1 files changed, 166 insertions, 115 deletions

diff --git a/drivers/staging/fsl-dpaa2/ethernet/dpaa2-eth.c b/drivers/staging/fsl-dpaa2/ethernet/dpaa2-eth.c
index 26017fe9df93..0d8ed002adcb 100644
--- a/drivers/staging/fsl-dpaa2/ethernet/dpaa2-eth.c
+++ b/drivers/staging/fsl-dpaa2/ethernet/dpaa2-eth.c
@@ -104,9 +104,11 @@ static void free_rx_fd(struct dpaa2_eth_priv *priv,
 		/* We don't support any other format */
 		return;
 
-	/* For S/G frames, we first need to free all SG entries */
+	/* For S/G frames, we first need to free all SG entries
+	 * except the first one, which was taken care of already
+	 */
 	sgt = vaddr + dpaa2_fd_get_offset(fd);
-	for (i = 0; i < DPAA2_ETH_MAX_SG_ENTRIES; i++) {
+	for (i = 1; i < DPAA2_ETH_MAX_SG_ENTRIES; i++) {
 		addr = dpaa2_sg_get_addr(&sgt[i]);
 		sg_vaddr = dpaa2_iova_to_virt(priv->iommu_domain, addr);
 		dma_unmap_single(dev, addr, DPAA2_ETH_RX_BUF_SIZE,
@@ -131,16 +133,15 @@ static struct sk_buff *build_linear_skb(struct dpaa2_eth_priv *priv,
 	u16 fd_offset = dpaa2_fd_get_offset(fd);
 	u32 fd_length = dpaa2_fd_get_len(fd);
 
-	skb = build_skb(fd_vaddr, DPAA2_ETH_RX_BUF_SIZE +
-			SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
+	ch->buf_count--;
+
+	skb = build_skb(fd_vaddr, DPAA2_ETH_SKB_SIZE);
 	if (unlikely(!skb))
 		return NULL;
 
 	skb_reserve(skb, fd_offset);
 	skb_put(skb, fd_length);
 
-	ch->buf_count--;
-
 	return skb;
 }
 
@@ -176,10 +177,21 @@ static struct sk_buff *build_frag_skb(struct dpaa2_eth_priv *priv,
 
 		if (i == 0) {
 			/* We build the skb around the first data buffer */
-			skb = build_skb(sg_vaddr, DPAA2_ETH_RX_BUF_SIZE +
-				SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
-			if (unlikely(!skb))
-				return NULL;
+			skb = build_skb(sg_vaddr, DPAA2_ETH_SKB_SIZE);
+			if (unlikely(!skb)) {
+				/* Free the first SG entry now, since we already
+				 * unmapped it and obtained the virtual address
+				 */
+				skb_free_frag(sg_vaddr);
+
+				/* We still need to subtract the buffers used
+				 * by this FD from our software counter
+				 */
+				while (!dpaa2_sg_is_final(&sgt[i]) &&
+				       i < DPAA2_ETH_MAX_SG_ENTRIES)
+					i++;
+				break;
+			}
 
 			sg_offset = dpaa2_sg_get_offset(sge);
 			skb_reserve(skb, sg_offset);
@@ -206,6 +218,8 @@ static struct sk_buff *build_frag_skb(struct dpaa2_eth_priv *priv,
 			break;
 	}
 
+	WARN_ONCE(i == DPAA2_ETH_MAX_SG_ENTRIES, "Final bit not set in SGT");
+
 	/* Count all data buffers + SG table buffer */
 	ch->buf_count -= i + 2;
 
@@ -557,10 +571,10 @@ static netdev_tx_t dpaa2_eth_tx(struct sk_buff *skb, struct net_device *net_dev)
 	percpu_stats = this_cpu_ptr(priv->percpu_stats);
 	percpu_extras = this_cpu_ptr(priv->percpu_extras);
 
-	if (unlikely(skb_headroom(skb) < DPAA2_ETH_NEEDED_HEADROOM(priv))) {
+	if (unlikely(skb_headroom(skb) < dpaa2_eth_needed_headroom(priv))) {
 		struct sk_buff *ns;
 
-		ns = skb_realloc_headroom(skb, DPAA2_ETH_NEEDED_HEADROOM(priv));
+		ns = skb_realloc_headroom(skb, dpaa2_eth_needed_headroom(priv));
 		if (unlikely(!ns)) {
 			percpu_stats->tx_dropped++;
 			goto err_alloc_headroom;
@@ -718,6 +732,23 @@ static int set_tx_csum(struct dpaa2_eth_priv *priv, bool enable)
 	return 0;
 }
 
