Merge branch 'thunderx-irq-hints'

Sunil Goutham says:

====================
net: thunderx: Setting IRQ affinity hints and other optimizations

This patch series contains changes
- To add support for virtual function's irq affinity hint
- Replace napi_schedule() with napi_schedule_irqoff()
- Reduce page allocation overhead by allocating pages
  of higher order when pagesize is 4KB.
- Add couple of stats which helps in debugging
- Some miscellaneous changes to BGX driver.

Changes from v1:
- As suggested changed MAC address invalid log message
  to dev_err() instead of dev_warn().
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/cavium/thunder/nic.h

@@ -248,10 +248,13 @@ struct nicvf_drv_stats {
 	u64 rx_frames_jumbo;
 	u64 rx_drops;
 
+	u64 rcv_buffer_alloc_failures;
+
 	/* Tx */
 	u64 tx_frames_ok;
 	u64 tx_drops;
 	u64 tx_tso;
+	u64 tx_timeout;
 	u64 txq_stop;
 	u64 txq_wake;
 };
@@ -306,6 +309,7 @@ struct nicvf {
 	struct msix_entry	msix_entries[NIC_VF_MSIX_VECTORS];
 	char			irq_name[NIC_VF_MSIX_VECTORS][20];
 	bool			irq_allocated[NIC_VF_MSIX_VECTORS];
+	cpumask_var_t		affinity_mask[NIC_VF_MSIX_VECTORS];
 
 	/* VF <-> PF mailbox communication */
 	bool			pf_acked;

drivers/net/ethernet/cavium/thunder/nicvf_ethtool.c

@@ -89,9 +89,11 @@ static const struct nicvf_stat nicvf_drv_stats[] = {
 	NICVF_DRV_STAT(rx_frames_1518),
 	NICVF_DRV_STAT(rx_frames_jumbo),
 	NICVF_DRV_STAT(rx_drops),
+	NICVF_DRV_STAT(rcv_buffer_alloc_failures),
 	NICVF_DRV_STAT(tx_frames_ok),
 	NICVF_DRV_STAT(tx_tso),
 	NICVF_DRV_STAT(tx_drops),
+	NICVF_DRV_STAT(tx_timeout),
 	NICVF_DRV_STAT(txq_stop),
 	NICVF_DRV_STAT(txq_wake),
 };

drivers/net/ethernet/cavium/thunder/nicvf_main.c

@@ -828,7 +828,7 @@ static irqreturn_t nicvf_intr_handler(int irq, void *cq_irq)
 	nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx);
 
 	/* Schedule NAPI */
-	napi_schedule(&cq_poll->napi);
+	napi_schedule_irqoff(&cq_poll->napi);
 
 	/* Clear interrupt */
 	nicvf_clear_intr(nic, NICVF_INTR_CQ, qidx);
@@ -899,6 +899,31 @@ static void nicvf_disable_msix(struct nicvf *nic)
 	}
 }
 
+static void nicvf_set_irq_affinity(struct nicvf *nic)
+{
+	int vec, cpu;
+	int irqnum;
+
+	for (vec = 0; vec < nic->num_vec; vec++) {
+		if (!nic->irq_allocated[vec])
+			continue;
+
+		if (!zalloc_cpumask_var(&nic->affinity_mask[vec], GFP_KERNEL))
+			return;
+		 /* CQ interrupts */
+		if (vec < NICVF_INTR_ID_SQ)
+			/* Leave CPU0 for RBDR and other interrupts */
+			cpu = nicvf_netdev_qidx(nic, vec) + 1;
+		else
+			cpu = 0;
+
+		cpumask_set_cpu(cpumask_local_spread(cpu, nic->node),
+				nic->affinity_mask[vec]);
+		irqnum = nic->msix_entries[vec].vector;
+		irq_set_affinity_hint(irqnum, nic->affinity_mask[vec]);
+	}
+}
+
 static int nicvf_register_interrupts(struct nicvf *nic)
 {
 	int irq, ret = 0;
@@ -944,8 +969,13 @@ static int nicvf_register_interrupts(struct nicvf *nic)
 	ret = request_irq(nic->msix_entries[irq].vector,
 			  nicvf_qs_err_intr_handler,
 			  0, nic->irq_name[irq], nic);
-	if (!ret)
-		nic->irq_allocated[irq] = true;
+	if (ret)
+		goto err;
+
+	nic->irq_allocated[irq] = true;
+
+	/* Set IRQ affinities */
+	nicvf_set_irq_affinity(nic);
 
 err:
 	if (ret)
@@ -963,6 +993,9 @@ static void nicvf_unregister_interrupts(struct nicvf *nic)
 		if (!nic->irq_allocated[irq])
 			continue;
 
+		irq_set_affinity_hint(nic->msix_entries[irq].vector, NULL);
+		free_cpumask_var(nic->affinity_mask[irq]);
+
 		if (irq < NICVF_INTR_ID_SQ)
 			free_irq(nic->msix_entries[irq].vector, nic->napi[irq]);
 		else
@@ -1394,6 +1427,7 @@ static void nicvf_tx_timeout(struct net_device *dev)
 		netdev_warn(dev, "%s: Transmit timed out, resetting\n",
 			    dev->name);
 
