dpaa_eth: add support for DPAA Ethernet

This introduces the Freescale Data Path Acceleration Architecture
(DPAA) Ethernet driver (dpaa_eth) that builds upon the DPAA QMan,
BMan, PAMU and FMan drivers to deliver Ethernet connectivity on
the Freescale DPAA QorIQ platforms.

Signed-off-by: Madalin Bucur <madalin.bucur@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

6 files changed, 2850 insertions, 0 deletions

diff --git a/drivers/net/ethernet/freescale/Kconfig b/drivers/net/ethernet/freescale/Kconfig
index d1ca45fbb164..aa3f615886b4 100644
--- a/drivers/net/ethernet/freescale/Kconfig
+++ b/drivers/net/ethernet/freescale/Kconfig
@@ -93,4 +93,6 @@ config GIANFAR
 	  and MPC86xx family of chips, the eTSEC on LS1021A and the FEC
 	  on the 8540.
 
+source "drivers/net/ethernet/freescale/dpaa/Kconfig"
+
 endif # NET_VENDOR_FREESCALE
diff --git a/drivers/net/ethernet/freescale/Makefile b/drivers/net/ethernet/freescale/Makefile
index cbe21dc7e37e..4a13115155c9 100644
--- a/drivers/net/ethernet/freescale/Makefile
+++ b/drivers/net/ethernet/freescale/Makefile
@@ -22,3 +22,4 @@ obj-$(CONFIG_UCC_GETH) += ucc_geth_driver.o
 ucc_geth_driver-objs := ucc_geth.o ucc_geth_ethtool.o
 
 obj-$(CONFIG_FSL_FMAN) += fman/
+obj-$(CONFIG_FSL_DPAA_ETH) += dpaa/
diff --git a/drivers/net/ethernet/freescale/dpaa/Kconfig b/drivers/net/ethernet/freescale/dpaa/Kconfig
new file mode 100644
index 000000000000..f3a3454805f9
--- /dev/null
+++ b/drivers/net/ethernet/freescale/dpaa/Kconfig
@@ -0,0 +1,10 @@
+menuconfig FSL_DPAA_ETH
+	tristate "DPAA Ethernet"
+	depends on FSL_SOC && FSL_DPAA && FSL_FMAN
+	select PHYLIB
+	select FSL_FMAN_MAC
+	---help---
+	  Data Path Acceleration Architecture Ethernet driver,
+	  supporting the Freescale QorIQ chips.
+	  Depends on Freescale Buffer Manager and Queue Manager
+	  driver and Frame Manager Driver.
diff --git a/drivers/net/ethernet/freescale/dpaa/Makefile b/drivers/net/ethernet/freescale/dpaa/Makefile
new file mode 100644
index 000000000000..fc76029281c7
--- /dev/null
+++ b/drivers/net/ethernet/freescale/dpaa/Makefile
@@ -0,0 +1,11 @@
+#
+# Makefile for the Freescale DPAA Ethernet controllers
+#
+
+# Include FMan headers
+FMAN        = $(srctree)/drivers/net/ethernet/freescale/fman
+ccflags-y += -I$(FMAN)
+
+obj-$(CONFIG_FSL_DPAA_ETH) += fsl_dpa.o
+
+fsl_dpa-objs += dpaa_eth.o
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
new file mode 100644
index 000000000000..8b9b0720f8b1
--- /dev/null
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
@@ -0,0 +1,2682 @@
+/* Copyright 2008 - 2016 Freescale Semiconductor Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *	 notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *	 notice, this list of conditions and the following disclaimer in the
+ *	 documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *	 names of its contributors may be used to endorse or promote products
+ *	 derived from this software without specific prior written permission.
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/of_platform.h>
+#include <linux/of_mdio.h>
+#include <linux/of_net.h>
+#include <linux/io.h>
+#include <linux/if_arp.h>
+#include <linux/if_vlan.h>
+#include <linux/icmp.h>
+#include <linux/ip.h>
+#include <linux/ipv6.h>
+#include <linux/udp.h>
+#include <linux/tcp.h>
+#include <linux/net.h>
+#include <linux/skbuff.h>
+#include <linux/etherdevice.h>
+#include <linux/if_ether.h>
+#include <linux/highmem.h>
+#include <linux/percpu.h>
+#include <linux/dma-mapping.h>
+#include <linux/sort.h>
+#include <soc/fsl/bman.h>
+#include <soc/fsl/qman.h>
+
+#include "fman.h"
+#include "fman_port.h"
+#include "mac.h"
+#include "dpaa_eth.h"
+
+static int debug = -1;
+module_param(debug, int, 0444);
+MODULE_PARM_DESC(debug, "Module/Driver verbosity level (0=none,...,16=all)");
+
+static u16 tx_timeout = 1000;
+module_param(tx_timeout, ushort, 0444);
+MODULE_PARM_DESC(tx_timeout, "The Tx timeout in ms");
+
+#define FM_FD_STAT_RX_ERRORS						\
+	(FM_FD_ERR_DMA | FM_FD_ERR_PHYSICAL	| \
+	 FM_FD_ERR_SIZE | FM_FD_ERR_CLS_DISCARD | \
+	 FM_FD_ERR_EXTRACTION | FM_FD_ERR_NO_SCHEME	| \
+	 FM_FD_ERR_PRS_TIMEOUT | FM_FD_ERR_PRS_ILL_INSTRUCT | \
+	 FM_FD_ERR_PRS_HDR_ERR)
+
+#define FM_FD_STAT_TX_ERRORS \
+	(FM_FD_ERR_UNSUPPORTED_FORMAT | \
+	 FM_FD_ERR_LENGTH | FM_FD_ERR_DMA)
+
+#define DPAA_MSG_DEFAULT (NETIF_MSG_DRV | NETIF_MSG_PROBE | \
+			  NETIF_MSG_LINK | NETIF_MSG_IFUP | \
+			  NETIF_MSG_IFDOWN)
+
+#define DPAA_INGRESS_CS_THRESHOLD 0x10000000
+/* Ingress congestion threshold on FMan ports
+ * The size in bytes of the ingress tail-drop threshold on FMan ports.
