[net-next v4 0/8] dpaa_eth: Add the Freescale DPAA Ethernet driver

[net-next v4 0/8] dpaa_eth: Add the Freescale DPAA Ethernet driver

[Madalin Bucur]

This patch series adds the Ethernet driver for the Freescale
QorIQ Data Path Acceleration Architecture (DPAA).

This version includes changes following the feedback received
on previous versions from Eric Dumazet, Bob Cochran, Joe Perches,
Paul Bolle, Joakim Tjernlund, Scott Wood, David Miller - thank you.

Together with the driver a managed version of alloc_percpu
is provided that simplifies the release of per-CPU memory.

The Freescale DPAA architecture consists in a series of hardware
blocks that support the Ethernet connectivity. The Ethernet driver
depends upon the following drivers that are currently in the Linux
kernel or in review (the underlying drivers are not inter-dependent):
 - Peripheral Access Memory Unit (PAMU)
    drivers/iommu/fsl_*
 - Frame Manager (FMan)
    drivers/net/ethernet/freescale/fman
 - Queue Manager (QMan), Buffer Manager (BMan)
    drivers/soc/fsl/qbman

dpaa_eth interfaces mapping to FMan MACs:

  dpaa_eth       /eth0\     ...       /ethN\
  driver        |      |             |      |
  -------------   ----   -----------   ----   -------------
       -Ports  / Tx  Rx \    ...    / Tx  Rx \
  FMan        |          |         |          |
       -MACs  |   MAC0   |         |   MACN   |
             /   dtsec0   \  ...  /   dtsecN   \ (or tgec)
            /              \     /              \(or memac)
  ---------  --------------  ---  --------------  ---------
      FMan, FMan Port, FMan SP, FMan MURAM drivers
  ---------------------------------------------------------
      FMan HW blocks: MURAM, MACs, Ports, SP
  ---------------------------------------------------------

dpaa_eth relation to QMan, FMan:
              ________________________________
  dpaa_eth   /            eth0                \
  driver    /                                  \
  ---------   -^-   -^-   -^-   ---    ---------
  QMan driver / \   / \   / \  \   /  | BMan    |
             |Rx | |Rx | |Tx | |Tx |  | driver  |
  ---------  |Dfl| |Err| |Cnf| |FQs|  |         |
  QMan HW    |FQ | |FQ | |FQ | |   |  |         |
             /   \ /   \ /   \  \ /   |         |
  ---------   ---   ---   ---   -v-    ---------
            |        FMan QMI         |         |
            | FMan HW       FMan BMI  | BMan HW |
              -----------------------   --------

where the acronyms used above (and in the code) are:
DPAA = Data Path Acceleration Architecture
FMan = DPAA Frame Manager
QMan = DPAA Queue Manager
BMan = DPAA Buffers Manager
QMI = QMan interface in FMan
BMI = BMan interface in FMan
FMan SP = FMan Storage Profiles
MURAM = Multi-user RAM in FMan
FQ = QMan Frame Queue
Rx Dfl FQ = default reception FQ
Rx Err FQ = Rx error frames FQ
Tx Cnf FQ = Tx confirmation FQ
Tx FQs = transmission frame queues
dtsec = datapath three speed Ethernet controller (10/100/1000 Mbps)
tgec = ten gigabit Ethernet controller (10 Gbps)
memac = multirate Ethernet MAC (10/100/1000/10000)

The latest FMan driver patches were submitted by Igal Liberman:
https://patchwork.ozlabs.org/project/netdev/list/?submitter=64715&state=*&q=[v7,

The latest Q/BMan drivers were submitted by Roy Pledge:
https://patchwork.ozlabs.org/project/linuxppc-dev/list/?submitter=66331&state=*

Changes from v3:
 - removed bogus delay and comment in .ndo_stop implementation
 - addressed minor issues reported by David Miller

Changes from v2:
 - removed debugfs, moved exports to ethtool statistics
 - removed congestion groups Kconfig params

Changes from v1:
 - bpool level Kconfig options removed
 - print format using pr_fmt, cleaned up prints
 - __hot/__cold removed
 - gratuitous unlikely() removed
 - code style aligned, consistent spacing for declarations
 - comment formatting

The complete patch set based on the latest net-next/master kernel
can be found in the public git at:
http://git.freescale.com/git/cgit.cgi/ppc/upstream/linux.git
under the tag ldup_public_git_20151102:
http://git.freescale.com/git/cgit.cgi/ppc/upstream/linux.git/log/?h=ldup_public_git_20151102

There is one u-boot patch that one needs to make sure it's applied
to align u-boot to the latest device tree binding document specification
used by the FMan driver. Please make sure your u-boot includes this patch:

	commit 97a8d010e029111e5711a45264a726bedbeb24c4
	Author: Igal Liberman <igal.liberman@freescale.com>
	Date:   Tue Aug 18 14:47:05 2015 +0300

	    net/fman: Support both new and legacy FMan Compatibles

The patch was included in u-boot in v2015.10-rc3.

Madalin Bucur (8):
  devres: add devm_alloc_percpu()
  dpaa_eth: add support for DPAA Ethernet
  dpaa_eth: add support for S/G frames
  dpaa_eth: add driver's Tx queue selection
  dpaa_eth: add ethtool functionality
  dpaa_eth: add ethtool statistics
  dpaa_eth: add sysfs exports
  dpaa_eth: add trace points

 Documentation/driver-model/devres.txt              |    4 +
 drivers/base/devres.c                              |   64 +
 drivers/net/ethernet/freescale/Kconfig             |    2 +
 drivers/net/ethernet/freescale/Makefile            |    1 +
 drivers/net/ethernet/freescale/dpaa/Kconfig        |   32 +
 drivers/net/ethernet/freescale/dpaa/Makefile       |   12 +
 drivers/net/ethernet/freescale/dpaa/dpaa_eth.c     |  854 ++++++++++++
 drivers/net/ethernet/freescale/dpaa/dpaa_eth.h     |  479 +++++++
 .../net/ethernet/freescale/dpaa/dpaa_eth_common.c  | 1395 ++++++++++++++++++++
 .../net/ethernet/freescale/dpaa/dpaa_eth_common.h  |  109 ++
 drivers/net/ethernet/freescale/dpaa/dpaa_eth_sg.c  |  706 ++++++++++
 .../net/ethernet/freescale/dpaa/dpaa_eth_sysfs.c   |  167 +++
 .../net/ethernet/freescale/dpaa/dpaa_eth_trace.h   |  141 ++
 drivers/net/ethernet/freescale/dpaa/dpaa_ethtool.c |  413 ++++++
 include/linux/device.h                             |   19 +
 15 files changed, 4398 insertions(+)
 create mode 100644 drivers/net/ethernet/freescale/dpaa/Kconfig
 create mode 100644 drivers/net/ethernet/freescale/dpaa/Makefile
 create mode 100644 drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
 create mode 100644 drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
 create mode 100644 drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c
 create mode 100644 drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h
 create mode 100644 drivers/net/ethernet/freescale/dpaa/dpaa_eth_sg.c
 create mode 100644 drivers/net/ethernet/freescale/dpaa/dpaa_eth_sysfs.c
 create mode 100644 drivers/net/ethernet/freescale/dpaa/dpaa_eth_trace.h
 create mode 100644 drivers/net/ethernet/freescale/dpaa/dpaa_ethtool.c

-- 
1.7.11.7

_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/linuxppc-dev

[net-next v4 1/8] devres: add devm_alloc_percpu()

[Madalin Bucur]

Introduce managed counterparts for alloc_percpu() and free_percpu().
Add devm_alloc_percpu() and devm_free_percpu() into the managed
interfaces list.

Signed-off-by: Madalin Bucur <madalin.bucur@freescale.com>
Tested-by: Madalin-Cristian Bucur <madalin.bucur@freescale.com>
---
 Documentation/driver-model/devres.txt |  4 +++
 drivers/base/devres.c                 | 64 +++++++++++++++++++++++++++++++++++
 include/linux/device.h                | 19 +++++++++++
 3 files changed, 87 insertions(+)

diff --git a/Documentation/driver-model/devres.txt b/Documentation/driver-model/devres.txt
index 831a536..595fd1b 100644
--- a/Documentation/driver-model/devres.txt
+++ b/Documentation/driver-model/devres.txt
@@ -312,6 +312,10 @@ MEM
   devm_kvasprintf()
   devm_kzalloc()
 
+PER-CPU MEM
+  devm_alloc_percpu()
+  devm_free_percpu()
+
 PCI
   pcim_enable_device()	: after success, all PCI ops become managed
   pcim_pin_device()	: keep PCI device enabled after release
diff --git a/drivers/base/devres.c b/drivers/base/devres.c
index 8754646..6c314cc 100644
--- a/drivers/base/devres.c
+++ b/drivers/base/devres.c
@@ -10,6 +10,7 @@
 #include <linux/device.h>
 #include <linux/module.h>
 #include <linux/slab.h>
+#include <linux/percpu.h>
 
 #include "base.h"
 
@@ -984,3 +985,66 @@ void devm_free_pages(struct device *dev, unsigned long addr)
 			       &devres));
 }
 EXPORT_SYMBOL_GPL(devm_free_pages);
+
+static void devm_percpu_release(struct device *dev, void *pdata)
+{
+	void __percpu *p;
+
+	p = *(void __percpu **)pdata;
+	free_percpu(p);
+}
+
+static int devm_percpu_match(struct device *dev, void *data, void *p)
+{
+	struct devres *devr = container_of(data, struct devres, data);
+
+	return *(void **)devr->data == p;
+}
+
+/**
+ * __devm_alloc_percpu - Resource-managed alloc_percpu
+ * @dev: Device to allocate per-cpu memory for
+ * @size: Size of per-cpu memory to allocate
+ * @align: Alignement of per-cpu memory to allocate
+ *
+ * Managed alloc_percpu. Per-cpu memory allocated with this function is
+ * automatically freed on driver detach.
+ *
+ * RETURNS:
+ * Pointer to allocated memory on success, NULL on failure.
+ */
+void __percpu *__devm_alloc_percpu(struct device *dev, size_t size,
+		size_t align)
+{
+	void *p;
+	void __percpu *pcpu;
+
+	pcpu = __alloc_percpu(size, align);
+	if (!pcpu)
+		return NULL;
+
+	p = devres_alloc(devm_percpu_release, sizeof(void *), GFP_KERNEL);
+	if (!p)
+		return NULL;
+
+	*(void __percpu **)p = pcpu;
+
+	devres_add(dev, p);
+
+	return pcpu;
+}
+EXPORT_SYMBOL_GPL(__devm_alloc_percpu);
+
+/**
+ * devm_free_percpu - Resource-managed free_percpu
+ * @dev: Device this memory belongs to
+ * @pdata: Per-cpu memory to free
+ *
+ * Free memory allocated with devm_alloc_percpu().
+ */
+void devm_free_percpu(struct device *dev, void __percpu *pdata)
+{
+	WARN_ON(devres_destroy(dev, devm_percpu_release, devm_percpu_match,
+			       (void *)pdata));
+}
+EXPORT_SYMBOL_GPL(devm_free_percpu);
diff --git a/include/linux/device.h b/include/linux/device.h
index 5d7bc63..b563cc5 100644
--- a/include/linux/device.h
+++ b/include/linux/device.h
@@ -673,6 +673,25 @@ void __iomem *devm_ioremap_resource(struct device *dev, struct resource *res);
 int devm_add_action(struct device *dev, void (*action)(void *), void *data);
 void devm_remove_action(struct device *dev, void (*action)(void *), void *data);
 
+/**
+ * devm_alloc_percpu - Resource-managed alloc_percpu
+ * @dev: Device to allocate per-cpu memory for
+ * @type: Type to allocate per-cpu memory for
+ *
+ * Managed alloc_percpu. Per-cpu memory allocated with this function is
+ * automatically freed on driver detach.
+ *
+ * RETURNS:
+ * Pointer to allocated memory on success, NULL on failure.
+ */
+#define devm_alloc_percpu(dev, type)      \
+	(typeof(type) __percpu *)__devm_alloc_percpu(dev, sizeof(type), \
+						     __alignof__(type))
+
+void __percpu *__devm_alloc_percpu(struct device *dev, size_t size,
+				   size_t align);
+void devm_free_percpu(struct device *dev, void __percpu *pdata);
+
 struct device_dma_parameters {
 	/*
 	 * a low level driver may set these to teach IOMMU code about
-- 
1.7.11.7

Re: [net-next v4 1/8] devres: add devm_alloc_percpu()

[Joe Perches]

On Mon, 2015-11-02 at 19:31 +0200, Madalin Bucur wrote:
> Introduce managed counterparts for alloc_percpu() and free_percpu().
> Add devm_alloc_percpu() and devm_free_percpu() into the managed
> interfaces list.

trivia, could be fixed later

> +/**
> + * __devm_alloc_percpu - Resource-managed alloc_percpu
> + * @dev: Device to allocate per-cpu memory for
> + * @size: Size of per-cpu memory to allocate
> + * @align: Alignement of per-cpu memory to allocate

French spelling?  alignment

[net-next v4 2/8] dpaa_eth: add support for DPAA Ethernet

[Madalin Bucur]

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@freescale.com>
---
 drivers/net/ethernet/freescale/Kconfig             |    2 +
 drivers/net/ethernet/freescale/Makefile            |    1 +
 drivers/net/ethernet/freescale/dpaa/Kconfig        |   22 +
 drivers/net/ethernet/freescale/dpaa/Makefile       |   11 +
 drivers/net/ethernet/freescale/dpaa/dpaa_eth.c     |  819 ++++++++++++
 drivers/net/ethernet/freescale/dpaa/dpaa_eth.h     |  432 +++++++
 .../net/ethernet/freescale/dpaa/dpaa_eth_common.c  | 1299 ++++++++++++++++++++
 .../net/ethernet/freescale/dpaa/dpaa_eth_common.h  |   98 ++
 drivers/net/ethernet/freescale/dpaa/dpaa_eth_sg.c  |  408 ++++++
 9 files changed, 3092 insertions(+)
 create mode 100644 drivers/net/ethernet/freescale/dpaa/Kconfig
 create mode 100644 drivers/net/ethernet/freescale/dpaa/Makefile
 create mode 100644 drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
 create mode 100644 drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
 create mode 100644 drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c
 create mode 100644 drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h
 create mode 100644 drivers/net/ethernet/freescale/dpaa/dpaa_eth_sg.c

