nuttx/drivers/net/encx24j600.c

2895 lines
83 KiB
C
Raw Normal View History

/****************************************************************************
* drivers/net/encx24j600.c
*
* Copyright (C) 2017-2018 Gregory Nutt. All rights reserved.
* Copyright (C) 2013-2014 UVC Ingenieure. All rights reserved.
2013-09-25 16:26:56 +02:00
* Author: Max Holtzberg <mh@uvc.de>
*
* References:
* - ENC424J600/624J600 Data Sheet, Stand-Alone 10/100 Ethernet Controller
* with SPI or Parallel Interface, DS39935C, 2010 Microchip Technology Inc.
*
* Derived from enc28j60 driver written by:
*
* Copyright (C) 2010-2012 Gregory Nutt. All rights reserved.
* Author: Gregory Nutt <gnutt@nuttx.org>
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. 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.
* 3. Neither the name NuttX nor the names of its contributors may be
* used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "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 THE
* COPYRIGHT OWNER OR CONTRIBUTORS 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.
*
****************************************************************************/
/****************************************************************************
* Included Files
****************************************************************************/
#include <nuttx/config.h>
#if defined(CONFIG_NET) && defined(CONFIG_ENCX24J600)
#include <stdint.h>
#include <stdbool.h>
#include <stdint.h>
#include <time.h>
#include <string.h>
#include <debug.h>
#include <errno.h>
#include <queue.h>
#include <arpa/inet.h>
#include <nuttx/arch.h>
#include <nuttx/irq.h>
#include <nuttx/wdog.h>
#include <nuttx/spi/spi.h>
#include <nuttx/wqueue.h>
#include <nuttx/clock.h>
2014-07-05 00:38:51 +02:00
#include <nuttx/net/net.h>
#include <nuttx/net/arp.h>
#include <nuttx/net/netdev.h>
#include <nuttx/net/encx24j600.h>
#ifdef CONFIG_NET_PKT
# include <nuttx/net/pkt.h>
#endif
#include "encx24j600.h"
/****************************************************************************
2013-09-25 16:26:56 +02:00
* Pre-processor Definitions
****************************************************************************/
/* Configuration ************************************************************/
/* ENCX24J600 Configuration Settings:
*
* CONFIG_ENCX24J600 - Enabled ENCX24J600 support
* CONFIG_ENCX24J600_SPIMODE - Controls the SPI mode
* CONFIG_ENCX24J600_FREQUENCY - Define to use a different bus frequency
* CONFIG_ENCX24J600_NINTERFACES - Specifies the number of physical ENCX24J600
* devices that will be supported.
*/
/* The ENCX24J600 spec says that it supports SPI mode 0,0 only: "The
* implementation used on this device supports SPI mode 0,0 only. In
* addition, the SPI port requires that SCK be at Idle in a low state;
* selectable clock polarity is not supported." However, sometimes you
* need to tinker with these things.
*/
#ifndef CONFIG_ENCX24J600_SPIMODE
# define CONFIG_ENCX24J600_SPIMODE SPIDEV_MODE0
#endif
/* CONFIG_ENCX24J600_NINTERFACES determines the number of physical interfaces
* that will be supported.
*/
#ifndef CONFIG_ENCX24J600_NINTERFACES
# define CONFIG_ENCX24J600_NINTERFACES 1
#endif
This commit attempts remove some long standard confusion in naming and some actual problems that result from the naming confusion. The basic problem is the standard MTU does not include the size of the Ethernet header. For clarity, I changed the naming of most things called MTU to PKTSIZE. For example, CONFIG_NET_ETH_MTU is now CONFIG_NET_ETH_PKTSIZE. This makes the user interface a little hostile. People thing of an MTU of 1500 bytes, but the corresponding packet is really 1514 bytes (including the 14 byte Ethernet header). A more friendly solution would configure the MTU (as before), but then derive the packet buffer size by adding the MAC header length. Instead, we define the packet buffer size then derive the MTU. The MTU is not common currency in networking. On the wire, the only real issue is the MSS which is derived from MTU by subtracting the IP header and TCP header sizes (for the case of TCP). Now it is derived for the PKTSIZE by subtracting the IP header, the TCP header, and the MAC header sizes. So we should be all good and without the recurring 14 byte error in MTU's and MSS's. Squashed commit of the following: Trivial update to fix some spacing issues. net/: Rename several macros containing _MTU to _PKTSIZE. net/: Rename CONFIG_NET_SLIP_MTU to CONFIG_NET_SLIP_PKTSIZE and similarly for CONFIG_NET_TUN_MTU. These are not the MTU which does not include the size of the link layer header. These are the full size of the packet buffer memory (minus any GUARD bytes). net/: Rename CONFIG_NET_6LOWPAN_MTU to CONFIG_NET_6LOWPAN_PKTSIZE and similarly for CONFIG_NET_TUN_MTU. These are not the MTU which does not include the size of the link layer header. These are the full size of the packet buffer memory (minus any GUARD bytes). net/: Rename CONFIG_NET_ETH_MTU to CONFIG_NET_ETH_PKTSIZE. This is not the MTU which does not include the size of the link layer header. This is the full size of the packet buffer memory (minus any GUARD bytes). net/: Rename the file d_mtu in the network driver structure to d_pktsize. That value saved there is not the MTU. The packetsize is the memory large enough to hold the maximum packet PLUS the size of the link layer header. The MTU does not include the link layer header.
2018-07-04 22:10:40 +02:00
/* CONFIG_NET_ETH_PKTSIZE must always be defined */
This commit attempts remove some long standard confusion in naming and some actual problems that result from the naming confusion. The basic problem is the standard MTU does not include the size of the Ethernet header. For clarity, I changed the naming of most things called MTU to PKTSIZE. For example, CONFIG_NET_ETH_MTU is now CONFIG_NET_ETH_PKTSIZE. This makes the user interface a little hostile. People thing of an MTU of 1500 bytes, but the corresponding packet is really 1514 bytes (including the 14 byte Ethernet header). A more friendly solution would configure the MTU (as before), but then derive the packet buffer size by adding the MAC header length. Instead, we define the packet buffer size then derive the MTU. The MTU is not common currency in networking. On the wire, the only real issue is the MSS which is derived from MTU by subtracting the IP header and TCP header sizes (for the case of TCP). Now it is derived for the PKTSIZE by subtracting the IP header, the TCP header, and the MAC header sizes. So we should be all good and without the recurring 14 byte error in MTU's and MSS's. Squashed commit of the following: Trivial update to fix some spacing issues. net/: Rename several macros containing _MTU to _PKTSIZE. net/: Rename CONFIG_NET_SLIP_MTU to CONFIG_NET_SLIP_PKTSIZE and similarly for CONFIG_NET_TUN_MTU. These are not the MTU which does not include the size of the link layer header. These are the full size of the packet buffer memory (minus any GUARD bytes). net/: Rename CONFIG_NET_6LOWPAN_MTU to CONFIG_NET_6LOWPAN_PKTSIZE and similarly for CONFIG_NET_TUN_MTU. These are not the MTU which does not include the size of the link layer header. These are the full size of the packet buffer memory (minus any GUARD bytes). net/: Rename CONFIG_NET_ETH_MTU to CONFIG_NET_ETH_PKTSIZE. This is not the MTU which does not include the size of the link layer header. This is the full size of the packet buffer memory (minus any GUARD bytes). net/: Rename the file d_mtu in the network driver structure to d_pktsize. That value saved there is not the MTU. The packetsize is the memory large enough to hold the maximum packet PLUS the size of the link layer header. The MTU does not include the link layer header.
2018-07-04 22:10:40 +02:00
#if !defined(CONFIG_NET_ETH_PKTSIZE) && (CONFIG_NET_ETH_PKTSIZE <= MAX_FRAMELEN)
# error "CONFIG_NET_ETH_PKTSIZE is not valid for the ENCX24J600"
#endif
/* We need to have the work queue to handle SPI interrupts */
#if !defined(CONFIG_SCHED_WORKQUEUE)
# error "Worker thread support is required (CONFIG_SCHED_WORKQUEUE)"
#endif
/* The low priority work queue is preferred. If it is not enabled, LPWORK
* will be the same as HPWORK.
*
* NOTE: However, the network should NEVER run on the high priority work
* queue! That queue is intended only to service short back end interrupt
* processing that never suspends. Suspending the high priority work queue
* may bring the system to its knees!
*/
#define ENCWORK LPWORK
/* CONFIG_ENCX24J600_DUMPPACKET will dump the contents of each packet to the console. */
#ifdef CONFIG_ENCX24J600_DUMPPACKET
# define enc_dumppacket(m,a,n) lib_dumpbuffer(m,a,n)
#else
# define enc_dumppacket(m,a,n)
#endif
/* Low-level register debug */
#if !defined(CONFIG_DEBUG_FEATURES) || !defined(CONFIG_DEBUG_NET)
# undef CONFIG_ENCX24J600_REGDEBUG
#endif
/* Timing *******************************************************************/
/* TX poll delay = 1 seconds. CLK_TCK is the number of clock ticks per second */
#define ENC_WDDELAY (1*CLK_TCK)
/* TX timeout = 1 minute */
#define ENC_TXTIMEOUT (60*CLK_TCK)
/* Poll timeout */
#define ENC_POLLTIMEOUT MSEC2TICK(50)
2013-09-25 16:26:56 +02:00
/* Register poll timeout */
#define ENC_REGPOLLTIMEOUT MSEC2TICK(5000)
/* Packet Memory ************************************************************/
/* Packet memory layout */
This commit attempts remove some long standard confusion in naming and some actual problems that result from the naming confusion. The basic problem is the standard MTU does not include the size of the Ethernet header. For clarity, I changed the naming of most things called MTU to PKTSIZE. For example, CONFIG_NET_ETH_MTU is now CONFIG_NET_ETH_PKTSIZE. This makes the user interface a little hostile. People thing of an MTU of 1500 bytes, but the corresponding packet is really 1514 bytes (including the 14 byte Ethernet header). A more friendly solution would configure the MTU (as before), but then derive the packet buffer size by adding the MAC header length. Instead, we define the packet buffer size then derive the MTU. The MTU is not common currency in networking. On the wire, the only real issue is the MSS which is derived from MTU by subtracting the IP header and TCP header sizes (for the case of TCP). Now it is derived for the PKTSIZE by subtracting the IP header, the TCP header, and the MAC header sizes. So we should be all good and without the recurring 14 byte error in MTU's and MSS's. Squashed commit of the following: Trivial update to fix some spacing issues. net/: Rename several macros containing _MTU to _PKTSIZE. net/: Rename CONFIG_NET_SLIP_MTU to CONFIG_NET_SLIP_PKTSIZE and similarly for CONFIG_NET_TUN_MTU. These are not the MTU which does not include the size of the link layer header. These are the full size of the packet buffer memory (minus any GUARD bytes). net/: Rename CONFIG_NET_6LOWPAN_MTU to CONFIG_NET_6LOWPAN_PKTSIZE and similarly for CONFIG_NET_TUN_MTU. These are not the MTU which does not include the size of the link layer header. These are the full size of the packet buffer memory (minus any GUARD bytes). net/: Rename CONFIG_NET_ETH_MTU to CONFIG_NET_ETH_PKTSIZE. This is not the MTU which does not include the size of the link layer header. This is the full size of the packet buffer memory (minus any GUARD bytes). net/: Rename the file d_mtu in the network driver structure to d_pktsize. That value saved there is not the MTU. The packetsize is the memory large enough to hold the maximum packet PLUS the size of the link layer header. The MTU does not include the link layer header.