+ * Traffic piling up above this value will be rejected by QMan and discarded
+ * by FMan.
+ */
+
+/* Size in bytes of the FQ taildrop threshold */
+#define DPAA_FQ_TD 0x200000
+
+#define DPAA_CS_THRESHOLD_1G 0x06000000
+/* Egress congestion threshold on 1G ports, range 0x1000 .. 0x10000000
+ * The size in bytes of the egress Congestion State notification threshold on
+ * 1G ports. The 1G dTSECs can quite easily be flooded by cores doing Tx in a
+ * tight loop (e.g. by sending UDP datagrams at "while(1) speed"),
+ * and the larger the frame size, the more acute the problem.
+ * So we have to find a balance between these factors:
+ * - avoiding the device staying congested for a prolonged time (risking
+ *   the netdev watchdog to fire - see also the tx_timeout module param);
+ * - affecting performance of protocols such as TCP, which otherwise
+ *   behave well under the congestion notification mechanism;
+ * - preventing the Tx cores from tightly-looping (as if the congestion
+ *   threshold was too low to be effective);
+ * - running out of memory if the CS threshold is set too high.
+ */
+
+#define DPAA_CS_THRESHOLD_10G 0x10000000
+/* The size in bytes of the egress Congestion State notification threshold on
+ * 10G ports, range 0x1000 .. 0x10000000
+ */
+
+/* Largest value that the FQD's OAL field can hold */
+#define FSL_QMAN_MAX_OAL	127
+
+/* Default alignment for start of data in an Rx FD */
+#define DPAA_FD_DATA_ALIGNMENT  16
+
+/* Values for the L3R field of the FM Parse Results
+ */
+/* L3 Type field: First IP Present IPv4 */
+#define FM_L3_PARSE_RESULT_IPV4	0x8000
+/* L3 Type field: First IP Present IPv6 */
+#define FM_L3_PARSE_RESULT_IPV6	0x4000
+/* Values for the L4R field of the FM Parse Results */
+/* L4 Type field: UDP */
+#define FM_L4_PARSE_RESULT_UDP	0x40
+/* L4 Type field: TCP */
+#define FM_L4_PARSE_RESULT_TCP	0x20
+
+#define DPAA_SGT_MAX_ENTRIES 16 /* maximum number of entries in SG Table */
+#define DPAA_BUFF_RELEASE_MAX 8 /* maximum number of buffers released at once */
+
+#define FSL_DPAA_BPID_INV		0xff
+#define FSL_DPAA_ETH_MAX_BUF_COUNT	128
+#define FSL_DPAA_ETH_REFILL_THRESHOLD	80
+
+#define DPAA_TX_PRIV_DATA_SIZE	16
+#define DPAA_PARSE_RESULTS_SIZE sizeof(struct fman_prs_result)
+#define DPAA_TIME_STAMP_SIZE 8
+#define DPAA_HASH_RESULTS_SIZE 8
+#define DPAA_RX_PRIV_DATA_SIZE	(u16)(DPAA_TX_PRIV_DATA_SIZE + \
+					dpaa_rx_extra_headroom)
+
+#define DPAA_ETH_RX_QUEUES	128
+
+#define DPAA_ENQUEUE_RETRIES	100000
+
+enum port_type {RX, TX};
+
+struct fm_port_fqs {
+	struct dpaa_fq *tx_defq;
+	struct dpaa_fq *tx_errq;
+	struct dpaa_fq *rx_defq;
+	struct dpaa_fq *rx_errq;
+};
+
+/* All the dpa bps in use at any moment */
+static struct dpaa_bp *dpaa_bp_array[BM_MAX_NUM_OF_POOLS];
+
+/* The raw buffer size must be cacheline aligned */
+#define DPAA_BP_RAW_SIZE 4096
+/* When using more than one buffer pool, the raw sizes are as follows:
+ * 1 bp: 4KB
+ * 2 bp: 2KB, 4KB
+ * 3 bp: 1KB, 2KB, 4KB
+ * 4 bp: 1KB, 2KB, 4KB, 8KB
+ */
+static inline size_t bpool_buffer_raw_size(u8 index, u8 cnt)
+{
+	size_t res = DPAA_BP_RAW_SIZE / 4;
+	u8 i;
+
+	for (i = (cnt < 3) ? cnt : 3; i < 3 + index; i++)
+		res *= 2;
+	return res;
+}
+
+/* FMan-DMA requires 16-byte alignment for Rx buffers, but SKB_DATA_ALIGN is
+ * even stronger (SMP_CACHE_BYTES-aligned), so we just get away with that,
+ * via SKB_WITH_OVERHEAD(). We can't rely on netdev_alloc_frag() giving us
+ * half-page-aligned buffers, so we reserve some more space for start-of-buffer
+ * alignment.
