linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* New FarSync T-Series driver
@ 2001-07-03 17:28 Robert J.Dunlop
  2001-07-04 14:18 ` François romieu
  0 siblings, 1 reply; 5+ messages in thread
From: Robert J.Dunlop @ 2001-07-03 17:28 UTC (permalink / raw)
  To: Krzysztof Halasa, Alan Cox; +Cc: linux-kernel

Hi guys,

Sorry for the big post, but I posted URLs for an earlier version of this
a couple of months back and got very little feedback.  I know sync card
drivers ain't sexy.

The driver has now completed in house testing at FarSite along with
positive feedback from several field testers, for which I'm very greatful.
Details of cards and the official company download at www.farsite.co.uk

Could someone please incorporate this into the official tree and/or tell
me what changes I need to make before this can happen.


This patch is against kernel 2.4.5-ac23
The full support package and alternate patches against 2.4.6-pre9 and a
patch that uses Krzysztof Halasa's Generic HDLC Layer are available at
http://www.xyzzy.clara.co.uk/farsync/index.html


--- linux/CREDITS.orig	Tue Jul  3 15:19:43 2001
+++ linux/CREDITS	Tue Jul  3 15:22:13 2001
@@ -740,6 +740,17 @@
 S: Hillsboro, Oregon 97124
 S: USA
 
+N: Bob Dunlop
+E: rjd@xyzzy.clara.co.uk
+E: bob.dunlop@farsite.co.uk
+W: www.farsite.co.uk
+D: FarSync card device driver
+S: FarSite Communications Ltd
+S: Tempus Business Centre
+S: 60 Kingsclere Road
+S: Basingstoke       RG21 6XG
+S: UK
+
 N: Cyrus Durgin
 E: cider@speakeasy.org
 W: http://www.speakeasy.org/~cider/
--- linux/Documentation/Configure.help.orig	Tue Jul  3 15:19:43 2001
+++ linux/Documentation/Configure.help	Tue Jul  3 15:22:13 2001
@@ -8595,6 +8595,21 @@
   of the Cisco HDLC/PPP driver (syncppp.c).
   The SyncLink WAN driver (in character devices) must also be enabled.
 
+FarSync T-Series support
+CONFIG_FARSYNC
+  This driver supports the FarSync T-Series X.21 (and V.35/V.24) cards
+  from FarSite Communications Ltd.
+  Synchronous communication is supported on all ports at speeds up to
+  8Mb/s (128K on V.24) using synchronous PPP or Cisco HDLC.
+
+  If you want to compile this driver as a module ( = code which can be
+  inserted in and removed from the running kernel whenever you want)
+  say M here and read Documentation/modules.txt.
+  The module will be called farsync.o and if you want the module to be
+  automatically loaded when the interface is referenced then you should
+  add "alias syncX farsync" to /etc/modules.conf for each interface,
+  where X is 0, 1, 2, ...
+
 Frame Relay (DLCI) support
 CONFIG_DLCI
   This is support for the frame relay protocol; frame relay is a fast
--- linux/drivers/net/wan/Config.in.orig	Tue Jul  3 15:19:39 2001
+++ linux/drivers/net/wan/Config.in	Tue Jul  3 15:22:13 2001
@@ -39,6 +39,10 @@
 
    dep_tristate '  Etinc PCISYNC serial board support (EXPERIMENTAL)' CONFIG_DSCC4 m
 
+# FarSite Communications' cards
+
+   tristate '  FarSync T-Series X.21 (and V.35/V.24) cards' CONFIG_FARSYNC
+
 #
 # Lan Media's board. Currently 1000, 1200, 5200, 5245
 #
--- linux/drivers/net/wan/Makefile.orig	Thu Apr 12 20:11:39 2001
+++ linux/drivers/net/wan/Makefile	Tue Jul  3 15:22:13 2001
@@ -33,6 +33,7 @@
 obj-$(CONFIG_COMX_PROTO_LAPB)	+=				comx-proto-lapb.o
 obj-$(CONFIG_COMX_PROTO_FR)	+=				comx-proto-fr.o
 obj-$(CONFIG_COSA)		+=		syncppp.o	cosa.o
+obj-$(CONFIG_FARSYNC)		+=		syncppp.o	farsync.o
 obj-$(CONFIG_DSCC4)             +=				dscc4.o
 obj-$(CONFIG_LANMEDIA)		+=		syncppp.o
 obj-$(CONFIG_SYNCLINK_SYNCPPP)	+=		syncppp.o
--- linux/drivers/pci/pci.ids.orig	Tue Jul  3 15:19:43 2001
+++ linux/drivers/pci/pci.ids	Tue Jul  3 15:22:13 2001
@@ -4266,6 +4266,9 @@
 1607  Lava Semiconductor Manufacturing Inc
 1608  Automated Wagering International
 1609  Scimetric Instruments Inc
+1619  FarSite Communications Ltd
+	0400  FarSync T2P (2 port X.21/V.35/V.24)
+	0440  FarSync T4P (4 port X.21/V.35/V.24)
 1668  Action Tec Electronics Inc
 1813  Ambient Technologies Inc
 1a08  Sierra semiconductor
--- linux/MAINTAINERS.orig	Tue Jul  3 15:19:34 2001
+++ linux/MAINTAINERS	Tue Jul  3 15:22:13 2001
@@ -500,6 +500,13 @@
 L:	linux-kernel@vger.kernel.org
 S:	Maintained
 
