Merge branch 'eliminate-config_nr_cpus-dependency-in-dpaa-eth-and-enable-compile_test-in-fsl_qbman'

Vladimir Oltean says:

====================
Eliminate CONFIG_NR_CPUS dependency in dpaa-eth and enable COMPILE_TEST in fsl_qbman

Breno's previous attempt at enabling COMPILE_TEST for the fsl_qbman
driver (now included here as patch 5/5) triggered compilation warnings
for large CONFIG_NR_CPUS values:
https://lore.kernel.org/all/202406261920.l5pzM1rj-lkp@intel.com/

Patch 1/5 switches two NR_CPUS arrays in the dpaa-eth driver to dynamic
allocation to avoid that warning. There is more NR_CPUS usage in the
fsl-qbman driver, but that looks relatively harmless and I couldn't find
a good reason to change it.

I noticed, while testing, that the driver doesn't actually work properly
with high CONFIG_NR_CPUS values, and patch 2/5 addresses that.

During code analysis, I have identified two places which treat
conditions that can never happen. Patches 3/5 and 4/5 simplify the
probing code - dpaa_fq_setup() - just a little bit.

Finally we have at 5/5 the patch that triggered all of this. There is
an okay from Herbert to take it via netdev, despite it being on soc/qbman:
https://lore.kernel.org/all/Zns%2FeVVBc7pdv0yM@gondor.apana.org.au/

Link to v1:
https://lore.kernel.org/netdev/20240710230025.46487-1-vladimir.oltean@nxp.com/
====================

Link: https://patch.msgid.link/20240713225336.1746343-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Author: kuba-moo

drivers/net/ethernet/freescale/dpaa/dpaa_eth.c

@@ -371,6 +371,7 @@ static int dpaa_setup_tc(struct net_device *net_dev, enum tc_setup_type type,
 			 void *type_data)
 {
 	struct dpaa_priv *priv = netdev_priv(net_dev);
+	int num_txqs_per_tc = dpaa_num_txqs_per_tc();
 	struct tc_mqprio_qopt *mqprio = type_data;
 	u8 num_tc;
 	int i;
@@ -398,12 +399,12 @@ static int dpaa_setup_tc(struct net_device *net_dev, enum tc_setup_type type,
 	netdev_set_num_tc(net_dev, num_tc);
 
 	for (i = 0; i < num_tc; i++)
-		netdev_set_tc_queue(net_dev, i, DPAA_TC_TXQ_NUM,
-				    i * DPAA_TC_TXQ_NUM);
+		netdev_set_tc_queue(net_dev, i, num_txqs_per_tc,
+				    i * num_txqs_per_tc);
 
 out:
 	priv->num_tc = num_tc ? : 1;
-	netif_set_real_num_tx_queues(net_dev, priv->num_tc * DPAA_TC_TXQ_NUM);
+	netif_set_real_num_tx_queues(net_dev, priv->num_tc * num_txqs_per_tc);
 	return 0;
 }
 
@@ -649,7 +650,7 @@ static inline void dpaa_assign_wq(struct dpaa_fq *fq, int idx)
 		fq->wq = 6;
 		break;
 	case FQ_TYPE_TX:
-		switch (idx / DPAA_TC_TXQ_NUM) {
+		switch (idx / dpaa_num_txqs_per_tc()) {
 		case 0:
 			/* Low priority (best effort) */
 			fq->wq = 6;
@@ -667,8 +668,8 @@ static inline void dpaa_assign_wq(struct dpaa_fq *fq, int idx)
 			fq->wq = 0;
 			break;
 		default:
-			WARN(1, "Too many TX FQs: more than %d!\n",
-			     DPAA_ETH_TXQ_NUM);
+			WARN(1, "Too many TX FQs: more than %zu!\n",
+			     dpaa_max_num_txqs());
 		}
 		break;
 	default:
@@ -740,7 +741,8 @@ static int dpaa_alloc_all_fqs(struct device *dev, struct list_head *list,
 
 	port_fqs->rx_pcdq = &dpaa_fq[0];
 
-	if (!dpaa_fq_alloc(dev, 0, DPAA_ETH_TXQ_NUM, list, FQ_TYPE_TX_CONF_MQ))
+	if (!dpaa_fq_alloc(dev, 0, dpaa_max_num_txqs(), list,
+			   FQ_TYPE_TX_CONF_MQ))
 		goto fq_alloc_failed;
 
 	dpaa_fq = dpaa_fq_alloc(dev, 0, 1, list, FQ_TYPE_TX_ERROR);
@@ -755,7 +757,7 @@ static int dpaa_alloc_all_fqs(struct device *dev, struct list_head *list,
 
 	port_fqs->tx_defq = &dpaa_fq[0];
 