diff --git a/drivers/net/ethernet/freescale/Kconfig b/drivers/net/ethernet/freescale/Kconfig
index f3f89cc..92198be 100644
--- a/drivers/net/ethernet/freescale/Kconfig
+++ b/drivers/net/ethernet/freescale/Kconfig
@@ -92,4 +92,6 @@ 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 4097c58..ae13dc5 100644
--- a/drivers/net/ethernet/freescale/Makefile
+++ b/drivers/net/ethernet/freescale/Makefile
@@ -12,6 +12,7 @@ obj-$(CONFIG_FS_ENET) += fs_enet/
 obj-$(CONFIG_FSL_PQ_MDIO) += fsl_pq_mdio.o
 obj-$(CONFIG_FSL_XGMAC_MDIO) += xgmac_mdio.o
 obj-$(CONFIG_GIANFAR) += gianfar_driver.o
+obj-$(CONFIG_FSL_DPAA_ETH) += dpaa/
 obj-$(CONFIG_PTP_1588_CLOCK_GIANFAR) += gianfar_ptp.o
 gianfar_driver-objs := gianfar.o \
 		gianfar_ethtool.o
diff --git a/drivers/net/ethernet/freescale/dpaa/Kconfig b/drivers/net/ethernet/freescale/dpaa/Kconfig
new file mode 100644
index 0000000..022d5aa
--- /dev/null
+++ b/drivers/net/ethernet/freescale/dpaa/Kconfig
@@ -0,0 +1,22 @@
+menuconfig FSL_DPAA_ETH
+	tristate "DPAA Ethernet"
+	depends on FSL_SOC && FSL_BMAN && FSL_QMAN && 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.
+
+if FSL_DPAA_ETH
+
+config FSL_DPAA_ETH_FRIENDLY_IF_NAME
+	bool "Use fmX-macY names for the DPAA interfaces"
+	default y
+	---help---
+	  The DPAA Ethernet netdevices are created for each FMan port available
+	  on a certain board. Enable this to get interface names derived from
+	  the underlying FMan hardware for a simple identification.
+
+endif # FSL_DPAA_ETH
diff --git a/drivers/net/ethernet/freescale/dpaa/Makefile b/drivers/net/ethernet/freescale/dpaa/Makefile
new file mode 100644
index 0000000..3847ec7
--- /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 dpaa_eth_sg.o dpaa_eth_common.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 0000000..8381616
--- /dev/null
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
@@ -0,0 +1,819 @@
+/* Copyright 2008 - 2015 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_mdio.h>
+#include <linux/of_net.h>
+#include <linux/kthread.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/if_ether.h>
+#include <linux/highmem.h>
+#include <linux/percpu.h>
+#include <linux/dma-mapping.h>
+#include <soc/fsl/bman.h>
+
+#include "fman.h"
+#include "fman_port.h"
+
+#include "mac.h"
+#include "dpaa_eth.h"
+#include "dpaa_eth_common.h"
+
+/* Valid checksum indication */
+#define DPA_CSUM_VALID		0xFFFF
+
+#define DPA_DESCRIPTION "FSL DPAA Ethernet driver"
+
+#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.
+ */
+
+static u8 debug = -1;
+module_param(debug, byte, S_IRUGO);
+MODULE_PARM_DESC(debug, "Module/Driver verbosity level");
+
+/* This has to work in tandem with the DPA_CS_THRESHOLD_xxx values. */
+static u16 tx_timeout = 1000;
+module_param(tx_timeout, ushort, S_IRUGO);
+MODULE_PARM_DESC(tx_timeout, "The Tx timeout in ms");
+
+/* BM */
+
+#define DPAA_ETH_MAX_PAD (L1_CACHE_BYTES * 8)
+
+static u8 dpa_priv_common_bpid;
+
+static void _dpa_rx_error(struct net_device *net_dev,
+			  const struct dpa_priv_s *priv,
+			  struct dpa_percpu_priv_s *percpu_priv,
+			  const struct qm_fd *fd,
+			  u32 fqid)
+{
+	/* limit common, possibly innocuous Rx FIFO Overflow errors'
+	 * interference with zero-loss convergence benchmark results.
+	 */
+	if (likely(fd->status & FM_FD_ERR_PHYSICAL))
+		pr_warn_once("non-zero error counters in fman statistics (sysfs)\n");
+	else
+		if (net_ratelimit())
+			netif_err(priv, hw, net_dev, "Err FD status = 0x%08x\n",
+				  fd->status & FM_FD_STAT_RX_ERRORS);
+
+	percpu_priv->stats.rx_errors++;
+
+	dpa_fd_release(net_dev, fd);
+}
+
+static void _dpa_tx_error(struct net_device *net_dev,
+			  const struct dpa_priv_s *priv,
+			  struct dpa_percpu_priv_s *percpu_priv,
+			  const struct qm_fd *fd,
+			  u32 fqid)
+{
+	struct sk_buff *skb;
+
+	if (net_ratelimit())
+		netif_warn(priv, hw, net_dev, "FD status = 0x%08x\n",
+			   fd->status & FM_FD_STAT_TX_ERRORS);
+
+	percpu_priv->stats.tx_errors++;
+
+	/* If we intended the buffers from this frame to go into the bpools
+	 * when the FMan transmit was done, we need to put it in manually.
+	 */
+	if (fd->bpid != 0xff) {
+		dpa_fd_release(net_dev, fd);
+		return;
+	}
+
+	skb = _dpa_cleanup_tx_fd(priv, fd);
+	dev_kfree_skb(skb);
+}
+
+static int dpaa_eth_poll(struct napi_struct *napi, int budget)
+{
+	struct dpa_napi_portal *np =
+			container_of(napi, struct dpa_napi_portal, napi);
+
+	int cleaned = qman_p_poll_dqrr(np->p, budget);
+
+	if (cleaned < budget) {
+		int tmp;
+
+		napi_complete(napi);
+		tmp = qman_p_irqsource_add(np->p, QM_PIRQ_DQRI);
+		DPA_ERR_ON(tmp);
+	} else if (np->down) {
+		qman_p_irqsource_add(np->p, QM_PIRQ_DQRI);
+	}
+
+	return cleaned;
+}
+
+static void _dpa_tx_conf(struct net_device *net_dev,
+			 const struct dpa_priv_s *priv,
+			 struct dpa_percpu_priv_s *percpu_priv,
+			 const struct qm_fd *fd,
+			 u32 fqid)
+{
+	struct sk_buff	*skb;
+
+	if (unlikely(fd->status & FM_FD_STAT_TX_ERRORS) != 0) {
+		if (net_ratelimit())
+			netif_warn(priv, hw, net_dev, "FD status = 0x%08x\n",
+				   fd->status & FM_FD_STAT_TX_ERRORS);
+
+		percpu_priv->stats.tx_errors++;
+	}
+
+	skb = _dpa_cleanup_tx_fd(priv, fd);
+
+	dev_kfree_skb(skb);
+}
+
+static enum qman_cb_dqrr_result
+priv_rx_error_dqrr(struct qman_portal *portal,
+		   struct qman_fq *fq,
+		   const struct qm_dqrr_entry *dq)
+{
+	struct net_device *net_dev;
+	struct dpa_priv_s *priv;
+	struct dpa_percpu_priv_s *percpu_priv;
+	int *count_ptr;
+
+	net_dev = ((struct dpa_fq *)fq)->net_dev;
+	priv = netdev_priv(net_dev);
+
+	percpu_priv = raw_cpu_ptr(priv->percpu_priv);
+	count_ptr = raw_cpu_ptr(priv->dpa_bp->percpu_count);
+
+	if (dpaa_eth_napi_schedule(percpu_priv, portal))
+		return qman_cb_dqrr_stop;
+
+	if (unlikely(dpaa_eth_refill_bpools(priv->dpa_bp, count_ptr)))
+		/* Unable to refill the buffer pool due to insufficient
+		 * system memory. Just release the frame back into the pool,
+		 * otherwise we'll soon end up with an empty buffer pool.
+		 */
+		dpa_fd_release(net_dev, &dq->fd);
+	else
+		_dpa_rx_error(net_dev, priv, percpu_priv, &dq->fd, fq->fqid);
+
+	return qman_cb_dqrr_consume;
+}
+
+static enum qman_cb_dqrr_result
+priv_rx_default_dqrr(struct qman_portal *portal,
+		     struct qman_fq *fq,
+		     const struct qm_dqrr_entry *dq)
+{
+	struct net_device *net_dev;
+	struct dpa_priv_s *priv;
+	struct dpa_percpu_priv_s *percpu_priv;
+	int *count_ptr;
+	struct dpa_bp *dpa_bp;
+
+	net_dev = ((struct dpa_fq *)fq)->net_dev;
+	priv = netdev_priv(net_dev);
+	dpa_bp = priv->dpa_bp;
+
+	/* IRQ handler, non-migratable; safe to use raw_cpu_ptr here */
+	percpu_priv = raw_cpu_ptr(priv->percpu_priv);
+	count_ptr = raw_cpu_ptr(dpa_bp->percpu_count);
+
+	if (unlikely(dpaa_eth_napi_schedule(percpu_priv, portal)))
+		return qman_cb_dqrr_stop;
+
+	/* Vale of plenty: make sure we didn't run out of buffers */
+
+	if (unlikely(dpaa_eth_refill_bpools(dpa_bp, count_ptr)))
+		/* Unable to refill the buffer pool due to insufficient
+		 * system memory. Just release the frame back into the pool,
+		 * otherwise we'll soon end up with an empty buffer pool.
+		 */
+		dpa_fd_release(net_dev, &dq->fd);
+	else
+		_dpa_rx(net_dev, portal, priv, percpu_priv, &dq->fd, fq->fqid,
+			count_ptr);
+
+	return qman_cb_dqrr_consume;
+}
+
+static enum qman_cb_dqrr_result
+priv_tx_conf_error_dqrr(struct qman_portal *portal,
+			struct qman_fq *fq,
+			const struct qm_dqrr_entry *dq)
+{
+	struct net_device *net_dev;
+	struct dpa_priv_s *priv;
+	struct dpa_percpu_priv_s *percpu_priv;
+
+	net_dev = ((struct dpa_fq *)fq)->net_dev;
+	priv = netdev_priv(net_dev);
+
+	percpu_priv = raw_cpu_ptr(priv->percpu_priv);
+
+	if (dpaa_eth_napi_schedule(percpu_priv, portal))
+		return qman_cb_dqrr_stop;
+
+	_dpa_tx_error(net_dev, priv, percpu_priv, &dq->fd, fq->fqid);
+
+	return qman_cb_dqrr_consume;
+}
+
+static enum qman_cb_dqrr_result
+priv_tx_conf_default_dqrr(struct qman_portal *portal,
+			  struct qman_fq *fq,
+			  const struct qm_dqrr_entry *dq)
+{
+	struct net_device *net_dev;
+	struct dpa_priv_s *priv;
+	struct dpa_percpu_priv_s *percpu_priv;
+
+	net_dev = ((struct dpa_fq *)fq)->net_dev;
+	priv = netdev_priv(net_dev);
+
+	/* Non-migratable context, safe to use raw_cpu_ptr */
+	percpu_priv = raw_cpu_ptr(priv->percpu_priv);
+
+	if (dpaa_eth_napi_schedule(percpu_priv, portal))
+		return qman_cb_dqrr_stop;
+
+	_dpa_tx_conf(net_dev, priv, percpu_priv, &dq->fd, fq->fqid);
+
+	return qman_cb_dqrr_consume;
+}
+
+static void priv_ern(struct qman_portal *portal,
+		     struct qman_fq *fq,
+		     const struct qm_mr_entry *msg)
+{
+	struct net_device *net_dev;
+	const struct dpa_priv_s *priv;
+	struct sk_buff *skb;
+	struct dpa_percpu_priv_s *percpu_priv;
+	const struct qm_fd *fd = &msg->ern.fd;
+
+	net_dev = ((struct dpa_fq *)fq)->net_dev;
+	priv = netdev_priv(net_dev);
+	/* Non-migratable context, safe to use raw_cpu_ptr */
+	percpu_priv = raw_cpu_ptr(priv->percpu_priv);
+
+	percpu_priv->stats.tx_dropped++;
+	percpu_priv->stats.tx_fifo_errors++;
+
+	/* If we intended this buffer to go into the pool
+	 * when the FM was done, we need to put it in
+	 * manually.
+	 */
+	if (msg->ern.fd.bpid != 0xff) {
+		dpa_fd_release(net_dev, fd);
+		return;
+	}
+
+	skb = _dpa_cleanup_tx_fd(priv, fd);
+	dev_kfree_skb_any(skb);
+}
+
+static const struct dpa_fq_cbs_t private_fq_cbs = {
+	.rx_defq = { .cb = { .dqrr = priv_rx_default_dqrr } },
+	.tx_defq = { .cb = { .dqrr = priv_tx_conf_default_dqrr } },
+	.rx_errq = { .cb = { .dqrr = priv_rx_error_dqrr } },
+	.tx_errq = { .cb = { .dqrr = priv_tx_conf_error_dqrr } },
+	.egress_ern = { .cb = { .ern = priv_ern } }
+};
+
+static void dpaa_eth_napi_enable(struct dpa_priv_s *priv)
+{
+	struct dpa_percpu_priv_s *percpu_priv;
+	int i, j;
+
+	for_each_possible_cpu(i) {
+		percpu_priv = per_cpu_ptr(priv->percpu_priv, i);
+
+		for (j = 0; j < qman_portal_max; j++) {
+			percpu_priv->np[j].down = 0;
+			napi_enable(&percpu_priv->np[j].napi);
+		}
+	}
+}
+
+static void dpaa_eth_napi_disable(struct dpa_priv_s *priv)
+{
+	struct dpa_percpu_priv_s *percpu_priv;
+	int i, j;
+
+	for_each_possible_cpu(i) {
+		percpu_priv = per_cpu_ptr(priv->percpu_priv, i);
+
+		for (j = 0; j < qman_portal_max; j++) {
+			percpu_priv->np[j].down = 1;
+			napi_disable(&percpu_priv->np[j].napi);
+		}
+	}
+}
+
+static int dpa_eth_priv_start(struct net_device *net_dev)
+{
+	struct dpa_priv_s *priv;
+	int err;
+
+	priv = netdev_priv(net_dev);
+	dpaa_eth_napi_enable(priv);
+
+	err = dpa_start(net_dev);
+	if (err < 0)
+		dpaa_eth_napi_disable(priv);
+
+	return err;
+}
+
+static int dpa_eth_priv_stop(struct net_device *net_dev)
+{
+	struct dpa_priv_s *priv;
+	int err;
+
+	err = dpa_stop(net_dev);
+
+	priv = netdev_priv(net_dev);
+	dpaa_eth_napi_disable(priv);
+
+	return err;
+}
+
+static const struct net_device_ops dpa_private_ops = {
+	.ndo_open = dpa_eth_priv_start,
+	.ndo_start_xmit = dpa_tx,
+	.ndo_stop = dpa_eth_priv_stop,
+	.ndo_tx_timeout = dpa_timeout,
+	.ndo_get_stats64 = dpa_get_stats64,
+	.ndo_set_mac_address = dpa_set_mac_address,
+	.ndo_validate_addr = eth_validate_addr,
+	.ndo_change_mtu = dpa_change_mtu,
+	.ndo_set_rx_mode = dpa_set_rx_mode,
+	.ndo_init = dpa_ndo_init,
+	.ndo_set_features = dpa_set_features,
+	.ndo_fix_features = dpa_fix_features,
+};
+
+static int dpa_private_napi_add(struct net_device *net_dev)
+{
+	struct dpa_priv_s *priv = netdev_priv(net_dev);
+	struct dpa_percpu_priv_s *percpu_priv;
+	int i, cpu;
+
+	for_each_possible_cpu(cpu) {
+		percpu_priv = per_cpu_ptr(priv->percpu_priv, cpu);
+
+		percpu_priv->np = devm_kzalloc(net_dev->dev.parent,
+			qman_portal_max * sizeof(struct dpa_napi_portal),
+			GFP_KERNEL);
+
+		if (!percpu_priv->np)
+			return -ENOMEM;
+
+		for (i = 0; i < qman_portal_max; i++)
+			netif_napi_add(net_dev, &percpu_priv->np[i].napi,
+				       dpaa_eth_poll, NAPI_POLL_WEIGHT);
+	}
+
+	return 0;
+}
+
+void dpa_private_napi_del(struct net_device *net_dev)
+{
+	struct dpa_priv_s *priv = netdev_priv(net_dev);
+	struct dpa_percpu_priv_s *percpu_priv;
+	int i, cpu;
+
+	for_each_possible_cpu(cpu) {
+		percpu_priv = per_cpu_ptr(priv->percpu_priv, cpu);
+
+		if (percpu_priv->np) {
+			for (i = 0; i < qman_portal_max; i++)
+				netif_napi_del(&percpu_priv->np[i].napi);
+
+			devm_kfree(net_dev->dev.parent, percpu_priv->np);
+		}
+	}
+}
+
+static int dpa_private_netdev_init(struct net_device *net_dev)
+{
+	int i;
+	struct dpa_priv_s *priv = netdev_priv(net_dev);
+	struct dpa_percpu_priv_s *percpu_priv;
+	const u8 *mac_addr;
+
+	/* 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 = &dpa_private_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->hw_features |= (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
+				 NETIF_F_LLTX);
+
+	net_dev->features |= NETIF_F_GSO;
+
+	return dpa_netdev_init(net_dev, mac_addr, tx_timeout);
+}
+
+static struct dpa_bp *dpa_priv_bp_probe(struct device *dev)
+{
+	struct dpa_bp *dpa_bp;
+
+	dpa_bp = devm_kzalloc(dev, sizeof(*dpa_bp), GFP_KERNEL);
+	if (!dpa_bp)
+		return ERR_PTR(-ENOMEM);
+
+	dpa_bp->percpu_count = devm_alloc_percpu(dev, *dpa_bp->percpu_count);
+	dpa_bp->config_count = FSL_DPAA_ETH_MAX_BUF_COUNT;
+
+	dpa_bp->seed_cb = dpa_bp_priv_seed;
+	dpa_bp->free_buf_cb = _dpa_bp_free_pf;
+
+	return dpa_bp;
+}
+
+/* Place all ingress FQs (Rx Default, Rx Error) in a dedicated CGR.
+ * We won't be sending congestion notifications to FMan; for now, we just use
+ * this CGR to generate enqueue rejections to FMan in order to drop the frames
+ * before they reach our ingress queues and eat up memory.
+ */
+static int dpaa_eth_priv_ingress_cgr_init(struct dpa_priv_s *priv)
+{
+	struct qm_mcc_initcgr initcgr;
+	u32 cs_th;
+	int err;
+
+	err = qman_alloc_cgrid(&priv->ingress_cgr.cgrid);
+	if (err < 0) {
+		pr_err("Error %d allocating CGR ID\n", err);
+		goto out_error;
+	}
+
+	/* Enable CS TD, but disable Congestion State Change Notifications. */
+	initcgr.we_mask = QM_CGR_WE_CS_THRES;
+	initcgr.cgr.cscn_en = QM_CGR_EN;
+	cs_th = DPAA_INGRESS_CS_THRESHOLD;
+	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;
+
+	/* This is actually a hack, because this CGR will be associated with
+	 * our affine SWP. However, we'll place our ingress FQs in it.
+	 */
+	err = qman_create_cgr(&priv->ingress_cgr, QMAN_CGR_FLAG_USE_INIT,
+			      &initcgr);
+	if (err < 0) {
+		pr_err("Error %d creating ingress CGR with ID %d\n", err,
+		       priv->ingress_cgr.cgrid);
+		qman_release_cgrid(priv->ingress_cgr.cgrid);
+		goto out_error;
+	}
+	pr_debug("Created ingress CGR %d for netdev with hwaddr %pM\n",
+		 priv->ingress_cgr.cgrid, priv->mac_dev->addr);
+
+	/* struct qman_cgr allows special cgrid values (i.e. outside the 0..255
+	 * range), but we have no common initialization path between the
+	 * different variants of the DPAA Eth driver, so we do it here rather
+	 * than modifying every other variant than "private Eth".
+	 */
+	priv->use_ingress_cgr = true;
+
+out_error:
+	return err;
+}
+
+static int dpa_priv_bp_create(struct net_device *net_dev, struct dpa_bp *dpa_bp,
+			      size_t count)
+{
+	struct dpa_priv_s *priv = netdev_priv(net_dev);
+	int i;
+
+	netif_dbg(priv, probe, net_dev,
+		  "Using private BM buffer pools\n");
+
+	priv->bp_count = count;
+
+	for (i = 0; i < count; i++) {
+		int err;
+
+		err = dpa_bp_alloc(&dpa_bp[i]);
+		if (err < 0) {
+			dpa_bp_free(priv);
+			priv->dpa_bp = NULL;
+			return err;
+		}
+
+		priv->dpa_bp = &dpa_bp[i];
+	}
+
+	dpa_priv_common_bpid = priv->dpa_bp->bpid;
+	return 0;
+}
+
+static const struct of_device_id dpa_match[];
+
+static int
+dpaa_eth_priv_probe(struct platform_device *pdev)
+{
+	int err = 0, i, channel;
+	struct device *dev;
+	struct dpa_bp *dpa_bp;
+	struct dpa_fq *dpa_fq, *tmp;
+	size_t count = 1;
+	struct net_device *net_dev = NULL;
+	struct dpa_priv_s *priv = NULL;
+	struct dpa_percpu_priv_s *percpu_priv;
+	struct fm_port_fqs port_fqs;
+	struct dpa_buffer_layout_s *buf_layout = NULL;
+	struct mac_device *mac_dev;
+	struct task_struct *kth;
+
+	dev = &pdev->dev;
+
+	/* Get the buffer pool assigned to this interface;
+	 * run only once the default pool probing code
+	 */
+	dpa_bp = (dpa_bpid2pool(dpa_priv_common_bpid)) ? :
+			dpa_priv_bp_probe(dev);
+	if (IS_ERR(dpa_bp))
+		return PTR_ERR(dpa_bp);
+
+	/* Allocate this early, so we can store relevant information in
+	 * the private area
+	 */
+	net_dev = alloc_etherdev_mq(sizeof(*priv), DPAA_ETH_TX_QUEUES);
+	if (!net_dev) {
+		dev_err(dev, "alloc_etherdev_mq() failed\n");
+		goto alloc_etherdev_mq_failed;
+	}
+
+#ifdef CONFIG_FSL_DPAA_ETH_FRIENDLY_IF_NAME
+	snprintf(net_dev->name, IFNAMSIZ, "fm%d-mac%d",
+		 dpa_mac_fman_index_get(pdev),
+		 dpa_mac_hw_index_get(pdev));
+#endif
+
+	/* Do this here, so we can be verbose early */
+	SET_NETDEV_DEV(net_dev, dev);
+	dev_set_drvdata(dev, net_dev);
+
+	priv = netdev_priv(net_dev);
+	priv->net_dev = net_dev;
+
+	priv->msg_enable = netif_msg_init(debug, -1);
+
+	mac_dev = dpa_mac_dev_get(pdev);
+	if (IS_ERR(mac_dev) || !mac_dev) {
+		err = PTR_ERR(mac_dev);
+		goto mac_probe_failed;
+	}
+
+	/* We have physical ports, so we need to establish
+	 * the buffer layout.
+	 */
+	buf_layout = devm_kzalloc(dev, 2 * sizeof(*buf_layout),
+				  GFP_KERNEL);
+	if (!buf_layout)
+		goto alloc_failed;
+
+	dpa_set_buffers_layout(mac_dev, buf_layout);
+
+	/* For private ports, need to compute the size of the default
+	 * buffer pool, based on FMan port buffer layout;also update
+	 * the maximum buffer size for private ports if necessary
+	 */
+	dpa_bp->size = dpa_bp_size(&buf_layout[RX]);
+
+	INIT_LIST_HEAD(&priv->dpa_fq_list);
+
+	memset(&port_fqs, 0, sizeof(port_fqs));
+
+	err = dpa_fq_probe_mac(dev, &priv->dpa_fq_list, &port_fqs, true, RX);
+	if (!err)
+		err = dpa_fq_probe_mac(dev, &priv->dpa_fq_list,
+				       &port_fqs, true, TX);
+
+	if (err < 0)
+		goto fq_probe_failed;
+
+	/* bp init */
+
+	err = dpa_priv_bp_create(net_dev, dpa_bp, count);
+
+	if (err < 0)
+		goto bp_create_failed;
+
+	priv->mac_dev = mac_dev;
+
+	channel = dpa_get_channel();
+
+	if (channel < 0) {
+		err = channel;
+		goto get_channel_failed;
+	}
+
+	priv->channel = (u16)channel;
+
+	/* Start a thread that will walk the cpus with affine portals
+	 * and add this pool channel to each's dequeue mask.
+	 */
+	kth = kthread_run(dpaa_eth_add_channel,
+			  (void *)(unsigned long)priv->channel,
+			  "dpaa_%p:%d", net_dev, priv->channel);
+	if (!kth) {
+		err = -ENOMEM;
+		goto add_channel_failed;
+	}
+
+	dpa_fq_setup(priv, &private_fq_cbs, priv->mac_dev->port[TX]);
+
+	/* Create a congestion group for this netdev, with
+	 * dynamically-allocated CGR ID.
+	 * Must be executed after probing the MAC, but before
+	 * assigning the egress FQs to the CGRs.
+	 */
+	err = dpaa_eth_cgr_init(priv);
+	if (err < 0) {
+		dev_err(dev, "Error initializing CGR\n");
+		goto tx_cgr_init_failed;
+	}
+	err = dpaa_eth_priv_ingress_cgr_init(priv);
+	if (err < 0) {
+		dev_err(dev, "Error initializing ingress CGR\n");
+		goto rx_cgr_init_failed;
+	}
+
+	/* Add the FQs to the interface, and make them active */
+	list_for_each_entry_safe(dpa_fq, tmp, &priv->dpa_fq_list, list) {
+		err = dpa_fq_init(dpa_fq, false);
+		if (err < 0)
+			goto fq_alloc_failed;
+	}
+
+	priv->buf_layout = buf_layout;
+	priv->tx_headroom = dpa_get_headroom(&priv->buf_layout[TX]);
+	priv->rx_headroom = dpa_get_headroom(&priv->buf_layout[RX]);
+
+	/* All real interfaces need their ports initialized */
+	dpaa_eth_init_ports(mac_dev, dpa_bp, count, &port_fqs,
+			    buf_layout, dev);
+
+	priv->percpu_priv = devm_alloc_percpu(dev, *priv->percpu_priv);
+
+	if (!priv->percpu_priv) {
+		dev_err(dev, "devm_alloc_percpu() failed\n");
+		err = -ENOMEM;
+		goto alloc_percpu_failed;
+	}
+	for_each_possible_cpu(i) {
+		percpu_priv = per_cpu_ptr(priv->percpu_priv, i);
+		memset(percpu_priv, 0, sizeof(*percpu_priv));
+	}
+
+	/* Initialize NAPI */
+	err = dpa_private_napi_add(net_dev);
+
+	if (err < 0)
+		goto napi_add_failed;
+
+	err = dpa_private_netdev_init(net_dev);
+
+	if (err < 0)
+		goto netdev_init_failed;
+
+	pr_info("Probed interface %s\n", net_dev->name);
+
+	return 0;
+
+netdev_init_failed:
+napi_add_failed:
+	dpa_private_napi_del(net_dev);
+alloc_percpu_failed:
+	dpa_fq_free(dev, &priv->dpa_fq_list);
+fq_alloc_failed:
+	qman_delete_cgr_safe(&priv->ingress_cgr);
+	qman_release_cgrid(priv->ingress_cgr.cgrid);
+rx_cgr_init_failed:
+	qman_delete_cgr_safe(&priv->cgr_data.cgr);
+	qman_release_cgrid(priv->cgr_data.cgr.cgrid);
+tx_cgr_init_failed:
+add_channel_failed:
+get_channel_failed:
+	dpa_bp_free(priv);
+bp_create_failed:
+fq_probe_failed:
+alloc_failed:
+mac_probe_failed:
+	dev_set_drvdata(dev, NULL);
+	free_netdev(net_dev);
+alloc_etherdev_mq_failed:
+	if (atomic_read(&dpa_bp->refs) == 0)
+		devm_kfree(dev, dpa_bp);
+
+	return err;
+}
+
+static struct platform_device_id dpa_devtype[] = {
+	{
+		.name = "dpaa-ethernet",
+		.driver_data = 0,
+	}, {
+	}
+};
+MODULE_DEVICE_TABLE(platform, dpa_devtype);
+
+static struct platform_driver dpa_driver = {
+	.driver = {
+		.name = KBUILD_MODNAME,
+	},
+	.id_table = dpa_devtype,
+	.probe = dpaa_eth_priv_probe,
+	.remove = dpa_remove
+};
+
+static int __init dpa_load(void)
+{
+	int err;
+
+	pr_info(DPA_DESCRIPTION "\n");
+
+	/* initialise dpaa_eth mirror values */
+	dpa_rx_extra_headroom = fman_get_rx_extra_headroom();
+	dpa_max_frm = fman_get_max_frm();
+
+	err = platform_driver_register(&dpa_driver);
+	if (err < 0)
+		pr_err("Error, platform_driver_register() = %d\n", err);
+
+	return err;
+}
+module_init(dpa_load);
+
+static void __exit dpa_unload(void)
+{
+	platform_driver_unregister(&dpa_driver);
+
+	/* Only one channel is used and needs to be relased after all
+	 * interfaces are removed
+	 */
+	dpa_release_channel();
+}
+module_exit(dpa_unload);
+
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_AUTHOR("Andy Fleming <afleming@freescale.com>");
+MODULE_DESCRIPTION(DPA_DESCRIPTION);
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
new file mode 100644
index 0000000..1cc8682
--- /dev/null
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
@@ -0,0 +1,432 @@
+/* Copyright 2008 - 2015 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.
+ */
+
+#ifndef __DPA_H
+#define __DPA_H
+
+#include <linux/netdevice.h>
+#include <soc/fsl/qman.h>
+
+#include "fman.h"
+#include "mac.h"
+
+extern int dpa_rx_extra_headroom;
+extern int dpa_max_frm;
+
+#define dpa_get_rx_extra_headroom() dpa_rx_extra_headroom
+#define dpa_get_max_frm() dpa_max_frm
+
+#define dpa_get_max_mtu()	\
+	(dpa_get_max_frm() - (VLAN_ETH_HLEN + ETH_FCS_LEN))
+
+/* Simple enum of FQ types - used for array indexing */
+enum port_type {RX, TX};
+
+struct dpa_buffer_layout_s {
+	u16 priv_data_size;
+	bool parse_results;
+	bool time_stamp;
+	bool hash_results;
+	u16 data_align;
+};
+
+#define DPA_ERR_ON(cond)
+
+#define DPA_TX_PRIV_DATA_SIZE	16
+#define DPA_PARSE_RESULTS_SIZE sizeof(struct fman_prs_result)
+#define DPA_TIME_STAMP_SIZE 8
+#define DPA_HASH_RESULTS_SIZE 8
+#define DPA_RX_PRIV_DATA_SIZE	(DPA_TX_PRIV_DATA_SIZE + \
+					dpa_get_rx_extra_headroom())
+
+#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)
+
+/* The raw buffer size must be cacheline aligned.
+ * Normally we use 2K buffers.
+ */
+#define DPA_BP_RAW_SIZE		2048
+
+/* This is what FMan is ever allowed to use.
+ * 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 (can we?), so we reserve some more space
+ * for start-of-buffer alignment.
+ */
+#define dpa_bp_size(buffer_layout)	(SKB_WITH_OVERHEAD(DPA_BP_RAW_SIZE) - \
+						SMP_CACHE_BYTES)
+/* We must ensure that skb_shinfo is always cacheline-aligned. */
+#define DPA_SKB_SIZE(size)	((size) & ~(SMP_CACHE_BYTES - 1))
+
+/* Largest value that the FQD's OAL field can hold.
+ * This is DPAA-1.x specific.
+ */
+#define FSL_QMAN_MAX_OAL	127
+
+/* Default alignment for start of data in an Rx FD */
+#define DPA_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
+ * See $8.8.4.7.20 - L4 HXS - L4 Results from DPAA-Rev2 Reference Manual.
+ */
+/* L4 Type field: UDP */
+#define FM_L4_PARSE_RESULT_UDP	0x40
+/* L4 Type field: TCP */
+#define FM_L4_PARSE_RESULT_TCP	0x20
+
+/* number of Tx queues to FMan */
+#define DPAA_ETH_TX_QUEUES	NR_CPUS
+
+#define DPAA_ETH_RX_QUEUES	128
+
+#define FSL_DPAA_ETH_MAX_BUF_COUNT	128
+#define FSL_DPAA_ETH_REFILL_THRESHOLD	80
+
+/* More detailed FQ types - used for fine-grained WQ assignments */
+enum dpa_fq_type {
+	FQ_TYPE_RX_DEFAULT = 1, /* Rx Default FQs */
+	FQ_TYPE_RX_ERROR,	/* Rx Error FQs */
+	FQ_TYPE_RX_PCD,		/* User-defined PCDs */
+	FQ_TYPE_TX,		/* "Real" Tx FQs */
+	FQ_TYPE_TX_CONFIRM,	/* Tx default Conf FQ (actually an Rx FQ) */
+	FQ_TYPE_TX_CONF_MQ,	/* Tx conf FQs (one for each Tx FQ) */
+	FQ_TYPE_TX_ERROR,	/* Tx Error FQs (these are actually Rx FQs) */
+};
+
+struct dpa_fq {
+	struct qman_fq fq_base;
+	struct list_head list;
+	struct net_device *net_dev;
+	bool init;
+	u32 fqid;
+	u32 flags;
+	u16 channel;
+	u8 wq;
+	enum dpa_fq_type fq_type;
+};
+
+struct dpa_fq_cbs_t {
+	struct qman_fq rx_defq;
+	struct qman_fq tx_defq;
+	struct qman_fq rx_errq;
+	struct qman_fq tx_errq;
+	struct qman_fq egress_ern;
+};
+
+struct fqid_cell {
+	u32 start;
+	u32 count;
+};
+
+struct dpa_bp {
+	struct bman_pool *pool;
+	u8 bpid;
+	struct device *dev;
+	/* the buffer pools used for the private ports are initialized
+	 * with config_count buffers for each CPU; at runtime the
+	 * number of buffers per CPU is constantly brought back to this
+	 * level
+	 */
+	int config_count;
+	size_t size;
+	bool seed_pool;
+	/* physical address of the contiguous memory used by the pool to store
+	 * the buffers
+	 */
+	dma_addr_t paddr;
+	/* virtual address of the contiguous memory used by the pool to store
+	 * the buffers
+	 */
+	void __iomem *vaddr;
+	/* current number of buffers in the bpool alloted to this CPU */
+	int __percpu *percpu_count;
+	atomic_t refs;
+	/* some bpools need to be seeded before use by this cb */
+	int (*seed_cb)(struct dpa_bp *);
+	/* some bpools need to be emptied before freeing; this cb is used
+	 * for freeing of individual buffers taken from the pool
+	 */
+	void (*free_buf_cb)(void *addr);
+};
+
+struct dpa_napi_portal {
+	struct napi_struct napi;
+	struct qman_portal *p;
+	bool down;
+};
+
+struct dpa_percpu_priv_s {
+	struct net_device *net_dev;
+	struct dpa_napi_portal *np;
+	struct rtnl_link_stats64 stats;
+};
+
+struct dpa_priv_s {
+	struct dpa_percpu_priv_s __percpu *percpu_priv;
+	struct dpa_bp *dpa_bp;
+	/* Store here the needed Tx headroom for convenience and speed
+	 * (even though it can be computed based on the fields of buf_layout)
+	 */
+	u16 tx_headroom;
+	struct net_device *net_dev;
+	struct mac_device *mac_dev;
+	struct qman_fq *egress_fqs[DPAA_ETH_TX_QUEUES];
+	struct qman_fq *conf_fqs[DPAA_ETH_TX_QUEUES];
+
+	size_t bp_count;
+
+	u16 channel;	/* "fsl,qman-channel-id" */
+	struct list_head dpa_fq_list;
+
+	u32 msg_enable;	/* net_device message level */
+
+	struct {
+		/* All egress queues to a given net device belong to one
+		 * (and the same) congestion group.
+		 */
+		struct qman_cgr cgr;
+	} cgr_data;
+	/* Use a per-port CGR for ingress traffic. */
+	bool use_ingress_cgr;
+	struct qman_cgr ingress_cgr;
+
+	struct dpa_buffer_layout_s *buf_layout;
+	u16 rx_headroom;
+};
+
+struct fm_port_fqs {
+	struct dpa_fq *tx_defq;
+	struct dpa_fq *tx_errq;
+	struct dpa_fq *rx_defq;
+	struct dpa_fq *rx_errq;
+};
+
+int dpa_bp_priv_seed(struct dpa_bp *dpa_bp);
+int dpaa_eth_refill_bpools(struct dpa_bp *dpa_bp, int *count_ptr);
+void _dpa_rx(struct net_device *net_dev,
+	     struct qman_portal *portal,
+	     const struct dpa_priv_s *priv,
+	     struct dpa_percpu_priv_s *percpu_priv,
+	     const struct qm_fd *fd,
+	     u32 fqid,
+	     int *count_ptr);
+int dpa_tx(struct sk_buff *skb, struct net_device *net_dev);
+struct sk_buff *_dpa_cleanup_tx_fd(const struct dpa_priv_s *priv,
+				   const struct qm_fd *fd);
+
+/* Turn on HW checksum computation for this outgoing frame.
+ * If the current protocol is not something we support in this regard
+ * (or if the stack has already computed the SW checksum), we do nothing.
+ *
+ * Returns 0 if all goes well (or HW csum doesn't apply), and a negative value
+ * otherwise.
+ *
+ * Note that this function may modify the fd->cmd field and the skb data buffer
+ * (the Parse Results area).
+ */
+int dpa_enable_tx_csum(struct dpa_priv_s *priv, struct sk_buff *skb,
+		       struct qm_fd *fd, char *parse_results);
+
+static inline int dpaa_eth_napi_schedule(struct dpa_percpu_priv_s *percpu_priv,
+					 struct qman_portal *portal)
+{
+	/* In case of threaded ISR for RT enable kernel,
+	 * in_irq() does not return appropriate value, so use
+	 * in_serving_softirq to distinguish softirq or irq context.
+	 */
+	if (unlikely(in_irq() || !in_serving_softirq())) {
+		/* Disable QMan IRQ and invoke NAPI */
+		int ret = qman_p_irqsource_remove(portal, QM_PIRQ_DQRI);
+
+		if (likely(!ret)) {
+			const struct qman_portal_config *pc =
+					qman_p_get_portal_config(portal);
+			struct dpa_napi_portal *np =
+					&percpu_priv->np[pc->channel];
+
+			np->p = portal;
+			napi_schedule(&np->napi);
+			return 1;
+		}
+	}
+	return 0;
+}
+
+static inline ssize_t __const dpa_fd_length(const struct qm_fd *fd)
+{
+	return fd->length20;
+}
+
+static inline ssize_t __const dpa_fd_offset(const struct qm_fd *fd)
+{
+	return fd->offset;
+}
+
+/* Verifies if the skb length is below the interface MTU */
+static inline int dpa_check_rx_mtu(struct sk_buff *skb, int mtu)
+{
+	if (unlikely(skb->len > mtu))
+		if ((skb->protocol != htons(ETH_P_8021Q)) ||
+		    (skb->len > mtu + 4))
+			return -1;
+
+	return 0;
+}
+
+static inline u16 dpa_get_headroom(struct dpa_buffer_layout_s *bl)
+{
+	u16 headroom;
+	/* The frame headroom must accommodate:
+	 * - the driver private data area
+	 * - parse results, hash results, timestamp if selected
+	 * If either hash results or time stamp are selected, both will
+	 * be copied to/from the frame headroom, as TS is located between PR and
+	 * HR in the IC and IC copy size has a granularity of 16bytes
+	 * (see description of FMBM_RICP and FMBM_TICP registers in DPAARM)
+	 *
+	 * Also make sure the headroom is a multiple of data_align bytes
+	 */
+	headroom = (u16)(bl->priv_data_size +
+		   (bl->parse_results ? DPA_PARSE_RESULTS_SIZE : 0) +
+		   (bl->hash_results || bl->time_stamp ?
+		    DPA_TIME_STAMP_SIZE + DPA_HASH_RESULTS_SIZE : 0));
+
+	return bl->data_align ? ALIGN(headroom, bl->data_align) : headroom;
+}
+
+void dpa_private_napi_del(struct net_device *net_dev);
+
+static inline void clear_fd(struct qm_fd *fd)
+{
+	fd->opaque_addr = 0;
+	fd->opaque = 0;
+	fd->cmd = 0;
+}
+
+static inline int _dpa_tx_fq_to_id(const struct dpa_priv_s *priv,
+				   struct qman_fq *tx_fq)
+{
+	int i;
+
+	for (i = 0; i < DPAA_ETH_TX_QUEUES; i++)
+		if (priv->egress_fqs[i] == tx_fq)
+			return i;
+
+	return -EINVAL;
+}
+
+static inline int dpa_xmit(struct dpa_priv_s *priv,
+			   struct rtnl_link_stats64 *percpu_stats,
+			   int queue,
+			   struct qm_fd *fd)
+{
+	int err, i;
+	struct qman_fq *egress_fq;
+
+	egress_fq = priv->egress_fqs[queue];
+	if (fd->bpid == 0xff)
+		fd->cmd |= qman_fq_fqid(priv->conf_fqs[queue]);
+
+	for (i = 0; i < 100000; i++) {
+		err = qman_enqueue(egress_fq, fd, 0);
+		if (err != -EBUSY)
+			break;
+	}
+
+	if (unlikely(err < 0)) {
+		percpu_stats->tx_errors++;
+		percpu_stats->tx_fifo_errors++;
+		return err;
+	}
+
+	percpu_stats->tx_packets++;
+	percpu_stats->tx_bytes += dpa_fd_length(fd);
+
+	return 0;
+}
+
+/* Use multiple WQs for FQ assignment:
+ *	- Tx Confirmation queues go to WQ1.
+ *	- Rx Default and Tx queues go to WQ3 (no differentiation between
+ *	  Rx and Tx traffic).
+ *	- 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).
+ * 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 _dpa_assign_wq(struct dpa_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_DEFAULT:
+	case FQ_TYPE_TX:
+		fq->wq = 3;
+		break;
+	case FQ_TYPE_RX_ERROR:
+	case FQ_TYPE_TX_ERROR:
+		fq->wq = 2;
+		break;
+	default:
+		WARN(1, "Invalid FQ type %d for FQID %d!\n",
+		     fq->fq_type, fq->fqid);
+	}
+}
+
+/* Use the queue selected by XPS */
+#define dpa_get_queue_mapping(skb) \
+	skb_get_queue_mapping(skb)
+
+static inline void _dpa_bp_free_pf(void *addr)
+{
+	put_page(virt_to_head_page(addr));
+}
+
+#endif	/* __DPA_H */
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c
new file mode 100644
index 0000000..963be4d8
--- /dev/null
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c
@@ -0,0 +1,1299 @@
+/* Copyright 2008 - 2015 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.
+ */
+
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/of_platform.h>
+#include <linux/of_net.h>
+#include <linux/etherdevice.h>
+#include <linux/kthread.h>
+#include <linux/percpu.h>
+#include <linux/highmem.h>
+#include <linux/sort.h>
+#include <soc/fsl/qman.h>
+#include <linux/ip.h>
+#include <linux/ipv6.h>
+#include <linux/if_vlan.h>
+#include "dpaa_eth.h"
+#include "dpaa_eth_common.h"
+#include "mac.h"
+
+/* Size in bytes of the FQ taildrop threshold */
+#define DPA_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
+ */
+
+static struct dpa_bp *dpa_bp_array[64];
+
+int dpa_max_frm;
+
+int dpa_rx_extra_headroom;
+
+static const struct fqid_cell tx_confirm_fqids[] = {
+	{0, DPAA_ETH_TX_QUEUES}
+};
+
+static const struct fqid_cell default_fqids[][3] = {
+	[RX] = { {0, 1}, {0, 1}, {0, DPAA_ETH_RX_QUEUES} },
+	[TX] = { {0, 1}, {0, 1}, {0, DPAA_ETH_TX_QUEUES} }
+};
+
+int dpa_netdev_init(struct net_device *net_dev,
+		    const u8 *mac_addr,
+		    u16 tx_timeout)
+{
+	int err;
+	struct dpa_priv_s *priv = netdev_priv(net_dev);
+	struct device *dev = net_dev->dev.parent;
+
+	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;
+}
+
+int dpa_start(struct net_device *net_dev)
+{
+	int err, i;
+	struct dpa_priv_s *priv;
+	struct mac_device *mac_dev;
+
+	priv = netdev_priv(net_dev);
+	mac_dev = priv->mac_dev;
+
+	err = mac_dev->init_phy(net_dev, priv->mac_dev);
+	if (err < 0) {
+		netif_err(priv, ifup, net_dev, "init_phy() = %d\n", err);
+		return err;
+	}
+
+	for (i = 0; i < ARRAY_SIZE(mac_dev->port); i++) {
+		err = fman_port_enable(mac_dev->port[i]);
+		if (err)
+			goto mac_start_failed;
+	}
+
+	err = priv->mac_dev->start(mac_dev);
+	if (err < 0) {
+		netif_err(priv, ifup, net_dev, "mac_dev->start() = %d\n", err);
+		goto mac_start_failed;
+	}
+
+	netif_tx_start_all_queues(net_dev);
+
+	return 0;
+
+mac_start_failed:
+	for (i = 0; i < ARRAY_SIZE(mac_dev->port); i++)
+		fman_port_disable(mac_dev->port[i]);
+
+	return err;
+}
+
+int dpa_stop(struct net_device *net_dev)
+{
+	int i, err, error;
+	struct dpa_priv_s *priv;
+	struct mac_device *mac_dev;
+
+	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 (mac_dev->phy_dev)
+		phy_disconnect(mac_dev->phy_dev);
+	mac_dev->phy_dev = NULL;
+
+	return err;
+}
+
+void dpa_timeout(struct net_device *net_dev)
+{
+	const struct dpa_priv_s	*priv;
+	struct dpa_percpu_priv_s *percpu_priv;
+
+	priv = netdev_priv(net_dev);
+	percpu_priv = raw_cpu_ptr(priv->percpu_priv);
+
+	netif_crit(priv, timer, net_dev, "Transmit timeout latency: %u ms\n",
+		   jiffies_to_msecs(jiffies - net_dev->trans_start));
+
+	percpu_priv->stats.tx_errors++;
+}
+
+/* Calculates the statistics for the given device by adding the statistics
+ * collected by each CPU.
+ */
+struct rtnl_link_stats64 *dpa_get_stats64(struct net_device *net_dev,
+					  struct rtnl_link_stats64 *stats)
+{
+	struct dpa_priv_s *priv = netdev_priv(net_dev);
+	u64 *cpustats;
+	u64 *netstats = (u64 *)stats;
+	int i, j;
+	struct dpa_percpu_priv_s *percpu_priv;
+	int numstats = sizeof(struct rtnl_link_stats64) / sizeof(u64);
+
+	for_each_possible_cpu(i) {
+		percpu_priv = per_cpu_ptr(priv->percpu_priv, i);
+
+		cpustats = (u64 *)&percpu_priv->stats;
+
+		for (j = 0; j < numstats; j++)
+			netstats[j] += cpustats[j];
+	}
+
+	return stats;
+}
+
+int dpa_change_mtu(struct net_device *net_dev, int new_mtu)
+{
+	const int max_mtu = dpa_get_max_mtu();
+
+	/* Make sure we don't exceed the Ethernet controller's MAXFRM */
+	if (new_mtu < 68 || new_mtu > max_mtu) {
+		netdev_err(net_dev, "Invalid L3 mtu %d (must be between %d and %d).\n",
+			   new_mtu, 68, max_mtu);
+		return -EINVAL;
+	}
+	net_dev->mtu = new_mtu;
+
+	return 0;
+}
+
