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 <linux/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(struct timer_list *t);
347 static void yellowfin_tx_timeout(struct net_device *dev, unsigned int txqueue);
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_validate_addr	= eth_validate_addr,
364 	.ndo_set_mac_address 	= eth_mac_addr,
365 	.ndo_do_ioctl 		= netdev_ioctl,
366 	.ndo_tx_timeout 	= yellowfin_tx_timeout,
367 };
368 
369 static int yellowfin_init_one(struct pci_dev *pdev,
370 			      const struct pci_device_id *ent)
371 {
372 	struct net_device *dev;
373 	struct yellowfin_private *np;
374 	int irq;
375 	int chip_idx = ent->driver_data;
376 	static int find_cnt;
377 	void __iomem *ioaddr;
378 	int i, option = find_cnt < MAX_UNITS ? options[find_cnt] : 0;
379 	int drv_flags = pci_id_tbl[chip_idx].drv_flags;
380         void *ring_space;
381         dma_addr_t ring_dma;
382 #ifdef USE_IO_OPS
383 	int bar = 0;
384 #else
385 	int bar = 1;
386 #endif
387 
388 /* when built into the kernel, we only print version if device is found */
389 #ifndef MODULE
390 	static int printed_version;
391 	if (!printed_version++)
392 		printk(version);
393 #endif
394 
395 	i = pci_enable_device(pdev);
396 	if (i) return i;
397 
398 	dev = alloc_etherdev(sizeof(*np));
399 	if (!dev)
400 		return -ENOMEM;
401 
402 	SET_NETDEV_DEV(dev, &pdev->dev);
403 
404 	np = netdev_priv(dev);
405 
406 	if (pci_request_regions(pdev, DRV_NAME))
407 		goto err_out_free_netdev;
408 
409 	pci_set_master (pdev);
410 
411 	ioaddr = pci_iomap(pdev, bar, YELLOWFIN_SIZE);
412 	if (!ioaddr)
413 		goto err_out_free_res;
414 
415 	irq = pdev->irq;
416 
417 	if (drv_flags & DontUseEeprom)
418 		for (i = 0; i < 6; i++)
419 			dev->dev_addr[i] = ioread8(ioaddr + StnAddr + i);
420 	else {
421 		int ee_offset = (read_eeprom(ioaddr, 6) == 0xff ? 0x100 : 0);
422 		for (i = 0; i < 6; i++)
423 			dev->dev_addr[i] = read_eeprom(ioaddr, ee_offset + i);
424 	}
425 
426 	/* Reset the chip. */
427 	iowrite32(0x80000000, ioaddr + DMACtrl);
428 
429 	pci_set_drvdata(pdev, dev);
430 	spin_lock_init(&np->lock);
431 
432 	np->pci_dev = pdev;
433 	np->chip_id = chip_idx;
434 	np->drv_flags = drv_flags;
435 	np->base = ioaddr;
436 
437 	ring_space = pci_alloc_consistent(pdev, TX_TOTAL_SIZE, &ring_dma);
438 	if (!ring_space)
439 		goto err_out_cleardev;
440 	np->tx_ring = ring_space;
441 	np->tx_ring_dma = ring_dma;
442 
443 	ring_space = pci_alloc_consistent(pdev, RX_TOTAL_SIZE, &ring_dma);
444 	if (!ring_space)
445 		goto err_out_unmap_tx;
446 	np->rx_ring = ring_space;
447 	np->rx_ring_dma = ring_dma;
448 