+/* Free buffers acquired from the buffer pool or which were meant to
+ * be released in the pool
+ */
+static void free_bufs(struct dpaa2_eth_priv *priv, u64 *buf_array, int count)
+{
+	struct device *dev = priv->net_dev->dev.parent;
+	void *vaddr;
+	int i;
+
+	for (i = 0; i < count; i++) {
+		vaddr = dpaa2_iova_to_virt(priv->iommu_domain, buf_array[i]);
+		dma_unmap_single(dev, buf_array[i], DPAA2_ETH_RX_BUF_SIZE,
+				 DMA_BIDIRECTIONAL);
+		skb_free_frag(vaddr);
+	}
+}
+
 /* Perform a single release command to add buffers
  * to the specified buffer pool
  */
@@ -727,17 +758,17 @@ static int add_bufs(struct dpaa2_eth_priv *priv, u16 bpid)
 	u64 buf_array[DPAA2_ETH_BUFS_PER_CMD];
 	void *buf;
 	dma_addr_t addr;
-	int i;
+	int i, err;
 
 	for (i = 0; i < DPAA2_ETH_BUFS_PER_CMD; i++) {
 		/* Allocate buffer visible to WRIOP + skb shared info +
 		 * alignment padding
 		 */
-		buf = napi_alloc_frag(DPAA2_ETH_BUF_RAW_SIZE);
+		buf = napi_alloc_frag(dpaa2_eth_buf_raw_size(priv));
 		if (unlikely(!buf))
 			goto err_alloc;
 
-		buf = PTR_ALIGN(buf, DPAA2_ETH_RX_BUF_ALIGN);
+		buf = PTR_ALIGN(buf, priv->rx_buf_align);
 
 		addr = dma_map_single(dev, buf, DPAA2_ETH_RX_BUF_SIZE,
 				      DMA_FROM_DEVICE);
@@ -748,28 +779,33 @@ static int add_bufs(struct dpaa2_eth_priv *priv, u16 bpid)
 
 		/* tracing point */
 		trace_dpaa2_eth_buf_seed(priv->net_dev,
-					 buf, DPAA2_ETH_BUF_RAW_SIZE,
+					 buf, dpaa2_eth_buf_raw_size(priv),
 					 addr, DPAA2_ETH_RX_BUF_SIZE,
 					 bpid);
 	}
 
 release_bufs:
-	/* In case the portal is busy, retry until successful.
-	 * The buffer release function would only fail if the QBMan portal
-	 * was busy, which implies portal contention (i.e. more CPUs than
-	 * portals, i.e. GPPs w/o affine DPIOs). For all practical purposes,
-	 * there is little we can realistically do, short of giving up -
-	 * in which case we'd risk depleting the buffer pool and never again
-	 * receiving the Rx interrupt which would kick-start the refill logic.
-	 * So just keep retrying, at the risk of being moved to ksoftirqd.
-	 */
-	while (dpaa2_io_service_release(NULL, bpid, buf_array, i))
+	/* In case the portal is busy, retry until successful */
+	while ((err = dpaa2_io_service_release(NULL, bpid,
+					       buf_array, i)) == -EBUSY)
 		cpu_relax();
+
+	/* If release command failed, clean up and bail out;
+	 * not much else we can do about it
+	 */
+	if (err) {
+		free_bufs(priv, buf_array, i);
+		return 0;
+	}
+
 	return i;
 
 err_map:
 	skb_free_frag(buf);
 err_alloc:
+	/* If we managed to allocate at least some buffers,
+	 * release them to hardware
+	 */
 	if (i)
 		goto release_bufs;
 
@@ -811,10 +847,8 @@ static int seed_pool(struct dpaa2_eth_priv *priv, u16 bpid)
  */
 static void drain_bufs(struct dpaa2_eth_priv *priv, int count)
 {
-	struct device *dev = priv->net_dev->dev.parent;
 	u64 buf_array[DPAA2_ETH_BUFS_PER_CMD];
-	void *vaddr;
-	int ret, i;
+	int ret;
 
 	do {
 		ret = dpaa2_io_service_acquire(NULL, priv->bpid,
@@ -823,15 +857,7 @@ static void drain_bufs(struct dpaa2_eth_priv *priv, int count)
 			netdev_err(priv->net_dev, "dpaa2_io_service_acquire() failed\n");
 			return;
 		}
-		for (i = 0; i < ret; i++) {
-			/* Same logic as on regular Rx path */
-			vaddr = dpaa2_iova_to_virt(priv->iommu_domain,
-						   buf_array[i]);
-			dma_unmap_single(dev, buf_array[i],
-					 DPAA2_ETH_RX_BUF_SIZE,
-					 DMA_FROM_DEVICE);
-			skb_free_frag(vaddr);
-		}
+		free_bufs(priv, buf_array, ret);
 	} while (ret);
 }
 