+/* .ndo_init callback */
+int dpa_ndo_init(struct net_device *net_dev)
+{
+	/* If fsl_fm_max_frm is set to a higher value than the all-common 1500,
+	 * we choose conservatively and let the user explicitly set a higher
+	 * MTU via ifconfig. Otherwise, the user may end up with different MTUs
+	 * in the same LAN.
+	 * If on the other hand fsl_fm_max_frm has been chosen below 1500,
+	 * start with the maximum allowed.
+	 */
+	int init_mtu = min(dpa_get_max_mtu(), ETH_DATA_LEN);
+
+	netdev_dbg(net_dev, "Setting initial MTU on net device: %d\n",
+		   init_mtu);
+	net_dev->mtu = init_mtu;
+
+	return 0;
+}
+
+int dpa_set_features(struct net_device *dev, netdev_features_t features)
+{
+	/* Not much to do here for now */
+	dev->features = features;
+	return 0;
+}
+
+netdev_features_t dpa_fix_features(struct net_device *dev,
+				   netdev_features_t features)
+{
+	netdev_features_t unsupported_features = 0;
+
+	/* In theory we should never be requested to enable features that
+	 * we didn't set in netdev->features and netdev->hw_features at probe
+	 * time, but double check just to be on the safe side.
+	 * We don't support enabling Rx csum through ethtool yet
+	 */
+	unsupported_features |= NETIF_F_RXCSUM;
+
+	features &= ~unsupported_features;
+
+	return features;
+}
+
+int dpa_remove(struct platform_device *pdev)
+{
+	int err;
+	struct device *dev;
+	struct net_device *net_dev;
+	struct dpa_priv_s *priv;
+
+	dev = &pdev->dev;
+	net_dev = dev_get_drvdata(dev);
+
+	priv = netdev_priv(net_dev);
+
+	dev_set_drvdata(dev, NULL);
+	unregister_netdev(net_dev);
+
+	err = dpa_fq_free(dev, &priv->dpa_fq_list);
+
+	qman_delete_cgr_safe(&priv->ingress_cgr);
+	qman_release_cgrid(priv->ingress_cgr.cgrid);
+	qman_delete_cgr_safe(&priv->cgr_data.cgr);
+	qman_release_cgrid(priv->cgr_data.cgr.cgrid);
+
+	dpa_private_napi_del(net_dev);
+
+	dpa_bp_free(priv);
+
+	if (priv->buf_layout)
+		devm_kfree(dev, priv->buf_layout);
+
+	free_netdev(net_dev);
+
+	return err;
+}
+
+struct mac_device *dpa_mac_dev_get(struct platform_device *pdev)
+{
+	struct device *dpa_dev, *dev;
+	struct device_node *mac_node;
+	struct platform_device *of_dev;
+	struct mac_device *mac_dev;
+	struct dpaa_eth_data *eth_data;
+
+	dpa_dev = &pdev->dev;
+	eth_data = dpa_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(dpa_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(dpa_dev, "dev_get_drvdata(%s) failed\n",
+			dev_name(dev));
+		return ERR_PTR(-EINVAL);
+	}
+
+	return mac_dev;
+}
+
+int dpa_mac_hw_index_get(struct platform_device *pdev)
+{
+	struct device *dpa_dev;
+	struct dpaa_eth_data *eth_data;
+
+	dpa_dev = &pdev->dev;
+	eth_data = dpa_dev->platform_data;
+
+	return eth_data->mac_hw_id;
+}
+
+int dpa_mac_fman_index_get(struct platform_device *pdev)
+{
+	struct device *dpa_dev;
+	struct dpaa_eth_data *eth_data;
+
+	dpa_dev = &pdev->dev;
+	eth_data = dpa_dev->platform_data;
+
+	return eth_data->fman_hw_id;
+}
+
+int dpa_set_mac_address(struct net_device *net_dev, void *addr)
+{
+	const struct dpa_priv_s	*priv;
+	int err;
+	struct mac_device *mac_dev;
+
+	priv = netdev_priv(net_dev);
+
+	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);
+		return err;
+	}
+
+	return 0;
+}
+
+void dpa_set_rx_mode(struct net_device *net_dev)
+{
+	int err;
+	const struct dpa_priv_s	*priv;
+
+	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);
+}
+
+void dpa_set_buffers_layout(struct mac_device *mac_dev,
+			    struct dpa_buffer_layout_s *layout)
+{
+	/* Rx */
+	layout[RX].priv_data_size = (u16)DPA_RX_PRIV_DATA_SIZE;
+	layout[RX].parse_results = true;
+	layout[RX].hash_results = true;
+	layout[RX].data_align = DPA_FD_DATA_ALIGNMENT;
+
+	/* Tx */
+	layout[TX].priv_data_size = DPA_TX_PRIV_DATA_SIZE;
+	layout[TX].parse_results = true;
+	layout[TX].hash_results = true;
+	layout[TX].data_align = DPA_FD_DATA_ALIGNMENT;
+}
+
+int dpa_bp_alloc(struct dpa_bp *dpa_bp)
+{
+	int err;
+	struct bman_pool_params bp_params;
+	struct platform_device *pdev;
+
+	if (dpa_bp->size == 0 || dpa_bp->config_count == 0) {
+		pr_err("Buffer pool is not properly initialized! Missing size or initial number of buffers");
+		return -EINVAL;
+	}
+
+	memset(&bp_params, 0, sizeof(struct bman_pool_params));
+
+	/* If the pool is already specified, we only create one per bpid */
+	if (dpa_bpid2pool_use(dpa_bp->bpid))
+		return 0;
+
+	if (dpa_bp->bpid == 0)
+		bp_params.flags |= BMAN_POOL_FLAG_DYNAMIC_BPID;
+	else
+		bp_params.bpid = dpa_bp->bpid;
+
+	dpa_bp->pool = bman_new_pool(&bp_params);
+	if (!dpa_bp->pool) {
+		pr_err("bman_new_pool() failed\n");
+		return -ENODEV;
+	}
+
+	dpa_bp->bpid = (u8)bman_get_params(dpa_bp->pool)->bpid;
+
+	pdev = platform_device_register_simple("DPAA_bpool",
+					       dpa_bp->bpid, NULL, 0);
+	if (IS_ERR(pdev)) {
+		err = PTR_ERR(pdev);
+		goto pdev_register_failed;
+	}
+
+	err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(40));
+	if (err)
+		goto pdev_mask_failed;
+
+	dpa_bp->dev = &pdev->dev;
+
+	if (dpa_bp->seed_cb) {
+		err = dpa_bp->seed_cb(dpa_bp);
+		if (err)
+			goto pool_seed_failed;
+	}
+
+	dpa_bpid2pool_map(dpa_bp->bpid, dpa_bp);
+
+	return 0;
+
+pool_seed_failed:
+pdev_mask_failed:
+	platform_device_unregister(pdev);
+pdev_register_failed:
+	bman_free_pool(dpa_bp->pool);
+
+	return err;
+}
+
+void dpa_bp_drain(struct dpa_bp *bp)
+{
+	int ret;
+	u8 num = 8;
+
+	do {
+		struct bm_buffer bmb[8];
+		int i;
+
+		ret = bman_acquire(bp->pool, bmb, num, 0);
+		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;
+			}
+		}
+
+		for (i = 0; i < num; i++) {
+			dma_addr_t addr = bm_buf_addr(&bmb[i]);
+
+			dma_unmap_single(bp->dev, addr, bp->size,
+					 DMA_BIDIRECTIONAL);
+
+			bp->free_buf_cb(phys_to_virt(addr));
+		}
+	} while (ret > 0);
+}
+
+static void _dpa_bp_free(struct dpa_bp *dpa_bp)
+{
+	struct dpa_bp *bp = dpa_bpid2pool(dpa_bp->bpid);
+
+	/* the mapping between bpid and dpa_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)
+		dpa_bp_drain(bp);
+
+	dpa_bp_array[bp->bpid] = NULL;
+	bman_free_pool(bp->pool);
+
+	if (bp->dev)
+		platform_device_unregister(to_platform_device(bp->dev));
+}
+
+void dpa_bp_free(struct dpa_priv_s *priv)
+{
+	int i;
+
+	for (i = 0; i < priv->bp_count; i++)
+		_dpa_bp_free(&priv->dpa_bp[i]);
+}
+
+struct dpa_bp *dpa_bpid2pool(int bpid)
+{
+	return dpa_bp_array[bpid];
+}
+
+void dpa_bpid2pool_map(int bpid, struct dpa_bp *dpa_bp)
+{
+	dpa_bp_array[bpid] = dpa_bp;
+	atomic_set(&dpa_bp->refs, 1);
+}
+
+bool dpa_bpid2pool_use(int bpid)
+{
+	if (dpa_bpid2pool(bpid)) {
+		atomic_inc(&dpa_bp_array[bpid]->refs);
+		return true;
+	}
+
+	return false;
+}
+
+struct dpa_fq *dpa_fq_alloc(struct device *dev,
+			    const struct fqid_cell *fqids,
+			    struct list_head *list,
+			    enum dpa_fq_type fq_type)
+{
+	int i;
+	struct dpa_fq *dpa_fq;
+
+	dpa_fq = devm_kzalloc(dev, sizeof(*dpa_fq) * fqids->count, GFP_KERNEL);
+	if (!dpa_fq)
+		return NULL;
+
+	for (i = 0; i < fqids->count; i++) {
+		dpa_fq[i].fq_type = fq_type;
+		dpa_fq[i].fqid = fqids->start ? fqids->start + i : 0;
+		list_add_tail(&dpa_fq[i].list, list);
+	}
+
+	for (i = 0; i < fqids->count; i++)
+		_dpa_assign_wq(dpa_fq + i);
+
+	return dpa_fq;
+}
+
+int dpa_fq_probe_mac(struct device *dev, struct list_head *list,
+		     struct fm_port_fqs *port_fqs,
+		     bool alloc_tx_conf_fqs,
+		     enum port_type ptype)
+{
+	const struct fqid_cell *fqids;
+	struct dpa_fq *dpa_fq;
+	int num_ranges;
+	int i;
+
+	if (ptype == TX && alloc_tx_conf_fqs) {
+		if (!dpa_fq_alloc(dev, tx_confirm_fqids, list,
+				  FQ_TYPE_TX_CONF_MQ))
+			goto fq_alloc_failed;
+	}
+
+	fqids = default_fqids[ptype];
+	num_ranges = 3;
+
+	for (i = 0; i < num_ranges; i++) {
+		switch (i) {
+		case 0:
+			/* The first queue is the error queue */
+			if (fqids[i].count != 1)
+				goto invalid_error_queue;
+
+			dpa_fq = dpa_fq_alloc(dev, &fqids[i], list,
+					      ptype == RX ?
+						FQ_TYPE_RX_ERROR :
+						FQ_TYPE_TX_ERROR);
+			if (!dpa_fq)
+				goto fq_alloc_failed;
+
+			if (ptype == RX)
+				port_fqs->rx_errq = &dpa_fq[0];
+			else
+				port_fqs->tx_errq = &dpa_fq[0];
+			break;
+		case 1:
+			/* the second queue is the default queue */
+			if (fqids[i].count != 1)
+				goto invalid_default_queue;
+
+			dpa_fq = dpa_fq_alloc(dev, &fqids[i], list,
+					      ptype == RX ?
+						FQ_TYPE_RX_DEFAULT :
+						FQ_TYPE_TX_CONFIRM);
+			if (!dpa_fq)
+				goto fq_alloc_failed;
+
+			if (ptype == RX)
+				port_fqs->rx_defq = &dpa_fq[0];
+			else
+				port_fqs->tx_defq = &dpa_fq[0];
+			break;
+		default:
+			/* all subsequent queues are Tx */
+			if (!dpa_fq_alloc(dev, &fqids[i], list, FQ_TYPE_TX))
+				goto fq_alloc_failed;
+			break;
+		}
+	}
+
+	return 0;
+
+fq_alloc_failed:
+	dev_err(dev, "dpa_fq_alloc() failed\n");
+	return -ENOMEM;
+
+invalid_default_queue:
+invalid_error_queue:
+	dev_err(dev, "Too many default or error queues\n");
+	return -EINVAL;
+}
+
+static u32 rx_pool_channel;
+static DEFINE_SPINLOCK(rx_pool_channel_init);
+
+int dpa_get_channel(void)
+{
+	spin_lock(&rx_pool_channel_init);
+	if (!rx_pool_channel) {
+		u32 pool;
+		int 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;
+}
+
+void dpa_release_channel(void)
+{
+	qman_release_pool(rx_pool_channel);
+}
+
+int dpaa_eth_add_channel(void *__arg)
+{
+	const cpumask_t *cpus = qman_affine_cpus();
+	u32 pool = QM_SDQCR_CHANNELS_POOL_CONV((u16)(unsigned long)__arg);
+	int cpu;
+	struct qman_portal *portal;
+
+	for_each_cpu(cpu, cpus) {
+		portal = (struct qman_portal *)qman_get_affine_portal(cpu);
+		qman_p_static_dequeue_add(portal, pool);
+	}
+	return 0;
+}
+
+/* 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 dpa_priv_s *priv = (struct dpa_priv_s *)container_of(cgr,
+		struct dpa_priv_s, cgr_data.cgr);
+
+	if (congested)
+		netif_tx_stop_all_queues(priv->net_dev);
+	else
+		netif_tx_wake_all_queues(priv->net_dev);
+}
+
+int dpaa_eth_cgr_init(struct dpa_priv_s *priv)
+{
+	struct qm_mcc_initcgr initcgr;
+	u32 cs_th;
+	int err;
+
+	err = qman_alloc_cgrid(&priv->cgr_data.cgr.cgrid);
+	if (err < 0) {
+		pr_err("Error %d allocating CGR ID\n", 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) {
+		pr_err("Error %d creating CGR with ID %d\n", err,
+		       priv->cgr_data.cgr.cgrid);
+		qman_release_cgrid(priv->cgr_data.cgr.cgrid);
+		goto out_error;
+	}
+	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 dpa_setup_ingress(const struct dpa_priv_s *priv,
+				     struct dpa_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 dpa_setup_egress(const struct dpa_priv_s *priv,
+				    struct dpa_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;
+	}
+}
+
+void dpa_fq_setup(struct dpa_priv_s *priv, const struct dpa_fq_cbs_t *fq_cbs,
+		  struct fman_port *tx_port)
+{
+	struct dpa_fq *fq;
+	u16 portals[NR_CPUS];
+	int cpu, num_portals = 0;
+	const cpumask_t *affine_cpus = qman_affine_cpus();
+	int egress_cnt = 0, conf_cnt = 0;
+
+	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->dpa_fq_list, list) {
+		switch (fq->fq_type) {
+		case FQ_TYPE_RX_DEFAULT:
+			DPA_ERR_ON(!priv->mac_dev);
+			dpa_setup_ingress(priv, fq, &fq_cbs->rx_defq);
+			break;
+		case FQ_TYPE_RX_ERROR:
+			DPA_ERR_ON(!priv->mac_dev);
+			dpa_setup_ingress(priv, fq, &fq_cbs->rx_errq);
+			break;
+		case FQ_TYPE_TX:
+			dpa_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_TX_QUEUES)
+				priv->egress_fqs[egress_cnt++] = &fq->fq_base;
+			break;
+		case FQ_TYPE_TX_CONFIRM:
+			DPA_ERR_ON(!priv->mac_dev);
+			dpa_setup_ingress(priv, fq, &fq_cbs->tx_defq);
+			break;
+		case FQ_TYPE_TX_CONF_MQ:
+			DPA_ERR_ON(!priv->mac_dev);
+			dpa_setup_ingress(priv, fq, &fq_cbs->tx_defq);
+			priv->conf_fqs[conf_cnt++] = &fq->fq_base;
+			break;
+		case FQ_TYPE_TX_ERROR:
+			DPA_ERR_ON(!priv->mac_dev);
+			dpa_setup_ingress(priv, fq, &fq_cbs->tx_errq);
+			break;
+		default:
+			dev_warn(priv->net_dev->dev.parent,
+				 "Unknown FQ type detected!\n");
+			break;
+		}
+	}
+
+	/* The number of Tx queues may be smaller than the number of cores, if
+	 * the Tx queue range is specified in the device tree instead of being
+	 * dynamically allocated.
+	 * Make sure all CPUs receive a corresponding Tx queue.
+	 */
+	while (egress_cnt < DPAA_ETH_TX_QUEUES) {
+		list_for_each_entry(fq, &priv->dpa_fq_list, list) {
+			if (fq->fq_type != FQ_TYPE_TX)
+				continue;
+			priv->egress_fqs[egress_cnt++] = &fq->fq_base;
+			if (egress_cnt == DPAA_ETH_TX_QUEUES)
+				break;
+		}
+	}
+}
+
+int dpa_fq_init(struct dpa_fq *dpa_fq, bool td_enable)
+{
+	int err;
+	const struct dpa_priv_s	*priv;
+	struct device *dev;
+	struct qman_fq *fq;
+	struct qm_mcc_initfq initfq;
+	struct qman_fq *confq = NULL;
+	int queue_id;
+
+	priv = netdev_priv(dpa_fq->net_dev);
+	dev = dpa_fq->net_dev->dev.parent;
+
+	if (dpa_fq->fqid == 0)
+		dpa_fq->flags |= QMAN_FQ_FLAG_DYNAMIC_FQID;
+
+	dpa_fq->init = !(dpa_fq->flags & QMAN_FQ_FLAG_NO_MODIFY);
+
+	err = qman_create_fq(dpa_fq->fqid, dpa_fq->flags, &dpa_fq->fq_base);
+	if (err) {
+		dev_err(dev, "qman_create_fq() failed\n");
+		return err;
+	}
+	fq = &dpa_fq->fq_base;
+
+	if (dpa_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 (dpa_fq->fq_type == FQ_TYPE_TX_CONFIRM)
+			initfq.fqd.fq_ctrl |= QM_FQCTRL_HOLDACTIVE;
+
+		/* FQ placement */
+		initfq.we_mask |= QM_INITFQ_WE_DESTWQ;
+
+		initfq.fqd.dest.channel	= dpa_fq->channel;
+		initfq.fqd.dest.wq = dpa_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 (dpa_fq->fq_type == FQ_TYPE_TX ||
+		    dpa_fq->fq_type == FQ_TYPE_TX_CONFIRM ||
+		    dpa_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;
+			initfq.fqd.oac_init.oac = QM_OAC_CG;
+			initfq.fqd.oac_init.oal =
+				(signed char)(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_taildrop_set(&initfq.fqd.td,
+					    DPA_FQ_TD, 1);
+			initfq.fqd.fq_ctrl = QM_FQCTRL_TDE;
+		}
+
+		/* Configure the Tx confirmation queue, now that we know
+		 * which Tx queue it pairs with.
+		 */
+		if (dpa_fq->fq_type == FQ_TYPE_TX) {
+			queue_id = _dpa_tx_fq_to_id(priv, &dpa_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 *private* ingress queues in our "ingress CGR". */
+		if (priv->use_ingress_cgr &&
+		    (dpa_fq->fq_type == FQ_TYPE_RX_DEFAULT ||
+		     dpa_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;
+			initfq.fqd.oac_init.oac = QM_OAC_CG;
+			initfq.fqd.oac_init.oal =
+				(signed char)(min(sizeof(struct sk_buff) +
+				priv->tx_headroom, (size_t)FSL_QMAN_MAX_OAL));
+		}
+
+		/* Initialization common to all ingress queues */
+		if (dpa_fq->flags & QMAN_FQ_FLAG_NO_ENQUEUE) {
+			initfq.we_mask |= QM_INITFQ_WE_CONTEXTA;
+			initfq.fqd.fq_ctrl |=
+				QM_FQCTRL_CTXASTASHING | QM_FQCTRL_AVOIDBLOCK;
+			initfq.fqd.context_a.stashing.exclusive =
+				QM_STASHING_EXCL_DATA | QM_STASHING_EXCL_CTX |
+				QM_STASHING_EXCL_ANNOTATION;
+			initfq.fqd.context_a.stashing.data_cl = 2;
+			initfq.fqd.context_a.stashing.annotation_cl = 1;
+			initfq.fqd.context_a.stashing.context_cl =
+				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, 0);
+			return err;
+		}
+	}
+
+	dpa_fq->fqid = qman_fq_fqid(fq);
+
+	return 0;
+}
+
+static int _dpa_fq_free(struct device *dev, struct qman_fq *fq)
+{
+	int err, error;
+	struct dpa_fq *dpa_fq;
+	const struct dpa_priv_s	*priv;
+
+	err = 0;
+
+	dpa_fq = container_of(fq, struct dpa_fq, fq_base);
+	priv = netdev_priv(dpa_fq->net_dev);
+
+	if (dpa_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)) {
+			dev_err(dev, "qman_oos_fq(%u) = %d\n",
+				qman_fq_fqid(fq), error);
+			if (err >= 0)
+				err = error;
+		}
+	}
+
+	qman_destroy_fq(fq, 0);
+	list_del(&dpa_fq->list);
+
+	return err;
+}
+
+int dpa_fq_free(struct device *dev, struct list_head *list)
+{
+	int err, error;
+	struct dpa_fq *dpa_fq, *tmp;
+
+	err = 0;
+	list_for_each_entry_safe(dpa_fq, tmp, list, list) {
+		error = _dpa_fq_free(dev, (struct qman_fq *)dpa_fq);
+		if (error < 0 && err >= 0)
+			err = error;
+	}
+
+	return err;
+}
+
+static void
+dpaa_eth_init_tx_port(struct fman_port *port, struct dpa_fq *errq,
+		      struct dpa_fq *defq,
+		      struct dpa_buffer_layout_s *buf_layout)
+{
+	struct fman_port_params params;
+	struct fman_buffer_prefix_content buf_prefix_content;
+	int err;
+
+	memset(&params, 0, sizeof(params));
+	memset(&buf_prefix_content, 0, sizeof(buf_prefix_content));
+
+	buf_prefix_content.priv_data_size = buf_layout->priv_data_size;
+	buf_prefix_content.pass_prs_result = buf_layout->parse_results;
+	buf_prefix_content.pass_hash_result = buf_layout->hash_results;
+	buf_prefix_content.pass_time_stamp = buf_layout->time_stamp;
+	buf_prefix_content.data_align = buf_layout->data_align;
+
+	params.specific_params.non_rx_params.err_fqid = errq->fqid;
+	params.specific_params.non_rx_params.dflt_fqid = defq->fqid;
+
+	err = fman_port_config(port, &params);
+	if (err)
+		pr_info("fman_port_config failed\n");
+
+	err = fman_port_cfg_buf_prefix_content(port, &buf_prefix_content);
+	if (err)
+		pr_info("fman_port_cfg_buf_prefix_content failed\n");
+
+	err = fman_port_init(port);
+	if (err)
+		pr_err("fm_port_init failed\n");
+}
+
+static void
+dpaa_eth_init_rx_port(struct fman_port *port, struct dpa_bp *bp,
+		      size_t count, struct dpa_fq *errq, struct dpa_fq *defq,
+		      struct dpa_buffer_layout_s *buf_layout)
+{
+	struct fman_port_params params;
+	struct fman_buffer_prefix_content buf_prefix_content;
+	struct fman_port_rx_params *rx_p;
+	int i, err;
+
+	memset(&params, 0, sizeof(params));
+	memset(&buf_prefix_content, 0, sizeof(buf_prefix_content));
+
+	buf_prefix_content.priv_data_size = buf_layout->priv_data_size;
+	buf_prefix_content.pass_prs_result = buf_layout->parse_results;
+	buf_prefix_content.pass_hash_result = buf_layout->hash_results;
+	buf_prefix_content.pass_time_stamp = buf_layout->time_stamp;
+	buf_prefix_content.data_align = buf_layout->data_align;
+
+	rx_p = &params.specific_params.rx_params;
+	rx_p->err_fqid = errq->fqid;
+	rx_p->dflt_fqid = defq->fqid;
+
+	count = min(ARRAY_SIZE(rx_p->ext_buf_pools.ext_buf_pool), count);
+	rx_p->ext_buf_pools.num_of_pools_used = (u8)count;
+	for (i = 0; i < count; i++) {
+		rx_p->ext_buf_pools.ext_buf_pool[i].id =  bp[i].bpid;
+		rx_p->ext_buf_pools.ext_buf_pool[i].size = (u16)bp[i].size;
+	}
+
+	err = fman_port_config(port, &params);
+	if (err)
+		pr_info("fman_port_config failed\n");
+
+	err = fman_port_cfg_buf_prefix_content(port, &buf_prefix_content);
+	if (err)
+		pr_info("fman_port_cfg_buf_prefix_content failed\n");
+
+	err = fman_port_init(port);
+	if (err)
+		pr_err("fm_port_init failed\n");
+}
+
+void dpaa_eth_init_ports(struct mac_device *mac_dev,
+			 struct dpa_bp *bp, size_t count,
+			 struct fm_port_fqs *port_fqs,
+			 struct dpa_buffer_layout_s *buf_layout,
+			 struct device *dev)
+{
+	struct fman_port *rxport = mac_dev->port[RX];
+	struct fman_port *txport = mac_dev->port[TX];
+
+	dpaa_eth_init_tx_port(txport, port_fqs->tx_errq,
+			      port_fqs->tx_defq, &buf_layout[TX]);
+	dpaa_eth_init_rx_port(rxport, bp, count, port_fqs->rx_errq,
+			      port_fqs->rx_defq, &buf_layout[RX]);
+}
+
+void dpa_fd_release(const struct net_device *net_dev, const struct qm_fd *fd)
+{
+	struct dpa_bp *dpa_bp;
+	struct bm_buffer bmb;
+
+	memset(&bmb, 0, sizeof(bmb));
+	bm_buffer_set64(&bmb, fd->addr);
+
+	dpa_bp = dpa_bpid2pool(fd->bpid);
+	DPA_ERR_ON(!dpa_bp);
+
+	DPA_ERR_ON(fd->format == qm_fd_sg);
+
+	while (bman_release(dpa_bp->pool, &bmb, 1, 0))
+		cpu_relax();
+}
+
+/* Turn on HW checksum computation for this outgoing frame.
+ * If the current protocol is not something we support in this regard
+ * (or if the stack has already computed the SW checksum), we do nothing.
+ *
+ * Returns 0 if all goes well (or HW csum doesn't apply), and a negative value
+ * otherwise.
+ *
+ * Note that this function may modify the fd->cmd field and the skb data buffer
+ * (the Parse Results area).
+ */
+int dpa_enable_tx_csum(struct dpa_priv_s *priv,
+		       struct sk_buff *skb,
+		       struct qm_fd *fd,
+		       char *parse_results)
+{
+	struct fman_prs_result *parse_result;
+	struct iphdr *iph;
+	struct ipv6hdr *ipv6h = NULL;
+	u8 l4_proto;
+	u16 ethertype = ntohs(skb->protocol);
+	int retval = 0;
+
+	if (skb->ip_summed != CHECKSUM_PARTIAL)
+		return 0;
+
+	/* Note: L3 csum seems to be already computed in sw, but we can't choose
+	 * L4 alone from the FM configuration anyway.
+	 */
+
+	/* Fill in some fields of the Parse Results array, so the FMan
+	 * can find them as if they came from the FMan Parser.
+	 */
+	parse_result = (struct fman_prs_result *)parse_results;
+
+	/* If we're dealing with VLAN, get the real Ethernet type */
+	if (ethertype == ETH_P_8021Q) {
+		/* We can't always assume the MAC header is set correctly
+		 * by the stack, so reset to beginning of skb->data
+		 */
+		skb_reset_mac_header(skb);
+		ethertype = ntohs(vlan_eth_hdr(skb)->h_vlan_encapsulated_proto);
+	}
+
+	/* Fill in the relevant L3 parse result fields
+	 * and read the L4 protocol type
+	 */
+	switch (ethertype) {
+	case ETH_P_IP:
+		parse_result->l3r = cpu_to_be16(FM_L3_PARSE_RESULT_IPV4);
+		iph = ip_hdr(skb);
+		DPA_ERR_ON(!iph);
+		l4_proto = iph->protocol;
+		break;
+	case ETH_P_IPV6:
+		parse_result->l3r = cpu_to_be16(FM_L3_PARSE_RESULT_IPV6);
+		ipv6h = ipv6_hdr(skb);
+		DPA_ERR_ON(!ipv6h);
+		l4_proto = ipv6h->nexthdr;
+		break;
+	default:
+		/* We shouldn't even be here */
+		if (net_ratelimit())
+			netif_alert(priv, tx_err, priv->net_dev,
+				    "Can't compute HW csum for L3 proto 0x%x\n",
+				    ntohs(skb->protocol));
+		retval = -EIO;
+		goto return_error;
+	}
+
+	/* Fill in the relevant L4 parse result fields */
+	switch (l4_proto) {
+	case IPPROTO_UDP:
+		parse_result->l4r = FM_L4_PARSE_RESULT_UDP;
+		break;
+	case IPPROTO_TCP:
+		parse_result->l4r = FM_L4_PARSE_RESULT_TCP;
+		break;
+	default:
+		/* This can as well be a BUG() */
+		if (net_ratelimit())
+			netif_alert(priv, tx_err, priv->net_dev,
+				    "Can't compute HW csum for L4 proto 0x%x\n",
+				    l4_proto);
+		retval = -EIO;
+		goto return_error;
+	}
+
+	/* At index 0 is IPOffset_1 as defined in the Parse Results */
+	parse_result->ip_off[0] = (u8)skb_network_offset(skb);
+	parse_result->l4_off = (u8)skb_transport_offset(skb);
+
+	/* Enable L3 (and L4, if TCP or UDP) HW checksum. */
+	fd->cmd |= FM_FD_CMD_RPD | FM_FD_CMD_DTC;
+
+	/* On P1023 and similar platforms fd->cmd interpretation could
+	 * be disabled by setting CONTEXT_A bit ICMD; currently this bit
+	 * is not set so we do not need to check; in the future, if/when
+	 * using context_a we need to check this bit
+	 */
+
+return_error:
+	return retval;
+}
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h
new file mode 100644
index 0000000..68843c0
--- /dev/null
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h
@@ -0,0 +1,98 @@
+/* Copyright 2008 - 2015 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.
+ */
+
+#ifndef __DPAA_ETH_COMMON_H
+#define __DPAA_ETH_COMMON_H
+
+#include <linux/etherdevice.h>
+#include <soc/fsl/bman.h>
+#include <linux/of_platform.h>
+
+#include "dpaa_eth.h"
+
+#define DPA_BUFF_RELEASE_MAX 8 /* maximum number of buffers released at once */
+
+/* used in napi related functions */
+extern u16 qman_portal_max;
+
+int dpa_netdev_init(struct net_device *net_dev,
+		    const u8 *mac_addr,
+		    u16 tx_timeout);
+int dpa_start(struct net_device *net_dev);
+int dpa_stop(struct net_device *net_dev);
+void dpa_timeout(struct net_device *net_dev);
+struct rtnl_link_stats64 *dpa_get_stats64(struct net_device *net_dev,
+					  struct rtnl_link_stats64 *stats);
+int dpa_change_mtu(struct net_device *net_dev, int new_mtu);
+int dpa_ndo_init(struct net_device *net_dev);
+int dpa_set_features(struct net_device *dev, netdev_features_t features);
+netdev_features_t dpa_fix_features(struct net_device *dev,
+				   netdev_features_t features);
+int dpa_remove(struct platform_device *pdev);
+struct mac_device *dpa_mac_dev_get(struct platform_device *pdev);
+int dpa_mac_hw_index_get(struct platform_device *pdev);
+int dpa_mac_fman_index_get(struct platform_device *pdev);
+int dpa_set_mac_address(struct net_device *net_dev, void *addr);
+void dpa_set_rx_mode(struct net_device *net_dev);
+void dpa_set_buffers_layout(struct mac_device *mac_dev,
+			    struct dpa_buffer_layout_s *layout);
+int dpa_bp_alloc(struct dpa_bp *dpa_bp);
+void dpa_bp_free(struct dpa_priv_s *priv);
+struct dpa_bp *dpa_bpid2pool(int bpid);
+void dpa_bpid2pool_map(int bpid, struct dpa_bp *dpa_bp);
+bool dpa_bpid2pool_use(int bpid);
+void dpa_bp_drain(struct dpa_bp *bp);
+struct dpa_fq *dpa_fq_alloc(struct device *dev,
+			    const struct fqid_cell *fqids,
+			    struct list_head *list,
+			    enum dpa_fq_type fq_type);
+int dpa_fq_probe_mac(struct device *dev, struct list_head *list,
+		     struct fm_port_fqs *port_fqs,
+		     bool tx_conf_fqs_per_core,
+		     enum port_type ptype);
+int dpa_get_channel(void);
+void dpa_release_channel(void);
+int dpaa_eth_add_channel(void *__arg);
+int dpaa_eth_cgr_init(struct dpa_priv_s *priv);
+void dpa_fq_setup(struct dpa_priv_s *priv, const struct dpa_fq_cbs_t *fq_cbs,
+		  struct fman_port *tx_port);
+int dpa_fq_init(struct dpa_fq *dpa_fq, bool td_enable);
+int dpa_fq_free(struct device *dev, struct list_head *list);
+void dpaa_eth_init_ports(struct mac_device *mac_dev,
+			 struct dpa_bp *bp, size_t count,
+			 struct fm_port_fqs *port_fqs,
+			 struct dpa_buffer_layout_s *buf_layout,
+			 struct device *dev);
+void dpa_fd_release(const struct net_device *net_dev, const struct qm_fd *fd);
+int dpa_enable_tx_csum(struct dpa_priv_s *priv,
+		       struct sk_buff *skb,
+		       struct qm_fd *fd,
+		       char *parse_results);
+#endif	/* __DPAA_ETH_COMMON_H */
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_sg.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_sg.c
new file mode 100644
index 0000000..115798b
--- /dev/null
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_sg.c
@@ -0,0 +1,408 @@
+/* Copyright 2012 - 2015 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/skbuff.h>
+#include <linux/highmem.h>
+#include <soc/fsl/bman.h>
+
+#include "dpaa_eth.h"
+#include "dpaa_eth_common.h"
+
+/* Convenience macros for storing/retrieving the skb back-pointers.
+ *
+ * NB: @off is an offset from a (struct sk_buff **) pointer!
+ */
+#define DPA_WRITE_SKB_PTR(skb, skbh, addr, off) \
+	{ \
+		skbh = (struct sk_buff **)addr; \
+		*(skbh + (off)) = skb; \
+	}
+#define DPA_READ_SKB_PTR(skb, skbh, addr, off) \
+	{ \
+		skbh = (struct sk_buff **)addr; \
+		skb = *(skbh + (off)); \
+	}
+
+static int _dpa_bp_add_8_bufs(const struct dpa_bp *dpa_bp)
+{
+	struct bm_buffer bmb[8];
+	void *new_buf;
+	dma_addr_t addr;
+	u8 i;
+	struct device *dev = dpa_bp->dev;
+	struct sk_buff *skb, **skbh;
+
+	memset(bmb, 0, sizeof(bmb));
+
+	for (i = 0; i < 8; i++) {
+		/* We'll prepend the skb back-pointer; can't use the DPA
+		 * priv space, because FMan will overwrite it (from offset 0)
+		 * if it ends up being the second, third, etc. fragment
+		 * in a S/G frame.
+		 *
+		 * We only need enough space to store a pointer, but allocate
+		 * an entire cacheline for performance reasons.
+		 */
+		new_buf = netdev_alloc_frag(SMP_CACHE_BYTES + DPA_BP_RAW_SIZE);
+		if (unlikely(!new_buf))
+			goto netdev_alloc_failed;
+		new_buf = PTR_ALIGN(new_buf + SMP_CACHE_BYTES, SMP_CACHE_BYTES);
+
+		skb = build_skb(new_buf, DPA_SKB_SIZE(dpa_bp->size) +
+			SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
+		if (unlikely(!skb)) {
+			put_page(virt_to_head_page(new_buf));
+			goto build_skb_failed;
+		}
+		DPA_WRITE_SKB_PTR(skb, skbh, new_buf, -1);
+
+		addr = dma_map_single(dev, new_buf,
+				      dpa_bp->size, DMA_BIDIRECTIONAL);
+		if (unlikely(dma_mapping_error(dev, addr)))
+			goto dma_map_failed;
+
+		bm_buffer_set64(&bmb[i], addr);
+	}
+
+release_bufs:
+	/* Release the buffers. In case bman is busy, keep trying
+	 * until successful. bman_release() is guaranteed to succeed
+	 * in a reasonable amount of time
+	 */
+	while (unlikely(bman_release(dpa_bp->pool, bmb, i, 0)))
+		cpu_relax();
+	return i;
+
+dma_map_failed:
+	kfree_skb(skb);
+
+build_skb_failed:
+netdev_alloc_failed:
+	net_err_ratelimited("dpa_bp_add_8_bufs() failed\n");
+	WARN_ONCE(1, "Memory allocation failure on Rx\n");
+
+	bm_buffer_set64(&bmb[i], 0);
+	/* Avoid releasing a completely null buffer; bman_release() requires
+	 * at least one buffer.
+	 */
+	if (likely(i))
+		goto release_bufs;
+
+	return 0;
+}
+
+/* Cold path wrapper over _dpa_bp_add_8_bufs(). */
+static void dpa_bp_add_8_bufs(const struct dpa_bp *dpa_bp, int cpu)
+{
+	int *count_ptr = per_cpu_ptr(dpa_bp->percpu_count, cpu);
+	*count_ptr += _dpa_bp_add_8_bufs(dpa_bp);
+}
+
+int dpa_bp_priv_seed(struct dpa_bp *dpa_bp)
+{
+	int i;
+
+	/* Give each CPU an allotment of "config_count" buffers */
+	for_each_possible_cpu(i) {
+		int j;
+
+		/* Although we access another CPU's counters here
+		 * we do it at boot time so it is safe
+		 */
+		for (j = 0; j < dpa_bp->config_count; j += 8)
+			dpa_bp_add_8_bufs(dpa_bp, i);
+	}
+	return 0;
+}
+
+/* Add buffers/(pages) for Rx processing whenever bpool count falls below
+ * REFILL_THRESHOLD.
+ */
+int dpaa_eth_refill_bpools(struct dpa_bp *dpa_bp, int *countptr)
+{
+	int count = *countptr;
+	int new_bufs;
+
+	if (unlikely(count < FSL_DPAA_ETH_REFILL_THRESHOLD)) {
+		do {
+			new_bufs = _dpa_bp_add_8_bufs(dpa_bp);
+			if (unlikely(!new_bufs)) {
+				/* Avoid looping forever if we've temporarily
+				 * run out of memory. We'll try again at the
+				 * next NAPI cycle.
+				 */
+				break;
+			}
+			count += new_bufs;
+		} while (count < FSL_DPAA_ETH_MAX_BUF_COUNT);
+
+		*countptr = count;
+		if (unlikely(count < FSL_DPAA_ETH_MAX_BUF_COUNT))
+			return -ENOMEM;
+	}
+
+	return 0;
+}
+
+/* Cleanup function for outgoing frame descriptors that were built on Tx path,
+ * either contiguous frames or scatter/gather ones.
+ * Skb freeing is not handled here.
+ *
+ * This function may be called on error paths in the Tx function, so guard
+ * against cases when not all fd relevant fields were filled in.
+ *
+ * Return the skb backpointer, since for S/G frames the buffer containing it
+ * gets freed here.
+ */
+struct sk_buff *_dpa_cleanup_tx_fd(const struct dpa_priv_s *priv,
+				   const struct qm_fd *fd)
+{
+	struct dpa_bp *dpa_bp = priv->dpa_bp;
+	dma_addr_t addr = qm_fd_addr(fd);
+	struct sk_buff **skbh;
+	struct sk_buff *skb = NULL;
+	const enum dma_data_direction dma_dir = DMA_TO_DEVICE;
+
+	/* retrieve skb back pointer */
+	DPA_READ_SKB_PTR(skb, skbh, phys_to_virt(addr), 0);
+	dma_unmap_single(dpa_bp->dev, addr,
+			 skb_tail_pointer(skb) - (u8 *)skbh, dma_dir);
+	return skb;
+}
+
+/* Build a linear skb around the received buffer.
+ * We are guaranteed there is enough room at the end of the data buffer to
+ * accommodate the shared info area of the skb.
+ */
+static struct sk_buff *contig_fd_to_skb(const struct dpa_priv_s *priv,
+					const struct qm_fd *fd)
+{
+	struct sk_buff *skb = NULL, **skbh;
+	ssize_t fd_off = dpa_fd_offset(fd);
+	dma_addr_t addr = qm_fd_addr(fd);
+	void *vaddr;
+
+	vaddr = phys_to_virt(addr);
+	DPA_ERR_ON(!IS_ALIGNED((unsigned long)vaddr, SMP_CACHE_BYTES));
+
+	/* Retrieve the skb and adjust data and tail pointers, to make sure
+	 * forwarded skbs will have enough space on Tx if extra headers
+	 * are added.
+	 */
+	DPA_READ_SKB_PTR(skb, skbh, vaddr, -1);
+
+	DPA_ERR_ON(fd_off != priv->rx_headroom);
+	skb_reserve(skb, fd_off);
+	skb_put(skb, dpa_fd_length(fd));
+
+	skb->ip_summed = CHECKSUM_NONE;
+
+	return skb;
+}
+
+void _dpa_rx(struct net_device *net_dev,
+	     struct qman_portal *portal,
+	     const struct dpa_priv_s *priv,
+	     struct dpa_percpu_priv_s *percpu_priv,
+	     const struct qm_fd *fd,
+	     u32 fqid,
+	     int *count_ptr)
+{
+	struct dpa_bp *dpa_bp;
+	struct sk_buff *skb;
+	dma_addr_t addr = qm_fd_addr(fd);
+	u32 fd_status = fd->status;
+	unsigned int skb_len;
+	struct rtnl_link_stats64 *percpu_stats = &percpu_priv->stats;
+
+	if (unlikely(fd_status & FM_FD_STAT_RX_ERRORS) != 0) {
+		if (net_ratelimit())
+			netif_warn(priv, hw, net_dev, "FD status = 0x%08x\n",
+				   fd_status & FM_FD_STAT_RX_ERRORS);
+
+		percpu_stats->rx_errors++;
+		goto _release_frame;
+	}
+
+	dpa_bp = priv->dpa_bp;
+	DPA_ERR_ON(dpa_bp != dpa_bpid2pool(fd->bpid));
+
+	/* prefetch the first 64 bytes of the frame */
+	dma_unmap_single(dpa_bp->dev, addr, dpa_bp->size, DMA_BIDIRECTIONAL);
+	prefetch(phys_to_virt(addr) + dpa_fd_offset(fd));
+
+	/* The only FD type that we may receive is contig */
+	DPA_ERR_ON((fd->format != qm_fd_contig));
+
+	skb = contig_fd_to_skb(priv, fd);
+
+	/* Account for the contig buffer
+	 * having been removed from the pool.
+	 */
+	(*count_ptr)--;
+	skb->protocol = eth_type_trans(skb, net_dev);
+
+	/* IP Reassembled frames are allowed to be larger than MTU */
+	if (unlikely(dpa_check_rx_mtu(skb, net_dev->mtu) &&
+		     !(fd_status & FM_FD_IPR))) {
+		percpu_stats->rx_dropped++;
+		goto drop_bad_frame;
+	}
+
+	skb_len = skb->len;
+
+	if (unlikely(netif_receive_skb(skb) == NET_RX_DROP))
+		goto packet_dropped;
+
+	percpu_stats->rx_packets++;
+	percpu_stats->rx_bytes += skb_len;
+
+packet_dropped:
+	return;
+
+drop_bad_frame:
+	dev_kfree_skb(skb);
+	return;
+
+_release_frame:
+	dpa_fd_release(net_dev, fd);
+}
+
+static int skb_to_contig_fd(struct dpa_priv_s *priv,
+			    struct sk_buff *skb, struct qm_fd *fd,
+			    int *count_ptr, int *offset)
+{
+	struct sk_buff **skbh;
+	dma_addr_t addr;
+	struct dpa_bp *dpa_bp = priv->dpa_bp;
+	struct net_device *net_dev = priv->net_dev;
+	int err;
+	enum dma_data_direction dma_dir;
+	unsigned char *buffer_start;
+
+	{
+		/* We are guaranteed to have at least tx_headroom bytes
+		 * available, so just use that for offset.
+		 */
+		fd->bpid = 0xff;
+		buffer_start = skb->data - priv->tx_headroom;
+		fd->offset = priv->tx_headroom;
+		dma_dir = DMA_TO_DEVICE;
+
+		DPA_WRITE_SKB_PTR(skb, skbh, buffer_start, 0);
+	}
+
+	/* Enable L3/L4 hardware checksum computation.
+	 *
+	 * We must do this before dma_map_single(DMA_TO_DEVICE), because we may
+	 * need to write into the skb.
+	 */
+	err = dpa_enable_tx_csum(priv, skb, fd,
+				 ((char *)skbh) + DPA_TX_PRIV_DATA_SIZE);
+	if (unlikely(err < 0)) {
+		if (net_ratelimit())
+			netif_err(priv, tx_err, net_dev, "HW csum error: %d\n",
+				  err);
+		return err;
+	}
+
+	/* Fill in the rest of the FD fields */
+	fd->format = qm_fd_contig;
+	fd->length20 = skb->len;
+	fd->cmd |= FM_FD_CMD_FCO;
+
+	/* Map the entire buffer size that may be seen by FMan, but no more */
+	addr = dma_map_single(dpa_bp->dev, skbh,
+			      skb_tail_pointer(skb) - buffer_start, dma_dir);
+	if (unlikely(dma_mapping_error(dpa_bp->dev, addr))) {
+		if (net_ratelimit())
+			netif_err(priv, tx_err, net_dev, "dma_map_single() failed\n");
+		return -EINVAL;
+	}
+	fd->addr_hi = (u8)upper_32_bits(addr);
+	fd->addr_lo = lower_32_bits(addr);
+
+	return 0;
+}
+
+int dpa_tx(struct sk_buff *skb, struct net_device *net_dev)
+{
+	struct dpa_priv_s *priv;
+	struct qm_fd fd;
+	struct dpa_percpu_priv_s *percpu_priv;
+	struct rtnl_link_stats64 *percpu_stats;
+	int err = 0;
+	const int queue_mapping = dpa_get_queue_mapping(skb);
+	int *countptr, offset = 0;
+
+	priv = netdev_priv(net_dev);
+	/* Non-migratable context, safe to use raw_cpu_ptr */
+	percpu_priv = raw_cpu_ptr(priv->percpu_priv);
+	percpu_stats = &percpu_priv->stats;
+	countptr = raw_cpu_ptr(priv->dpa_bp->percpu_count);
+
+	clear_fd(&fd);
+
+	/* We're going to store the skb backpointer at the beginning
+	 * of the data buffer, so we need a privately owned skb
+	 *
+	 * We've made sure skb is not shared in dev->priv_flags,
+	 * we need to verify the skb head is not cloned
+	 */
+	if (skb_cow_head(skb, priv->tx_headroom))
+		goto enomem;
+
+	DPA_ERR_ON(skb_is_nonlinear(skb));
+
+	/* Finally, create a contig FD from this skb */
+	err = skb_to_contig_fd(priv, skb, &fd, countptr, &offset);
+	if (unlikely(err < 0))
+		goto skb_to_fd_failed;
+
+	if (likely(dpa_xmit(priv, percpu_stats, queue_mapping, &fd) == 0))
+		return NETDEV_TX_OK;
+
+	/* dpa_xmit failed */
+	if (fd.bpid != 0xff) {
+		(*countptr)--;
+		dpa_fd_release(net_dev, &fd);
+		percpu_stats->tx_errors++;
+		return NETDEV_TX_OK;
+	}
+	_dpa_cleanup_tx_fd(priv, &fd);
+skb_to_fd_failed:
+enomem:
+	percpu_stats->tx_errors++;
+	dev_kfree_skb(skb);
+	return NETDEV_TX_OK;
+}
-- 
1.7.11.7