+FARSYNC SYNCHRONOUS DRIVER
+P:	Bob Dunlop
+M:	rjd@xyzzy.clara.co.uk
+M:	bob.dunlop@farsite.co.uk
+W:	http://www.farsite.co.uk/
+S:	Supported
+
 FILE LOCKING (flock() and fcntl()/lockf())
 P:	Matthew Wilcox
 M:	matthew@wil.cx
--- /dev/null	Sat Mar 24 04:37:44 2001
+++ linux/drivers/net/wan/farsync.c	Tue Jul  3 14:28:51 2001
@@ -0,0 +1,1834 @@
+/*
+ *      FarSync X21 driver for Linux (2.4.x kernel version)
+ *
+ *      Actually sync driver for X.21, V.35 and V.24 on FarSync T-series cards
+ *
+ *      Copyright (C) 2001 FarSite Communications Ltd.
+ *      www.farsite.co.uk
+ *
+ *      This program is free software; you can redistribute it and/or
+ *      modify it under the terms of the GNU General Public License
+ *      as published by the Free Software Foundation; either version
+ *      2 of the License, or (at your option) any later version.
+ *
+ *      Author: R.J.Dunlop      <bob.dunlop@farsite.co.uk>
+ */
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/pci.h>
+#include <linux/ioport.h>
+#include <linux/netdevice.h>
+#include <linux/init.h>
+#include <linux/if_arp.h>
+#include <asm/uaccess.h>
+#include <net/syncppp.h>
+
+#include "farsync.h"
+
+
+/*
+ *      Module info
+ */
+MODULE_AUTHOR("R.J.Dunlop <bob.dunlop@farsite.co.uk>");
+MODULE_DESCRIPTION("FarSync T-Series X21 driver. FarSite Communications Ltd.");
+
+EXPORT_NO_SYMBOLS;
+
+
+/*      Driver configuration and global parameters
+ *      ==========================================
+ */
+
+/*      Number of ports (per card) supported
+ */
+#define FST_MAX_PORTS           4
+
+
+/*      PCI vendor and device IDs
+ */
+#define FSC_PCI_VENDOR_ID       0x1619  /* FarSite Communications Ltd */
+#define T2P_PCI_DEVICE_ID       0x0400  /* T2P X21 2 port card */
+#define T4P_PCI_DEVICE_ID       0x0440  /* T4P X21 4 port card */
+
+
+/*      Default parameters for the link
+ */
+#define FST_TX_QUEUE_LEN        100     /* At 8Mbps a longer queue length is
+                                         * useful, the syncppp module forces
+                                         * this down assuming a slower line I
+                                         * guess.
+                                         */
+#define FST_MAX_MTU             8000    /* Huge but possible */
+#define FST_DEF_MTU             1500    /* Common sane value */
+
+#define FST_TX_TIMEOUT          (2*HZ)
+
+
+#ifdef ARPHRD_RAWHDLC
+#define ARPHRD_MYTYPE   ARPHRD_RAWHDLC  /* Raw frames */
+#else
+#define ARPHRD_MYTYPE   ARPHRD_HDLC     /* Cisco-HDLC (keepalives etc) */
+#endif
+
+
+/*      Card shared memory layout
+ *      =========================
+ */
+#pragma pack(1)
+
+/*      This information is derived in part from the FarSite FarSync Smc.h
+ *      file. Unfortunately various name clashes and the non-portability of the
+ *      bit field declarations in that file have meant that I have chosen to
+ *      recreate the information here.
+ *      Note for FarSite developers: I have switched "unsigned long" to
+ *      "unsigned int" because the size of an int is stable (32bit) under
+ *      Linux but long varies between 32 and 64 bit architectures.
+ *
+ *      The SMC (Shared Memory Configuration) has a version number that is
+ *      incremented every time there is a significant change. This number can
+ *      be used to check that we have not got out of step with the firmware
+ *      contained in the .CDE files.
+ */
+#define SMC_VERSION 11
+
+#define FST_MEMSIZE 0x100000    /* Size of card memory (1Mb) */
+
+#define SMC_BASE 0x00002000L    /* Base offset of the shared memory window main
+                                 * configuration structure */
+#define BFM_BASE 0x00010000L    /* Base offset of the shared memory window DMA
+                                 * buffers */
+
+#define LEN_TX_BUFFER 8192      /* Size of packet buffers */
+#define LEN_RX_BUFFER 8192
+
+#define LEN_SMALL_TX_BUFFER 256 /* Size of obsolete buffs used for DOS diags */
+#define LEN_SMALL_RX_BUFFER 256
+
+#define NUM_TX_BUFFER 2         /* Must be power of 2. Fixed by firmware */
+#define NUM_RX_BUFFER 8
+
+/* Interrupt retry time in milliseconds */
+#define INT_RETRY_TIME 2
+
+
+/*      The Am186CH/CC processors support a SmartDMA mode using circular pools
+ *      of buffer descriptors. The structure is almost identical to that used
+ *      in the LANCE Ethernet controllers. Details available as PDF from the
+ *      AMD web site: http://www.amd.com/products/epd/processors/\
+ *                    2.16bitcont/3.am186cxfa/a21914/21914.pdf
+ */
+struct txdesc {                         /* Transmit descriptor */
+        volatile unsigned short ladr;   /* Low order address of packet. This is
+                                         * a linear address in the Am186 memory
+                                         * space
+                                         */
+        volatile unsigned char  hadr;   /* High order address. Low 4 bits only,
+                                         * high 4 bits must be zero
+                                         */
+        volatile unsigned char  bits;   /* Status and config */
+        volatile unsigned short bcnt;   /* 2s complement of packet size in low
+                                         * 15 bits. Transmit terminal count
+                                         * interrupt enable in top bit.
+                                         */
+                 unsigned short unused; /* Not used in Tx */
+};
+
+struct rxdesc {                         /* Receive descriptor */
+        volatile unsigned short ladr;   /* Low order address of packet */
+        volatile unsigned char  hadr;   /* High order address */
+        volatile unsigned char  bits;   /* Status and config */
+        volatile unsigned short bcnt;   /* 2s complement of buffer size in low
+                                         * 15 bits. Receive terminal count
+                                         * interrupt enable in top bit.
+                                         */
+        volatile unsigned short mcnt;   /* Message byte count (15 bits) */
+};
+
+/* Convert a length into the 15 bit 2's complement */
+/* #define cnv_bcnt(len)   (( ~(len) + 1 ) & 0x7FFF ) */
+/* Since we need to set the high bit to enable the completion interrupt this
+ * can be made a lot simpler
+ */
+#define cnv_bcnt(len)   (-(len))
+
+/* Status and config bits for the above */
+#define DMA_OWN         0x80            /* SmartDMA owns the descriptor */
+#define TX_STP          0x02            /* Tx: start of packet */
+#define TX_ENP          0x01            /* Tx: end of packet */
+#define RX_ERR          0x40            /* Rx: error (OR of next 4 bits) */
+#define RX_FRAM         0x20            /* Rx: framing error */
+#define RX_OFLO         0x10            /* Rx: overflow error */
+#define RX_CRC          0x08            /* Rx: CRC error */
+#define RX_HBUF         0x04            /* Rx: buffer error */
+#define RX_STP          0x02            /* Rx: start of packet */
+#define RX_ENP          0x01            /* Rx: end of packet */
+
+
+/* Interrupts from the card are caused by various events and these are presented
+ * in a circular buffer as several events may be processed on one physical int
+ */
+#define MAX_CIRBUFF     32
+
+struct cirbuff {
+        unsigned char rdindex;          /* read, then increment and wrap */
+        unsigned char wrindex;          /* write, then increment and wrap */
+        unsigned char evntbuff[MAX_CIRBUFF];
+};
+
+/* Interrupt event codes.
+ * Where appropriate the two low order bits indicate the port number
+ */
+#define CTLA_CHG        0x18    /* Control signal changed */
+#define CTLB_CHG        0x19
+#define CTLC_CHG        0x1A
+#define CTLD_CHG        0x1B
+
+#define INIT_CPLT       0x20    /* Initialisation complete */
+#define INIT_FAIL       0x21    /* Initialisation failed */
+
+#define ABTA_SENT       0x24    /* Abort sent */
+#define ABTB_SENT       0x25
+#define ABTC_SENT       0x26
+#define ABTD_SENT       0x27
+
+#define TXA_UNDF        0x28    /* Transmission underflow */
+#define TXB_UNDF        0x29
+#define TXC_UNDF        0x2A
+#define TXD_UNDF        0x2B
+
+
+/* Port physical configuration. See farsync.h for field values */
+struct port_cfg {
+        unsigned short  lineInterface;  /* Physical interface type */
+        unsigned char   x25op;          /* Unused at present */
+        unsigned char   internalClock;  /* 1 => internal clock, 0 => external */
+        unsigned int    lineSpeed;      /* Speed in bps */
+};
+
+/* Finally sling all the above together into the shared memory structure.
+ * Sorry it's a hodge podge of arrays, structures and unused bits, it's been
+ * evolving under NT for some time so I guess we're stuck with it.
+ * The structure starts at offset SMC_BASE.
+ * See farsync.h for some field values.
+ */
+struct fst_shared {
+                                /* DMA descriptor rings */
+        struct rxdesc rxDescrRing[FST_MAX_PORTS][NUM_RX_BUFFER];
+        struct txdesc txDescrRing[FST_MAX_PORTS][NUM_TX_BUFFER];
+
+                                /* Obsolete small buffers */
+        char smallRxBuffer[FST_MAX_PORTS][NUM_RX_BUFFER][LEN_SMALL_RX_BUFFER];
+        char smallTxBuffer[FST_MAX_PORTS][NUM_TX_BUFFER][LEN_SMALL_TX_BUFFER];
+
+                                /* 0x00 => initialising, 0x01 => running,
+                                 * 0xFF => halted
+                                 */
+        unsigned char  taskStatus;
+
+                                /* Set to 0x01 by adapter to signal interrupt,
+                                 * set to 0xEE by host to acknowledge interrupt
+                                 */
+        unsigned char  interruptHandshake;
+
+        unsigned short smcVersion;      /* Must match SMC_VERSION */
+
+                                /* 0xIIVVRRBB where II = product ID, VV = major
+                                 * version, RR = revision and BB = build
+                                 */
+        unsigned int   smcFirmwareVersion;
+
+        unsigned short txa_done;        /* Obsolete completion flags */
+        unsigned short rxa_done;
+        unsigned short txb_done;
+        unsigned short rxb_done;
+        unsigned short txc_done;
+        unsigned short rxc_done;
+        unsigned short txd_done;
+        unsigned short rxd_done;
+
+        unsigned short mailbox[4];      /* Diagnostics mailbox. Not used */
+
+        struct cirbuff interruptEvent;  /* interrupt causes */
+
+        unsigned int v24IpSts[FST_MAX_PORTS];   /* V.24 control input status */
+        unsigned int v24OpSts[FST_MAX_PORTS];   /* V.24 control output status */
+
+        struct port_cfg portConfig[FST_MAX_PORTS];
+
+                                        /* lsb: 0=> present, 1=> absent */
+        unsigned short clockStatus[FST_MAX_PORTS];
+
+        unsigned short cableStatus;     /* lsb: 0=> present, 1=> absent */
+
+                                        /* transmit descriptor ring index */
+        unsigned short txDescrIndex[FST_MAX_PORTS];
+                                        /* receive descriptor ring index */
+        unsigned short rxDescrIndex[FST_MAX_PORTS];
+
+        unsigned short portMailbox[FST_MAX_PORTS][2];   /* command, modifier */
+        unsigned short cardMailbox[4];  /* Not used */
+
+                                /* Number of times that card thinks the host has
+                                 * missed an interrupt by not acknowledging
+                                 * within 2mS (I guess NT has problems)
+                                 */
+        unsigned int   interruptRetryCount;
+
+                                /* Driver private data used as an ID. We'll not
+                                 * use this on Linux I'd rather keep such things
+                                 * in main memory rather than on the PCI bus
+                                 */
+        unsigned int   portHandle[FST_MAX_PORTS];
+
+                                /* Count of Tx underflows for stats */
+        unsigned int   transmitBufferUnderflow[FST_MAX_PORTS];
+
+                                /* Debounced V.24 control input status */
+        unsigned int   v24DebouncedSts[FST_MAX_PORTS];
+
+                                /* Adapter debounce timers. Don't touch */
+        unsigned int   ctsTimer[FST_MAX_PORTS];
+        unsigned int   ctsTimerRun[FST_MAX_PORTS];
+        unsigned int   dcdTimer[FST_MAX_PORTS];
+        unsigned int   dcdTimerRun[FST_MAX_PORTS];
+
+                                /* Number of ports detected at startup */
+        unsigned int   numberOfPorts;
+
+        unsigned short _reserved[64];
+
+                                /* Bit-mask to enable features:
+                                 * Bit 0: 1 enables LED identify mode
+                                 */
+        unsigned short cardMode;
+
+        unsigned short portScheduleOffset;
+
+                                /* endOfSmcSignature MUST be the last member of
+                                 * the structure and marks the end of the shared
+                                 * memory. Adapter code initializes its value as
+                                 * END_SIG.
+                                 */
+        unsigned int   endOfSmcSignature;
+};
+
+/* endOfSmcSignature value */
+#define END_SIG                 0x12345678
+
+/* Mailbox values. (portMailbox) */
+#define NOP             0       /* No operation */
+#define ACK             1       /* Positive acknowledgement to PC driver */
+#define NAK             2       /* Negative acknowledgement to PC driver */
+#define STARTPORT       3       /* Start an HDLC port */
+#define STOPPORT        4       /* Stop an HDLC port */
+#define ABORTTX         5       /* Abort the transmitter for a port */
+#define SETV24O         6       /* Set V24 outputs */
+
+
+/* Larger buffers are positioned in memory at offset BFM_BASE */
+struct buf_window {
+  char txBuffer[FST_MAX_PORTS][NUM_TX_BUFFER][LEN_TX_BUFFER];
+  char rxBuffer[FST_MAX_PORTS][NUM_RX_BUFFER][LEN_RX_BUFFER];
+};
+
+/* Calculate offset of a buffer object within the shared memory window */
+#define BUF_OFFSET(X)   ((unsigned int)&(((struct buf_window *)BFM_BASE)->X))
+
+#pragma pack()
+
+
+/*      Device driver private information
+ *      =================================
+ */
+/*      Per port (line or channel) information
+ */
+struct fst_port_info {
+        void                   *if_ptr; /* Some drivers describe this as a
+                                         * general purpose pointer. However if
+                                         * using syncPPP it has a very specific
+                                         * purpose: it must be the first item in
+                                         * the structure pointed to by dev->priv
+                                         * and must in turn point to the
+                                         * associated ppp_device structure.