@@ -927,13 +953,14 @@ static int dpaa2_eth_poll(struct napi_struct *napi, int budget)
 			break;
 	}
 
-	if (cleaned < budget) {
-		napi_complete_done(napi, cleaned);
+	if (cleaned < budget && napi_complete_done(napi, cleaned)) {
 		/* Re-enable data available notifications */
 		do {
 			err = dpaa2_io_service_rearm(NULL, &ch->nctx);
 			cpu_relax();
 		} while (err == -EBUSY);
+		WARN_ONCE(err, "CDAN notifications rearm failed on core %d",
+			  ch->nctx.desired_cpu);
 	}
 
 	ch->stats.frames += cleaned;
@@ -1415,34 +1442,32 @@ static struct fsl_mc_device *setup_dpcon(struct dpaa2_eth_priv *priv)
 	err = dpcon_open(priv->mc_io, 0, dpcon->obj_desc.id, &dpcon->mc_handle);
 	if (err) {
 		dev_err(dev, "dpcon_open() failed\n");
-		goto err_open;
+		goto free;
 	}
 
 	err = dpcon_reset(priv->mc_io, 0, dpcon->mc_handle);
 	if (err) {
 		dev_err(dev, "dpcon_reset() failed\n");
-		goto err_reset;
+		goto close;
 	}
 
 	err = dpcon_get_attributes(priv->mc_io, 0, dpcon->mc_handle, &attrs);
 	if (err) {
 		dev_err(dev, "dpcon_get_attributes() failed\n");
-		goto err_get_attr;
+		goto close;
 	}
 
 	err = dpcon_enable(priv->mc_io, 0, dpcon->mc_handle);
 	if (err) {
 		dev_err(dev, "dpcon_enable() failed\n");
-		goto err_enable;
+		goto close;
 	}
 
 	return dpcon;
 
-err_enable:
-err_get_attr:
-err_reset:
+close:
 	dpcon_close(priv->mc_io, 0, dpcon->mc_handle);
-err_open:
+free:
 	fsl_mc_object_free(dpcon);
 
 	return NULL;
@@ -1749,66 +1774,32 @@ static void free_dpbp(struct dpaa2_eth_priv *priv)
 	fsl_mc_object_free(priv->dpbp_dev);
 }
 
-/* Configure the DPNI object this interface is associated with */
-static int setup_dpni(struct fsl_mc_device *ls_dev)
+static int set_buffer_layout(struct dpaa2_eth_priv *priv)
 {
-	struct device *dev = &ls_dev->dev;
-	struct dpaa2_eth_priv *priv;
-	struct net_device *net_dev;
+	struct device *dev = priv->net_dev->dev.parent;
 	struct dpni_buffer_layout buf_layout = {0};
 	int err;
 
-	net_dev = dev_get_drvdata(dev);
-	priv = netdev_priv(net_dev);
-
-	/* get a handle for the DPNI object */
-	err = dpni_open(priv->mc_io, 0, ls_dev->obj_desc.id, &priv->mc_token);
-	if (err) {
-		dev_err(dev, "dpni_open() failed\n");
-		goto err_open;
-	}
-
-	ls_dev->mc_io = priv->mc_io;
-	ls_dev->mc_handle = priv->mc_token;
-
-	err = dpni_reset(priv->mc_io, 0, priv->mc_token);
-	if (err) {
-		dev_err(dev, "dpni_reset() failed\n");
-		goto err_reset;
-	}
-
-	err = dpni_get_attributes(priv->mc_io, 0, priv->mc_token,
-				  &priv->dpni_attrs);
-	if (err) {
-		dev_err(dev, "dpni_get_attributes() failed (err=%d)\n", err);
-		goto err_get_attr;
-	}
+	/* We need to check for WRIOP version 1.0.0, but depending on the MC
+	 * version, this number is not always provided correctly on rev1.
+	 * We need to check for both alternatives in this situation.
+	 */
+	if (priv->dpni_attrs.wriop_version == DPAA2_WRIOP_VERSION(0, 0, 0) ||
+	    priv->dpni_attrs.wriop_version == DPAA2_WRIOP_VERSION(1, 0, 0))
+		priv->rx_buf_align = DPAA2_ETH_RX_BUF_ALIGN_REV1;
+	else
+		priv->rx_buf_align = DPAA2_ETH_RX_BUF_ALIGN;
 