_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/linuxppc-dev

Re: [net-next v4 2/8] dpaa_eth: add support for DPAA Ethernet

[Joe Perches]

On Mon, 2015-11-02 at 19:31 +0200, Madalin Bucur wrote:
> 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.
[]
> diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
[]
> +static void _dpa_rx_error(struct net_device *net_dev,
> +			  const struct dpa_priv_s *priv,
> +			  struct dpa_percpu_priv_s *percpu_priv,
> +			  const struct qm_fd *fd,
> +			  u32 fqid)
> +{
> +	/* limit common, possibly innocuous Rx FIFO Overflow errors'
> +	 * interference with zero-loss convergence benchmark results.
> +	 */
> +	if (likely(fd->status & FM_FD_ERR_PHYSICAL))
> +		pr_warn_once("non-zero error counters in fman statistics (sysfs)\n");
> +	else
> +		if (net_ratelimit())
> +			netif_err(priv, hw, net_dev, "Err FD status = 0x%08x\n",
> +				  fd->status & FM_FD_STAT_RX_ERRORS);

It's a bit of a pity the logging message code is
a mix of pr_<level>, dev_<level>, netdev_<level>
and netif_<level>

Perhaps netif_<foo>_ratelimited macros should be added.

Something like:

---
 include/linux/netdevice.h | 54 +++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 54 insertions(+)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 210d11a..555471d 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -4025,6 +4025,60 @@ do {								\
 })
 #endif
 
+#define netif_level_ratelimited(level, priv, type, dev, fmt, args...)	\
+do {									\
+	if (netif_msg_##type(priv) && net_ratelimit())			\
+		netdev_##level(dev, fmt, ##args);			\
+} while (0)
+
+#define netif_emerg_ratelimited(priv, type, dev, fmt, args...)		\
+	netif_level_ratelimited(emerg, priv, type, dev, fmt, ##args)
+#define netif_alert_ratelimited(priv, type, dev, fmt, args...)		\
+	netif_level_ratelimited(alert, priv, type, dev, fmt, ##args)
+#define netif_crit_ratelimited(priv, type, dev, fmt, args...)		\
+	netif_level_ratelimited(crit, priv, type, dev, fmt, ##args)
+#define netif_err_ratelimited(priv, type, dev, fmt, args...)		\
+	netif_level_ratelimited(err, priv, type, dev, fmt, ##args)
+#define netif_warn_ratelimited(priv, type, dev, fmt, args...)		\
+	netif_level_ratelimited(warn, priv, type, dev, fmt, ##args)
+#define netif_notice_ratelimited(priv, type, dev, fmt, args...)		\
+	netif_level_ratelimited(notice, priv, type, dev, fmt, ##args)
+#define netif_info_ratelimited(priv, type, dev, fmt, args...)		\
+	netif_level_ratelimited(info, priv, type, dev, fmt, ##args)
+
+#if defined(CONFIG_DYNAMIC_DEBUG)
+/* descriptor check is first to prevent flooding with "callbacks suppressed" */
+#define netif_dbg_ratelimited(priv, type, dev, fmt, args...)		\
+do {									\
+	DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt);			\
+	if (unlikely(descriptor.flags & _DPRINTK_FLAGS_PRINT) &&	\
+	    netif_msg_##type(priv) && net_ratelimit())			\
+		__dynamic_netdev_dbg(&descriptor, dev, fmt, ##args);	\
+} while (0)
+#elif defined(DEBUG)
+#define netif_dbg_ratelimited(priv, type, dev, fmt, args...)		\
+do {									\
+	if (netif_msg_##type(priv) && net_ratelimit())			\
+		netif_printk(priv, type, KERN_DEBUG, dev, fmt, ##args);	\
+} while (0)
+#else
+#define netif_dbg_ratelimited(priv, type, dev, fmt, args...)		\
+do {									\
+	if (0)								\
+		netif_printk(priv, type, KERN_DEBUG, dev, fmt, ##args);	\
+} while (0)
+#endif
+
+#if defined(VERBOSE_DEBUG)
+#define netif_vdbg_ratelimited	netif_dbg_ratelimited
+#else
+#define netif_vdbg(priv, type, dev, fmt, args...)			\
+do {									\
+	if (0)								\
+		netif_printk(priv, type, KERN_DEBUG, dev, fmt, ##args);	\
+} while (0)
+#endif
+
 /*
  *	The list of packet types we will receive (as opposed to discard)
  *	and the routines to invoke.

Re: [net-next v4 2/8] dpaa_eth: add support for DPAA Ethernet

[Joakim Tjernlund]

On Mon, 2015-11-02 at 19:31 +0200, Madalin Bucur wrote:
> +       if (unlikely(fd_status & FM_FD_STAT_RX_ERRORS) != 0) {
> +               if (net_ratelimit())
> +                       netif_warn(priv, hw, net_dev, "FD status = 0x%08x\n",
> +                                  fd_status & FM_FD_STAT_RX_ERRORS);
> +
> +               percpu_stats->rx_errors++;
> +               goto _release_frame;
> +       }

I cannot find any detailed error accounting(maybe I am not looking hard enough) but I
would appreciate if both TX and RX errors where better accounted(rx_length_errors, rx_frame_errors,
rx_crc_errors, rx_fifo_errors etc.). This has helped me many times in the past diagnosing
board HW problems.

 Jocke
_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/linuxppc-dev

RE: [net-next v4 2/8] dpaa_eth: add support for DPAA Ethernet

[Madalin-Cristian Bucur]

> -----Original Message-----
> From: Joakim Tjernlund [mailto:joakim.tjernlund@transmode.se]
> 
> On Mon, 2015-11-02 at 19:31 +0200, Madalin Bucur wrote:
> > +       if (unlikely(fd_status & FM_FD_STAT_RX_ERRORS) != 0) {
> > +               if (net_ratelimit())
> > +                       netif_warn(priv, hw, net_dev, "FD status =
> 0x%08x\n",
> > +                                  fd_status & FM_FD_STAT_RX_ERRORS);
> > +
> > +               percpu_stats->rx_errors++;
> > +               goto _release_frame;
> > +       }
> 
> I cannot find any detailed error accounting(maybe I am not looking hard
> enough) but I
> would appreciate if both TX and RX errors where better
> accounted(rx_length_errors, rx_frame_errors,
> rx_crc_errors, rx_fifo_errors etc.). This has helped me many times in the
> past diagnosing
> board HW problems.
> 
>  Jocke

Hi Jocke,

There are some error counters exported through ethtool (used to be debugfs).
FMan HW provides more debug information than we currently export, that will be
improved in the future but given the current priority of having a codebase as
small and reviewable as possible we had to drop some things from the initial
submission.