+ */
+#define dpaa_bp_size(raw_size) SKB_WITH_OVERHEAD((raw_size) - SMP_CACHE_BYTES)
+
+static int dpaa_max_frm;
+
+static int dpaa_rx_extra_headroom;
+
+#define dpaa_get_max_mtu()	\
+	(dpaa_max_frm - (VLAN_ETH_HLEN + ETH_FCS_LEN))
+
+static int dpaa_netdev_init(struct net_device *net_dev,
+			    const struct net_device_ops *dpaa_ops,
+			    u16 tx_timeout)
+{
+	struct dpaa_priv *priv = netdev_priv(net_dev);
+	struct device *dev = net_dev->dev.parent;
+	struct dpaa_percpu_priv *percpu_priv;
+	const u8 *mac_addr;
+	int i, err;
+
+	/* Although we access another CPU's private data here
+	 * we do it at initialization so it is safe
+	 */
+	for_each_possible_cpu(i) {
+		percpu_priv = per_cpu_ptr(priv->percpu_priv, i);
+		percpu_priv->net_dev = net_dev;
+	}
+
+	net_dev->netdev_ops = dpaa_ops;
+	mac_addr = priv->mac_dev->addr;
+
+	net_dev->mem_start = priv->mac_dev->res->start;
+	net_dev->mem_end = priv->mac_dev->res->end;
+
+	net_dev->min_mtu = ETH_MIN_MTU;
+	net_dev->max_mtu = dpaa_get_max_mtu();
+
+	net_dev->hw_features |= (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
+				 NETIF_F_LLTX);
+
+	net_dev->hw_features |= NETIF_F_SG | NETIF_F_HIGHDMA;
+	/* The kernels enables GSO automatically, if we declare NETIF_F_SG.
+	 * For conformity, we'll still declare GSO explicitly.
+	 */
+	net_dev->features |= NETIF_F_GSO;
+
+	net_dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
+	/* we do not want shared skbs on TX */
+	net_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
+
+	net_dev->features |= net_dev->hw_features;
+	net_dev->vlan_features = net_dev->features;
+
+	memcpy(net_dev->perm_addr, mac_addr, net_dev->addr_len);
+	memcpy(net_dev->dev_addr, mac_addr, net_dev->addr_len);
+
+	net_dev->needed_headroom = priv->tx_headroom;
+	net_dev->watchdog_timeo = msecs_to_jiffies(tx_timeout);
+
+	/* start without the RUNNING flag, phylib controls it later */
+	netif_carrier_off(net_dev);
+
+	err = register_netdev(net_dev);
+	if (err < 0) {
+		dev_err(dev, "register_netdev() = %d\n", err);
+		return err;
+	}
+
+	return 0;
+}
+
+static int dpaa_stop(struct net_device *net_dev)
+{
+	struct mac_device *mac_dev;
+	struct dpaa_priv *priv;
+	int i, err, error;
+
+	priv = netdev_priv(net_dev);
+	mac_dev = priv->mac_dev;
+
+	netif_tx_stop_all_queues(net_dev);
+	/* Allow the Fman (Tx) port to process in-flight frames before we
+	 * try switching it off.
+	 */
+	usleep_range(5000, 10000);
+
+	err = mac_dev->stop(mac_dev);
+	if (err < 0)
+		netif_err(priv, ifdown, net_dev, "mac_dev->stop() = %d\n",
+			  err);
+
+	for (i = 0; i < ARRAY_SIZE(mac_dev->port); i++) {
+		error = fman_port_disable(mac_dev->port[i]);
+		if (error)
+			err = error;
+	}
+
+	if (net_dev->phydev)
+		phy_disconnect(net_dev->phydev);
+	net_dev->phydev = NULL;
+
+	return err;
+}
+
+static void dpaa_tx_timeout(struct net_device *net_dev)
+{
+	struct dpaa_percpu_priv *percpu_priv;
+	const struct dpaa_priv	*priv;
+
+	priv = netdev_priv(net_dev);
+	percpu_priv = this_cpu_ptr(priv->percpu_priv);
+
+	netif_crit(priv, timer, net_dev, "Transmit timeout latency: %u ms\n",
+		   jiffies_to_msecs(jiffies - dev_trans_start(net_dev)));
+
+	percpu_priv->stats.tx_errors++;
+}
+
+/* Calculates the statistics for the given device by adding the statistics
+ * collected by each CPU.