+                                         */
+        struct fst_card_info   *card;   /* Card we're associated with */
+        int                     index;  /* Port index on the card */
+        int                     proto;  /* Protocol we are running */
+        int                     hwif;   /* Line hardware (lineInterface copy) */
+        int                     run;    /* Port is running */
+        int                     rxpos;  /* Next Rx buffer to use */
+        int                     txpos;  /* Next Tx buffer to use */
+        int                     txipos; /* Next Tx buffer to check for free */
+        int                     txcnt;  /* Count of Tx buffers in use */
+        struct net_device      *dev;    /* Kernel network device entry */
+        struct net_device_stats stats;  /* Standard statistics */
+        struct ppp_device       pppdev; /* Link to syncPPP */
+};
+
+/*      Per card information
+ */
+struct fst_card_info {
+        char          *mem;             /* Card memory mapped to kernel space */
+        char          *ctlmem;          /* Control memory for PCI cards */
+        unsigned int   phys_mem;        /* Physical memory window address */
+        unsigned int   phys_ctlmem;     /* Physical control memory address */
+        unsigned int   irq;             /* Interrupt request line number */
+        unsigned int   nports;          /* Number of serial ports */
+        unsigned int   type;            /* Type index of card */
+        unsigned int   state;           /* State of card */
+        spinlock_t     card_lock;       /* Lock for SMP access */
+        unsigned short pci_conf;        /* PCI card config in I/O space */
+                                        /* Per port info */
+        struct fst_port_info ports[ FST_MAX_PORTS ];
+};
+
+/*
+ *      Card access locking
+ *      Note locking only covers the essentials, if someone wants to run two
+ *      copies of the download task for the same card on different processors
+ *      then they will get what they deserve. Likewise I see little point in
+ *      locking things like stats access where corruption is basically
+ *      harmless.
+ */
+#define FST_LOCK(C,F)   spin_lock_irqsave ( &C->card_lock, F )
+#define FST_UNLOCK(C,F) spin_unlock_irqrestore ( &C->card_lock, F )
+
+
+/*
+ *      Shared memory window access macros
+ *
+ *      We have a nice memory based structure above, which could be directly
+ *      mapped on i386 but might not work on other architectures unless we use
+ *      the readb,w,l and writeb,w,l macros. Unfortunately these macros take
+ *      physical offsets so we have to convert. The only saving grace is that
+ *      this should all collapse back to a simple indirection eventually.
+ */
+#define WIN_OFFSET(X)   ((long)&(((struct fst_shared *)SMC_BASE)->X))
+
+#define FST_RDB(C,E)    readb ((C)->mem + WIN_OFFSET(E))
+#define FST_RDW(C,E)    readw ((C)->mem + WIN_OFFSET(E))
+#define FST_RDL(C,E)    readl ((C)->mem + WIN_OFFSET(E))
+
+#define FST_WRB(C,E,B)  writeb ((B), (C)->mem + WIN_OFFSET(E))
+#define FST_WRW(C,E,W)  writew ((W), (C)->mem + WIN_OFFSET(E))
+#define FST_WRL(C,E,L)  writel ((L), (C)->mem + WIN_OFFSET(E))
+
+
+/*
+ *      Debug support
+ */
+#if FST_DEBUG
+
+static int fst_debug_mask = { FST_DEBUG };
+
+/* Most common debug activity is to print something if the corresponding bit
+ * is set in the debug mask. Note: this uses a non-ANSI extension in GCC to
+ * support variable numbers of macro parameters. The inverted if prevents us
+ * eating someone else's else clause.
+ */
+#define dbg(F,fmt,A...) if ( ! ( fst_debug_mask & (F))) \
+                                ; \
+                        else \
+                                printk ( KERN_DEBUG FST_NAME ": " fmt, ## A )
+
+#else
+# define dbg(X...)      /* NOP */
+#endif
+
+
+/*      Printing short cuts
+ */
+#define printk_err(fmt,A...)    printk ( KERN_ERR     FST_NAME ": " fmt, ## A )
+#define printk_warn(fmt,A...)   printk ( KERN_WARNING FST_NAME ": " fmt, ## A )
+#define printk_info(fmt,A...)   printk ( KERN_INFO    FST_NAME ": " fmt, ## A )
+
+
+/*
+ *      PCI ID lookup table
+ */
+static struct pci_device_id fst_pci_dev_id[] __devinitdata = {
+        { FSC_PCI_VENDOR_ID, T2P_PCI_DEVICE_ID, PCI_ANY_ID, PCI_ANY_ID, 0, 0,
+                                        FST_TYPE_T2P },
+        { FSC_PCI_VENDOR_ID, T4P_PCI_DEVICE_ID, PCI_ANY_ID, PCI_ANY_ID, 0, 0,
+                                        FST_TYPE_T4P },
+        { 0, }                          /* End */
+};
+
+MODULE_DEVICE_TABLE ( pci, fst_pci_dev_id );
+
+
+/*      Card control functions
+ *      ======================
+ */
+/*      Place the processor in reset state
+ *
+ * Used to be a simple write to card control space but a glitch in the latest
+ * AMD Am186CH processor means that we now have to do it by asserting and de-
+ * asserting the PLX chip PCI Adapter Software Reset. Bit 30 in CNTRL register
+ * at offset 0x50.
+ */
+static inline void
+fst_cpureset ( struct fst_card_info *card )
+{
+        unsigned int regval;
+
+        regval = inl ( card->pci_conf + 0x50 );
+
+        outl ( regval |  0x40000000, card->pci_conf + 0x50 );
+        outl ( regval & ~0x40000000, card->pci_conf + 0x50 );
+}
+
+/*      Release the processor from reset
+ */
+static inline void
+fst_cpurelease ( struct fst_card_info *card )
+{
+        (void) readb ( card->ctlmem );
+}
+
+/*      Clear the cards interrupt flag
+ */
+static inline void
+fst_clear_intr ( struct fst_card_info *card )
+{
+        /* Poke the appropriate PLX chip register (same as enabling interrupts)
+         */
+        outw ( 0x0543, card->pci_conf + 0x4C );
+}
+
+/*      Disable card interrupts
+ */
+static inline void
+fst_disable_intr ( struct fst_card_info *card )
+{
+        outw ( 0x0000, card->pci_conf + 0x4C );
+}
+
+
+/*      Issue a Mailbox command for a port.
+ *      Note we issue them on a fire and forget basis, not expecting to see an
+ *      error and not waiting for completion.
+ */
+static void
+fst_issue_cmd ( struct fst_port_info *port, unsigned short cmd )
+{
+        unsigned short mbval;
+        unsigned long flags;
+        int safety;
+
+        FST_LOCK ( port->card, flags );
+        mbval = FST_RDW ( port->card, portMailbox[port->index][0]);
+
+        safety = 0;
+        /* Wait for any previous command to complete */
+        while ( mbval > NAK )
+        {
+                FST_UNLOCK ( port->card, flags );
+                schedule_timeout ( 1 );
+                FST_LOCK ( port->card, flags );
+
+                if ( ++safety > 1000 )
+                {
+                        printk_err ("Mailbox safety timeout\n");
+                        break;
+                }
+
+                mbval = FST_RDW ( port->card, portMailbox[port->index][0]);
+        }
+        if ( safety > 0 )
+        {
+                dbg ( DBG_CMD,"Mailbox clear after %d jiffies\n", safety );
+        }
+        if ( mbval == NAK )
+        {
+                dbg ( DBG_CMD,"issue_cmd: previous command was NAK'd\n");
+        }
+
+        FST_WRW ( port->card, portMailbox[port->index][0], cmd );
+
+        if ( cmd == ABORTTX || cmd == STARTPORT )
+        {
+                port->txpos  = 0;
+                port->txipos = 0;
+                port->txcnt  = 0;
+        }
+
+        FST_UNLOCK ( port->card, flags );
+}
+
+
+/*      Port output signals control
+ */
+static inline void
+fst_op_raise ( struct fst_port_info *port, unsigned int outputs )
+{
+        outputs |= FST_RDL ( port->card, v24OpSts[port->index]);
+        FST_WRL ( port->card, v24OpSts[port->index], outputs );
+
+        if ( port->run )
+                fst_issue_cmd ( port, SETV24O );
+}
+
+static inline void
+fst_op_lower ( struct fst_port_info *port, unsigned int outputs )
+{
+        outputs = ~outputs & FST_RDL ( port->card, v24OpSts[port->index]);
+        FST_WRL ( port->card, v24OpSts[port->index], outputs );
+
+        if ( port->run )
+                fst_issue_cmd ( port, SETV24O );
+}
+
+
+/*
+ *      Setup port Rx buffers
+ */
+static void
+fst_rx_config ( struct fst_port_info *port )
+{
+        int i;
+        int pi;
+        unsigned int offset;
+        unsigned long flags;
+        struct fst_card_info *card;
+
+        pi   = port->index;
+        card = port->card;
+        FST_LOCK ( card, flags );
+        for ( i = 0 ; i < NUM_RX_BUFFER ; i++ )
+        {
+                offset = BUF_OFFSET ( rxBuffer[pi][i]);
+
+                FST_WRW ( card, rxDescrRing[pi][i].ladr,
+                                        (unsigned short) offset );
+                FST_WRB ( card, rxDescrRing[pi][i].hadr,
+                                        (unsigned char)( offset >> 16 ));
+                FST_WRW ( card, rxDescrRing[pi][i].bcnt,
+                                        cnv_bcnt ( LEN_RX_BUFFER ));
+                FST_WRW ( card, rxDescrRing[pi][i].mcnt, 0 );
+                FST_WRB ( card, rxDescrRing[pi][i].bits, DMA_OWN );
+        }
+        port->rxpos  = 0;
+        FST_UNLOCK ( card, flags );
+}
+
+
+/*
+ *      Setup port Tx buffers
+ */
+static void
+fst_tx_config ( struct fst_port_info *port )
+{
+        int i;
+        int pi;
+        unsigned int offset;
+        unsigned long flags;
+        struct fst_card_info *card;
+
+        pi   = port->index;
+        card = port->card;
+        FST_LOCK ( card, flags );
+        for ( i = 0 ; i < NUM_TX_BUFFER ; i++ )
+        {
+                offset = BUF_OFFSET ( txBuffer[pi][i]);
+
+                FST_WRW ( card, txDescrRing[pi][i].ladr,
+                                        (unsigned short) offset );
+                FST_WRB ( card, txDescrRing[pi][i].hadr,
+                                        (unsigned char)( offset >> 16 ));
+                FST_WRW ( card, txDescrRing[pi][i].bcnt, 0 );
+                FST_WRB ( card, txDescrRing[pi][i].bits, 0 );
+        }
+        port->txpos  = 0;
+        port->txipos = 0;
+        port->txcnt  = 0;
+        FST_UNLOCK ( card, flags );
+}
+
+
+/*      Control signal change interrupt event
+ */
+static void
+fst_intr_ctlchg ( struct fst_card_info *card, struct fst_port_info *port )
+{
+        int signals;
+
+        signals = FST_RDL ( card, v24DebouncedSts[port->index]);
+
+        if ( signals & (( port->hwif == X21 ) ? IPSTS_INDICATE : IPSTS_DCD ))
+        {
+                if ( ! netif_carrier_ok ( port->dev ))
+                {
+                        dbg ( DBG_INTR,"DCD active\n");
+
+                        /* Poke sPPP to renegotiate */
+                        if ( port->proto == FST_HDLC || port->proto == FST_PPP )
+                        {
+                                sppp_reopen ( port->dev );
+                        }
+
+                        netif_carrier_on ( port->dev );
+                }
+        }
+        else
+        {
+                if ( netif_carrier_ok ( port->dev ))
+                {
+                        dbg ( DBG_INTR,"DCD lost\n");
+                        netif_carrier_off ( port->dev );
+                }
+        }
+}
+
+
+/*      Rx complete interrupt
+ */
+static void
+fst_intr_rx ( struct fst_card_info *card, struct fst_port_info *port )
+{
+        unsigned char dmabits;
+        int pi;
+        int rxp;
+        unsigned short len;
+        struct sk_buff *skb;
+        int i;
+
+
+        /* Check we have a buffer to process */
+        pi  = port->index;
+        rxp = port->rxpos;
+        dmabits = FST_RDB ( card, rxDescrRing[pi][rxp].bits );
+        if ( dmabits & DMA_OWN )
+        {
+                dbg ( DBG_RX | DBG_INTR,"intr_rx: No buffer port %d pos %d\n",
+                                        pi, rxp );
+                return;
+        }
+
+        /* Get buffer length */
+        len = FST_RDW ( card, rxDescrRing[pi][rxp].mcnt );
+        /* Discard the CRC */
+        len -= 2;
+
+        /* Check buffer length and for other errors. We insist on one packet
+         * in one buffer. This simplifies things greatly and since we've
+         * allocated 8K it shouldn't be a real world limitation
+         */
+        dbg ( DBG_RX,"intr_rx: %d,%d: flags %x len %d\n", pi, rxp, dmabits,
+                                        len );
+        if ( dmabits != ( RX_STP | RX_ENP ) || len > LEN_RX_BUFFER - 2 )
+        {
+                port->stats.rx_errors++;
+
+                /* Update error stats and discard buffer */
+                if ( dmabits & RX_OFLO )
+                {
+                        port->stats.rx_fifo_errors++;
+                }
+                if ( dmabits & RX_CRC )
+                {
+                        port->stats.rx_crc_errors++;
+                }
+                if ( dmabits & RX_FRAM )
+                {
+                        port->stats.rx_frame_errors++;
+                }
+                if ( dmabits == ( RX_STP | RX_ENP ))
+                {
+                        port->stats.rx_length_errors++;
+                }
+
+                /* Discard buffer descriptors until we see the end of packet
+                 * marker
+                 */
+                i = 0;
+                while (( dmabits & ( DMA_OWN | RX_ENP )) == 0 )
+                {
+                        FST_WRB ( card, rxDescrRing[pi][rxp].bits, DMA_OWN );
+                        if ( ++rxp >= NUM_RX_BUFFER )
+                                rxp = 0;
+                        if ( ++i > NUM_RX_BUFFER )
+                        {
+                                dbg ( DBG_ASS,"intr_rx: Discarding more bufs"
+                                                " than we have\n");
+                                break;
+                        }
+                        dmabits = FST_RDB ( card, rxDescrRing[pi][rxp].bits );
+                }
+
+                /* Discard the terminal buffer */
+                if ( ! ( dmabits & DMA_OWN ))
+                {
+                        FST_WRB ( card, rxDescrRing[pi][rxp].bits, DMA_OWN );
+                        if ( ++rxp >= NUM_RX_BUFFER )
+                                rxp = 0;
+                }
+                port->rxpos = rxp;
+                return;
+        }
+
+        /* Allocate SKB */
+        if (( skb = dev_alloc_skb ( len )) == NULL )
+        {
+                dbg ( DBG_RX,"intr_rx: can't allocate buffer\n");
+
+                port->stats.rx_dropped++;
+
+                /* Return descriptor to card */
+                FST_WRB ( card, rxDescrRing[pi][rxp].bits, DMA_OWN );
+
+                if ( ++rxp >= NUM_RX_BUFFER )
+                        port->rxpos = 0;
+                else
+                        port->rxpos = rxp;
+                return;
+        }
+
+        memcpy_fromio ( skb_put ( skb, len ),
+                                card->mem + BUF_OFFSET ( rxBuffer[pi][rxp][0]),
+                                len );
+
+        /* Reset buffer descriptor */