2018-07-04 22:10:40 +02:00
#define PKTMEM_ALIGNED_BUFSIZE ((CONFIG_NET_ETH_PKTSIZE + 1) & ~1)
#define PKTMEM_RX_START (PKTMEM_START + PKTMEM_SIZE / 2) /* Followed by RX buffer */
#define PKTMEM_RX_SIZE (PKTMEM_SIZE - PKTMEM_RX_START)
#define PKTMEM_RX_END (PKTMEM_START + PKTMEM_SIZE) /* RX buffer goes to the end of SRAM */
/* We use preinitialized TX descriptors */
#define ENC_NTXDESCR ((PKTMEM_RX_START - PKTMEM_START) / PKTMEM_ALIGNED_BUFSIZE)
/* This is a helper pointer for accessing the contents of the Ethernet header */
2014-06-25 17:57:52 +02:00
#define BUF ((struct eth_hdr_s *)priv->dev.d_buf)
/* Debug ********************************************************************/
#ifdef CONFIG_ENCX24J600_REGDEBUG
# define enc_wrdump(a,v) \
syslog(LOG_DEBUG, "ENCX24J600: %02x<-%04x\n", a, v);
# define enc_rddump(a,v) \
syslog(LOG_DEBUG, "ENCX24J600: %02x->%04x\n", a, v);
# define enc_bfsdump(a,m) \
syslog(LOG_DEBUG, "ENCX24J600: %02x|=%04x\n", a, m);
# define enc_bfcdump(a,m) \
syslog(LOG_DEBUG, "ENCX24J600: %02x&=~%04x\n", a, m);
# define enc_cmddump(c) \
syslog(LOG_DEBUG, "ENCX24J600: CMD: %02x\n", c);
# define enc_bmdump(c,b,s) \
syslog(LOG_DEBUG, "ENCX24J600: CMD: %02x buffer: %p length: %d\n", c, b, s);
#else
# define enc_wrdump(a,v)
# define enc_rddump(a,v)
# define enc_bfsdump(a,m)
# define enc_bfcdump(a,m)
# define enc_cmddump(c)
# define enc_bmdump(c,b,s)
#endif
/****************************************************************************
* Private Types
****************************************************************************/
/* The state of the interface */
enum enc_state_e
{
ENCSTATE_UNINIT = 0, /* The interface is in an uninitialized state */
ENCSTATE_DOWN, /* The interface is down */
ENCSTATE_UP, /* The interface is up */
ENCSTATE_RUNNING /* The interface is has a cable plugged in and is ready to use */
};
struct enc_descr_s
{
struct enc_descr_next *flink;
uint16_t addr;
uint16_t len;
};
/* The enc_driver_s encapsulates all state information for a single hardware
* interface
*/
struct enc_driver_s
{
/* Device control */
uint8_t ifstate; /* Interface state: See ENCSTATE_* */
uint8_t bank; /* Currently selected bank command */
uint16_t nextpkt; /* Next packet address */
FAR const struct enc_lower_s *lower; /* Low-level MCU-specific support */
/* Timing */
WDOG_ID txpoll; /* TX poll timer */
WDOG_ID txtimeout; /* TX timeout timer */
2016-01-24 15:21:55 +01:00
/* Avoid SPI accesses from the interrupt handler by using the work queue */
struct work_s irqwork; /* Interrupt continuation work queue support */
struct work_s towork; /* Tx timeout work queue support */
struct work_s pollwork; /* Poll timeout work queue support */
struct enc_descr_s txdescralloc[ENC_NTXDESCR];
struct enc_descr_s rxdescralloc[CONFIG_ENCX24J600_NRXDESCR];
sq_queue_t txfreedescr; /* Free inititialized TX descriptors */
sq_queue_t rxfreedescr; /* Free RX descriptors */
sq_queue_t txqueue; /* Enqueued descriptors waiting for transmission */
sq_queue_t rxqueue; /* Unhandled incoming packets waiting for reception */
/* This is the contained SPI driver intstance */
FAR struct spi_dev_s *spi;
/* This holds the information visible to the NuttX network */
struct net_driver_s dev; /* Interface understood by the network */
};
/****************************************************************************
* Private Data
****************************************************************************/
/* A single packet buffer is used */
This commit attempts remove some long standard confusion in naming and some actual problems that result from the naming confusion. The basic problem is the standard MTU does not include the size of the Ethernet header. For clarity, I changed the naming of most things called MTU to PKTSIZE. For example, CONFIG_NET_ETH_MTU is now CONFIG_NET_ETH_PKTSIZE. This makes the user interface a little hostile. People thing of an MTU of 1500 bytes, but the corresponding packet is really 1514 bytes (including the 14 byte Ethernet header). A more friendly solution would configure the MTU (as before), but then derive the packet buffer size by adding the MAC header length. Instead, we define the packet buffer size then derive the MTU. The MTU is not common currency in networking. On the wire, the only real issue is the MSS which is derived from MTU by subtracting the IP header and TCP header sizes (for the case of TCP). Now it is derived for the PKTSIZE by subtracting the IP header, the TCP header, and the MAC header sizes. So we should be all good and without the recurring 14 byte error in MTU's and MSS's. Squashed commit of the following: Trivial update to fix some spacing issues. net/: Rename several macros containing _MTU to _PKTSIZE. net/: Rename CONFIG_NET_SLIP_MTU to CONFIG_NET_SLIP_PKTSIZE and similarly for CONFIG_NET_TUN_MTU. These are not the MTU which does not include the size of the link layer header. These are the full size of the packet buffer memory (minus any GUARD bytes). net/: Rename CONFIG_NET_6LOWPAN_MTU to CONFIG_NET_6LOWPAN_PKTSIZE and similarly for CONFIG_NET_TUN_MTU. These are not the MTU which does not include the size of the link layer header. These are the full size of the packet buffer memory (minus any GUARD bytes). net/: Rename CONFIG_NET_ETH_MTU to CONFIG_NET_ETH_PKTSIZE. This is not the MTU which does not include the size of the link layer header. This is the full size of the packet buffer memory (minus any GUARD bytes). net/: Rename the file d_mtu in the network driver structure to d_pktsize. That value saved there is not the MTU. The packetsize is the memory large enough to hold the maximum packet PLUS the size of the link layer header. The MTU does not include the link layer header.
2018-07-04 22:10:40 +02:00
static uint8_t g_pktbuf[MAX_NETDEV_PKTSIZE + CONFIG_NET_GUARDSIZE];
/* Driver status structure */
static struct enc_driver_s g_encx24j600[CONFIG_ENCX24J600_NINTERFACES];
/****************************************************************************
* Private Function Prototypes
****************************************************************************/
/* Low-level SPI helpers */
static void enc_lock(FAR struct enc_driver_s *priv);
static inline void enc_unlock(FAR struct enc_driver_s *priv);
/* SPI control register access */
static inline void enc_setethrst(FAR struct enc_driver_s *priv);
static void enc_setbank(FAR struct enc_driver_s *priv, uint8_t bank);
static uint16_t enc_rdreg(FAR struct enc_driver_s *priv, uint16_t ctrlreg);
static void enc_wrreg(FAR struct enc_driver_s *priv, uint16_t ctrlreg,
uint16_t wrdata);
static int enc_waitreg(FAR struct enc_driver_s *priv, uint16_t ctrlreg,
uint16_t bits, uint16_t value);
static void enc_bfs(FAR struct enc_driver_s *priv, uint16_t ctrlreg,
uint16_t bits);
static void enc_bfc(FAR struct enc_driver_s *priv, uint16_t ctrlreg,
uint16_t bits);
static void enc_cmd(FAR struct enc_driver_s *priv, uint8_t cmd, uint16_t arg);
#if 0 /* Sometimes useful */
static void enc_rxdump(FAR struct enc_driver_s *priv);
static void enc_txdump(FAR struct enc_driver_s *priv);
#endif
/* SPI buffer transfers */
static void enc_rdbuffer(FAR struct enc_driver_s *priv, FAR uint8_t *buffer,
size_t buflen);
static inline void enc_wrbuffer(FAR struct enc_driver_s *priv,
FAR const uint8_t *buffer, size_t buflen);
/* PHY register access */
static uint16_t enc_rdphy(FAR struct enc_driver_s *priv, uint8_t phyaddr);
static void enc_wrphy(FAR struct enc_driver_s *priv, uint8_t phyaddr,
uint16_t phydata);
/* Common TX logic */
static int enc_txenqueue(FAR struct enc_driver_s *priv);
static int enc_transmit(FAR struct enc_driver_s *priv);
static int enc_txpoll(struct net_driver_s *dev);
/* Common RX logic */
static struct enc_descr_s *enc_rxgetdescr(FAR struct enc_driver_s *priv);
static void enc_rxldpkt(FAR struct enc_driver_s *priv,
FAR struct enc_descr_s *descr);
static void enc_rxrmpkt(FAR struct enc_driver_s *priv,
FAR struct enc_descr_s *descr);
static void enc_rxdispatch(FAR struct enc_driver_s *priv);
/* Interrupt handling */
static void enc_linkstatus(FAR struct enc_driver_s *priv);
static void enc_txif(FAR struct enc_driver_s *priv);
static void enc_pktif(FAR struct enc_driver_s *priv);
static void enc_rxabtif(FAR struct enc_driver_s *priv);
static void enc_irqworker(FAR void *arg);
static int enc_interrupt(int irq, FAR void *context, FAR void *arg);
/* Watchdog timer expirations */
static void enc_toworker(FAR void *arg);
static void enc_txtimeout(int argc, uint32_t arg, ...);
static void enc_pollworker(FAR void *arg);
static void enc_polltimer(int argc, uint32_t arg, ...);
/* NuttX callback functions */
2014-06-28 00:48:12 +02:00
static int enc_ifup(struct net_driver_s *dev);
static int enc_ifdown(struct net_driver_s *dev);
static int enc_txavail(struct net_driver_s *dev);
#ifdef CONFIG_NET_MCASTGROUP
2014-06-28 00:48:12 +02:00
static int enc_addmac(struct net_driver_s *dev, FAR const uint8_t *mac);
static int enc_rmmac(struct net_driver_s *dev, FAR const uint8_t *mac);
#endif
/* Initialization */
static void enc_pwrsave(FAR struct enc_driver_s *priv);
static void enc_setmacaddr(FAR struct enc_driver_s *priv);
static void enc_resetbuffers(FAR struct enc_driver_s *priv);
static int enc_reset(FAR struct enc_driver_s *priv);
/****************************************************************************
* Private Functions
****************************************************************************/
/****************************************************************************
* Name: enc_lock
*
* Description:
* Select the SPI, locking and re-configuring if necessary
*
* Input Parameters:
* spi - Reference to the SPI driver structure
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
static void enc_lock(FAR struct enc_driver_s *priv)
{
/* Lock the SPI bus in case there are multiple devices competing for the SPI
* bus.
*/
SPI_LOCK(priv->spi, true);
/* Now make sure that the SPI bus is configured for the ENCX24J600 (it
* might have gotten configured for a different device while unlocked)
*/
SPI_SETMODE(priv->spi, CONFIG_ENCX24J600_SPIMODE);
SPI_SETBITS(priv->spi, 8);
SPI_HWFEATURES(priv->spi, 0);
SPI_SETFREQUENCY(priv->spi, CONFIG_ENCX24J600_FREQUENCY);
}
/****************************************************************************
* Name: enc_unlock
*
* Description:
* De-select the SPI
*
* Input Parameters:
* spi - Reference to the SPI driver structure
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
static inline void enc_unlock(FAR struct enc_driver_s *priv)
{
/* Relinquish the lock on the bus. */
SPI_LOCK(priv->spi, false);
}
/****************************************************************************
* Name: enc_cmd
*
* Description:
* Execute two byte command.
