Linux Audio

Check our new training course

Loading...
Note: File does not exist in v3.1.
   1/* hamachi.c: A Packet Engines GNIC-II Gigabit Ethernet driver for Linux. */
   2/*
   3	Written 1998-2000 by Donald Becker.
   4	Updates 2000 by Keith Underwood.
   5
   6	This software may be used and distributed according to the terms of
   7	the GNU General Public License (GPL), incorporated herein by reference.
   8	Drivers based on or derived from this code fall under the GPL and must
   9	retain the authorship, copyright and license notice.  This file is not
  10	a complete program and may only be used when the entire operating
  11	system is licensed under the GPL.
  12
  13	The author may be reached as becker@scyld.com, or C/O
  14	Scyld Computing Corporation
  15	410 Severn Ave., Suite 210
  16	Annapolis MD 21403
  17
  18	This driver is for the Packet Engines GNIC-II PCI Gigabit Ethernet
  19	adapter.
  20
  21	Support and updates available at
  22	http://www.scyld.com/network/hamachi.html
  23	[link no longer provides useful info -jgarzik]
  24	or
  25	http://www.parl.clemson.edu/~keithu/hamachi.html
  26
  27*/
  28
  29#define DRV_NAME	"hamachi"
  30#define DRV_VERSION	"2.1"
  31#define DRV_RELDATE	"Sept 11, 2006"
  32
  33
  34/* A few user-configurable values. */
  35
  36static int debug = 1;		/* 1 normal messages, 0 quiet .. 7 verbose.  */
  37#define final_version
  38#define hamachi_debug debug
  39/* Maximum events (Rx packets, etc.) to handle at each interrupt. */
  40static int max_interrupt_work = 40;
  41static int mtu;
  42/* Default values selected by testing on a dual processor PIII-450 */
  43/* These six interrupt control parameters may be set directly when loading the
  44 * module, or through the rx_params and tx_params variables
  45 */
  46static int max_rx_latency = 0x11;
  47static int max_rx_gap = 0x05;
  48static int min_rx_pkt = 0x18;
  49static int max_tx_latency = 0x00;
  50static int max_tx_gap = 0x00;
  51static int min_tx_pkt = 0x30;
  52
  53/* Set the copy breakpoint for the copy-only-tiny-frames scheme.
  54   -Setting to > 1518 causes all frames to be copied
  55	-Setting to 0 disables copies
  56*/
  57static int rx_copybreak;
  58
  59/* An override for the hardware detection of bus width.
  60	Set to 1 to force 32 bit PCI bus detection.  Set to 4 to force 64 bit.
  61	Add 2 to disable parity detection.
  62*/
  63static int force32;
  64
  65
  66/* Used to pass the media type, etc.
  67   These exist for driver interoperability.
  68   No media types are currently defined.
  69		- The lower 4 bits are reserved for the media type.
  70		- The next three bits may be set to one of the following:
  71			0x00000000 : Autodetect PCI bus
  72			0x00000010 : Force 32 bit PCI bus
  73			0x00000020 : Disable parity detection
  74			0x00000040 : Force 64 bit PCI bus
  75			Default is autodetect
  76		- The next bit can be used to force half-duplex.  This is a bad
  77		  idea since no known implementations implement half-duplex, and,
  78		  in general, half-duplex for gigabit ethernet is a bad idea.
  79			0x00000080 : Force half-duplex
  80			Default is full-duplex.
  81		- In the original driver, the ninth bit could be used to force
  82		  full-duplex.  Maintain that for compatibility
  83		   0x00000200 : Force full-duplex
  84*/
  85#define MAX_UNITS 8				/* More are supported, limit only on options */
  86static int options[MAX_UNITS] = {-1, -1, -1, -1, -1, -1, -1, -1};
  87static int full_duplex[MAX_UNITS] = {-1, -1, -1, -1, -1, -1, -1, -1};
  88/* The Hamachi chipset supports 3 parameters each for Rx and Tx
  89 * interruput management.  Parameters will be loaded as specified into
  90 * the TxIntControl and RxIntControl registers.
  91 *
  92 * The registers are arranged as follows:
  93 *     23 - 16   15 -  8   7    -    0
  94 *    _________________________________
  95 *   | min_pkt | max_gap | max_latency |
  96 *    ---------------------------------
  97 *   min_pkt      : The minimum number of packets processed between
  98 *                  interrupts.
  99 *   max_gap      : The maximum inter-packet gap in units of 8.192 us
 100 *   max_latency  : The absolute time between interrupts in units of 8.192 us
 101 *
 102 */
 103static int rx_params[MAX_UNITS] = {-1, -1, -1, -1, -1, -1, -1, -1};
 104static int tx_params[MAX_UNITS] = {-1, -1, -1, -1, -1, -1, -1, -1};
 105
 106/* Operational parameters that are set at compile time. */
 107
 108/* Keep the ring sizes a power of two for compile efficiency.
 109	The compiler will convert <unsigned>'%'<2^N> into a bit mask.
 110   Making the Tx ring too large decreases the effectiveness of channel
 111   bonding and packet priority.
 112   There are no ill effects from too-large receive rings, except for
 113	excessive memory usage */
 114/* Empirically it appears that the Tx ring needs to be a little bigger
 115   for these Gbit adapters or you get into an overrun condition really
 116   easily.  Also, things appear to work a bit better in back-to-back
 117   configurations if the Rx ring is 8 times the size of the Tx ring
 118*/
 119#define TX_RING_SIZE	64
 120#define RX_RING_SIZE	512
 121#define TX_TOTAL_SIZE	TX_RING_SIZE*sizeof(struct hamachi_desc)
 122#define RX_TOTAL_SIZE	RX_RING_SIZE*sizeof(struct hamachi_desc)
 123
 124/*
 125 * Enable netdev_ioctl.  Added interrupt coalescing parameter adjustment.
 126 * 2/19/99 Pete Wyckoff <wyckoff@ca.sandia.gov>
 127 */
 128
 129/* play with 64-bit addrlen; seems to be a teensy bit slower  --pw */
 130/* #define ADDRLEN 64 */
 131
 132/*
 133 * RX_CHECKSUM turns on card-generated receive checksum generation for
 134 *   TCP and UDP packets.  Otherwise the upper layers do the calculation.
 135 * 3/10/1999 Pete Wyckoff <wyckoff@ca.sandia.gov>
 136 */
 137#define RX_CHECKSUM
 138
 139/* Operational parameters that usually are not changed. */
 140/* Time in jiffies before concluding the transmitter is hung. */
 141#define TX_TIMEOUT  (5*HZ)
 142
 143#include <linux/capability.h>
 144#include <linux/module.h>
 145#include <linux/kernel.h>
 146#include <linux/string.h>
 147#include <linux/timer.h>
 148#include <linux/time.h>
 149#include <linux/errno.h>
 150#include <linux/ioport.h>
 151#include <linux/interrupt.h>
 152#include <linux/pci.h>
 153#include <linux/init.h>
 154#include <linux/ethtool.h>
 155#include <linux/mii.h>
 156#include <linux/netdevice.h>
 157#include <linux/etherdevice.h>
 158#include <linux/skbuff.h>
 159#include <linux/ip.h>
 160#include <linux/delay.h>
 161#include <linux/bitops.h>
 162
 163#include <linux/uaccess.h>
 164#include <asm/processor.h>	/* Processor type for cache alignment. */
 165#include <asm/io.h>
 166#include <asm/unaligned.h>
 167#include <asm/cache.h>
 168
 169static const char version[] =
 170KERN_INFO DRV_NAME ".c:v" DRV_VERSION " " DRV_RELDATE "  Written by Donald Becker\n"
 171"   Some modifications by Eric kasten <kasten@nscl.msu.edu>\n"
 172"   Further modifications by Keith Underwood <keithu@parl.clemson.edu>\n";
 173
 174
 175/* IP_MF appears to be only defined in <netinet/ip.h>, however,
 176   we need it for hardware checksumming support.  FYI... some of
 177   the definitions in <netinet/ip.h> conflict/duplicate those in
 178   other linux headers causing many compiler warnings.
 179*/
 180#ifndef IP_MF
 181  #define IP_MF 0x2000   /* IP more frags from <netinet/ip.h> */
 182#endif
 183
 184/* Define IP_OFFSET to be IPOPT_OFFSET */
 185#ifndef IP_OFFSET
 186  #ifdef IPOPT_OFFSET
 187    #define IP_OFFSET IPOPT_OFFSET
 188  #else
 189    #define IP_OFFSET 2
 190  #endif
 191#endif
 192
 193#define RUN_AT(x) (jiffies + (x))
 194
 195#ifndef ADDRLEN
 196#define ADDRLEN 32
 197#endif
 198
 199/* Condensed bus+endian portability operations. */
 200#if ADDRLEN == 64
 201#define cpu_to_leXX(addr)	cpu_to_le64(addr)
 202#define leXX_to_cpu(addr)	le64_to_cpu(addr)
 203#else
 204#define cpu_to_leXX(addr)	cpu_to_le32(addr)
 205#define leXX_to_cpu(addr)	le32_to_cpu(addr)
 206#endif
 207
 208
 209/*
 210				Theory of Operation
 211
 212I. Board Compatibility
 213
 214This device driver is designed for the Packet Engines "Hamachi"
 215Gigabit Ethernet chip.  The only PCA currently supported is the GNIC-II 64-bit
 21666Mhz PCI card.
 217
 218II. Board-specific settings
 219
 220No jumpers exist on the board.  The chip supports software correction of
 221various motherboard wiring errors, however this driver does not support
 222that feature.
 223
 224III. Driver operation
 225
 226IIIa. Ring buffers
 227
 228The Hamachi uses a typical descriptor based bus-master architecture.
 229The descriptor list is similar to that used by the Digital Tulip.
 230This driver uses two statically allocated fixed-size descriptor lists
 231formed into rings by a branch from the final descriptor to the beginning of
 232the list.  The ring sizes are set at compile time by RX/TX_RING_SIZE.
 233
 234This driver uses a zero-copy receive and transmit scheme similar my other
 235network drivers.
 236The driver allocates full frame size skbuffs for the Rx ring buffers at
 237open() time and passes the skb->data field to the Hamachi as receive data
 238buffers.  When an incoming frame is less than RX_COPYBREAK bytes long,
 239a fresh skbuff is allocated and the frame is copied to the new skbuff.
 240When the incoming frame is larger, the skbuff is passed directly up the
 241protocol stack and replaced by a newly allocated skbuff.
 242
 243The RX_COPYBREAK value is chosen to trade-off the memory wasted by
 244using a full-sized skbuff for small frames vs. the copying costs of larger
 245frames.  Gigabit cards are typically used on generously configured machines
 246and the underfilled buffers have negligible impact compared to the benefit of
 247a single allocation size, so the default value of zero results in never
 248copying packets.
 249
 250IIIb/c. Transmit/Receive Structure
 251
 252The Rx and Tx descriptor structure are straight-forward, with no historical
 253baggage that must be explained.  Unlike the awkward DBDMA structure, there
 254are no unused fields or option bits that had only one allowable setting.
 255
 256Two details should be noted about the descriptors: The chip supports both 32
 257bit and 64 bit address structures, and the length field is overwritten on
 258the receive descriptors.  The descriptor length is set in the control word
 259for each channel. The development driver uses 32 bit addresses only, however
 26064 bit addresses may be enabled for 64 bit architectures e.g. the Alpha.