+        FST_WRB ( card, rxDescrRing[pi][rxp].bits, DMA_OWN );
+        if ( ++rxp >= NUM_RX_BUFFER )
+                port->rxpos = 0;
+        else
+                port->rxpos = rxp;
+
+        /* Update stats */
+        port->stats.rx_packets++;
+        port->stats.rx_bytes += len;
+
+        /* Push upstream */
+        if ( port->proto == FST_HDLC || port->proto == FST_PPP )
+        {
+                /* Mark for further processing by sPPP module */
+                skb->protocol = htons ( ETH_P_WAN_PPP );
+        }
+        else
+        {
+                /* DEC customer specific protocol (since nothing defined for
+                 * marking raw data), at least one other driver uses this value
+                 * for this purpose.
+                 */
+                skb->protocol = htons ( ETH_P_CUST );
+                skb->pkt_type = PACKET_HOST;
+        }
+        skb->mac.raw = skb->data;
+        skb->dev = port->dev;
+        netif_rx ( skb );
+
+        port->dev->last_rx = jiffies;
+}
+
+
+/*
+ *      The interrupt service routine
+ *      Dev_id is our fst_card_info pointer
+ */
+static void
+fst_intr ( int irq, void *dev_id, struct pt_regs *regs )
+{
+        struct fst_card_info *card;
+        struct fst_port_info *port;
+        int rdidx;                      /* Event buffer indices */
+        int wridx;
+        int event;                      /* Actual event for processing */
+        int pi;
+
+        if (( card = dev_id ) == NULL )
+        {
+                dbg ( DBG_INTR,"intr: spurious %d\n", irq );
+                return;
+        }
+
+        dbg ( DBG_INTR,"intr: %d %p\n", irq, card );
+
+        spin_lock ( &card->card_lock );
+
+        /* Clear and reprime the interrupt source */
+        fst_clear_intr ( card );
+
+        /* Set the software acknowledge */
+        FST_WRB ( card, interruptHandshake, 0xEE );
+
+        /* Drain the event queue */
+        rdidx = FST_RDB ( card, interruptEvent.rdindex );
+        wridx = FST_RDB ( card, interruptEvent.wrindex );
+        while ( rdidx != wridx )
+        {
+                event = FST_RDB ( card, interruptEvent.evntbuff[rdidx]);
+
+                port = &card->ports[event & 0x03];
+
+                dbg ( DBG_INTR,"intr: %x\n", event );
+
+                switch ( event )
+                {
+                case CTLA_CHG:
+                case CTLB_CHG:
+                case CTLC_CHG:
+                case CTLD_CHG:
+                        if ( port->run && port->dev != NULL )
+                                fst_intr_ctlchg ( card, port );
+                        break;
+
+                case ABTA_SENT:
+                case ABTB_SENT:
+                case ABTC_SENT:
+                case ABTD_SENT:
+                        dbg ( DBG_TX,"Abort complete port %d\n", event & 0x03 );
+                        break;
+
+                case TXA_UNDF:
+                case TXB_UNDF:
+                case TXC_UNDF:
+                case TXD_UNDF:
+                        /* Difficult to see how we'd get this given that we
+                         * always load up the entire packet for DMA.
+                         */
+                        dbg ( DBG_TX,"Tx underflow port %d\n", event & 0x03 );
+                        break;
+
+                case INIT_CPLT:
+                        dbg ( DBG_INIT,"Card init OK intr\n");
+                        break;
+
+                case INIT_FAIL:
+                        dbg ( DBG_INIT,"Card init FAILED intr\n");
+                        card->state = FST_IFAILED;
+                        break;
+
+                default:
+                        printk_err ("intr: unknown card event code. ignored\n");
+                        break;
+                }
+
+                /* Bump and wrap the index */
+                if ( ++rdidx >= MAX_CIRBUFF )
+                        rdidx = 0;
+        }
+        FST_WRB ( card, interruptEvent.rdindex, rdidx );
+
+        for ( pi = 0, port = card->ports ; pi < card->nports ; pi++, port++ )
+        {
+                if ( port->dev == NULL || ! port->run )
+                        continue;
+
+                /* Check for rx completions */
+                while ( ! ( FST_RDB ( card, rxDescrRing[pi][port->rxpos].bits )
+                                                                & DMA_OWN ))
+                {
+                        fst_intr_rx ( card, port );
+                }
+
+                /* Check for Tx completions */
+                while ( port->txcnt > 0 && ! ( FST_RDB ( card,
+                        txDescrRing[pi][port->txipos].bits ) & DMA_OWN ))
+                {
+                        --port->txcnt;
+                        if ( ++port->txipos >= NUM_TX_BUFFER )
+                                port->txipos = 0;
+                        netif_wake_queue ( port->dev );
+                }
+        }
+
+        spin_unlock ( &card->card_lock );
+}
+
+
+/*      Check that the shared memory configuration is one that we can handle
+ *      and that some basic parameters are correct
+ */
+static void
+check_started_ok ( struct fst_card_info *card )
+{
+        int i;
+
+        /* Check structure version and end marker */
+        if ( FST_RDW ( card, smcVersion ) != SMC_VERSION )
+        {
+                printk_err ("Bad shared memory version %d expected %d\n",
+                                FST_RDW ( card, smcVersion ), SMC_VERSION );
+                card->state = FST_BADVERSION;
+                return;
+        }
+        if ( FST_RDL ( card, endOfSmcSignature ) != END_SIG )
+        {
+                printk_err ("Missing shared memory signature\n");
+                card->state = FST_BADVERSION;
+                return;
+        }
+        /* Firmware status flag, 0x00 = initialising, 0x01 = OK, 0xFF = fail */
+        if (( i = FST_RDB ( card, taskStatus )) == 0x01 )
+        {
+                card->state = FST_RUNNING;
+        }
+        else if ( i == 0xFF )
+        {
+                printk_err ("Firmware initialisation failed. Card halted\n");
+                card->state = FST_HALTED;
+                return;
+        }
+        else if ( i != 0x00 )
+        {
+                printk_err ("Unknown firmware status 0x%x\n", i );
+                card->state = FST_HALTED;
+                return;
+        }
+
+        /* Finally check the number of ports reported by firmware against the
+         * number we assumed at card detection. Should never happen with
+         * existing firmware etc so we just report it for the moment.
+         */
+        if ( FST_RDL ( card, numberOfPorts ) != card->nports )
+        {
+                printk_warn ("Port count mismatch."
+                                " Firmware thinks %d we say %d\n",
+                                FST_RDL ( card, numberOfPorts ), card->nports );
+        }
+}
+
+
+static int
+fst_change_mtu ( struct net_device *dev, int new_mtu )
+{
+        if ( new_mtu < 128 || new_mtu > FST_MAX_MTU )
+                return -EINVAL;
+        dev->mtu = new_mtu;
+        return 0;
+}
+
+
+/* Sooner or later you can't avoid a forward declaration */
+static int fst_ioctl ( struct net_device *dev, struct ifreq *ifr, int cmd );
+
+static int
+switch_proto ( struct fst_port_info *port, int new_proto )
+{
+        int err;
+        int orig_mtu;
+        struct net_device *dev;
+
+        dev = port->dev;
+
+        /* Turn off sPPP module ? */
+        if (( new_proto != FST_HDLC && new_proto != FST_PPP )
+                && ( port->proto == FST_HDLC || port->proto == FST_PPP ))
+        {
+                sppp_close ( port->pppdev.dev );
+                sppp_detach ( port->pppdev.dev );
+
+                /* Reset some fields overwritten by sPPP */
+                dev->hard_header     = NULL;
+                dev->rebuild_header  = NULL;
+                dev->tx_queue_len    = FST_TX_QUEUE_LEN;
+                dev->type            = ARPHRD_MYTYPE;
+                dev->addr_len        = 0;
+                dev->hard_header_len = 0;
+                dev->do_ioctl        = fst_ioctl;
+                dev->change_mtu      = fst_change_mtu;
+                dev->flags           = IFF_POINTOPOINT|IFF_NOARP;
+
+                return 0;
+        }
+
+        /* Turn on sPPP ? */
+        if (( new_proto == FST_HDLC || new_proto == FST_PPP )
+                && ( port->proto != FST_HDLC && port->proto != FST_PPP ))
+        {
+                orig_mtu = dev->mtu;
+
+                /* Attach to sync PPP module */
+                port->pppdev.dev = dev;
+                sppp_attach ( &port->pppdev );
+
+                if ( orig_mtu < dev->mtu )
+                        dev->change_mtu ( dev, orig_mtu );
+
+                /* Claw back the ioctl routine. We promise to be good and call
+                 * the sync PPP routines once we've eliminated our functions.
+                 */
+                dev->do_ioctl = fst_ioctl;
+
+                /* Set the mode */
+                if ( new_proto == FST_HDLC )
+                {
+                        err = sppp_do_ioctl ( port->pppdev.dev, NULL,
+                                                                SPPPIOCCISCO );
+                }
+                else
+                {
+                        err = sppp_do_ioctl ( port->pppdev.dev, NULL,
+                                                                SPPPIOCPPP );
+                }
+
+                /* Open the device */
+                if ( err == 0 )
+                {
+                        (void) sppp_open ( port->dev );
+                }
+
+                return err;
+        }
+
+        /* Switch sPPP mode to that desired */
+        err = 0;
+        if ( new_proto == FST_HDLC && port->pppdev.dev != NULL )
+        {
+                err = sppp_do_ioctl ( port->pppdev.dev, NULL, SPPPIOCCISCO );
+        }
+        else if ( new_proto == FST_PPP && port->pppdev.dev != NULL )
+        {
+                err = sppp_do_ioctl ( port->pppdev.dev, NULL, SPPPIOCPPP );
+        }
+
+        /* Anything else is switching from one raw mode to another which is
+         * basically a NOP
+         */
+
+        return err;
+}
+
+
+static int
+set_conf_from_info ( struct fst_card_info *card, struct fst_port_info *port,
+                struct fstioc_info *info )
+{
+        int err;
+
+        /* Set things according to the user set valid flags */
+        err = 0;
+        if ( info->valid & FSTVAL_PROTO )
+        {
+                if ( port->proto != info->proto )
+                {
+                        err = switch_proto ( port, info->proto );
+                        if ( err == 0 )
+                                port->proto = info->proto;
+                }
+        }
+        if ( info->valid & FSTVAL_CABLE )
+        {
+                FST_WRB ( card, portConfig[port->index].lineInterface,
+                                        info->lineInterface );
+                port->hwif = info->lineInterface;
+        }
+        if ( info->valid & FSTVAL_SPEED )
+        {
+                FST_WRL ( card, portConfig[port->index].lineSpeed,
+                                        info->lineSpeed );
+                FST_WRB ( card, portConfig[port->index].internalClock,
+                                        info->internalClock );
+        }
+        if ( info->valid & FSTVAL_MODE )
+        {
+                FST_WRW ( card, cardMode, info->cardMode );
+        }
+#if FST_DEBUG
+        if ( info->valid & FSTVAL_DEBUG )
+        {
+                fst_debug_mask = info->debug;
+        }
+#endif
+
+        return err;
+}
+
+static void
+gather_conf_info ( struct fst_card_info *card, struct fst_port_info *port,
+                struct fstioc_info *info )
+{
+        int i;
+
+        memset ( info, 0, sizeof ( struct fstioc_info ));
+
+        i = port->index;
+        info->nports = card->nports;
+        info->type   = card->type;
+        info->state  = card->state;
+        info->proto  = port->proto;
+        info->index  = i;
+#if FST_DEBUG
+        info->debug  = fst_debug_mask;
+#endif
+
+        /* Only mark information as valid if card is running.
+         * Copy the data anyway in case it is useful for diagnostics
+         */
+        info->valid
+                = (( card->state == FST_RUNNING ) ? FSTVAL_ALL : FSTVAL_CARD )
+#if FST_DEBUG
+                | FSTVAL_DEBUG
+#endif
+                ;
+
+        info->lineInterface = FST_RDW ( card, portConfig[i].lineInterface );
+        info->internalClock = FST_RDB ( card, portConfig[i].internalClock );
+        info->lineSpeed     = FST_RDL ( card, portConfig[i].lineSpeed );
+        info->v24IpSts      = FST_RDL ( card, v24IpSts[i] );
+        info->v24OpSts      = FST_RDL ( card, v24OpSts[i] );
+        info->clockStatus   = FST_RDW ( card, clockStatus[i] );
+        info->cableStatus   = FST_RDW ( card, cableStatus );
+        info->cardMode      = FST_RDW ( card, cardMode );
+        info->smcFirmwareVersion  = FST_RDL ( card, smcFirmwareVersion );
+}
+
+
+static int
+fst_ioctl ( struct net_device *dev, struct ifreq *ifr, int cmd )
+{
+        struct fst_card_info *card;
+        struct fst_port_info *port;
+        struct fstioc_write wrthdr;
+        struct fstioc_info info;
+        unsigned long flags;
+
+        dbg ( DBG_IOCTL,"ioctl: %x, %p\n", cmd, ifr->ifr_data );
+
+        port = dev->priv;
+        card = port->card;
+
+        if ( !capable ( CAP_NET_ADMIN ))
+                return -EPERM;
+
+        switch ( cmd )
+        {
+        case FSTCPURESET:
+                fst_cpureset ( card );
+                card->state = FST_RESET;
+                return 0;
+
+        case FSTCPURELEASE:
+                fst_cpurelease ( card );
+                card->state = FST_STARTING;
+                return 0;
+
+        case FSTWRITE:                  /* Code write (download) */
+
+                /* First copy in the header with the length and offset of data
+                 * to write
+                 */
+                if ( ifr->ifr_data == NULL )
+                {
+                        return -EINVAL;
+                }
+                if ( copy_from_user ( &wrthdr, ifr->ifr_data,
+                                        sizeof ( struct fstioc_write )))
+                {
+                        return -EFAULT;
+                }
+
+                /* Sanity check the parameters. We don't support partial writes
+                 * when going over the top
+                 */
+                if ( wrthdr.size + wrthdr.offset > FST_MEMSIZE )
+                {
+                        return -ENXIO;
+                }
+
+                /* Now copy the data to the card.
+                 * This will probably break on some architectures.
+                 * I'll fix it when I have something to test on.
+                 */
+                if ( copy_from_user ( card->mem + wrthdr.offset,
+                                ifr->ifr_data + sizeof ( struct fstioc_write ),
