1 /* yellowfin.c: A Packet Engines G-NIC ethernet driver for linux. */
2 /*
3 	Written 1997-2001 by Donald Becker.
4 
5 	This software may be used and distributed according to the terms of
6 	the GNU General Public License (GPL), incorporated herein by reference.
7 	Drivers based on or derived from this code fall under the GPL and must
8 	retain the authorship, copyright and license notice.  This file is not
9 	a complete program and may only be used when the entire operating
10 	system is licensed under the GPL.
11 
12 	This driver is for the Packet Engines G-NIC PCI Gigabit Ethernet adapter.
13 	It also supports the Symbios Logic version of the same chip core.
14 
15 	The author may be reached as becker@scyld.com, or C/O
16 	Scyld Computing Corporation
17 	410 Severn Ave., Suite 210
18 	Annapolis MD 21403
19 
20 	Support and updates available at
21 	http://www.scyld.com/network/yellowfin.html
22 	[link no longer provides useful info -jgarzik]
23 
24 */
25 
26 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
27 
28 #define DRV_NAME	"yellowfin"
29 #define DRV_VERSION	"2.1"
30 #define DRV_RELDATE	"Sep 11, 2006"
31 
32 /* The user-configurable values.
33    These may be modified when a driver module is loaded.*/
34 
35 static int debug = 1;			/* 1 normal messages, 0 quiet .. 7 verbose. */
36 /* Maximum events (Rx packets, etc.) to handle at each interrupt. */
37 static int max_interrupt_work = 20;
38 static int mtu;
39 #ifdef YF_PROTOTYPE			/* Support for prototype hardware errata. */
40 /* System-wide count of bogus-rx frames. */
41 static int bogus_rx;
42 static int dma_ctrl = 0x004A0263; 			/* Constrained by errata */
43 static int fifo_cfg = 0x0020;				/* Bypass external Tx FIFO. */
44 #elif defined(YF_NEW)					/* A future perfect board :->.  */
45 static int dma_ctrl = 0x00CAC277;			/* Override when loading module! */
46 static int fifo_cfg = 0x0028;
47 #else
48 static const int dma_ctrl = 0x004A0263; 			/* Constrained by errata */
49 static const int fifo_cfg = 0x0020;				/* Bypass external Tx FIFO. */
50 #endif
51 
52 /* Set the copy breakpoint for the copy-only-tiny-frames scheme.
53    Setting to > 1514 effectively disables this feature. */
54 static int rx_copybreak;
55 
56 /* Used to pass the media type, etc.
57    No media types are currently defined.  These exist for driver
58    interoperability.
59 */
60 #define MAX_UNITS 8				/* More are supported, limit only on options */
61 static int options[MAX_UNITS] = {-1, -1, -1, -1, -1, -1, -1, -1};
62 static int full_duplex[MAX_UNITS] = {-1, -1, -1, -1, -1, -1, -1, -1};
63 
64 /* Do ugly workaround for GX server chipset errata. */
65 static int gx_fix;
66 
67 /* Operational parameters that are set at compile time. */
68 
69 /* Keep the ring sizes a power of two for efficiency.
70    Making the Tx ring too long decreases the effectiveness of channel
71    bonding and packet priority.
72    There are no ill effects from too-large receive rings. */
73 #define TX_RING_SIZE	16
74 #define TX_QUEUE_SIZE	12		/* Must be > 4 && <= TX_RING_SIZE */
75 #define RX_RING_SIZE	64
76 #define STATUS_TOTAL_SIZE	TX_RING_SIZE*sizeof(struct tx_status_words)
77 #define TX_TOTAL_SIZE		2*TX_RING_SIZE*sizeof(struct yellowfin_desc)
78 #define RX_TOTAL_SIZE		RX_RING_SIZE*sizeof(struct yellowfin_desc)
79 
80 /* Operational parameters that usually are not changed. */
81 /* Time in jiffies before concluding the transmitter is hung. */
82 #define TX_TIMEOUT  (2*HZ)
83 #define PKT_BUF_SZ		1536			/* Size of each temporary Rx buffer.*/
84 
85 #define yellowfin_debug debug
86 
87 #include <linux/module.h>
88 #include <linux/kernel.h>
89 #include <linux/string.h>
90 #include <linux/timer.h>
91 #include <linux/errno.h>
92 #include <linux/ioport.h>
93 #include <linux/interrupt.h>
94 #include <linux/pci.h>
95 #include <linux/init.h>
96 #include <linux/mii.h>
97 #include <linux/netdevice.h>
98 #include <linux/etherdevice.h>
99 #include <linux/skbuff.h>
100 #include <linux/ethtool.h>
101 #include <linux/crc32.h>
102 #include <linux/bitops.h>
103 #include <asm/uaccess.h>
104 #include <asm/processor.h>		/* Processor type for cache alignment. */
105 #include <asm/unaligned.h>
106 #include <asm/io.h>
107 
108 /* These identify the driver base version and may not be removed. */
109 static const char version[] =
110   KERN_INFO DRV_NAME ".c:v1.05  1/09/2001  Written by Donald Becker <becker@scyld.com>\n"
111   "  (unofficial 2.4.x port, " DRV_VERSION ", " DRV_RELDATE ")\n";
112 
113 MODULE_AUTHOR("Donald Becker <becker@scyld.com>");
114 MODULE_DESCRIPTION("Packet Engines Yellowfin G-NIC Gigabit Ethernet driver");
115 MODULE_LICENSE("GPL");
116 
117 module_param(max_interrupt_work, int, 0);
118 module_param(mtu, int, 0);
119 module_param(debug, int, 0);
120 module_param(rx_copybreak, int, 0);
121 module_param_array(options, int, NULL, 0);
122 module_param_array(full_duplex, int, NULL, 0);
123 module_param(gx_fix, int, 0);
124 MODULE_PARM_DESC(max_interrupt_work, "G-NIC maximum events handled per interrupt");
125 MODULE_PARM_DESC(mtu, "G-NIC MTU (all boards)");
126 MODULE_PARM_DESC(debug, "G-NIC debug level (0-7)");
127 MODULE_PARM_DESC(rx_copybreak, "G-NIC copy breakpoint for copy-only-tiny-frames");
128 MODULE_PARM_DESC(options, "G-NIC: Bits 0-3: media type, bit 17: full duplex");
129 MODULE_PARM_DESC(full_duplex, "G-NIC full duplex setting(s) (1)");
130 MODULE_PARM_DESC(gx_fix, "G-NIC: enable GX server chipset bug workaround (0-1)");
131 
132 /*
133 				Theory of Operation
134 
135 I. Board Compatibility
136 
137 This device driver is designed for the Packet Engines "Yellowfin" Gigabit
138 Ethernet adapter.  The G-NIC 64-bit PCI card is supported, as well as the
139 Symbios 53C885E dual function chip.
140 
141 II. Board-specific settings
142 
143 PCI bus devices are configured by the system at boot time, so no jumpers
144 need to be set on the board.  The system BIOS preferably should assign the
145 PCI INTA signal to an otherwise unused system IRQ line.
146 Note: Kernel versions earlier than 1.3.73 do not support shared PCI
147 interrupt lines.
148 
149 III. Driver operation
150 
151 IIIa. Ring buffers
152 
153 The Yellowfin uses the Descriptor Based DMA Architecture specified by Apple.
154 This is a descriptor list scheme similar to that used by the EEPro100 and
155 Tulip.  This driver uses two statically allocated fixed-size descriptor lists
156 formed into rings by a branch from the final descriptor to the beginning of
157 the list.  The ring sizes are set at compile time by RX/TX_RING_SIZE.
158 
159 The driver allocates full frame size skbuffs for the Rx ring buffers at
160 open() time and passes the skb->data field to the Yellowfin as receive data
161 buffers.  When an incoming frame is less than RX_COPYBREAK bytes long,
162 a fresh skbuff is allocated and the frame is copied to the new skbuff.
163 When the incoming frame is larger, the skbuff is passed directly up the
164 protocol stack and replaced by a newly allocated skbuff.
165 
166 The RX_COPYBREAK value is chosen to trade-off the memory wasted by
167 using a full-sized skbuff for small frames vs. the copying costs of larger
168 frames.  For small frames the copying cost is negligible (esp. considering
169 that we are pre-loading the cache with immediately useful header
170 information).  For large frames the copying cost is non-trivial, and the
171 larger copy might flush the cache of useful data.
172 
173 IIIC. Synchronization
174 
175 The driver runs as two independent, single-threaded flows of control.  One
176 is the send-packet routine, which enforces single-threaded use by the
177 dev->tbusy flag.  The other thread is the interrupt handler, which is single
178 threaded by the hardware and other software.