Madalin
_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/linuxppc-dev

Re: [net-next v4 2/8] dpaa_eth: add support for DPAA Ethernet

[Joakim Tjernlund]

On Tue, 2015-11-03 at 09:37 +0000, Madalin-Cristian Bucur wrote:
> > -----Original Message-----
> > From: Joakim Tjernlund [mailto:joakim.tjernlund@transmode.se]
> > 
> > On Mon, 2015-11-02 at 19:31 +0200, Madalin Bucur wrote:
> > > +       if (unlikely(fd_status & FM_FD_STAT_RX_ERRORS) != 0) {
> > > +               if (net_ratelimit())
> > > +                       netif_warn(priv, hw, net_dev, "FD status =
> > 0x%08x\n",
> > > +                                  fd_status & FM_FD_STAT_RX_ERRORS);
> > > +
> > > +               percpu_stats->rx_errors++;
> > > +               goto _release_frame;
> > > +       }
> > 
> > I cannot find any detailed error accounting(maybe I am not looking hard
> > enough) but I
> > would appreciate if both TX and RX errors where better
> > accounted(rx_length_errors, rx_frame_errors,
> > rx_crc_errors, rx_fifo_errors etc.). This has helped me many times in the
> > past diagnosing
> > board HW problems.
> > 
> >  Jocke
> 
> Hi Jocke,
> 
> There are some error counters exported through ethtool (used to be debugfs).
> FMan HW provides more debug information than we currently export, that will be
> improved in the future but given the current priority of having a codebase as
> small and reviewable as possible we had to drop some things from the initial
> submission.

I know, but ethtool is not always available.
Even the old fec_main.c has it:
 		if (status & (BD_ENET_RX_LG | BD_ENET_RX_SH | BD_ENET_RX_NO |
			   BD_ENET_RX_CR | BD_ENET_RX_OV)) {
			ndev->stats.rx_errors++;
			if (status & (BD_ENET_RX_LG | BD_ENET_RX_SH)) {
				/* Frame too long or too short. */
				ndev->stats.rx_length_errors++;
			}
			if (status & BD_ENET_RX_NO)	/* Frame alignment */
				ndev->stats.rx_frame_errors++;
			if (status & BD_ENET_RX_CR)	/* CRC Error */
				ndev->stats.rx_crc_errors++;
			if (status & BD_ENET_RX_OV)	/* FIFO overrun */
				ndev->stats.rx_fifo_errors++;
		}
so it is just a few more lines ... Pretty please ? :)

 Jocke

RE: [net-next v4 2/8] dpaa_eth: add support for DPAA Ethernet

[Madalin-Cristian Bucur]

> -----Original Message-----
> From: Joakim Tjernlund [mailto:joakim.tjernlund@transmode.se]
> 
> On Tue, 2015-11-03 at 09:37 +0000, Madalin-Cristian Bucur wrote:
> > > -----Original Message-----
> > > From: Joakim Tjernlund [mailto:joakim.tjernlund@transmode.se]
> > >
> > > On Mon, 2015-11-02 at 19:31 +0200, Madalin Bucur wrote:
> > > > +       if (unlikely(fd_status & FM_FD_STAT_RX_ERRORS) != 0) {
> > > > +               if (net_ratelimit())
> > > > +                       netif_warn(priv, hw, net_dev, "FD status =
> > > 0x%08x\n",
> > > > +                                  fd_status &
> FM_FD_STAT_RX_ERRORS);
> > > > +
> > > > +               percpu_stats->rx_errors++;
> > > > +               goto _release_frame;
> > > > +       }
> > >
> > > I cannot find any detailed error accounting(maybe I am not looking
> hard
> > > enough) but I
> > > would appreciate if both TX and RX errors where better
> > > accounted(rx_length_errors, rx_frame_errors,
> > > rx_crc_errors, rx_fifo_errors etc.). This has helped me many times in
> the
> > > past diagnosing
> > > board HW problems.
> > >
> > >  Jocke
> >
> > Hi Jocke,
> >
> > There are some error counters exported through ethtool (used to be
> debugfs).
> > FMan HW provides more debug information than we currently export, that
> will be
> > improved in the future but given the current priority of having a
> codebase as
> > small and reviewable as possible we had to drop some things from the
> initial
> > submission.
> 
> I know, but ethtool is not always available.
> Even the old fec_main.c has it:
>  		if (status & (BD_ENET_RX_LG | BD_ENET_RX_SH | BD_ENET_RX_NO |
> 			   BD_ENET_RX_CR | BD_ENET_RX_OV)) {
> 			ndev->stats.rx_errors++;
> 			if (status & (BD_ENET_RX_LG | BD_ENET_RX_SH)) {
> 				/* Frame too long or too short. */
> 				ndev->stats.rx_length_errors++;
> 			}
> 			if (status & BD_ENET_RX_NO)	/* Frame alignment */
> 				ndev->stats.rx_frame_errors++;
> 			if (status & BD_ENET_RX_CR)	/* CRC Error */
> 				ndev->stats.rx_crc_errors++;
> 			if (status & BD_ENET_RX_OV)	/* FIFO overrun */
> 				ndev->stats.rx_fifo_errors++;
> 		}
> so it is just a few more lines ... Pretty please ? :)
> 
>  Jocke

It may be more that just a few lines to add complete debug details.
Your request is noted and will be among the first features to work on
after the driver is accepted upstream.

Thanks,
Madalin

Re: [net-next v4 2/8] dpaa_eth: add support for DPAA Ethernet

[Scott Wood]

On Mon, Nov 02, 2015 at 07:31:34PM +0200, Madalin Bucur wrote:
> diff --git a/drivers/net/ethernet/freescale/dpaa/Makefile b/drivers/net/ethernet/freescale/dpaa/Makefile
> new file mode 100644
> index 0000000..3847ec7
> --- /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)

...or just put the two parts of the same driver into the same directory.

> +#define DPA_DESCRIPTION "FSL DPAA Ethernet driver"

Avoiding duplicating those four words is not worth the obfuscation of
moving it somewhere else.

> +static u8 debug = -1;

-1 does not fit in a u8.  Usually this is declared as int.

> +module_param(debug, byte, S_IRUGO);
> +MODULE_PARM_DESC(debug, "Module/Driver verbosity level");

It would be good to document the range of values here.

> +
> +/* This has to work in tandem with the DPA_CS_THRESHOLD_xxx values. */
> +static u16 tx_timeout = 1000;
> +module_param(tx_timeout, ushort, S_IRUGO);
> +MODULE_PARM_DESC(tx_timeout, "The Tx timeout in ms");

Could you elaborate on the relationship with DPA_CS_THRESHOLD_xxx?
Or even tell me where I can find such a symbol?

> +
> +/* BM */

BM?

> +#define DPAA_ETH_MAX_PAD (L1_CACHE_BYTES * 8)

This isn't used.

> +static u8 dpa_priv_common_bpid;
> +
> +static void _dpa_rx_error(struct net_device *net_dev,
> +			  const struct dpa_priv_s *priv,
> +			  struct dpa_percpu_priv_s *percpu_priv,
> +			  const struct qm_fd *fd,
> +			  u32 fqid)

Why the leading underscore?  Likewise elsewhere.

> +{
> +	/* limit common, possibly innocuous Rx FIFO Overflow errors'
> +	 * interference with zero-loss convergence benchmark results.
> +	 */

Spamming the console is bad regardless of whether you're running a
certain benchmark...

> +	if (likely(fd->status & FM_FD_ERR_PHYSICAL))
> +		pr_warn_once("non-zero error counters in fman statistics (sysfs)\n");
> +	else
> +		if (net_ratelimit())
> +			netif_err(priv, hw, net_dev, "Err FD status = 0x%08x\n",
> +				  fd->status & FM_FD_STAT_RX_ERRORS);

Why are you using likely in error paths?
Why do we need to print this error at all?  Do other net drivers do that?

> +	percpu_priv->stats.rx_errors++;
> +
> +	dpa_fd_release(net_dev, fd);

Can we reserve "fd" for file descriptors and call this something else?

> +}
> +
> +static void _dpa_tx_error(struct net_device *net_dev,
> +			  const struct dpa_priv_s *priv,
> +			  struct dpa_percpu_priv_s *percpu_priv,
> +			  const struct qm_fd *fd,
> +			  u32 fqid)
> +{
> +	struct sk_buff *skb;
> +
> +	if (net_ratelimit())
> +		netif_warn(priv, hw, net_dev, "FD status = 0x%08x\n",
> +			   fd->status & FM_FD_STAT_TX_ERRORS);
> +
> +	percpu_priv->stats.tx_errors++;
> +
> +	/* If we intended the buffers from this frame to go into the bpools
> +	 * when the FMan transmit was done, we need to put it in manually.
> +	 */
> +	if (fd->bpid != 0xff) {
> +		dpa_fd_release(net_dev, fd);
> +		return;
> +	}

Define a symbolic constant for this special 0xff value.

> +static enum qman_cb_dqrr_result
> +priv_rx_error_dqrr(struct qman_portal *portal,
> +		   struct qman_fq *fq,
> +		   const struct qm_dqrr_entry *dq)
> +{
> +	struct net_device *net_dev;
> +	struct dpa_priv_s *priv;
> +	struct dpa_percpu_priv_s *percpu_priv;
> +	int *count_ptr;
> +
> +	net_dev = ((struct dpa_fq *)fq)->net_dev;

Don't open-code the casting of one struct to another like this.  Use
container_of().

> +static enum qman_cb_dqrr_result
> +priv_rx_default_dqrr(struct qman_portal *portal,
> +		     struct qman_fq *fq,
> +		     const struct qm_dqrr_entry *dq)
> +{
> +	struct net_device *net_dev;
> +	struct dpa_priv_s *priv;
> +	struct dpa_percpu_priv_s *percpu_priv;
> +	int *count_ptr;
> +	struct dpa_bp *dpa_bp;
> +
> +	net_dev = ((struct dpa_fq *)fq)->net_dev;
> +	priv = netdev_priv(net_dev);
> +	dpa_bp = priv->dpa_bp;
> +
> +	/* IRQ handler, non-migratable; safe to use raw_cpu_ptr here */
> +	percpu_priv = raw_cpu_ptr(priv->percpu_priv);
> +	count_ptr = raw_cpu_ptr(dpa_bp->percpu_count);

But why do you *need* to use raw?

> +
> +	if (unlikely(dpaa_eth_napi_schedule(percpu_priv, portal)))
> +		return qman_cb_dqrr_stop;

If this is always an "IRQ handler" then why is it "unlikely" that this
will return non-zero?

> +static const struct dpa_fq_cbs_t private_fq_cbs = {
> +	.rx_defq = { .cb = { .dqrr = priv_rx_default_dqrr } },
> +	.tx_defq = { .cb = { .dqrr = priv_tx_conf_default_dqrr } },
> +	.rx_errq = { .cb = { .dqrr = priv_rx_error_dqrr } },
> +	.tx_errq = { .cb = { .dqrr = priv_tx_conf_error_dqrr } },
> +	.egress_ern = { .cb = { .ern = priv_ern } }
> +};
> +
> +static void dpaa_eth_napi_enable(struct dpa_priv_s *priv)
> +{
> +	struct dpa_percpu_priv_s *percpu_priv;
> +	int i, j;
> +
> +	for_each_possible_cpu(i) {
> +		percpu_priv = per_cpu_ptr(priv->percpu_priv, i);
> +
> +		for (j = 0; j < qman_portal_max; j++) {
> +			percpu_priv->np[j].down = 0;
> +			napi_enable(&percpu_priv->np[j].napi);
> +		}
> +	}
> +}
> +
> +static void dpaa_eth_napi_disable(struct dpa_priv_s *priv)
> +{
> +	struct dpa_percpu_priv_s *percpu_priv;
> +	int i, j;
> +
> +	for_each_possible_cpu(i) {
> +		percpu_priv = per_cpu_ptr(priv->percpu_priv, i);
> +
> +		for (j = 0; j < qman_portal_max; j++) {
> +			percpu_priv->np[j].down = 1;
> +			napi_disable(&percpu_priv->np[j].napi);
> +		}
> +	}
> +}

Why ss there a NAPI instance for every combination of portal and cpu,
even though a portal is normally bound to a single cpu?

> +static int
> +dpaa_eth_priv_probe(struct platform_device *pdev)
> +{

Why "priv"?  That makes sense when talking about the data structure, not
in function names.

Also no need for newline after "static int".

> +	int err = 0, i, channel;
> +	struct device *dev;
> +	struct dpa_bp *dpa_bp;
> +	struct dpa_fq *dpa_fq, *tmp;
> +	size_t count = 1;
> +	struct net_device *net_dev = NULL;
> +	struct dpa_priv_s *priv = NULL;
> +	struct dpa_percpu_priv_s *percpu_priv;

What is this "_s"?  If it means "_struct", drop it.

> +	struct fm_port_fqs port_fqs;
> +	struct dpa_buffer_layout_s *buf_layout = NULL;
> +	struct mac_device *mac_dev;
> +	struct task_struct *kth;
> +
> +	dev = &pdev->dev;
> +
> +	/* Get the buffer pool assigned to this interface;
> +	 * run only once the default pool probing code
> +	 */
> +	dpa_bp = (dpa_bpid2pool(dpa_priv_common_bpid)) ? :
> +			dpa_priv_bp_probe(dev);

This is an awkward/non-idiomatic way of saying:

        dpa_bp = dpa_bpid2pool(dpa_priv_common_bpid);
        if (!dpa_bp)
                dpa_bp = dpa_priv_bp_probe(dev);

> +	if (IS_ERR(dpa_bp))
> +		return PTR_ERR(dpa_bp);
> +
> +	/* Allocate this early, so we can store relevant information in
> +	 * the private area
> +	 */
> +	net_dev = alloc_etherdev_mq(sizeof(*priv), DPAA_ETH_TX_QUEUES);
> +	if (!net_dev) {
> +		dev_err(dev, "alloc_etherdev_mq() failed\n");
> +		goto alloc_etherdev_mq_failed;
> +	}
> +
> +#ifdef CONFIG_FSL_DPAA_ETH_FRIENDLY_IF_NAME
> +	snprintf(net_dev->name, IFNAMSIZ, "fm%d-mac%d",
> +		 dpa_mac_fman_index_get(pdev),
> +		 dpa_mac_hw_index_get(pdev));
> +#endif

Is this sort of in-kernel renaming considered acceptable?
Why is it a compile time option?

> +
> +	/* Do this here, so we can be verbose early */
> +	SET_NETDEV_DEV(net_dev, dev);
> +	dev_set_drvdata(dev, net_dev);
> +
> +	priv = netdev_priv(net_dev);
> +	priv->net_dev = net_dev;
> +
> +	priv->msg_enable = netif_msg_init(debug, -1);

This is the only driver that passes -1 as the second argument of
netif_msg_init().  Why?

> +
> +	mac_dev = dpa_mac_dev_get(pdev);
> +	if (IS_ERR(mac_dev) || !mac_dev) {
> +		err = PTR_ERR(mac_dev);
> +		goto mac_probe_failed;
> +	}

When can dpa_mac_dev_get() return NULL rather than an error value?

> +
> +	/* We have physical ports, so we need to establish
> +	 * the buffer layout.
> +	 */
> +	buf_layout = devm_kzalloc(dev, 2 * sizeof(*buf_layout),
> +				  GFP_KERNEL);
> +	if (!buf_layout)
> +		goto alloc_failed;

Why not just make buf_layout a static array in the priv struct?

> +
> +	dpa_set_buffers_layout(mac_dev, buf_layout);
> +
> +	/* For private ports, need to compute the size of the default
> +	 * buffer pool, based on FMan port buffer layout;also update
> +	 * the maximum buffer size for private ports if necessary
> +	 */
> +	dpa_bp->size = dpa_bp_size(&buf_layout[RX]);

What is a "private port", what is the opposite of a private port, and
what does this code do differently for one versus the other?

> +static int __init dpa_load(void)
> +{
> +	int err;
> +
> +	pr_info(DPA_DESCRIPTION "\n");
> +

Don't spam the console just because a driver has been loaded.  Wait until
a device has actually been found.

> +	/* initialise dpaa_eth mirror values */
> +	dpa_rx_extra_headroom = fman_get_rx_extra_headroom();
> +	dpa_max_frm = fman_get_max_frm();
> +
> +	err = platform_driver_register(&dpa_driver);
> +	if (err < 0)
> +		pr_err("Error, platform_driver_register() = %d\n", err);

A user seeing this would wonder what driver's registration failed.

It's a good habit to use __func__ on all output, especially if it's using
pr_xxx rather than dev_xxx.

> +MODULE_LICENSE("Dual BSD/GPL");
> +MODULE_AUTHOR("Andy Fleming <afleming@freescale.com>");
> +MODULE_DESCRIPTION(DPA_DESCRIPTION);

Andy hasn't touched this code in years, and lots of others have worked on
it.  I'm not sure that citing any individual person here makes sense.

> +#define dpa_get_rx_extra_headroom() dpa_rx_extra_headroom
> +#define dpa_get_max_frm() dpa_max_frm

Why?

> +#define DPA_ERR_ON(cond)

So all the DPA_ERR_ON() users are dead code?

> +/* This is what FMan is ever allowed to use.

-EPARSE

> + * 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 (can we?), so we reserve some more space
> + * for start-of-buffer alignment.
> + */
> +#define dpa_bp_size(buffer_layout)	(SKB_WITH_OVERHEAD(DPA_BP_RAW_SIZE) - \
> +						SMP_CACHE_BYTES)

The buffer_layout parameter is not used.

> +/* number of Tx queues to FMan */
> +#define DPAA_ETH_TX_QUEUES	NR_CPUS

It would be better to use nr_cpu_ids in the places where dynamic
allocations are possible (and consider reworking the places that use
static arrays to be dynamic).

> +static inline int dpaa_eth_napi_schedule(struct dpa_percpu_priv_s *percpu_priv,
> +					 struct qman_portal *portal)
> +{
> +	/* In case of threaded ISR for RT enable kernel,
> +	 * in_irq() does not return appropriate value, so use
> +	 * in_serving_softirq to distinguish softirq or irq context.
> +	 */
> +	if (unlikely(in_irq() || !in_serving_softirq())) {
> +		/* Disable QMan IRQ and invoke NAPI */
> +		int ret = qman_p_irqsource_remove(portal, QM_PIRQ_DQRI);
> +
> +		if (likely(!ret)) {
> +			const struct qman_portal_config *pc =
> +					qman_p_get_portal_config(portal);
> +			struct dpa_napi_portal *np =
> +					&percpu_priv->np[pc->channel];
> +
> +			np->p = portal;
> +			napi_schedule(&np->napi);
> +			return 1;
> +		}
> +	}
> +	return 0;
> +}

If nearly the entire function is "unlikely", why is anything other than
the if-statement inlined?

s/for RT enable kernel//

> +/* Use the queue selected by XPS */
> +#define dpa_get_queue_mapping(skb) \
> +	skb_get_queue_mapping(skb)

Why is this wrapper needed?

> +int dpa_remove(struct platform_device *pdev)
> +{
> +	int err;
> +	struct device *dev;
> +	struct net_device *net_dev;
> +	struct dpa_priv_s *priv;
> +
> +	dev = &pdev->dev;
> +	net_dev = dev_get_drvdata(dev);
> +
> +	priv = netdev_priv(net_dev);
> +
> +	dev_set_drvdata(dev, NULL);
> +	unregister_netdev(net_dev);
> +
> +	err = dpa_fq_free(dev, &priv->dpa_fq_list);
> +
> +	qman_delete_cgr_safe(&priv->ingress_cgr);
> +	qman_release_cgrid(priv->ingress_cgr.cgrid);
> +	qman_delete_cgr_safe(&priv->cgr_data.cgr);
> +	qman_release_cgrid(priv->cgr_data.cgr.cgrid);
> +
> +	dpa_private_napi_del(net_dev);
> +
> +	dpa_bp_free(priv);
> +
> +	if (priv->buf_layout)
> +		devm_kfree(dev, priv->buf_layout);
> +
> +	free_netdev(net_dev);
> +
> +	return err;
> +}

You don't need to check for NULL before calling kfree(), and you don't
need to call devm_kfree() in a .remove() function.

> +void dpa_set_buffers_layout(struct mac_device *mac_dev,
> +			    struct dpa_buffer_layout_s *layout)
> +{
> +	/* Rx */
> +	layout[RX].priv_data_size = (u16)DPA_RX_PRIV_DATA_SIZE;

Unnecessary cast.

> +	layout[RX].parse_results = true;
> +	layout[RX].hash_results = true;
> +	layout[RX].data_align = DPA_FD_DATA_ALIGNMENT;
> +
> +	/* Tx */
> +	layout[TX].priv_data_size = DPA_TX_PRIV_DATA_SIZE;

...and not even consistent.

> +	layout[TX].parse_results = true;
> +	layout[TX].hash_results = true;
> +	layout[TX].data_align = DPA_FD_DATA_ALIGNMENT;
> +}

When are parse_results/hash_results ever false?

If the answer is in a future patch, why is the mechanism being introduced
in this patch, resulting both in this patch being cluttered, and the
functionality introduced in the future patch being non-self-contained and
thus harder to review?

> +int dpa_fq_probe_mac(struct device *dev, struct list_head *list,
> +		     struct fm_port_fqs *port_fqs,
> +		     bool alloc_tx_conf_fqs,
> +		     enum port_type ptype)
> +{
> +	const struct fqid_cell *fqids;
> +	struct dpa_fq *dpa_fq;
> +	int num_ranges;
> +	int i;
> +
> +	if (ptype == TX && alloc_tx_conf_fqs) {
> +		if (!dpa_fq_alloc(dev, tx_confirm_fqids, list,
> +				  FQ_TYPE_TX_CONF_MQ))
> +			goto fq_alloc_failed;
> +	}
> +
> +	fqids = default_fqids[ptype];
> +	num_ranges = 3;
> +
> +	for (i = 0; i < num_ranges; i++) {
> +		switch (i) {
> +		case 0:
> +			/* The first queue is the error queue */
> +			if (fqids[i].count != 1)
> +				goto invalid_error_queue;
> +
> +			dpa_fq = dpa_fq_alloc(dev, &fqids[i], list,
> +					      ptype == RX ?
> +						FQ_TYPE_RX_ERROR :
> +						FQ_TYPE_TX_ERROR);
> +			if (!dpa_fq)
> +				goto fq_alloc_failed;
> +
> +			if (ptype == RX)
> +				port_fqs->rx_errq = &dpa_fq[0];
> +			else
> +				port_fqs->tx_errq = &dpa_fq[0];
> +			break;
> +		case 1:
> +			/* the second queue is the default queue */
> +			if (fqids[i].count != 1)
> +				goto invalid_default_queue;
> +
> +			dpa_fq = dpa_fq_alloc(dev, &fqids[i], list,
> +					      ptype == RX ?
> +						FQ_TYPE_RX_DEFAULT :
> +						FQ_TYPE_TX_CONFIRM);
> +			if (!dpa_fq)
> +				goto fq_alloc_failed;
> +
> +			if (ptype == RX)
> +				port_fqs->rx_defq = &dpa_fq[0];
> +			else
> +				port_fqs->tx_defq = &dpa_fq[0];
> +			break;
> +		default:
> +			/* all subsequent queues are Tx */
> +			if (!dpa_fq_alloc(dev, &fqids[i], list, FQ_TYPE_TX))
> +				goto fq_alloc_failed;
> +			break;

num_ranges can only be three, so there's only one "subsequent queue" and
this is overly complicated.  Even if eventually num_queues might be
greater than 3, there's no reason to shove the first two queues into a
loop only to use a switch statement to execute completely separate code
for them.

> +	/* Fill in the relevant L4 parse result fields */
> +	switch (l4_proto) {
> +	case IPPROTO_UDP:
> +		parse_result->l4r = FM_L4_PARSE_RESULT_UDP;
> +		break;
> +	case IPPROTO_TCP:
> +		parse_result->l4r = FM_L4_PARSE_RESULT_TCP;
> +		break;
> +	default:
> +		/* This can as well be a BUG() */

No, it can't.  I'm guessing the assumption is that for other protocols,
skb->ip_summed will != CHECKSUM_PARTIAL, but this should at worst be a
rate-limited WARN.   BUG() is for situations where there's no reasonable
path to error out.

> +/* Convenience macros for storing/retrieving the skb back-pointers.
> + *
> + * NB: @off is an offset from a (struct sk_buff **) pointer!
> + */
> +#define DPA_WRITE_SKB_PTR(skb, skbh, addr, off) \
> +	{ \
> +		skbh = (struct sk_buff **)addr; \
> +		*(skbh + (off)) = skb; \
> +	}
> +#define DPA_READ_SKB_PTR(skb, skbh, addr, off) \
> +	{ \
> +		skbh = (struct sk_buff **)addr; \
> +		skb = *(skbh + (off)); \
> +	}
> +

Why can these not be functions?

> +release_bufs:
> +	/* Release the buffers. In case bman is busy, keep trying
> +	 * until successful. bman_release() is guaranteed to succeed
> +	 * in a reasonable amount of time
> +	 */
> +	while (unlikely(bman_release(dpa_bp->pool, bmb, i, 0)))
> +		cpu_relax();

What is a "reasonable amount of time" and what happens if this guarantee
fails?  Usually we have timeouts on such waiting-on-hardware loops so   
that a device failure doesn't turn into a kernel failure.

> +	/* The only FD type that we may receive is contig */
> +	DPA_ERR_ON((fd->format != qm_fd_contig));

Unnecessary parens.

> +_release_frame:
> +	dpa_fd_release(net_dev, fd);
> +}

No leading underscores on goto labels.

> +
> +static int skb_to_contig_fd(struct dpa_priv_s *priv,
> +			    struct sk_buff *skb, struct qm_fd *fd,
> +			    int *count_ptr, int *offset)
> +{
> +	struct sk_buff **skbh;
> +	dma_addr_t addr;
> +	struct dpa_bp *dpa_bp = priv->dpa_bp;
> +	struct net_device *net_dev = priv->net_dev;
> +	int err;
> +	enum dma_data_direction dma_dir;
> +	unsigned char *buffer_start;
> +
> +	{
> +		/* We are guaranteed to have at least tx_headroom bytes
> +		 * available, so just use that for offset.
> +		 */
> +		fd->bpid = 0xff;
> +		buffer_start = skb->data - priv->tx_headroom;
> +		fd->offset = priv->tx_headroom;
> +		dma_dir = DMA_TO_DEVICE;
> +
> +		DPA_WRITE_SKB_PTR(skb, skbh, buffer_start, 0);
> +	}

Why is this block of code inside braces?

> +
> +	/* Enable L3/L4 hardware checksum computation.
> +	 *
> +	 * We must do this before dma_map_single(DMA_TO_DEVICE), because we may
> +	 * need to write into the skb.
> +	 */
> +	err = dpa_enable_tx_csum(priv, skb, fd,
> +				 ((char *)skbh) + DPA_TX_PRIV_DATA_SIZE);
> +	if (unlikely(err < 0)) {
> +		if (net_ratelimit())
> +			netif_err(priv, tx_err, net_dev, "HW csum error: %d\n",
> +				  err);
> +		return err;
> +	}
> +
> +	/* Fill in the rest of the FD fields */
> +	fd->format = qm_fd_contig;
> +	fd->length20 = skb->len;
> +	fd->cmd |= FM_FD_CMD_FCO;
> +
> +	/* Map the entire buffer size that may be seen by FMan, but no more */
> +	addr = dma_map_single(dpa_bp->dev, skbh,
> +			      skb_tail_pointer(skb) - buffer_start, dma_dir);
> +	if (unlikely(dma_mapping_error(dpa_bp->dev, addr))) {
> +		if (net_ratelimit())
> +			netif_err(priv, tx_err, net_dev, "dma_map_single() failed\n");
> +		return -EINVAL;
> +	}
> +	fd->addr_hi = (u8)upper_32_bits(addr);
> +	fd->addr_lo = lower_32_bits(addr);

Why is addr_hi limited to 8 bits?  E.g. t4240 has 40-bit physical
addresses...

-Scott

[net-next v4 3/8] dpaa_eth: add support for S/G frames

[Madalin Bucur]

Add support for Scater/Gather (S/G) frames. The FMan can place
the frame content into multiple buffers and provide a S/G Table
(SGT) into one first buffer with references to the others.