+ */
+static struct rtnl_link_stats64 *dpaa_get_stats64(struct net_device *net_dev,
+						  struct rtnl_link_stats64 *s)
+{
+	int numstats = sizeof(struct rtnl_link_stats64) / sizeof(u64);
+	struct dpaa_priv *priv = netdev_priv(net_dev);
+	struct dpaa_percpu_priv *percpu_priv;
+	u64 *netstats = (u64 *)s;
+	u64 *cpustats;
+	int i, j;
+
+	for_each_possible_cpu(i) {
+		percpu_priv = per_cpu_ptr(priv->percpu_priv, i);
+
+		cpustats = (u64 *)&percpu_priv->stats;
+
+		/* add stats from all CPUs */
+		for (j = 0; j < numstats; j++)
+			netstats[j] += cpustats[j];
+	}
+
+	return s;
+}
+
+static struct mac_device *dpaa_mac_dev_get(struct platform_device *pdev)
+{
+	struct platform_device *of_dev;
+	struct dpaa_eth_data *eth_data;
+	struct device *dpaa_dev, *dev;
+	struct device_node *mac_node;
+	struct mac_device *mac_dev;
+
+	dpaa_dev = &pdev->dev;
+	eth_data = dpaa_dev->platform_data;
+	if (!eth_data)
+		return ERR_PTR(-ENODEV);
+
+	mac_node = eth_data->mac_node;
+
+	of_dev = of_find_device_by_node(mac_node);
+	if (!of_dev) {
+		dev_err(dpaa_dev, "of_find_device_by_node(%s) failed\n",
+			mac_node->full_name);
+		of_node_put(mac_node);
+		return ERR_PTR(-EINVAL);
+	}
+	of_node_put(mac_node);
+
+	dev = &of_dev->dev;
+
+	mac_dev = dev_get_drvdata(dev);
+	if (!mac_dev) {
+		dev_err(dpaa_dev, "dev_get_drvdata(%s) failed\n",
+			dev_name(dev));
+		return ERR_PTR(-EINVAL);
+	}
+
+	return mac_dev;
+}
+
+static int dpaa_set_mac_address(struct net_device *net_dev, void *addr)
+{
+	const struct dpaa_priv *priv;
+	struct mac_device *mac_dev;
+	struct sockaddr old_addr;
+	int err;
+
+	priv = netdev_priv(net_dev);
+
+	memcpy(old_addr.sa_data, net_dev->dev_addr,  ETH_ALEN);
+
+	err = eth_mac_addr(net_dev, addr);
+	if (err < 0) {
+		netif_err(priv, drv, net_dev, "eth_mac_addr() = %d\n", err);
+		return err;
+	}
+
+	mac_dev = priv->mac_dev;
+
+	err = mac_dev->change_addr(mac_dev->fman_mac,
+				   (enet_addr_t *)net_dev->dev_addr);
+	if (err < 0) {
+		netif_err(priv, drv, net_dev, "mac_dev->change_addr() = %d\n",
+			  err);
+		/* reverting to previous address */
+		eth_mac_addr(net_dev, &old_addr);
+
+		return err;
+	}
+
+	return 0;
+}
+
+static void dpaa_set_rx_mode(struct net_device *net_dev)
+{
+	const struct dpaa_priv	*priv;
+	int err;
+
+	priv = netdev_priv(net_dev);
+
+	if (!!(net_dev->flags & IFF_PROMISC) != priv->mac_dev->promisc) {
+		priv->mac_dev->promisc = !priv->mac_dev->promisc;
+		err = priv->mac_dev->set_promisc(priv->mac_dev->fman_mac,
+						 priv->mac_dev->promisc);
+		if (err < 0)
+			netif_err(priv, drv, net_dev,
+				  "mac_dev->set_promisc() = %d\n",
+				  err);
+	}
+
+	err = priv->mac_dev->set_multi(net_dev, priv->mac_dev);
+	if (err < 0)
+		netif_err(priv, drv, net_dev, "mac_dev->set_multi() = %d\n",
+			  err);
+}
+
+static struct dpaa_bp *dpaa_bpid2pool(int bpid)
+{
+	if (WARN_ON(bpid < 0 || bpid >= BM_MAX_NUM_OF_POOLS))
+		return NULL;
+
+	return dpaa_bp_array[bpid];
+}
+
+/* checks if this bpool is already allocated */
+static bool dpaa_bpid2pool_use(int bpid)
+{
+	if (dpaa_bpid2pool(bpid)) {
+		atomic_inc(&dpaa_bp_array[bpid]->refs);
+		return true;
+	}
+
+	return false;
+}
+
+/* called only once per bpid by dpaa_bp_alloc_pool() */
+static void dpaa_bpid2pool_map(int bpid, struct dpaa_bp *dpaa_bp)
+{
+	dpaa_bp_array[bpid] = dpaa_bp;
+	atomic_set(&dpaa_bp->refs, 1);
+}
+
+static int dpaa_bp_alloc_pool(struct dpaa_bp *dpaa_bp)
+{
+	int err;
+