-	if (!dpaa_fq_alloc(dev, 0, DPAA_ETH_TXQ_NUM, list, FQ_TYPE_TX))
+	if (!dpaa_fq_alloc(dev, 0, dpaa_max_num_txqs(), list, FQ_TYPE_TX))
 		goto fq_alloc_failed;
 
 	return 0;
@@ -931,14 +933,18 @@ static inline void dpaa_setup_egress(const struct dpaa_priv *priv,
 	}
 }
 
-static void dpaa_fq_setup(struct dpaa_priv *priv,
-			  const struct dpaa_fq_cbs *fq_cbs,
-			  struct fman_port *tx_port)
+static int dpaa_fq_setup(struct dpaa_priv *priv,
+			 const struct dpaa_fq_cbs *fq_cbs,
+			 struct fman_port *tx_port)
 {
 	int egress_cnt = 0, conf_cnt = 0, num_portals = 0, portal_cnt = 0, cpu;
 	const cpumask_t *affine_cpus = qman_affine_cpus();
-	u16 channels[NR_CPUS];
 	struct dpaa_fq *fq;
+	u16 *channels;
+
+	channels = kcalloc(num_possible_cpus(), sizeof(u16), GFP_KERNEL);
+	if (!channels)
+		return -ENOMEM;
 
 	for_each_cpu_and(cpu, affine_cpus, cpu_online_mask)
 		channels[num_portals++] = qman_affine_channel(cpu);
@@ -965,11 +971,7 @@ static void dpaa_fq_setup(struct dpaa_priv *priv,
 		case FQ_TYPE_TX:
 			dpaa_setup_egress(priv, fq, tx_port,
 					  &fq_cbs->egress_ern);
-			/* If we have more Tx queues than the number of cores,
-			 * just ignore the extra ones.
-			 */
-			if (egress_cnt < DPAA_ETH_TXQ_NUM)
-				priv->egress_fqs[egress_cnt++] = &fq->fq_base;
+			priv->egress_fqs[egress_cnt++] = &fq->fq_base;
 			break;
 		case FQ_TYPE_TX_CONF_MQ:
 			priv->conf_fqs[conf_cnt++] = &fq->fq_base;
@@ -987,24 +989,17 @@ static void dpaa_fq_setup(struct dpaa_priv *priv,
 		}
 	}
 
-	 /* Make sure all CPUs receive a corresponding Tx queue. */
-	while (egress_cnt < DPAA_ETH_TXQ_NUM) {
-		list_for_each_entry(fq, &priv->dpaa_fq_list, list) {
-			if (fq->fq_type != FQ_TYPE_TX)
-				continue;
-			priv->egress_fqs[egress_cnt++] = &fq->fq_base;
-			if (egress_cnt == DPAA_ETH_TXQ_NUM)
-				break;
-		}
-	}
+	kfree(channels);
+
+	return 0;
 }
 
 static inline int dpaa_tx_fq_to_id(const struct dpaa_priv *priv,
 				   struct qman_fq *tx_fq)
 {
 	int i;
 
-	for (i = 0; i < DPAA_ETH_TXQ_NUM; i++)
+	for (i = 0; i < dpaa_max_num_txqs(); i++)
 		if (priv->egress_fqs[i] == tx_fq)
 			return i;
 
@@ -3324,7 +3319,7 @@ static int dpaa_eth_probe(struct platform_device *pdev)
 	/* Allocate this early, so we can store relevant information in
 	 * the private area
 	 */
-	net_dev = alloc_etherdev_mq(sizeof(*priv), DPAA_ETH_TXQ_NUM);
+	net_dev = alloc_etherdev_mq(sizeof(*priv), dpaa_max_num_txqs());
 	if (!net_dev) {
 		dev_err(dev, "alloc_etherdev_mq() failed\n");
 		return -ENOMEM;
@@ -3339,6 +3334,22 @@ static int dpaa_eth_probe(struct platform_device *pdev)
 
 	priv->msg_enable = netif_msg_init(debug, DPAA_MSG_DEFAULT);
 
+	priv->egress_fqs = devm_kcalloc(dev, dpaa_max_num_txqs(),
+					sizeof(*priv->egress_fqs),
+					GFP_KERNEL);
+	if (!priv->egress_fqs) {
+		err = -ENOMEM;
+		goto free_netdev;
+	}
+
+	priv->conf_fqs = devm_kcalloc(dev, dpaa_max_num_txqs(),
+				      sizeof(*priv->conf_fqs),
+				      GFP_KERNEL);
+	if (!priv->conf_fqs) {
+		err = -ENOMEM;
+		goto free_netdev;
+	}
+
 	mac_dev = dpaa_mac_dev_get(pdev);
 	if (IS_ERR(mac_dev)) {
 		netdev_err(net_dev, "dpaa_mac_dev_get() failed\n");
@@ -3416,7 +3427,9 @@ static int dpaa_eth_probe(struct platform_device *pdev)
 	 */
 	dpaa_eth_add_channel(priv->channel, &pdev->dev);
 