Signed-off-by: Madalin Bucur <madalin.bucur@freescale.com>
---
 drivers/net/ethernet/freescale/dpaa/dpaa_eth.c     |   6 +
 drivers/net/ethernet/freescale/dpaa/dpaa_eth.h     |   2 +-
 .../net/ethernet/freescale/dpaa/dpaa_eth_common.c  |  50 ++-
 .../net/ethernet/freescale/dpaa/dpaa_eth_common.h  |   2 +
 drivers/net/ethernet/freescale/dpaa/dpaa_eth_sg.c  | 335 +++++++++++++++++++--
 5 files changed, 374 insertions(+), 21 deletions(-)

diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
index 8381616..31d55b4 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
@@ -463,6 +463,12 @@ static int dpa_private_netdev_init(struct net_device *net_dev)
 	net_dev->hw_features |= (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
 				 NETIF_F_LLTX);
 
+	/* Advertise S/G and HIGHDMA support for private interfaces */
+	net_dev->hw_features |= NETIF_F_SG | NETIF_F_HIGHDMA;
+	/* Recent kernels enable GSO automatically, if
+	 * we declare NETIF_F_SG. For conformity, we'll
+	 * still declare GSO explicitly.
+	 */
 	net_dev->features |= NETIF_F_GSO;
 
 	return dpa_netdev_init(net_dev, mac_addr, tx_timeout);
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
index 1cc8682..1ba6617 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
@@ -347,7 +347,7 @@ static inline void clear_fd(struct qm_fd *fd)
 }
 
 static inline int _dpa_tx_fq_to_id(const struct dpa_priv_s *priv,
-				   struct qman_fq *tx_fq)
+		struct qman_fq *tx_fq)
 {
 	int i;
 
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c
index 963be4d8..b36cbca 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c
@@ -1177,10 +1177,42 @@ void dpaa_eth_init_ports(struct mac_device *mac_dev,
 			      port_fqs->rx_defq, &buf_layout[RX]);
 }
 
+void dpa_release_sgt(struct qm_sg_entry *sgt)
+{
+	struct dpa_bp *dpa_bp;
+	struct bm_buffer bmb[DPA_BUFF_RELEASE_MAX];
+	u8 i = 0, j;
+
+	memset(bmb, 0, sizeof(bmb));
+
+	do {
+		dpa_bp = dpa_bpid2pool(sgt[i].bpid);
+		DPA_ERR_ON(!dpa_bp);
+
+		j = 0;
+		do {
+			DPA_ERR_ON(sgt[i].extension);
+
+			bmb[j].hi = sgt[i].addr_hi;
+			bmb[j].lo = be32_to_cpu(sgt[i].addr_lo);
+
+			j++; i++;
+		} while (j < ARRAY_SIZE(bmb) &&
+				!sgt[i - 1].final &&
+				sgt[i - 1].bpid == sgt[i].bpid);
+
+		while (bman_release(dpa_bp->pool, bmb, j, 0))
+			cpu_relax();
+	} while (!sgt[i - 1].final);
+}
+
 void dpa_fd_release(const struct net_device *net_dev, const struct qm_fd *fd)
 {
+	struct qm_sg_entry *sgt;
 	struct dpa_bp *dpa_bp;
 	struct bm_buffer bmb;
+	dma_addr_t addr;
+	void *vaddr;
 
 	memset(&bmb, 0, sizeof(bmb));
 	bm_buffer_set64(&bmb, fd->addr);
@@ -1188,7 +1220,23 @@ void dpa_fd_release(const struct net_device *net_dev, const struct qm_fd *fd)
 	dpa_bp = dpa_bpid2pool(fd->bpid);
 	DPA_ERR_ON(!dpa_bp);
 
-	DPA_ERR_ON(fd->format == qm_fd_sg);
+	if (fd->format == qm_fd_sg) {
+		vaddr = phys_to_virt(fd->addr);
+		sgt = vaddr + dpa_fd_offset(fd);
+
+		dma_unmap_single(dpa_bp->dev, qm_fd_addr(fd), dpa_bp->size,
+				 DMA_BIDIRECTIONAL);
+
+		dpa_release_sgt(sgt);
+
+		addr = dma_map_single(dpa_bp->dev, vaddr, dpa_bp->size,
+				      DMA_BIDIRECTIONAL);
+		if (dma_mapping_error(dpa_bp->dev, addr)) {
+			dev_err(dpa_bp->dev, "DMA mapping failed");
+			return;
+		}
+		bm_buffer_set64(&bmb, addr);
+	}
 
 	while (bman_release(dpa_bp->pool, &bmb, 1, 0))
 		cpu_relax();
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h
index 68843c0..9df8f14 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h
@@ -37,6 +37,7 @@
 
 #include "dpaa_eth.h"
 
+#define DPA_SGT_MAX_ENTRIES 16 /* maximum number of entries in SG Table */
 #define DPA_BUFF_RELEASE_MAX 8 /* maximum number of buffers released at once */
 
 /* used in napi related functions */
@@ -90,6 +91,7 @@ void dpaa_eth_init_ports(struct mac_device *mac_dev,
 			 struct fm_port_fqs *port_fqs,
 			 struct dpa_buffer_layout_s *buf_layout,
 			 struct device *dev);
+void dpa_release_sgt(struct qm_sg_entry *sgt);
 void dpa_fd_release(const struct net_device *net_dev, const struct qm_fd *fd);
 int dpa_enable_tx_csum(struct dpa_priv_s *priv,
 		       struct sk_buff *skb,
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_sg.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_sg.c
index 115798b..af83ac2c 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_sg.c
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_sg.c
@@ -53,6 +53,31 @@
 		skb = *(skbh + (off)); \
 	}
 
+/* DMA map and add a page frag back into the bpool.
+ * @vaddr fragment must have been allocated with netdev_alloc_frag(),
+ * specifically for fitting into @dpa_bp.
+ */
+static void dpa_bp_recycle_frag(struct dpa_bp *dpa_bp, unsigned long vaddr,
+				int *count_ptr)
+{
+	struct bm_buffer bmb;
+	dma_addr_t addr;
+
+	addr = dma_map_single(dpa_bp->dev, (void *)vaddr, dpa_bp->size,
+			      DMA_BIDIRECTIONAL);
+	if (unlikely(dma_mapping_error(dpa_bp->dev, addr))) {
+		dev_err(dpa_bp->dev, "DMA mapping failed");
+		return;
+	}
+
+	bm_buffer_set64(&bmb, addr);
+
+	while (bman_release(dpa_bp->pool, &bmb, 1, 0))
+		cpu_relax();
+
+	(*count_ptr)++;
+}
+
 static int _dpa_bp_add_8_bufs(const struct dpa_bp *dpa_bp)
 {
 	struct bm_buffer bmb[8];
@@ -187,16 +212,49 @@ int dpaa_eth_refill_bpools(struct dpa_bp *dpa_bp, int *countptr)
 struct sk_buff *_dpa_cleanup_tx_fd(const struct dpa_priv_s *priv,
 				   const struct qm_fd *fd)
 {
+	const struct qm_sg_entry *sgt;
+	int i;
 	struct dpa_bp *dpa_bp = priv->dpa_bp;
 	dma_addr_t addr = qm_fd_addr(fd);
 	struct sk_buff **skbh;
 	struct sk_buff *skb = NULL;
 	const enum dma_data_direction dma_dir = DMA_TO_DEVICE;
+	int nr_frags;
+
 
 	/* retrieve skb back pointer */
 	DPA_READ_SKB_PTR(skb, skbh, phys_to_virt(addr), 0);
-	dma_unmap_single(dpa_bp->dev, addr,
-			 skb_tail_pointer(skb) - (u8 *)skbh, dma_dir);
+
+	if (unlikely(fd->format == qm_fd_sg)) {
+		nr_frags = skb_shinfo(skb)->nr_frags;
+		dma_unmap_single(dpa_bp->dev, addr, dpa_fd_offset(fd) +
+				 sizeof(struct qm_sg_entry) * (1 + nr_frags),
+				 dma_dir);
+
+		/* The sgt buffer has been allocated with netdev_alloc_frag(),
+		 * it's from lowmem.
+		 */
+		sgt = phys_to_virt(addr + dpa_fd_offset(fd));
+
+		/* sgt[0] is from lowmem, was dma_map_single()-ed */
+		dma_unmap_single(dpa_bp->dev, (dma_addr_t)sgt[0].addr,
+				 sgt[0].length, dma_dir);
+
+		/* remaining pages were mapped with dma_map_page() */
+		for (i = 1; i < nr_frags; i++) {
+			DPA_ERR_ON(sgt[i].extension);
+
+			dma_unmap_page(dpa_bp->dev, (dma_addr_t)sgt[i].addr,
+				       sgt[i].length, dma_dir);
+		}
+
+		/* Free the page frag that we allocated on Tx */
+		put_page(virt_to_head_page(sgt));
+	} else {
+		dma_unmap_single(dpa_bp->dev, addr,
+				 skb_tail_pointer(skb) - (u8 *)skbh, dma_dir);
+	}
+
 	return skb;
 }
 
@@ -230,6 +288,107 @@ static struct sk_buff *contig_fd_to_skb(const struct dpa_priv_s *priv,
 	return skb;
 }
 
+/* Build an skb with the data of the first S/G entry in the linear portion and
+ * the rest of the frame as skb fragments.
+ *
+ * The page fragment holding the S/G Table is recycled here.
+ */
+static struct sk_buff *sg_fd_to_skb(const struct dpa_priv_s *priv,
+				    const struct qm_fd *fd,
+				    int *count_ptr)
+{
+	const struct qm_sg_entry *sgt;
+	dma_addr_t addr = qm_fd_addr(fd);
+	ssize_t fd_off = dpa_fd_offset(fd);
+	dma_addr_t sg_addr;
+	void *vaddr, *sg_vaddr;
+	struct dpa_bp *dpa_bp;
+	struct page *page, *head_page;
+	int frag_offset, frag_len;
+	int page_offset;
+	int i;
+	struct sk_buff *skb = NULL, *skb_tmp, **skbh;
+
+	vaddr = phys_to_virt(addr);
+	DPA_ERR_ON(!IS_ALIGNED((unsigned long)vaddr, SMP_CACHE_BYTES));
+
+	dpa_bp = priv->dpa_bp;
+	/* Iterate through the SGT entries and add data buffers to the skb */
+	sgt = vaddr + fd_off;
+	for (i = 0; i < DPA_SGT_MAX_ENTRIES; i++) {
+		/* Extension bit is not supported */
+		DPA_ERR_ON(sgt[i].extension);
+
+		/* We use a single global Rx pool */
+		DPA_ERR_ON(dpa_bp != dpa_bpid2pool(sgt[i].bpid));
+
+		sg_addr = qm_sg_addr(&sgt[i]);
+		sg_vaddr = phys_to_virt(sg_addr);
+		DPA_ERR_ON(!IS_ALIGNED((unsigned long)sg_vaddr,
+				       SMP_CACHE_BYTES));
+
+		dma_unmap_single(dpa_bp->dev, sg_addr, dpa_bp->size,
+				 DMA_BIDIRECTIONAL);
+		if (i == 0) {
+			DPA_READ_SKB_PTR(skb, skbh, sg_vaddr, -1);
+			DPA_ERR_ON(skb->head != sg_vaddr);
+
+			skb->ip_summed = CHECKSUM_NONE;
+
+			/* Make sure forwarded skbs will have enough space
+			 * on Tx, if extra headers are added.
+			 */
+			DPA_ERR_ON(fd_off != priv->rx_headroom);
+			skb_reserve(skb, fd_off);
+			skb_put(skb, sgt[i].length);
+		} else {
+			/* Not the first S/G entry; all data from buffer will
+			 * be added in an skb fragment; fragment index is offset
+			 * by one since first S/G entry was incorporated in the
+			 * linear part of the skb.
+			 *
+			 * Caution: 'page' may be a tail page.
+			 */
+			DPA_READ_SKB_PTR(skb_tmp, skbh, sg_vaddr, -1);
+			page = virt_to_page(sg_vaddr);
+			head_page = virt_to_head_page(sg_vaddr);
+
+			/* Free (only) the skbuff shell because its data buffer
+			 * is already a frag in the main skb.
+			 */
+			get_page(head_page);
+			dev_kfree_skb(skb_tmp);
+
+			/* Compute offset in (possibly tail) page */
+			page_offset = ((unsigned long)sg_vaddr &
+					(PAGE_SIZE - 1)) +
+				(page_address(page) - page_address(head_page));
+			/* page_offset only refers to the beginning of sgt[i];
+			 * but the buffer itself may have an internal offset.
+			 */
+			frag_offset = sgt[i].offset + page_offset;
+			frag_len = sgt[i].length;
+			/* skb_add_rx_frag() does no checking on the page; if
+			 * we pass it a tail page, we'll end up with
+			 * bad page accounting and eventually with segafults.
+			 */
+			skb_add_rx_frag(skb, i - 1, head_page, frag_offset,
+					frag_len, dpa_bp->size);
+		}
+		/* Update the pool count for the current {cpu x bpool} */
+		(*count_ptr)--;
+
+		if (sgt[i].final)
+			break;
+	}
+	WARN_ONCE(i == DPA_SGT_MAX_ENTRIES, "No final bit on SGT\n");
+
+	/* recycle the SGT fragment */
+	DPA_ERR_ON(dpa_bp != dpa_bpid2pool(fd->bpid));
+	dpa_bp_recycle_frag(dpa_bp, (unsigned long)vaddr, count_ptr);
+	return skb;
+}
+
 void _dpa_rx(struct net_device *net_dev,
 	     struct qman_portal *portal,
 	     const struct dpa_priv_s *priv,
@@ -257,17 +416,20 @@ void _dpa_rx(struct net_device *net_dev,
 	dpa_bp = priv->dpa_bp;
 	DPA_ERR_ON(dpa_bp != dpa_bpid2pool(fd->bpid));
 
-	/* prefetch the first 64 bytes of the frame */
+	/* prefetch the first 64 bytes of the frame or the SGT start */
 	dma_unmap_single(dpa_bp->dev, addr, dpa_bp->size, DMA_BIDIRECTIONAL);
 	prefetch(phys_to_virt(addr) + dpa_fd_offset(fd));
 
-	/* The only FD type that we may receive is contig */
-	DPA_ERR_ON((fd->format != qm_fd_contig));
+	/* The only FD types that we may receive are contig and S/G */
+	DPA_ERR_ON((fd->format != qm_fd_contig) && (fd->format != qm_fd_sg));
 
-	skb = contig_fd_to_skb(priv, fd);
+	if (likely(fd->format == qm_fd_contig))
+		skb = contig_fd_to_skb(priv, fd);
+	else
+		skb = sg_fd_to_skb(priv, fd, count_ptr);
 
-	/* Account for the contig buffer
-	 * having been removed from the pool.
+	/* Account for either the contig buffer or the SGT buffer (depending on
+	 * which case we were in) having been removed from the pool.
 	 */
 	(*count_ptr)--;
 	skb->protocol = eth_type_trans(skb, net_dev);
@@ -355,6 +517,121 @@ static int skb_to_contig_fd(struct dpa_priv_s *priv,
 	return 0;
 }
 
+static int skb_to_sg_fd(struct dpa_priv_s *priv,
+			struct sk_buff *skb, struct qm_fd *fd)
+{
+	struct dpa_bp *dpa_bp = priv->dpa_bp;
+	dma_addr_t addr;
+	struct sk_buff **skbh;
+	struct net_device *net_dev = priv->net_dev;
+	int err;
+
+	struct qm_sg_entry *sgt;
+	void *sgt_buf;
+	void *buffer_start;
+	skb_frag_t *frag;
+	int i, j;
+	const enum dma_data_direction dma_dir = DMA_TO_DEVICE;
+	const int nr_frags = skb_shinfo(skb)->nr_frags;
+
+	fd->format = qm_fd_sg;
+
+	/* get a page frag to store the SGTable */
+	sgt_buf = netdev_alloc_frag(priv->tx_headroom +
+		sizeof(struct qm_sg_entry) * (1 + nr_frags));
+	if (unlikely(!sgt_buf)) {
+		netdev_err(net_dev, "netdev_alloc_frag() failed\n");
+		return -ENOMEM;
+	}
+
+	/* Enable L3/L4 hardware checksum computation.
+	 *
+	 * We must do this before dma_map_single(DMA_TO_DEVICE), because we may
+	 * need to write into the skb.
+	 */
+	err = dpa_enable_tx_csum(priv, skb, fd,
+				 sgt_buf + DPA_TX_PRIV_DATA_SIZE);
+	if (unlikely(err < 0)) {
+		if (net_ratelimit())
+			netif_err(priv, tx_err, net_dev, "HW csum error: %d\n",
+				  err);
+		goto csum_failed;
+	}
+
+	sgt = (struct qm_sg_entry *)(sgt_buf + priv->tx_headroom);
+	sgt[0].bpid = 0xff;
+	sgt[0].offset = 0;
+	sgt[0].length = cpu_to_be32(skb_headlen(skb));
+	sgt[0].extension = 0;
+	sgt[0].final = 0;
+	addr = dma_map_single(dpa_bp->dev, skb->data, sgt[0].length, dma_dir);
+	if (unlikely(dma_mapping_error(dpa_bp->dev, addr))) {
+		dev_err(dpa_bp->dev, "DMA mapping failed");
+		err = -EINVAL;
+		goto sg0_map_failed;
+	}
+	sgt[0].addr_hi = (u8)upper_32_bits(addr);
+	sgt[0].addr_lo = cpu_to_be32(lower_32_bits(addr));
+
+	/* populate the rest of SGT entries */
+	for (i = 1; i <= nr_frags; i++) {
+		frag = &skb_shinfo(skb)->frags[i - 1];
+		sgt[i].bpid = 0xff;
+		sgt[i].offset = 0;
+		sgt[i].length = cpu_to_be32(frag->size);
+		sgt[i].extension = 0;
+		sgt[i].final = 0;
+
+		DPA_ERR_ON(!skb_frag_page(frag));
+		addr = skb_frag_dma_map(dpa_bp->dev, frag, 0, sgt[i].length,
+					dma_dir);
+		if (unlikely(dma_mapping_error(dpa_bp->dev, addr))) {
+			dev_err(dpa_bp->dev, "DMA mapping failed");
+			err = -EINVAL;
+			goto sg_map_failed;
+		}
+
+		/* keep the offset in the address */
+		sgt[i].addr_hi = (u8)upper_32_bits(addr);
+		sgt[i].addr_lo = cpu_to_be32(lower_32_bits(addr));
+	}
+	sgt[i - 1].final = 1;
+
+	fd->length20 = skb->len;
+	fd->offset = priv->tx_headroom;
+
+	/* DMA map the SGT page */
+	buffer_start = (void *)sgt - priv->tx_headroom;
+	DPA_WRITE_SKB_PTR(skb, skbh, buffer_start, 0);
+
+	addr = dma_map_single(dpa_bp->dev, buffer_start, priv->tx_headroom +
+			      sizeof(struct qm_sg_entry) * (1 + nr_frags),
+			      dma_dir);
+	if (unlikely(dma_mapping_error(dpa_bp->dev, addr))) {
+		dev_err(dpa_bp->dev, "DMA mapping failed");
+		err = -EINVAL;
+		goto sgt_map_failed;
+	}
+
+	fd->bpid = 0xff;
+	fd->cmd |= FM_FD_CMD_FCO;
+	fd->addr_hi = (u8)upper_32_bits(addr);
+	fd->addr_lo = lower_32_bits(addr);
+
+	return 0;
+
+sgt_map_failed:
+sg_map_failed:
+	for (j = 0; j < i; j++)
+		dma_unmap_page(dpa_bp->dev, qm_sg_addr(&sgt[j]),
+			       cpu_to_be32(sgt[j].length), dma_dir);
+sg0_map_failed:
+csum_failed:
+	put_page(virt_to_head_page(sgt_buf));
+
+	return err;
+}
+
 int dpa_tx(struct sk_buff *skb, struct net_device *net_dev)
 {
 	struct dpa_priv_s *priv;
@@ -363,6 +640,7 @@ int dpa_tx(struct sk_buff *skb, struct net_device *net_dev)
 	struct rtnl_link_stats64 *percpu_stats;
 	int err = 0;
 	const int queue_mapping = dpa_get_queue_mapping(skb);
+	bool nonlinear = skb_is_nonlinear(skb);
 	int *countptr, offset = 0;
 
 	priv = netdev_priv(net_dev);
@@ -373,19 +651,38 @@ int dpa_tx(struct sk_buff *skb, struct net_device *net_dev)
 
 	clear_fd(&fd);
 
-	/* We're going to store the skb backpointer at the beginning
-	 * of the data buffer, so we need a privately owned skb
-	 *
-	 * We've made sure skb is not shared in dev->priv_flags,
-	 * we need to verify the skb head is not cloned
-	 */
-	if (skb_cow_head(skb, priv->tx_headroom))
-		goto enomem;
+	if (!nonlinear) {
+		/* We're going to store the skb backpointer at the beginning
+		 * of the data buffer, so we need a privately owned skb
+		 *
+		 * We've made sure skb is not shared in dev->priv_flags,
+		 * we need to verify the skb head is not cloned
+		 */
+		if (skb_cow_head(skb, priv->tx_headroom))
+			goto enomem;
+
+		WARN_ON(skb_is_nonlinear(skb));
+	}
 
-	DPA_ERR_ON(skb_is_nonlinear(skb));
+	/* MAX_SKB_FRAGS is equal or larger than our DPA_SGT_MAX_ENTRIES;
+	 * make sure we don't feed FMan with more fragments than it supports.
+	 * Btw, we're using the first sgt entry to store the linear part of
+	 * the skb, so we're one extra frag short.
+	 */
+	if (nonlinear &&
+	    likely(skb_shinfo(skb)->nr_frags < DPA_SGT_MAX_ENTRIES)) {
+		/* Just create a S/G fd based on the skb */
+		err = skb_to_sg_fd(priv, skb, &fd);
+	} else {
+		/* If the egress skb contains more fragments than we support
+		 * we have no choice but to linearize it ourselves.
+		 */
+		if (unlikely(nonlinear) && __skb_linearize(skb))
+			goto enomem;
 
-	/* Finally, create a contig FD from this skb */
-	err = skb_to_contig_fd(priv, skb, &fd, countptr, &offset);
+		/* Finally, create a contig FD from this skb */
+		err = skb_to_contig_fd(priv, skb, &fd, countptr, &offset);
+	}
 	if (unlikely(err < 0))
 		goto skb_to_fd_failed;
 
-- 
1.7.11.7

Re: [net-next v4 3/8] dpaa_eth: add support for S/G frames

[Joe Perches]

On Mon, 2015-11-02 at 19:31 +0200, Madalin Bucur wrote:
> Add support for Scater/Gather (S/G) frames. The FMan can place
> the frame content into multiple buffers and provide a S/G Table
> (SGT) into one first buffer with references to the others.

trivia: scatter

> diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c
[]
> @@ -1177,10 +1177,42 @@ void dpaa_eth_init_ports(struct mac_device *mac_dev,
>  			      port_fqs->rx_defq, &buf_layout[RX]);
>  }
>  
> +void dpa_release_sgt(struct qm_sg_entry *sgt)
> +{
> +	struct dpa_bp *dpa_bp;
> +	struct bm_buffer bmb[DPA_BUFF_RELEASE_MAX];

Where is "struct bm_buffer" defined?

RE: [net-next v4 3/8] dpaa_eth: add support for S/G frames

[Madalin-Cristian Bucur]

> -----Original Message-----
> From: Joe Perches [mailto:joe@perches.com]
> 
> On Mon, 2015-11-02 at 19:31 +0200, Madalin Bucur wrote:
> > Add support for Scater/Gather (S/G) frames. The FMan can place
> > the frame content into multiple buffers and provide a S/G Table
> > (SGT) into one first buffer with references to the others.
> 
> trivia: scatter
> 
> > diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c
> b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c
> []
> > @@ -1177,10 +1177,42 @@ void dpaa_eth_init_ports(struct mac_device
> *mac_dev,
> >  			      port_fqs->rx_defq, &buf_layout[RX]);
> >  }
> >
> > +void dpa_release_sgt(struct qm_sg_entry *sgt)
> > +{
> > +	struct dpa_bp *dpa_bp;
> > +	struct bm_buffer bmb[DPA_BUFF_RELEASE_MAX];
> 
> Where is "struct bm_buffer" defined?
> 

Thank you, I'll address this and the other observations you have sent.
The struct bm_buffer is defined in the Buffer Manager driver header file,
in include/soc/fsl/bman.h.

Madalin
_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/linuxppc-dev

Re: [net-next v4 3/8] dpaa_eth: add support for S/G frames

[Joe Perches]

On Mon, 2015-11-02 at 19:31 +0200, Madalin Bucur wrote:
> Add support for Scater/Gather (S/G) frames. The FMan can place
> the frame content into multiple buffers and provide a S/G Table
> (SGT) into one first buffer with references to the others.

trivia:

> diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
[]
> @@ -347,7 +347,7 @@ static inline void clear_fd(struct qm_fd *fd)
>  }
>  
>  static inline int _dpa_tx_fq_to_id(const struct dpa_priv_s *priv,
> -				   struct qman_fq *tx_fq)
> +		struct qman_fq *tx_fq)

superfluous change?

> +void dpa_release_sgt(struct qm_sg_entry *sgt)
> +{
> +	struct dpa_bp *dpa_bp;
> +	struct bm_buffer bmb[DPA_BUFF_RELEASE_MAX];
> +	u8 i = 0, j;

Using int may be better than u8 for indexing

[net-next v4 4/8] dpaa_eth: add driver's Tx queue selection

[Madalin Bucur]

Allow the selection of the transmission queue based on the CPU id.

Signed-off-by: Madalin Bucur <madalin.bucur@freescale.com>
---
 drivers/net/ethernet/freescale/dpaa/Kconfig           | 10 ++++++++++
 drivers/net/ethernet/freescale/dpaa/dpaa_eth.c        |  3 +++
 drivers/net/ethernet/freescale/dpaa/dpaa_eth.h        |  6 ++++++
 drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c |  8 ++++++++
 drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h |  4 ++++
 5 files changed, 31 insertions(+)

diff --git a/drivers/net/ethernet/freescale/dpaa/Kconfig b/drivers/net/ethernet/freescale/dpaa/Kconfig
index 022d5aa..2577aac 100644
--- a/drivers/net/ethernet/freescale/dpaa/Kconfig
+++ b/drivers/net/ethernet/freescale/dpaa/Kconfig
@@ -11,6 +11,16 @@ menuconfig FSL_DPAA_ETH
 
 if FSL_DPAA_ETH
 
+config FSL_DPAA_ETH_USE_NDO_SELECT_QUEUE
+	bool "Use driver's Tx queue selection mechanism"
+	default y
+	---help---
+	  The DPAA Ethernet driver defines a ndo_select_queue() callback for optimal selection
+	  of the egress FQ. That will override the XPS support for this netdevice.
+	  If for whatever reason you want to be in control of the egress FQ-to-CPU selection and mapping,
+	  or simply don't want to use the driver's ndo_select_queue() callback, then unselect this
+	  and use the standard XPS support instead.
+
 config FSL_DPAA_ETH_FRIENDLY_IF_NAME
 	bool "Use fmX-macY names for the DPAA interfaces"
 	default y
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
index 31d55b4..894f1a7 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
@@ -390,6 +390,9 @@ static const struct net_device_ops dpa_private_ops = {
 	.ndo_get_stats64 = dpa_get_stats64,
 	.ndo_set_mac_address = dpa_set_mac_address,
 	.ndo_validate_addr = eth_validate_addr,
+#ifdef CONFIG_FSL_DPAA_ETH_USE_NDO_SELECT_QUEUE
+	.ndo_select_queue = dpa_select_queue,
+#endif
 	.ndo_change_mtu = dpa_change_mtu,
 	.ndo_set_rx_mode = dpa_set_rx_mode,
 	.ndo_init = dpa_ndo_init,
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
index 1ba6617..87577cf 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
@@ -420,9 +420,15 @@ static inline void _dpa_assign_wq(struct dpa_fq *fq)
 	}
 }
 
+#ifdef CONFIG_FSL_DPAA_ETH_USE_NDO_SELECT_QUEUE
+/* Use in lieu of skb_get_queue_mapping() */
+#define dpa_get_queue_mapping(skb) \
+	raw_smp_processor_id()
+#else
 /* Use the queue selected by XPS */
 #define dpa_get_queue_mapping(skb) \
 	skb_get_queue_mapping(skb)
+#endif
 
 static inline void _dpa_bp_free_pf(void *addr)
 {
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c
index b36cbca..89f3b1f 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c
@@ -593,6 +593,14 @@ bool dpa_bpid2pool_use(int bpid)
 	return false;
 }
 
+#ifdef CONFIG_FSL_DPAA_ETH_USE_NDO_SELECT_QUEUE
+u16 dpa_select_queue(struct net_device *net_dev, struct sk_buff *skb,
+		     void *accel_priv, select_queue_fallback_t fallback)
+{
+	return dpa_get_queue_mapping(skb);
+}
+#endif
+
 struct dpa_fq *dpa_fq_alloc(struct device *dev,
 			    const struct fqid_cell *fqids,
 			    struct list_head *list,
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h
index 9df8f14..2e9471d 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h
@@ -70,6 +70,10 @@ struct dpa_bp *dpa_bpid2pool(int bpid);
 void dpa_bpid2pool_map(int bpid, struct dpa_bp *dpa_bp);
 bool dpa_bpid2pool_use(int bpid);
 void dpa_bp_drain(struct dpa_bp *bp);
+#ifdef CONFIG_FSL_DPAA_ETH_USE_NDO_SELECT_QUEUE
+u16 dpa_select_queue(struct net_device *net_dev, struct sk_buff *skb,
+		     void *accel_priv, select_queue_fallback_t fallback);
+#endif
 struct dpa_fq *dpa_fq_alloc(struct device *dev,
 			    const struct fqid_cell *fqids,
 			    struct list_head *list,
-- 
1.7.11.7

Re: [net-next v4 4/8] dpaa_eth: add driver's Tx queue selection

[Scott Wood]

On Mon, 2015-11-02 at 19:31 +0200, Madalin Bucur wrote:
> Allow the selection of the transmission queue based on the CPU id.