*
* Input Parameters:
* priv - Reference to the driver state structure
* cmd - ENCX24J600 two-byte command
* arg - Two byte argument to the command
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
static void enc_cmd(FAR struct enc_driver_s *priv, uint8_t cmd, uint16_t arg)
{
DEBUGASSERT(priv && priv->spi);
/* Select ENCX24J600 chip */
SPI_SELECT(priv->spi, SPIDEV_ETHERNET(0), true);
SPI_SEND(priv->spi, cmd); /* Clock out the command */
SPI_SEND(priv->spi, arg & 0xff); /* Clock out the low byte */
SPI_SEND(priv->spi, arg >> 8); /* Clock out the high byte */
/* De-select ENCX24J600 chip. */
SPI_SELECT(priv->spi, SPIDEV_ETHERNET(0), false);
enc_wrdump(cmd, arg);
}
/****************************************************************************
* Name: enc_setethrst
*
* Description:
* Issues System Reset by setting ETHRST (ECON2<4>)
*
* Input Parameters:
* priv - Reference to the driver state structure
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
static inline void enc_setethrst(FAR struct enc_driver_s *priv)
{
DEBUGASSERT(priv && priv->spi);
/* Select ENCX24J600 chip */
SPI_SELECT(priv->spi, SPIDEV_ETHERNET(0), true);
/* Send the system reset command. */
SPI_SEND(priv->spi, ENC_SETETHRST);
up_udelay(25);
/* De-select ENCX24J600 chip. */
SPI_SELECT(priv->spi, SPIDEV_ETHERNET(0), false);
enc_cmddump(ENC_SETETHRST);
}
/****************************************************************************
* Name: enc_setbank
*
* Description:
* Set the bank for the next control register access.
*
* Assumption:
* The caller has exclusive access to the SPI bus
*
* Input Parameters:
* priv - Reference to the driver state structure
* bank - SPI command to select the bank with
*
* Returned Value:
* None
*
* Assumptions:
* The chip is selected and SPI is ready for communication.
*
****************************************************************************/
static void enc_setbank(FAR struct enc_driver_s *priv, uint8_t bank)
{
/* Check if a bank has to be set and if the bank setting has changed.
* For registers that are available on all banks, the bank command is set
* to 0.
*/
if (bank != 0 && bank != priv->bank)
{
/* Select bank with supplied command */
SPI_SEND(priv->spi, bank);
/* Then remember the bank setting */
priv->bank = bank;
}
}
/****************************************************************************
* Name: enc_rdreg
*
* Description:
* Read one word from a control register using the RCR command.
*
* Input Parameters:
* priv - Reference to the driver state structure
* ctrlreg - Bit encoded address of banked register to read
*
* Returned Value:
* The byte read from the banked register
*
* Assumptions:
*
****************************************************************************/
static uint16_t enc_rdreg(FAR struct enc_driver_s *priv, uint16_t ctrlreg)
{
uint16_t rddata;
DEBUGASSERT(priv && priv->spi);
DEBUGASSERT((ctrlreg & 0xe0) == 0); /* banked regeitsers only */
SPI_SELECT(priv->spi, SPIDEV_ETHERNET(0), true);
enc_setbank(priv, GETBANK(ctrlreg));
SPI_SEND(priv->spi, ENC_RCR | GETADDR(ctrlreg));
2013-09-25 16:26:56 +02:00
rddata = SPI_SEND(priv->spi, 0); /* Clock in the low byte */
rddata |= SPI_SEND(priv->spi, 0) << 8; /* Clock in the high byte */
SPI_SELECT(priv->spi, SPIDEV_ETHERNET(0), false);
enc_rddump(GETADDR(ctrlreg), rddata);
return rddata;
}
/****************************************************************************
* Name: enc_wrreg
*
* Description:
* Write one word to a control register using the WCR command.
*
* Input Parameters:
* priv - Reference to the driver state structure
* ctrlreg - Bit encoded address of banked register to write
* wrdata - The data to send
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
static void enc_wrreg(FAR struct enc_driver_s *priv, uint16_t ctrlreg,
uint16_t wrdata)
{
DEBUGASSERT(priv && priv->spi);
DEBUGASSERT((ctrlreg & 0xe0) == 0); /* banked regeitsers only */
SPI_SELECT(priv->spi, SPIDEV_ETHERNET(0), true);
enc_setbank(priv, GETBANK(ctrlreg));
SPI_SEND(priv->spi, ENC_WCR | GETADDR(ctrlreg));
2013-09-25 16:26:56 +02:00
SPI_SEND(priv->spi, wrdata & 0xff); /* Clock out the low byte */
SPI_SEND(priv->spi, wrdata >> 8); /* Clock out the high byte */
SPI_SELECT(priv->spi, SPIDEV_ETHERNET(0), false);
enc_wrdump(GETADDR(ctrlreg), wrdata);
}
/****************************************************************************
* Name: enc_waitbreg
*
* Description:
* Wait until banked register bit(s) take a specific value (or a timeout
* occurs).
*
* Input Parameters:
* priv - Reference to the driver state structure
* ctrlreg - Bit encoded address of banked register to check
* bits - The bits to check (a mask)
* value - The value of the bits to return (value under mask)
*
* Returned Value:
* OK on success, negated errno on failure
*
* Assumptions:
*
****************************************************************************/
static int enc_waitreg(FAR struct enc_driver_s *priv, uint16_t ctrlreg,
uint16_t bits, uint16_t value)
{
clock_t start = clock_systimer();
clock_t elapsed;
uint16_t rddata;
/* Loop until the exit condition is met */
do
{
/* Read the byte from the requested banked register */
rddata = enc_rdreg(priv, ctrlreg);
elapsed = clock_systimer() - start;
}
2013-09-25 16:26:56 +02:00
while ((rddata & bits) != value && elapsed < ENC_REGPOLLTIMEOUT);
return (rddata & bits) == value ? OK : -ETIMEDOUT;
}
/****************************************************************************
* Name: enc_bfs
*
* Description:
* Bit Field Set.
*
* Input Parameters:
* priv - Reference to the driver state structure
* ctrlreg - Bit encoded address of banked register to set bits in
* bits - The bits to set (a mask)
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
static void enc_bfs(FAR struct enc_driver_s *priv, uint16_t ctrlreg,
uint16_t bits)
{
DEBUGASSERT(priv && priv->spi);
/* Select ENCX24J600 chip */
SPI_SELECT(priv->spi, SPIDEV_ETHERNET(0), true);
/* Set the bank */
enc_setbank(priv, GETBANK(ctrlreg));
/* Send the BFS command and data. The sequence requires 24-clocks:
* 8 to clock out the cmd + 16 to clock out the data.
*/
SPI_SEND(priv->spi, ENC_BFS | GETADDR(ctrlreg)); /* Clock out the command */
SPI_SEND(priv->spi, bits & 0xff); /* Clock out the low byte */
SPI_SEND(priv->spi, bits >> 8); /* Clock out the high byte */
/* De-select ENCX24J600 chip. */
SPI_SELECT(priv->spi, SPIDEV_ETHERNET(0), false);
enc_bfsdump(GETADDR(ctrlreg), bits);
}
/****************************************************************************
* Name: enc_bfc
*
* Description:
* Bit Field Clear.
*
* Input Parameters:
* priv - Reference to the driver state structure
* ctrlreg - Bit encoded address of banked register to clear bits in
* bits - The bits to clear (a mask)
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
static void enc_bfc(FAR struct enc_driver_s *priv, uint16_t ctrlreg,
uint16_t bits)
{
DEBUGASSERT(priv && priv->spi);
/* Select ENCX24J600 chip */
SPI_SELECT(priv->spi, SPIDEV_ETHERNET(0), true);
/* Set the bank */
enc_setbank(priv, GETBANK(ctrlreg));
/* Send the BFC command and data. The sequence requires 24-clocks:
* 8 to clock out the cmd + 16 to clock out the data.
*/
SPI_SEND(priv->spi, ENC_BFC | GETADDR(ctrlreg)); /* Clock out the command */
SPI_SEND(priv->spi, bits & 0xff); /* Clock out the low byte */
SPI_SEND(priv->spi, bits >> 8); /* Clock out the high byte */
/* De-select ENCX24J600 chip. */
SPI_SELECT(priv->spi, SPIDEV_ETHERNET(0), false);
enc_bfcdump(GETADDR(ctrlreg), bits);
}
/****************************************************************************
* Name: enc_txdump enc_rxdump
*
* Description:
* Dump registers associated with receiving or sending packets.
*
* Input Parameters:
* priv - Reference to the driver state structure
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
#if 0 /* Sometimes useful */
static void enc_rxdump(FAR struct enc_driver_s *priv)
{
syslog(LOG_DEBUG, "Rx Registers:\n");
syslog(LOG_DEBUG, " EIE: %02x EIR: %02x\n",
enc_rdgreg(priv, ENC_EIE), enc_rdgreg(priv, ENC_EIR));
syslog(LOG_DEBUG, " ESTAT: %02x ECON1: %02x ECON2: %02x\n",
enc_rdgreg(priv, ENC_ESTAT), enc_rdgreg(priv, ENC_ECON1),
enc_rdgreg(priv, ENC_ECON2));
syslog(LOG_DEBUG, " ERXST: %02x %02x\n",
enc_rdbreg(priv, ENC_ERXSTH), enc_rdbreg(priv, ENC_ERXSTL));
syslog(LOG_DEBUG, " ERXND: %02x %02x\n",
enc_rdbreg(priv, ENC_ERXNDH), enc_rdbreg(priv, ENC_ERXNDL));
syslog(LOG_DEBUG, " ERXRDPT: %02x %02x\n",
enc_rdbreg(priv, ENC_ERXRDPTH), enc_rdbreg(priv, ENC_ERXRDPTL));
syslog(LOG_DEBUG, " ERXFCON: %02x EPKTCNT: %02x\n",
enc_rdbreg(priv, ENC_ERXFCON), enc_rdbreg(priv, ENC_EPKTCNT));
syslog(LOG_DEBUG, " MACON1: %02x MACON3: %02x\n",
enc_rdbreg(priv, ENC_MACON1), enc_rdbreg(priv, ENC_MACON3));
syslog(LOG_DEBUG, " MAMXFL: %02x %02x\n",
enc_rdbreg(priv, ENC_MAMXFLH), enc_rdbreg(priv, ENC_MAMXFLL));
syslog(LOG_DEBUG, " MAADR: %02x:%02x:%02x:%02x:%02x:%02x\n",
enc_rdbreg(priv, ENC_MAADR1), enc_rdbreg(priv, ENC_MAADR2),
enc_rdbreg(priv, ENC_MAADR3), enc_rdbreg(priv, ENC_MAADR4),
enc_rdbreg(priv, ENC_MAADR5), enc_rdbreg(priv, ENC_MAADR6));
}
#endif
#if 0 /* Sometimes useful */
static void enc_txdump(FAR struct enc_driver_s *priv)
{
syslog(LOG_DEBUG, "Tx Registers:\n");
syslog(LOG_DEBUG, " EIE: %02x EIR: %02x\n",
enc_rdgreg(priv, ENC_EIE), enc_rdgreg(priv, ENC_EIR));
syslog(LOG_DEBUG, " ESTAT: %02x ECON1: %02x\n",
enc_rdgreg(priv, ENC_ESTAT), enc_rdgreg(priv, ENC_ECON1));
syslog(LOG_DEBUG, " ETXST: %02x %02x\n",
enc_rdbreg(priv, ENC_ETXSTH), enc_rdbreg(priv, ENC_ETXSTL));
syslog(LOG_DEBUG, " ETXND: %02x %02x\n",
enc_rdbreg(priv, ENC_ETXNDH), enc_rdbreg(priv, ENC_ETXNDL));
syslog(LOG_DEBUG, " MACON1: %02x MACON3: %02x MACON4: %02x\n",
enc_rdbreg(priv, ENC_MACON1), enc_rdbreg(priv, ENC_MACON3),
enc_rdbreg(priv, ENC_MACON4));
syslog(LOG_DEBUG, " MACON1: %02x MACON3: %02x MACON4: %02x\n",
enc_rdbreg(priv, ENC_MACON1), enc_rdbreg(priv, ENC_MACON3),
enc_rdbreg(priv, ENC_MACON4));
syslog(LOG_DEBUG, " MABBIPG: %02x MAIPG %02x %02x\n",
enc_rdbreg(priv, ENC_MABBIPG), enc_rdbreg(priv, ENC_MAIPGH),
enc_rdbreg(priv, ENC_MAIPGL));
syslog(LOG_DEBUG, " MACLCON1: %02x MACLCON2: %02x\n",
enc_rdbreg(priv, ENC_MACLCON1), enc_rdbreg(priv, ENC_MACLCON2));
syslog(LOG_DEBUG, " MAMXFL: %02x %02x\n",
enc_rdbreg(priv, ENC_MAMXFLH), enc_rdbreg(priv, ENC_MAMXFLL));
}
#endif
/****************************************************************************
* Name: enc_rdbuffer
*
* Description:
* Read a buffer of data from RX Data Buffer.