179 
180 The send packet thread has partial control over the Tx ring and 'dev->tbusy'
181 flag.  It sets the tbusy flag whenever it's queuing a Tx packet. If the next
182 queue slot is empty, it clears the tbusy flag when finished otherwise it sets
183 the 'yp->tx_full' flag.
184 
185 The interrupt handler has exclusive control over the Rx ring and records stats
186 from the Tx ring.  After reaping the stats, it marks the Tx queue entry as
187 empty by incrementing the dirty_tx mark. Iff the 'yp->tx_full' flag is set, it
188 clears both the tx_full and tbusy flags.
189 
190 IV. Notes
191 
192 Thanks to Kim Stearns of Packet Engines for providing a pair of G-NIC boards.
193 Thanks to Bruce Faust of Digitalscape for providing both their SYM53C885 board
194 and an AlphaStation to verifty the Alpha port!
195 
196 IVb. References
197 
198 Yellowfin Engineering Design Specification, 4/23/97 Preliminary/Confidential
199 Symbios SYM53C885 PCI-SCSI/Fast Ethernet Multifunction Controller Preliminary
200    Data Manual v3.0
201 http://cesdis.gsfc.nasa.gov/linux/misc/NWay.html
202 http://cesdis.gsfc.nasa.gov/linux/misc/100mbps.html
203 
204 IVc. Errata
205 
206 See Packet Engines confidential appendix (prototype chips only).
207 */
208 
209 
210 
211 enum capability_flags {
212 	HasMII=1, FullTxStatus=2, IsGigabit=4, HasMulticastBug=8, FullRxStatus=16,
213 	HasMACAddrBug=32, /* Only on early revs.  */
214 	DontUseEeprom=64, /* Don't read the MAC from the EEPROm. */
215 };
216 
217 /* The PCI I/O space extent. */
218 enum {
219 	YELLOWFIN_SIZE	= 0x100,
220 };
221 
222 struct pci_id_info {
223         const char *name;
224         struct match_info {
225                 int     pci, pci_mask, subsystem, subsystem_mask;
226                 int revision, revision_mask;                            /* Only 8 bits. */
227         } id;
228         int drv_flags;                          /* Driver use, intended as capability flags. */
229 };
230 
231 static const struct pci_id_info pci_id_tbl[] = {
232 	{"Yellowfin G-NIC Gigabit Ethernet", { 0x07021000, 0xffffffff},
233 	 FullTxStatus | IsGigabit | HasMulticastBug | HasMACAddrBug | DontUseEeprom},
234 	{"Symbios SYM83C885", { 0x07011000, 0xffffffff},
235 	  HasMII | DontUseEeprom },
236 	{ }
237 };
238 
239 static const struct pci_device_id yellowfin_pci_tbl[] = {
240 	{ 0x1000, 0x0702, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 },
241 	{ 0x1000, 0x0701, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 1 },
242 	{ }
243 };
244 MODULE_DEVICE_TABLE (pci, yellowfin_pci_tbl);
245 
246 
247 /* Offsets to the Yellowfin registers.  Various sizes and alignments. */
248 enum yellowfin_offsets {
249 	TxCtrl=0x00, TxStatus=0x04, TxPtr=0x0C,
250 	TxIntrSel=0x10, TxBranchSel=0x14, TxWaitSel=0x18,
251 	RxCtrl=0x40, RxStatus=0x44, RxPtr=0x4C,
252 	RxIntrSel=0x50, RxBranchSel=0x54, RxWaitSel=0x58,
253 	EventStatus=0x80, IntrEnb=0x82, IntrClear=0x84, IntrStatus=0x86,
254 	ChipRev=0x8C, DMACtrl=0x90, TxThreshold=0x94,
255 	Cnfg=0xA0, FrameGap0=0xA2, FrameGap1=0xA4,
256 	MII_Cmd=0xA6, MII_Addr=0xA8, MII_Wr_Data=0xAA, MII_Rd_Data=0xAC,
257 	MII_Status=0xAE,
258 	RxDepth=0xB8, FlowCtrl=0xBC,
259 	AddrMode=0xD0, StnAddr=0xD2, HashTbl=0xD8, FIFOcfg=0xF8,
260 	EEStatus=0xF0, EECtrl=0xF1, EEAddr=0xF2, EERead=0xF3, EEWrite=0xF4,
261 	EEFeature=0xF5,
262 };
263 
264 /* The Yellowfin Rx and Tx buffer descriptors.
265    Elements are written as 32 bit for endian portability. */
266 struct yellowfin_desc {
267 	__le32 dbdma_cmd;
268 	__le32 addr;
269 	__le32 branch_addr;
270 	__le32 result_status;
271 };
272 
273 struct tx_status_words {
274 #ifdef __BIG_ENDIAN
275 	u16 tx_errs;
276 	u16 tx_cnt;
277 	u16 paused;
278 	u16 total_tx_cnt;
279 #else  /* Little endian chips. */
280 	u16 tx_cnt;
281 	u16 tx_errs;
282 	u16 total_tx_cnt;
283 	u16 paused;
284 #endif /* __BIG_ENDIAN */
285 };
286 
287 /* Bits in yellowfin_desc.cmd */
288 enum desc_cmd_bits {
289 	CMD_TX_PKT=0x10000000, CMD_RX_BUF=0x20000000, CMD_TXSTATUS=0x30000000,
290 	CMD_NOP=0x60000000, CMD_STOP=0x70000000,
291 	BRANCH_ALWAYS=0x0C0000, INTR_ALWAYS=0x300000, WAIT_ALWAYS=0x030000,
292 	BRANCH_IFTRUE=0x040000,
293 };
294 
295 /* Bits in yellowfin_desc.status */
296 enum desc_status_bits { RX_EOP=0x0040, };
297 
298 /* Bits in the interrupt status/mask registers. */
299 enum intr_status_bits {
300 	IntrRxDone=0x01, IntrRxInvalid=0x02, IntrRxPCIFault=0x04,IntrRxPCIErr=0x08,
301 	IntrTxDone=0x10, IntrTxInvalid=0x20, IntrTxPCIFault=0x40,IntrTxPCIErr=0x80,
302 	IntrEarlyRx=0x100, IntrWakeup=0x200, };
303 
304 #define PRIV_ALIGN	31 	/* Required alignment mask */
305 #define MII_CNT		4
306 struct yellowfin_private {
307 	/* Descriptor rings first for alignment.