+	if (dpaa_bp->size == 0 || dpaa_bp->config_count == 0) {
+		pr_err("%s: Buffer pool is not properly initialized! Missing size or initial number of buffers\n",
+		       __func__);
+		return -EINVAL;
+	}
+
+	/* If the pool is already specified, we only create one per bpid */
+	if (dpaa_bp->bpid != FSL_DPAA_BPID_INV &&
+	    dpaa_bpid2pool_use(dpaa_bp->bpid))
+		return 0;
+
+	if (dpaa_bp->bpid == FSL_DPAA_BPID_INV) {
+		dpaa_bp->pool = bman_new_pool();
+		if (!dpaa_bp->pool) {
+			pr_err("%s: bman_new_pool() failed\n",
+			       __func__);
+			return -ENODEV;
+		}
+
+		dpaa_bp->bpid = (u8)bman_get_bpid(dpaa_bp->pool);
+	}
+
+	if (dpaa_bp->seed_cb) {
+		err = dpaa_bp->seed_cb(dpaa_bp);
+		if (err)
+			goto pool_seed_failed;
+	}
+
+	dpaa_bpid2pool_map(dpaa_bp->bpid, dpaa_bp);
+
+	return 0;
+
+pool_seed_failed:
+	pr_err("%s: pool seeding failed\n", __func__);
+	bman_free_pool(dpaa_bp->pool);
+
+	return err;
+}
+
+/* remove and free all the buffers from the given buffer pool */
+static void dpaa_bp_drain(struct dpaa_bp *bp)
+{
+	u8 num = 8;
+	int ret;
+
+	do {
+		struct bm_buffer bmb[8];
+		int i;
+
+		ret = bman_acquire(bp->pool, bmb, num);
+		if (ret < 0) {
+			if (num == 8) {
+				/* we have less than 8 buffers left;
+				 * drain them one by one
+				 */
+				num = 1;
+				ret = 1;
+				continue;
+			} else {
+				/* Pool is fully drained */
+				break;
+			}
+		}
+
+		if (bp->free_buf_cb)
+			for (i = 0; i < num; i++)
+				bp->free_buf_cb(bp, &bmb[i]);
+	} while (ret > 0);
+}
+
+static void dpaa_bp_free(struct dpaa_bp *dpaa_bp)
+{
+	struct dpaa_bp *bp = dpaa_bpid2pool(dpaa_bp->bpid);
+
+	/* the mapping between bpid and dpaa_bp is done very late in the
+	 * allocation procedure; if something failed before the mapping, the bp
+	 * was not configured, therefore we don't need the below instructions
+	 */
+	if (!bp)
+		return;
+
+	if (!atomic_dec_and_test(&bp->refs))
+		return;
+
+	if (bp->free_buf_cb)
+		dpaa_bp_drain(bp);
+
+	dpaa_bp_array[bp->bpid] = NULL;
+	bman_free_pool(bp->pool);
+}
+
+static void dpaa_bps_free(struct dpaa_priv *priv)
+{
+	int i;
+
+	for (i = 0; i < DPAA_BPS_NUM; i++)
+		dpaa_bp_free(priv->dpaa_bps[i]);
+}
+
+/* Use multiple WQs for FQ assignment:
+ *	- Tx Confirmation queues go to WQ1.
+ *	- Rx Error and Tx Error queues go to WQ2 (giving them a better chance
+ *	  to be scheduled, in case there are many more FQs in WQ3).
+ *	- Rx Default and Tx queues go to WQ3 (no differentiation between
+ *	  Rx and Tx traffic).
+ * This ensures that Tx-confirmed buffers are timely released. In particular,
+ * it avoids congestion on the Tx Confirm FQs, which can pile up PFDRs if they
+ * are greatly outnumbered by other FQs in the system, while
+ * dequeue scheduling is round-robin.
+ */
+static inline void dpaa_assign_wq(struct dpaa_fq *fq)
+{
+	switch (fq->fq_type) {
+	case FQ_TYPE_TX_CONFIRM:
+	case FQ_TYPE_TX_CONF_MQ:
+		fq->wq = 1;
+		break;
+	case FQ_TYPE_RX_ERROR:
+	case FQ_TYPE_TX_ERROR:
+		fq->wq = 2;
+		break;
+	case FQ_TYPE_RX_DEFAULT:
+	case FQ_TYPE_TX:
+		fq->wq = 3;
+		break;
+	default:
+		WARN(1, "Invalid FQ type %d for FQID %d!\n",
+		     fq->fq_type, fq->fqid);
+	}
+}
+
+static struct dpaa_fq *dpaa_fq_alloc(struct device *dev,
+				     u32 start, u32 count,
+				     struct list_head *list,
+				     enum dpaa_fq_type fq_type)
+{
+	struct dpaa_fq *dpaa_fq;
+	int i;
+
+	dpaa_fq = devm_kzalloc(dev, sizeof(*dpaa_fq) * count,