*
* Input Parameters:
* priv - Reference to the driver state structure
* buffer - A pointer to the buffer to read into
* buflen - The number of bytes to read
*
* Returned Value:
* None
*
* Assumptions:
* RX Data pointer is set to the correct address
*
****************************************************************************/
static void enc_rdbuffer(FAR struct enc_driver_s *priv, FAR uint8_t *buffer,
size_t buflen)
{
DEBUGASSERT(priv && priv->spi);
/* Select ENCX24J600 chip */
SPI_SELECT(priv->spi, SPIDEV_ETHERNET(0), true);
/* Send the read buffer memory command (ignoring the response) */
SPI_SEND(priv->spi, ENC_RRXDATA);
/* Then read the buffer data */
SPI_RECVBLOCK(priv->spi, buffer, buflen);
/* De-select ENCX24J600 chip. */
SPI_SELECT(priv->spi, SPIDEV_ETHERNET(0), false);
enc_bmdump(ENC_RRXDATA, buffer, buflen);
}
/****************************************************************************
* Name: enc_wrbuffer
*
* Description:
* Write a buffer of data.
*
* Input Parameters:
* priv - Reference to the driver state structure
* buffer - A pointer to the buffer to write from
* buflen - The number of bytes to write
*
* Returned Value:
* None
*
* Assumptions:
* General Purpose Write pointer is set to the correct address
*
****************************************************************************/
static inline void enc_wrbuffer(FAR struct enc_driver_s *priv,
FAR const uint8_t *buffer, size_t buflen)
{
DEBUGASSERT(priv && priv->spi);
SPI_SELECT(priv->spi, SPIDEV_ETHERNET(0), true);
SPI_SEND(priv->spi, ENC_WGPDATA);
SPI_SNDBLOCK(priv->spi, buffer, buflen);
SPI_SELECT(priv->spi, SPIDEV_ETHERNET(0), false);
enc_bmdump(ENC_WGPDATA, buffer, buflen);
}
/****************************************************************************
* Name: enc_rdphy
*
* Description:
* Read 16-bits of PHY data.
*
* Input Parameters:
* priv - Reference to the driver state structure
* phyaddr - The PHY register address
*
* Returned Value:
* 16-bit value read from the PHY
*
* Assumptions:
*
****************************************************************************/
static uint16_t enc_rdphy(FAR struct enc_driver_s *priv, uint8_t phyaddr)
{
uint16_t data = 0;
/* "To read from a PHY register:
* 1. Write the address of the PHY register to read from into the MIREGADR
* register (Register 3-1). Make sure to also set reserved bit 8 of
* this register.
*/
enc_wrreg(priv, ENC_MIREGADR, phyaddr);
/* 2. Set the MIIRD bit (MICMD<0>, Register 3-2). The read operation
* begins and the BUSY bit (MISTAT<0>, Register 3-3) is automatically
* set by hardware.
*/
enc_bfs(priv, ENC_MICMD, MICMD_MIIRD);
/* 3. Wait 25.6 μs. Poll the BUSY (MISTAT<0>) bit to be certain that the
* operation is complete. While busy, the host controller should not
* start any MIISCAN operations or write to the MIWR register. When the
* MAC has obtained the register contents, the BUSY bit will clear
* itself.
*/
up_udelay(26);
if (enc_waitreg(priv, ENC_MISTAT, MISTAT_BUSY, 0x00) == OK)
{
/* 4. Clear the MIIRD (MICMD<0>) bit. */
enc_bfc(priv, ENC_MICMD, MICMD_MIIRD);
/* 5. Read the desired data from the MIRD register. For 8-bit
* interfaces, the order that these bytes are read is unimportant."
*/
data = enc_rdreg(priv, ENC_MIRD);
}
return data;
}
/****************************************************************************
* Name: enc_wrphy
*
* Description:
* write 16-bits of PHY data.
*
* Input Parameters:
* priv - Reference to the driver state structure
* phyaddr - The PHY register address
* phydata - 16-bit data to write to the PHY
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
static void enc_wrphy(FAR struct enc_driver_s *priv, uint8_t phyaddr,
uint16_t phydata)
{
/* "To write to a PHY register:
*
* 1. Write the address of the PHY register to write to into the MIREGADR
* register. Make sure to also set reserved bit 8 of this register.
*/
enc_wrreg(priv, ENC_MIREGADR, 0x0100 | phyaddr);
/* 2. Write the 16 bits of data into the MIWR register. The low byte must
* be written first, followed by the high byte.
*/
enc_wrreg(priv, ENC_MIWR, phydata);
/* 3. Writing to the high byte of MIWR begins the MIIM transaction and
* the BUSY (MISTAT<0>) bit is automatically set by hardware.
*
* The PHY register is written after the MIIM operation completes, which
* takes 25.6 μs. When the write operation has completed, the BUSY bit
* clears itself. The host controller should not start any MIISCAN, MIWR
* or MIIRD operations while the BUSY bit is set.
*/
up_udelay(26);
enc_waitreg(priv, ENC_MISTAT, MISTAT_BUSY, 0);
}
/****************************************************************************
* Name: enc_transmit
*
* Description:
* Start hardware transmission. Called either from:
*
* - pkif interrupt when an application responds to the receipt of data
* by trying to send something, or
* - From watchdog based polling.
*
* Input Parameters:
* priv - Reference to the driver state structure
*
* Returned Value:
* OK on success; a negated errno on failure
*
* Assumptions:
*
****************************************************************************/
static int enc_transmit(FAR struct enc_driver_s *priv)
{
FAR struct enc_descr_s *descr;
/* dequeue next packet to transmit */
descr = (FAR struct enc_descr_s *)sq_remfirst(&priv->txqueue);
DEBUGASSERT(descr != NULL);
/* Verify that the hardware is ready to send another packet. The driver
* starts a transmission process by setting ECON1.TXRTS. When the packet is
* finished transmitting or is aborted due to an error/cancellation, the
* ECON1.TXRTS bit will be cleared.
*
* NOTE: If we got here, then we have committed to sending a packet.
* higher level logic must have assured that (1) there is no transmission
* in progress, and that (2) TX-related interrupts are disabled.
*/
DEBUGASSERT((enc_rdreg(priv, ENC_ECON1) & ECON1_TXRTS) == 0);
/* Set TXStart and TXLen registers. */
enc_wrreg(priv, ENC_ETXST, descr->addr);
enc_wrreg(priv, ENC_ETXLEN, descr->len);
/* Set TXRTS to send the packet in the transmit buffer */
enc_bfs(priv, ENC_ECON1, ECON1_TXRTS);
/* Setup the TX timeout watchdog (perhaps restarting the timer). Note:
* Is there a race condition. Could the TXIF interrupt occur before
* the timer is started?
*/
wd_start(priv->txtimeout, ENC_TXTIMEOUT, enc_txtimeout, 1,
(wdparm_t)priv);
/* free the descriptor */
sq_addlast((FAR sq_entry_t *)descr, &priv->txfreedescr);
return OK;
}
/****************************************************************************
* Name: enc_txenqueue
*
* Description:
* Write packet from d_buf to the enc's SRAM if a free descriptor is
* available. The filled descriptor is enqueued for transmission.
*
* Input Parameters:
* dev - Reference to the NuttX driver state structure
*
* Returned Value:
* OK on success; a negated errno on failure
*
* Assumptions:
* A packet is available in d_buf.
* Interrupts are enabled but the caller holds the network lock.
*
****************************************************************************/
static int enc_txenqueue(FAR struct enc_driver_s *priv)
{
int ret = OK;
FAR struct enc_descr_s *descr;
DEBUGASSERT(priv->dev.d_len > 0);
/* Increment statistics */
NETDEV_TXPACKETS(&priv->dev);
descr = (FAR struct enc_descr_s *)sq_remfirst(&priv->txfreedescr);
if (descr != NULL)
{
enc_dumppacket("Write packet to enc SRAM", priv->dev.d_buf,
priv->dev.d_len);
2013-09-25 16:26:56 +02:00
/* Copy the packet into the transmit buffer described by the current
* tx descriptor
*/
enc_cmd(priv, ENC_WGPWRPT, descr->addr);
enc_wrbuffer(priv, priv->dev.d_buf, priv->dev.d_len);
/* store packet length */
descr->len = priv->dev.d_len;
/* enqueue packet */
sq_addlast((FAR sq_entry_t *)descr, &priv->txqueue);
/* if currently no transmission is active, trigger the transmission */
if ((enc_rdreg(priv, ENC_ECON1) & ECON1_TXRTS) == 0)
{
enc_transmit(priv);
}
}
else
{
nerr("ERROR: no free descriptors\n");
ret = -ENOMEM;
}
return ret;
}
/****************************************************************************
* Name: enc_txpoll
*
* Description:
* Enqueues network packets if available.
* This is a callback from devif_poll(). devif_poll() may be called:
*
* 1. When the preceding TX packet send is complete,
* 2. When the preceding TX packet send timedout and the interface is reset
* 3. During normal TX polling
*
* Input Parameters:
* dev - Reference to the NuttX driver state structure
*
* Returned Value:
* OK on success; a negated errno on failure
*
* Assumptions:
* Interrupts are enabled but the caller holds the network lock.
*
****************************************************************************/
static int enc_txpoll(struct net_driver_s *dev)
{
FAR struct enc_driver_s *priv = (FAR struct enc_driver_s *)dev->d_private;
int ret = OK;
/* If the polling resulted in data that should be sent out on the network,
* the field d_len is set to a value > 0.
*/
ninfo("Poll result: d_len=%d\n", priv->dev.d_len);
if (priv->dev.d_len > 0)
{
/* Look up the destination MAC address and add it to the Ethernet
* header.
*/
#ifdef CONFIG_NET_IPv4
#ifdef CONFIG_NET_IPv6
if (IFF_IS_IPv4(priv->dev.d_flags))
#endif
{
arp_out(&priv->dev);
}
#endif /* CONFIG_NET_IPv4 */
#ifdef CONFIG_NET_IPv6
#ifdef CONFIG_NET_IPv4
else
#endif
{
neighbor_out(&priv->dev);
}
#endif /* CONFIG_NET_IPv6 */
if (!devif_loopback(&priv->dev))
{
/* Send the packet */
ret = enc_txenqueue(priv);
}
}
/* If zero is returned, the polling will continue until all connections have
* been examined.
*/
return ret;
}
/****************************************************************************
* Name: enc_linkstatus
*
* Description:
* The current link status can be obtained from the PHSTAT1.LLSTAT or
* PHSTAT2.LSTAT.
*
* Input Parameters:
* priv - Reference to the driver state structure
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
static void enc_linkstatus(FAR struct enc_driver_s *priv)
{
uint16_t regval;
/* Before transmitting the first packet after link establishment or
* auto-negotiation, the MAC duplex configuration must be manually set to
* match the duplex configuration of the PHY. To do this, configure
* FULDPX (MACON2<0>) to match PHYDPX (ESTAT<10>).
