+ 	.ndo_vlan_rx_add_vid	= mvpp2_vlan_rx_add_vid,
+@@ -7138,6 +7165,9 @@ static void mvpp2_port_remove(struct mvpp2_port *port)
  {
  	int i;
 
@@ -68,7 +93,7 @@ diff -u a/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c b/drivers/net/ethernet
  	unregister_netdev(port->dev);
  	if (port->phylink)
  		phylink_destroy(port->phylink);
-@@ -7609,6 +7631,14 @@ static int mvpp2_probe(struct platform_device *pdev)
+@@ -7743,6 +7773,14 @@ static int mvpp2_probe(struct platform_device *pdev)
  	mvpp2_dbgfs_init(priv, pdev->name);
 
  	platform_set_drvdata(pdev, priv);

LINUX/mvpp2_netmap.h

@@ -43,6 +43,8 @@ struct mvpp2_nm_adapter {
 /* Extra netmap buffers required for each of the 4 netmap pools for each MVPP2 instance */
 #define MVPP2_NETMAP_POOL_NUM_BUF		2048
 
+#define MVPP2_NETMAP_MAX_RX_PKT_SIZE(na) NETMAP_BUF_SIZE(na) - NETMAP_SLOT_HEADROOM - MVPP2_SKB_SHINFO_SIZE
+
 /* Used functions from below netmap header include */
 static u32 mvpp2_thread_read(struct mvpp2 *priv, unsigned int thread, u32 offset);
 static u32 mvpp2_thread_read_relaxed(struct mvpp2 *priv, unsigned int thread, u32 offset);
@@ -71,6 +73,8 @@ static void mvpp2_rxq_short_pool_set(struct mvpp2_port *port, int lrxq, int shor
 static void mvpp2_rxq_offset_set(struct mvpp2_port *port, int prxq, int offset);
 static void mvpp2_rxq_drop_pkts(struct mvpp2_port *port, struct mvpp2_rx_queue *rxq);
 static void mvpp2_bm_bufs_get_addrs(struct device *dev, struct mvpp2 *priv, struct mvpp2_bm_pool *bm_pool, dma_addr_t *dma_addr, phys_addr_t *phys_addr);
+static void mvpp2_ingress_enable(struct mvpp2_port *port);
+static void mvpp2_ingress_disable(struct mvpp2_port *port);
 
 
 /* Mask/unmask TX interupts - we will process tx completion before sending new packets */
@@ -323,12 +327,50 @@ static int mvpp2_netmap_rxsync(struct netmap_kring *kring, int flags)
 	return 0;
 }
 
+static int mvpp2_netmap_change_mtu(struct net_device *dev, int mtu)
+{
+	struct netmap_adapter *na = NA(dev);
+
+	/* The native drivers behaviour when setting the MTU shuffles queue
+	 * pools in several circumstances and switches between per_cpu and
+	 * shared pools.  This is problematic for the zero copy support in the
+	 * netmap driver as it forces the native driver into shared pool mode
+	 * and then changes the netmap enabled interface pools.  Thus having
+	 * other code change pools associated with rx queues will break invariants
+	 * expected by the netmap driver.
+	 *
+	 * Thus only allow changing the MTU below the buffer size alloted by netmap
+	 * and don't actually alter any hardware settings based on this.
+	 */
+	if (!IS_ALIGNED(MVPP2_RX_PKT_SIZE(mtu), 8)) {
+		netdev_info(dev, "illegal MTU value %d, round to %d\n", mtu,
+			    ALIGN(MVPP2_RX_PKT_SIZE(mtu), 8));
+		mtu = ALIGN(MVPP2_RX_PKT_SIZE(mtu), 8);
+	}
+
+	if (na && mtu > MVPP2_NETMAP_MAX_RX_PKT_SIZE(na)) {
+		netdev_err(dev, "Illegal MTU value %d (> %d) for netmap\n",
+			   mtu, (int)MVPP2_NETMAP_MAX_RX_PKT_SIZE(na));
+		return -EINVAL;
+	}
+
+	/* Update mtu field in netdev
+	 * Should be fine to leave the buffer size at the netmap size this should
+	 * only mean some packets are received which are larger than the mtu.
+	 * This also does not do any of the flow control config which the native
+	 * driver does, this may need to be revisited.
+	 */
+	dev->mtu = mtu;
+
+	return 0;
+}
+
 static int mvpp2_netmap_users(struct mvpp2 *priv)
 {
 	int count = 0;
 	int i;
 	for (i = 0; i < priv->port_count; i++) {
-		if (priv->port_list[i] && NA(priv->port_list[i]->dev))
+		if (priv->port_list[i] && NA(priv->port_list[i]->dev) && nm_netmap_on(NA(priv->port_list[i]->dev)))
 			count++;
 	}
 	return count;
@@ -404,11 +446,21 @@ static void mvpp2_netmap_destroy_pools(struct netmap_adapter *na)
 	BUG_ON(!priv->percpu_pools);
 }
 