+			       GFP_KERNEL);
+	if (!dpaa_fq)
+		return NULL;
+
+	for (i = 0; i < count; i++) {
+		dpaa_fq[i].fq_type = fq_type;
+		dpaa_fq[i].fqid = start ? start + i : 0;
+		list_add_tail(&dpaa_fq[i].list, list);
+	}
+
+	for (i = 0; i < count; i++)
+		dpaa_assign_wq(dpaa_fq + i);
+
+	return dpaa_fq;
+}
+
+static int dpaa_alloc_all_fqs(struct device *dev, struct list_head *list,
+			      struct fm_port_fqs *port_fqs)
+{
+	struct dpaa_fq *dpaa_fq;
+
+	dpaa_fq = dpaa_fq_alloc(dev, 0, 1, list, FQ_TYPE_RX_ERROR);
+	if (!dpaa_fq)
+		goto fq_alloc_failed;
+
+	port_fqs->rx_errq = &dpaa_fq[0];
+
+	dpaa_fq = dpaa_fq_alloc(dev, 0, 1, list, FQ_TYPE_RX_DEFAULT);
+	if (!dpaa_fq)
+		goto fq_alloc_failed;
+
+	port_fqs->rx_defq = &dpaa_fq[0];
+
+	if (!dpaa_fq_alloc(dev, 0, DPAA_ETH_TXQ_NUM, list, FQ_TYPE_TX_CONF_MQ))
+		goto fq_alloc_failed;
+
+	dpaa_fq = dpaa_fq_alloc(dev, 0, 1, list, FQ_TYPE_TX_ERROR);
+	if (!dpaa_fq)
+		goto fq_alloc_failed;
+
+	port_fqs->tx_errq = &dpaa_fq[0];
+
+	dpaa_fq = dpaa_fq_alloc(dev, 0, 1, list, FQ_TYPE_TX_CONFIRM);
+	if (!dpaa_fq)
+		goto fq_alloc_failed;
+
+	port_fqs->tx_defq = &dpaa_fq[0];
+
+	if (!dpaa_fq_alloc(dev, 0, DPAA_ETH_TXQ_NUM, list, FQ_TYPE_TX))
+		goto fq_alloc_failed;
+
+	return 0;
+
+fq_alloc_failed:
+	dev_err(dev, "dpaa_fq_alloc() failed\n");
+	return -ENOMEM;
+}
+
+static u32 rx_pool_channel;
+static DEFINE_SPINLOCK(rx_pool_channel_init);
+
+static int dpaa_get_channel(void)
+{
+	spin_lock(&rx_pool_channel_init);
+	if (!rx_pool_channel) {
+		u32 pool;
+		int ret;
+
+		ret = qman_alloc_pool(&pool);
+
+		if (!ret)
+			rx_pool_channel = pool;
+	}
+	spin_unlock(&rx_pool_channel_init);
+	if (!rx_pool_channel)
+		return -ENOMEM;
+	return rx_pool_channel;
+}
+
+static void dpaa_release_channel(void)
+{
+	qman_release_pool(rx_pool_channel);
+}
+
+static void dpaa_eth_add_channel(u16 channel)
+{
+	u32 pool = QM_SDQCR_CHANNELS_POOL_CONV(channel);
+	const cpumask_t *cpus = qman_affine_cpus();
+	struct qman_portal *portal;
+	int cpu;
+
+	for_each_cpu(cpu, cpus) {
+		portal = qman_get_affine_portal(cpu);
+		qman_p_static_dequeue_add(portal, pool);
+	}
+}
+
+/* Congestion group state change notification callback.
+ * Stops the device's egress queues while they are congested and
+ * wakes them upon exiting congested state.
+ * Also updates some CGR-related stats.
+ */
+static void dpaa_eth_cgscn(struct qman_portal *qm, struct qman_cgr *cgr,
+			   int congested)
+{
+	struct dpaa_priv *priv = (struct dpaa_priv *)container_of(cgr,
+		struct dpaa_priv, cgr_data.cgr);
+
+	if (congested)
+		netif_tx_stop_all_queues(priv->net_dev);
+	else
+		netif_tx_wake_all_queues(priv->net_dev);
+}
+
+static int dpaa_eth_cgr_init(struct dpaa_priv *priv)
+{
+	struct qm_mcc_initcgr initcgr;
+	u32 cs_th;
+	int err;
+
+	err = qman_alloc_cgrid(&priv->cgr_data.cgr.cgrid);
+	if (err < 0) {
+		if (netif_msg_drv(priv))
+			pr_err("%s: Error %d allocating CGR ID\n",
+			       __func__, err);
+		goto out_error;
+	}
+	priv->cgr_data.cgr.cb = dpaa_eth_cgscn;
+
+	/* Enable Congestion State Change Notifications and CS taildrop */
+	initcgr.we_mask = QM_CGR_WE_CSCN_EN | QM_CGR_WE_CS_THRES;
+	initcgr.cgr.cscn_en = QM_CGR_EN;
+
+	/* Set different thresholds based on the MAC speed.
+	 * This may turn suboptimal if the MAC is reconfigured at a speed
+	 * lower than its max, e.g. if a dTSEC later negotiates a 100Mbps link.
+	 * In such cases, we ought to reconfigure the threshold, too.