308 	   Tx requires a second descriptor for status. */
309 	struct yellowfin_desc *rx_ring;
310 	struct yellowfin_desc *tx_ring;
311 	struct sk_buff* rx_skbuff[RX_RING_SIZE];
312 	struct sk_buff* tx_skbuff[TX_RING_SIZE];
313 	dma_addr_t rx_ring_dma;
314 	dma_addr_t tx_ring_dma;
315 
316 	struct tx_status_words *tx_status;
317 	dma_addr_t tx_status_dma;
318 
319 	struct timer_list timer;	/* Media selection timer. */
320 	/* Frequently used and paired value: keep adjacent for cache effect. */
321 	int chip_id, drv_flags;
322 	struct pci_dev *pci_dev;
323 	unsigned int cur_rx, dirty_rx;		/* Producer/consumer ring indices */
324 	unsigned int rx_buf_sz;				/* Based on MTU+slack. */
325 	struct tx_status_words *tx_tail_desc;
326 	unsigned int cur_tx, dirty_tx;
327 	int tx_threshold;
328 	unsigned int tx_full:1;				/* The Tx queue is full. */
329 	unsigned int full_duplex:1;			/* Full-duplex operation requested. */
330 	unsigned int duplex_lock:1;
331 	unsigned int medialock:1;			/* Do not sense media. */
332 	unsigned int default_port:4;		/* Last dev->if_port value. */
333 	/* MII transceiver section. */
334 	int mii_cnt;						/* MII device addresses. */
335 	u16 advertising;					/* NWay media advertisement */
336 	unsigned char phys[MII_CNT];		/* MII device addresses, only first one used */
337 	spinlock_t lock;
338 	void __iomem *base;
339 };
340 
341 static int read_eeprom(void __iomem *ioaddr, int location);
342 static int mdio_read(void __iomem *ioaddr, int phy_id, int location);
343 static void mdio_write(void __iomem *ioaddr, int phy_id, int location, int value);
344 static int netdev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd);
345 static int yellowfin_open(struct net_device *dev);
346 static void yellowfin_timer(unsigned long data);
347 static void yellowfin_tx_timeout(struct net_device *dev);
348 static int yellowfin_init_ring(struct net_device *dev);
349 static netdev_tx_t yellowfin_start_xmit(struct sk_buff *skb,
350 					struct net_device *dev);
351 static irqreturn_t yellowfin_interrupt(int irq, void *dev_instance);
352 static int yellowfin_rx(struct net_device *dev);
353 static void yellowfin_error(struct net_device *dev, int intr_status);
354 static int yellowfin_close(struct net_device *dev);
355 static void set_rx_mode(struct net_device *dev);
356 static const struct ethtool_ops ethtool_ops;
357 
358 static const struct net_device_ops netdev_ops = {
359 	.ndo_open 		= yellowfin_open,
360 	.ndo_stop 		= yellowfin_close,
361 	.ndo_start_xmit 	= yellowfin_start_xmit,
362 	.ndo_set_rx_mode	= set_rx_mode,
363 	.ndo_change_mtu		= eth_change_mtu,
364 	.ndo_validate_addr	= eth_validate_addr,
365 	.ndo_set_mac_address 	= eth_mac_addr,
366 	.ndo_do_ioctl 		= netdev_ioctl,
367 	.ndo_tx_timeout 	= yellowfin_tx_timeout,
368 };
369 
370 static int yellowfin_init_one(struct pci_dev *pdev,
371 			      const struct pci_device_id *ent)
372 {
373 	struct net_device *dev;
374 	struct yellowfin_private *np;
375 	int irq;
376 	int chip_idx = ent->driver_data;
377 	static int find_cnt;
378 	void __iomem *ioaddr;
379 	int i, option = find_cnt < MAX_UNITS ? options[find_cnt] : 0;
380 	int drv_flags = pci_id_tbl[chip_idx].drv_flags;
381         void *ring_space;
382         dma_addr_t ring_dma;
383 #ifdef USE_IO_OPS
384 	int bar = 0;
385 #else
386 	int bar = 1;
387 #endif
388 
389 /* when built into the kernel, we only print version if device is found */
390 #ifndef MODULE
391 	static int printed_version;
392 	if (!printed_version++)
393 		printk(version);
394 #endif
395 
396 	i = pci_enable_device(pdev);
397 	if (i) return i;
398 
399 	dev = alloc_etherdev(sizeof(*np));
400 	if (!dev)
401 		return -ENOMEM;
402 
403 	SET_NETDEV_DEV(dev, &pdev->dev);
404 
405 	np = netdev_priv(dev);
406 
407 	if (pci_request_regions(pdev, DRV_NAME))
408 		goto err_out_free_netdev;
409 
410 	pci_set_master (pdev);
411 
412 	ioaddr = pci_iomap(pdev, bar, YELLOWFIN_SIZE);
413 	if (!ioaddr)
414 		goto err_out_free_res;
415 
416 	irq = pdev->irq;
417 
418 	if (drv_flags & DontUseEeprom)
419 		for (i = 0; i < 6; i++)
420 			dev->dev_addr[i] = ioread8(ioaddr + StnAddr + i);
421 	else {
422 		int ee_offset = (read_eeprom(ioaddr, 6) == 0xff ? 0x100 : 0);
423 		for (i = 0; i < 6; i++)
424 			dev->dev_addr[i] = read_eeprom(ioaddr, ee_offset + i);
425 	}
426 
427 	/* Reset the chip. */
428 	iowrite32(0x80000000, ioaddr + DMACtrl);
429 
430 	pci_set_drvdata(pdev, dev);
431 	spin_lock_init(&np->lock);
432 
433 	np->pci_dev = pdev;
434 	np->chip_id = chip_idx;
435 	np->drv_flags = drv_flags;
436 	np->base = ioaddr;
437 
438 	ring_space = pci_alloc_consistent(pdev, TX_TOTAL_SIZE, &ring_dma);
439 	if (!ring_space)
440 		goto err_out_cleardev;
441 	np->tx_ring = ring_space;
442 	np->tx_ring_dma = ring_dma;
443 
444 	ring_space = pci_alloc_consistent(pdev, RX_TOTAL_SIZE, &ring_dma);
445 	if (!ring_space)
446 		goto err_out_unmap_tx;
447 	np->rx_ring = ring_space;
448 	np->rx_ring_dma = ring_dma;
449 
450 	ring_space = pci_alloc_consistent(pdev, STATUS_TOTAL_SIZE, &ring_dma);
451 	if (!ring_space)
452 		goto err_out_unmap_rx;
453 	np->tx_status = ring_space;
454 	np->tx_status_dma = ring_dma;
455 
456 	if (dev->mem_start)
457 		option = dev->mem_start;
458 
459 	/* The lower four bits are the media type. */
460 	if (option > 0) {
461 		if (option & 0x200)
462 			np->full_duplex = 1;
463 		np->default_port = option & 15;
464 		if (np->default_port)
465 			np->medialock = 1;
466 	}
467 	if (find_cnt < MAX_UNITS  &&  full_duplex[find_cnt] > 0)
468 		np->full_duplex = 1;
469 
470 	if (np->full_duplex)
471 		np->duplex_lock = 1;
472 
473 	/* The Yellowfin-specific entries in the device structure. */
474 	dev->netdev_ops = &netdev_ops;
475 	dev->ethtool_ops = &ethtool_ops;
476 	dev->watchdog_timeo = TX_TIMEOUT;
477 
478 	if (mtu)
479 		dev->mtu = mtu;
480 
481 	i = register_netdev(dev);
482 	if (i)
483 		goto err_out_unmap_status;
484 
485 	netdev_info(dev, "%s type %8x at %p, %pM, IRQ %d\n",
486 		    pci_id_tbl[chip_idx].name,
487 		    ioread32(ioaddr + ChipRev), ioaddr,
488 		    dev->dev_addr, irq);
489 
490 	if (np->drv_flags & HasMII) {
491 		int phy, phy_idx = 0;
492 		for (phy = 0; phy < 32 && phy_idx < MII_CNT; phy++) {
493 			int mii_status = mdio_read(ioaddr, phy, 1);
494 			if (mii_status != 0xffff  &&  mii_status != 0x0000) {
495 				np->phys[phy_idx++] = phy;
496 				np->advertising = mdio_read(ioaddr, phy, 4);
497 				netdev_info(dev, "MII PHY found at address %d, status 0x%04x advertising %04x\n",
498 					    phy, mii_status, np->advertising);
499 			}
500 		}
501 		np->mii_cnt = phy_idx;
502 	}
503 
504 	find_cnt++;
505 
506 	return 0;
507 
508 err_out_unmap_status:
509         pci_free_consistent(pdev, STATUS_TOTAL_SIZE, np->tx_status,
510 		np->tx_status_dma);
511 err_out_unmap_rx:
512         pci_free_consistent(pdev, RX_TOTAL_SIZE, np->rx_ring, np->rx_ring_dma);
513 err_out_unmap_tx:
514         pci_free_consistent(pdev, TX_TOTAL_SIZE, np->tx_ring, np->tx_ring_dma);
515 err_out_cleardev:
516 	pci_iounmap(pdev, ioaddr);
517 err_out_free_res:
518 	pci_release_regions(pdev);
519 err_out_free_netdev:
520 	free_netdev (dev);
521 	return -ENODEV;
522 }
523 
524 static int read_eeprom(void __iomem *ioaddr, int location)
525 {
526 	int bogus_cnt = 10000;		/* Typical 33Mhz: 1050 ticks */
527 
528 	iowrite8(location, ioaddr + EEAddr);
529 	iowrite8(0x30 | ((location >> 8) & 7), ioaddr + EECtrl);
530 	while ((ioread8(ioaddr + EEStatus) & 0x80)  &&  --bogus_cnt > 0)
531 		;
532 	return ioread8(ioaddr + EERead);
533 }
534 
535 /* MII Managemen Data I/O accesses.