*/
regval = enc_rdreg(priv, ENC_ESTAT);
if (regval & ESTAT_PHYLNK)
{
if (regval & ESTAT_PHYDPX)
{
/* Configure full-duplex */
enc_wrreg(priv, ENC_MABBIPG, 0x15);
enc_bfs(priv, ENC_MACON2, MACON2_FULDPX);
}
else
{
/* Configure half-duplex */
enc_wrreg(priv, ENC_MABBIPG, 0x12);
enc_bfc(priv, ENC_MACON2, MACON2_FULDPX);
}
netdev_carrier_on(&priv->dev);
priv->ifstate = ENCSTATE_RUNNING;
}
else
{
netdev_carrier_off(&priv->dev);
priv->ifstate = ENCSTATE_UP;
}
}
/****************************************************************************
* Name: enc_txif
*
* Description:
* An TXIF interrupt was received indicating that the last TX packet(s) is
* done
*
* Input Parameters:
* priv - Reference to the driver state structure
*
* Returned Value:
* None
*
* Assumptions:
* Interrupts are enabled but the caller holds the network lock.
*
****************************************************************************/
static void enc_txif(FAR struct enc_driver_s *priv)
{
NETDEV_TXDONE(&priv->dev);
if (sq_empty(&priv->txqueue))
{
/* If no further xmits are pending, then cancel the TX timeout */
wd_cancel(priv->txtimeout);
2014-07-03 01:23:25 +02:00
/* Poll for TX packets from the networking layer */
devif_poll(&priv->dev, enc_txpoll);
}
else
{
/* Process txqueue */
enc_transmit(priv);
}
}
/****************************************************************************
* Name: enc_rxldpkt
*
* Description:
2014-07-03 01:23:25 +02:00
* Load packet from the enc's RX buffer to the driver d_buf.
*
* Input Parameters:
* priv - Reference to the driver state structure
* descr - Reference to the descriptor that should be loaded
*
* Returned Value:
* None
*
* Assumptions:
* Interrupts are enabled but the caller holds the network lock.
*
****************************************************************************/
static void enc_rxldpkt(FAR struct enc_driver_s *priv,
FAR struct enc_descr_s *descr)
{
DEBUGASSERT(priv != NULL && descr != NULL);
ninfo("load packet @%04x len: %d\n", descr->addr, descr->len);
/* Set the rx data pointer to the start of the received packet (ERXRDPT) */
enc_cmd(priv, ENC_WRXRDPT, descr->addr);
/* Save the packet length (without the 4 byte CRC) in priv->dev.d_len */
priv->dev.d_len = descr->len - 4;
/* Copy the data data from the receive buffer to priv->dev.d_buf */
enc_rdbuffer(priv, priv->dev.d_buf, priv->dev.d_len);
enc_dumppacket("loaded RX packet", priv->dev.d_buf, priv->dev.d_len);
}
/****************************************************************************
* Name: enc_rxgetdescr
*
* Description:
* Check for a free descriptor in the free list. If no free descriptor is
* available a pending descriptor will be freed and returned
*
* Input Parameters:
* priv - Reference to the driver state structure
*
* Returned Value:
* A free rx descriptor
*
* Assumptions:
* Interrupts are enabled but the caller holds the network lock.
*
****************************************************************************/
static struct enc_descr_s *enc_rxgetdescr(FAR struct enc_driver_s *priv)
{
if (sq_empty(&priv->rxfreedescr))
{
DEBUGASSERT(sq_peek(&priv->rxqueue) != NULL);
/* Packets are held in the enc's SRAM until the space is needed */
enc_rxrmpkt(priv, (FAR struct enc_descr_s *)sq_peek(&priv->rxqueue));
}
return (FAR struct enc_descr_s *)sq_remfirst(&priv->rxfreedescr);
}
/****************************************************************************
* Name: enc_rxrmpkt
*
* Description:
* Remove packet from the RX queue and free the block of memory in the enc's
* SRAM.
*
* Input Parameters:
* priv - Reference to the driver state structure
* descr - Reference to the descriptor that should be freed
*
* Returned Value:
* None
*
* Assumptions:
* Interrupts are enabled but the caller holds the network lock.
*
****************************************************************************/
static void enc_rxrmpkt(FAR struct enc_driver_s *priv,
FAR struct enc_descr_s *descr)
{
uint16_t addr;
ninfo("free descr: %p\n", descr);
/* If it is the last descriptor in the queue, advance ERXTAIL.
* This way it is possible that gaps occur. Maybe pending packets
* can be reordered th enc's DMA to free RX space?
*/
if (descr != NULL)
{
if (descr == (FAR struct enc_descr_s *)sq_peek(&priv->rxqueue))
{
/* Wrap address properly around */
addr = (descr->addr - PKTMEM_RX_START + descr->len - 2 +
PKTMEM_RX_SIZE)
% PKTMEM_RX_SIZE + PKTMEM_RX_START;
DEBUGASSERT(addr >= PKTMEM_RX_START && addr < PKTMEM_RX_END);
ninfo("ERXTAIL %04x\n", addr);
enc_wrreg(priv, ENC_ERXTAIL, addr);
/* Remove packet from RX queue */
sq_remfirst(&priv->rxqueue);
}
else
{
/* Remove packet from RX queue */
sq_rem((FAR sq_entry_t *)descr, &priv->rxqueue);
}
sq_addlast((FAR sq_entry_t *)descr, &priv->rxfreedescr);
}
}
/****************************************************************************
* Name: enc_rxdispatch
*
* Description:
* Give the newly received packet to the network.
*
* Input Parameters:
* priv - Reference to the driver state structure
*
* Returned Value:
* None
*
* Assumptions:
* Interrupts are enabled but the caller holds the network lock.
*
****************************************************************************/
static void enc_rxdispatch(FAR struct enc_driver_s *priv)
{
FAR struct enc_descr_s *descr;
struct enc_descr_s *next;
/* Process the RX queue */
descr = (FAR struct enc_descr_s *)sq_peek(&priv->rxqueue);
while (descr != NULL)
{
/* Store the next pointer, because removing the item from list will set
* flink to NULL
*/
next = (FAR struct enc_descr_s *)sq_next(descr);
/* Load the packet from the enc's SRAM */
enc_rxldpkt(priv, descr);
#ifdef CONFIG_NET_PKT
/* When packet sockets are enabled, feed the frame into the packet tap */
pkt_input(&priv->dev);
#endif
/* We only accept IP packets of the configured type and ARP packets */
#ifdef CONFIG_NET_IPv4
2014-07-04 23:40:49 +02:00
if (BUF->type == HTONS(ETHTYPE_IP))
{
ninfo("IPv4 frame\n");
NETDEV_RXIPV4(&priv->dev);
/* Handle ARP on input then give the IPv4 packet to the network
* layer
*/
arp_ipin(&priv->dev);
ipv4_input(&priv->dev);
/* Free the packet */
enc_rxrmpkt(priv, descr);
/* If the above function invocation resulted in data that should be
* sent out on the network, the field d_len will set to a value
* > 0.
*/
if (priv->dev.d_len > 0)
{
/* Update the Ethernet header with the correct MAC address */
#ifdef CONFIG_NET_IPv6
if (IFF_IS_IPv4(priv->dev.d_flags))
#endif
{
arp_out(&priv->dev);
}
#ifdef CONFIG_NET_IPv6
else
{
neighbor_out(&priv->dev);
}
#endif
/* And send the packet */
enc_txenqueue(priv);
}
}
else
#endif
#ifdef CONFIG_NET_IPv6
if (BUF->type == HTONS(ETHTYPE_IP6))
{
2020-03-03 16:11:57 +01:00
ninfo("IPv6 frame\n");
NETDEV_RXIPV6(&priv->dev);
/* Give the IPv6 packet to the network layer */
ipv6_input(&priv->dev);
/* Free the packet */
enc_rxrmpkt(priv, descr);
/* If the above function invocation resulted in data that should be
* sent out on the network, the field d_len will set to a value
* > 0.
*/
if (priv->dev.d_len > 0)
{
/* Update the Ethernet header with the correct MAC address */
#ifdef CONFIG_NET_IPv4
if (IFF_IS_IPv4(priv->dev.d_flags))
{
arp_out(&priv->dev);
}
else
#endif
#ifdef CONFIG_NET_IPv6
{
neighbor_out(&priv->dev);
}
#endif
/* And send the packet */
enc_txenqueue(priv);
}
}
else
#endif
#ifdef CONFIG_NET_ARP
if (BUF->type == htons(ETHTYPE_ARP))
{
ninfo("ARP packet received (%02x)\n", BUF->type);
NETDEV_RXARP(&priv->dev);
arp_arpin(&priv->dev);
/* ARP packets are freed immediately */
enc_rxrmpkt(priv, descr);
/* If the above function invocation resulted in data that should be
* sent out on the network, the field d_len will set to a value
* > 0.
*/
if (priv->dev.d_len > 0)
{
enc_txenqueue(priv);
}
}
else
#endif
{
/* free unsupported packet */
enc_rxrmpkt(priv, descr);
nerr("ERROR: Unsupported packet type dropped (%02x)\n",
htons(BUF->type));
NETDEV_RXDROPPED(&priv->dev);
}
descr = next;
}
}
/****************************************************************************
* Name: enc_pktif
*
* Description:
* An interrupt was received indicating the availability of a new RX packet
*
* Input Parameters:
* priv - Reference to the driver state structure
*
* Returned Value:
* None
*
* Assumptions:
* Interrupts are enabled but the caller holds the network lock.
*
****************************************************************************/
static void enc_pktif(FAR struct enc_driver_s *priv)
{
FAR struct enc_descr_s *descr;
uint8_t rsv[8];
uint16_t pktlen;
uint32_t rxstat;
uint16_t curpkt;
int pktcnt;
DEBUGASSERT(priv->nextpkt >= PKTMEM_RX_START &&
priv->nextpkt < PKTMEM_RX_END);
/* Enqueue all pending packets to the RX queue until PKTCNT == 0 or
* no more descriptors are available.
*/
pktcnt = (enc_rdreg(priv, ENC_ESTAT) & ESTAT_PKTCNT_MASK) >>
ESTAT_PKTCNT_SHIFT;
while (pktcnt > 0)
{
curpkt = priv->nextpkt;
/* Set the rx data pointer to the start of the received packet (ERXRDPT) */
enc_cmd(priv, ENC_WRXRDPT, curpkt);
/* Read the next packet pointer and the 6 byte read status vector (RSV)
* at the beginning of the received packet. (ERXRDPT should auto-
* increment and wrap to the beginning of the read buffer as necessary)
*/
enc_rdbuffer(priv, rsv, 8);
/* Decode the new next packet pointer, and the RSV. The
* RSV is encoded as:
*
* Bits 0-15: Indicates length of the received frame. This includes
* the destination address, source address, type/length,
* data, padding and CRC fields. This field is stored in
* little-endian format.
* Bits 16-47: Bit encoded RX status.
*/
priv->nextpkt = (uint16_t)rsv[1] << 8 | (uint16_t)rsv[0];
pktlen = (uint16_t)rsv[3] << 8 | (uint16_t)rsv[2];
rxstat = (uint32_t)rsv[7] << 24 | (uint32_t)rsv[6] << 16 |
(uint32_t)rsv[5] << 8 | (uint32_t)rsv[4];
ninfo("Receiving packet, nextpkt: %04x pktlen: %d rxstat: %08x "
"pktcnt: %d\n",
priv->nextpkt, pktlen, rxstat, pktcnt);
/* We enqueue the packet first and remove it later if its faulty.
* This way we avoid freeing packets that are not processed yet.