 261
 262IIId. Synchronization
 263
 264This driver is very similar to my other network drivers.
 265The driver runs as two independent, single-threaded flows of control.  One
 266is the send-packet routine, which enforces single-threaded use by the
 267dev->tbusy flag.  The other thread is the interrupt handler, which is single
 268threaded by the hardware and other software.
 269
 270The send packet thread has partial control over the Tx ring and 'dev->tbusy'
 271flag.  It sets the tbusy flag whenever it's queuing a Tx packet. If the next
 272queue slot is empty, it clears the tbusy flag when finished otherwise it sets
 273the 'hmp->tx_full' flag.
 274
 275The interrupt handler has exclusive control over the Rx ring and records stats
 276from the Tx ring.  After reaping the stats, it marks the Tx queue entry as
 277empty by incrementing the dirty_tx mark. Iff the 'hmp->tx_full' flag is set, it
 278clears both the tx_full and tbusy flags.
 279
 280IV. Notes
 281
 282Thanks to Kim Stearns of Packet Engines for providing a pair of GNIC-II boards.
 283
 284IVb. References
 285
 286Hamachi Engineering Design Specification, 5/15/97
 287(Note: This version was marked "Confidential".)
 288
 289IVc. Errata
 290
 291None noted.
 292
 293V.  Recent Changes
 294
 29501/15/1999 EPK  Enlargement of the TX and RX ring sizes.  This appears
 296    to help avoid some stall conditions -- this needs further research.
 297
 29801/15/1999 EPK  Creation of the hamachi_tx function.  This function cleans
 299    the Tx ring and is called from hamachi_start_xmit (this used to be
 300    called from hamachi_interrupt but it tends to delay execution of the
 301    interrupt handler and thus reduce bandwidth by reducing the latency
 302    between hamachi_rx()'s).  Notably, some modification has been made so
 303    that the cleaning loop checks only to make sure that the DescOwn bit
 304    isn't set in the status flag since the card is not required
 305    to set the entire flag to zero after processing.
 306
 30701/15/1999 EPK In the hamachi_start_tx function, the Tx ring full flag is
 308    checked before attempting to add a buffer to the ring.  If the ring is full
 309    an attempt is made to free any dirty buffers and thus find space for
 310    the new buffer or the function returns non-zero which should case the
 311    scheduler to reschedule the buffer later.
 312
 31301/15/1999 EPK Some adjustments were made to the chip initialization.
 314    End-to-end flow control should now be fully active and the interrupt
 315    algorithm vars have been changed.  These could probably use further tuning.
 316
 31701/15/1999 EPK Added the max_{rx,tx}_latency options.  These are used to
 318    set the rx and tx latencies for the Hamachi interrupts. If you're having
 319    problems with network stalls, try setting these to higher values.
 320    Valid values are 0x00 through 0xff.
 321
 32201/15/1999 EPK In general, the overall bandwidth has increased and
 323    latencies are better (sometimes by a factor of 2).  Stalls are rare at
 324    this point, however there still appears to be a bug somewhere between the
 325    hardware and driver.  TCP checksum errors under load also appear to be
 326    eliminated at this point.
 327
 32801/18/1999 EPK Ensured that the DescEndRing bit was being set on both the
 329    Rx and Tx rings.  This appears to have been affecting whether a particular
 330    peer-to-peer connection would hang under high load.  I believe the Rx
 331    rings was typically getting set correctly, but the Tx ring wasn't getting
 332    the DescEndRing bit set during initialization. ??? Does this mean the
 333    hamachi card is using the DescEndRing in processing even if a particular
 334    slot isn't in use -- hypothetically, the card might be searching the
 335    entire Tx ring for slots with the DescOwn bit set and then processing
 336    them.  If the DescEndRing bit isn't set, then it might just wander off
 337    through memory until it hits a chunk of data with that bit set
 338    and then looping back.
 339
 34002/09/1999 EPK Added Michel Mueller's TxDMA Interrupt and Tx-timeout
 341    problem (TxCmd and RxCmd need only to be set when idle or stopped.
 342
 34302/09/1999 EPK Added code to check/reset dev->tbusy in hamachi_interrupt.
 344    (Michel Mueller pointed out the ``permanently busy'' potential
 345    problem here).
 346
 34702/22/1999 EPK Added Pete Wyckoff's ioctl to control the Tx/Rx latencies.
 348
 34902/23/1999 EPK Verified that the interrupt status field bits for Tx were
 350    incorrectly defined and corrected (as per Michel Mueller).
 351
 35202/23/1999 EPK Corrected the Tx full check to check that at least 4 slots
 353    were available before resetting the tbusy and tx_full flags
 354    (as per Michel Mueller).
 355
 35603/11/1999 EPK Added Pete Wyckoff's hardware checksumming support.
 357
 35812/31/1999 KDU Cleaned up assorted things and added Don's code to force
 35932 bit.
 360
 36102/20/2000 KDU Some of the control was just plain odd.  Cleaned up the
 362hamachi_start_xmit() and hamachi_interrupt() code.  There is still some
 363re-structuring I would like to do.
 364
 36503/01/2000 KDU Experimenting with a WIDE range of interrupt mitigation
 366parameters on a dual P3-450 setup yielded the new default interrupt
 367mitigation parameters.  Tx should interrupt VERY infrequently due to
 368Eric's scheme.  Rx should be more often...
 369
 37003/13/2000 KDU Added a patch to make the Rx Checksum code interact
 371nicely with non-linux machines.
 372
 37303/13/2000 KDU Experimented with some of the configuration values:
 374
 375	-It seems that enabling PCI performance commands for descriptors
 376	(changing RxDMACtrl and TxDMACtrl lower nibble from 5 to D) has minimal
 377	performance impact for any of my tests. (ttcp, netpipe, netperf)  I will
 378	leave them that way until I hear further feedback.
 379
 380	-Increasing the PCI_LATENCY_TIMER to 130
 381	(2 + (burst size of 128 * (0 wait states + 1))) seems to slightly
 382	degrade performance.  Leaving default at 64 pending further information.
 383
 38403/14/2000 KDU Further tuning:
 385
 386	-adjusted boguscnt in hamachi_rx() to depend on interrupt
 387	mitigation parameters chosen.
 388
 389	-Selected a set of interrupt parameters based on some extensive testing.
 390	These may change with more testing.
 391
 392TO DO:
 393
 394-Consider borrowing from the acenic driver code to check PCI_COMMAND for
 395PCI_COMMAND_INVALIDATE.  Set maximum burst size to cache line size in
 396that case.
 397
 398-fix the reset procedure.  It doesn't quite work.
 399*/
 400
 401/* A few values that may be tweaked. */
 402/* Size of each temporary Rx buffer, calculated as:
 403 * 1518 bytes (ethernet packet) + 2 bytes (to get 8 byte alignment for
 404 * the card) + 8 bytes of status info + 8 bytes for the Rx Checksum
 405 */
 406#define PKT_BUF_SZ		1536
 407
 408/* For now, this is going to be set to the maximum size of an ethernet
 409 * packet.  Eventually, we may want to make it a variable that is
 410 * related to the MTU
 411 */
 412#define MAX_FRAME_SIZE  1518
 413
 414/* The rest of these values should never change. */
 415
 416static void hamachi_timer(unsigned long data);
 417
 418enum capability_flags {CanHaveMII=1, };
 419static const struct chip_info {
 420	u16	vendor_id, device_id, device_id_mask, pad;
 421	const char *name;
 422	void (*media_timer)(unsigned long data);
 423	int flags;
 424} chip_tbl[] = {
 425	{0x1318, 0x0911, 0xffff, 0, "Hamachi GNIC-II", hamachi_timer, 0},
 426	{0,},
 427};
 428
 429/* Offsets to the Hamachi registers.  Various sizes. */
 430enum hamachi_offsets {
 431	TxDMACtrl=0x00, TxCmd=0x04, TxStatus=0x06, TxPtr=0x08, TxCurPtr=0x10,
 432	RxDMACtrl=0x20, RxCmd=0x24, RxStatus=0x26, RxPtr=0x28, RxCurPtr=0x30,
 433	PCIClkMeas=0x060, MiscStatus=0x066, ChipRev=0x68, ChipReset=0x06B,
 434	LEDCtrl=0x06C, VirtualJumpers=0x06D, GPIO=0x6E,
 435	TxChecksum=0x074, RxChecksum=0x076,
 436	TxIntrCtrl=0x078, RxIntrCtrl=0x07C,
 437	InterruptEnable=0x080, InterruptClear=0x084, IntrStatus=0x088,
 438	EventStatus=0x08C,
 439	MACCnfg=0x0A0, FrameGap0=0x0A2, FrameGap1=0x0A4,
 440	/* See enum MII_offsets below. */
 441	MACCnfg2=0x0B0, RxDepth=0x0B8, FlowCtrl=0x0BC, MaxFrameSize=0x0CE,
 442	AddrMode=0x0D0, StationAddr=0x0D2,
 443	/* Gigabit AutoNegotiation. */
 444	ANCtrl=0x0E0, ANStatus=0x0E2, ANXchngCtrl=0x0E4, ANAdvertise=0x0E8,
 445	ANLinkPartnerAbility=0x0EA,
 446	EECmdStatus=0x0F0, EEData=0x0F1, EEAddr=0x0F2,
 447	FIFOcfg=0x0F8,
 448};
 449
 450/* Offsets to the MII-mode registers. */
 451enum MII_offsets {
 452	MII_Cmd=0xA6, MII_Addr=0xA8, MII_Wr_Data=0xAA, MII_Rd_Data=0xAC,
 453	MII_Status=0xAE,
 454};
 455
 456/* Bits in the interrupt status/mask registers. */
 457enum intr_status_bits {
 458	IntrRxDone=0x01, IntrRxPCIFault=0x02, IntrRxPCIErr=0x04,
 459	IntrTxDone=0x100, IntrTxPCIFault=0x200, IntrTxPCIErr=0x400,
 460	LinkChange=0x10000, NegotiationChange=0x20000, StatsMax=0x40000, };
 461
 462/* The Hamachi Rx and Tx buffer descriptors. */
 463struct hamachi_desc {
 464	__le32 status_n_length;
 465#if ADDRLEN == 64
 466	u32 pad;
 467	__le64 addr;
 468#else
 469	__le32 addr;
 470#endif
 471};
 472
 473/* Bits in hamachi_desc.status_n_length */
 474enum desc_status_bits {
 475	DescOwn=0x80000000, DescEndPacket=0x40000000, DescEndRing=0x20000000,
 476	DescIntr=0x10000000,
 477};
 478
 479#define PRIV_ALIGN	15  			/* Required alignment mask */
 480#define MII_CNT		4
 481struct hamachi_private {
 482	/* Descriptor rings first for alignment.  Tx requires a second descriptor
 483	   for status. */