+                                wrthdr.size ))
+                {
+                        return -EFAULT;
+                }
+
+                /* Writes to the memory of a card in the reset state constitute
+                 * a download
+                 */
+                if ( card->state == FST_RESET )
+                {
+                        card->state = FST_DOWNLOAD;
+                }
+                return 0;
+
+        case FSTGETCONF:
+
+                /* If card has just been started check the shared memory config
+                 * version and marker
+                 */
+                if ( card->state == FST_STARTING )
+                {
+                        check_started_ok ( card );
+
+                        /* If everything checked out enable card interrupts */
+                        if ( card->state == FST_RUNNING )
+                        {
+                                FST_LOCK ( port->card, flags );
+                                fst_clear_intr ( card );
+                                FST_WRB ( card, interruptHandshake, 0xEE );
+                                FST_UNLOCK ( port->card, flags );
+                        }
+                }
+
+                if ( ifr->ifr_data == NULL )
+                {
+                        return -EINVAL;
+                }
+
+                gather_conf_info ( card, port, &info );
+
+                if ( copy_to_user ( ifr->ifr_data, &info, sizeof ( info )))
+                {
+                        return -EFAULT;
+                }
+                return 0;
+
+        case FSTSETCONF:
+
+                if ( copy_from_user ( &info,  ifr->ifr_data, sizeof ( info )))
+                {
+                        return -EFAULT;
+                }
+
+                return set_conf_from_info ( card, port, &info );
+
+        default:
+                /* Not one of ours. Pass it through to sPPP package */
+                if ( port->proto == FST_HDLC || port->proto == FST_PPP )
+                        return sppp_do_ioctl ( dev, ifr, cmd );
+                else
+                        return -EINVAL;
+        }
+}
+
+
+static void
+fst_openport ( struct fst_port_info *port )
+{
+        int signals;
+
+        /* Only init things if card is actually running. This allows open to
+         * succeed for downloads etc.
+         */
+        if ( port->card->state == FST_RUNNING )
+        {
+                if ( port->run )
+                {
+                        dbg ( DBG_OPEN,"open: found port already running\n");
+
+                        fst_issue_cmd ( port, STOPPORT );
+                        port->run = 0;
+                }
+
+                fst_rx_config ( port );
+                fst_tx_config ( port );
+                fst_op_raise ( port, OPSTS_RTS | OPSTS_DTR );
+
+                fst_issue_cmd ( port, STARTPORT );
+                port->run = 1;
+
+                signals = FST_RDL ( port->card, v24DebouncedSts[port->index]);
+                if ( signals & (( port->hwif == X21 ) ? IPSTS_INDICATE
+                                                      : IPSTS_DCD ))
+                        netif_carrier_on ( port->dev );
+                else
+                        netif_carrier_off ( port->dev );
+        }
+}
+
+static void
+fst_closeport ( struct fst_port_info *port )
+{
+        if ( port->card->state == FST_RUNNING )
+        {
+                if ( port->run )
+                {
+                        port->run = 0;
+                        fst_op_lower ( port, OPSTS_RTS | OPSTS_DTR );
+
+                        fst_issue_cmd ( port, STOPPORT );
+                }
+                else
+                {
+                        dbg ( DBG_OPEN,"close: port not running\n");
+                }
+        }
+}
+
+
+static int
+fst_open ( struct net_device *dev )
+{
+        struct fst_card_info *card;
+        struct fst_port_info *port;
+        int orig_mtu;
+        int err;
+
+        MOD_INC_USE_COUNT;
+
+        port = dev->priv;
+        card = port->card;
+
+        switch ( port->proto )
+        {
+        case FST_HDLC:
+        case FST_PPP:
+
+                orig_mtu = dev->mtu;
+
+                /* Attach to sync PPP module */
+                port->pppdev.dev = dev;
+                sppp_attach ( &port->pppdev );
+
+                if ( orig_mtu < dev->mtu )
+                        dev->change_mtu ( dev, orig_mtu );
+
+                /* Claw back the ioctl routine. We promise to be good and call
+                 * the sync PPP routines once we've eliminated our functions.
+                 */
+                dev->do_ioctl = fst_ioctl;
+
+                err = sppp_do_ioctl ( dev, NULL, port->proto == FST_HDLC
+                                                ? SPPPIOCCISCO : SPPPIOCPPP );
+                if ( err )
+                        goto error_1;
+
+                err = sppp_open ( dev );
+                if ( err )
+                        goto error_1;
+                break;
+
+        case FST_MONITOR:
+        case FST_RAW:
+                break;
+
+        default:
+                dbg ( DBG_OPEN,"open: Unknown proto %d\n", port->proto );
+                break;
+        }
+
+        fst_openport ( port );
+
+        netif_wake_queue ( dev );
+        return 0;
+
+error_1:
+        sppp_detach ( dev );
+        MOD_DEC_USE_COUNT;
+        return err;
+}
+
+static int
+fst_close ( struct net_device *dev )
+{
+        struct fst_port_info *port;
+
+        port = dev->priv;
+
+        netif_stop_queue ( dev );
+        switch ( port->proto )
+        {
+        case FST_HDLC:
+        case FST_PPP:
+                sppp_close ( dev );
+                sppp_detach ( dev );
+                break;
+
+        case FST_MONITOR:
+        case FST_RAW:
+                break;
+
+        default:
+                dbg ( DBG_OPEN,"close: Unknown proto %d\n", port->proto );
+                break;
+        }
+
+        fst_closeport ( port );
+
+        MOD_DEC_USE_COUNT;
+        return 0;
+}
+
+
+static void
+fst_tx_timeout ( struct net_device *dev )
+{
+        struct fst_port_info *port;
+
+        dbg ( DBG_INTR | DBG_TX,"tx_timeout\n");
+
+        port = dev->priv;
+
+        port->stats.tx_errors++;
+        port->stats.tx_aborted_errors++;
+
+        if ( port->txcnt > 0 )
+                fst_issue_cmd ( port, ABORTTX );
+
+        dev->trans_start = jiffies;
+        netif_wake_queue ( dev );
+}
+
+
+static int
+fst_start_xmit ( struct sk_buff *skb, struct net_device *dev )
+{
+        struct fst_card_info *card;
+        struct fst_port_info *port;
+        unsigned char dmabits;
+        unsigned long flags;
+        int pi;
+        int txp;
+
+        port = dev->priv;
+        card = port->card;
+
+        /* Drop packet if in monitor (rx only) mode */
+        if ( port->proto == FST_MONITOR )
+        {
+                dev_kfree_skb ( skb );
+                return 0;
+        }
+
+        /* Drop packet with error if we don't have carrier */
+        if ( ! netif_carrier_ok ( dev ))
+        {
+                dev_kfree_skb ( skb );
+                port->stats.tx_errors++;
+                port->stats.tx_carrier_errors++;
+                return 0;
+        }
+
+        /* Drop it if it's too big! MTU failure ? */
+        if ( skb->len > LEN_TX_BUFFER )
+        {
+                dbg ( DBG_TX,"Packet too large %d vs %d\n", skb->len,
+                                                LEN_TX_BUFFER );
+                dev_kfree_skb ( skb );
+                port->stats.tx_errors++;
+                return 0;
+        }
+
+        /* Check we have a buffer */
+        pi = port->index;
+        FST_LOCK ( card, flags );
+        txp = port->txpos;
+        dmabits = FST_RDB ( card, txDescrRing[pi][txp].bits );
+        if ( dmabits & DMA_OWN )
+        {
+                FST_UNLOCK ( card, flags );
+                dbg ( DBG_TX,"Out of Tx buffers\n");
+                dev_kfree_skb ( skb );
+                port->stats.tx_errors++;
+                return 0;
+        }
+        if ( ++port->txpos >= NUM_TX_BUFFER )
+                port->txpos = 0;
+
+        if ( ++port->txcnt >= NUM_TX_BUFFER )
+                netif_stop_queue ( dev );
+
+        /* Release the card lock before we copy the data as we now have
+         * exclusive access to the buffer.
+         */
+        FST_UNLOCK ( card, flags );
+
+        /* Enqueue the packet */
+        memcpy_toio ( card->mem + BUF_OFFSET ( txBuffer[pi][txp][0]),
+                                                skb->data, skb->len );
+        FST_WRW ( card, txDescrRing[pi][txp].bcnt, cnv_bcnt ( skb->len ));
+        FST_WRB ( card, txDescrRing[pi][txp].bits, DMA_OWN | TX_STP | TX_ENP );
+
+        port->stats.tx_packets++;
+        port->stats.tx_bytes += skb->len;
+
+        dev_kfree_skb ( skb );
+
+        dev->trans_start = jiffies;
+        return 0;
+}
+
+
+static struct net_device_stats *
+fst_get_stats ( struct net_device *dev )
+{
+        struct fst_port_info *port;
+
+        if (( port = dev->priv ) != NULL )
+                return &port->stats;
+        else
+                return NULL;
+}
+
+
+/*
+ *      Card setup having checked hardware resources.
+ *      Should be pretty bizarre if we get an error here (kernel memory
+ *      exhaustion is one possibility). If we do see a problem we report it
+ *      via a printk and leave the corresponding interface and all that follow
+ *      disabled.
+ */
+static char *type_strings[] __devinitdata = {
+        "no hardware",                  /* Should never be seen */
+        "FarSync T2P",
+        "FarSync T4P"
+};
+
+static void __devinit
+fst_init_card ( struct fst_card_info *card )
+{
+        int i;
+        int err;
+        struct net_device *dev;
+
+        /* We're working on a number of ports based on the card ID. If the
+         * firmware detects something different later (should never happen)
+         * we'll have to revise it in some way then.
+         */
+        for ( i = 0 ; i < card->nports ; i++ )
+        {
+                card->ports[i].if_ptr = &card->ports[i].pppdev;
+                card->ports[i].card   = card;
+                card->ports[i].index  = i;
+                card->ports[i].proto  = FST_HDLC;
+                card->ports[i].run    = 0;
+
+                dev = kmalloc ( sizeof ( struct net_device ), GFP_KERNEL );
+                if ( dev == NULL )
+                {
+                        printk_err ("Cannot allocate net_device for port %d\n",
+                                        i );
+                        /* No point going any further */
+                        card->nports = i;
+                        break;
+                }
+                memset ( dev, 0, sizeof ( struct net_device ));
+                card->ports[i].dev = dev;
+
+                if ( dev_alloc_name ( dev, FST_NDEV_NAME "%d") < 0 )
+                {
+                        printk_err ("Cannot allocate i/f name for port %d\n",
+                                        i );
+                        kfree ( dev );
+                        card->nports = i;
+                        break;
+                }
+
+                /* Fill in remainder of the net device info */
+                                /* Since this is a PCI setup this is purely
+                                 * informational. Give them the buffer addresses
+                                 * and basic card I/O.
+                                 */
+                dev->mem_start   = card->phys_mem
+                                 + BUF_OFFSET ( txBuffer[i][0][0]);
+                dev->mem_end     = card->phys_mem
+                                 + BUF_OFFSET ( txBuffer[i][NUM_TX_BUFFER][0]);
+                dev->rmem_start  = card->phys_mem
+                                 + BUF_OFFSET ( rxBuffer[i][0][0]);
+                dev->rmem_end    = card->phys_mem
+                                 + BUF_OFFSET ( rxBuffer[i][NUM_RX_BUFFER][0]);
+                dev->base_addr   = card->pci_conf;
+                dev->irq         = card->irq;
+
+                dev->get_stats          = fst_get_stats;
+                dev->mtu                = FST_DEF_MTU;
+                dev->change_mtu         = fst_change_mtu;
+                dev->priv               = &card->ports[i];
+                dev->tx_queue_len       = FST_TX_QUEUE_LEN;
+                dev->type               = ARPHRD_MYTYPE;
+                dev->addr_len           = 0;
+                dev->open               = fst_open;
+                dev->stop               = fst_close;
+                dev->hard_start_xmit    = fst_start_xmit;
+                dev->do_ioctl           = fst_ioctl;
+                dev->watchdog_timeo     = FST_TX_TIMEOUT;
+                dev->tx_timeout         = fst_tx_timeout;
+                dev->flags              = IFF_POINTOPOINT|IFF_NOARP;
+
+                if (( err = register_netdev ( dev )) < 0 )
+                {
+                        printk_err ("Cannot register %s (errno %d)\n",
+                                        dev->name, -err );
+                        kfree ( dev );
+                        card->nports = i;
+                        break;
+                }
+        }
+
+        spin_lock_init ( &card->card_lock );
+
+        printk ( KERN_INFO "%s-%s: %s IRQ%d, %d ports\n",
+                        card->ports[0].dev->name,
+                        card->ports[card->nports-1].dev->name,
+                        type_strings[card->type], card->irq, card->nports );
+}
+
+
+/*
+ *      Initialise card when detected.
+ *      Returns 0 to indicate success, or errno otherwise.
+ */
+static int __devinit
+fst_add_one ( struct pci_dev *pdev, const struct pci_device_id *ent )
+{
+        static int firsttime_done = 0;
+        struct fst_card_info *card;
+        int err = 0;
+
+        if ( ! firsttime_done )
+        {
+                printk ( KERN_INFO "FarSync X21 driver " FST_USER_VERSION
+                                " (c) 2001 FarSite Communications Ltd.\n");
+                firsttime_done = 1;
+        }
+
+        /* Allocate driver private data */
+        card = kmalloc ( sizeof ( struct fst_card_info ),  GFP_KERNEL);
+        if (card == NULL)
+        {
+                printk_err ("FarSync card found but insufficient memory for"
+                                " driver storage\n");
+                return -ENOMEM;
+        }
+        memset ( card, 0, sizeof ( struct fst_card_info ));
+
+        /* Record info we need*/
+        card->irq         = pdev->irq;
+        card->pci_conf    = pci_resource_start ( pdev, 1 );
+        card->phys_mem    = pci_resource_start ( pdev, 2 );
+        card->phys_ctlmem = pci_resource_start ( pdev, 3 );
+
+        card->type        = ent->driver_data;
+        card->nports      = ( ent->driver_data == FST_TYPE_T2P ) ? 2 : 4;
+
+        card->state       = FST_UNINIT;