Explain why.

> 
> Signed-off-by: Madalin Bucur <madalin.bucur@freescale.com>
> ---
>  drivers/net/ethernet/freescale/dpaa/Kconfig           | 10 ++++++++++
>  drivers/net/ethernet/freescale/dpaa/dpaa_eth.c        |  3 +++
>  drivers/net/ethernet/freescale/dpaa/dpaa_eth.h        |  6 ++++++
>  drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c |  8 ++++++++
>  drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h |  4 ++++
>  5 files changed, 31 insertions(+)
> 
> diff --git a/drivers/net/ethernet/freescale/dpaa/Kconfig
> b/drivers/net/ethernet/freescale/dpaa/Kconfig
> index 022d5aa..2577aac 100644
> --- a/drivers/net/ethernet/freescale/dpaa/Kconfig
> +++ b/drivers/net/ethernet/freescale/dpaa/Kconfig
> @@ -11,6 +11,16 @@ menuconfig FSL_DPAA_ETH
>  
>  if FSL_DPAA_ETH
>  
> +config FSL_DPAA_ETH_USE_NDO_SELECT_QUEUE
> +	bool "Use driver's Tx queue selection mechanism"
> +	default y
> +	---help---
> +	  The DPAA Ethernet driver defines a ndo_select_queue() callback
> for optimal selection
> +	  of the egress FQ. That will override the XPS support for this
> netdevice.
> +	  If for whatever reason you want to be in control of the egress FQ
> -to-CPU selection and mapping,
> +	  or simply don't want to use the driver's ndo_select_queue()
> callback, then unselect this
> +	  and use the standard XPS support instead.

Is there a use case for needing this to be configurable?

> diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
> b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
> index 31d55b4..894f1a7 100644
> --- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
> +++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
> @@ -390,6 +390,9 @@ static const struct net_device_ops dpa_private_ops = {
>  	.ndo_get_stats64 = dpa_get_stats64,
>  	.ndo_set_mac_address = dpa_set_mac_address,
>  	.ndo_validate_addr = eth_validate_addr,
> +#ifdef CONFIG_FSL_DPAA_ETH_USE_NDO_SELECT_QUEUE
> +	.ndo_select_queue = dpa_select_queue,
> +#endif
>  	.ndo_change_mtu = dpa_change_mtu,
>  	.ndo_set_rx_mode = dpa_set_rx_mode,
>  	.ndo_init = dpa_ndo_init,
> diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
> b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
> index 1ba6617..87577cf 100644
> --- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
> +++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
> @@ -420,9 +420,15 @@ static inline void _dpa_assign_wq(struct dpa_fq *fq)
>  	}
>  }
>  
> +#ifdef CONFIG_FSL_DPAA_ETH_USE_NDO_SELECT_QUEUE
> +/* Use in lieu of skb_get_queue_mapping() */
> +#define dpa_get_queue_mapping(skb) \
> +	raw_smp_processor_id()
> +#else
>  /* Use the queue selected by XPS */
>  #define dpa_get_queue_mapping(skb) \
>  	skb_get_queue_mapping(skb)
> +#endif

Why is this necessary?  Shouldn't providing a custom .ndo_select_queue() be
sufficient to ensure that skb_get_queue_mapping() returns the same thing?

And if this goes away, it's just a matter of a function pointer, so if it does
need to be configurable it could be a runtime option.

-Scott

_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/linuxppc-dev

RE: [net-next v4 4/8] dpaa_eth: add driver's Tx queue selection

[Madalin-Cristian Bucur]

> -----Original Message-----
> From: Wood Scott-B07421
> Sent: Wednesday, December 02, 2015 11:41 PM
> 
> On Mon, 2015-11-02 at 19:31 +0200, Madalin Bucur wrote:
> > Allow the selection of the transmission queue based on the CPU id.
> 
> Explain why.

I'll add more details in the commit log. This is a customer generated
feature. Bypassing the standard XPS can increase performance by making use
of the DPAA HW particularities.

> >
> > Signed-off-by: Madalin Bucur <madalin.bucur@freescale.com>
> > ---
> >  drivers/net/ethernet/freescale/dpaa/Kconfig           | 10 ++++++++++
> >  drivers/net/ethernet/freescale/dpaa/dpaa_eth.c        |  3 +++
> >  drivers/net/ethernet/freescale/dpaa/dpaa_eth.h        |  6 ++++++
> >  drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c |  8 ++++++++
> >  drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h |  4 ++++
> >  5 files changed, 31 insertions(+)
> >
> > diff --git a/drivers/net/ethernet/freescale/dpaa/Kconfig
> > b/drivers/net/ethernet/freescale/dpaa/Kconfig
> > index 022d5aa..2577aac 100644
> > --- a/drivers/net/ethernet/freescale/dpaa/Kconfig
> > +++ b/drivers/net/ethernet/freescale/dpaa/Kconfig
> > @@ -11,6 +11,16 @@ menuconfig FSL_DPAA_ETH
> >
> >  if FSL_DPAA_ETH
> >
> > +config FSL_DPAA_ETH_USE_NDO_SELECT_QUEUE
> > +	bool "Use driver's Tx queue selection mechanism"
> > +	default y
> > +	---help---
> > +	  The DPAA Ethernet driver defines a ndo_select_queue() callback
> > for optimal selection
> > +	  of the egress FQ. That will override the XPS support for this
> > netdevice.
> > +	  If for whatever reason you want to be in control of the egress FQ
> > -to-CPU selection and mapping,
> > +	  or simply don't want to use the driver's ndo_select_queue()
> > callback, then unselect this
> > +	  and use the standard XPS support instead.
> 
> Is there a use case for needing this to be configurable?

If the standard XPS is desired, the Kconfig option allows the driver user to
select that.

> > diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
> > b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
> > index 31d55b4..894f1a7 100644
> > --- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
> > +++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
> > @@ -390,6 +390,9 @@ static const struct net_device_ops dpa_private_ops =
> {
> >  	.ndo_get_stats64 = dpa_get_stats64,
> >  	.ndo_set_mac_address = dpa_set_mac_address,
> >  	.ndo_validate_addr = eth_validate_addr,
> > +#ifdef CONFIG_FSL_DPAA_ETH_USE_NDO_SELECT_QUEUE
> > +	.ndo_select_queue = dpa_select_queue,
> > +#endif
> >  	.ndo_change_mtu = dpa_change_mtu,
> >  	.ndo_set_rx_mode = dpa_set_rx_mode,
> >  	.ndo_init = dpa_ndo_init,
> > diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
> > b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
> > index 1ba6617..87577cf 100644
> > --- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
> > +++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
> > @@ -420,9 +420,15 @@ static inline void _dpa_assign_wq(struct dpa_fq
> *fq)
> >  	}
> >  }
> >
> > +#ifdef CONFIG_FSL_DPAA_ETH_USE_NDO_SELECT_QUEUE
> > +/* Use in lieu of skb_get_queue_mapping() */
> > +#define dpa_get_queue_mapping(skb) \
> > +	raw_smp_processor_id()
> > +#else
> >  /* Use the queue selected by XPS */
> >  #define dpa_get_queue_mapping(skb) \
> >  	skb_get_queue_mapping(skb)
> > +#endif
> 
> Why is this necessary?  Shouldn't providing a custom .ndo_select_queue()
> be
> sufficient to ensure that skb_get_queue_mapping() returns the same thing?

dpa_get_queue_mapping() is used in more than one place, the ndo function cannot
be used directly in all places, the current setup is justified.

> And if this goes away, it's just a matter of a function pointer, so if it
> does
> need to be configurable it could be a runtime option.
> 
> -Scott

It's used on the hot path, adding an extra indirection layer to make it selectable
at runtime would defeat the purpose...

Madalin

Re: [net-next v4 4/8] dpaa_eth: add driver's Tx queue selection

[Scott Wood]

On Thu, 2015-12-03 at 04:02 -0600, Bucur Madalin-Cristian-B32716 wrote:
> > -----Original Message-----
> > From: Wood Scott-B07421
> > Sent: Wednesday, December 02, 2015 11:41 PM
> > 
> > On Mon, 2015-11-02 at 19:31 +0200, Madalin Bucur wrote:
> > > Allow the selection of the transmission queue based on the CPU id.
> > 
> > Explain why.
> 
> I'll add more details in the commit log. This is a customer generated
> feature. Bypassing the standard XPS can increase performance by making use
> of the DPAA HW particularities.
> 
> > > 
> > > Signed-off-by: Madalin Bucur <madalin.bucur@freescale.com>
> > > ---
> > >  drivers/net/ethernet/freescale/dpaa/Kconfig           | 10 ++++++++++
> > >  drivers/net/ethernet/freescale/dpaa/dpaa_eth.c        |  3 +++
> > >  drivers/net/ethernet/freescale/dpaa/dpaa_eth.h        |  6 ++++++
> > >  drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c |  8 ++++++++
> > >  drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h |  4 ++++
> > >  5 files changed, 31 insertions(+)
> > > 
> > > diff --git a/drivers/net/ethernet/freescale/dpaa/Kconfig
> > > b/drivers/net/ethernet/freescale/dpaa/Kconfig
> > > index 022d5aa..2577aac 100644
> > > --- a/drivers/net/ethernet/freescale/dpaa/Kconfig
> > > +++ b/drivers/net/ethernet/freescale/dpaa/Kconfig
> > > @@ -11,6 +11,16 @@ menuconfig FSL_DPAA_ETH
> > > 
> > >  if FSL_DPAA_ETH
> > > 
> > > +config FSL_DPAA_ETH_USE_NDO_SELECT_QUEUE
> > > +	bool "Use driver's Tx queue selection mechanism"
> > > +	default y
> > > +	---help---
> > > +	  The DPAA Ethernet driver defines a ndo_select_queue()
> > > callback
> > > for optimal selection
> > > +	  of the egress FQ. That will override the XPS support for this
> > > netdevice.
> > > +	  If for whatever reason you want to be in control of the
> > > egress FQ
> > > -to-CPU selection and mapping,
> > > +	  or simply don't want to use the driver's ndo_select_queue()
> > > callback, then unselect this
> > > +	  and use the standard XPS support instead.
> > 
> > Is there a use case for needing this to be configurable?
> 
> If the standard XPS is desired, the Kconfig option allows the driver user to
> select that.

Under what circumstances does it make sense to desire this?  Could you put the
answer to that in the help text?

> > > diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
> > > b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
> > > index 31d55b4..894f1a7 100644
> > > --- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
> > > +++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
> > > @@ -390,6 +390,9 @@ static const struct net_device_ops dpa_private_ops =
> > {
> > >  	.ndo_get_stats64 = dpa_get_stats64,
> > >  	.ndo_set_mac_address = dpa_set_mac_address,
> > >  	.ndo_validate_addr = eth_validate_addr,
> > > +#ifdef CONFIG_FSL_DPAA_ETH_USE_NDO_SELECT_QUEUE
> > > +	.ndo_select_queue = dpa_select_queue,
> > > +#endif
> > >  	.ndo_change_mtu = dpa_change_mtu,
> > >  	.ndo_set_rx_mode = dpa_set_rx_mode,
> > >  	.ndo_init = dpa_ndo_init,
> > > diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
> > > b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
> > > index 1ba6617..87577cf 100644
> > > --- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
> > > +++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
> > > @@ -420,9 +420,15 @@ static inline void _dpa_assign_wq(struct dpa_fq
> > *fq)
> > >  	}
> > >  }
> > > 
> > > +#ifdef CONFIG_FSL_DPAA_ETH_USE_NDO_SELECT_QUEUE
> > > +/* Use in lieu of skb_get_queue_mapping() */
> > > +#define dpa_get_queue_mapping(skb) \
> > > +	raw_smp_processor_id()
> > > +#else
> > >  /* Use the queue selected by XPS */
> > >  #define dpa_get_queue_mapping(skb) \
> > >  	skb_get_queue_mapping(skb)
> > > +#endif
> > 
> > Why is this necessary?  Shouldn't providing a custom .ndo_select_queue()
> > be
> > sufficient to ensure that skb_get_queue_mapping() returns the same thing?
> 
> dpa_get_queue_mapping() is used in more than one place, the ndo function
> cannot
> be used directly in all places, the current setup is justified.

It's called in two places that I see.  For the call in dpa_tx(), when will
skb_get_queue_mapping() not return the correct answer?  For the call in
dpa_select_queue(), that's already called via function pointer so it would not
be a new indirection layer.

> > And if this goes away, it's just a matter of a function pointer, so if it
> > does
> > need to be configurable it could be a runtime option.
> > 
> > -Scott
> 
> It's used on the hot path, adding an extra indirection layer to make it
> selectable
> at runtime would defeat the purpose...

"if this goes away"

I wasn't asking for a new indirection layer.

-Scott

[net-next v4 5/8] dpaa_eth: add ethtool functionality

[Madalin Bucur]

Add support for basic ethtool operations.

Signed-off-by: Madalin Bucur <madalin.bucur@freescale.com>
---
 drivers/net/ethernet/freescale/dpaa/Makefile       |   2 +-
 .../net/ethernet/freescale/dpaa/dpaa_eth_common.c  |   2 +
 .../net/ethernet/freescale/dpaa/dpaa_eth_common.h  |   3 +
 drivers/net/ethernet/freescale/dpaa/dpaa_ethtool.c | 230 +++++++++++++++++++++
 4 files changed, 236 insertions(+), 1 deletion(-)
 create mode 100644 drivers/net/ethernet/freescale/dpaa/dpaa_ethtool.c

diff --git a/drivers/net/ethernet/freescale/dpaa/Makefile b/drivers/net/ethernet/freescale/dpaa/Makefile
index 3847ec7..9b75d52 100644
--- a/drivers/net/ethernet/freescale/dpaa/Makefile
+++ b/drivers/net/ethernet/freescale/dpaa/Makefile
@@ -8,4 +8,4 @@ ccflags-y += -I$(FMAN)
 
 obj-$(CONFIG_FSL_DPAA_ETH) += fsl_dpa.o
 
-fsl_dpa-objs += dpaa_eth.o dpaa_eth_sg.o dpaa_eth_common.o
+fsl_dpa-objs += dpaa_eth.o dpaa_eth_sg.o dpaa_eth_common.o dpaa_ethtool.o
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c
index 89f3b1f..2b95696 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c
@@ -102,6 +102,8 @@ int dpa_netdev_init(struct net_device *net_dev,
 	memcpy(net_dev->perm_addr, mac_addr, net_dev->addr_len);
 	memcpy(net_dev->dev_addr, mac_addr, net_dev->addr_len);
 
+	net_dev->ethtool_ops = &dpa_ethtool_ops;
+
 	net_dev->needed_headroom = priv->tx_headroom;
 	net_dev->watchdog_timeo = msecs_to_jiffies(tx_timeout);
 
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h
index 2e9471d..160a018 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h
@@ -43,6 +43,9 @@
 /* used in napi related functions */
 extern u16 qman_portal_max;
 
+/* from dpa_ethtool.c */
+extern const struct ethtool_ops dpa_ethtool_ops;
+
 int dpa_netdev_init(struct net_device *net_dev,
 		    const u8 *mac_addr,
 		    u16 tx_timeout);
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_ethtool.c b/drivers/net/ethernet/freescale/dpaa/dpaa_ethtool.c
new file mode 100644
index 0000000..fa8ba69
--- /dev/null
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_ethtool.c
@@ -0,0 +1,230 @@
+/* Copyright 2008-2015 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/string.h>
+
+#include "dpaa_eth.h"
+#include "mac.h"
+#include "dpaa_eth_common.h"
+
+static int dpa_get_settings(struct net_device *net_dev,
+			    struct ethtool_cmd *et_cmd)
+{
+	int err;
+	struct dpa_priv_s *priv;
+
+	priv = netdev_priv(net_dev);
+
+	if (!priv->mac_dev->phy_dev) {
+		netdev_dbg(net_dev, "phy device not initialized\n");
+		return 0;
+	}
+
+	err = phy_ethtool_gset(priv->mac_dev->phy_dev, et_cmd);
+
+	return err;
+}
+
+static int dpa_set_settings(struct net_device *net_dev,
+			    struct ethtool_cmd *et_cmd)
+{
+	int err;
+	struct dpa_priv_s *priv;
+
+	priv = netdev_priv(net_dev);
+
+	if (!priv->mac_dev->phy_dev) {
+		netdev_err(net_dev, "phy device not initialized\n");
+		return -ENODEV;
+	}
+
+	err = phy_ethtool_sset(priv->mac_dev->phy_dev, et_cmd);
+	if (err < 0)
+		netdev_err(net_dev, "phy_ethtool_sset() = %d\n", err);
+
+	return err;
+}
+
+static void dpa_get_drvinfo(struct net_device *net_dev,
+			    struct ethtool_drvinfo *drvinfo)
+{
+	int len;
+
+	strlcpy(drvinfo->driver, KBUILD_MODNAME,
+		sizeof(drvinfo->driver));
+	len = snprintf(drvinfo->version, sizeof(drvinfo->version),
+		       "%X", 0);
+	len = snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version),
+		       "%X", 0);
+
+	if (len >= sizeof(drvinfo->fw_version)) {
+		/* Truncated output */
+		netdev_notice(net_dev, "snprintf() = %d\n", len);
+	}
+	strlcpy(drvinfo->bus_info, dev_name(net_dev->dev.parent->parent),
+		sizeof(drvinfo->bus_info));
+}
+
+static u32 dpa_get_msglevel(struct net_device *net_dev)
+{
+	return ((struct dpa_priv_s *)netdev_priv(net_dev))->msg_enable;
+}
+
+static void dpa_set_msglevel(struct net_device *net_dev,
+			     u32 msg_enable)
+{
+	((struct dpa_priv_s *)netdev_priv(net_dev))->msg_enable = msg_enable;
+}
+
+static int dpa_nway_reset(struct net_device *net_dev)
+{
+	int err;
+	struct dpa_priv_s *priv;
+
+	priv = netdev_priv(net_dev);
+
+	if (!priv->mac_dev->phy_dev) {
+		netdev_err(net_dev, "phy device not initialized\n");
+		return -ENODEV;
+	}
+
+	err = 0;
+	if (priv->mac_dev->phy_dev->autoneg) {
+		err = phy_start_aneg(priv->mac_dev->phy_dev);
+		if (err < 0)
+			netdev_err(net_dev, "phy_start_aneg() = %d\n",
+				   err);
+	}
+
+	return err;
+}
+
+static void dpa_get_pauseparam(struct net_device *net_dev,
+			       struct ethtool_pauseparam *epause)
+{
+	struct dpa_priv_s *priv;
+	struct mac_device *mac_dev;
+	struct phy_device *phy_dev;
+
+	priv = netdev_priv(net_dev);
+	mac_dev = priv->mac_dev;
+
+	phy_dev = mac_dev->phy_dev;
+	if (!phy_dev) {
+		netdev_err(net_dev, "phy device not initialized\n");
+		return;
+	}
+
+	epause->autoneg = mac_dev->autoneg_pause;
+	epause->rx_pause = mac_dev->rx_pause_active;
+	epause->tx_pause = mac_dev->tx_pause_active;
+}
+
+static int dpa_set_pauseparam(struct net_device *net_dev,
+			      struct ethtool_pauseparam *epause)
+{
+	struct dpa_priv_s *priv;
+	struct mac_device *mac_dev;
+	struct phy_device *phy_dev;
+	int err;
+	u32 newadv, oldadv;
+	bool rx_pause, tx_pause;
+
+	priv = netdev_priv(net_dev);
+	mac_dev = priv->mac_dev;
+
+	phy_dev = mac_dev->phy_dev;
+	if (!phy_dev) {
+		netdev_err(net_dev, "phy device not initialized\n");
+		return -ENODEV;
+	}
+
+	if (!(phy_dev->supported & SUPPORTED_Pause) ||
+	    (!(phy_dev->supported & SUPPORTED_Asym_Pause) &&
+	    (epause->rx_pause != epause->tx_pause)))
+		return -EINVAL;
+
+	/* The MAC should know how to handle PAUSE frame autonegotiation before
+	 * adjust_link is triggered by a forced renegotiation of sym/asym PAUSE
+	 * settings.
+	 */
+	mac_dev->autoneg_pause = !!epause->autoneg;
+	mac_dev->rx_pause_req = !!epause->rx_pause;
+	mac_dev->tx_pause_req = !!epause->tx_pause;
+
+	/* Determine the sym/asym advertised PAUSE capabilities from the desired
+	 * rx/tx pause settings.
+	 */
+	newadv = 0;
+	if (epause->rx_pause)
+		newadv = ADVERTISED_Pause | ADVERTISED_Asym_Pause;
+	if (epause->tx_pause)
+		newadv |= ADVERTISED_Asym_Pause;
+
+	oldadv = phy_dev->advertising &
+			(ADVERTISED_Pause | ADVERTISED_Asym_Pause);
+
+	/* If there are differences between the old and the new advertised
+	 * values, restart PHY autonegotiation and advertise the new values.
+	 */
+	if (oldadv != newadv) {
+		phy_dev->advertising &= ~(ADVERTISED_Pause
+				| ADVERTISED_Asym_Pause);
+		phy_dev->advertising |= newadv;
+		if (phy_dev->autoneg) {
+			err = phy_start_aneg(phy_dev);
+			if (err < 0)
+				netdev_err(net_dev, "phy_start_aneg() = %d\n",
+					   err);
+		}
+	}
+
+	fman_get_pause_cfg(mac_dev, &rx_pause, &tx_pause);
+	err = fman_set_mac_active_pause(mac_dev, rx_pause, tx_pause);
+	if (err < 0)
+		netdev_err(net_dev, "set_mac_active_pause() = %d\n", err);
+
+	return err;
+}
+
+const struct ethtool_ops dpa_ethtool_ops = {
+	.get_settings = dpa_get_settings,
+	.set_settings = dpa_set_settings,
+	.get_drvinfo = dpa_get_drvinfo,
+	.get_msglevel = dpa_get_msglevel,
+	.set_msglevel = dpa_set_msglevel,
+	.nway_reset = dpa_nway_reset,
+	.get_pauseparam = dpa_get_pauseparam,
+	.set_pauseparam = dpa_set_pauseparam,
+	.get_link = ethtool_op_get_link,
+};
-- 
1.7.11.7

[net-next v4 6/8] dpaa_eth: add ethtool statistics

[Madalin Bucur]

Add a series of counters to be exported through ethtool:
- add detailed counters for reception errors;
- add detailed counters for QMan enqueue reject events;
- count the number of fragmented skbs received from the stack;
- count all frames received on the Tx confirmation path;
- add congestion group statistics;
- count the number of interrupts for each CPU.

Signed-off-by: Ioana Ciornei <ioana.ciornei@freescale.com>
Signed-off-by: Madalin Bucur <madalin.bucur@freescale.com>
---
 drivers/net/ethernet/freescale/dpaa/dpaa_eth.c     |  12 ++
 drivers/net/ethernet/freescale/dpaa/dpaa_eth.h     |  34 ++++
 .../net/ethernet/freescale/dpaa/dpaa_eth_common.c  |  40 ++++-
 .../net/ethernet/freescale/dpaa/dpaa_eth_common.h  |   2 +
 drivers/net/ethernet/freescale/dpaa/dpaa_eth_sg.c  |   1 +
 drivers/net/ethernet/freescale/dpaa/dpaa_ethtool.c | 183 +++++++++++++++++++++
 6 files changed, 270 insertions(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
index 894f1a7..0b3332a 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
@@ -102,6 +102,15 @@ static void _dpa_rx_error(struct net_device *net_dev,
 
 	percpu_priv->stats.rx_errors++;
 
+	if (fd->status & FM_FD_ERR_DMA)
+		percpu_priv->rx_errors.dme++;
+	if (fd->status & FM_FD_ERR_PHYSICAL)
+		percpu_priv->rx_errors.fpe++;
+	if (fd->status & FM_FD_ERR_SIZE)
+		percpu_priv->rx_errors.fse++;
+	if (fd->status & FM_FD_ERR_PRS_HDR_ERR)
+		percpu_priv->rx_errors.phe++;
+
 	dpa_fd_release(net_dev, fd);
 }
 
@@ -167,6 +176,8 @@ static void _dpa_tx_conf(struct net_device *net_dev,
 		percpu_priv->stats.tx_errors++;
 	}
 
+	percpu_priv->tx_confirm++;
+
 	skb = _dpa_cleanup_tx_fd(priv, fd);
 
 	dev_kfree_skb(skb);
@@ -302,6 +313,7 @@ static void priv_ern(struct qman_portal *portal,
 
 	percpu_priv->stats.tx_dropped++;
 	percpu_priv->stats.tx_fifo_errors++;
+	count_ern(percpu_priv, msg);
 
 	/* If we intended this buffer to go into the pool
 	 * when the FM was done, we need to put it in
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
index 87577cf..ccaadd9 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
@@ -192,6 +192,25 @@ struct dpa_bp {
 	void (*free_buf_cb)(void *addr);
 };
 
+struct dpa_rx_errors {
+	u64 dme;		/* DMA Error */
+	u64 fpe;		/* Frame Physical Error */
+	u64 fse;		/* Frame Size Error */
+	u64 phe;		/* Header Error */
+};
+
+/* Counters for QMan ERN frames - one counter per rejection code */
+struct dpa_ern_cnt {
+	u64 cg_tdrop;		/* Congestion group taildrop */
+	u64 wred;		/* WRED congestion */
+	u64 err_cond;		/* Error condition */
+	u64 early_window;	/* Order restoration, frame too early */
+	u64 late_window;	/* Order restoration, frame too late */
+	u64 fq_tdrop;		/* FQ taildrop */
+	u64 fq_retired;		/* FQ is retired */
+	u64 orp_zero;		/* ORP disabled */
+};
+
 struct dpa_napi_portal {
 	struct napi_struct napi;
 	struct qman_portal *p;
@@ -201,7 +220,13 @@ struct dpa_napi_portal {
 struct dpa_percpu_priv_s {
 	struct net_device *net_dev;
 	struct dpa_napi_portal *np;
+	u64 in_interrupt;
+	u64 tx_confirm;
+	/* fragmented (non-linear) skbuffs received from the stack */
+	u64 tx_frag_skbuffs;
 	struct rtnl_link_stats64 stats;
+	struct dpa_rx_errors rx_errors;
+	struct dpa_ern_cnt ern_cnt;
 };
 
 struct dpa_priv_s {
@@ -228,6 +253,14 @@ struct dpa_priv_s {
 		 * (and the same) congestion group.
 		 */
 		struct qman_cgr cgr;
+		/* If congested, when it began. Used for performance stats. */
+		u32 congestion_start_jiffies;
+		/* Number of jiffies the Tx port was congested. */
+		u32 congested_jiffies;
+		/* Counter for the number of times the CGR
+		 * entered congestion state
+		 */
+		u32 cgr_congested_count;
 	} cgr_data;
 	/* Use a per-port CGR for ingress traffic. */
 	bool use_ingress_cgr;
@@ -289,6 +322,7 @@ static inline int dpaa_eth_napi_schedule(struct dpa_percpu_priv_s *percpu_priv,
 
 			np->p = portal;
 			napi_schedule(&np->napi);
+			percpu_priv->in_interrupt++;
 			return 1;
 		}
 	}
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c
index 2b95696..4947cb9 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c
@@ -751,10 +751,15 @@ static void dpaa_eth_cgscn(struct qman_portal *qm, struct qman_cgr *cgr,
 	struct dpa_priv_s *priv = (struct dpa_priv_s *)container_of(cgr,
 		struct dpa_priv_s, cgr_data.cgr);
 
-	if (congested)
+	if (congested) {
+		priv->cgr_data.congestion_start_jiffies = jiffies;
 		netif_tx_stop_all_queues(priv->net_dev);
-	else
+		priv->cgr_data.cgr_congested_count++;
+	} else {
+		priv->cgr_data.congested_jiffies +=
+			(jiffies - priv->cgr_data.congestion_start_jiffies);
 		netif_tx_wake_all_queues(priv->net_dev);
+	}
 }
 
 int dpaa_eth_cgr_init(struct dpa_priv_s *priv)
@@ -1252,6 +1257,37 @@ void dpa_fd_release(const struct net_device *net_dev, const struct qm_fd *fd)
 		cpu_relax();
 }
 
