3 files changed, 268 insertions, 107 deletions

diff --git a/drivers/net/ethernet/marvell/mvpp2/mvpp2.h b/drivers/net/ethernet/marvell/mvpp2/mvpp2.h
index 4d9564ba68f6..543a310ec102 100644
--- a/drivers/net/ethernet/marvell/mvpp2/mvpp2.h
+++ b/drivers/net/ethernet/marvell/mvpp2/mvpp2.h
@@ -683,6 +683,7 @@ enum mvpp2_prs_l3_cast {
 #define MVPP2_BM_SHORT_BUF_NUM		2048
 #define MVPP2_BM_POOL_SIZE_MAX		(16*1024 - MVPP2_BM_POOL_PTR_ALIGN/4)
 #define MVPP2_BM_POOL_PTR_ALIGN		128
+#define MVPP2_BM_MAX_POOLS		8
 
 /* BM cookie (32 bits) definition */
 #define MVPP2_BM_COOKIE_POOL_OFFS	8
@@ -787,6 +788,9 @@ struct mvpp2 {
 	/* Aggregated TXQs */
 	struct mvpp2_tx_queue *aggr_txqs;
 
+	/* Are we using page_pool with per-cpu pools? */
+	int percpu_pools;
+
 	/* BM pools */
 	struct mvpp2_bm_pool *bm_pools;
 
@@ -829,9 +833,8 @@ struct mvpp2_pcpu_stats {
 /* Per-CPU port control */
 struct mvpp2_port_pcpu {
 	struct hrtimer tx_done_timer;
+	struct net_device *dev;
 	bool timer_scheduled;
-	/* Tasklet for egress finalization */
-	struct tasklet_struct tx_done_tasklet;
 };
 
 struct mvpp2_queue_vector {
diff --git a/drivers/net/ethernet/marvell/mvpp2/mvpp2_debugfs.c b/drivers/net/ethernet/marvell/mvpp2/mvpp2_debugfs.c
index 274fb07362cb..4a3baa7e0142 100644
--- a/drivers/net/ethernet/marvell/mvpp2/mvpp2_debugfs.c
+++ b/drivers/net/ethernet/marvell/mvpp2/mvpp2_debugfs.c
@@ -452,8 +452,6 @@ static int mvpp2_dbgfs_flow_port_init(struct dentry *parent,
 	struct dentry *port_dir;
 
 	port_dir = debugfs_create_dir(port->dev->name, parent);
-	if (IS_ERR(port_dir))
-		return PTR_ERR(port_dir);
 
 	port_entry = &port->priv->dbgfs_entries->port_flow_entries[port->id];
 
@@ -480,8 +478,6 @@ static int mvpp2_dbgfs_flow_entry_init(struct dentry *parent,
 	sprintf(flow_entry_name, "%02d", flow);
 
 	flow_entry_dir = debugfs_create_dir(flow_entry_name, parent);
-	if (!flow_entry_dir)
-		return -ENOMEM;
 
 	entry = &priv->dbgfs_entries->flow_entries[flow];
 
@@ -514,8 +510,6 @@ static int mvpp2_dbgfs_flow_init(struct dentry *parent, struct mvpp2 *priv)
 	int i, ret;
 
 	flow_dir = debugfs_create_dir("flows", parent);
-	if (!flow_dir)
-		return -ENOMEM;
 
 	for (i = 0; i < MVPP2_N_PRS_FLOWS; i++) {
 		ret = mvpp2_dbgfs_flow_entry_init(flow_dir, priv, i);
@@ -539,8 +533,6 @@ static int mvpp2_dbgfs_prs_entry_init(struct dentry *parent,
 	sprintf(prs_entry_name, "%03d", tid);
 
 	prs_entry_dir = debugfs_create_dir(prs_entry_name, parent);
-	if (!prs_entry_dir)
-		return -ENOMEM;
 
 	entry = &priv->dbgfs_entries->prs_entries[tid];
 