+	 */
+	if (priv->mac_dev->if_support & SUPPORTED_10000baseT_Full)
+		cs_th = DPAA_CS_THRESHOLD_10G;
+	else
+		cs_th = DPAA_CS_THRESHOLD_1G;
+	qm_cgr_cs_thres_set64(&initcgr.cgr.cs_thres, cs_th, 1);
+
+	initcgr.we_mask |= QM_CGR_WE_CSTD_EN;
+	initcgr.cgr.cstd_en = QM_CGR_EN;
+
+	err = qman_create_cgr(&priv->cgr_data.cgr, QMAN_CGR_FLAG_USE_INIT,
+			      &initcgr);
+	if (err < 0) {
+		if (netif_msg_drv(priv))
+			pr_err("%s: Error %d creating CGR with ID %d\n",
+			       __func__, err, priv->cgr_data.cgr.cgrid);
+		qman_release_cgrid(priv->cgr_data.cgr.cgrid);
+		goto out_error;
+	}
+	if (netif_msg_drv(priv))
+		pr_debug("Created CGR %d for netdev with hwaddr %pM on QMan channel %d\n",
+			 priv->cgr_data.cgr.cgrid, priv->mac_dev->addr,
+			 priv->cgr_data.cgr.chan);
+
+out_error:
+	return err;
+}
+
+static inline void dpaa_setup_ingress(const struct dpaa_priv *priv,
+				      struct dpaa_fq *fq,
+				      const struct qman_fq *template)
+{
+	fq->fq_base = *template;
+	fq->net_dev = priv->net_dev;
+
+	fq->flags = QMAN_FQ_FLAG_NO_ENQUEUE;
+	fq->channel = priv->channel;
+}
+
+static inline void dpaa_setup_egress(const struct dpaa_priv *priv,
+				     struct dpaa_fq *fq,
+				     struct fman_port *port,
+				     const struct qman_fq *template)
+{
+	fq->fq_base = *template;
+	fq->net_dev = priv->net_dev;
+
+	if (port) {
+		fq->flags = QMAN_FQ_FLAG_TO_DCPORTAL;
+		fq->channel = (u16)fman_port_get_qman_channel_id(port);
+	} else {
+		fq->flags = QMAN_FQ_FLAG_NO_MODIFY;
+	}
+}
+
+static void 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, cpu;
+	const cpumask_t *affine_cpus = qman_affine_cpus();
+	u16 portals[NR_CPUS];
+	struct dpaa_fq *fq;
+
+	for_each_cpu(cpu, affine_cpus)
+		portals[num_portals++] = qman_affine_channel(cpu);
+	if (num_portals == 0)
+		dev_err(priv->net_dev->dev.parent,
+			"No Qman software (affine) channels found");
+
+	/* Initialize each FQ in the list */
+	list_for_each_entry(fq, &priv->dpaa_fq_list, list) {
+		switch (fq->fq_type) {
+		case FQ_TYPE_RX_DEFAULT:
+			dpaa_setup_ingress(priv, fq, &fq_cbs->rx_defq);
+			break;
+		case FQ_TYPE_RX_ERROR:
+			dpaa_setup_ingress(priv, fq, &fq_cbs->rx_errq);
+			break;
+		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;
+			break;
+		case FQ_TYPE_TX_CONF_MQ:
+			priv->conf_fqs[conf_cnt++] = &fq->fq_base;
+			/* fall through */
+		case FQ_TYPE_TX_CONFIRM:
+			dpaa_setup_ingress(priv, fq, &fq_cbs->tx_defq);
+			break;
+		case FQ_TYPE_TX_ERROR:
+			dpaa_setup_ingress(priv, fq, &fq_cbs->tx_errq);
+			break;
+		default:
+			dev_warn(priv->net_dev->dev.parent,
+				 "Unknown FQ type detected!\n");
+			break;
+		}
+	}
+
+	 /* 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;
+		}
+	}
+}
+
+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++)
+		if (priv->egress_fqs[i] == tx_fq)
+			return i;
+
+	return -EINVAL;
+}
+
+static int dpaa_fq_init(struct dpaa_fq *dpaa_fq, bool td_enable)
+{
+	const struct dpaa_priv	*priv;
+	struct qman_fq *confq = NULL;
+	struct qm_mcc_initfq initfq;
+	struct device *dev;
+	struct qman_fq *fq;
+	int queue_id;
+	int err;
+
+	priv = netdev_priv(dpaa_fq->net_dev);
+	dev = dpaa_fq->net_dev->dev.parent;
+
+	if (dpaa_fq->fqid == 0)
+		dpaa_fq->flags |= QMAN_FQ_FLAG_DYNAMIC_FQID;
+
+	dpaa_fq->init = !(dpaa_fq->flags & QMAN_FQ_FLAG_NO_MODIFY);
+
+	err = qman_create_fq(dpaa_fq->fqid, dpaa_fq->flags, &dpaa_fq->fq_base);
+	if (err) {
+		dev_err(dev, "qman_create_fq() failed\n");
+		return err;
+	}
+	fq = &dpaa_fq->fq_base;
+
+	if (dpaa_fq->init) {
+		memset(&initfq, 0, sizeof(initfq));
+
+		initfq.we_mask = QM_INITFQ_WE_FQCTRL;
+		/* Note: we may get to keep an empty FQ in cache */
+		initfq.fqd.fq_ctrl = QM_FQCTRL_PREFERINCACHE;
+
+		/* Try to reduce the number of portal interrupts for
+		 * Tx Confirmation FQs.
+		 */
+		if (dpaa_fq->fq_type == FQ_TYPE_TX_CONFIRM)
+			initfq.fqd.fq_ctrl |= QM_FQCTRL_HOLDACTIVE;
+
+		/* FQ placement */
+		initfq.we_mask |= QM_INITFQ_WE_DESTWQ;
+
+		qm_fqd_set_destwq(&initfq.fqd, dpaa_fq->channel, dpaa_fq->wq);
+
+		/* Put all egress queues in a congestion group of their own.
+		 * Sensu stricto, the Tx confirmation queues are Rx FQs,
+		 * rather than Tx - but they nonetheless account for the
+		 * memory footprint on behalf of egress traffic. We therefore
+		 * place them in the netdev's CGR, along with the Tx FQs.