536    These routines assume the MDIO controller is idle, and do not exit until
537    the command is finished. */
538 
539 static int mdio_read(void __iomem *ioaddr, int phy_id, int location)
540 {
541 	int i;
542 
543 	iowrite16((phy_id<<8) + location, ioaddr + MII_Addr);
544 	iowrite16(1, ioaddr + MII_Cmd);
545 	for (i = 10000; i >= 0; i--)
546 		if ((ioread16(ioaddr + MII_Status) & 1) == 0)
547 			break;
548 	return ioread16(ioaddr + MII_Rd_Data);
549 }
550 
551 static void mdio_write(void __iomem *ioaddr, int phy_id, int location, int value)
552 {
553 	int i;
554 
555 	iowrite16((phy_id<<8) + location, ioaddr + MII_Addr);
556 	iowrite16(value, ioaddr + MII_Wr_Data);
557 
558 	/* Wait for the command to finish. */
559 	for (i = 10000; i >= 0; i--)
560 		if ((ioread16(ioaddr + MII_Status) & 1) == 0)
561 			break;
562 }
563 
564 
565 static int yellowfin_open(struct net_device *dev)
566 {
567 	struct yellowfin_private *yp = netdev_priv(dev);
568 	const int irq = yp->pci_dev->irq;
569 	void __iomem *ioaddr = yp->base;
570 	int i, rc;
571 
572 	/* Reset the chip. */
573 	iowrite32(0x80000000, ioaddr + DMACtrl);
574 
575 	rc = request_irq(irq, yellowfin_interrupt, IRQF_SHARED, dev->name, dev);
576 	if (rc)
577 		return rc;
578 
579 	rc = yellowfin_init_ring(dev);
580 	if (rc < 0)
581 		goto err_free_irq;
582 
583 	iowrite32(yp->rx_ring_dma, ioaddr + RxPtr);
584 	iowrite32(yp->tx_ring_dma, ioaddr + TxPtr);
585 
586 	for (i = 0; i < 6; i++)
587 		iowrite8(dev->dev_addr[i], ioaddr + StnAddr + i);
588 
589 	/* Set up various condition 'select' registers.
590 	   There are no options here. */
591 	iowrite32(0x00800080, ioaddr + TxIntrSel); 	/* Interrupt on Tx abort */
592 	iowrite32(0x00800080, ioaddr + TxBranchSel);	/* Branch on Tx abort */
593 	iowrite32(0x00400040, ioaddr + TxWaitSel); 	/* Wait on Tx status */
594 	iowrite32(0x00400040, ioaddr + RxIntrSel);	/* Interrupt on Rx done */
595 	iowrite32(0x00400040, ioaddr + RxBranchSel);	/* Branch on Rx error */
596 	iowrite32(0x00400040, ioaddr + RxWaitSel);	/* Wait on Rx done */
597 
598 	/* Initialize other registers: with so many this eventually this will
599 	   converted to an offset/value list. */
600 	iowrite32(dma_ctrl, ioaddr + DMACtrl);
601 	iowrite16(fifo_cfg, ioaddr + FIFOcfg);
602 	/* Enable automatic generation of flow control frames, period 0xffff. */
603 	iowrite32(0x0030FFFF, ioaddr + FlowCtrl);
604 
605 	yp->tx_threshold = 32;
606 	iowrite32(yp->tx_threshold, ioaddr + TxThreshold);
607 
608 	if (dev->if_port == 0)
609 		dev->if_port = yp->default_port;
610 
611 	netif_start_queue(dev);
612 
613 	/* Setting the Rx mode will start the Rx process. */
614 	if (yp->drv_flags & IsGigabit) {
615 		/* We are always in full-duplex mode with gigabit! */
616 		yp->full_duplex = 1;
617 		iowrite16(0x01CF, ioaddr + Cnfg);
618 	} else {
619 		iowrite16(0x0018, ioaddr + FrameGap0); /* 0060/4060 for non-MII 10baseT */
620 		iowrite16(0x1018, ioaddr + FrameGap1);
621 		iowrite16(0x101C | (yp->full_duplex ? 2 : 0), ioaddr + Cnfg);
622 	}
623 	set_rx_mode(dev);
624 
625 	/* Enable interrupts by setting the interrupt mask. */
626 	iowrite16(0x81ff, ioaddr + IntrEnb);			/* See enum intr_status_bits */
627 	iowrite16(0x0000, ioaddr + EventStatus);		/* Clear non-interrupting events */
628 	iowrite32(0x80008000, ioaddr + RxCtrl);		/* Start Rx and Tx channels. */
629 	iowrite32(0x80008000, ioaddr + TxCtrl);
630 
631 	if (yellowfin_debug > 2) {
632 		netdev_printk(KERN_DEBUG, dev, "Done %s()\n", __func__);
633 	}
634 
635 	/* Set the timer to check for link beat. */
636 	init_timer(&yp->timer);
637 	yp->timer.expires = jiffies + 3*HZ;
638 	yp->timer.data = (unsigned long)dev;
639 	yp->timer.function = yellowfin_timer;				/* timer handler */
640 	add_timer(&yp->timer);
641 out:
642 	return rc;
643 
644 err_free_irq:
645 	free_irq(irq, dev);
646 	goto out;
647 }
648 
649 static void yellowfin_timer(unsigned long data)
650 {
651 	struct net_device *dev = (struct net_device *)data;
652 	struct yellowfin_private *yp = netdev_priv(dev);
653 	void __iomem *ioaddr = yp->base;
654 	int next_tick = 60*HZ;
655 
656 	if (yellowfin_debug > 3) {
657 		netdev_printk(KERN_DEBUG, dev, "Yellowfin timer tick, status %08x\n",
658 			      ioread16(ioaddr + IntrStatus));
659 	}
660 
661 	if (yp->mii_cnt) {
662 		int bmsr = mdio_read(ioaddr, yp->phys[0], MII_BMSR);
663 		int lpa = mdio_read(ioaddr, yp->phys[0], MII_LPA);
664 		int negotiated = lpa & yp->advertising;
665 		if (yellowfin_debug > 1)
666 			netdev_printk(KERN_DEBUG, dev, "MII #%d status register is %04x, link partner capability %04x\n",
667 				      yp->phys[0], bmsr, lpa);
668 
669 		yp->full_duplex = mii_duplex(yp->duplex_lock, negotiated);
670 
671 		iowrite16(0x101C | (yp->full_duplex ? 2 : 0), ioaddr + Cnfg);
672 
673 		if (bmsr & BMSR_LSTATUS)
674 			next_tick = 60*HZ;
675 		else
676 			next_tick = 3*HZ;
677 	}
678 
679 	yp->timer.expires = jiffies + next_tick;
680 	add_timer(&yp->timer);
681 }
682 
683 static void yellowfin_tx_timeout(struct net_device *dev)
684 {
685 	struct yellowfin_private *yp = netdev_priv(dev);
686 	void __iomem *ioaddr = yp->base;
687 
688 	netdev_warn(dev, "Yellowfin transmit timed out at %d/%d Tx status %04x, Rx status %04x, resetting...\n",
689 		    yp->cur_tx, yp->dirty_tx,
690 		    ioread32(ioaddr + TxStatus),
691 		    ioread32(ioaddr + RxStatus));
692 
693 	/* Note: these should be KERN_DEBUG. */
694 	if (yellowfin_debug) {
695 		int i;
696 		pr_warn("  Rx ring %p: ", yp->rx_ring);
697 		for (i = 0; i < RX_RING_SIZE; i++)
698 			pr_cont(" %08x", yp->rx_ring[i].result_status);
699 		pr_cont("\n");
700 		pr_warn("  Tx ring %p: ", yp->tx_ring);
701 		for (i = 0; i < TX_RING_SIZE; i++)
702 			pr_cont(" %04x /%08x",
703 			       yp->tx_status[i].tx_errs,
704 			       yp->tx_ring[i].result_status);
705 		pr_cont("\n");
706 	}
707 
708 	/* If the hardware is found to hang regularly, we will update the code
709 	   to reinitialize the chip here. */
710 	dev->if_port = 0;
711 
712 	/* Wake the potentially-idle transmit channel. */
713 	iowrite32(0x10001000, yp->base + TxCtrl);
714 	if (yp->cur_tx - yp->dirty_tx < TX_QUEUE_SIZE)
715 		netif_wake_queue (dev);		/* Typical path */
716 
717 	dev->trans_start = jiffies; /* prevent tx timeout */
718 	dev->stats.tx_errors++;
719 }
720 