@@ -578,8 +570,6 @@ static int mvpp2_dbgfs_prs_init(struct dentry *parent, struct mvpp2 *priv)
 	int i, ret;
 
 	prs_dir = debugfs_create_dir("parser", parent);
-	if (!prs_dir)
-		return -ENOMEM;
 
 	for (i = 0; i < MVPP2_PRS_TCAM_SRAM_SIZE; i++) {
 		ret = mvpp2_dbgfs_prs_entry_init(prs_dir, priv, i);
@@ -688,8 +678,6 @@ static int mvpp2_dbgfs_port_init(struct dentry *parent,
 	struct dentry *port_dir;
 
 	port_dir = debugfs_create_dir(port->dev->name, parent);
-	if (IS_ERR(port_dir))
-		return PTR_ERR(port_dir);
 
 	debugfs_create_file("parser_entries", 0444, port_dir, port,
 			    &mvpp2_dbgfs_port_parser_fops);
@@ -716,15 +704,10 @@ void mvpp2_dbgfs_init(struct mvpp2 *priv, const char *name)
 	int ret, i;
 
 	mvpp2_root = debugfs_lookup(MVPP2_DRIVER_NAME, NULL);
-	if (!mvpp2_root) {
+	if (!mvpp2_root)
 		mvpp2_root = debugfs_create_dir(MVPP2_DRIVER_NAME, NULL);
-		if (IS_ERR(mvpp2_root))
-			return;
-	}
 
 	mvpp2_dir = debugfs_create_dir(name, mvpp2_root);
-	if (IS_ERR(mvpp2_dir))
-		return;
 
 	priv->dbgfs_dir = mvpp2_dir;
 	priv->dbgfs_entries = kzalloc(sizeof(*priv->dbgfs_entries), GFP_KERNEL);
diff --git a/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c b/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
index ccdd47f3b8fb..111b3b8239e1 100644
--- a/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
+++ b/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
@@ -292,6 +292,26 @@ static void mvpp2_txq_inc_put(struct mvpp2_port *port,
 		txq_pcpu->txq_put_index = 0;
 }
 
+/* Get number of maximum RXQ */
+static int mvpp2_get_nrxqs(struct mvpp2 *priv)
+{
+	unsigned int nrxqs;
+
+	if (priv->hw_version == MVPP22 && queue_mode == MVPP2_QDIST_SINGLE_MODE)
+		return 1;
+
+	/* According to the PPv2.2 datasheet and our experiments on
+	 * PPv2.1, RX queues have an allocation granularity of 4 (when
+	 * more than a single one on PPv2.2).
+	 * Round up to nearest multiple of 4.
+	 */
+	nrxqs = (num_possible_cpus() + 3) & ~0x3;
+	if (nrxqs > MVPP2_PORT_MAX_RXQ)
+		nrxqs = MVPP2_PORT_MAX_RXQ;
+
+	return nrxqs;
+}
+
 /* Get number of physical egress port */
 static inline int mvpp2_egress_port(struct mvpp2_port *port)
 {
@@ -323,8 +343,7 @@ static void mvpp2_frag_free(const struct mvpp2_bm_pool *pool, void *data)
 /* Buffer Manager configuration routines */
 
 /* Create pool */
-static int mvpp2_bm_pool_create(struct platform_device *pdev,
-				struct mvpp2 *priv,
+static int mvpp2_bm_pool_create(struct device *dev, struct mvpp2 *priv,
 				struct mvpp2_bm_pool *bm_pool, int size)
 {
 	u32 val;
@@ -343,7 +362,7 @@ static int mvpp2_bm_pool_create(struct platform_device *pdev,
 	else
 		bm_pool->size_bytes = 2 * sizeof(u64) * size;
 
-	bm_pool->virt_addr = dma_alloc_coherent(&pdev->dev, bm_pool->size_bytes,
+	bm_pool->virt_addr = dma_alloc_coherent(dev, bm_pool->size_bytes,
 						&bm_pool->dma_addr,
 						GFP_KERNEL);
 	if (!bm_pool->virt_addr)
@@ -351,9 +370,9 @@ static int mvpp2_bm_pool_create(struct platform_device *pdev,
 
 	if (!IS_ALIGNED((unsigned long)bm_pool->virt_addr,
 			MVPP2_BM_POOL_PTR_ALIGN)) {
-		dma_free_coherent(&pdev->dev, bm_pool->size_bytes,
+		dma_free_coherent(dev, bm_pool->size_bytes,
 				  bm_pool->virt_addr, bm_pool->dma_addr);
-		dev_err(&pdev->dev, "BM pool %d is not %d bytes aligned\n",
+		dev_err(dev, "BM pool %d is not %d bytes aligned\n",
 			bm_pool->id, MVPP2_BM_POOL_PTR_ALIGN);
 		return -ENOMEM;
 	}
@@ -468,15 +487,14 @@ static int mvpp2_check_hw_buf_num(struct mvpp2 *priv, struct mvpp2_bm_pool *bm_p
 }
 