+
+        dbg ( DBG_PCI,"type %d nports %d irq %d\n", card->type,
+                        card->nports, card->irq );
+        dbg ( DBG_PCI,"conf %04x mem %08x ctlmem %08x\n",
+                        card->pci_conf, card->phys_mem, card->phys_ctlmem );
+
+        /* Check we can get access to the memory and I/O regions */
+        if ( ! request_region ( card->pci_conf, 0x80,"PLX config regs"))
+        {
+                printk_err ("Unable to get config I/O @ 0x%04X\n",
+                                                card->pci_conf );
+                err = -ENODEV;
+                goto error_1;
+        }
+        if ( ! request_mem_region ( card->phys_mem, FST_MEMSIZE,"Shared RAM"))
+        {
+                printk_err ("Unable to get main memory @ 0x%08X\n",
+                                                card->phys_mem );
+                err = -ENODEV;
+                goto error_2;
+        }
+        if ( ! request_mem_region ( card->phys_ctlmem, 0x10,"Control memory"))
+        {
+                printk_err ("Unable to get control memory @ 0x%08X\n",
+                                                card->phys_ctlmem );
+                err = -ENODEV;
+                goto error_3;
+        }
+
+        /* Try to enable the device */
+        if (( err = pci_enable_device ( pdev )) != 0 )
+        {
+                printk_err ("Failed to enable card. Err %d\n", -err );
+                goto error_4;
+        }
+
+        /* Get virtual addresses of memory regions */
+        card->mem = ioremap ( card->phys_mem, FST_MEMSIZE );
+        card->ctlmem = ioremap ( card->phys_ctlmem, 0x10 );
+        dbg ( DBG_PCI,"kernel mem %p, ctlmem %p\n", card->mem, card->ctlmem);
+
+        /* Reset the card's processor */
+        fst_cpureset ( card );
+        card->state = FST_RESET;
+
+        /* Register the interrupt handler */
+        if ( request_irq ( card->irq, fst_intr, SA_SHIRQ, FST_DEV_NAME, card ))
+        {
+
+                printk_err ("Unable to register interrupt %d\n", card->irq );
+                err = -ENODEV;
+                goto error_5;
+        }
+
+        /* Record driver data for later use */
+        pdev->driver_data = card;
+
+        /* Remainder of card setup */
+        fst_init_card ( card );
+
+        return 0;                       /* Success */
+
+
+                                        /* Failure. Release resources */
+error_5:
+        iounmap ( card->ctlmem );
+        iounmap ( card->mem );
+error_4:
+        release_mem_region ( card->phys_ctlmem, 0x10 );
+error_3:
+        release_mem_region ( card->phys_mem, FST_MEMSIZE );
+error_2:
+        release_region ( card->pci_conf, 0x80 );
+error_1:
+        kfree ( card );
+        return err;
+}
+
+
+/*
+ *      Cleanup and close down a card
+ */
+static void __devexit
+fst_remove_one ( struct pci_dev *pdev )
+{
+        struct fst_card_info *card;
+        int i;
+
+        card = pdev->driver_data;
+
+        for ( i = 0 ; i < card->nports ; i++ )
+        {
+                unregister_netdev ( card->ports[i].dev );
+                kfree ( card->ports[i].dev );
+        }
+
+        fst_disable_intr ( card );
+        free_irq ( card->irq, card );
+
+        iounmap ( card->ctlmem );
+        iounmap ( card->mem );
+
+        release_mem_region ( card->phys_ctlmem, 0x10 );
+        release_mem_region ( card->phys_mem, FST_MEMSIZE );
+        release_region ( card->pci_conf, 0x80 );
+
+        kfree ( card );
+}
+
+static struct pci_driver fst_driver = {
+        name:           FST_NAME,
+        id_table:       fst_pci_dev_id,
+        probe:          fst_add_one,
+        remove:         fst_remove_one,
+        suspend:        NULL,
+        resume:         NULL,
+};
+
+static int __init
+fst_init(void)
+{
+        return pci_module_init ( &fst_driver );
+}
+
+static void __exit
+fst_cleanup_module(void)
+{
+        pci_unregister_driver ( &fst_driver );
+}
+
+module_init ( fst_init );
+module_exit ( fst_cleanup_module );
+
--- /dev/null	Sat Mar 24 04:37:44 2001
+++ linux/drivers/net/wan/farsync.h	Tue Jul  3 14:14:29 2001
@@ -0,0 +1,226 @@
+/*
+ *      FarSync X21 driver for Linux
+ *
+ *      Actually sync driver for X.21, V.35 and V.24 on FarSync T-series cards
+ *
+ *      Copyright (C) 2001 FarSite Communications Ltd.
+ *      www.farsite.co.uk
+ *
+ *      This program is free software; you can redistribute it and/or
+ *      modify it under the terms of the GNU General Public License
+ *      as published by the Free Software Foundation; either version
+ *      2 of the License, or (at your option) any later version.
+ *
+ *      Author: R.J.Dunlop      <bob.dunlop@farsite.co.uk>
+ *
+ *      For the most part this file only contains structures and information
+ *      that is visible to applications outside the driver. Shared memory
+ *      layout etc is internal to the driver and described within farsync.c.
+ *      Overlap exists in that the values used for some fields within the
+ *      ioctl interface extend into the cards firmware interface so values in
+ *      this file may not be changed arbitrarily.
+ */
+
+/*      What's in a name
+ *
+ *      The project name for this driver is Oscar. The driver is intended to be
+ *      used with the FarSite T-Series cards (T2P & T4P) running in the high
+ *      speed frame shifter mode. This is sometimes referred to as X.21 mode
+ *      which is a complete misnomer as the card continues to support V.24 and
+ *      V.35 as well as X.21.
+ *
+ *      A short common prefix is useful for routines within the driver to avoid
+ *      conflict with other similar drivers and I chosen to use "fst_" for this
+ *      purpose (FarSite T-series).
+ *
+ *      Finally the device driver needs a short network interface name. Since
+ *      "hdlc" is already in use I've chosen the even less informative "sync"
+ *      for the present.
+ */
+#define FST_NAME                "fst"           /* In debug/info etc */
+#define FST_NDEV_NAME           "sync"          /* For net interface */
+#define FST_DEV_NAME            "farsync"       /* For misc interfaces */
+
+
+/*      User version number
+ *
+ *      This version number is incremented with each official release of the
+ *      package and is a simplified number for normal user reference.
+ *      Individual files are tracked by the version control system and may
+ *      have individual versions (or IDs) that move much faster than the
+ *      the release version as individual updates are tracked.
+ */
+#define FST_USER_VERSION        "0.09"
+
+
+/*      Ioctl call command values
+ *
+ *      The first three private ioctls are used by the sync-PPP module,
+ *      allowing a little room for expansion we start our numbering at 10.
+ */
+#define FSTWRITE        (SIOCDEVPRIVATE+10)
+#define FSTCPURESET     (SIOCDEVPRIVATE+11)
+#define FSTCPURELEASE   (SIOCDEVPRIVATE+12)
+#define FSTGETCONF      (SIOCDEVPRIVATE+13)
+#define FSTSETCONF      (SIOCDEVPRIVATE+14)
+
+
+/*      FSTWRITE
+ *
+ *      Used to write a block of data (firmware etc) before the card is running
+ */
+struct fstioc_write {
+        unsigned int  size;
+        unsigned int  offset;
+        unsigned char data[0];
+};
+
+
+/*      FSTCPURESET and FSTCPURELEASE
+ *
+ *      These take no additional data.
+ *      FSTCPURESET forces the cards CPU into a reset state and holds it there.
+ *      FSTCPURELEASE releases the CPU from this reset state allowing it to run,
+ *      the reset vector should be setup before this ioctl is run.
+ */
+
+/*      FSTGETCONF and FSTSETCONF
+ *
+ *      Get and set a card/ports configuration.
+ *      In order to allow selective setting of items and for the kernel to
+ *      indicate a partial status response the first field "valid" is a bitmask
+ *      indicating which other fields in the structure are valid.
+ *      Many of the field names in this structure match those used in the
+ *      firmware shared memory configuration interface and come originally from
+ *      the NT header file Smc.h
+ *
+ *      When used with FSTGETCONF this structure should be zeroed before use.
+ *      This is to allow for possible future expansion when some of the fields
+ *      might be used to indicate a different (expanded) structure.
+ */
+struct fstioc_info {
+        unsigned int   valid;           /* Bits of structure that are valid */
+        unsigned int   nports;          /* Number of serial ports */
+        unsigned int   type;            /* Type index of card */
+        unsigned int   state;           /* State of card */
+        unsigned int   index;           /* Index of port ioctl was issued on */
+        unsigned int   smcFirmwareVersion;
+        unsigned short lineInterface;   /* Physical interface type */
+        unsigned char  proto;           /* Line protocol */
+        unsigned char  internalClock;   /* 1 => internal clock, 0 => external */
+        unsigned int   lineSpeed;       /* Speed in bps */
+        unsigned int   v24IpSts;        /* V.24 control input status */
+        unsigned int   v24OpSts;        /* V.24 control output status */
+        unsigned short clockStatus;     /* lsb: 0=> present, 1=> absent */
+        unsigned short cableStatus;     /* lsb: 0=> present, 1=> absent */
+        unsigned short cardMode;        /* lsb: LED id mode */
+        unsigned short debug;           /* Debug flags */
+};
+
+/* "valid" bitmask */
+#define FSTVAL_NONE     0x00000000      /* Nothing valid (firmware not running).
+                                         * Slight misnomer. In fact nports,
+                                         * type, state and index will be set
+                                         * based on hardware detected.
+                                         */
+#define FSTVAL_OMODEM   0x0000001F      /* First 5 bits correspond to the
+                                         * output status bits defined for
+                                         * v24OpSts
+                                         */
+#define FSTVAL_SPEED    0x00000020      /* internalClock, lineSpeed, clockStatus
+                                         */
+#define FSTVAL_CABLE    0x00000040      /* lineInterface, cableStatus */
+#define FSTVAL_IMODEM   0x00000080      /* v24IpSts */
+#define FSTVAL_CARD     0x00000100      /* nports, type, state, index,
+                                         * smcFirmwareVersion
+                                         */
+#define FSTVAL_PROTO    0x00000200      /* proto */
+#define FSTVAL_MODE     0x00000400      /* cardMode */
+#define FSTVAL_DEBUG    0x80000000      /* debug */
+#define FSTVAL_ALL      0x000007FF      /* Note: does not include DEBUG flag */
+
+/* "type" */
+#define FST_TYPE_NONE   0               /* Probably should never happen */
+#define FST_TYPE_T2P    1               /* T2P X21 2 port card */
+#define FST_TYPE_T4P    2               /* T4P X21 4 port card */
+
+/* "state" */
+#define FST_UNINIT      0               /* Raw uninitialised state following
+                                         * system startup */
+#define FST_RESET       1               /* Processor held in reset state */
+#define FST_DOWNLOAD    2               /* Card being downloaded */
+#define FST_STARTING    3               /* Released following download */
+#define FST_RUNNING     4               /* Processor running */
+#define FST_BADVERSION  5               /* Bad shared memory version detected */
+#define FST_HALTED      6               /* Processor flagged a halt */
+#define FST_IFAILED     7               /* Firmware issued initialisation failed
+                                         * interrupt
+                                         */
+/* "lineInterface" */
+#define V24             1
+#define X21             2
+#define V35             3
+
+/* "proto" */
+#define FST_HDLC        1               /* Cisco compatible HDLC */
+#define FST_PPP         2               /* Sync PPP */
+#define FST_MONITOR     3               /* Monitor only (raw packet reception) */
+#define FST_RAW         4               /* Two way raw packets */
+#define FST_GEN_HDLC    5               /* Using "Generic HDLC" module */
+
+/* "internalClock" */
+#define INTCLK          1
+#define EXTCLK          0
+
+/* "v24IpSts" bitmask */
+#define IPSTS_CTS       0x00000001      /* Clear To Send (Indicate for X.21) */
+#define IPSTS_INDICATE  IPSTS_CTS
+#define IPSTS_DSR       0x00000002      /* Data Set Ready (T2P Port A) */
+#define IPSTS_DCD       0x00000004      /* Data Carrier Detect */
+#define IPSTS_RI        0x00000008      /* Ring Indicator (T2P Port A) */
+#define IPSTS_TMI       0x00000010      /* Test Mode Indicator (Not Supported)*/
+
+/* "v24OpSts" bitmask */
+#define OPSTS_RTS       0x00000001      /* Request To Send (Control for X.21) */
+#define OPSTS_CONTROL   OPSTS_RTS
+#define OPSTS_DTR       0x00000002      /* Data Terminal Ready */
+#define OPSTS_DSRS      0x00000004      /* Data Signalling Rate Select (Not
+                                         * Supported) */
+#define OPSTS_SS        0x00000008      /* Select Standby (Not Supported) */
+#define OPSTS_LL        0x00000010      /* Maintenance Test (Not Supported) */
+
+/* "cardMode" bitmask */
+#define CARD_MODE_IDENTIFY      0x0001
+
+
+/*      Debug support
+ *
+ *      These should only be enabled for development kernels, production code
+ *      should define FST_DEBUG=0 in order to exclude the code.
+ *      Setting FST_DEBUG=1 will include all the debug code but in a disabled
+ *      state, use the FSTSETCONF ioctl to enable specific debug actions, or
+ *      FST_DEBUG can be set to prime the debug selection.
+ */
+#define FST_DEBUG       0x0000
+#if FST_DEBUG
+
+extern int fst_debug_mask;              /* Bit mask of actions to debug, bits
+                                         * listed below. Note: Bit 0 is used
+                                         * to trigger the inclusion of this
+                                         * code, without enabling any actions.
+                                         */
+#define DBG_INIT        0x0002          /* Card detection and initialisation */
+#define DBG_OPEN        0x0004          /* Open and close sequences */
+#define DBG_PCI         0x0008          /* PCI config operations */
+#define DBG_IOCTL       0x0010          /* Ioctls and other config */
+#define DBG_INTR        0x0020          /* Interrupt routines (be careful) */
+#define DBG_TX          0x0040          /* Packet transmission */
+#define DBG_RX          0x0080          /* Packet reception */
+#define DBG_CMD         0x0100          /* Port command issuing */
+
+#define DBG_ASS         0xFFFF          /* Assert like statements. Code that
+                                         * should never be reached, if you see
+                                         * one of these then I've been an ass
+                                         */
+#endif  /* FST_DEBUG */
+