-	/* Configure buffer layouts */
-	/* rx buffer */
-	buf_layout.pass_parser_result = true;
+	/* tx buffer */
 	buf_layout.pass_frame_status = true;
 	buf_layout.private_data_size = DPAA2_ETH_SWA_SIZE;
-	buf_layout.data_align = DPAA2_ETH_RX_BUF_ALIGN;
-	buf_layout.options = DPNI_BUF_LAYOUT_OPT_PARSER_RESULT |
-			     DPNI_BUF_LAYOUT_OPT_FRAME_STATUS |
-			     DPNI_BUF_LAYOUT_OPT_PRIVATE_DATA_SIZE |
-			     DPNI_BUF_LAYOUT_OPT_DATA_ALIGN;
-	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
-				     DPNI_QUEUE_RX, &buf_layout);
-	if (err) {
-		dev_err(dev, "dpni_set_buffer_layout(RX) failed\n");
-		goto err_buf_layout;
-	}
-
-	/* tx buffer */
 	buf_layout.options = DPNI_BUF_LAYOUT_OPT_FRAME_STATUS |
 			     DPNI_BUF_LAYOUT_OPT_PRIVATE_DATA_SIZE;
 	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
 				     DPNI_QUEUE_TX, &buf_layout);
 	if (err) {
 		dev_err(dev, "dpni_set_buffer_layout(TX) failed\n");
-		goto err_buf_layout;
+		return err;
 	}
 
 	/* tx-confirm buffer */
@@ -1817,7 +1808,7 @@ static int setup_dpni(struct fsl_mc_device *ls_dev)
 				     DPNI_QUEUE_TX_CONFIRM, &buf_layout);
 	if (err) {
 		dev_err(dev, "dpni_set_buffer_layout(TX_CONF) failed\n");
-		goto err_buf_layout;
+		return err;
 	}
 
 	/* Now that we've set our tx buffer layout, retrieve the minimum
@@ -1827,24 +1818,76 @@ static int setup_dpni(struct fsl_mc_device *ls_dev)
 				      &priv->tx_data_offset);
 	if (err) {
 		dev_err(dev, "dpni_get_tx_data_offset() failed\n");
-		goto err_data_offset;
+		return err;
 	}
 
 	if ((priv->tx_data_offset % 64) != 0)
 		dev_warn(dev, "Tx data offset (%d) not a multiple of 64B\n",
 			 priv->tx_data_offset);
 
-	/* Accommodate software annotation space (SWA) */
-	priv->tx_data_offset += DPAA2_ETH_SWA_SIZE;
+	/* rx buffer */
+	buf_layout.pass_parser_result = true;
+	buf_layout.data_align = priv->rx_buf_align;
+	buf_layout.data_head_room = dpaa2_eth_rx_head_room(priv);
+	buf_layout.private_data_size = 0;
+	buf_layout.options = DPNI_BUF_LAYOUT_OPT_PARSER_RESULT |
+			     DPNI_BUF_LAYOUT_OPT_FRAME_STATUS |
+			     DPNI_BUF_LAYOUT_OPT_DATA_ALIGN |
+			     DPNI_BUF_LAYOUT_OPT_DATA_HEAD_ROOM;
+	err = dpni_set_buffer_layout(priv->mc_io, 0, priv->mc_token,
+				     DPNI_QUEUE_RX, &buf_layout);
+	if (err) {
+		dev_err(dev, "dpni_set_buffer_layout(RX) failed\n");
+		return err;
+	}
 