 /* Cleanup pool */
-static int mvpp2_bm_pool_destroy(struct platform_device *pdev,
-				 struct mvpp2 *priv,
+static int mvpp2_bm_pool_destroy(struct device *dev, struct mvpp2 *priv,
 				 struct mvpp2_bm_pool *bm_pool)
 {
 	int buf_num;
 	u32 val;
 
 	buf_num = mvpp2_check_hw_buf_num(priv, bm_pool);
-	mvpp2_bm_bufs_free(&pdev->dev, priv, bm_pool, buf_num);
+	mvpp2_bm_bufs_free(dev, priv, bm_pool, buf_num);
 
 	/* Check buffer counters after free */
 	buf_num = mvpp2_check_hw_buf_num(priv, bm_pool);
@@ -490,24 +508,26 @@ static int mvpp2_bm_pool_destroy(struct platform_device *pdev,
 	val |= MVPP2_BM_STOP_MASK;
 	mvpp2_write(priv, MVPP2_BM_POOL_CTRL_REG(bm_pool->id), val);
 
-	dma_free_coherent(&pdev->dev, bm_pool->size_bytes,
+	dma_free_coherent(dev, bm_pool->size_bytes,
 			  bm_pool->virt_addr,
 			  bm_pool->dma_addr);
 	return 0;
 }
 
-static int mvpp2_bm_pools_init(struct platform_device *pdev,
-			       struct mvpp2 *priv)
+static int mvpp2_bm_pools_init(struct device *dev, struct mvpp2 *priv)
 {
-	int i, err, size;
+	int i, err, size, poolnum = MVPP2_BM_POOLS_NUM;
 	struct mvpp2_bm_pool *bm_pool;
 
+	if (priv->percpu_pools)
+		poolnum = mvpp2_get_nrxqs(priv) * 2;
+
 	/* Create all pools with maximum size */
 	size = MVPP2_BM_POOL_SIZE_MAX;
-	for (i = 0; i < MVPP2_BM_POOLS_NUM; i++) {
+	for (i = 0; i < poolnum; i++) {
 		bm_pool = &priv->bm_pools[i];
 		bm_pool->id = i;
-		err = mvpp2_bm_pool_create(pdev, priv, bm_pool, size);
+		err = mvpp2_bm_pool_create(dev, priv, bm_pool, size);
 		if (err)
 			goto err_unroll_pools;
 		mvpp2_bm_pool_bufsize_set(priv, bm_pool, 0);
@@ -515,17 +535,23 @@ static int mvpp2_bm_pools_init(struct platform_device *pdev,
 	return 0;
 
 err_unroll_pools:
-	dev_err(&pdev->dev, "failed to create BM pool %d, size %d\n", i, size);
+	dev_err(dev, "failed to create BM pool %d, size %d\n", i, size);
 	for (i = i - 1; i >= 0; i--)
-		mvpp2_bm_pool_destroy(pdev, priv, &priv->bm_pools[i]);
+		mvpp2_bm_pool_destroy(dev, priv, &priv->bm_pools[i]);
 	return err;
 }
 
-static int mvpp2_bm_init(struct platform_device *pdev, struct mvpp2 *priv)
+static int mvpp2_bm_init(struct device *dev, struct mvpp2 *priv)
 {
-	int i, err;
+	int i, err, poolnum = MVPP2_BM_POOLS_NUM;
 
-	for (i = 0; i < MVPP2_BM_POOLS_NUM; i++) {
+	if (priv->percpu_pools)
+		poolnum = mvpp2_get_nrxqs(priv) * 2;
+
+	dev_info(dev, "using %d %s buffers\n", poolnum,
+		 priv->percpu_pools ? "per-cpu" : "shared");
+
+	for (i = 0; i < poolnum; i++) {
 		/* Mask BM all interrupts */
 		mvpp2_write(priv, MVPP2_BM_INTR_MASK_REG(i), 0);
 		/* Clear BM cause register */
@@ -533,12 +559,12 @@ static int mvpp2_bm_init(struct platform_device *pdev, struct mvpp2 *priv)
 	}
 