-- 
        Bob Dunlop                      FarSite Communications
        rjd@xyzzy.clara.co.uk           bob.dunlop@farsite.co.uk
        www.xyzzy.clara.co.uk           www.farsite.co.uk

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: New FarSync T-Series driver
  2001-07-03 17:28 New FarSync T-Series driver Robert J.Dunlop
@ 2001-07-04 14:18 ` François romieu
  2001-07-09 15:19   ` Robert J.Dunlop
  0 siblings, 1 reply; 5+ messages in thread
From: François romieu @ 2001-07-04 14:18 UTC (permalink / raw)
  To: Robert J.Dunlop; +Cc: linux-kernel

The Tue, Jul 03, 2001 at 06:28:03PM +0100, Robert J.Dunlop wrote :
[...]
> Sorry for the big post, but I posted URLs for an earlier version of this
> a couple of months back and got very little feedback.  I know sync card
> drivers ain't sexy.

Just my HO:
* error_1, error_2... error_n labels are ugly;
* ioremap may fail;
* mix of spin_lock and FST_LOCK isn't nice (kill the latter ?);
* 
+                offset = BUF_OFFSET ( rxBuffer[pi][i]);
[...]
+                                card->mem + BUF_OFFSET ( rxBuffer[pi][rxp][0]),