*/
descr = enc_rxgetdescr(priv);
/* Store the start address of the frame without the enc's header */
descr->addr = curpkt + 8;
descr->len = pktlen;
sq_addlast((FAR sq_entry_t *)descr, &priv->rxqueue);
/* Check if the packet was received OK */
if ((rxstat & RXSTAT_OK) == 0)
{
nerr("ERROR: RXSTAT: %08x\n", rxstat);
/* Discard packet */
enc_rxrmpkt(priv, descr);
NETDEV_RXERRORS(&priv->dev);
}
/* Check for a usable packet length (4 added for the CRC) */
else if (pktlen > (CONFIG_NET_ETH_PKTSIZE + 4) ||
pktlen <= (ETH_HDRLEN + 4))
{
nerr("ERROR: Bad packet size dropped (%d)\n", pktlen);
/* Discard packet */
enc_rxrmpkt(priv, descr);
NETDEV_RXERRORS(&priv->dev);
}
/* Decrement PKTCNT */
enc_bfs(priv, ENC_ECON1, ECON1_PKTDEC);
/* Try to process the packet */
enc_rxdispatch(priv);
/* Read out again, maybe there has another packet arrived */
pktcnt = (enc_rdreg(priv, ENC_ESTAT) & ESTAT_PKTCNT_MASK) >>
ESTAT_PKTCNT_SHIFT;
}
}
/****************************************************************************
* Name: enc_rxabtif
*
* Description:
* An interrupt was received indicating the abortion of an RX packet
*
* "The receive abort interrupt occurs when the reception of a frame has
* been aborted. A frame being received is aborted when the Head Pointer
* attempts to overrun the Tail Pointer, or when the packet counter has
* reached FFh. In either case, the receive buffer is full and cannot
* fit the incoming frame, so the packet has been dropped. This
* interrupt does not occur when packets are dropped due to the receive
* filters rejecting a packet. The interrupt should be cleared by
* software once it has been serviced."
*
* Input Parameters:
* priv - Reference to the driver state structure
*
* Returned Value:
* None
*
* Assumptions:
* Interrupts are enabled but the caller holds the network lock.
*
****************************************************************************/
static void enc_rxabtif(FAR struct enc_driver_s *priv)
{
FAR struct enc_descr_s *descr;
#if 0
/* Free the last received packet from the RX queue */
ninfo("rx abort\n");
ninfo("ESTAT: %04x\n", enc_rdreg(priv, ENC_ESTAT));
ninfo("EIR: %04x\n", enc_rdreg(priv, ENC_EIR));
ninfo("ERXTAIL: %04x\n", enc_rdreg(priv, ENC_ERXTAIL));
ninfo("ERXHAED: %04x\n", enc_rdreg(priv, ENC_ERXHEAD));
descr = (FAR struct enc_descr_s *)sq_peek(&priv->rxqueue);
while (descr != NULL)
{
ninfo("addr: %04x len: %d\n", descr->addr, descr->len);
descr = (FAR struct enc_descr_s *)sq_next(descr);
}
DEBUGASSERT(false);
#endif
descr = (FAR struct enc_descr_s *)sq_peek(&priv->rxqueue);
if (descr != NULL)
{
enc_rxrmpkt(priv, descr);
ninfo("pending packet freed\n");
}
else
{
/* If no pending packet blocks the reception, reset all buffers */
enc_resetbuffers(priv);
}
}
/****************************************************************************
* Name: enc_irqworker
*
* Description:
* Perform interrupt handling logic outside of the interrupt handler (on
* the work queue thread).
*
* Input Parameters:
* arg - The reference to the driver structure (case to void*)
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
static void enc_irqworker(FAR void *arg)
{
FAR struct enc_driver_s *priv = (FAR struct enc_driver_s *)arg;
uint16_t eir;
DEBUGASSERT(priv);
/* Get exclusive access to both the network and the SPI bus. */
net_lock();
enc_lock(priv);
/* A good practice is for the host controller to clear the Global Interrupt
* Enable bit, INTIE (EIE<15>), immediately after an interrupt event. This
* causes the interrupt pin to return to the non-asserted (high) state. Once
* the interrupt has been serviced, the INTIE bit is set again to re-enable
* interrupts. If a new interrupt occurs while servicing another, the act of
* resetting the global enable bit will cause a new falling edge to occur on
* the interrupt pin and ensure that the host does not miss any events
*/
enc_bfc(priv, ENC_EIE, EIE_INTIE);
/* Loop until all interrupts have been processed (EIR==0). Note that
* there is no infinite loop check... if there are always pending
* interrupts, we are just broken.
*/
while ((eir = enc_rdreg(priv, ENC_EIR) & EIR_ALLINTS) != 0)
{
/* Handle interrupts according to interrupt register register bit
* settings.
*/
ninfo("EIR: %04x\n", eir);
if ((eir & EIR_DMAIF) != 0) /* DMA interrupt */
{
/* Not used by this driver. Just clear the interrupt request. */
enc_bfc(priv, ENC_EIR, EIR_DMAIF);
}
/* LINKIF: The link change interrupt occurs when the PHY link status
* changes. This flag is set by hardware when a link has either been
* established or broken between the device and a remote Ethernet
* partner. The current link status can be read from PHYLNK
* (ESTAT<8>). The interrupt should be cleared by software once it
* has been serviced.
*
* To enable the link change interrupt, set LINKIE (EIE<11>).
*/
if ((eir & EIR_LINKIF) != 0) /* PHY Link Status Change */
{
enc_linkstatus(priv); /* Get current link status */
enc_bfc(priv, ENC_EIR, EIR_LINKIF); /* Clear the LINKIF interrupt */
}
/* The transmit complete interrupt occurs when the transmission of a
* frame has ended (whether or not it was successful). This flag is set
* when TXRTS (ECON1<1>) is cleared. The interrupt should be cleared by
* software once it has been serviced.
*/
if ((eir & EIR_TXIF) != 0) /* Transmit Done */
{
enc_txif(priv);
enc_bfc(priv, ENC_EIR, EIR_TXIF);
}
/* The receive abort interrupt occurs when the reception of a frame has
* been aborted. A frame being received is aborted when the Head Pointer
* attempts to overrun the Tail Pointer, or when the packet counter has
* reached FFh. In either case, the receive buffer is full and cannot
* fit the incoming frame, so the packet has been dropped. This
* interrupt does not occur when packets are dropped due to the receive
* filters rejecting a packet. The interrupt should be cleared by
* software once it has been serviced.
*
* To enable the receive abort interrupt, set RXABTIE (EIE<1>).
* The corresponding interrupt flag is RXABTIF (EIR<1>).
*/
if ((eir & EIR_RXABTIF) != 0) /* Receive Abort */
{
NETDEV_RXERRORS(&priv->dev);
enc_rxabtif(priv);
enc_bfc(priv, ENC_EIR, EIR_RXABTIF); /* Clear the RXABTIF interrupt */
}
/* The received packet pending interrupt occurs when one or more frames
* have been received and are ready for software processing. This flag
* is set when the PKTCNT<7:0> (ESTAT<7:0>) bits are non-zero. This
* interrupt flag is read-only and will automatically clear when the
* PKTCNT bits are decremented to zero. For more details about
* receiving and processing incoming frames, refer to Section 9.0
* "Transmitting and Receiving Packets".
*
* To enable the received packet pending interrupt, set PKTIE (EIE<6>).
* The corresponding interrupt flag is PKTIF (EIR<6>).
*/
if ((eir & EIR_PKTIF) != 0 /* RX Packet Pending */
&& (enc_rdreg(priv, ENC_ESTAT) & ESTAT_PKTCNT_MASK) != 0)
{
enc_pktif(priv); /* Handle packet receipt */
/* No clearing necessary, after PKTCNT == 0 the bit is automatically
* cleared. This means we will loop until all packets are processed.
*/
}
#ifdef CONFIG_NETDEV_STATISTICS
/* The transmit abort interrupt occurs when the transmission of a frame
* has been aborted. An abort can occur for any of the following
* reasons:
*
* * Excessive collisions occurred as defined by the Retransmission
* Maximum, MAXRET<3:0> bits (MACLCON<3:0>), setting. If this occurs,
* the COLCNT bits (ETXSTAT<3:0>) will indicate the number of
* collisions that occurred.
*
* * A late collision occurred after 63 bytes were transmitted. If this
* occurs, LATECOL (ETXSTAT<10>) will be set.
*
* * The medium was busy and the packet was deferred. If this occurs,
* EXDEFER (ETXSTAT<8>) will be set.
*
* * The application aborted the transmission by clearing TXRTS
* (ECON1<1>).
*
* The interrupt should be cleared by software once it has been
* serviced. To enable the transmit abort interrupt, set TXABTIE
* (EIE<2>).
*/
if ((eir & EIR_TXABTIF) != 0) /* Transmit Abort */
{
NETDEV_TXERRORS(&priv->dev);
enc_bfc(priv, ENC_EIR, EIR_TXABTIF); /* Clear the TXABTIF interrupt */
}
#endif
}
/* Enable GPIO interrupts */
priv->lower->enable(priv->lower);
/* Enable Ethernet interrupts */
enc_bfs(priv, ENC_EIE, EIE_INTIE);
/* Release lock on the SPI bus and the network */
enc_unlock(priv);
net_unlock();
}
/****************************************************************************
* Name: enc_interrupt
*
* Description:
* Hardware interrupt handler
*
* Input Parameters:
* irq - Number of the IRQ that generated the interrupt
* context - Interrupt register state save info (architecture-specific)
*
* Returned Value:
* OK on success
*
* Assumptions:
*
****************************************************************************/
static int enc_interrupt(int irq, FAR void *context, FAR void *arg)
{
FAR struct enc_driver_s *priv;
DEBUGASSERT(arg != NULL);
priv = (FAR struct enc_driver_s *)arg;
/* In complex environments, we cannot do SPI transfers from the interrupt
* handler because semaphores are probably used to lock the SPI bus. In
* this case, we will defer processing to the worker thread. This is also
* much kinder in the use of system resources and is, therefore, probably
* a good thing to do in any event.
*/
DEBUGASSERT(work_available(&priv->irqwork));
/* Notice that further GPIO interrupts are disabled until the work is
* actually performed. This is to prevent overrun of the worker thread.
* Interrupts are re-enabled in enc_irqworker() when the work is completed.
*/
priv->lower->disable(priv->lower);
return work_queue(ENCWORK, &priv->irqwork, enc_irqworker,
(FAR void *)priv, 0);
}
/****************************************************************************
* Name: enc_toworker
*
* Description:
* Our TX watchdog timed out. This is the worker thread continuation of
* the watchdog timer interrupt. Reset the hardware and start again.
*
* Input Parameters:
* arg - The reference to the driver structure (case to void*)
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
static void enc_toworker(FAR void *arg)
{
FAR struct enc_driver_s *priv = (FAR struct enc_driver_s *)arg;
int ret;
nerr("ERROR: Tx timeout\n");
DEBUGASSERT(priv);
/* Get exclusive access to the network. */
net_lock();
/* Increment statistics and dump debug info */
NETDEV_TXTIMEOUTS(&priv->dev);
/* Then reset the hardware: Take the interface down, then bring it
* back up
*/
ret = enc_ifdown(&priv->dev);
DEBUGASSERT(ret == OK);
ret = enc_ifup(&priv->dev);
DEBUGASSERT(ret == OK);
UNUSED(ret);
/* Then poll the network for new XMIT data */
devif_poll(&priv->dev, enc_txpoll);
/* Release the network */
net_unlock();
}
/****************************************************************************
* Name: enc_txtimeout
*
* Description:
* Our TX watchdog timed out. Called from the timer interrupt handler.
* The last TX never completed. Perform work on the worker thread.
*
* Input Parameters:
* argc - The number of available arguments
* arg - The first argument
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
static void enc_txtimeout(int argc, uint32_t arg, ...)
{
FAR struct enc_driver_s *priv = (FAR struct enc_driver_s *)arg;
int ret;
/* In complex environments, we cannot do SPI transfers from the timeout
* handler because semaphores are probably used to lock the SPI bus. In
* this case, we will defer processing to the worker thread. This is also
* much kinder in the use of system resources and is, therefore, probably
* a good thing to do in any event.
*/
DEBUGASSERT(priv && work_available(&priv->towork));
/* Notice that Tx timeout watchdog is not active so further Tx timeouts
* can occur until we restart the Tx timeout watchdog.