 	return 0;
+}
 
-err_data_offset:
-err_buf_layout:
-err_get_attr:
-err_reset:
+/* Configure the DPNI object this interface is associated with */
+static int setup_dpni(struct fsl_mc_device *ls_dev)
+{
+	struct device *dev = &ls_dev->dev;
+	struct dpaa2_eth_priv *priv;
+	struct net_device *net_dev;
+	int err;
+
+	net_dev = dev_get_drvdata(dev);
+	priv = netdev_priv(net_dev);
+
+	/* get a handle for the DPNI object */
+	err = dpni_open(priv->mc_io, 0, ls_dev->obj_desc.id, &priv->mc_token);
+	if (err) {
+		dev_err(dev, "dpni_open() failed\n");
+		return err;
+	}
+
+	ls_dev->mc_io = priv->mc_io;
+	ls_dev->mc_handle = priv->mc_token;
+
+	err = dpni_reset(priv->mc_io, 0, priv->mc_token);
+	if (err) {
+		dev_err(dev, "dpni_reset() failed\n");
+		goto close;
+	}
+
+	err = dpni_get_attributes(priv->mc_io, 0, priv->mc_token,
+				  &priv->dpni_attrs);
+	if (err) {
+		dev_err(dev, "dpni_get_attributes() failed (err=%d)\n", err);
+		goto close;
+	}
+
+	err = set_buffer_layout(priv);
+	if (err)
+		goto close;
+
+
+	return 0;
+
+close:
 	dpni_close(priv->mc_io, 0, priv->mc_token);
-err_open:
+
 	return err;
 }
 
@@ -2085,7 +2128,7 @@ static int bind_dpni(struct dpaa2_eth_priv *priv)
 	 */
 	err = dpaa2_eth_set_hash(net_dev, DPAA2_RXH_SUPPORTED);
 	if (err)
-		netdev_err(net_dev, "Failed to configure hashing\n");
+		dev_err(dev, "Failed to configure hashing\n");
 
 	/* Configure handling of error frames */
 	err_cfg.errors = DPAA2_FAS_RX_ERR_MASK;
@@ -2230,6 +2273,7 @@ static int netdev_init(struct net_device *net_dev)
 {
 	struct device *dev = net_dev->dev.parent;
 	struct dpaa2_eth_priv *priv = netdev_priv(net_dev);
+	u16 rx_headroom, req_headroom;
 	u8 bcast_addr[ETH_ALEN];
 	u8 num_queues;
 	int err;
@@ -2251,7 +2295,20 @@ static int netdev_init(struct net_device *net_dev)
 	/* Reserve enough space to align buffer as per hardware requirement;
 	 * NOTE: priv->tx_data_offset MUST be initialized at this point.
 	 */
-	net_dev->needed_headroom = DPAA2_ETH_NEEDED_HEADROOM(priv);
+	net_dev->needed_headroom = dpaa2_eth_needed_headroom(priv);
+
+	/* If headroom guaranteed by hardware in the Rx frame buffer is
+	 * smaller than the Tx headroom required by the stack, issue a
+	 * one time warning. This will most likely mean skbs forwarded to
+	 * another DPAA2 network interface will get reallocated, with a
+	 * significant performance impact.
+	 */
+	req_headroom = LL_RESERVED_SPACE(net_dev) - ETH_HLEN;
+	rx_headroom = ALIGN(DPAA2_ETH_RX_HWA_SIZE +
+			    dpaa2_eth_rx_head_room(priv), priv->rx_buf_align);
+	if (req_headroom > rx_headroom)
+		dev_info_once(dev, "Required headroom (%d) greater than available (%d)\n",
+			      req_headroom, rx_headroom);
 
 	/* Set MTU limits */
 	net_dev->min_mtu = 68;
@@ -2303,7 +2360,7 @@ static irqreturn_t dpni_irq0_handler(int irq_num, void *arg)
 
 static irqreturn_t dpni_irq0_handler_thread(int irq_num, void *arg)
 {
-	u32 status = 0, clear = 0;
+	u32 status = ~0;
 	struct device *dev = (struct device *)arg;
 	struct fsl_mc_device *dpni_dev = to_fsl_mc_device(dev);
 	struct net_device *net_dev = dev_get_drvdata(dev);
@@ -2313,18 +2370,12 @@ static irqreturn_t dpni_irq0_handler_thread(int irq_num, void *arg)
 				  DPNI_IRQ_INDEX, &status);
 	if (unlikely(err)) {
 		netdev_err(net_dev, "Can't get irq status (err %d)\n", err);
-		clear = 0xffffffff;
-		goto out;
+		return IRQ_HANDLED;
 	}
 
-	if (status & DPNI_IRQ_EVENT_LINK_CHANGED) {
-		clear |= DPNI_IRQ_EVENT_LINK_CHANGED;
+	if (status & DPNI_IRQ_EVENT_LINK_CHANGED)
 		link_state_update(netdev_priv(net_dev));
-	}
 
-out:
-	dpni_clear_irq_status(dpni_dev->mc_io, 0, dpni_dev->mc_handle,
-			      DPNI_IRQ_INDEX, clear);
 	return IRQ_HANDLED;
 }