A bit of a macro abuse imho.

*
+        if ( ++port->txpos >= NUM_TX_BUFFER )
+                port->txpos = 0;

Why not:
port->txpos++;
foo = port->txpos%NUM_TX_BUFFER;

--
Ueimor

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: New FarSync T-Series driver
  2001-07-04 14:18 ` François romieu
@ 2001-07-09 15:19   ` Robert J.Dunlop
  2001-07-26 14:51     ` FarSync T-Series tweak and question Robert J.Dunlop
  0 siblings, 1 reply; 5+ messages in thread
From: Robert J.Dunlop @ 2001-07-09 15:19 UTC (permalink / raw)
  To: François romieu; +Cc: linux-kernel, Alan Cox


Thanks for the freedback guys.

Revised versions of the driver patches against 2.4.6-ac2 and 2.4.7-pre3
are available at:
http://www.xyzzy.clara.co.uk/farsync/farsync-patch-2.4.6ac2.gz and
http://www.xyzzy.clara.co.uk/farsync/farsync-patch-2.4.7pre3.gz



On Wed, Jul 04, 2001 at 04:18:45PM +0200, François romieu wrote:
> Just my HO:
Hey it's all good feedback.


> * error_1, error_2... error_n labels are ugly;
I don't like goto's either but in this function the structure kinda demands
it. Lot's of points where you fail and need to release an increasing list of
resources. Not sure a list of labels like this is any better.
    err_out_free_iqrx:
    err_out_free_iqtx:
    err_out_free_irq:
    err_out_iounmap:
    err_out_free_mmio_region:
    err_out_free_mmio_region0:
    err_out:

I've left this one as it is for the present. I have removed the gotos from
fst_open() however.


> * ioremap may fail;
Fixed that one, although it did add a "goto error_6" :-(


> * mix of spin_lock and FST_LOCK isn't nice (kill the latter ?);
Quite correct. Un-helpful information hiding. I've unwound the macros in
place and removed them.


> +                offset = BUF_OFFSET ( rxBuffer[pi][i]);
> [...]
> +                                card->mem + BUF_OFFSET ( rxBuffer[pi][rxp][0]),
> A bit of a macro abuse imho.
Not quite sure what you mean here. I have corrected the code so that BUF_OFFSET
sees a consistent argument.


> +        if ( ++port->txpos >= NUM_TX_BUFFER )
> +                port->txpos = 0;
> 
> Why not:
> port->txpos++;
> foo = port->txpos%NUM_TX_BUFFER;
I think mine is clearer but then I've always bumped and wrapped pointers and
indexs that way. Another alternative would be:
    port->txpos = ( port->txpos + 1 ) % NUM_TX_BUFFER;

Looks cleaner than both and lets the compiler decide, but if the % operator
produces a div operation instead of a mask on even one processor type then I
think we lose. If anyone can prove there's an optimal way to do this sort of
bump and wrap operation I'll gladly adopt it.


And finally one from Alan Cox:
> Don't assume short is 16bits. It is so far but that might change
> Use u8/u16/u32 (or for user exposed structs __u16/__u32 etc)
Oh God! how did I miss that one!
And after I'd taken the NT driver writers to task over the same issue :(
Still corrected now.

-- 
        Bob Dunlop                      FarSite Communications
        rjd@xyzzy.clara.co.uk           bob.dunlop@farsite.co.uk
        www.xyzzy.clara.co.uk           www.farsite.co.uk

^ permalink raw reply	[flat|nested] 5+ messages in thread

* FarSync T-Series tweak and question
  2001-07-09 15:19   ` Robert J.Dunlop
@ 2001-07-26 14:51     ` Robert J.Dunlop
  2001-07-27 16:11       ` François romieu
  0 siblings, 1 reply; 5+ messages in thread
From: Robert J.Dunlop @ 2001-07-26 14:51 UTC (permalink / raw)
  To: François romieu, Alan Cox; +Cc: linux-kernel

Hi,

Having thought about it a while I've decided that I agree with
François romieu's comment.

> * error_1, error_2... error_n labels are ugly;

Should have done it earlier but I've now cleaned up my code.
Alan could you please apply the patch at the end of this mail.


Second I have a question.  We had a bit of debate about how best to code
the bumping and wrapping of an index into a circular buffer.

> > +        if ( ++port->txpos >= NUM_TX_BUFFER )
> > +                port->txpos = 0;
> > 
> > Why not:
> > port->txpos++;
> > foo = port->txpos%NUM_TX_BUFFER;
> I think mine is clearer but then I've always bumped and wrapped pointers and
> indexs that way. Another alternative would be:
>     port->txpos = ( port->txpos + 1 ) % NUM_TX_BUFFER;

Having taken the time that the % operation wouldn't be detrimental on other
processor architectures and having checked with a small test that the alternate
form produced tighter code on Intel I applied the change to my driver.  The
code that had shrunk in a small test got bigger (and slower?) in the driver :-(
I think what happened is that register usage changed and something else had
to be fetched from the stack.  I've decided for the moment not to adopt this
change on the grounds that if it ain't broke don't patch it.

Now the question. Is there a sensible way of coding this so that as the
compilers get better we give them the maximum number of clues as to what we
are actually trying to do?



--- linux/drivers/net/wan/farsync.c.orig	Wed Jul 25 07:42:28 2001
+++ linux/drivers/net/wan/farsync.c	Wed Jul 25 09:24:01 2001
@@ -1689,28 +1689,28 @@
                 printk_err ("Unable to get config I/O @ 0x%04X\n",
                                                 card->pci_conf );
                 err = -ENODEV;
-                goto error_1;
+                goto error_free_card;
         }
         if ( ! request_mem_region ( card->phys_mem, FST_MEMSIZE,"Shared RAM"))
         {
                 printk_err ("Unable to get main memory @ 0x%08X\n",
                                                 card->phys_mem );
                 err = -ENODEV;
-                goto error_2;
+                goto error_release_io;
         }
         if ( ! request_mem_region ( card->phys_ctlmem, 0x10,"Control memory"))
         {
                 printk_err ("Unable to get control memory @ 0x%08X\n",
                                                 card->phys_ctlmem );
                 err = -ENODEV;
-                goto error_3;
+                goto error_release_mem;
         }
 
         /* Try to enable the device */
         if (( err = pci_enable_device ( pdev )) != 0 )
         {
                 printk_err ("Failed to enable card. Err %d\n", -err );
-                goto error_4;
+                goto error_release_ctlmem;
         }
 
         /* Get virtual addresses of memory regions */
@@ -1718,13 +1718,13 @@
         {
                 printk_err ("Physical memory remap failed\n");
                 err = -ENODEV;
-                goto error_4;
+                goto error_release_ctlmem;
         }
         if (( card->ctlmem = ioremap ( card->phys_ctlmem, 0x10 )) == NULL )
         {
                 printk_err ("Control memory remap failed\n");
                 err = -ENODEV;
-                goto error_5;
+                goto error_unmap_mem;
         }
         dbg ( DBG_PCI,"kernel mem %p, ctlmem %p\n", card->mem, card->ctlmem);
 
@@ -1738,7 +1738,7 @@
 
                 printk_err ("Unable to register interrupt %d\n", card->irq );
                 err = -ENODEV;
-                goto error_6;
+                goto error_unmap_ctlmem;
         }
 
         /* Record driver data for later use */
@@ -1751,17 +1751,22 @@
 
 
                                         /* Failure. Release resources */
-error_6:
+error_unmap_ctlmem:
         iounmap ( card->ctlmem );
-error_5:
+
+error_unmap_mem:
         iounmap ( card->mem );
-error_4:
+
+error_release_ctlmem:
         release_mem_region ( card->phys_ctlmem, 0x10 );
-error_3:
+
+error_release_mem:
         release_mem_region ( card->phys_mem, FST_MEMSIZE );
-error_2:
+
+error_release_io:
         release_region ( card->pci_conf, 0x80 );
-error_1:
+
+error_free_card:
         kfree ( card );
         return err;
 }

-- 
        Bob Dunlop                      FarSite Communications
        rjd@xyzzy.clara.co.uk           bob.dunlop@farsite.co.uk
        www.xyzzy.clara.co.uk           www.farsite.co.uk

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: FarSync T-Series tweak and question
  2001-07-26 14:51     ` FarSync T-Series tweak and question Robert J.Dunlop
@ 2001-07-27 16:11       ` François romieu
  0 siblings, 0 replies; 5+ messages in thread
From: François romieu @ 2001-07-27 16:11 UTC (permalink / raw)
  To: Robert J.Dunlop; +Cc: linux-kernel

The Thu, Jul 26, 2001 at 03:51:23PM +0100, Robert J.Dunlop wrote :
[...]
> > > +        if ( ++port->txpos >= NUM_TX_BUFFER )
> > > +                port->txpos = 0;
[...]
> > I think mine is clearer but then I've always bumped and wrapped pointers and
> > indexs that way. Another alternative would be:
> >     port->txpos = ( port->txpos + 1 ) % NUM_TX_BUFFER;
> 
> Having taken the time that the % operation wouldn't be detrimental on other
> processor architectures and having checked with a small test that the
> alternate form produced tighter code on Intel I applied the change to my 
> driver.  The code that had shrunk in a small test got bigger (and slower?) 
> in the driver :-(

Have you the .o at hand ? I'd be curious to objdump them.
Btw, I don't believe either expression impacts the perfs. During init,
it's not time critical and while processing data, it looks like the driver 
does himself the dma with the adapter.

--
Ueimor

^ permalink raw reply	[flat|nested] 5+ messages in thread

end of thread, other threads:[~2001-07-27 16:13 UTC | newest]

Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2001-07-03 17:28 New FarSync T-Series driver Robert J.Dunlop
2001-07-04 14:18 ` François romieu
2001-07-09 15:19   ` Robert J.Dunlop
2001-07-26 14:51     ` FarSync T-Series tweak and question Robert J.Dunlop
2001-07-27 16:11       ` François romieu

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).