 484	struct hamachi_desc *rx_ring;
 485	struct hamachi_desc *tx_ring;
 486	struct sk_buff* rx_skbuff[RX_RING_SIZE];
 487	struct sk_buff* tx_skbuff[TX_RING_SIZE];
 488	dma_addr_t tx_ring_dma;
 489	dma_addr_t rx_ring_dma;
 490	struct timer_list timer;		/* Media selection timer. */
 491	/* Frequently used and paired value: keep adjacent for cache effect. */
 492	spinlock_t lock;
 493	int chip_id;
 494	unsigned int cur_rx, dirty_rx;		/* Producer/consumer ring indices */
 495	unsigned int cur_tx, dirty_tx;
 496	unsigned int rx_buf_sz;			/* Based on MTU+slack. */
 497	unsigned int tx_full:1;			/* The Tx queue is full. */
 498	unsigned int duplex_lock:1;
 499	unsigned int default_port:4;		/* Last dev->if_port value. */
 500	/* MII transceiver section. */
 501	int mii_cnt;								/* MII device addresses. */
 502	struct mii_if_info mii_if;		/* MII lib hooks/info */
 503	unsigned char phys[MII_CNT];		/* MII device addresses, only first one used. */
 504	u32 rx_int_var, tx_int_var;	/* interrupt control variables */
 505	u32 option;							/* Hold on to a copy of the options */
 506	struct pci_dev *pci_dev;
 507	void __iomem *base;
 508};
 509
 510MODULE_AUTHOR("Donald Becker <becker@scyld.com>, Eric Kasten <kasten@nscl.msu.edu>, Keith Underwood <keithu@parl.clemson.edu>");
 511MODULE_DESCRIPTION("Packet Engines 'Hamachi' GNIC-II Gigabit Ethernet driver");
 512MODULE_LICENSE("GPL");
 513
 514module_param(max_interrupt_work, int, 0);
 515module_param(mtu, int, 0);
 516module_param(debug, int, 0);
 517module_param(min_rx_pkt, int, 0);
 518module_param(max_rx_gap, int, 0);
 519module_param(max_rx_latency, int, 0);
 520module_param(min_tx_pkt, int, 0);
 521module_param(max_tx_gap, int, 0);
 522module_param(max_tx_latency, int, 0);
 523module_param(rx_copybreak, int, 0);
 524module_param_array(rx_params, int, NULL, 0);
 525module_param_array(tx_params, int, NULL, 0);
 526module_param_array(options, int, NULL, 0);
 527module_param_array(full_duplex, int, NULL, 0);
 528module_param(force32, int, 0);
 529MODULE_PARM_DESC(max_interrupt_work, "GNIC-II maximum events handled per interrupt");
 530MODULE_PARM_DESC(mtu, "GNIC-II MTU (all boards)");
 531MODULE_PARM_DESC(debug, "GNIC-II debug level (0-7)");
 532MODULE_PARM_DESC(min_rx_pkt, "GNIC-II minimum Rx packets processed between interrupts");
 533MODULE_PARM_DESC(max_rx_gap, "GNIC-II maximum Rx inter-packet gap in 8.192 microsecond units");
 534MODULE_PARM_DESC(max_rx_latency, "GNIC-II time between Rx interrupts in 8.192 microsecond units");
 535MODULE_PARM_DESC(min_tx_pkt, "GNIC-II minimum Tx packets processed between interrupts");
 536MODULE_PARM_DESC(max_tx_gap, "GNIC-II maximum Tx inter-packet gap in 8.192 microsecond units");
 537MODULE_PARM_DESC(max_tx_latency, "GNIC-II time between Tx interrupts in 8.192 microsecond units");
 538MODULE_PARM_DESC(rx_copybreak, "GNIC-II copy breakpoint for copy-only-tiny-frames");
 539MODULE_PARM_DESC(rx_params, "GNIC-II min_rx_pkt+max_rx_gap+max_rx_latency");
 540MODULE_PARM_DESC(tx_params, "GNIC-II min_tx_pkt+max_tx_gap+max_tx_latency");
 541MODULE_PARM_DESC(options, "GNIC-II Bits 0-3: media type, bits 4-6: as force32, bit 7: half duplex, bit 9 full duplex");
 542MODULE_PARM_DESC(full_duplex, "GNIC-II full duplex setting(s) (1)");
 543MODULE_PARM_DESC(force32, "GNIC-II: Bit 0: 32 bit PCI, bit 1: disable parity, bit 2: 64 bit PCI (all boards)");
 544
 545static int read_eeprom(void __iomem *ioaddr, int location);
 546static int mdio_read(struct net_device *dev, int phy_id, int location);
 547static void mdio_write(struct net_device *dev, int phy_id, int location, int value);
 548static int hamachi_open(struct net_device *dev);
 549static int netdev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd);
 550static void hamachi_timer(unsigned long data);
 551static void hamachi_tx_timeout(struct net_device *dev);
 552static void hamachi_init_ring(struct net_device *dev);
 553static netdev_tx_t hamachi_start_xmit(struct sk_buff *skb,
 554				      struct net_device *dev);
 555static irqreturn_t hamachi_interrupt(int irq, void *dev_instance);
 556static int hamachi_rx(struct net_device *dev);
 557static inline int hamachi_tx(struct net_device *dev);
 558static void hamachi_error(struct net_device *dev, int intr_status);
 559static int hamachi_close(struct net_device *dev);
 560static struct net_device_stats *hamachi_get_stats(struct net_device *dev);
 561static void set_rx_mode(struct net_device *dev);
 562static const struct ethtool_ops ethtool_ops;
 563static const struct ethtool_ops ethtool_ops_no_mii;
 564
 565static const struct net_device_ops hamachi_netdev_ops = {
 566	.ndo_open		= hamachi_open,
 567	.ndo_stop		= hamachi_close,
 568	.ndo_start_xmit		= hamachi_start_xmit,
 569	.ndo_get_stats		= hamachi_get_stats,
 570	.ndo_set_rx_mode	= set_rx_mode,
 571	.ndo_validate_addr	= eth_validate_addr,
 572	.ndo_set_mac_address 	= eth_mac_addr,
 573	.ndo_tx_timeout		= hamachi_tx_timeout,
 574	.ndo_do_ioctl		= netdev_ioctl,
 575};
 576
 577
 578static int hamachi_init_one(struct pci_dev *pdev,
 579			    const struct pci_device_id *ent)
 580{
 581	struct hamachi_private *hmp;
 582	int option, i, rx_int_var, tx_int_var, boguscnt;
 583	int chip_id = ent->driver_data;
 584	int irq;
 585	void __iomem *ioaddr;
 586	unsigned long base;
 587	static int card_idx;
 588	struct net_device *dev;
 589	void *ring_space;
 590	dma_addr_t ring_dma;
 591	int ret = -ENOMEM;
 592
 593/* when built into the kernel, we only print version if device is found */
 594#ifndef MODULE
 595	static int printed_version;
 596	if (!printed_version++)
 597		printk(version);
 598#endif
 599
 600	if (pci_enable_device(pdev)) {
 601		ret = -EIO;
 602		goto err_out;
 603	}
 604
 605	base = pci_resource_start(pdev, 0);
 606#ifdef __alpha__				/* Really "64 bit addrs" */
 607	base |= (pci_resource_start(pdev, 1) << 32);
 608#endif
 609
 610	pci_set_master(pdev);
 611
 612	i = pci_request_regions(pdev, DRV_NAME);
 613	if (i)
 614		return i;
 615
 616	irq = pdev->irq;
 617	ioaddr = ioremap(base, 0x400);
 618	if (!ioaddr)
 619		goto err_out_release;
 620
 621	dev = alloc_etherdev(sizeof(struct hamachi_private));
 622	if (!dev)
 623		goto err_out_iounmap;
 624
 625	SET_NETDEV_DEV(dev, &pdev->dev);
 626
 627	for (i = 0; i < 6; i++)
 628		dev->dev_addr[i] = 1 ? read_eeprom(ioaddr, 4 + i)
 629			: readb(ioaddr + StationAddr + i);
 630
 631#if ! defined(final_version)
 632	if (hamachi_debug > 4)
 633		for (i = 0; i < 0x10; i++)
 634			printk("%2.2x%s",
 635				   read_eeprom(ioaddr, i), i % 16 != 15 ? " " : "\n");
 636#endif
 637
 638	hmp = netdev_priv(dev);
 639	spin_lock_init(&hmp->lock);
 640
 641	hmp->mii_if.dev = dev;
 642	hmp->mii_if.mdio_read = mdio_read;
 643	hmp->mii_if.mdio_write = mdio_write;
 644	hmp->mii_if.phy_id_mask = 0x1f;
 645	hmp->mii_if.reg_num_mask = 0x1f;
 646
 647	ring_space = pci_alloc_consistent(pdev, TX_TOTAL_SIZE, &ring_dma);
 648	if (!ring_space)
 649		goto err_out_cleardev;
 650	hmp->tx_ring = ring_space;
 651	hmp->tx_ring_dma = ring_dma;
 652
 653	ring_space = pci_alloc_consistent(pdev, RX_TOTAL_SIZE, &ring_dma);
 654	if (!ring_space)
 655		goto err_out_unmap_tx;
 656	hmp->rx_ring = ring_space;
 657	hmp->rx_ring_dma = ring_dma;
 658
 659	/* Check for options being passed in */
 660	option = card_idx < MAX_UNITS ? options[card_idx] : 0;
 661	if (dev->mem_start)
 662		option = dev->mem_start;
 663
 664	/* If the bus size is misidentified, do the following. */
 665	force32 = force32 ? force32 :
 666		((option  >= 0) ? ((option & 0x00000070) >> 4) : 0 );
 667	if (force32)
 668		writeb(force32, ioaddr + VirtualJumpers);
 669
 670	/* Hmmm, do we really need to reset the chip???. */
 671	writeb(0x01, ioaddr + ChipReset);
 672
 673	/* After a reset, the clock speed measurement of the PCI bus will not
 674	 * be valid for a moment.  Wait for a little while until it is.  If
 675	 * it takes more than 10ms, forget it.
 676	 */
 677	udelay(10);
 678	i = readb(ioaddr + PCIClkMeas);
 679	for (boguscnt = 0; (!(i & 0x080)) && boguscnt < 1000; boguscnt++){
 680		udelay(10);
 681		i = readb(ioaddr + PCIClkMeas);
 682	}
 683
 684	hmp->base = ioaddr;
 685	pci_set_drvdata(pdev, dev);
 686
 687	hmp->chip_id = chip_id;
 688	hmp->pci_dev = pdev;
 689
 690	/* The lower four bits are the media type. */
 691	if (option > 0) {
 692		hmp->option = option;
 693		if (option & 0x200)
 694			hmp->mii_if.full_duplex = 1;
 695		else if (option & 0x080)
 696			hmp->mii_if.full_duplex = 0;
 697		hmp->default_port = option & 15;
 698		if (hmp->default_port)
 699			hmp->mii_if.force_media = 1;
 700	}
 701	if (card_idx < MAX_UNITS  &&  full_duplex[card_idx] > 0)
 702		hmp->mii_if.full_duplex = 1;
 703
 704	/* lock the duplex mode if someone specified a value */
 705	if (hmp->mii_if.full_duplex || (option & 0x080))
 706		hmp->duplex_lock = 1;
 707
 708	/* Set interrupt tuning parameters */
 709	max_rx_latency = max_rx_latency & 0x00ff;
 710	max_rx_gap = max_rx_gap & 0x00ff;
 711	min_rx_pkt = min_rx_pkt & 0x00ff;
 712	max_tx_latency = max_tx_latency & 0x00ff;
 713	max_tx_gap = max_tx_gap & 0x00ff;
 714	min_tx_pkt = min_tx_pkt & 0x00ff;
 715
 716	rx_int_var = card_idx < MAX_UNITS ? rx_params[card_idx] : -1;
 717	tx_int_var = card_idx < MAX_UNITS ? tx_params[card_idx] : -1;
 718	hmp->rx_int_var = rx_int_var >= 0 ? rx_int_var :
 719		(min_rx_pkt << 16 | max_rx_gap << 8 | max_rx_latency);
 720	hmp->tx_int_var = tx_int_var >= 0 ? tx_int_var :
 721		(min_tx_pkt << 16 | max_tx_gap << 8 | max_tx_latency);
 722
 723
 724	/* The Hamachi-specific entries in the device structure. */
 725	dev->netdev_ops = &hamachi_netdev_ops;
 726	dev->ethtool_ops = (chip_tbl[hmp->chip_id].flags & CanHaveMII) ?