721 /* Initialize the Rx and Tx rings, along with various 'dev' bits. */
722 static int yellowfin_init_ring(struct net_device *dev)
723 {
724 	struct yellowfin_private *yp = netdev_priv(dev);
725 	int i, j;
726 
727 	yp->tx_full = 0;
728 	yp->cur_rx = yp->cur_tx = 0;
729 	yp->dirty_tx = 0;
730 
731 	yp->rx_buf_sz = (dev->mtu <= 1500 ? PKT_BUF_SZ : dev->mtu + 32);
732 
733 	for (i = 0; i < RX_RING_SIZE; i++) {
734 		yp->rx_ring[i].dbdma_cmd =
735 			cpu_to_le32(CMD_RX_BUF | INTR_ALWAYS | yp->rx_buf_sz);
736 		yp->rx_ring[i].branch_addr = cpu_to_le32(yp->rx_ring_dma +
737 			((i+1)%RX_RING_SIZE)*sizeof(struct yellowfin_desc));
738 	}
739 
740 	for (i = 0; i < RX_RING_SIZE; i++) {
741 		struct sk_buff *skb = netdev_alloc_skb(dev, yp->rx_buf_sz + 2);
742 		yp->rx_skbuff[i] = skb;
743 		if (skb == NULL)
744 			break;
745 		skb_reserve(skb, 2);	/* 16 byte align the IP header. */
746 		yp->rx_ring[i].addr = cpu_to_le32(pci_map_single(yp->pci_dev,
747 			skb->data, yp->rx_buf_sz, PCI_DMA_FROMDEVICE));
748 	}
749 	if (i != RX_RING_SIZE) {
750 		for (j = 0; j < i; j++)
751 			dev_kfree_skb(yp->rx_skbuff[j]);
752 		return -ENOMEM;
753 	}
754 	yp->rx_ring[i-1].dbdma_cmd = cpu_to_le32(CMD_STOP);
755 	yp->dirty_rx = (unsigned int)(i - RX_RING_SIZE);
756 
757 #define NO_TXSTATS
758 #ifdef NO_TXSTATS
759 	/* In this mode the Tx ring needs only a single descriptor. */
760 	for (i = 0; i < TX_RING_SIZE; i++) {
761 		yp->tx_skbuff[i] = NULL;
762 		yp->tx_ring[i].dbdma_cmd = cpu_to_le32(CMD_STOP);
763 		yp->tx_ring[i].branch_addr = cpu_to_le32(yp->tx_ring_dma +
764 			((i+1)%TX_RING_SIZE)*sizeof(struct yellowfin_desc));
765 	}
766 	/* Wrap ring */
767 	yp->tx_ring[--i].dbdma_cmd = cpu_to_le32(CMD_STOP | BRANCH_ALWAYS);
768 #else
769 {
770 	/* Tx ring needs a pair of descriptors, the second for the status. */
771 	for (i = 0; i < TX_RING_SIZE; i++) {
772 		j = 2*i;
773 		yp->tx_skbuff[i] = 0;
774 		/* Branch on Tx error. */
775 		yp->tx_ring[j].dbdma_cmd = cpu_to_le32(CMD_STOP);
776 		yp->tx_ring[j].branch_addr = cpu_to_le32(yp->tx_ring_dma +
777 			(j+1)*sizeof(struct yellowfin_desc));
778 		j++;
779 		if (yp->flags & FullTxStatus) {
780 			yp->tx_ring[j].dbdma_cmd =
781 				cpu_to_le32(CMD_TXSTATUS | sizeof(*yp->tx_status));
782 			yp->tx_ring[j].request_cnt = sizeof(*yp->tx_status);
783 			yp->tx_ring[j].addr = cpu_to_le32(yp->tx_status_dma +
784 				i*sizeof(struct tx_status_words));
785 		} else {
786 			/* Symbios chips write only tx_errs word. */
787 			yp->tx_ring[j].dbdma_cmd =
788 				cpu_to_le32(CMD_TXSTATUS | INTR_ALWAYS | 2);
789 			yp->tx_ring[j].request_cnt = 2;
790 			/* Om pade ummmmm... */
791 			yp->tx_ring[j].addr = cpu_to_le32(yp->tx_status_dma +
792 				i*sizeof(struct tx_status_words) +
793 				&(yp->tx_status[0].tx_errs) -
794 				&(yp->tx_status[0]));
795 		}
796 		yp->tx_ring[j].branch_addr = cpu_to_le32(yp->tx_ring_dma +
797 			((j+1)%(2*TX_RING_SIZE))*sizeof(struct yellowfin_desc));
798 	}
799 	/* Wrap ring */
800 	yp->tx_ring[++j].dbdma_cmd |= cpu_to_le32(BRANCH_ALWAYS | INTR_ALWAYS);
801 }
802 #endif
803 	yp->tx_tail_desc = &yp->tx_status[0];
804 	return 0;
805 }
806 
807 static netdev_tx_t yellowfin_start_xmit(struct sk_buff *skb,
808 					struct net_device *dev)
809 {
810 	struct yellowfin_private *yp = netdev_priv(dev);
811 	unsigned entry;
812 	int len = skb->len;
813 
814 	netif_stop_queue (dev);
815 
816 	/* Note: Ordering is important here, set the field with the
817 	   "ownership" bit last, and only then increment cur_tx. */
818 
819 	/* Calculate the next Tx descriptor entry. */
820 	entry = yp->cur_tx % TX_RING_SIZE;
821 
822 	if (gx_fix) {	/* Note: only works for paddable protocols e.g.  IP. */
823 		int cacheline_end = ((unsigned long)skb->data + skb->len) % 32;
824 		/* Fix GX chipset errata. */
825 		if (cacheline_end > 24  || cacheline_end == 0) {
826 			len = skb->len + 32 - cacheline_end + 1;
827 			if (skb_padto(skb, len)) {
828 				yp->tx_skbuff[entry] = NULL;
829 				netif_wake_queue(dev);
830 				return NETDEV_TX_OK;
831 			}
832 		}
833 	}
834 	yp->tx_skbuff[entry] = skb;
835 
836 #ifdef NO_TXSTATS
837 	yp->tx_ring[entry].addr = cpu_to_le32(pci_map_single(yp->pci_dev,
838 		skb->data, len, PCI_DMA_TODEVICE));
839 	yp->tx_ring[entry].result_status = 0;
840 	if (entry >= TX_RING_SIZE-1) {
841 		/* New stop command. */
842 		yp->tx_ring[0].dbdma_cmd = cpu_to_le32(CMD_STOP);
843 		yp->tx_ring[TX_RING_SIZE-1].dbdma_cmd =
844 			cpu_to_le32(CMD_TX_PKT|BRANCH_ALWAYS | len);
845 	} else {
846 		yp->tx_ring[entry+1].dbdma_cmd = cpu_to_le32(CMD_STOP);
847 		yp->tx_ring[entry].dbdma_cmd =
848 			cpu_to_le32(CMD_TX_PKT | BRANCH_IFTRUE | len);
849 	}
850 	yp->cur_tx++;
851 #else
852 	yp->tx_ring[entry<<1].request_cnt = len;
853 	yp->tx_ring[entry<<1].addr = cpu_to_le32(pci_map_single(yp->pci_dev,
854 		skb->data, len, PCI_DMA_TODEVICE));
855 	/* The input_last (status-write) command is constant, but we must
856 	   rewrite the subsequent 'stop' command. */
857 
858 	yp->cur_tx++;
859 	{
860 		unsigned next_entry = yp->cur_tx % TX_RING_SIZE;
861 		yp->tx_ring[next_entry<<1].dbdma_cmd = cpu_to_le32(CMD_STOP);
862 	}
863 	/* Final step -- overwrite the old 'stop' command. */
864 
865 	yp->tx_ring[entry<<1].dbdma_cmd =
866 		cpu_to_le32( ((entry % 6) == 0 ? CMD_TX_PKT|INTR_ALWAYS|BRANCH_IFTRUE :
867 					  CMD_TX_PKT | BRANCH_IFTRUE) | len);
868 #endif
869 
870 	/* Non-x86 Todo: explicitly flush cache lines here. */
871 
872 	/* Wake the potentially-idle transmit channel. */
873 	iowrite32(0x10001000, yp->base + TxCtrl);
874 
875 	if (yp->cur_tx - yp->dirty_tx < TX_QUEUE_SIZE)
876 		netif_start_queue (dev);		/* Typical path */
877 	else
878 		yp->tx_full = 1;
879 
880 	if (yellowfin_debug > 4) {
881 		netdev_printk(KERN_DEBUG, dev, "Yellowfin transmit frame #%d queued in slot %d\n",
882 			      yp->cur_tx, entry);
883 	}
884 	return NETDEV_TX_OK;
885 }
886 
887 /* The interrupt handler does all of the Rx thread work and cleans up
888    after the Tx thread. */
889 static irqreturn_t yellowfin_interrupt(int irq, void *dev_instance)
890 {
891 	struct net_device *dev = dev_instance;
892 	struct yellowfin_private *yp;
893 	void __iomem *ioaddr;
894 	int boguscnt = max_interrupt_work;
895 	unsigned int handled = 0;
896 
897 	yp = netdev_priv(dev);
898 	ioaddr = yp->base;
899 
900 	spin_lock (&yp->lock);
901 
902 	do {