+	nic->drv_stats.tx_timeout++;
 	schedule_work(&nic->reset_task);
 }
 

drivers/net/ethernet/cavium/thunder/nicvf_queues.c

@@ -78,7 +78,7 @@ static void nicvf_free_q_desc_mem(struct nicvf *nic, struct q_desc_mem *dmem)
 static inline int nicvf_alloc_rcv_buffer(struct nicvf *nic, gfp_t gfp,
 					 u32 buf_len, u64 **rbuf)
 {
-	int order = get_order(buf_len);
+	int order = (PAGE_SIZE <= 4096) ?  PAGE_ALLOC_COSTLY_ORDER : 0;
 
 	/* Check if request can be accomodated in previous allocated page */
 	if (nic->rb_page) {
@@ -96,8 +96,7 @@ static inline int nicvf_alloc_rcv_buffer(struct nicvf *nic, gfp_t gfp,
 		nic->rb_page = alloc_pages(gfp | __GFP_COMP | __GFP_NOWARN,
 					   order);
 		if (!nic->rb_page) {
-			netdev_err(nic->netdev,
-				   "Failed to allocate new rcv buffer\n");
+			nic->drv_stats.rcv_buffer_alloc_failures++;
 			return -ENOMEM;
 		}
 		nic->rb_page_offset = 0;

drivers/net/ethernet/cavium/thunder/thunder_bgx.c

@@ -886,7 +886,8 @@ static void bgx_get_qlm_mode(struct bgx *bgx)
 
 #ifdef CONFIG_ACPI
 
-static int acpi_get_mac_address(struct acpi_device *adev, u8 *dst)
+static int acpi_get_mac_address(struct device *dev, struct acpi_device *adev,
+				u8 *dst)
 {
 	u8 mac[ETH_ALEN];
 	int ret;
@@ -897,10 +898,13 @@ static int acpi_get_mac_address(struct acpi_device *adev, u8 *dst)
 		goto out;
 
 	if (!is_valid_ether_addr(mac)) {
+		dev_err(dev, "MAC address invalid: %pM\n", mac);
 		ret = -EINVAL;
 		goto out;
 	}
 
+	dev_info(dev, "MAC address set to: %pM\n", mac);
+
 	memcpy(dst, mac, ETH_ALEN);
 out:
 	return ret;
@@ -911,14 +915,15 @@ static acpi_status bgx_acpi_register_phy(acpi_handle handle,
 					 u32 lvl, void *context, void **rv)
 {
 	struct bgx *bgx = context;
+	struct device *dev = &bgx->pdev->dev;
 	struct acpi_device *adev;
 
 	if (acpi_bus_get_device(handle, &adev))
 		goto out;
 
-	acpi_get_mac_address(adev, bgx->lmac[bgx->lmac_count].mac);
+	acpi_get_mac_address(dev, adev, bgx->lmac[bgx->lmac_count].mac);
 
-	SET_NETDEV_DEV(&bgx->lmac[bgx->lmac_count].netdev, &bgx->pdev->dev);
+	SET_NETDEV_DEV(&bgx->lmac[bgx->lmac_count].netdev, dev);
 
 	bgx->lmac[bgx->lmac_count].lmacid = bgx->lmac_count;
 out:
@@ -968,26 +973,27 @@ static int bgx_init_acpi_phy(struct bgx *bgx)
 
 static int bgx_init_of_phy(struct bgx *bgx)
 {
-	struct device_node *np;
-	struct device_node *np_child;
+	struct fwnode_handle *fwn;
 	u8 lmac = 0;
-	char bgx_sel[5];
 	const char *mac;
 
-	/* Get BGX node from DT */
-	snprintf(bgx_sel, 5, "bgx%d", bgx->bgx_id);
-	np = of_find_node_by_name(NULL, bgx_sel);
-	if (!np)
-		return -ENODEV;
+	device_for_each_child_node(&bgx->pdev->dev, fwn) {
+		struct device_node *phy_np;
+		struct device_node *node = to_of_node(fwn);
 
-	for_each_child_of_node(np, np_child) {
-		struct device_node *phy_np = of_parse_phandle(np_child,
-							      "phy-handle", 0);
+		/* If it is not an OF node we cannot handle it yet, so
+		 * exit the loop.
+		 */
+		if (!node)
+			break;
+
+		phy_np = of_parse_phandle(node, "phy-handle", 0);
 		if (!phy_np)
 			continue;
+
 		bgx->lmac[lmac].phydev = of_phy_find_device(phy_np);
 
-		mac = of_get_mac_address(np_child);
+		mac = of_get_mac_address(node);
 		if (mac)
 			ether_addr_copy(bgx->lmac[lmac].mac, mac);
 
@@ -995,7 +1001,7 @@ static int bgx_init_of_phy(struct bgx *bgx)
 		bgx->lmac[lmac].lmacid = lmac;
 		lmac++;
 		if (lmac == MAX_LMAC_PER_BGX) {
-			of_node_put(np_child);
+			of_node_put(node);
 			break;
 		}
 	}