 	/* Allocate and initialize BM pools */
-	priv->bm_pools = devm_kcalloc(&pdev->dev, MVPP2_BM_POOLS_NUM,
+	priv->bm_pools = devm_kcalloc(dev, poolnum,
 				      sizeof(*priv->bm_pools), GFP_KERNEL);
 	if (!priv->bm_pools)
 		return -ENOMEM;
 
-	err = mvpp2_bm_pools_init(pdev, priv);
+	err = mvpp2_bm_pools_init(dev, priv);
 	if (err < 0)
 		return err;
 	return 0;
@@ -679,6 +705,13 @@ static int mvpp2_bm_bufs_add(struct mvpp2_port *port,
 	phys_addr_t phys_addr;
 	void *buf;
 
+	if (port->priv->percpu_pools &&
+	    bm_pool->pkt_size > MVPP2_BM_LONG_PKT_SIZE) {
+		netdev_err(port->dev,
+			   "attempted to use jumbo frames with per-cpu pools");
+		return 0;
+	}
+
 	buf_size = MVPP2_RX_BUF_SIZE(bm_pool->pkt_size);
 	total_size = MVPP2_RX_TOTAL_SIZE(buf_size);
 
@@ -722,7 +755,64 @@ mvpp2_bm_pool_use(struct mvpp2_port *port, unsigned pool, int pkt_size)
 	struct mvpp2_bm_pool *new_pool = &port->priv->bm_pools[pool];
 	int num;
 
-	if (pool >= MVPP2_BM_POOLS_NUM) {
+	if ((port->priv->percpu_pools && pool > mvpp2_get_nrxqs(port->priv) * 2) ||
+	    (!port->priv->percpu_pools && pool >= MVPP2_BM_POOLS_NUM)) {
+		netdev_err(port->dev, "Invalid pool %d\n", pool);
+		return NULL;
+	}
+
+	/* Allocate buffers in case BM pool is used as long pool, but packet
+	 * size doesn't match MTU or BM pool hasn't being used yet
+	 */
+	if (new_pool->pkt_size == 0) {
+		int pkts_num;
+
+		/* Set default buffer number or free all the buffers in case
+		 * the pool is not empty
+		 */
+		pkts_num = new_pool->buf_num;
+		if (pkts_num == 0) {
+			if (port->priv->percpu_pools) {
+				if (pool < port->nrxqs)
+					pkts_num = mvpp2_pools[MVPP2_BM_SHORT].buf_num;
+				else
+					pkts_num = mvpp2_pools[MVPP2_BM_LONG].buf_num;
+			} else {
+				pkts_num = mvpp2_pools[pool].buf_num;
+			}
+		} else {
+			mvpp2_bm_bufs_free(port->dev->dev.parent,
+					   port->priv, new_pool, pkts_num);
+		}
+
+		new_pool->pkt_size = pkt_size;
+		new_pool->frag_size =
+			SKB_DATA_ALIGN(MVPP2_RX_BUF_SIZE(pkt_size)) +
+			MVPP2_SKB_SHINFO_SIZE;
+
+		/* Allocate buffers for this pool */
+		num = mvpp2_bm_bufs_add(port, new_pool, pkts_num);
+		if (num != pkts_num) {
+			WARN(1, "pool %d: %d of %d allocated\n",
+			     new_pool->id, num, pkts_num);
+			return NULL;
+		}
+	}
+
+	mvpp2_bm_pool_bufsize_set(port->priv, new_pool,
+				  MVPP2_RX_BUF_SIZE(new_pool->pkt_size));
+
+	return new_pool;
+}
+
+static struct mvpp2_bm_pool *
+mvpp2_bm_pool_use_percpu(struct mvpp2_port *port, int type,
+			 unsigned int pool, int pkt_size)
+{
+	struct mvpp2_bm_pool *new_pool = &port->priv->bm_pools[pool];
+	int num;
+
+	if (pool > port->nrxqs * 2) {
 		netdev_err(port->dev, "Invalid pool %d\n", pool);
 		return NULL;
 	}
@@ -738,7 +828,7 @@ mvpp2_bm_pool_use(struct mvpp2_port *port, unsigned pool, int pkt_size)
 		 */
 		pkts_num = new_pool->buf_num;
 		if (pkts_num == 0)
-			pkts_num = mvpp2_pools[pool].buf_num;
+			pkts_num = mvpp2_pools[type].buf_num;
 		else
 			mvpp2_bm_bufs_free(port->dev->dev.parent,
 					   port->priv, new_pool, pkts_num);
@@ -763,11 +853,11 @@ mvpp2_bm_pool_use(struct mvpp2_port *port, unsigned pool, int pkt_size)
 	return new_pool;
 }
 