903 		u16 intr_status = ioread16(ioaddr + IntrClear);
904 
905 		if (yellowfin_debug > 4)
906 			netdev_printk(KERN_DEBUG, dev, "Yellowfin interrupt, status %04x\n",
907 				      intr_status);
908 
909 		if (intr_status == 0)
910 			break;
911 		handled = 1;
912 
913 		if (intr_status & (IntrRxDone | IntrEarlyRx)) {
914 			yellowfin_rx(dev);
915 			iowrite32(0x10001000, ioaddr + RxCtrl);		/* Wake Rx engine. */
916 		}
917 
918 #ifdef NO_TXSTATS
919 		for (; yp->cur_tx - yp->dirty_tx > 0; yp->dirty_tx++) {
920 			int entry = yp->dirty_tx % TX_RING_SIZE;
921 			struct sk_buff *skb;
922 
923 			if (yp->tx_ring[entry].result_status == 0)
924 				break;
925 			skb = yp->tx_skbuff[entry];
926 			dev->stats.tx_packets++;
927 			dev->stats.tx_bytes += skb->len;
928 			/* Free the original skb. */
929 			pci_unmap_single(yp->pci_dev, le32_to_cpu(yp->tx_ring[entry].addr),
930 				skb->len, PCI_DMA_TODEVICE);
931 			dev_kfree_skb_irq(skb);
932 			yp->tx_skbuff[entry] = NULL;
933 		}
934 		if (yp->tx_full &&
935 		    yp->cur_tx - yp->dirty_tx < TX_QUEUE_SIZE - 4) {
936 			/* The ring is no longer full, clear tbusy. */
937 			yp->tx_full = 0;
938 			netif_wake_queue(dev);
939 		}
940 #else
941 		if ((intr_status & IntrTxDone) || (yp->tx_tail_desc->tx_errs)) {
942 			unsigned dirty_tx = yp->dirty_tx;
943 
944 			for (dirty_tx = yp->dirty_tx; yp->cur_tx - dirty_tx > 0;
945 				 dirty_tx++) {
946 				/* Todo: optimize this. */
947 				int entry = dirty_tx % TX_RING_SIZE;
948 				u16 tx_errs = yp->tx_status[entry].tx_errs;
949 				struct sk_buff *skb;
950 
951 #ifndef final_version
952 				if (yellowfin_debug > 5)
953 					netdev_printk(KERN_DEBUG, dev, "Tx queue %d check, Tx status %04x %04x %04x %04x\n",
954 						      entry,
955 						      yp->tx_status[entry].tx_cnt,
956 						      yp->tx_status[entry].tx_errs,
957 						      yp->tx_status[entry].total_tx_cnt,
958 						      yp->tx_status[entry].paused);
959 #endif
960 				if (tx_errs == 0)
961 					break;	/* It still hasn't been Txed */
962 				skb = yp->tx_skbuff[entry];
963 				if (tx_errs & 0xF810) {
964 					/* There was an major error, log it. */
965 #ifndef final_version
966 					if (yellowfin_debug > 1)
967 						netdev_printk(KERN_DEBUG, dev, "Transmit error, Tx status %04x\n",
968 							      tx_errs);
969 #endif
970 					dev->stats.tx_errors++;
971 					if (tx_errs & 0xF800) dev->stats.tx_aborted_errors++;
972 					if (tx_errs & 0x0800) dev->stats.tx_carrier_errors++;
973 					if (tx_errs & 0x2000) dev->stats.tx_window_errors++;
974 					if (tx_errs & 0x8000) dev->stats.tx_fifo_errors++;
975 				} else {
976 #ifndef final_version
977 					if (yellowfin_debug > 4)
978 						netdev_printk(KERN_DEBUG, dev, "Normal transmit, Tx status %04x\n",
979 							      tx_errs);
980 #endif
981 					dev->stats.tx_bytes += skb->len;
982 					dev->stats.collisions += tx_errs & 15;
983 					dev->stats.tx_packets++;
984 				}
985 				/* Free the original skb. */
986 				pci_unmap_single(yp->pci_dev,
987 					yp->tx_ring[entry<<1].addr, skb->len,
988 					PCI_DMA_TODEVICE);
989 				dev_kfree_skb_irq(skb);
990 				yp->tx_skbuff[entry] = 0;
991 				/* Mark status as empty. */
992 				yp->tx_status[entry].tx_errs = 0;
993 			}
994 
995 #ifndef final_version
996 			if (yp->cur_tx - dirty_tx > TX_RING_SIZE) {
997 				netdev_err(dev, "Out-of-sync dirty pointer, %d vs. %d, full=%d\n",
998 					   dirty_tx, yp->cur_tx, yp->tx_full);
999 				dirty_tx += TX_RING_SIZE;
1000 			}
1001 #endif
1002 
1003 			if (yp->tx_full &&
1004 			    yp->cur_tx - dirty_tx < TX_QUEUE_SIZE - 2) {
1005 				/* The ring is no longer full, clear tbusy. */
1006 				yp->tx_full = 0;
1007 				netif_wake_queue(dev);
1008 			}
1009 
1010 			yp->dirty_tx = dirty_tx;
1011 			yp->tx_tail_desc = &yp->tx_status[dirty_tx % TX_RING_SIZE];
1012 		}
1013 #endif
1014 
1015 		/* Log errors and other uncommon events. */
1016 		if (intr_status & 0x2ee)	/* Abnormal error summary. */
1017 			yellowfin_error(dev, intr_status);
1018 
1019 		if (--boguscnt < 0) {
1020 			netdev_warn(dev, "Too much work at interrupt, status=%#04x\n",
1021 				    intr_status);
1022 			break;
1023 		}
1024 	} while (1);
1025 
1026 	if (yellowfin_debug > 3)
1027 		netdev_printk(KERN_DEBUG, dev, "exiting interrupt, status=%#04x\n",
1028 			      ioread16(ioaddr + IntrStatus));
1029 
1030 	spin_unlock (&yp->lock);
1031 	return IRQ_RETVAL(handled);
1032 }
1033 
1034 /* This routine is logically part of the interrupt handler, but separated
1035    for clarity and better register allocation. */
1036 static int yellowfin_rx(struct net_device *dev)
1037 {
1038 	struct yellowfin_private *yp = netdev_priv(dev);
1039 	int entry = yp->cur_rx % RX_RING_SIZE;
1040 	int boguscnt = yp->dirty_rx + RX_RING_SIZE - yp->cur_rx;
1041 
1042 	if (yellowfin_debug > 4) {
1043 		printk(KERN_DEBUG " In yellowfin_rx(), entry %d status %08x\n",
1044 			   entry, yp->rx_ring[entry].result_status);
1045 		printk(KERN_DEBUG "   #%d desc. %08x %08x %08x\n",
1046 			   entry, yp->rx_ring[entry].dbdma_cmd, yp->rx_ring[entry].addr,
1047 			   yp->rx_ring[entry].result_status);
1048 	}
1049 
1050 	/* If EOP is set on the next entry, it's a new packet. Send it up. */
1051 	while (1) {
1052 		struct yellowfin_desc *desc = &yp->rx_ring[entry];
1053 		struct sk_buff *rx_skb = yp->rx_skbuff[entry];
1054 		s16 frame_status;
1055 		u16 desc_status;
1056 		int data_size, yf_size;
1057 		u8 *buf_addr;
1058 
1059 		if(!desc->result_status)
1060 			break;
1061 		pci_dma_sync_single_for_cpu(yp->pci_dev, le32_to_cpu(desc->addr),
1062 			yp->rx_buf_sz, PCI_DMA_FROMDEVICE);
1063 		desc_status = le32_to_cpu(desc->result_status) >> 16;
1064 		buf_addr = rx_skb->data;
1065 		data_size = (le32_to_cpu(desc->dbdma_cmd) -
1066 			le32_to_cpu(desc->result_status)) & 0xffff;
1067 		frame_status = get_unaligned_le16(&(buf_addr[data_size - 2]));
1068 		if (yellowfin_debug > 4)
1069 			printk(KERN_DEBUG "  %s() status was %04x\n",
1070 			       __func__, frame_status);
1071 		if (--boguscnt < 0)
1072 			break;
1073 
1074 		yf_size = sizeof(struct yellowfin_desc);
1075 
1076 		if ( ! (desc_status & RX_EOP)) {
1077 			if (data_size != 0)
1078 				netdev_warn(dev, "Oversized Ethernet frame spanned multiple buffers, status %04x, data_size %d!\n",
1079 					    desc_status, data_size);
1080 			dev->stats.rx_length_errors++;