+/* Set the correct rxq offset based on whether netmap is enabled or not */
+static void mvpp2_netmap_rxq_offset_set(struct mvpp2_port *port, int prxq)
+{
+	if (NA(port->dev) && nm_netmap_on(NA(port->dev)))
+		mvpp2_rxq_offset_set(port, prxq, NETMAP_SLOT_HEADROOM);
+	else
+		mvpp2_rxq_offset_set(port, prxq, MVPP2_SKB_HEADROOM);
+}
+
 static void mvpp2_netmap_start(struct netmap_adapter *na)
 {
 	struct mvpp2_port *port = netdev_priv(na->ifp);
 	struct mvpp2 *priv = port->priv;
 	int i;
+	bool running = netif_running(port->dev);
 
 	rtnl_lock();
 
@@ -417,6 +469,10 @@ static void mvpp2_netmap_start(struct netmap_adapter *na)
 		mvpp2_netmap_create_pools (na, MVPP2_NETMAP_POOL_NUM_BUF);
 	}
 
+	/* Disable ingress while emptying rings */
+	if (running)
+		mvpp2_ingress_disable(port);
+
 	/* Use netmap BM pools for this port */
 	for (i = 0; i < port->nrxqs; i++) {
 		mvpp2_rxq_drop_pkts(port, port->rxqs[i]);
@@ -425,6 +481,11 @@ static void mvpp2_netmap_start(struct netmap_adapter *na)
 		mvpp2_rxq_offset_set(port, i, NETMAP_SLOT_HEADROOM);
 	}
 
+	/* Re-enable ingress if interface was running */
+	if (running)
+		mvpp2_ingress_enable(port);
+
+
 	rtnl_unlock();
 }
 
@@ -434,9 +495,14 @@ static void mvpp2_netmap_stop(struct netmap_adapter *na)
 	struct mvpp2_port *port = netdev_priv(na->ifp);
 	struct mvpp2 *priv = port->priv;
 	int i;
+	bool running = netif_running(port->dev);
 
 	rtnl_lock();
 
+	/* Disable ingress while emptying rings */
+	if (running)
+		mvpp2_ingress_disable(port);
+
 	/* Use native BM pools for this port */
 	for (i = 0; i < port->nrxqs; i++) {
 		mvpp2_rxq_drop_pkts(port, port->rxqs[i]);
@@ -445,6 +511,10 @@ static void mvpp2_netmap_stop(struct netmap_adapter *na)
 		mvpp2_rxq_offset_set(port, i, MVPP2_SKB_HEADROOM);
 	}
 
+	/* Re-enable ingress if interface was running */
+	if (running)
+		mvpp2_ingress_enable(port);
+
 	/* Free all netmap buffers and pools if no longer needed */
 	if (mvpp2_netmap_users(priv) <= 1) {
 		mvpp2_netmap_destroy_pools (na);