-/* Initialize pools for swf */
-static int mvpp2_swf_bm_pool_init(struct mvpp2_port *port)
+/* Initialize pools for swf, shared buffers variant */
+static int mvpp2_swf_bm_pool_init_shared(struct mvpp2_port *port)
 {
-	int rxq;
 	enum mvpp2_bm_pool_log_num long_log_pool, short_log_pool;
+	int rxq;
 
 	/* If port pkt_size is higher than 1518B:
 	 * HW Long pool - SW Jumbo pool, HW Short pool - SW Long pool
@@ -811,6 +901,47 @@ static int mvpp2_swf_bm_pool_init(struct mvpp2_port *port)
 	return 0;
 }
 
+/* Initialize pools for swf, percpu buffers variant */
+static int mvpp2_swf_bm_pool_init_percpu(struct mvpp2_port *port)
+{
+	struct mvpp2_bm_pool *p;
+	int i;
+
+	for (i = 0; i < port->nrxqs; i++) {
+		p = mvpp2_bm_pool_use_percpu(port, MVPP2_BM_SHORT, i,
+					     mvpp2_pools[MVPP2_BM_SHORT].pkt_size);
+		if (!p)
+			return -ENOMEM;
+
+		port->priv->bm_pools[i].port_map |= BIT(port->id);
+		mvpp2_rxq_short_pool_set(port, i, port->priv->bm_pools[i].id);
+	}
+
+	for (i = 0; i < port->nrxqs; i++) {
+		p = mvpp2_bm_pool_use_percpu(port, MVPP2_BM_LONG, i + port->nrxqs,
+					     mvpp2_pools[MVPP2_BM_LONG].pkt_size);
+		if (!p)
+			return -ENOMEM;
+
+		port->priv->bm_pools[i + port->nrxqs].port_map |= BIT(port->id);
+		mvpp2_rxq_long_pool_set(port, i,
+					port->priv->bm_pools[i + port->nrxqs].id);
+	}
+
+	port->pool_long = NULL;
+	port->pool_short = NULL;
+
+	return 0;
+}
+
+static int mvpp2_swf_bm_pool_init(struct mvpp2_port *port)
+{
+	if (port->priv->percpu_pools)
+		return mvpp2_swf_bm_pool_init_percpu(port);
+	else
+		return mvpp2_swf_bm_pool_init_shared(port);
+}
+
 static void mvpp2_set_hw_csum(struct mvpp2_port *port,
 			      enum mvpp2_bm_pool_log_num new_long_pool)
 {
@@ -837,6 +968,9 @@ static int mvpp2_bm_update_mtu(struct net_device *dev, int mtu)
 	enum mvpp2_bm_pool_log_num new_long_pool;
 	int pkt_size = MVPP2_RX_PKT_SIZE(mtu);
 
+	if (port->priv->percpu_pools)
+		goto out_set;
+
 	/* If port MTU is higher than 1518B:
 	 * HW Long pool - SW Jumbo pool, HW Short pool - SW Long pool
 	 * else: HW Long pool - SW Long pool, HW Short pool - SW Short pool
@@ -866,6 +1000,7 @@ static int mvpp2_bm_update_mtu(struct net_device *dev, int mtu)
 		mvpp2_set_hw_csum(port, new_long_pool);
 	}
 
+out_set:
 	dev->mtu = mtu;
 	dev->wanted_features = dev->features;
 