*/
ret = work_queue(ENCWORK, &priv->towork, enc_toworker, (FAR void *)priv, 0);
UNUSED(ret);
DEBUGASSERT(ret == OK);
}
/****************************************************************************
* Name: enc_pollworker
*
* Description:
* Periodic timer handler continuation.
*
* Input Parameters:
* argc - The number of available arguments
* arg - The first argument
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
static void enc_pollworker(FAR void *arg)
{
FAR struct enc_driver_s *priv = (FAR struct enc_driver_s *)arg;
DEBUGASSERT(priv);
/* Get exclusive access to both the network and the SPI bus. */
net_lock();
enc_lock(priv);
/* Verify that the hardware is ready to send another packet. The driver
* start a transmission process by setting ECON1.TXRTS. When the packet is
* finished transmitting or is aborted due to an error/cancellation, the
* ECON1.TXRTS bit will be cleared.
*/
if ((enc_rdreg(priv, ENC_ECON1) & ECON1_TXRTS) == 0)
{
/* Yes.. update TCP timing states and poll the network for new XMIT
* data. Hmmm.. looks like a bug here to me. Does this mean if there
* is a transmit in progress, we will missing TCP time state updates?
*/
devif_timer(&priv->dev, ENC_WDDELAY, enc_txpoll);
}
/* Release lock on the SPI bus and the network */
enc_unlock(priv);
net_unlock();
/* Setup the watchdog poll timer again */
wd_start(priv->txpoll, ENC_WDDELAY, enc_polltimer, 1, (wdparm_t)arg);
}
/****************************************************************************
* Name: enc_polltimer
*
* Description:
* Periodic timer handler. Called from the timer interrupt handler.
*
* Input Parameters:
* argc - The number of available arguments
* arg - The first argument
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
static void enc_polltimer(int argc, uint32_t arg, ...)
{
FAR struct enc_driver_s *priv = (FAR struct enc_driver_s *)arg;
int ret;
/* In complex environments, we cannot do SPI transfers from the timeout
* handler because semaphores are probably used to lock the SPI bus. In
* this case, we will defer processing to the worker thread. This is also
* much kinder in the use of system resources and is, therefore, probably
* a good thing to do in any event.
*/
DEBUGASSERT(priv && work_available(&priv->pollwork));
/* Notice that poll watchdog is not active so further poll timeouts can
* occur until we restart the poll timeout watchdog.
*/
ret = work_queue(ENCWORK, &priv->pollwork, enc_pollworker,
(FAR void *)priv, 0);
DEBUGASSERT(ret == OK);
UNUSED(ret);
}
/****************************************************************************
* Name: enc_ifup
*
* Description:
* NuttX Callback: Bring up the Ethernet interface when an IP address is
* provided
*
* Input Parameters:
* dev - Reference to the NuttX driver state structure
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
2014-06-28 00:48:12 +02:00
static int enc_ifup(struct net_driver_s *dev)
{
FAR struct enc_driver_s *priv = (FAR struct enc_driver_s *)dev->d_private;
int ret;
ninfo("Bringing up: %d.%d.%d.%d\n",
dev->d_ipaddr & 0xff, (dev->d_ipaddr >> 8) & 0xff,
(dev->d_ipaddr >> 16) & 0xff, dev->d_ipaddr >> 24);
/* Lock the SPI bus so that we have exclusive access */
enc_lock(priv);
/* Initialize Ethernet interface, set the MAC address, and make sure that
* the ENC28J80 is not in power save mode.
*/
ret = enc_reset(priv);
if (ret == OK)
{
enc_setmacaddr(priv);
/* Enable interrupts at the ENCX24J600. Interrupts are still disabled
* at the interrupt controller.
*/
enc_bfc(priv, ENC_EIR, EIR_ALLINTS);
enc_bfs(priv, ENC_EIE, EIE_INTIE | EIE_LINKIE |
EIE_PKTIE | EIE_RXABTIE |
EIE_TXIE);
#ifdef CONFIG_NETDEV_STATISTICS
enc_bfs(priv, ENC_EIE, EIE_TXABTIE);
#endif
/* Enable the receiver */
enc_bfs(priv, ENC_ECON1, ECON1_RXEN);
/* Set and activate a timer process */
wd_start(priv->txpoll, ENC_WDDELAY, enc_polltimer, 1,
(wdparm_t)priv);
/* Mark the interface up and enable the Ethernet interrupt at the
* controller
*/
priv->ifstate = ENCSTATE_UP;
priv->lower->enable(priv->lower);
}
/* Un-lock the SPI bus */
enc_unlock(priv);
return ret;
}
/****************************************************************************
* Name: enc_ifdown
*
* Description:
* NuttX Callback: Stop the interface.
*
* Input Parameters:
* dev - Reference to the NuttX driver state structure
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
2014-06-28 00:48:12 +02:00
static int enc_ifdown(struct net_driver_s *dev)
{
FAR struct enc_driver_s *priv = (FAR struct enc_driver_s *)dev->d_private;
irqstate_t flags;
int ret;
ninfo("Taking down: %d.%d.%d.%d\n",
dev->d_ipaddr & 0xff, (dev->d_ipaddr >> 8) & 0xff,
(dev->d_ipaddr >> 16) & 0xff, dev->d_ipaddr >> 24);
/* Lock the SPI bus so that we have exclusive access */
enc_lock(priv);
/* Disable the Ethernet interrupt */
flags = enter_critical_section();
priv->lower->disable(priv->lower);
/* Cancel the TX poll timer and TX timeout timers */
wd_cancel(priv->txpoll);
wd_cancel(priv->txtimeout);
/* Reset the device and leave in the power save state */
ret = enc_reset(priv);
enc_pwrsave(priv);
priv->ifstate = ENCSTATE_DOWN;
leave_critical_section(flags);
/* Un-lock the SPI bus */
enc_unlock(priv);
return ret;
}
/****************************************************************************
* Name: enc_txavail
*
* Description:
* Driver callback invoked when new TX data is available. This is a
* stimulus perform an out-of-cycle poll and, thereby, reduce the TX
* latency.
*
* Input Parameters:
* dev - Reference to the NuttX driver state structure
*
* Returned Value:
* None
*
* Assumptions:
* Called in normal user mode
*
****************************************************************************/
2014-06-28 00:48:12 +02:00
static int enc_txavail(struct net_driver_s *dev)
{
FAR struct enc_driver_s *priv = (FAR struct enc_driver_s *)dev->d_private;
irqstate_t flags;
/* Lock the SPI bus so that we have exclusive access */
enc_lock(priv);
/* Ignore the notification if the interface is not yet up */
flags = enter_critical_section();
if (priv->ifstate == ENCSTATE_RUNNING)
{
/* Check if the hardware is ready to send another packet. The driver
* starts a transmission process by setting ECON1.TXRTS. When the
* packet is finished transmitting or is aborted due to an error/
* cancellation, the ECON1.TXRTS bit will be cleared.
*/
if ((enc_rdreg(priv, ENC_ECON1) & ECON1_TXRTS) == 0)
{
/* The interface is up and TX is idle; poll the network for new XMIT data */
devif_poll(&priv->dev, enc_txpoll);
}
}
/* Un-lock the SPI bus */
leave_critical_section(flags);
enc_unlock(priv);
return OK;
}
/****************************************************************************
* Name: enc_addmac
*
* Description:
* NuttX Callback: Add the specified MAC address to the hardware multicast
* address filtering
*
* Input Parameters:
* dev - Reference to the NuttX driver state structure
* mac - The MAC address to be added
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
#ifdef CONFIG_NET_MCASTGROUP
2014-06-28 00:48:12 +02:00
static int enc_addmac(struct net_driver_s *dev, FAR const uint8_t *mac)
{
FAR struct enc_driver_s *priv = (FAR struct enc_driver_s *)dev->d_private;
/* Lock the SPI bus so that we have exclusive access */
enc_lock(priv);
/* Add the MAC address to the hardware multicast routing table */
#warning "Multicast MAC support not implemented"
/* Un-lock the SPI bus */
enc_unlock(priv);
return OK;
}
#endif
/****************************************************************************
* Name: enc_rmmac
*
* Description:
* NuttX Callback: Remove the specified MAC address from the hardware
* multicast address filtering
*
* Input Parameters:
* dev - Reference to the NuttX driver state structure
* mac - The MAC address to be removed
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
#ifdef CONFIG_NET_MCASTGROUP
2014-06-28 00:48:12 +02:00
static int enc_rmmac(struct net_driver_s *dev, FAR const uint8_t *mac)
{
FAR struct enc_driver_s *priv = (FAR struct enc_driver_s *)dev->d_private;
/* Lock the SPI bus so that we have exclusive access */
enc_lock(priv);
/* Add the MAC address to the hardware multicast routing table */
#warning "Multicast MAC support not implemented"
/* Un-lock the SPI bus */
enc_unlock(priv);
return OK;
}
#endif
/****************************************************************************
* Name: enc_pwrsave
*
* Description:
* The ENCX24J600 may be placed in Power-Down mode through the command
* interface. In this mode, the device will no longer be able to transmit
* or receive any packets or perform DMA operations. However, most
* registers, and all buffer memories, retain their states and remain
* accessible by the host controller. The clock driver also remains
* operational, leaving the CLKOUT function unaffected. However, the
* MAC/MII and PHY registers all become inaccessible, and the PHY
* registers lose their current states.
*
* 1. Turn off the Modular Exponentiation and AES engines by clearing
* CRYPTEN (EIR<15>).
* 2. Turn off packet reception by clearing RXEN (ECON1<0>).
* 3. Wait for any in-progress receptions to complete by polling
* RXBUSY (ESTAT<13>) until it is clear.
* 4. Wait for any current transmission operation to complete by verifying
* that TXRTS (ECON1<1>) is clear.
* 5. Power-down the PHY by setting the PSLEEP bit (PHCON1<11>).
* 6. Power-down the Ethernet interface by clearing
* ETHEN and STRCH (ECON2<15,14>). Disabling the LED stretching behavior
* is necessary to ensure no LEDs get trapped in a perpetually
* illuminated state in the event they are being stretched on when ETHEN
* is cleared.
*
* Note:
* Instead of providing a powerup function, the job is done by enc_reset.
* enc_ifup calls it anyway.
*
* Input Parameters:
* priv - Reference to the driver state structure
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
static void enc_pwrsave(FAR struct enc_driver_s *priv)
{
uint16_t regval;
ninfo("Set PWRSV\n");
/* 1. Turn off AES */
enc_bfc(priv, ENC_EIR, EIR_CRYPTEN);
/* 2. Turn off packet reception */
enc_bfc(priv, ENC_ECON1, ECON1_RXEN);
/* 3. Wait for pending reception to complete */
enc_waitreg(priv, ENC_ESTAT, ESTAT_RXBUSY, 0);
/* 4. Wait for any current transmissions to complete */
enc_waitreg(priv, ENC_ECON1, ECON1_TXRTS, 0);
/* 5. Power down the PHY */
regval = enc_rdphy(priv, ENC_PHCON1);
regval |= PHCON1_PSLEEP;
enc_wrphy(priv, ENC_PHCON1, regval);
/* 6. Power down the Ethernet interface */
enc_bfc(priv, ENC_ECON2, ECON2_ETHEN | ECON2_STRCH);
}
2013-09-25 16:26:56 +02:00
/****************************************************************************
* Name: enc_ldmacaddr
2013-09-25 16:26:56 +02:00
*
* Description:
* Load the MAC address from the ENCX24j600 and write it to the device
* structure.