1081 		} else if ((yp->drv_flags & IsGigabit)  &&  (frame_status & 0x0038)) {
1082 			/* There was a error. */
1083 			if (yellowfin_debug > 3)
1084 				printk(KERN_DEBUG "  %s() Rx error was %04x\n",
1085 				       __func__, frame_status);
1086 			dev->stats.rx_errors++;
1087 			if (frame_status & 0x0060) dev->stats.rx_length_errors++;
1088 			if (frame_status & 0x0008) dev->stats.rx_frame_errors++;
1089 			if (frame_status & 0x0010) dev->stats.rx_crc_errors++;
1090 			if (frame_status < 0) dev->stats.rx_dropped++;
1091 		} else if ( !(yp->drv_flags & IsGigabit)  &&
1092 				   ((buf_addr[data_size-1] & 0x85) || buf_addr[data_size-2] & 0xC0)) {
1093 			u8 status1 = buf_addr[data_size-2];
1094 			u8 status2 = buf_addr[data_size-1];
1095 			dev->stats.rx_errors++;
1096 			if (status1 & 0xC0) dev->stats.rx_length_errors++;
1097 			if (status2 & 0x03) dev->stats.rx_frame_errors++;
1098 			if (status2 & 0x04) dev->stats.rx_crc_errors++;
1099 			if (status2 & 0x80) dev->stats.rx_dropped++;
1100 #ifdef YF_PROTOTYPE		/* Support for prototype hardware errata. */
1101 		} else if ((yp->flags & HasMACAddrBug)  &&
1102 			!ether_addr_equal(le32_to_cpu(yp->rx_ring_dma +
1103 						      entry * yf_size),
1104 					  dev->dev_addr) &&
1105 			!ether_addr_equal(le32_to_cpu(yp->rx_ring_dma +
1106 						      entry * yf_size),
1107 					  "\377\377\377\377\377\377")) {
1108 			if (bogus_rx++ == 0)
1109 				netdev_warn(dev, "Bad frame to %pM\n",
1110 					    buf_addr);
1111 #endif
1112 		} else {
1113 			struct sk_buff *skb;
1114 			int pkt_len = data_size -
1115 				(yp->chip_id ? 7 : 8 + buf_addr[data_size - 8]);
1116 			/* To verify: Yellowfin Length should omit the CRC! */
1117 
1118 #ifndef final_version
1119 			if (yellowfin_debug > 4)
1120 				printk(KERN_DEBUG "  %s() normal Rx pkt length %d of %d, bogus_cnt %d\n",
1121 				       __func__, pkt_len, data_size, boguscnt);
1122 #endif
1123 			/* Check if the packet is long enough to just pass up the skbuff
1124 			   without copying to a properly sized skbuff. */
1125 			if (pkt_len > rx_copybreak) {
1126 				skb_put(skb = rx_skb, pkt_len);
1127 				pci_unmap_single(yp->pci_dev,
1128 					le32_to_cpu(yp->rx_ring[entry].addr),
1129 					yp->rx_buf_sz,
1130 					PCI_DMA_FROMDEVICE);
1131 				yp->rx_skbuff[entry] = NULL;
1132 			} else {
1133 				skb = netdev_alloc_skb(dev, pkt_len + 2);
1134 				if (skb == NULL)
1135 					break;
1136 				skb_reserve(skb, 2);	/* 16 byte align the IP header */
1137 				skb_copy_to_linear_data(skb, rx_skb->data, pkt_len);
1138 				skb_put(skb, pkt_len);
1139 				pci_dma_sync_single_for_device(yp->pci_dev,
1140 								le32_to_cpu(desc->addr),
1141 								yp->rx_buf_sz,
1142 								PCI_DMA_FROMDEVICE);
1143 			}
1144 			skb->protocol = eth_type_trans(skb, dev);
1145 			netif_rx(skb);
1146 			dev->stats.rx_packets++;
1147 			dev->stats.rx_bytes += pkt_len;
1148 		}
1149 		entry = (++yp->cur_rx) % RX_RING_SIZE;
1150 	}
1151 
1152 	/* Refill the Rx ring buffers. */
1153 	for (; yp->cur_rx - yp->dirty_rx > 0; yp->dirty_rx++) {
1154 		entry = yp->dirty_rx % RX_RING_SIZE;
1155 		if (yp->rx_skbuff[entry] == NULL) {
1156 			struct sk_buff *skb = netdev_alloc_skb(dev, yp->rx_buf_sz + 2);
1157 			if (skb == NULL)
1158 				break;				/* Better luck next round. */
1159 			yp->rx_skbuff[entry] = skb;
1160 			skb_reserve(skb, 2);	/* Align IP on 16 byte boundaries */
1161 			yp->rx_ring[entry].addr = cpu_to_le32(pci_map_single(yp->pci_dev,
1162 				skb->data, yp->rx_buf_sz, PCI_DMA_FROMDEVICE));
1163 		}
1164 		yp->rx_ring[entry].dbdma_cmd = cpu_to_le32(CMD_STOP);
1165 		yp->rx_ring[entry].result_status = 0;	/* Clear complete bit. */
1166 		if (entry != 0)
1167 			yp->rx_ring[entry - 1].dbdma_cmd =
1168 				cpu_to_le32(CMD_RX_BUF | INTR_ALWAYS | yp->rx_buf_sz);
1169 		else
1170 			yp->rx_ring[RX_RING_SIZE - 1].dbdma_cmd =
1171 				cpu_to_le32(CMD_RX_BUF | INTR_ALWAYS | BRANCH_ALWAYS
1172 							| yp->rx_buf_sz);
1173 	}
1174 
1175 	return 0;
1176 }
1177 
1178 static void yellowfin_error(struct net_device *dev, int intr_status)
1179 {
1180 	netdev_err(dev, "Something Wicked happened! %04x\n", intr_status);
1181 	/* Hmmmmm, it's not clear what to do here. */
1182 	if (intr_status & (IntrTxPCIErr | IntrTxPCIFault))
1183 		dev->stats.tx_errors++;
1184 	if (intr_status & (IntrRxPCIErr | IntrRxPCIFault))
1185 		dev->stats.rx_errors++;
1186 }
1187 
1188 static int yellowfin_close(struct net_device *dev)
1189 {
1190 	struct yellowfin_private *yp = netdev_priv(dev);
1191 	void __iomem *ioaddr = yp->base;
1192 	int i;
1193 
1194 	netif_stop_queue (dev);
1195 
1196 	if (yellowfin_debug > 1) {
1197 		netdev_printk(KERN_DEBUG, dev, "Shutting down ethercard, status was Tx %04x Rx %04x Int %02x\n",
1198 			      ioread16(ioaddr + TxStatus),
1199 			      ioread16(ioaddr + RxStatus),
1200 			      ioread16(ioaddr + IntrStatus));
1201 		netdev_printk(KERN_DEBUG, dev, "Queue pointers were Tx %d / %d,  Rx %d / %d\n",
1202 			      yp->cur_tx, yp->dirty_tx,
1203 			      yp->cur_rx, yp->dirty_rx);
1204 	}
1205 
1206 	/* Disable interrupts by clearing the interrupt mask. */
1207 	iowrite16(0x0000, ioaddr + IntrEnb);
1208 
1209 	/* Stop the chip's Tx and Rx processes. */
1210 	iowrite32(0x80000000, ioaddr + RxCtrl);
1211 	iowrite32(0x80000000, ioaddr + TxCtrl);
1212 
1213 	del_timer(&yp->timer);
1214 
1215 #if defined(__i386__)
1216 	if (yellowfin_debug > 2) {
1217 		printk(KERN_DEBUG "  Tx ring at %08llx:\n",
1218 				(unsigned long long)yp->tx_ring_dma);
1219 		for (i = 0; i < TX_RING_SIZE*2; i++)
1220 			printk(KERN_DEBUG " %c #%d desc. %08x %08x %08x %08x\n",
1221 				   ioread32(ioaddr + TxPtr) == (long)&yp->tx_ring[i] ? '>' : ' ',
1222 				   i, yp->tx_ring[i].dbdma_cmd, yp->tx_ring[i].addr,
1223 				   yp->tx_ring[i].branch_addr, yp->tx_ring[i].result_status);
1224 		printk(KERN_DEBUG "  Tx status %p:\n", yp->tx_status);
1225 		for (i = 0; i < TX_RING_SIZE; i++)
1226 			printk(KERN_DEBUG "   #%d status %04x %04x %04x %04x\n",
1227 				   i, yp->tx_status[i].tx_cnt, yp->tx_status[i].tx_errs,
1228 				   yp->tx_status[i].total_tx_cnt, yp->tx_status[i].paused);
1229 
1230 		printk(KERN_DEBUG "  Rx ring %08llx:\n",
1231 				(unsigned long long)yp->rx_ring_dma);
1232 		for (i = 0; i < RX_RING_SIZE; i++) {
1233 			printk(KERN_DEBUG " %c #%d desc. %08x %08x %08x\n",