@@ -2651,31 +2786,21 @@ handled:
 	return IRQ_HANDLED;
 }
 
-static void mvpp2_timer_set(struct mvpp2_port_pcpu *port_pcpu)
-{
-	ktime_t interval;
-
-	if (!port_pcpu->timer_scheduled) {
-		port_pcpu->timer_scheduled = true;
-		interval = MVPP2_TXDONE_HRTIMER_PERIOD_NS;
-		hrtimer_start(&port_pcpu->tx_done_timer, interval,
-			      HRTIMER_MODE_REL_PINNED);
-	}
-}
-
-static void mvpp2_tx_proc_cb(unsigned long data)
+static enum hrtimer_restart mvpp2_hr_timer_cb(struct hrtimer *timer)
 {
-	struct net_device *dev = (struct net_device *)data;
-	struct mvpp2_port *port = netdev_priv(dev);
+	struct net_device *dev;
+	struct mvpp2_port *port;
 	struct mvpp2_port_pcpu *port_pcpu;
 	unsigned int tx_todo, cause;
 
-	port_pcpu = per_cpu_ptr(port->pcpu,
-				mvpp2_cpu_to_thread(port->priv, smp_processor_id()));
+	port_pcpu = container_of(timer, struct mvpp2_port_pcpu, tx_done_timer);
+	dev = port_pcpu->dev;
 
 	if (!netif_running(dev))
-		return;
+		return HRTIMER_NORESTART;
+
 	port_pcpu->timer_scheduled = false;
+	port = netdev_priv(dev);
 
 	/* Process all the Tx queues */
 	cause = (1 << port->ntxqs) - 1;
@@ -2683,18 +2808,13 @@ static void mvpp2_tx_proc_cb(unsigned long data)
 				mvpp2_cpu_to_thread(port->priv, smp_processor_id()));
 
 	/* Set the timer in case not all the packets were processed */
-	if (tx_todo)
-		mvpp2_timer_set(port_pcpu);
-}
-
-static enum hrtimer_restart mvpp2_hr_timer_cb(struct hrtimer *timer)
-{
-	struct mvpp2_port_pcpu *port_pcpu = container_of(timer,
-							 struct mvpp2_port_pcpu,
-							 tx_done_timer);
-
-	tasklet_schedule(&port_pcpu->tx_done_tasklet);
+	if (tx_todo && !port_pcpu->timer_scheduled) {
+		port_pcpu->timer_scheduled = true;
+		hrtimer_forward_now(&port_pcpu->tx_done_timer,
+				    MVPP2_TXDONE_HRTIMER_PERIOD_NS);
 
+		return HRTIMER_RESTART;
+	}
 	return HRTIMER_NORESTART;
 }
 
@@ -2923,14 +3043,15 @@ static int mvpp2_tx_frag_process(struct mvpp2_port *port, struct sk_buff *skb,
 
 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
-		void *addr = page_address(frag->page.p) + frag->page_offset;
+		void *addr = skb_frag_address(frag);
 
 		tx_desc = mvpp2_txq_next_desc_get(aggr_txq);
 		mvpp2_txdesc_txq_set(port, tx_desc, txq->id);
-		mvpp2_txdesc_size_set(port, tx_desc, frag->size);
+		mvpp2_txdesc_size_set(port, tx_desc, skb_frag_size(frag));
 
 		buf_dma_addr = dma_map_single(port->dev->dev.parent, addr,
-					      frag->size, DMA_TO_DEVICE);
+					      skb_frag_size(frag),
+					      DMA_TO_DEVICE);
 		if (dma_mapping_error(port->dev->dev.parent, buf_dma_addr)) {
 			mvpp2_txq_desc_put(txq);
 			goto cleanup;
@@ -3181,7 +3302,12 @@ out:
 	    txq_pcpu->count > 0) {
 		struct mvpp2_port_pcpu *port_pcpu = per_cpu_ptr(port->pcpu, thread);
 
-		mvpp2_timer_set(port_pcpu);
+		if (!port_pcpu->timer_scheduled) {
+			port_pcpu->timer_scheduled = true;
+			hrtimer_start(&port_pcpu->tx_done_timer,
+				      MVPP2_TXDONE_HRTIMER_PERIOD_NS,
+				      HRTIMER_MODE_REL_PINNED_SOFT);
+		}
 	}
 
 	if (test_bit(thread, &port->priv->lock_map))
@@ -3618,7 +3744,6 @@ static int mvpp2_stop(struct net_device *dev)
 
 			hrtimer_cancel(&port_pcpu->tx_done_timer);
 			port_pcpu->timer_scheduled = false;
-			tasklet_kill(&port_pcpu->tx_done_tasklet);
 		}
 	}
 	mvpp2_cleanup_rxqs(port);
@@ -3709,10 +3834,48 @@ static int mvpp2_set_mac_address(struct net_device *dev, void *p)
 	return err;
 }
 