449 	ring_space = pci_alloc_consistent(pdev, STATUS_TOTAL_SIZE, &ring_dma);
450 	if (!ring_space)
451 		goto err_out_unmap_rx;
452 	np->tx_status = ring_space;
453 	np->tx_status_dma = ring_dma;
454 
455 	if (dev->mem_start)
456 		option = dev->mem_start;
457 
458 	/* The lower four bits are the media type. */
459 	if (option > 0) {
460 		if (option & 0x200)
461 			np->full_duplex = 1;
462 		np->default_port = option & 15;
463 		if (np->default_port)
464 			np->medialock = 1;
465 	}
466 	if (find_cnt < MAX_UNITS  &&  full_duplex[find_cnt] > 0)
467 		np->full_duplex = 1;
468 
469 	if (np->full_duplex)
470 		np->duplex_lock = 1;
471 
472 	/* The Yellowfin-specific entries in the device structure. */
473 	dev->netdev_ops = &netdev_ops;
474 	dev->ethtool_ops = &ethtool_ops;
475 	dev->watchdog_timeo = TX_TIMEOUT;
476 
477 	if (mtu)
478 		dev->mtu = mtu;
479 
480 	i = register_netdev(dev);
481 	if (i)
482 		goto err_out_unmap_status;
483 
484 	netdev_info(dev, "%s type %8x at %p, %pM, IRQ %d\n",
485 		    pci_id_tbl[chip_idx].name,
486 		    ioread32(ioaddr + ChipRev), ioaddr,
487 		    dev->dev_addr, irq);
488 
489 	if (np->drv_flags & HasMII) {
490 		int phy, phy_idx = 0;
491 		for (phy = 0; phy < 32 && phy_idx < MII_CNT; phy++) {
492 			int mii_status = mdio_read(ioaddr, phy, 1);
493 			if (mii_status != 0xffff  &&  mii_status != 0x0000) {
494 				np->phys[phy_idx++] = phy;
495 				np->advertising = mdio_read(ioaddr, phy, 4);
496 				netdev_info(dev, "MII PHY found at address %d, status 0x%04x advertising %04x\n",
497 					    phy, mii_status, np->advertising);
498 			}
499 		}
500 		np->mii_cnt = phy_idx;
501 	}
502 
503 	find_cnt++;
504 
505 	return 0;
506 
507 err_out_unmap_status:
508         pci_free_consistent(pdev, STATUS_TOTAL_SIZE, np->tx_status,
509 		np->tx_status_dma);
510 err_out_unmap_rx:
511         pci_free_consistent(pdev, RX_TOTAL_SIZE, np->rx_ring, np->rx_ring_dma);
512 err_out_unmap_tx:
513         pci_free_consistent(pdev, TX_TOTAL_SIZE, np->tx_ring, np->tx_ring_dma);
514 err_out_cleardev:
515 	pci_iounmap(pdev, ioaddr);
516 err_out_free_res:
517 	pci_release_regions(pdev);
518 err_out_free_netdev:
519 	free_netdev (dev);
520 	return -ENODEV;
521 }
522 
523 static int read_eeprom(void __iomem *ioaddr, int location)
524 {
525 	int bogus_cnt = 10000;		/* Typical 33Mhz: 1050 ticks */
526 
527 	iowrite8(location, ioaddr + EEAddr);
528 	iowrite8(0x30 | ((location >> 8) & 7), ioaddr + EECtrl);
529 	while ((ioread8(ioaddr + EEStatus) & 0x80)  &&  --bogus_cnt > 0)
530 		;
531 	return ioread8(ioaddr + EERead);
532 }
533 
534 /* MII Managemen Data I/O accesses.