-	dpaa_fq_setup(priv, &dpaa_fq_cbs, priv->mac_dev->port[TX]);
+	err = dpaa_fq_setup(priv, &dpaa_fq_cbs, priv->mac_dev->port[TX]);
+	if (err)
+		goto free_dpaa_bps;
 
 	/* Create a congestion group for this netdev, with
 	 * dynamically-allocated CGR ID.
@@ -3462,7 +3475,8 @@ static int dpaa_eth_probe(struct platform_device *pdev)
 	}
 
 	priv->num_tc = 1;
-	netif_set_real_num_tx_queues(net_dev, priv->num_tc * DPAA_TC_TXQ_NUM);
+	netif_set_real_num_tx_queues(net_dev,
+				     priv->num_tc * dpaa_num_txqs_per_tc());
 
 	/* Initialize NAPI */
 	err = dpaa_napi_add(net_dev);

drivers/net/ethernet/freescale/dpaa/dpaa_eth.h

@@ -18,10 +18,6 @@
 
 /* Number of prioritised traffic classes */
 #define DPAA_TC_NUM		4
-/* Number of Tx queues per traffic class */
-#define DPAA_TC_TXQ_NUM		NR_CPUS
-/* Total number of Tx queues */
-#define DPAA_ETH_TXQ_NUM	(DPAA_TC_NUM * DPAA_TC_TXQ_NUM)
 
 /* More detailed FQ types - used for fine-grained WQ assignments */
 enum dpaa_fq_type {
@@ -142,8 +138,8 @@ struct dpaa_priv {
 	struct mac_device *mac_dev;
 	struct device *rx_dma_dev;
 	struct device *tx_dma_dev;
-	struct qman_fq *egress_fqs[DPAA_ETH_TXQ_NUM];
-	struct qman_fq *conf_fqs[DPAA_ETH_TXQ_NUM];
+	struct qman_fq **egress_fqs;
+	struct qman_fq **conf_fqs;
 
 	u16 channel;
 	struct list_head dpaa_fq_list;
@@ -185,4 +181,16 @@ extern const struct ethtool_ops dpaa_ethtool_ops;
 /* from dpaa_eth_sysfs.c */
 void dpaa_eth_sysfs_remove(struct device *dev);
 void dpaa_eth_sysfs_init(struct device *dev);
+
+static inline size_t dpaa_num_txqs_per_tc(void)
+{
+	return num_possible_cpus();
+}
+
+/* Total number of Tx queues */
+static inline size_t dpaa_max_num_txqs(void)
+{
+	return DPAA_TC_NUM * dpaa_num_txqs_per_tc();
+}
+
 #endif	/* __DPAA_H */

drivers/net/ethernet/freescale/dpaa/dpaa_ethtool.c

@@ -457,12 +457,16 @@ static int dpaa_set_coalesce(struct net_device *dev,
 			     struct netlink_ext_ack *extack)
 {
 	const cpumask_t *cpus = qman_affine_cpus();
-	bool needs_revert[NR_CPUS] = {false};
 	struct qman_portal *portal;
 	u32 period, prev_period;
 	u8 thresh, prev_thresh;
+	bool *needs_revert;
 	int cpu, res;
 
+	needs_revert = kcalloc(num_possible_cpus(), sizeof(bool), GFP_KERNEL);
+	if (!needs_revert)
+		return -ENOMEM;
+
 	period = c->rx_coalesce_usecs;
 	thresh = c->rx_max_coalesced_frames;
 
@@ -485,6 +489,8 @@ static int dpaa_set_coalesce(struct net_device *dev,
 		needs_revert[cpu] = true;
 	}
 
+	kfree(needs_revert);
+
 	return 0;
 
 revert_values:
@@ -498,6 +504,8 @@ static int dpaa_set_coalesce(struct net_device *dev,
 		qman_dqrr_set_ithresh(portal, prev_thresh);
 	}
 
+	kfree(needs_revert);
+
 	return res;
 }
 

drivers/soc/fsl/qbman/Kconfig

@@ -1,7 +1,7 @@
 # SPDX-License-Identifier: GPL-2.0-only
 menuconfig FSL_DPAA
 	bool "QorIQ DPAA1 framework support"
-	depends on ((FSL_SOC_BOOKE || ARCH_LAYERSCAPE) && ARCH_DMA_ADDR_T_64BIT)
+	depends on ((FSL_SOC_BOOKE || ARCH_LAYERSCAPE || COMPILE_TEST) && ARCH_DMA_ADDR_T_64BIT)
 	select GENERIC_ALLOCATOR
 	help
 	  The Freescale Data Path Acceleration Architecture (DPAA) is a set of