 727		&ethtool_ops : &ethtool_ops_no_mii;
 728	dev->watchdog_timeo = TX_TIMEOUT;
 729	if (mtu)
 730		dev->mtu = mtu;
 731
 732	i = register_netdev(dev);
 733	if (i) {
 734		ret = i;
 735		goto err_out_unmap_rx;
 736	}
 737
 738	printk(KERN_INFO "%s: %s type %x at %p, %pM, IRQ %d.\n",
 739		   dev->name, chip_tbl[chip_id].name, readl(ioaddr + ChipRev),
 740		   ioaddr, dev->dev_addr, irq);
 741	i = readb(ioaddr + PCIClkMeas);
 742	printk(KERN_INFO "%s:  %d-bit %d Mhz PCI bus (%d), Virtual Jumpers "
 743		   "%2.2x, LPA %4.4x.\n",
 744		   dev->name, readw(ioaddr + MiscStatus) & 1 ? 64 : 32,
 745		   i ? 2000/(i&0x7f) : 0, i&0x7f, (int)readb(ioaddr + VirtualJumpers),
 746		   readw(ioaddr + ANLinkPartnerAbility));
 747
 748	if (chip_tbl[hmp->chip_id].flags & CanHaveMII) {
 749		int phy, phy_idx = 0;
 750		for (phy = 0; phy < 32 && phy_idx < MII_CNT; phy++) {
 751			int mii_status = mdio_read(dev, phy, MII_BMSR);
 752			if (mii_status != 0xffff  &&
 753				mii_status != 0x0000) {
 754				hmp->phys[phy_idx++] = phy;
 755				hmp->mii_if.advertising = mdio_read(dev, phy, MII_ADVERTISE);
 756				printk(KERN_INFO "%s: MII PHY found at address %d, status "
 757					   "0x%4.4x advertising %4.4x.\n",
 758					   dev->name, phy, mii_status, hmp->mii_if.advertising);
 759			}
 760		}
 761		hmp->mii_cnt = phy_idx;
 762		if (hmp->mii_cnt > 0)
 763			hmp->mii_if.phy_id = hmp->phys[0];
 764		else
 765			memset(&hmp->mii_if, 0, sizeof(hmp->mii_if));
 766	}
 767	/* Configure gigabit autonegotiation. */
 768	writew(0x0400, ioaddr + ANXchngCtrl);	/* Enable legacy links. */
 769	writew(0x08e0, ioaddr + ANAdvertise);	/* Set our advertise word. */
 770	writew(0x1000, ioaddr + ANCtrl);			/* Enable negotiation */
 771
 772	card_idx++;
 773	return 0;
 774
 775err_out_unmap_rx:
 776	pci_free_consistent(pdev, RX_TOTAL_SIZE, hmp->rx_ring,
 777		hmp->rx_ring_dma);
 778err_out_unmap_tx:
 779	pci_free_consistent(pdev, TX_TOTAL_SIZE, hmp->tx_ring,
 780		hmp->tx_ring_dma);
 781err_out_cleardev:
 782	free_netdev (dev);
 783err_out_iounmap:
 784	iounmap(ioaddr);
 785err_out_release:
 786	pci_release_regions(pdev);
 787err_out:
 788	return ret;
 789}
 790
 791static int read_eeprom(void __iomem *ioaddr, int location)
 792{
 793	int bogus_cnt = 1000;
 794
 795	/* We should check busy first - per docs -KDU */
 796	while ((readb(ioaddr + EECmdStatus) & 0x40)  && --bogus_cnt > 0);
 797	writew(location, ioaddr + EEAddr);
 798	writeb(0x02, ioaddr + EECmdStatus);
 799	bogus_cnt = 1000;
 800	while ((readb(ioaddr + EECmdStatus) & 0x40)  && --bogus_cnt > 0);
 801	if (hamachi_debug > 5)
 802		printk("   EEPROM status is %2.2x after %d ticks.\n",
 803			   (int)readb(ioaddr + EECmdStatus), 1000- bogus_cnt);
 804	return readb(ioaddr + EEData);
 805}
 806
 807/* MII Managemen Data I/O accesses.
 808   These routines assume the MDIO controller is idle, and do not exit until
 809   the command is finished. */
 810
 811static int mdio_read(struct net_device *dev, int phy_id, int location)
 812{
 813	struct hamachi_private *hmp = netdev_priv(dev);
 814	void __iomem *ioaddr = hmp->base;
 815	int i;
 816
 817	/* We should check busy first - per docs -KDU */
 818	for (i = 10000; i >= 0; i--)
 819		if ((readw(ioaddr + MII_Status) & 1) == 0)
 820			break;
 821	writew((phy_id<<8) + location, ioaddr + MII_Addr);
 822	writew(0x0001, ioaddr + MII_Cmd);
 823	for (i = 10000; i >= 0; i--)
 824		if ((readw(ioaddr + MII_Status) & 1) == 0)
 825			break;
 826	return readw(ioaddr + MII_Rd_Data);
 827}
 828
 829static void mdio_write(struct net_device *dev, int phy_id, int location, int value)
 830{
 831	struct hamachi_private *hmp = netdev_priv(dev);
 832	void __iomem *ioaddr = hmp->base;
 833	int i;
 834
 835	/* We should check busy first - per docs -KDU */
 836	for (i = 10000; i >= 0; i--)
 837		if ((readw(ioaddr + MII_Status) & 1) == 0)
 838			break;
 839	writew((phy_id<<8) + location, ioaddr + MII_Addr);
 840	writew(value, ioaddr + MII_Wr_Data);
 841
 842	/* Wait for the command to finish. */
 843	for (i = 10000; i >= 0; i--)
 844		if ((readw(ioaddr + MII_Status) & 1) == 0)
 845			break;
 846}
 847
 848
 849static int hamachi_open(struct net_device *dev)
 850{
 851	struct hamachi_private *hmp = netdev_priv(dev);
 852	void __iomem *ioaddr = hmp->base;
 853	int i;
 854	u32 rx_int_var, tx_int_var;
 855	u16 fifo_info;
 856
 857	i = request_irq(hmp->pci_dev->irq, hamachi_interrupt, IRQF_SHARED,
 858			dev->name, dev);
 859	if (i)
 860		return i;
 861
 862	hamachi_init_ring(dev);
 863
 864#if ADDRLEN == 64
 865	/* writellll anyone ? */
 866	writel(hmp->rx_ring_dma, ioaddr + RxPtr);
 867	writel(hmp->rx_ring_dma >> 32, ioaddr + RxPtr + 4);
 868	writel(hmp->tx_ring_dma, ioaddr + TxPtr);
 869	writel(hmp->tx_ring_dma >> 32, ioaddr + TxPtr + 4);
 870#else
 871	writel(hmp->rx_ring_dma, ioaddr + RxPtr);
 872	writel(hmp->tx_ring_dma, ioaddr + TxPtr);
 873#endif
 874
 875	/* TODO:  It would make sense to organize this as words since the card
 876	 * documentation does. -KDU
 877	 */
 878	for (i = 0; i < 6; i++)
 879		writeb(dev->dev_addr[i], ioaddr + StationAddr + i);
 880
 881	/* Initialize other registers: with so many this eventually this will
 882	   converted to an offset/value list. */
 883
 884	/* Configure the FIFO */
 885	fifo_info = (readw(ioaddr + GPIO) & 0x00C0) >> 6;
 886	switch (fifo_info){
 887		case 0 :
 888			/* No FIFO */
 889			writew(0x0000, ioaddr + FIFOcfg);
 890			break;
 891		case 1 :
 892			/* Configure the FIFO for 512K external, 16K used for Tx. */
 893			writew(0x0028, ioaddr + FIFOcfg);
 894			break;
 895		case 2 :
 896			/* Configure the FIFO for 1024 external, 32K used for Tx. */
 897			writew(0x004C, ioaddr + FIFOcfg);
 898			break;
 899		case 3 :
 900			/* Configure the FIFO for 2048 external, 32K used for Tx. */
 901			writew(0x006C, ioaddr + FIFOcfg);
 902			break;
 903		default :
 904			printk(KERN_WARNING "%s:  Unsupported external memory config!\n",
 905				dev->name);
 906			/* Default to no FIFO */
 907			writew(0x0000, ioaddr + FIFOcfg);
 908			break;
 909	}
 910
 911	if (dev->if_port == 0)
 912		dev->if_port = hmp->default_port;
 913
 914
 915	/* Setting the Rx mode will start the Rx process. */
 916	/* If someone didn't choose a duplex, default to full-duplex */
 917	if (hmp->duplex_lock != 1)
 918		hmp->mii_if.full_duplex = 1;
 919
 920	/* always 1, takes no more time to do it */
 921	writew(0x0001, ioaddr + RxChecksum);
 922	writew(0x0000, ioaddr + TxChecksum);
 923	writew(0x8000, ioaddr + MACCnfg); /* Soft reset the MAC */
 924	writew(0x215F, ioaddr + MACCnfg);
 925	writew(0x000C, ioaddr + FrameGap0);
 926	/* WHAT?!?!?  Why isn't this documented somewhere? -KDU */
 927	writew(0x1018, ioaddr + FrameGap1);
 928	/* Why do we enable receives/transmits here? -KDU */
 929	writew(0x0780, ioaddr + MACCnfg2); /* Upper 16 bits control LEDs. */
 930	/* Enable automatic generation of flow control frames, period 0xffff. */
 931	writel(0x0030FFFF, ioaddr + FlowCtrl);
 932	writew(MAX_FRAME_SIZE, ioaddr + MaxFrameSize); 	/* dev->mtu+14 ??? */
 933
 934	/* Enable legacy links. */
 935	writew(0x0400, ioaddr + ANXchngCtrl);	/* Enable legacy links. */
 936	/* Initial Link LED to blinking red. */
 937	writeb(0x03, ioaddr + LEDCtrl);
 938
 939	/* Configure interrupt mitigation.  This has a great effect on
 940	   performance, so systems tuning should start here!. */
 941
 942	rx_int_var = hmp->rx_int_var;
 943	tx_int_var = hmp->tx_int_var;
 944
 945	if (hamachi_debug > 1) {
 946		printk("max_tx_latency: %d, max_tx_gap: %d, min_tx_pkt: %d\n",
 947			tx_int_var & 0x00ff, (tx_int_var & 0x00ff00) >> 8,
 948			(tx_int_var & 0x00ff0000) >> 16);
 949		printk("max_rx_latency: %d, max_rx_gap: %d, min_rx_pkt: %d\n",
 950			rx_int_var & 0x00ff, (rx_int_var & 0x00ff00) >> 8,
 951			(rx_int_var & 0x00ff0000) >> 16);
 952		printk("rx_int_var: %x, tx_int_var: %x\n", rx_int_var, tx_int_var);
 953	}
 954
 955	writel(tx_int_var, ioaddr + TxIntrCtrl);
 956	writel(rx_int_var, ioaddr + RxIntrCtrl);
 957
 958	set_rx_mode(dev);
 959
 960	netif_start_queue(dev);
 961
 962	/* Enable interrupts by setting the interrupt mask. */
 963	writel(0x80878787, ioaddr + InterruptEnable);
 964	writew(0x0000, ioaddr + EventStatus);	/* Clear non-interrupting events */
 965
 966	/* Configure and start the DMA channels. */
 967	/* Burst sizes are in the low three bits: size = 4<<(val&7) */
 968#if ADDRLEN == 64
 969	writew(0x005D, ioaddr + RxDMACtrl); 		/* 128 dword bursts */
 970	writew(0x005D, ioaddr + TxDMACtrl);
 971#else
 972	writew(0x001D, ioaddr + RxDMACtrl);