1234 				   ioread32(ioaddr + RxPtr) == (long)&yp->rx_ring[i] ? '>' : ' ',
1235 				   i, yp->rx_ring[i].dbdma_cmd, yp->rx_ring[i].addr,
1236 				   yp->rx_ring[i].result_status);
1237 			if (yellowfin_debug > 6) {
1238 				if (get_unaligned((u8*)yp->rx_ring[i].addr) != 0x69) {
1239 					int j;
1240 
1241 					printk(KERN_DEBUG);
1242 					for (j = 0; j < 0x50; j++)
1243 						pr_cont(" %04x",
1244 							get_unaligned(((u16*)yp->rx_ring[i].addr) + j));
1245 					pr_cont("\n");
1246 				}
1247 			}
1248 		}
1249 	}
1250 #endif /* __i386__ debugging only */
1251 
1252 	free_irq(yp->pci_dev->irq, dev);
1253 
1254 	/* Free all the skbuffs in the Rx queue. */
1255 	for (i = 0; i < RX_RING_SIZE; i++) {
1256 		yp->rx_ring[i].dbdma_cmd = cpu_to_le32(CMD_STOP);
1257 		yp->rx_ring[i].addr = cpu_to_le32(0xBADF00D0); /* An invalid address. */
1258 		if (yp->rx_skbuff[i]) {
1259 			dev_kfree_skb(yp->rx_skbuff[i]);
1260 		}
1261 		yp->rx_skbuff[i] = NULL;
1262 	}
1263 	for (i = 0; i < TX_RING_SIZE; i++) {
1264 		if (yp->tx_skbuff[i])
1265 			dev_kfree_skb(yp->tx_skbuff[i]);
1266 		yp->tx_skbuff[i] = NULL;
1267 	}
1268 
1269 #ifdef YF_PROTOTYPE			/* Support for prototype hardware errata. */
1270 	if (yellowfin_debug > 0) {
1271 		netdev_printk(KERN_DEBUG, dev, "Received %d frames that we should not have\n",
1272 			      bogus_rx);
1273 	}
1274 #endif
1275 
1276 	return 0;
1277 }
1278 
1279 /* Set or clear the multicast filter for this adaptor. */
1280 
1281 static void set_rx_mode(struct net_device *dev)
1282 {
1283 	struct yellowfin_private *yp = netdev_priv(dev);
1284 	void __iomem *ioaddr = yp->base;
1285 	u16 cfg_value = ioread16(ioaddr + Cnfg);
1286 
1287 	/* Stop the Rx process to change any value. */
1288 	iowrite16(cfg_value & ~0x1000, ioaddr + Cnfg);
1289 	if (dev->flags & IFF_PROMISC) {			/* Set promiscuous. */
1290 		iowrite16(0x000F, ioaddr + AddrMode);
1291 	} else if ((netdev_mc_count(dev) > 64) ||
1292 		   (dev->flags & IFF_ALLMULTI)) {
1293 		/* Too many to filter well, or accept all multicasts. */
1294 		iowrite16(0x000B, ioaddr + AddrMode);
1295 	} else if (!netdev_mc_empty(dev)) { /* Must use the multicast hash table. */
1296 		struct netdev_hw_addr *ha;
1297 		u16 hash_table[4];
1298 		int i;
1299 
1300 		memset(hash_table, 0, sizeof(hash_table));
1301 		netdev_for_each_mc_addr(ha, dev) {
1302 			unsigned int bit;
1303 
1304 			/* Due to a bug in the early chip versions, multiple filter
1305 			   slots must be set for each address. */
1306 			if (yp->drv_flags & HasMulticastBug) {
1307 				bit = (ether_crc_le(3, ha->addr) >> 3) & 0x3f;
1308 				hash_table[bit >> 4] |= (1 << bit);
1309 				bit = (ether_crc_le(4, ha->addr) >> 3) & 0x3f;
1310 				hash_table[bit >> 4] |= (1 << bit);
1311 				bit = (ether_crc_le(5, ha->addr) >> 3) & 0x3f;
1312 				hash_table[bit >> 4] |= (1 << bit);
1313 			}
1314 			bit = (ether_crc_le(6, ha->addr) >> 3) & 0x3f;
1315 			hash_table[bit >> 4] |= (1 << bit);
1316 		}
1317 		/* Copy the hash table to the chip. */
1318 		for (i = 0; i < 4; i++)
1319 			iowrite16(hash_table[i], ioaddr + HashTbl + i*2);
1320 		iowrite16(0x0003, ioaddr + AddrMode);
1321 	} else {					/* Normal, unicast/broadcast-only mode. */
1322 		iowrite16(0x0001, ioaddr + AddrMode);
1323 	}
1324 	/* Restart the Rx process. */
1325 	iowrite16(cfg_value | 0x1000, ioaddr + Cnfg);
1326 }
1327 
1328 static void yellowfin_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1329 {
1330 	struct yellowfin_private *np = netdev_priv(dev);
1331 
1332 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1333 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1334 	strlcpy(info->bus_info, pci_name(np->pci_dev), sizeof(info->bus_info));
1335 }
1336 
1337 static const struct ethtool_ops ethtool_ops = {
1338 	.get_drvinfo = yellowfin_get_drvinfo
1339 };
1340 
1341 static int netdev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
1342 {
1343 	struct yellowfin_private *np = netdev_priv(dev);
1344 	void __iomem *ioaddr = np->base;
1345 	struct mii_ioctl_data *data = if_mii(rq);
1346 
1347 	switch(cmd) {
1348 	case SIOCGMIIPHY:		/* Get address of MII PHY in use. */
1349 		data->phy_id = np->phys[0] & 0x1f;
1350 		/* Fall Through */
1351 
1352 	case SIOCGMIIREG:		/* Read MII PHY register. */
1353 		data->val_out = mdio_read(ioaddr, data->phy_id & 0x1f, data->reg_num & 0x1f);
1354 		return 0;
1355 
1356 	case SIOCSMIIREG:		/* Write MII PHY register. */
1357 		if (data->phy_id == np->phys[0]) {
1358 			u16 value = data->val_in;
1359 			switch (data->reg_num) {
1360 			case 0:
1361 				/* Check for autonegotiation on or reset. */
1362 				np->medialock = (value & 0x9000) ? 0 : 1;
1363 				if (np->medialock)
1364 					np->full_duplex = (value & 0x0100) ? 1 : 0;
1365 				break;
1366 			case 4: np->advertising = value; break;
1367 			}
1368 			/* Perhaps check_duplex(dev), depending on chip semantics. */
1369 		}
1370 		mdio_write(ioaddr, data->phy_id & 0x1f, data->reg_num & 0x1f, data->val_in);
1371 		return 0;
1372 	default:
1373 		return -EOPNOTSUPP;
1374 	}
1375 }
1376 
1377 
1378 static void yellowfin_remove_one(struct pci_dev *pdev)
1379 {
1380 	struct net_device *dev = pci_get_drvdata(pdev);
1381 	struct yellowfin_private *np;
1382 
1383 	BUG_ON(!dev);
1384 	np = netdev_priv(dev);
1385 
1386         pci_free_consistent(pdev, STATUS_TOTAL_SIZE, np->tx_status,
1387 		np->tx_status_dma);
1388 	pci_free_consistent(pdev, RX_TOTAL_SIZE, np->rx_ring, np->rx_ring_dma);
1389 	pci_free_consistent(pdev, TX_TOTAL_SIZE, np->tx_ring, np->tx_ring_dma);
1390 	unregister_netdev (dev);
1391 
1392 	pci_iounmap(pdev, np->base);
1393 
1394 	pci_release_regions (pdev);
1395 
1396 	free_netdev (dev);
1397 }
1398 
1399 
1400 static struct pci_driver yellowfin_driver = {
1401 	.name		= DRV_NAME,
1402 	.id_table	= yellowfin_pci_tbl,
1403 	.probe		= yellowfin_init_one,
1404 	.remove		= yellowfin_remove_one,
1405 };
1406 
1407 
1408 static int __init yellowfin_init (void)
1409 {
1410 /* when a module, this is printed whether or not devices are found in probe */
1411 #ifdef MODULE
1412 	printk(version);
1413 #endif
1414 	return pci_register_driver(&yellowfin_driver);
1415 }
1416 
1417 
1418 static void __exit yellowfin_cleanup (void)
1419 {
1420 	pci_unregister_driver (&yellowfin_driver);
1421 }
1422 
1423 
1424 module_init(yellowfin_init);
1425 module_exit(yellowfin_cleanup);
1426