+/* Shut down all the ports, reconfigure the pools as percpu or shared,
+ * then bring up again all ports.
+ */
+static int mvpp2_bm_switch_buffers(struct mvpp2 *priv, bool percpu)
+{
+	int numbufs = MVPP2_BM_POOLS_NUM, i;
+	struct mvpp2_port *port = NULL;
+	bool status[MVPP2_MAX_PORTS];
+
+	for (i = 0; i < priv->port_count; i++) {
+		port = priv->port_list[i];
+		status[i] = netif_running(port->dev);
+		if (status[i])
+			mvpp2_stop(port->dev);
+	}
+
+	/* nrxqs is the same for all ports */
+	if (priv->percpu_pools)
+		numbufs = port->nrxqs * 2;
+
+	for (i = 0; i < numbufs; i++)
+		mvpp2_bm_pool_destroy(port->dev->dev.parent, priv, &priv->bm_pools[i]);
+
+	devm_kfree(port->dev->dev.parent, priv->bm_pools);
+	priv->percpu_pools = percpu;
+	mvpp2_bm_init(port->dev->dev.parent, priv);
+
+	for (i = 0; i < priv->port_count; i++) {
+		port = priv->port_list[i];
+		mvpp2_swf_bm_pool_init(port);
+		if (status[i])
+			mvpp2_open(port->dev);
+	}
+
+	return 0;
+}
+
 static int mvpp2_change_mtu(struct net_device *dev, int mtu)
 {
 	struct mvpp2_port *port = netdev_priv(dev);
 	bool running = netif_running(dev);
+	struct mvpp2 *priv = port->priv;
 	int err;
 
 	if (!IS_ALIGNED(MVPP2_RX_PKT_SIZE(mtu), 8)) {
@@ -3721,6 +3884,31 @@ static int mvpp2_change_mtu(struct net_device *dev, int mtu)
 		mtu = ALIGN(MVPP2_RX_PKT_SIZE(mtu), 8);
 	}
 
+	if (MVPP2_RX_PKT_SIZE(mtu) > MVPP2_BM_LONG_PKT_SIZE) {
+		if (priv->percpu_pools) {
+			netdev_warn(dev, "mtu %d too high, switching to shared buffers", mtu);
+			mvpp2_bm_switch_buffers(priv, false);
+		}
+	} else {
+		bool jumbo = false;
+		int i;
+
+		for (i = 0; i < priv->port_count; i++)
+			if (priv->port_list[i] != port &&
+			    MVPP2_RX_PKT_SIZE(priv->port_list[i]->dev->mtu) >
+			    MVPP2_BM_LONG_PKT_SIZE) {
+				jumbo = true;
+				break;
+			}
+
+		/* No port is using jumbo frames */
+		if (!jumbo) {
+			dev_info(port->dev->dev.parent,
+				 "all ports have a low MTU, switching to per-cpu buffers");
+			mvpp2_bm_switch_buffers(priv, true);
+		}
+	}
+
 	if (running)
 		mvpp2_stop_dev(port);
 
@@ -5010,7 +5198,6 @@ static int mvpp2_port_probe(struct platform_device *pdev,
 	struct device_node *port_node = to_of_node(port_fwnode);
 	netdev_features_t features;
 	struct net_device *dev;
-	struct resource *res;
 	struct phylink *phylink;
 	char *mac_from = "";
 	unsigned int ntxqs, nrxqs, thread;
@@ -5028,18 +5215,7 @@ static int mvpp2_port_probe(struct platform_device *pdev,
 	}
 
 	ntxqs = MVPP2_MAX_TXQ;
-	if (priv->hw_version == MVPP22 && queue_mode == MVPP2_QDIST_SINGLE_MODE) {
-		nrxqs = 1;
-	} else {
-		/* According to the PPv2.2 datasheet and our experiments on
-		 * PPv2.1, RX queues have an allocation granularity of 4 (when
-		 * more than a single one on PPv2.2).
-		 * Round up to nearest multiple of 4.
-		 */
-		nrxqs = (num_possible_cpus() + 3) & ~0x3;
-		if (nrxqs > MVPP2_PORT_MAX_RXQ)
-			nrxqs = MVPP2_PORT_MAX_RXQ;
-	}
+	nrxqs = mvpp2_get_nrxqs(priv);
 