 973	writew(0x001D, ioaddr + TxDMACtrl);
 974#endif
 975	writew(0x0001, ioaddr + RxCmd);
 976
 977	if (hamachi_debug > 2) {
 978		printk(KERN_DEBUG "%s: Done hamachi_open(), status: Rx %x Tx %x.\n",
 979			   dev->name, readw(ioaddr + RxStatus), readw(ioaddr + TxStatus));
 980	}
 981	/* Set the timer to check for link beat. */
 982	init_timer(&hmp->timer);
 983	hmp->timer.expires = RUN_AT((24*HZ)/10);			/* 2.4 sec. */
 984	hmp->timer.data = (unsigned long)dev;
 985	hmp->timer.function = hamachi_timer;				/* timer handler */
 986	add_timer(&hmp->timer);
 987
 988	return 0;
 989}
 990
 991static inline int hamachi_tx(struct net_device *dev)
 992{
 993	struct hamachi_private *hmp = netdev_priv(dev);
 994
 995	/* Update the dirty pointer until we find an entry that is
 996		still owned by the card */
 997	for (; hmp->cur_tx - hmp->dirty_tx > 0; hmp->dirty_tx++) {
 998		int entry = hmp->dirty_tx % TX_RING_SIZE;
 999		struct sk_buff *skb;
1000
1001		if (hmp->tx_ring[entry].status_n_length & cpu_to_le32(DescOwn))
1002			break;
1003		/* Free the original skb. */
1004		skb = hmp->tx_skbuff[entry];
1005		if (skb) {
1006			pci_unmap_single(hmp->pci_dev,
1007				leXX_to_cpu(hmp->tx_ring[entry].addr),
1008				skb->len, PCI_DMA_TODEVICE);
1009			dev_kfree_skb(skb);
1010			hmp->tx_skbuff[entry] = NULL;
1011		}
1012		hmp->tx_ring[entry].status_n_length = 0;
1013		if (entry >= TX_RING_SIZE-1)
1014			hmp->tx_ring[TX_RING_SIZE-1].status_n_length |=
1015				cpu_to_le32(DescEndRing);
1016		dev->stats.tx_packets++;
1017	}
1018
1019	return 0;
1020}
1021
1022static void hamachi_timer(unsigned long data)
1023{
1024	struct net_device *dev = (struct net_device *)data;
1025	struct hamachi_private *hmp = netdev_priv(dev);
1026	void __iomem *ioaddr = hmp->base;
1027	int next_tick = 10*HZ;
1028
1029	if (hamachi_debug > 2) {
1030		printk(KERN_INFO "%s: Hamachi Autonegotiation status %4.4x, LPA "
1031			   "%4.4x.\n", dev->name, readw(ioaddr + ANStatus),
1032			   readw(ioaddr + ANLinkPartnerAbility));
1033		printk(KERN_INFO "%s: Autonegotiation regs %4.4x %4.4x %4.4x "
1034		       "%4.4x %4.4x %4.4x.\n", dev->name,
1035		       readw(ioaddr + 0x0e0),
1036		       readw(ioaddr + 0x0e2),
1037		       readw(ioaddr + 0x0e4),
1038		       readw(ioaddr + 0x0e6),
1039		       readw(ioaddr + 0x0e8),
1040		       readw(ioaddr + 0x0eA));
1041	}
1042	/* We could do something here... nah. */
1043	hmp->timer.expires = RUN_AT(next_tick);
1044	add_timer(&hmp->timer);
1045}
1046
1047static void hamachi_tx_timeout(struct net_device *dev)
1048{
1049	int i;
1050	struct hamachi_private *hmp = netdev_priv(dev);
1051	void __iomem *ioaddr = hmp->base;
1052
1053	printk(KERN_WARNING "%s: Hamachi transmit timed out, status %8.8x,"
1054		   " resetting...\n", dev->name, (int)readw(ioaddr + TxStatus));
1055
1056	{
1057		printk(KERN_DEBUG "  Rx ring %p: ", hmp->rx_ring);
1058		for (i = 0; i < RX_RING_SIZE; i++)
1059			printk(KERN_CONT " %8.8x",
1060			       le32_to_cpu(hmp->rx_ring[i].status_n_length));
1061		printk(KERN_CONT "\n");
1062		printk(KERN_DEBUG"  Tx ring %p: ", hmp->tx_ring);
1063		for (i = 0; i < TX_RING_SIZE; i++)
1064			printk(KERN_CONT " %4.4x",
1065			       le32_to_cpu(hmp->tx_ring[i].status_n_length));
1066		printk(KERN_CONT "\n");
1067	}
1068
1069	/* Reinit the hardware and make sure the Rx and Tx processes
1070		are up and running.
1071	 */
1072	dev->if_port = 0;
1073	/* The right way to do Reset. -KDU
1074	 *		-Clear OWN bit in all Rx/Tx descriptors
1075	 *		-Wait 50 uS for channels to go idle
1076	 *		-Turn off MAC receiver
1077	 *		-Issue Reset
1078	 */
1079
1080	for (i = 0; i < RX_RING_SIZE; i++)
1081		hmp->rx_ring[i].status_n_length &= cpu_to_le32(~DescOwn);
1082
1083	/* Presume that all packets in the Tx queue are gone if we have to
1084	 * re-init the hardware.
1085	 */
1086	for (i = 0; i < TX_RING_SIZE; i++){
1087		struct sk_buff *skb;
1088
1089		if (i >= TX_RING_SIZE - 1)
1090			hmp->tx_ring[i].status_n_length =
1091				cpu_to_le32(DescEndRing) |
1092				(hmp->tx_ring[i].status_n_length &
1093				 cpu_to_le32(0x0000ffff));
1094		else
1095			hmp->tx_ring[i].status_n_length &= cpu_to_le32(0x0000ffff);
1096		skb = hmp->tx_skbuff[i];
1097		if (skb){
1098			pci_unmap_single(hmp->pci_dev, leXX_to_cpu(hmp->tx_ring[i].addr),
1099				skb->len, PCI_DMA_TODEVICE);
1100			dev_kfree_skb(skb);
1101			hmp->tx_skbuff[i] = NULL;
1102		}
1103	}
1104
1105	udelay(60); /* Sleep 60 us just for safety sake */
1106	writew(0x0002, ioaddr + RxCmd); /* STOP Rx */
1107
1108	writeb(0x01, ioaddr + ChipReset);  /* Reinit the hardware */
1109
1110	hmp->tx_full = 0;
1111	hmp->cur_rx = hmp->cur_tx = 0;
1112	hmp->dirty_rx = hmp->dirty_tx = 0;
1113	/* Rx packets are also presumed lost; however, we need to make sure a
1114	 * ring of buffers is in tact. -KDU
1115	 */
1116	for (i = 0; i < RX_RING_SIZE; i++){
1117		struct sk_buff *skb = hmp->rx_skbuff[i];
1118
1119		if (skb){
1120			pci_unmap_single(hmp->pci_dev,
1121				leXX_to_cpu(hmp->rx_ring[i].addr),
1122				hmp->rx_buf_sz, PCI_DMA_FROMDEVICE);
1123			dev_kfree_skb(skb);
1124			hmp->rx_skbuff[i] = NULL;
1125		}
1126	}
1127	/* Fill in the Rx buffers.  Handle allocation failure gracefully. */
1128	for (i = 0; i < RX_RING_SIZE; i++) {
1129		struct sk_buff *skb;
1130
1131		skb = netdev_alloc_skb_ip_align(dev, hmp->rx_buf_sz);
1132		hmp->rx_skbuff[i] = skb;
1133		if (skb == NULL)
1134			break;
1135
1136                hmp->rx_ring[i].addr = cpu_to_leXX(pci_map_single(hmp->pci_dev,
1137			skb->data, hmp->rx_buf_sz, PCI_DMA_FROMDEVICE));
1138		hmp->rx_ring[i].status_n_length = cpu_to_le32(DescOwn |
1139			DescEndPacket | DescIntr | (hmp->rx_buf_sz - 2));
1140	}
1141	hmp->dirty_rx = (unsigned int)(i - RX_RING_SIZE);
1142	/* Mark the last entry as wrapping the ring. */
1143	hmp->rx_ring[RX_RING_SIZE-1].status_n_length |= cpu_to_le32(DescEndRing);
1144
1145	/* Trigger an immediate transmit demand. */
1146	netif_trans_update(dev); /* prevent tx timeout */
1147	dev->stats.tx_errors++;
1148
1149	/* Restart the chip's Tx/Rx processes . */
1150	writew(0x0002, ioaddr + TxCmd); /* STOP Tx */
1151	writew(0x0001, ioaddr + TxCmd); /* START Tx */
1152	writew(0x0001, ioaddr + RxCmd); /* START Rx */
1153
1154	netif_wake_queue(dev);
1155}
1156
1157
1158/* Initialize the Rx and Tx rings, along with various 'dev' bits. */
1159static void hamachi_init_ring(struct net_device *dev)
1160{
1161	struct hamachi_private *hmp = netdev_priv(dev);
1162	int i;
1163
1164	hmp->tx_full = 0;
1165	hmp->cur_rx = hmp->cur_tx = 0;
1166	hmp->dirty_rx = hmp->dirty_tx = 0;
1167
1168	/* +26 gets the maximum ethernet encapsulation, +7 & ~7 because the
1169	 * card needs room to do 8 byte alignment, +2 so we can reserve
1170	 * the first 2 bytes, and +16 gets room for the status word from the
1171	 * card.  -KDU
1172	 */
1173	hmp->rx_buf_sz = (dev->mtu <= 1492 ? PKT_BUF_SZ :
1174		(((dev->mtu+26+7) & ~7) + 16));
1175
1176	/* Initialize all Rx descriptors. */
1177	for (i = 0; i < RX_RING_SIZE; i++) {
1178		hmp->rx_ring[i].status_n_length = 0;
1179		hmp->rx_skbuff[i] = NULL;
1180	}
1181	/* Fill in the Rx buffers.  Handle allocation failure gracefully. */
1182	for (i = 0; i < RX_RING_SIZE; i++) {
1183		struct sk_buff *skb = netdev_alloc_skb(dev, hmp->rx_buf_sz + 2);
1184		hmp->rx_skbuff[i] = skb;
1185		if (skb == NULL)
1186			break;
1187		skb_reserve(skb, 2); /* 16 byte align the IP header. */
1188                hmp->rx_ring[i].addr = cpu_to_leXX(pci_map_single(hmp->pci_dev,
1189			skb->data, hmp->rx_buf_sz, PCI_DMA_FROMDEVICE));
1190		/* -2 because it doesn't REALLY have that first 2 bytes -KDU */
1191		hmp->rx_ring[i].status_n_length = cpu_to_le32(DescOwn |
1192			DescEndPacket | DescIntr | (hmp->rx_buf_sz -2));
1193	}
1194	hmp->dirty_rx = (unsigned int)(i - RX_RING_SIZE);
1195	hmp->rx_ring[RX_RING_SIZE-1].status_n_length |= cpu_to_le32(DescEndRing);
1196
1197	for (i = 0; i < TX_RING_SIZE; i++) {
1198		hmp->tx_skbuff[i] = NULL;
1199		hmp->tx_ring[i].status_n_length = 0;
1200	}
1201	/* Mark the last entry of the ring */
1202	hmp->tx_ring[TX_RING_SIZE-1].status_n_length |= cpu_to_le32(DescEndRing);
1203}
1204
1205
1206static netdev_tx_t hamachi_start_xmit(struct sk_buff *skb,
1207				      struct net_device *dev)
1208{
1209	struct hamachi_private *hmp = netdev_priv(dev);
1210	unsigned entry;
1211	u16 status;
1212
1213	/* Ok, now make sure that the queue has space before trying to
1214		add another skbuff.  if we return non-zero the scheduler
1215		should interpret this as a queue full and requeue the buffer
1216		for later.