*
* Input Parameters:
2013-09-25 16:26:56 +02:00
* priv - Reference to the driver state structure
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
static void enc_ldmacaddr(FAR struct enc_driver_s *priv)
{
uint16_t regval;
uint8_t *mac = priv->dev.d_mac.ether.ether_addr_octet;
2013-09-25 16:26:56 +02:00
ninfo("Using ENCX24J600's built in MAC address\n");
2013-09-25 16:26:56 +02:00
regval = enc_rdreg(priv, ENC_MAADR1);
mac[0] = regval & 0xff;
mac[1] = regval >> 8;
regval = enc_rdreg(priv, ENC_MAADR2);
mac[2] = regval & 0xff;
mac[3] = regval >> 8;
regval = enc_rdreg(priv, ENC_MAADR3);
mac[4] = regval & 0xff;
mac[5] = regval >> 8;
}
/****************************************************************************
* Name: enc_setmacaddr
*
* Description:
* Set the MAC address to the configured value. This is done after ifup
* or after a TX timeout. Note that this means that the interface must
* be down before configuring the MAC addr.
* If the MAC address is 0 in all digits, the ENCX24J600's MAC is read out.
*
* Input Parameters:
* priv - Reference to the driver state structure
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
static void enc_setmacaddr(FAR struct enc_driver_s *priv)
{
uint8_t *mac = priv->dev.d_mac.ether.ether_addr_octet;
struct ether_addr zmac;
memset(&zmac, 0, sizeof(zmac));
if (memcmp(&priv->dev.d_mac.ether, &zmac, sizeof(zmac)) == 0)
{
/* No user defined MAC address. Read it from the device. */
2013-09-25 16:26:56 +02:00
enc_ldmacaddr(priv);
}
else
{
/* There is a user defined mac address. Write it to the ENCXJ600 */
ninfo("Using an user defined MAC address\n");
enc_wrreg(priv, ENC_MAADR1, (uint16_t)mac[1] << 8 | (uint16_t)mac[0]);
enc_wrreg(priv, ENC_MAADR2, (uint16_t)mac[3] << 8 | (uint16_t)mac[2]);
enc_wrreg(priv, ENC_MAADR3, (uint16_t)mac[5] << 8 | (uint16_t)mac[4]);
}
}
/****************************************************************************
* Name: enc_resetbuffers
*
* Description:
* Initializes the RX/TX queues and configures the enc's RX/TX buffers.
* Called on general reset and on rxabt interrupt.
*
* Input Parameters:
* priv - Reference to the driver state structure
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
static void enc_resetbuffers(FAR struct enc_driver_s *priv)
{
int i;
/* Initialize receive and transmit buffers */
priv->nextpkt = PKTMEM_RX_START;
enc_wrreg(priv, ENC_ERXST, PKTMEM_RX_START);
/* Program the Tail Pointer, ERXTAIL, to the last even address of the buffer */
enc_wrreg(priv, ENC_ERXTAIL, PKTMEM_RX_END - 2);
sq_init(&priv->txfreedescr);
sq_init(&priv->rxfreedescr);
sq_init(&priv->txqueue);
sq_init(&priv->rxqueue);
/* For transmission we preinitialize the descriptors to aligned NET_BUFFSIZE */
for (i = 0; i < ENC_NTXDESCR; i++)
{
priv->txdescralloc[i].addr = PKTMEM_START +
PKTMEM_ALIGNED_BUFSIZE * i;
sq_addlast((FAR sq_entry_t *)&priv->txdescralloc[i],
&priv->txfreedescr);
}
/* Receive descriptors addresses are set on reception */
for (i = 0; i < CONFIG_ENCX24J600_NRXDESCR; i++)
{
sq_addlast((FAR sq_entry_t *)&priv->rxdescralloc[i],
&priv->rxfreedescr);
}
}
/****************************************************************************
* Name: enc_reset
*
* Description:
* Stop, reset, re-initialize, and restart the ENCX24J600. This is done
* initially, on ifup, and after a TX timeout.
*
* Input Parameters:
* priv - Reference to the driver state structure
*
* Returned Value:
* None
*
* Assumptions:
*
****************************************************************************/
static int enc_reset(FAR struct enc_driver_s *priv)
{
int ret;
uint16_t regval;
ninfo("Reset\n");
do
{
enc_wrreg(priv, ENC_EUDAST, 0x1234);
}
while (enc_rdreg(priv, ENC_EUDAST) != 0x1234);
2013-09-25 16:26:56 +02:00
/* Wait for clock to become ready */
ret = enc_waitreg(priv, ENC_ESTAT, ESTAT_CLKRDY, ESTAT_CLKRDY);
if (ret != OK)
{
nerr("ERROR: encx24j600 clock failed to become ready\n");
return -ENODEV;
}
2013-09-25 16:26:56 +02:00
/* Reset the ENCX24J600 */
enc_setethrst(priv);
2013-09-25 16:26:56 +02:00
/* Check if EUDAST has been reset to 0 */
regval = enc_rdreg(priv, ENC_EUDAST);
if (regval != 0x0000)
{
nerr("ERROR: encx24j600 seems not to be reset properly\n");
return -ENODEV;
}
/* Wait at least 256 μs for the PHY registers and PHY status bits to become
* available.
*/
up_udelay(256);
/* Initialize RX/TX buffers */
enc_resetbuffers(priv);
2013-09-25 16:26:56 +02:00
#if 0
/* When restarting auto-negotiation, the ESTAT_PHYLINK gets set but the link
* seems not to be ready. Because auto-negotiation is enabled by default
* (but with different PHANA_* settings) I did not investigate that further.
*/
/* "Typically, when using auto-negotiation, users should write 0x05e1 to
* PHANA to advertise flow control capability."
*/
enc_wrphy(priv, ENC_PHANA, PHANA_ADPAUS0 | PHANA_AD10FD | PHANA_AD10 |
PHANA_AD100FD | PHANA_AD100 | PHANA_ADIEEE0);
2013-09-25 16:26:56 +02:00
/* Restart auto-negotiation */
enc_wrphy(priv, ENC_PHCON1, PHCON1_RENEG);
do
{
regval = enc_rdphy(priv, ENC_PHSTAT1);
}
while ((regval & PHSTAT1_ANDONE) != 0);
ninfo("Auto-negotiation completed\n");
2013-09-25 16:26:56 +02:00
#endif
enc_linkstatus(priv);
/* Set the maximum packet size which the controller will accept */
This commit attempts remove some long standard confusion in naming and some actual problems that result from the naming confusion. The basic problem is the standard MTU does not include the size of the Ethernet header. For clarity, I changed the naming of most things called MTU to PKTSIZE. For example, CONFIG_NET_ETH_MTU is now CONFIG_NET_ETH_PKTSIZE. This makes the user interface a little hostile. People thing of an MTU of 1500 bytes, but the corresponding packet is really 1514 bytes (including the 14 byte Ethernet header). A more friendly solution would configure the MTU (as before), but then derive the packet buffer size by adding the MAC header length. Instead, we define the packet buffer size then derive the MTU. The MTU is not common currency in networking. On the wire, the only real issue is the MSS which is derived from MTU by subtracting the IP header and TCP header sizes (for the case of TCP). Now it is derived for the PKTSIZE by subtracting the IP header, the TCP header, and the MAC header sizes. So we should be all good and without the recurring 14 byte error in MTU's and MSS's. Squashed commit of the following: Trivial update to fix some spacing issues. net/: Rename several macros containing _MTU to _PKTSIZE. net/: Rename CONFIG_NET_SLIP_MTU to CONFIG_NET_SLIP_PKTSIZE and similarly for CONFIG_NET_TUN_MTU. These are not the MTU which does not include the size of the link layer header. These are the full size of the packet buffer memory (minus any GUARD bytes). net/: Rename CONFIG_NET_6LOWPAN_MTU to CONFIG_NET_6LOWPAN_PKTSIZE and similarly for CONFIG_NET_TUN_MTU. These are not the MTU which does not include the size of the link layer header. These are the full size of the packet buffer memory (minus any GUARD bytes). net/: Rename CONFIG_NET_ETH_MTU to CONFIG_NET_ETH_PKTSIZE. This is not the MTU which does not include the size of the link layer header. This is the full size of the packet buffer memory (minus any GUARD bytes). net/: Rename the file d_mtu in the network driver structure to d_pktsize. That value saved there is not the MTU. The packetsize is the memory large enough to hold the maximum packet PLUS the size of the link layer header. The MTU does not include the link layer header.
2018-07-04 22:10:40 +02:00
enc_wrreg(priv, ENC_MAMXFL, CONFIG_NET_ETH_PKTSIZE + 4);
2013-09-25 16:26:56 +02:00
ret = enc_waitreg(priv, ENC_ESTAT, ESTAT_PHYLNK, ESTAT_PHYLNK);
if (ret == OK)
{
enc_linkstatus(priv);
}
#if 0
2013-09-25 16:26:56 +02:00
if (ret != OK)
{
nerr("ERROR: encx24j600 failed to establish link\n");
2013-09-25 16:26:56 +02:00
return -ENODEV;
}
#endif
2013-09-25 16:26:56 +02:00
return OK;
}
/****************************************************************************
* Public Functions
****************************************************************************/
/****************************************************************************
* Name: enc_initialize
*
* Description:
* Initialize the Ethernet driver. The ENCX24J600 device is assumed to be
* in the post-reset state upon entry to this function.
*
* Input Parameters:
* spi - A reference to the platform's SPI driver for the ENCX24J600
* lower - The MCU-specific interrupt used to control low-level MCU
* functions (i.e., ENCX24J600 GPIO interrupts).
* devno - If more than one ENCX24J600 is supported, then this is the
* zero based number that identifies the ENCX24J600;
*
* Returned Value:
* OK on success; Negated errno on failure.
*
* Assumptions:
*
****************************************************************************/
int enc_initialize(FAR struct spi_dev_s *spi,
FAR const struct enc_lower_s *lower,
unsigned int devno)
{
FAR struct enc_driver_s *priv;
DEBUGASSERT(devno < CONFIG_ENCX24J600_NINTERFACES);
priv = &g_encx24j600[devno];
/* Initialize the driver structure */
memset(g_encx24j600, 0,
CONFIG_ENCX24J600_NINTERFACES * sizeof(struct enc_driver_s));
priv->dev.d_buf = g_pktbuf; /* Single packet buffer */
priv->dev.d_ifup = enc_ifup; /* I/F up (new IP address) callback */
priv->dev.d_ifdown = enc_ifdown; /* I/F down callback */
priv->dev.d_txavail = enc_txavail; /* New TX data callback */
#ifdef CONFIG_NET_MCASTGROUP
priv->dev.d_addmac = enc_addmac; /* Add multicast MAC address */
priv->dev.d_rmmac = enc_rmmac; /* Remove multicast MAC address */
#endif
priv->dev.d_private = priv; /* Used to recover private state from dev */
/* Create a watchdog for timing polling for and timing of transmissions */
priv->txpoll = wd_create(); /* Create periodic poll timer */
priv->txtimeout = wd_create(); /* Create TX timeout timer */
priv->spi = spi; /* Save the SPI instance */
priv->lower = lower; /* Save the low-level MCU interface */
/* The interface should be in the down state. However, this function is
2019-08-21 17:32:59 +02:00
* called too early in initialization to perform the ENCX24J600 reset in
* enc_ifdown. We are depending upon the fact that the application level
* logic will call enc_ifdown later to reset the ENCX24J600.
*/
priv->ifstate = ENCSTATE_UNINIT;
/* Attach the interrupt to the driver (but don't enable it yet) */
if (lower->attach(lower, enc_interrupt, priv) < 0)
{
/* We could not attach the ISR to the interrupt */
return -EAGAIN;
}
2013-09-25 16:26:56 +02:00
/* Lock the SPI bus so that we have exclusive access */
enc_lock(priv);
/* Load the MAC address */
enc_ldmacaddr(priv);
/* Power down the device */
enc_pwrsave(priv);
2013-09-25 16:26:56 +02:00
/* Unlock the SPI bus */
enc_unlock(priv);
/* Register the device with the OS so that socket IOCTLs can be performed */
return netdev_register(&priv->dev, NET_LL_ETHERNET);
}
#endif /* CONFIG_NET && CONFIG_ENCX24J600_NET */