+void count_ern(struct dpa_percpu_priv_s *percpu_priv,
+	       const struct qm_mr_entry *msg)
+{
+	switch (msg->ern.rc & QM_MR_RC_MASK) {
+	case QM_MR_RC_CGR_TAILDROP:
+		percpu_priv->ern_cnt.cg_tdrop++;
+		break;
+	case QM_MR_RC_WRED:
+		percpu_priv->ern_cnt.wred++;
+		break;
+	case QM_MR_RC_ERROR:
+		percpu_priv->ern_cnt.err_cond++;
+		break;
+	case QM_MR_RC_ORPWINDOW_EARLY:
+		percpu_priv->ern_cnt.early_window++;
+		break;
+	case QM_MR_RC_ORPWINDOW_LATE:
+		percpu_priv->ern_cnt.late_window++;
+		break;
+	case QM_MR_RC_FQ_TAILDROP:
+		percpu_priv->ern_cnt.fq_tdrop++;
+		break;
+	case QM_MR_RC_ORPWINDOW_RETIRED:
+		percpu_priv->ern_cnt.fq_retired++;
+		break;
+	case QM_MR_RC_ORP_ZERO:
+		percpu_priv->ern_cnt.orp_zero++;
+		break;
+	}
+}
+
 /* Turn on HW checksum computation for this outgoing frame.
  * If the current protocol is not something we support in this regard
  * (or if the stack has already computed the SW checksum), we do nothing.
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h
index 160a018..262b9b4 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.h
@@ -100,6 +100,8 @@ void dpaa_eth_init_ports(struct mac_device *mac_dev,
 			 struct device *dev);
 void dpa_release_sgt(struct qm_sg_entry *sgt);
 void dpa_fd_release(const struct net_device *net_dev, const struct qm_fd *fd);
+void count_ern(struct dpa_percpu_priv_s *percpu_priv,
+	       const struct qm_mr_entry *msg);
 int dpa_enable_tx_csum(struct dpa_priv_s *priv,
 		       struct sk_buff *skb,
 		       struct qm_fd *fd,
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_sg.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_sg.c
index af83ac2c..b4c2cee 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_sg.c
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_sg.c
@@ -673,6 +673,7 @@ int dpa_tx(struct sk_buff *skb, struct net_device *net_dev)
 	    likely(skb_shinfo(skb)->nr_frags < DPA_SGT_MAX_ENTRIES)) {
 		/* Just create a S/G fd based on the skb */
 		err = skb_to_sg_fd(priv, skb, &fd);
+		percpu_priv->tx_frag_skbuffs++;
 	} else {
 		/* If the egress skb contains more fragments than we support
 		 * we have no choice but to linearize it ourselves.
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_ethtool.c b/drivers/net/ethernet/freescale/dpaa/dpaa_ethtool.c
index fa8ba69..356857b 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_ethtool.c
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_ethtool.c
@@ -37,6 +37,43 @@
 #include "mac.h"
 #include "dpaa_eth_common.h"
 
+static const char dpa_stats_percpu[][ETH_GSTRING_LEN] = {
+	"interrupts",
+	"rx packets",
+	"tx packets",
+	"tx confirm",
+	"tx S/G",
+	"tx error",
+	"rx error",
+	"bp count"
+};
+
+static char dpa_stats_global[][ETH_GSTRING_LEN] = {
+	/* dpa rx errors */
+	"rx dma error",
+	"rx frame physical error",
+	"rx frame size error",
+	"rx header error",
+
+	/* demultiplexing errors */
+	"qman cg_tdrop",
+	"qman wred",
+	"qman error cond",
+	"qman early window",
+	"qman late window",
+	"qman fq tdrop",
+	"qman fq retired",
+	"qman orp disabled",
+
+	/* congestion related stats */
+	"congestion time (ms)",
+	"entered congestion",
+	"congested (0/1)"
+};
+
+#define DPA_STATS_PERCPU_LEN ARRAY_SIZE(dpa_stats_percpu)
+#define DPA_STATS_GLOBAL_LEN ARRAY_SIZE(dpa_stats_global)
+
 static int dpa_get_settings(struct net_device *net_dev,
 			    struct ethtool_cmd *et_cmd)
 {
@@ -217,6 +254,149 @@ static int dpa_set_pauseparam(struct net_device *net_dev,
 	return err;
 }
 
+static int dpa_get_sset_count(struct net_device *net_dev, int type)
+{
+	unsigned int total_stats, num_stats;
+
+	num_stats   = num_online_cpus() + 1;
+	total_stats = num_stats * DPA_STATS_PERCPU_LEN + DPA_STATS_GLOBAL_LEN;
+
+	switch (type) {
+	case ETH_SS_STATS:
+		return total_stats;
+	default:
+		return -EOPNOTSUPP;
+	}
+}
+
+static void copy_stats(struct dpa_percpu_priv_s *percpu_priv, int num_cpus,
+		       int crr_cpu, u64 bp_count, u64 *data)
+{
+	int num_values = num_cpus + 1;
+	int crr = 0;
+
+	/* update current CPU's stats and also add them to the total values */
+	data[crr * num_values + crr_cpu] = percpu_priv->in_interrupt;
+	data[crr++ * num_values + num_cpus] += percpu_priv->in_interrupt;
+
+	data[crr * num_values + crr_cpu] = percpu_priv->stats.rx_packets;
+	data[crr++ * num_values + num_cpus] += percpu_priv->stats.rx_packets;
+
+	data[crr * num_values + crr_cpu] = percpu_priv->stats.tx_packets;
+	data[crr++ * num_values + num_cpus] += percpu_priv->stats.tx_packets;
+
+	data[crr * num_values + crr_cpu] = percpu_priv->tx_confirm;
+	data[crr++ * num_values + num_cpus] += percpu_priv->tx_confirm;
+
+	data[crr * num_values + crr_cpu] = percpu_priv->tx_frag_skbuffs;
+	data[crr++ * num_values + num_cpus] += percpu_priv->tx_frag_skbuffs;
+
+	data[crr * num_values + crr_cpu] = percpu_priv->stats.tx_errors;
+	data[crr++ * num_values + num_cpus] += percpu_priv->stats.tx_errors;
+
+	data[crr * num_values + crr_cpu] = percpu_priv->stats.rx_errors;
+	data[crr++ * num_values + num_cpus] += percpu_priv->stats.rx_errors;
+
+	data[crr * num_values + crr_cpu] = bp_count;
+	data[crr++ * num_values + num_cpus] += bp_count;
+}
+
+static void dpa_get_ethtool_stats(struct net_device *net_dev,
+				  struct ethtool_stats *stats, u64 *data)
+{
+	u64 bp_count, cg_time, cg_num, cg_status;
+	struct dpa_percpu_priv_s *percpu_priv;
+	struct qm_mcr_querycgr query_cgr;
+	struct dpa_rx_errors rx_errors;
+	struct dpa_ern_cnt ern_cnt;
+	struct dpa_priv_s *priv;
+	unsigned int num_cpus, offset;
+	struct dpa_bp *dpa_bp;
+	int total_stats, i;
+
+	total_stats = dpa_get_sset_count(net_dev, ETH_SS_STATS);
+	priv     = netdev_priv(net_dev);
+	dpa_bp   = priv->dpa_bp;
+	num_cpus = num_online_cpus();
+	bp_count = 0;
+
+	memset(&rx_errors, 0, sizeof(struct dpa_rx_errors));
+	memset(&ern_cnt, 0, sizeof(struct dpa_ern_cnt));
+	memset(data, 0, total_stats * sizeof(u64));
+
+	for_each_online_cpu(i) {
+		percpu_priv = per_cpu_ptr(priv->percpu_priv, i);
+
+		if (dpa_bp->percpu_count)
+			bp_count = *(per_cpu_ptr(dpa_bp->percpu_count, i));
+
+		rx_errors.dme += percpu_priv->rx_errors.dme;
+		rx_errors.fpe += percpu_priv->rx_errors.fpe;
+		rx_errors.fse += percpu_priv->rx_errors.fse;
+		rx_errors.phe += percpu_priv->rx_errors.phe;
+
+		ern_cnt.cg_tdrop     += percpu_priv->ern_cnt.cg_tdrop;
+		ern_cnt.wred         += percpu_priv->ern_cnt.wred;
+		ern_cnt.err_cond     += percpu_priv->ern_cnt.err_cond;
+		ern_cnt.early_window += percpu_priv->ern_cnt.early_window;
+		ern_cnt.late_window  += percpu_priv->ern_cnt.late_window;
+		ern_cnt.fq_tdrop     += percpu_priv->ern_cnt.fq_tdrop;
+		ern_cnt.fq_retired   += percpu_priv->ern_cnt.fq_retired;
+		ern_cnt.orp_zero     += percpu_priv->ern_cnt.orp_zero;
+
+		copy_stats(percpu_priv, num_cpus, i, bp_count, data);
+	}
+
+	offset = (num_cpus + 1) * DPA_STATS_PERCPU_LEN;
+	memcpy(data + offset, &rx_errors, sizeof(struct dpa_rx_errors));
+
+	offset += sizeof(struct dpa_rx_errors) / sizeof(u64);
+	memcpy(data + offset, &ern_cnt, sizeof(struct dpa_ern_cnt));
+
+	/* gather congestion related counters */
+	cg_num    = 0;
+	cg_status = 0;
+	cg_time   = jiffies_to_msecs(priv->cgr_data.congested_jiffies);
+	if (qman_query_cgr(&priv->cgr_data.cgr, &query_cgr) == 0) {
+		cg_num    = priv->cgr_data.cgr_congested_count;
+		cg_status = query_cgr.cgr.cs;
+
+		/* reset congestion stats (like QMan API does */
+		priv->cgr_data.congested_jiffies   = 0;
+		priv->cgr_data.cgr_congested_count = 0;
+	}
+
+	offset += sizeof(struct dpa_ern_cnt) / sizeof(u64);
+	data[offset++] = cg_time;
+	data[offset++] = cg_num;
+	data[offset++] = cg_status;
+}
+
+static void dpa_get_strings(struct net_device *net_dev, u32 stringset, u8 *data)
+{
+	unsigned int i, j, num_cpus, size;
+	char string_cpu[ETH_GSTRING_LEN];
+	u8 *strings;
+
+	strings   = data;
+	num_cpus  = num_online_cpus();
+	size      = DPA_STATS_GLOBAL_LEN * ETH_GSTRING_LEN;
+
+	for (i = 0; i < DPA_STATS_PERCPU_LEN; i++) {
+		for (j = 0; j < num_cpus; j++) {
+			snprintf(string_cpu, ETH_GSTRING_LEN, "%s [CPU %d]",
+				 dpa_stats_percpu[i], j);
+			memcpy(strings, string_cpu, ETH_GSTRING_LEN);
+			strings += ETH_GSTRING_LEN;
+		}
+		snprintf(string_cpu, ETH_GSTRING_LEN, "%s [TOTAL]",
+			 dpa_stats_percpu[i]);
+		memcpy(strings, string_cpu, ETH_GSTRING_LEN);
+		strings += ETH_GSTRING_LEN;
+	}
+	memcpy(strings, dpa_stats_global, size);
+}
+
 const struct ethtool_ops dpa_ethtool_ops = {
 	.get_settings = dpa_get_settings,
 	.set_settings = dpa_set_settings,
@@ -227,4 +407,7 @@ const struct ethtool_ops dpa_ethtool_ops = {
 	.get_pauseparam = dpa_get_pauseparam,
 	.set_pauseparam = dpa_set_pauseparam,
 	.get_link = ethtool_op_get_link,
+	.get_sset_count = dpa_get_sset_count,
+	.get_ethtool_stats = dpa_get_ethtool_stats,
+	.get_strings = dpa_get_strings,
 };
-- 
1.7.11.7

Re: [net-next v4 6/8] dpaa_eth: add ethtool statistics

[Joe Perches]

On Mon, 2015-11-02 at 19:31 +0200, Madalin Bucur wrote:
> Add a series of counters to be exported through ethtool:
> - add detailed counters for reception errors;
> - add detailed counters for QMan enqueue reject events;
> - count the number of fragmented skbs received from the stack;
> - count all frames received on the Tx confirmation path;
> - add congestion group statistics;
> - count the number of interrupts for each CPU.
[]
> diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_ethtool.c b/drivers/net/ethernet/freescale/dpaa/dpaa_ethtool.c
[]
> +static void dpa_get_strings(struct net_device *net_dev, u32 stringset, u8 *data)
> +{
> +	unsigned int i, j, num_cpus, size;
> +	char string_cpu[ETH_GSTRING_LEN];
> +	u8 *strings;
> +
> +	strings   = data;
> +	num_cpus  = num_online_cpus();
> +	size      = DPA_STATS_GLOBAL_LEN * ETH_GSTRING_LEN;
> +
> +	for (i = 0; i < DPA_STATS_PERCPU_LEN; i++) {
> +		for (j = 0; j < num_cpus; j++) {
> +			snprintf(string_cpu, ETH_GSTRING_LEN, "%s [CPU %d]",
> +				 dpa_stats_percpu[i], j);
> +			memcpy(strings, string_cpu, ETH_GSTRING_LEN);
> +			strings += ETH_GSTRING_LEN;
> +		}
> +		snprintf(string_cpu, ETH_GSTRING_LEN, "%s [TOTAL]",
> +			 dpa_stats_percpu[i]);
> +		memcpy(strings, string_cpu, ETH_GSTRING_LEN);
> +		strings += ETH_GSTRING_LEN;
> +	}
> +	memcpy(strings, dpa_stats_global, size);
> +}

This leaks uninitialized stack via a memcpy of uninitialized
string_cpu bytes into user-space.

Using
	char string_cpu[ETH_GSTRING_LEN] = {};
or a memset before each snprintf would fix it.

[net-next v4 7/8] dpaa_eth: add sysfs exports

[Madalin Bucur]

Export Frame Queue and Buffer Pool IDs through sysfs.

Signed-off-by: Madalin Bucur <madalin.bucur@freescale.com>
---
 drivers/net/ethernet/freescale/dpaa/Makefile       |   2 +-
 drivers/net/ethernet/freescale/dpaa/dpaa_eth.c     |   2 +
 drivers/net/ethernet/freescale/dpaa/dpaa_eth.h     |   3 +
 .../net/ethernet/freescale/dpaa/dpaa_eth_common.c  |   2 +
 .../net/ethernet/freescale/dpaa/dpaa_eth_sysfs.c   | 167 +++++++++++++++++++++
 5 files changed, 175 insertions(+), 1 deletion(-)
 create mode 100644 drivers/net/ethernet/freescale/dpaa/dpaa_eth_sysfs.c

diff --git a/drivers/net/ethernet/freescale/dpaa/Makefile b/drivers/net/ethernet/freescale/dpaa/Makefile
index 9b75d52..141ade4 100644
--- a/drivers/net/ethernet/freescale/dpaa/Makefile
+++ b/drivers/net/ethernet/freescale/dpaa/Makefile
@@ -8,4 +8,4 @@ ccflags-y += -I$(FMAN)
 
 obj-$(CONFIG_FSL_DPAA_ETH) += fsl_dpa.o
 
-fsl_dpa-objs += dpaa_eth.o dpaa_eth_sg.o dpaa_eth_common.o dpaa_ethtool.o
+fsl_dpa-objs += dpaa_eth.o dpaa_eth_sg.o dpaa_eth_common.o dpaa_ethtool.o dpaa_eth_sysfs.o
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
index 0b3332a..3cd03f5 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
@@ -756,6 +756,8 @@ dpaa_eth_priv_probe(struct platform_device *pdev)
 	if (err < 0)
 		goto netdev_init_failed;
 
+	dpaa_eth_sysfs_init(&net_dev->dev);
+
 	pr_info("Probed interface %s\n", net_dev->name);
 
 	return 0;
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
index ccaadd9..cdc7595 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
@@ -371,6 +371,9 @@ static inline u16 dpa_get_headroom(struct dpa_buffer_layout_s *bl)
 	return bl->data_align ? ALIGN(headroom, bl->data_align) : headroom;
 }
 
+void dpaa_eth_sysfs_remove(struct device *dev);
+void dpaa_eth_sysfs_init(struct device *dev);
+
 void dpa_private_napi_del(struct net_device *net_dev);
 
 static inline void clear_fd(struct qm_fd *fd)
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c
index 4947cb9..2cf4565 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_common.c
@@ -299,6 +299,8 @@ int dpa_remove(struct platform_device *pdev)
 
 	priv = netdev_priv(net_dev);
 
+	dpaa_eth_sysfs_remove(dev);
+
 	dev_set_drvdata(dev, NULL);
 	unregister_netdev(net_dev);
 
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_sysfs.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_sysfs.c
new file mode 100644
index 0000000..a6c71b1
--- /dev/null
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_sysfs.c
@@ -0,0 +1,167 @@
+/* Copyright 2008-2015 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.
+ */
+
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/kthread.h>
+#include <linux/io.h>
+#include <linux/of_net.h>
+#include "dpaa_eth.h"
+#include "mac.h"
+
+static ssize_t dpaa_eth_show_addr(struct device *dev,
+				  struct device_attribute *attr, char *buf)
+{
+	struct dpa_priv_s *priv = netdev_priv(to_net_dev(dev));
+	struct mac_device *mac_dev = priv->mac_dev;
+
+	if (mac_dev)
+		return sprintf(buf, "%llx",
+				(unsigned long long)mac_dev->res->start);
+	else
+		return sprintf(buf, "none");
+}
+
+static ssize_t dpaa_eth_show_fqids(struct device *dev,
+				   struct device_attribute *attr, char *buf)
+{
+	struct dpa_priv_s *priv = netdev_priv(to_net_dev(dev));
+	ssize_t bytes = 0;
+	int i = 0;
+	char *str;
+	struct dpa_fq *fq;
+	struct dpa_fq *tmp;
+	struct dpa_fq *prev = NULL;
+	u32 first_fqid = 0;
+	u32 last_fqid = 0;
+	char *prevstr = NULL;
+
+	list_for_each_entry_safe(fq, tmp, &priv->dpa_fq_list, list) {
+		switch (fq->fq_type) {
+		case FQ_TYPE_RX_DEFAULT:
+			str = "Rx default";
+			break;
+		case FQ_TYPE_RX_ERROR:
+			str = "Rx error";
+			break;
+		case FQ_TYPE_TX_CONFIRM:
+			str = "Tx default confirmation";
+			break;
+		case FQ_TYPE_TX_CONF_MQ:
+			str = "Tx confirmation (mq)";
+			break;
+		case FQ_TYPE_TX_ERROR:
+			str = "Tx error";
+			break;
+		case FQ_TYPE_TX:
+			str = "Tx";
+			break;
+		default:
+			str = "Unknown";
+		}
+
+		if (prev && (abs(fq->fqid - prev->fqid) != 1 ||
+			     str != prevstr)) {
+			if (last_fqid == first_fqid)
+				bytes += sprintf(buf + bytes,
+					"%s: %d\n", prevstr, prev->fqid);
+			else
+				bytes += sprintf(buf + bytes,
+					"%s: %d - %d\n", prevstr,
+					first_fqid, last_fqid);
+		}
+
+		if (prev && abs(fq->fqid - prev->fqid) == 1 &&
+		    str == prevstr) {
+			last_fqid = fq->fqid;
+		} else {
+			first_fqid = fq->fqid;
+			last_fqid = fq->fqid;
+		}
+
+		prev = fq;
+		prevstr = str;
+		i++;
+	}
+
+	if (prev) {
+		if (last_fqid == first_fqid)
+			bytes += sprintf(buf + bytes, "%s: %d\n", prevstr,
+					prev->fqid);
+		else
+			bytes += sprintf(buf + bytes, "%s: %d - %d\n", prevstr,
+					first_fqid, last_fqid);
+	}
+
+	return bytes;
+}
+
+static ssize_t dpaa_eth_show_bpids(struct device *dev,
+				   struct device_attribute *attr, char *buf)
+{
+	ssize_t bytes = 0;
+	struct dpa_priv_s *priv = netdev_priv(to_net_dev(dev));
+	struct dpa_bp *dpa_bp = priv->dpa_bp;
+	int i = 0;
+
+	for (i = 0; i < priv->bp_count; i++)
+		bytes += snprintf(buf + bytes, PAGE_SIZE - bytes, "%u\n",
+				dpa_bp[i].bpid);
+
+	return bytes;
+}
+
+static struct device_attribute dpaa_eth_attrs[] = {
+	__ATTR(device_addr, S_IRUGO, dpaa_eth_show_addr, NULL),
+	__ATTR(fqids, S_IRUGO, dpaa_eth_show_fqids, NULL),
+	__ATTR(bpids, S_IRUGO, dpaa_eth_show_bpids, NULL),
+};
+
+void dpaa_eth_sysfs_init(struct device *dev)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(dpaa_eth_attrs); i++)
+		if (device_create_file(dev, &dpaa_eth_attrs[i])) {
+			dev_err(dev, "Error creating sysfs file\n");
+			while (i > 0)
+				device_remove_file(dev, &dpaa_eth_attrs[--i]);
+			return;
+		}
+}
+
+void dpaa_eth_sysfs_remove(struct device *dev)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(dpaa_eth_attrs); i++)
+		device_remove_file(dev, &dpaa_eth_attrs[i]);
+}
-- 
1.7.11.7

_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/linuxppc-dev

[net-next v4 8/8] dpaa_eth: add trace points

[Madalin Bucur]

Add trace points on the hot processing path.

Signed-off-by: Ruxandra Ioana Radulescu <ruxandra.radulescu@freescale.com>
---
 drivers/net/ethernet/freescale/dpaa/Makefile       |   1 +
 drivers/net/ethernet/freescale/dpaa/dpaa_eth.c     |  12 ++
 drivers/net/ethernet/freescale/dpaa/dpaa_eth.h     |   4 +
 .../net/ethernet/freescale/dpaa/dpaa_eth_trace.h   | 141 +++++++++++++++++++++
 4 files changed, 158 insertions(+)
 create mode 100644 drivers/net/ethernet/freescale/dpaa/dpaa_eth_trace.h

diff --git a/drivers/net/ethernet/freescale/dpaa/Makefile b/drivers/net/ethernet/freescale/dpaa/Makefile
index 141ade4..15ed1c4 100644
--- a/drivers/net/ethernet/freescale/dpaa/Makefile
+++ b/drivers/net/ethernet/freescale/dpaa/Makefile
@@ -9,3 +9,4 @@ ccflags-y += -I$(FMAN)
 obj-$(CONFIG_FSL_DPAA_ETH) += fsl_dpa.o
 
 fsl_dpa-objs += dpaa_eth.o dpaa_eth_sg.o dpaa_eth_common.o dpaa_ethtool.o dpaa_eth_sysfs.o
+CFLAGS_dpaa_eth.o := -I$(src)
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
index 3cd03f5..b939d9d 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.c
@@ -57,6 +57,12 @@
 #include "dpaa_eth.h"
 #include "dpaa_eth_common.h"
 
+/* CREATE_TRACE_POINTS only needs to be defined once. Other dpa files
+ * using trace events only need to #include <trace/events/sched.h>
+ */
+#define CREATE_TRACE_POINTS
+#include "dpaa_eth_trace.h"
+
 /* Valid checksum indication */
 #define DPA_CSUM_VALID		0xFFFF
 
@@ -229,6 +235,9 @@ priv_rx_default_dqrr(struct qman_portal *portal,
 	priv = netdev_priv(net_dev);
 	dpa_bp = priv->dpa_bp;
 
+	/* Trace the Rx fd */
+	trace_dpa_rx_fd(net_dev, fq, &dq->fd);
+
 	/* IRQ handler, non-migratable; safe to use raw_cpu_ptr here */
 	percpu_priv = raw_cpu_ptr(priv->percpu_priv);
 	count_ptr = raw_cpu_ptr(dpa_bp->percpu_count);
@@ -285,6 +294,9 @@ priv_tx_conf_default_dqrr(struct qman_portal *portal,
 	net_dev = ((struct dpa_fq *)fq)->net_dev;
 	priv = netdev_priv(net_dev);
 
+	/* Trace the fd */
+	trace_dpa_tx_conf_fd(net_dev, fq, &dq->fd);
+
 	/* Non-migratable context, safe to use raw_cpu_ptr */
 	percpu_priv = raw_cpu_ptr(priv->percpu_priv);
 
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
index cdc7595..7dee8de 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth.h
@@ -36,6 +36,7 @@
 
 #include "fman.h"
 #include "mac.h"
+#include "dpaa_eth_trace.h"
 
 extern int dpa_rx_extra_headroom;
 extern int dpa_max_frm;
@@ -407,6 +408,9 @@ static inline int dpa_xmit(struct dpa_priv_s *priv,
 	if (fd->bpid == 0xff)
 		fd->cmd |= qman_fq_fqid(priv->conf_fqs[queue]);
 
+	/* Trace this Tx fd */
+	trace_dpa_tx_fd(priv->net_dev, egress_fq, fd);
+
 	for (i = 0; i < 100000; i++) {
 		err = qman_enqueue(egress_fq, fd, 0);
 		if (err != -EBUSY)
diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_eth_trace.h b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_trace.h
new file mode 100644
index 0000000..3b67477
--- /dev/null
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_eth_trace.h
@@ -0,0 +1,141 @@
+/* Copyright 2013-2015 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.
+ */
+
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM	dpaa_eth
+
+#if !defined(_DPAA_ETH_TRACE_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _DPAA_ETH_TRACE_H
+
+#include <linux/skbuff.h>
+#include <linux/netdevice.h>
+#include "dpaa_eth.h"
+#include <linux/tracepoint.h>
+
+#define fd_format_name(format)	{ qm_fd_##format, #format }
+#define fd_format_list	\
+	fd_format_name(contig),	\
+	fd_format_name(sg)
+
+/* This is used to declare a class of events.
+ * individual events of this type will be defined below.
+ */
+
+/* Store details about a frame descriptor and the FQ on which it was
+ * transmitted/received.
+ */
+DECLARE_EVENT_CLASS(dpaa_eth_fd,
+	/* Trace function prototype */
+	TP_PROTO(struct net_device *netdev,
+		 struct qman_fq *fq,
+		 const struct qm_fd *fd),
+
+	/* Repeat argument list here */
+	TP_ARGS(netdev, fq, fd),
+
+	/* A structure containing the relevant information we want to record.
+	 * Declare name and type for each normal element, name, type and size
+	 * for arrays. Use __string for variable length strings.
+	 */
+	TP_STRUCT__entry(
+		__field(u32,	fqid)
+		__field(u64,	fd_addr)
+		__field(u8,	fd_format)
+		__field(u16,	fd_offset)
+		__field(u32,	fd_length)
+		__field(u32,	fd_status)
+		__string(name,	netdev->name)
+	),
+
+	/* The function that assigns values to the above declared fields */
+	TP_fast_assign(
+		__entry->fqid = fq->fqid;
+		__entry->fd_addr = qm_fd_addr_get64(fd);
+		__entry->fd_format = fd->format;
+		__entry->fd_offset = dpa_fd_offset(fd);
+		__entry->fd_length = dpa_fd_length(fd);
+		__entry->fd_status = fd->status;
+		__assign_str(name, netdev->name);
+	),
+
+	/* This is what gets printed when the trace event is triggered */
+	TP_printk("[%s] fqid=%d, fd: addr=0x%llx, format=%s, off=%u, len=%u, status=0x%08x",
+		  __get_str(name), __entry->fqid, __entry->fd_addr,
+		  __print_symbolic(__entry->fd_format, fd_format_list),
+		  __entry->fd_offset, __entry->fd_length, __entry->fd_status)
+);
+
+/* Now declare events of the above type. Format is:
+ * DEFINE_EVENT(class, name, proto, args), with proto and args same as for class
+ */
+
+/* Tx (egress) fd */
+DEFINE_EVENT(dpaa_eth_fd, dpa_tx_fd,
+
+	TP_PROTO(struct net_device *netdev,
+		 struct qman_fq *fq,
+		 const struct qm_fd *fd),
+
+	TP_ARGS(netdev, fq, fd)
+);
+
+/* Rx fd */
+DEFINE_EVENT(dpaa_eth_fd, dpa_rx_fd,
+
+	TP_PROTO(struct net_device *netdev,
+		 struct qman_fq *fq,
+		 const struct qm_fd *fd),
+
+	TP_ARGS(netdev, fq, fd)
+);
+
+/* Tx confirmation fd */
+DEFINE_EVENT(dpaa_eth_fd, dpa_tx_conf_fd,
+
+	TP_PROTO(struct net_device *netdev,
+		 struct qman_fq *fq,
+		 const struct qm_fd *fd),
+
+	TP_ARGS(netdev, fq, fd)
+);
+
+/* If only one event of a certain type needs to be declared, use TRACE_EVENT().
+ * The syntax is the same as for DECLARE_EVENT_CLASS().
+ */
+
+#endif /* _DPAA_ETH_TRACE_H */
+
+/* This must be outside ifdef _DPAA_ETH_TRACE_H */
+#undef TRACE_INCLUDE_PATH
+#define TRACE_INCLUDE_PATH .
+#undef TRACE_INCLUDE_FILE
+#define TRACE_INCLUDE_FILE	dpaa_eth_trace
+#include <trace/define_trace.h>
-- 
1.7.11.7

_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/linuxppc-dev