1217	 */
1218	if (hmp->tx_full) {
1219		/* We should NEVER reach this point -KDU */
1220		printk(KERN_WARNING "%s: Hamachi transmit queue full at slot %d.\n",dev->name, hmp->cur_tx);
1221
1222		/* Wake the potentially-idle transmit channel. */
1223		/* If we don't need to read status, DON'T -KDU */
1224		status=readw(hmp->base + TxStatus);
1225		if( !(status & 0x0001) || (status & 0x0002))
1226			writew(0x0001, hmp->base + TxCmd);
1227		return NETDEV_TX_BUSY;
1228	}
1229
1230	/* Caution: the write order is important here, set the field
1231	   with the "ownership" bits last. */
1232
1233	/* Calculate the next Tx descriptor entry. */
1234	entry = hmp->cur_tx % TX_RING_SIZE;
1235
1236	hmp->tx_skbuff[entry] = skb;
1237
1238        hmp->tx_ring[entry].addr = cpu_to_leXX(pci_map_single(hmp->pci_dev,
1239		skb->data, skb->len, PCI_DMA_TODEVICE));
1240
1241	/* Hmmmm, could probably put a DescIntr on these, but the way
1242		the driver is currently coded makes Tx interrupts unnecessary
1243		since the clearing of the Tx ring is handled by the start_xmit
1244		routine.  This organization helps mitigate the interrupts a
1245		bit and probably renders the max_tx_latency param useless.
1246
1247		Update: Putting a DescIntr bit on all of the descriptors and
1248		mitigating interrupt frequency with the tx_min_pkt parameter. -KDU
1249	*/
1250	if (entry >= TX_RING_SIZE-1)		 /* Wrap ring */
1251		hmp->tx_ring[entry].status_n_length = cpu_to_le32(DescOwn |
1252			DescEndPacket | DescEndRing | DescIntr | skb->len);
1253	else
1254		hmp->tx_ring[entry].status_n_length = cpu_to_le32(DescOwn |
1255			DescEndPacket | DescIntr | skb->len);
1256	hmp->cur_tx++;
1257
1258	/* Non-x86 Todo: explicitly flush cache lines here. */
1259
1260	/* Wake the potentially-idle transmit channel. */
1261	/* If we don't need to read status, DON'T -KDU */
1262	status=readw(hmp->base + TxStatus);
1263	if( !(status & 0x0001) || (status & 0x0002))
1264		writew(0x0001, hmp->base + TxCmd);
1265
1266	/* Immediately before returning, let's clear as many entries as we can. */
1267	hamachi_tx(dev);
1268
1269	/* We should kick the bottom half here, since we are not accepting
1270	 * interrupts with every packet.  i.e. realize that Gigabit ethernet
1271	 * can transmit faster than ordinary machines can load packets;
1272	 * hence, any packet that got put off because we were in the transmit
1273	 * routine should IMMEDIATELY get a chance to be re-queued. -KDU
1274	 */
1275	if ((hmp->cur_tx - hmp->dirty_tx) < (TX_RING_SIZE - 4))
1276		netif_wake_queue(dev);  /* Typical path */
1277	else {
1278		hmp->tx_full = 1;
1279		netif_stop_queue(dev);
1280	}
1281
1282	if (hamachi_debug > 4) {
1283		printk(KERN_DEBUG "%s: Hamachi transmit frame #%d queued in slot %d.\n",
1284			   dev->name, hmp->cur_tx, entry);
1285	}
1286	return NETDEV_TX_OK;
1287}
1288
1289/* The interrupt handler does all of the Rx thread work and cleans up
1290   after the Tx thread. */
1291static irqreturn_t hamachi_interrupt(int irq, void *dev_instance)
1292{
1293	struct net_device *dev = dev_instance;
1294	struct hamachi_private *hmp = netdev_priv(dev);
1295	void __iomem *ioaddr = hmp->base;
1296	long boguscnt = max_interrupt_work;
1297	int handled = 0;
1298
1299#ifndef final_version			/* Can never occur. */
1300	if (dev == NULL) {
1301		printk (KERN_ERR "hamachi_interrupt(): irq %d for unknown device.\n", irq);
1302		return IRQ_NONE;
1303	}
1304#endif
1305
1306	spin_lock(&hmp->lock);
1307
1308	do {
1309		u32 intr_status = readl(ioaddr + InterruptClear);
1310
1311		if (hamachi_debug > 4)
1312			printk(KERN_DEBUG "%s: Hamachi interrupt, status %4.4x.\n",
1313				   dev->name, intr_status);
1314
1315		if (intr_status == 0)
1316			break;
1317
1318		handled = 1;
1319
1320		if (intr_status & IntrRxDone)
1321			hamachi_rx(dev);
1322
1323		if (intr_status & IntrTxDone){
1324			/* This code should RARELY need to execute. After all, this is
1325			 * a gigabit link, it should consume packets as fast as we put
1326			 * them in AND we clear the Tx ring in hamachi_start_xmit().
1327			 */
1328			if (hmp->tx_full){
1329				for (; hmp->cur_tx - hmp->dirty_tx > 0; hmp->dirty_tx++){
1330					int entry = hmp->dirty_tx % TX_RING_SIZE;
1331					struct sk_buff *skb;
1332
1333					if (hmp->tx_ring[entry].status_n_length & cpu_to_le32(DescOwn))
1334						break;
1335					skb = hmp->tx_skbuff[entry];
1336					/* Free the original skb. */
1337					if (skb){
1338						pci_unmap_single(hmp->pci_dev,
1339							leXX_to_cpu(hmp->tx_ring[entry].addr),
1340							skb->len,
1341							PCI_DMA_TODEVICE);
1342						dev_kfree_skb_irq(skb);
1343						hmp->tx_skbuff[entry] = NULL;
1344					}
1345					hmp->tx_ring[entry].status_n_length = 0;
1346					if (entry >= TX_RING_SIZE-1)
1347						hmp->tx_ring[TX_RING_SIZE-1].status_n_length |=
1348							cpu_to_le32(DescEndRing);
1349					dev->stats.tx_packets++;
1350				}
1351				if (hmp->cur_tx - hmp->dirty_tx < TX_RING_SIZE - 4){
1352					/* The ring is no longer full */
1353					hmp->tx_full = 0;
1354					netif_wake_queue(dev);
1355				}
1356			} else {
1357				netif_wake_queue(dev);
1358			}
1359		}
1360
1361
1362		/* Abnormal error summary/uncommon events handlers. */
1363		if (intr_status &
1364			(IntrTxPCIFault | IntrTxPCIErr | IntrRxPCIFault | IntrRxPCIErr |
1365			 LinkChange | NegotiationChange | StatsMax))
1366			hamachi_error(dev, intr_status);
1367
1368		if (--boguscnt < 0) {
1369			printk(KERN_WARNING "%s: Too much work at interrupt, status=0x%4.4x.\n",
1370				   dev->name, intr_status);
1371			break;
1372		}
1373	} while (1);
1374
1375	if (hamachi_debug > 3)
1376		printk(KERN_DEBUG "%s: exiting interrupt, status=%#4.4x.\n",
1377			   dev->name, readl(ioaddr + IntrStatus));
1378
1379#ifndef final_version
1380	/* Code that should never be run!  Perhaps remove after testing.. */
1381	{
1382		static int stopit = 10;
1383		if (dev->start == 0  &&  --stopit < 0) {
1384			printk(KERN_ERR "%s: Emergency stop, looping startup interrupt.\n",
1385				   dev->name);
1386			free_irq(irq, dev);
1387		}
1388	}
1389#endif
1390
1391	spin_unlock(&hmp->lock);
1392	return IRQ_RETVAL(handled);
1393}
1394
1395/* This routine is logically part of the interrupt handler, but separated
1396   for clarity and better register allocation. */
1397static int hamachi_rx(struct net_device *dev)
1398{
1399	struct hamachi_private *hmp = netdev_priv(dev);
1400	int entry = hmp->cur_rx % RX_RING_SIZE;
1401	int boguscnt = (hmp->dirty_rx + RX_RING_SIZE) - hmp->cur_rx;
1402
1403	if (hamachi_debug > 4) {
1404		printk(KERN_DEBUG " In hamachi_rx(), entry %d status %4.4x.\n",
1405			   entry, hmp->rx_ring[entry].status_n_length);
1406	}
1407
1408	/* If EOP is set on the next entry, it's a new packet. Send it up. */
1409	while (1) {
1410		struct hamachi_desc *desc = &(hmp->rx_ring[entry]);
1411		u32 desc_status = le32_to_cpu(desc->status_n_length);
1412		u16 data_size = desc_status;	/* Implicit truncate */
1413		u8 *buf_addr;
1414		s32 frame_status;
1415
1416		if (desc_status & DescOwn)
1417			break;
1418		pci_dma_sync_single_for_cpu(hmp->pci_dev,
1419					    leXX_to_cpu(desc->addr),
1420					    hmp->rx_buf_sz,
1421					    PCI_DMA_FROMDEVICE);
1422		buf_addr = (u8 *) hmp->rx_skbuff[entry]->data;
1423		frame_status = get_unaligned_le32(&(buf_addr[data_size - 12]));
1424		if (hamachi_debug > 4)
1425			printk(KERN_DEBUG "  hamachi_rx() status was %8.8x.\n",
1426				frame_status);
1427		if (--boguscnt < 0)
1428			break;
1429		if ( ! (desc_status & DescEndPacket)) {
1430			printk(KERN_WARNING "%s: Oversized Ethernet frame spanned "
1431				   "multiple buffers, entry %#x length %d status %4.4x!\n",
1432				   dev->name, hmp->cur_rx, data_size, desc_status);
1433			printk(KERN_WARNING "%s: Oversized Ethernet frame %p vs %p.\n",
1434				   dev->name, desc, &hmp->rx_ring[hmp->cur_rx % RX_RING_SIZE]);
1435			printk(KERN_WARNING "%s: Oversized Ethernet frame -- next status %x/%x last status %x.\n",
1436				   dev->name,
1437				   le32_to_cpu(hmp->rx_ring[(hmp->cur_rx+1) % RX_RING_SIZE].status_n_length) & 0xffff0000,
1438				   le32_to_cpu(hmp->rx_ring[(hmp->cur_rx+1) % RX_RING_SIZE].status_n_length) & 0x0000ffff,
1439				   le32_to_cpu(hmp->rx_ring[(hmp->cur_rx-1) % RX_RING_SIZE].status_n_length));
1440			dev->stats.rx_length_errors++;
1441		} /* else  Omit for prototype errata??? */