535    These routines assume the MDIO controller is idle, and do not exit until
536    the command is finished. */
537 
538 static int mdio_read(void __iomem *ioaddr, int phy_id, int location)
539 {
540 	int i;
541 
542 	iowrite16((phy_id<<8) + location, ioaddr + MII_Addr);
543 	iowrite16(1, ioaddr + MII_Cmd);
544 	for (i = 10000; i >= 0; i--)
545 		if ((ioread16(ioaddr + MII_Status) & 1) == 0)
546 			break;
547 	return ioread16(ioaddr + MII_Rd_Data);
548 }
549 
550 static void mdio_write(void __iomem *ioaddr, int phy_id, int location, int value)
551 {
552 	int i;
553 
554 	iowrite16((phy_id<<8) + location, ioaddr + MII_Addr);
555 	iowrite16(value, ioaddr + MII_Wr_Data);
556 
557 	/* Wait for the command to finish. */
558 	for (i = 10000; i >= 0; i--)
559 		if ((ioread16(ioaddr + MII_Status) & 1) == 0)
560 			break;
561 }
562 
563 
564 static int yellowfin_open(struct net_device *dev)
565 {
566 	struct yellowfin_private *yp = netdev_priv(dev);
567 	const int irq = yp->pci_dev->irq;
568 	void __iomem *ioaddr = yp->base;
569 	int i, rc;
570 
571 	/* Reset the chip. */
572 	iowrite32(0x80000000, ioaddr + DMACtrl);
573 
574 	rc = request_irq(irq, yellowfin_interrupt, IRQF_SHARED, dev->name, dev);
575 	if (rc)
576 		return rc;
577 
578 	rc = yellowfin_init_ring(dev);
579 	if (rc < 0)
580 		goto err_free_irq;
581 
582 	iowrite32(yp->rx_ring_dma, ioaddr + RxPtr);
583 	iowrite32(yp->tx_ring_dma, ioaddr + TxPtr);
584 
585 	for (i = 0; i < 6; i++)
586 		iowrite8(dev->dev_addr[i], ioaddr + StnAddr + i);
587 
588 	/* Set up various condition 'select' registers.
589 	   There are no options here. */
590 	iowrite32(0x00800080, ioaddr + TxIntrSel); 	/* Interrupt on Tx abort */
591 	iowrite32(0x00800080, ioaddr + TxBranchSel);	/* Branch on Tx abort */
592 	iowrite32(0x00400040, ioaddr + TxWaitSel); 	/* Wait on Tx status */
593 	iowrite32(0x00400040, ioaddr + RxIntrSel);	/* Interrupt on Rx done */
594 	iowrite32(0x00400040, ioaddr + RxBranchSel);	/* Branch on Rx error */
595 	iowrite32(0x00400040, ioaddr + RxWaitSel);	/* Wait on Rx done */
596 
597 	/* Initialize other registers: with so many this eventually this will
598 	   converted to an offset/value list. */
599 	iowrite32(dma_ctrl, ioaddr + DMACtrl);
600 	iowrite16(fifo_cfg, ioaddr + FIFOcfg);
601 	/* Enable automatic generation of flow control frames, period 0xffff. */
602 	iowrite32(0x0030FFFF, ioaddr + FlowCtrl);
603 
604 	yp->tx_threshold = 32;
605 	iowrite32(yp->tx_threshold, ioaddr + TxThreshold);
606 
607 	if (dev->if_port == 0)
608 		dev->if_port = yp->default_port;
609 
610 	netif_start_queue(dev);
611 
612 	/* Setting the Rx mode will start the Rx process. */
613 	if (yp->drv_flags & IsGigabit) {
614 		/* We are always in full-duplex mode with gigabit! */
615 		yp->full_duplex = 1;
616 		iowrite16(0x01CF, ioaddr + Cnfg);
617 	} else {
618 		iowrite16(0x0018, ioaddr + FrameGap0); /* 0060/4060 for non-MII 10baseT */
619 		iowrite16(0x1018, ioaddr + FrameGap1);
620 		iowrite16(0x101C | (yp->full_duplex ? 2 : 0), ioaddr + Cnfg);
621 	}
622 	set_rx_mode(dev);
623 
624 	/* Enable interrupts by setting the interrupt mask. */
625 	iowrite16(0x81ff, ioaddr + IntrEnb);			/* See enum intr_status_bits */