+		 */
+		if (dpaa_fq->fq_type == FQ_TYPE_TX ||
+		    dpaa_fq->fq_type == FQ_TYPE_TX_CONFIRM ||
+		    dpaa_fq->fq_type == FQ_TYPE_TX_CONF_MQ) {
+			initfq.we_mask |= QM_INITFQ_WE_CGID;
+			initfq.fqd.fq_ctrl |= QM_FQCTRL_CGE;
+			initfq.fqd.cgid = (u8)priv->cgr_data.cgr.cgrid;
+			/* Set a fixed overhead accounting, in an attempt to
+			 * reduce the impact of fixed-size skb shells and the
+			 * driver's needed headroom on system memory. This is
+			 * especially the case when the egress traffic is
+			 * composed of small datagrams.
+			 * Unfortunately, QMan's OAL value is capped to an
+			 * insufficient value, but even that is better than
+			 * no overhead accounting at all.
+			 */
+			initfq.we_mask |= QM_INITFQ_WE_OAC;
+			qm_fqd_set_oac(&initfq.fqd, QM_OAC_CG);
+			qm_fqd_set_oal(&initfq.fqd,
+				       min(sizeof(struct sk_buff) +
+				       priv->tx_headroom,
+				       (size_t)FSL_QMAN_MAX_OAL));
+		}
+
+		if (td_enable) {
+			initfq.we_mask |= QM_INITFQ_WE_TDTHRESH;
+			qm_fqd_set_taildrop(&initfq.fqd, DPAA_FQ_TD, 1);
+			initfq.fqd.fq_ctrl = QM_FQCTRL_TDE;
+		}
+
+		if (dpaa_fq->fq_type == FQ_TYPE_TX) {
+			queue_id = dpaa_tx_fq_to_id(priv, &dpaa_fq->fq_base);
+			if (queue_id >= 0)
+				confq = priv->conf_fqs[queue_id];
+			if (confq) {
+				initfq.we_mask |= QM_INITFQ_WE_CONTEXTA;
+			/* ContextA: OVOM=1(use contextA2 bits instead of ICAD)
+			 *	     A2V=1 (contextA A2 field is valid)
+			 *	     A0V=1 (contextA A0 field is valid)
+			 *	     B0V=1 (contextB field is valid)
+			 * ContextA A2: EBD=1 (deallocate buffers inside FMan)
+			 * ContextB B0(ASPID): 0 (absolute Virtual Storage ID)
+			 */
+				initfq.fqd.context_a.hi = 0x1e000000;
+				initfq.fqd.context_a.lo = 0x80000000;
+			}
+		}
+
+		/* Put all the ingress queues in our "ingress CGR". */
+		if (priv->use_ingress_cgr &&
+		    (dpaa_fq->fq_type == FQ_TYPE_RX_DEFAULT ||
+		     dpaa_fq->fq_type == FQ_TYPE_RX_ERROR)) {
+			initfq.we_mask |= QM_INITFQ_WE_CGID;
+			initfq.fqd.fq_ctrl |= QM_FQCTRL_CGE;
+			initfq.fqd.cgid = (u8)priv->ingress_cgr.cgrid;
+			/* Set a fixed overhead accounting, just like for the
+			 * egress CGR.
+			 */
+			initfq.we_mask |= QM_INITFQ_WE_OAC;
+			qm_fqd_set_oac(&initfq.fqd, QM_OAC_CG);
+			qm_fqd_set_oal(&initfq.fqd,
+				       min(sizeof(struct sk_buff) +
+				       priv->tx_headroom,
+				       (size_t)FSL_QMAN_MAX_OAL));
+		}
+
+		/* Initialization common to all ingress queues */
+		if (dpaa_fq->flags & QMAN_FQ_FLAG_NO_ENQUEUE) {
+			initfq.we_mask |= QM_INITFQ_WE_CONTEXTA;
+			initfq.fqd.fq_ctrl |=
+				QM_FQCTRL_HOLDACTIVE;
+			initfq.fqd.context_a.stashing.exclusive =
+				QM_STASHING_EXCL_DATA | QM_STASHING_EXCL_CTX |
+				QM_STASHING_EXCL_ANNOTATION;
+			qm_fqd_set_stashing(&initfq.fqd, 1, 2,
+					    DIV_ROUND_UP(sizeof(struct qman_fq),
+							 64));
+		}
+
+		err = qman_init_fq(fq, QMAN_INITFQ_FLAG_SCHED, &initfq);
+		if (err < 0) {
+			dev_err(dev, "qman_init_fq(%u) = %d\n",
+				qman_fq_fqid(fq), err);
+			qman_destroy_fq(fq);
+			return err;
+		}
+	}
+
+	dpaa_fq->fqid = qman_fq_fqid(fq);
+
+	return 0;
+}
+
+static int dpaa_fq_free_entry(struct device *dev, struct qman_fq *fq)
+{
+	const struct dpaa_priv  *priv;
+	struct dpaa_fq *dpaa_fq;
+	int err, error;
+
+	err = 0;
+
+	dpaa_fq = container_of(fq, struct dpaa_fq, fq_base);
+	priv = netdev_priv(dpaa_fq->net_dev);
+
+	if (dpaa_fq->init) {
+		err = qman_retire_fq(fq, NULL);
+		if (err < 0 && netif_msg_drv(priv))
+			dev_err(dev, "qman_retire_fq(%u) = %d\n",
+				qman_fq_fqid(fq), err);
+
+		error = qman_oos_fq(fq);
+		if (error < 0 && netif_msg_drv(priv)) {