 	dev = alloc_etherdev_mqs(sizeof(*port), ntxqs, nrxqs);
 	if (!dev)
@@ -5114,8 +5290,7 @@ static int mvpp2_port_probe(struct platform_device *pdev,
 	port->comphy = comphy;
 
 	if (priv->hw_version == MVPP21) {
-		res = platform_get_resource(pdev, IORESOURCE_MEM, 2 + id);
-		port->base = devm_ioremap_resource(&pdev->dev, res);
+		port->base = devm_platform_ioremap_resource(pdev, 2 + id);
 		if (IS_ERR(port->base)) {
 			err = PTR_ERR(port->base);
 			goto err_free_irq;
@@ -5184,13 +5359,10 @@ static int mvpp2_port_probe(struct platform_device *pdev,
 			port_pcpu = per_cpu_ptr(port->pcpu, thread);
 
 			hrtimer_init(&port_pcpu->tx_done_timer, CLOCK_MONOTONIC,
-				     HRTIMER_MODE_REL_PINNED);
+				     HRTIMER_MODE_REL_PINNED_SOFT);
 			port_pcpu->tx_done_timer.function = mvpp2_hr_timer_cb;
 			port_pcpu->timer_scheduled = false;
-
-			tasklet_init(&port_pcpu->tx_done_tasklet,
-				     mvpp2_tx_proc_cb,
-				     (unsigned long)dev);
+			port_pcpu->dev = dev;
 		}
 	}
 
@@ -5205,7 +5377,8 @@ static int mvpp2_port_probe(struct platform_device *pdev,
 		dev->features |= NETIF_F_NTUPLE;
 	}
 
-	mvpp2_set_hw_csum(port, port->pool_long->id);
+	if (!port->priv->percpu_pools)
+		mvpp2_set_hw_csum(port, port->pool_long->id);
 
 	dev->vlan_features |= features;
 	dev->gso_max_segs = MVPP2_MAX_TSO_SEGS;
@@ -5497,7 +5670,7 @@ static int mvpp2_init(struct platform_device *pdev, struct mvpp2 *priv)
 	mvpp2_write(priv, MVPP2_TX_SNOOP_REG, 0x1);
 
 	/* Buffer Manager initialization */
-	err = mvpp2_bm_init(pdev, priv);
+	err = mvpp2_bm_init(&pdev->dev, priv);
 	if (err < 0)
 		return err;
 
@@ -5544,14 +5717,12 @@ static int mvpp2_probe(struct platform_device *pdev)
 	if (priv->hw_version == MVPP21)
 		queue_mode = MVPP2_QDIST_SINGLE_MODE;
 
-	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
-	base = devm_ioremap_resource(&pdev->dev, res);
+	base = devm_platform_ioremap_resource(pdev, 0);
 	if (IS_ERR(base))
 		return PTR_ERR(base);
 
 	if (priv->hw_version == MVPP21) {
-		res = platform_get_resource(pdev, IORESOURCE_MEM, 1);
-		priv->lms_base = devm_ioremap_resource(&pdev->dev, res);
+		priv->lms_base = devm_platform_ioremap_resource(pdev, 1);
 		if (IS_ERR(priv->lms_base))
 			return PTR_ERR(priv->lms_base);
 	} else {
@@ -5585,6 +5756,10 @@ static int mvpp2_probe(struct platform_device *pdev)
 			priv->sysctrl_base = NULL;
 	}
 
+	if (priv->hw_version == MVPP22 &&
+	    mvpp2_get_nrxqs(priv) * 2 <= MVPP2_BM_MAX_POOLS)
+		priv->percpu_pools = 1;
+
 	mvpp2_setup_bm_pool();
 
 
@@ -5766,7 +5941,7 @@ static int mvpp2_remove(struct platform_device *pdev)
 	for (i = 0; i < MVPP2_BM_POOLS_NUM; i++) {
 		struct mvpp2_bm_pool *bm_pool = &priv->bm_pools[i];
 
-		mvpp2_bm_pool_destroy(pdev, priv, bm_pool);
+		mvpp2_bm_pool_destroy(&pdev->dev, priv, bm_pool);
 	}
 
 	for (i = 0; i < MVPP2_MAX_THREADS; i++) {