1442		if (frame_status & 0x00380000) {
1443			/* There was an error. */
1444			if (hamachi_debug > 2)
1445				printk(KERN_DEBUG "  hamachi_rx() Rx error was %8.8x.\n",
1446					   frame_status);
1447			dev->stats.rx_errors++;
1448			if (frame_status & 0x00600000)
1449				dev->stats.rx_length_errors++;
1450			if (frame_status & 0x00080000)
1451				dev->stats.rx_frame_errors++;
1452			if (frame_status & 0x00100000)
1453				dev->stats.rx_crc_errors++;
1454			if (frame_status < 0)
1455				dev->stats.rx_dropped++;
1456		} else {
1457			struct sk_buff *skb;
1458			/* Omit CRC */
1459			u16 pkt_len = (frame_status & 0x07ff) - 4;
1460#ifdef RX_CHECKSUM
1461			u32 pfck = *(u32 *) &buf_addr[data_size - 8];
1462#endif
1463
1464
1465#ifndef final_version
1466			if (hamachi_debug > 4)
1467				printk(KERN_DEBUG "  hamachi_rx() normal Rx pkt length %d"
1468					   " of %d, bogus_cnt %d.\n",
1469					   pkt_len, data_size, boguscnt);
1470			if (hamachi_debug > 5)
1471				printk(KERN_DEBUG"%s:  rx status %8.8x %8.8x %8.8x %8.8x %8.8x.\n",
1472					   dev->name,
1473					   *(s32*)&(buf_addr[data_size - 20]),
1474					   *(s32*)&(buf_addr[data_size - 16]),
1475					   *(s32*)&(buf_addr[data_size - 12]),
1476					   *(s32*)&(buf_addr[data_size - 8]),
1477					   *(s32*)&(buf_addr[data_size - 4]));
1478#endif
1479			/* Check if the packet is long enough to accept without copying
1480			   to a minimally-sized skbuff. */
1481			if (pkt_len < rx_copybreak &&
1482			    (skb = netdev_alloc_skb(dev, pkt_len + 2)) != NULL) {
1483#ifdef RX_CHECKSUM
1484				printk(KERN_ERR "%s: rx_copybreak non-zero "
1485				  "not good with RX_CHECKSUM\n", dev->name);
1486#endif
1487				skb_reserve(skb, 2);	/* 16 byte align the IP header */
1488				pci_dma_sync_single_for_cpu(hmp->pci_dev,
1489							    leXX_to_cpu(hmp->rx_ring[entry].addr),
1490							    hmp->rx_buf_sz,
1491							    PCI_DMA_FROMDEVICE);
1492				/* Call copy + cksum if available. */
1493#if 1 || USE_IP_COPYSUM
1494				skb_copy_to_linear_data(skb,
1495					hmp->rx_skbuff[entry]->data, pkt_len);
1496				skb_put(skb, pkt_len);
1497#else
1498				memcpy(skb_put(skb, pkt_len), hmp->rx_ring_dma
1499					+ entry*sizeof(*desc), pkt_len);
1500#endif
1501				pci_dma_sync_single_for_device(hmp->pci_dev,
1502							       leXX_to_cpu(hmp->rx_ring[entry].addr),
1503							       hmp->rx_buf_sz,
1504							       PCI_DMA_FROMDEVICE);
1505			} else {
1506				pci_unmap_single(hmp->pci_dev,
1507						 leXX_to_cpu(hmp->rx_ring[entry].addr),
1508						 hmp->rx_buf_sz, PCI_DMA_FROMDEVICE);
1509				skb_put(skb = hmp->rx_skbuff[entry], pkt_len);
1510				hmp->rx_skbuff[entry] = NULL;
1511			}
1512			skb->protocol = eth_type_trans(skb, dev);
1513
1514
1515#ifdef RX_CHECKSUM
1516			/* TCP or UDP on ipv4, DIX encoding */
1517			if (pfck>>24 == 0x91 || pfck>>24 == 0x51) {
1518				struct iphdr *ih = (struct iphdr *) skb->data;
1519				/* Check that IP packet is at least 46 bytes, otherwise,
1520				 * there may be pad bytes included in the hardware checksum.
1521				 * This wouldn't happen if everyone padded with 0.
1522				 */
1523				if (ntohs(ih->tot_len) >= 46){
1524					/* don't worry about frags */
1525					if (!(ih->frag_off & cpu_to_be16(IP_MF|IP_OFFSET))) {
1526						u32 inv = *(u32 *) &buf_addr[data_size - 16];
1527						u32 *p = (u32 *) &buf_addr[data_size - 20];
1528						register u32 crc, p_r, p_r1;
1529
1530						if (inv & 4) {
1531							inv &= ~4;
1532							--p;
1533						}
1534						p_r = *p;
1535						p_r1 = *(p-1);
1536						switch (inv) {
1537							case 0:
1538								crc = (p_r & 0xffff) + (p_r >> 16);
1539								break;
1540							case 1:
1541								crc = (p_r >> 16) + (p_r & 0xffff)
1542									+ (p_r1 >> 16 & 0xff00);
1543								break;
1544							case 2:
1545								crc = p_r + (p_r1 >> 16);
1546								break;
1547							case 3:
1548								crc = p_r + (p_r1 & 0xff00) + (p_r1 >> 16);
1549								break;
1550							default:	/*NOTREACHED*/ crc = 0;
1551						}
1552						if (crc & 0xffff0000) {
1553							crc &= 0xffff;
1554							++crc;
1555						}
1556						/* tcp/udp will add in pseudo */
1557						skb->csum = ntohs(pfck & 0xffff);
1558						if (skb->csum > crc)
1559							skb->csum -= crc;
1560						else
1561							skb->csum += (~crc & 0xffff);
1562						/*
1563						* could do the pseudo myself and return
1564						* CHECKSUM_UNNECESSARY
1565						*/
1566						skb->ip_summed = CHECKSUM_COMPLETE;
1567					}
1568				}
1569			}
1570#endif  /* RX_CHECKSUM */
1571
1572			netif_rx(skb);
1573			dev->stats.rx_packets++;
1574		}
1575		entry = (++hmp->cur_rx) % RX_RING_SIZE;
1576	}
1577
1578	/* Refill the Rx ring buffers. */
1579	for (; hmp->cur_rx - hmp->dirty_rx > 0; hmp->dirty_rx++) {
1580		struct hamachi_desc *desc;
1581
1582		entry = hmp->dirty_rx % RX_RING_SIZE;
1583		desc = &(hmp->rx_ring[entry]);
1584		if (hmp->rx_skbuff[entry] == NULL) {
1585			struct sk_buff *skb = netdev_alloc_skb(dev, hmp->rx_buf_sz + 2);
1586
1587			hmp->rx_skbuff[entry] = skb;
1588			if (skb == NULL)
1589				break;		/* Better luck next round. */
1590			skb_reserve(skb, 2);	/* Align IP on 16 byte boundaries */
1591                	desc->addr = cpu_to_leXX(pci_map_single(hmp->pci_dev,
1592				skb->data, hmp->rx_buf_sz, PCI_DMA_FROMDEVICE));
1593		}
1594		desc->status_n_length = cpu_to_le32(hmp->rx_buf_sz);
1595		if (entry >= RX_RING_SIZE-1)
1596			desc->status_n_length |= cpu_to_le32(DescOwn |
1597				DescEndPacket | DescEndRing | DescIntr);
1598		else
1599			desc->status_n_length |= cpu_to_le32(DescOwn |
1600				DescEndPacket | DescIntr);
1601	}
1602
1603	/* Restart Rx engine if stopped. */
1604	/* If we don't need to check status, don't. -KDU */
1605	if (readw(hmp->base + RxStatus) & 0x0002)
1606		writew(0x0001, hmp->base + RxCmd);
1607
1608	return 0;
1609}
1610
1611/* This is more properly named "uncommon interrupt events", as it covers more
1612   than just errors. */
1613static void hamachi_error(struct net_device *dev, int intr_status)
1614{
1615	struct hamachi_private *hmp = netdev_priv(dev);
1616	void __iomem *ioaddr = hmp->base;
1617
1618	if (intr_status & (LinkChange|NegotiationChange)) {
1619		if (hamachi_debug > 1)
1620			printk(KERN_INFO "%s: Link changed: AutoNegotiation Ctrl"
1621				   " %4.4x, Status %4.4x %4.4x Intr status %4.4x.\n",
1622				   dev->name, readw(ioaddr + 0x0E0), readw(ioaddr + 0x0E2),
1623				   readw(ioaddr + ANLinkPartnerAbility),
1624				   readl(ioaddr + IntrStatus));
1625		if (readw(ioaddr + ANStatus) & 0x20)
1626			writeb(0x01, ioaddr + LEDCtrl);
1627		else
1628			writeb(0x03, ioaddr + LEDCtrl);
1629	}
1630	if (intr_status & StatsMax) {
1631		hamachi_get_stats(dev);
1632		/* Read the overflow bits to clear. */
1633		readl(ioaddr + 0x370);
1634		readl(ioaddr + 0x3F0);
1635	}
1636	if ((intr_status & ~(LinkChange|StatsMax|NegotiationChange|IntrRxDone|IntrTxDone)) &&
1637	    hamachi_debug)
1638		printk(KERN_ERR "%s: Something Wicked happened! %4.4x.\n",
1639		       dev->name, intr_status);
1640	/* Hmmmmm, it's not clear how to recover from PCI faults. */
1641	if (intr_status & (IntrTxPCIErr | IntrTxPCIFault))
1642		dev->stats.tx_fifo_errors++;
1643	if (intr_status & (IntrRxPCIErr | IntrRxPCIFault))
1644		dev->stats.rx_fifo_errors++;
1645}
1646
1647static int hamachi_close(struct net_device *dev)
1648{
1649	struct hamachi_private *hmp = netdev_priv(dev);
1650	void __iomem *ioaddr = hmp->base;
1651	struct sk_buff *skb;
1652	int i;
1653
1654	netif_stop_queue(dev);
1655
1656	if (hamachi_debug > 1) {
1657		printk(KERN_DEBUG "%s: Shutting down ethercard, status was Tx %4.4x Rx %4.4x Int %2.2x.\n",
1658			   dev->name, readw(ioaddr + TxStatus),
1659			   readw(ioaddr + RxStatus), readl(ioaddr + IntrStatus));
1660		printk(KERN_DEBUG "%s: Queue pointers were Tx %d / %d,  Rx %d / %d.\n",
1661			   dev->name, hmp->cur_tx, hmp->dirty_tx, hmp->cur_rx, hmp->dirty_rx);
1662	}
1663
1664	/* Disable interrupts by clearing the interrupt mask. */
1665	writel(0x0000, ioaddr + InterruptEnable);
1666
1667	/* Stop the chip's Tx and Rx processes. */
1668	writel(2, ioaddr + RxCmd);
1669	writew(2, ioaddr + TxCmd);
1670
1671#ifdef __i386__
1672	if (hamachi_debug > 2) {
1673		printk(KERN_DEBUG "  Tx ring at %8.8x:\n",
1674			   (int)hmp->tx_ring_dma);
1675		for (i = 0; i < TX_RING_SIZE; i++)
1676			printk(KERN_DEBUG " %c #%d desc. %8.8x %8.8x.\n",
1677				   readl(ioaddr + TxCurPtr) == (long)&hmp->tx_ring[i] ? '>' : ' ',
1678				   i, hmp->tx_ring[i].status_n_length, hmp->tx_ring[i].addr);
1679		printk(KERN_DEBUG "  Rx ring %8.8x:\n",
1680			   (int)hmp->rx_ring_dma);
1681		for (i = 0; i < RX_RING_SIZE; i++) {
1682			printk(KERN_DEBUG " %c #%d desc. %4.4x %8.8x\n",
1683				   readl(ioaddr + RxCurPtr) == (long)&hmp->rx_ring[i] ? '>' : ' ',
1684				   i, hmp->rx_ring[i].status_n_length, hmp->rx_ring[i].addr);
1685			if (hamachi_debug > 6) {
1686				if (*(u8*)hmp->rx_skbuff[i]->data != 0x69) {
1687					u16 *addr = (u16 *)
1688						hmp->rx_skbuff[i]->data;
1689					int j;
1690					printk(KERN_DEBUG "Addr: ");
1691					for (j = 0; j < 0x50; j++)
1692						printk(" %4.4x", addr[j]);
1693					printk("\n");
1694				}
1695			}
1696		}
1697	}
1698#endif /* __i386__ debugging only */
1699
1700	free_irq(hmp->pci_dev->irq, dev);
1701
1702	del_timer_sync(&hmp->timer);
1703
1704	/* Free all the skbuffs in the Rx queue. */
1705	for (i = 0; i < RX_RING_SIZE; i++) {
1706		skb = hmp->rx_skbuff[i];
1707		hmp->rx_ring[i].status_n_length = 0;
1708		if (skb) {
1709			pci_unmap_single(hmp->pci_dev,
1710				leXX_to_cpu(hmp->rx_ring[i].addr),
1711				hmp->rx_buf_sz, PCI_DMA_FROMDEVICE);
1712			dev_kfree_skb(skb);
1713			hmp->rx_skbuff[i] = NULL;
1714		}
1715		hmp->rx_ring[i].addr = cpu_to_leXX(0xBADF00D0); /* An invalid address. */
1716	}
1717	for (i = 0; i < TX_RING_SIZE; i++) {
1718		skb = hmp->tx_skbuff[i];
1719		if (skb) {
1720			pci_unmap_single(hmp->pci_dev,
1721				leXX_to_cpu(hmp->tx_ring[i].addr),
1722				skb->len, PCI_DMA_TODEVICE);
1723			dev_kfree_skb(skb);
1724			hmp->tx_skbuff[i] = NULL;
1725		}
1726	}
1727
1728	writeb(0x00, ioaddr + LEDCtrl);
1729
1730	return 0;
1731}
1732
1733static struct net_device_stats *hamachi_get_stats(struct net_device *dev)
1734{
1735	struct hamachi_private *hmp = netdev_priv(dev);
1736	void __iomem *ioaddr = hmp->base;
1737
1738	/* We should lock this segment of code for SMP eventually, although
1739	   the vulnerability window is very small and statistics are
1740	   non-critical. */
1741        /* Ok, what goes here?  This appears to be stuck at 21 packets
1742           according to ifconfig.  It does get incremented in hamachi_tx(),
1743           so I think I'll comment it out here and see if better things
1744           happen.
1745        */
1746	/* dev->stats.tx_packets	= readl(ioaddr + 0x000); */
1747
1748	/* Total Uni+Brd+Multi */
1749	dev->stats.rx_bytes = readl(ioaddr + 0x330);
1750	/* Total Uni+Brd+Multi */
1751	dev->stats.tx_bytes = readl(ioaddr + 0x3B0);
1752	/* Multicast Rx */
1753	dev->stats.multicast = readl(ioaddr + 0x320);
1754
1755	/* Over+Undersized */
1756	dev->stats.rx_length_errors = readl(ioaddr + 0x368);
1757	/* Jabber */
1758	dev->stats.rx_over_errors = readl(ioaddr + 0x35C);
1759	/* Jabber */
1760	dev->stats.rx_crc_errors = readl(ioaddr + 0x360);
1761	/* Symbol Errs */
1762	dev->stats.rx_frame_errors = readl(ioaddr + 0x364);
1763	/* Dropped */
1764	dev->stats.rx_missed_errors = readl(ioaddr + 0x36C);
1765
1766	return &dev->stats;
1767}
1768
1769static void set_rx_mode(struct net_device *dev)
1770{
1771	struct hamachi_private *hmp = netdev_priv(dev);
1772	void __iomem *ioaddr = hmp->base;
1773
1774	if (dev->flags & IFF_PROMISC) {			/* Set promiscuous. */
1775		writew(0x000F, ioaddr + AddrMode);
1776	} else if ((netdev_mc_count(dev) > 63) || (dev->flags & IFF_ALLMULTI)) {
1777		/* Too many to match, or accept all multicasts. */
1778		writew(0x000B, ioaddr + AddrMode);
1779	} else if (!netdev_mc_empty(dev)) { /* Must use the CAM filter. */
1780		struct netdev_hw_addr *ha;
1781		int i = 0;
1782
1783		netdev_for_each_mc_addr(ha, dev) {
1784			writel(*(u32 *)(ha->addr), ioaddr + 0x100 + i*8);
1785			writel(0x20000 | (*(u16 *)&ha->addr[4]),
1786				   ioaddr + 0x104 + i*8);
1787			i++;
1788		}
1789		/* Clear remaining entries. */
1790		for (; i < 64; i++)
1791			writel(0, ioaddr + 0x104 + i*8);
1792		writew(0x0003, ioaddr + AddrMode);
1793	} else {					/* Normal, unicast/broadcast-only mode. */
1794		writew(0x0001, ioaddr + AddrMode);
1795	}
1796}
1797
1798static int check_if_running(struct net_device *dev)
1799{
1800	if (!netif_running(dev))
1801		return -EINVAL;
1802	return 0;
1803}
1804
1805static void hamachi_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1806{
1807	struct hamachi_private *np = netdev_priv(dev);
1808
1809	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1810	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1811	strlcpy(info->bus_info, pci_name(np->pci_dev), sizeof(info->bus_info));
1812}
1813
1814static int hamachi_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
1815{
1816	struct hamachi_private *np = netdev_priv(dev);
1817	spin_lock_irq(&np->lock);
1818	mii_ethtool_gset(&np->mii_if, ecmd);
1819	spin_unlock_irq(&np->lock);
1820	return 0;
1821}
1822
1823static int hamachi_set_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
1824{
1825	struct hamachi_private *np = netdev_priv(dev);
1826	int res;
1827	spin_lock_irq(&np->lock);
1828	res = mii_ethtool_sset(&np->mii_if, ecmd);
1829	spin_unlock_irq(&np->lock);
1830	return res;
1831}
1832
1833static int hamachi_nway_reset(struct net_device *dev)
1834{
1835	struct hamachi_private *np = netdev_priv(dev);
1836	return mii_nway_restart(&np->mii_if);
1837}
1838
1839static u32 hamachi_get_link(struct net_device *dev)
1840{
1841	struct hamachi_private *np = netdev_priv(dev);
1842	return mii_link_ok(&np->mii_if);
1843}
1844
1845static const struct ethtool_ops ethtool_ops = {
1846	.begin = check_if_running,
1847	.get_drvinfo = hamachi_get_drvinfo,
1848	.get_settings = hamachi_get_settings,
1849	.set_settings = hamachi_set_settings,
1850	.nway_reset = hamachi_nway_reset,
1851	.get_link = hamachi_get_link,
1852};
1853
1854static const struct ethtool_ops ethtool_ops_no_mii = {
1855	.begin = check_if_running,
1856	.get_drvinfo = hamachi_get_drvinfo,
1857};
1858
1859static int netdev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
1860{
1861	struct hamachi_private *np = netdev_priv(dev);
1862	struct mii_ioctl_data *data = if_mii(rq);
1863	int rc;
1864
1865	if (!netif_running(dev))
1866		return -EINVAL;
1867
1868	if (cmd == (SIOCDEVPRIVATE+3)) { /* set rx,tx intr params */
1869		u32 *d = (u32 *)&rq->ifr_ifru;
1870		/* Should add this check here or an ordinary user can do nasty
1871		 * things. -KDU
1872		 *
1873		 * TODO: Shut down the Rx and Tx engines while doing this.
1874		 */
1875		if (!capable(CAP_NET_ADMIN))
1876			return -EPERM;
1877		writel(d[0], np->base + TxIntrCtrl);
1878		writel(d[1], np->base + RxIntrCtrl);
1879		printk(KERN_NOTICE "%s: tx %08x, rx %08x intr\n", dev->name,
1880		  (u32) readl(np->base + TxIntrCtrl),
1881		  (u32) readl(np->base + RxIntrCtrl));
1882		rc = 0;
1883	}
1884
1885	else {
1886		spin_lock_irq(&np->lock);
1887		rc = generic_mii_ioctl(&np->mii_if, data, cmd, NULL);
1888		spin_unlock_irq(&np->lock);
1889	}
1890
1891	return rc;
1892}
1893
1894
1895static void hamachi_remove_one(struct pci_dev *pdev)
1896{
1897	struct net_device *dev = pci_get_drvdata(pdev);
1898
1899	if (dev) {
1900		struct hamachi_private *hmp = netdev_priv(dev);
1901
1902		pci_free_consistent(pdev, RX_TOTAL_SIZE, hmp->rx_ring,
1903			hmp->rx_ring_dma);
1904		pci_free_consistent(pdev, TX_TOTAL_SIZE, hmp->tx_ring,
1905			hmp->tx_ring_dma);
1906		unregister_netdev(dev);
1907		iounmap(hmp->base);
1908		free_netdev(dev);
1909		pci_release_regions(pdev);
1910	}
1911}
1912
1913static const struct pci_device_id hamachi_pci_tbl[] = {
1914	{ 0x1318, 0x0911, PCI_ANY_ID, PCI_ANY_ID, },
1915	{ 0, }
1916};
1917MODULE_DEVICE_TABLE(pci, hamachi_pci_tbl);
1918
1919static struct pci_driver hamachi_driver = {
1920	.name		= DRV_NAME,
1921	.id_table	= hamachi_pci_tbl,
1922	.probe		= hamachi_init_one,
1923	.remove		= hamachi_remove_one,
1924};
1925
1926static int __init hamachi_init (void)
1927{
1928/* when a module, this is printed whether or not devices are found in probe */
1929#ifdef MODULE
1930	printk(version);
1931#endif
1932	return pci_register_driver(&hamachi_driver);
1933}
1934
1935static void __exit hamachi_exit (void)
1936{
1937	pci_unregister_driver(&hamachi_driver);
1938}
1939
1940
1941module_init(hamachi_init);
1942module_exit(hamachi_exit);