626 	iowrite16(0x0000, ioaddr + EventStatus);		/* Clear non-interrupting events */
627 	iowrite32(0x80008000, ioaddr + RxCtrl);		/* Start Rx and Tx channels. */
628 	iowrite32(0x80008000, ioaddr + TxCtrl);
629 
630 	if (yellowfin_debug > 2) {
631 		netdev_printk(KERN_DEBUG, dev, "Done %s()\n", __func__);
632 	}
633 
634 	/* Set the timer to check for link beat. */
635 	timer_setup(&yp->timer, yellowfin_timer, 0);
636 	yp->timer.expires = jiffies + 3*HZ;
637 	add_timer(&yp->timer);
638 out:
639 	return rc;
640 
641 err_free_irq:
642 	free_irq(irq, dev);
643 	goto out;
644 }
645 
646 static void yellowfin_timer(struct timer_list *t)
647 {
648 	struct yellowfin_private *yp = from_timer(yp, t, timer);
649 	struct net_device *dev = pci_get_drvdata(yp->pci_dev);
650 	void __iomem *ioaddr = yp->base;
651 	int next_tick = 60*HZ;
652 
653 	if (yellowfin_debug > 3) {
654 		netdev_printk(KERN_DEBUG, dev, "Yellowfin timer tick, status %08x\n",
655 			      ioread16(ioaddr + IntrStatus));
656 	}
657 
658 	if (yp->mii_cnt) {
659 		int bmsr = mdio_read(ioaddr, yp->phys[0], MII_BMSR);
660 		int lpa = mdio_read(ioaddr, yp->phys[0], MII_LPA);
661 		int negotiated = lpa & yp->advertising;
662 		if (yellowfin_debug > 1)
663 			netdev_printk(KERN_DEBUG, dev, "MII #%d status register is %04x, link partner capability %04x\n",
664 				      yp->phys[0], bmsr, lpa);
665 
666 		yp->full_duplex = mii_duplex(yp->duplex_lock, negotiated);
667 
668 		iowrite16(0x101C | (yp->full_duplex ? 2 : 0), ioaddr + Cnfg);
669 
670 		if (bmsr & BMSR_LSTATUS)
671 			next_tick = 60*HZ;
672 		else
673 			next_tick = 3*HZ;
674 	}
675 
676 	yp->timer.expires = jiffies + next_tick;
677 	add_timer(&yp->timer);
678 }
679 
680 static void yellowfin_tx_timeout(struct net_device *dev, unsigned int txqueue)
681 {
682 	struct yellowfin_private *yp = netdev_priv(dev);
683 	void __iomem *ioaddr = yp->base;
684 
685 	netdev_warn(dev, "Yellowfin transmit timed out at %d/%d Tx status %04x, Rx status %04x, resetting...\n",
686 		    yp->cur_tx, yp->dirty_tx,
687 		    ioread32(ioaddr + TxStatus),
688 		    ioread32(ioaddr + RxStatus));
689 
690 	/* Note: these should be KERN_DEBUG. */
691 	if (yellowfin_debug) {
692 		int i;
693 		pr_warn("  Rx ring %p: ", yp->rx_ring);
694 		for (i = 0; i < RX_RING_SIZE; i++)
695 			pr_cont(" %08x", yp->rx_ring[i].result_status);
696 		pr_cont("\n");
697 		pr_warn("  Tx ring %p: ", yp->tx_ring);
698 		for (i = 0; i < TX_RING_SIZE; i++)
699 			pr_cont(" %04x /%08x",
700 			       yp->tx_status[i].tx_errs,
701 			       yp->tx_ring[i].result_status);
702 		pr_cont("\n");
703 	}
704 
705 	/* If the hardware is found to hang regularly, we will update the code
706 	   to reinitialize the chip here. */
707 	dev->if_port = 0;
708 
709 	/* Wake the potentially-idle transmit channel. */
710 	iowrite32(0x10001000, yp->base + TxCtrl);
711 	if (yp->cur_tx - yp->dirty_tx < TX_QUEUE_SIZE)
712 		netif_wake_queue (dev);		/* Typical path */
713 
714 	netif_trans_update(dev); /* prevent tx timeout */
715 	dev->stats.tx_errors++;
716 }
717 
718 /* Initialize the Rx and Tx rings, along with various 'dev' bits. */
719 static int yellowfin_init_ring(struct net_device *dev)
720 {
721 	struct yellowfin_private *yp = netdev_priv(dev);
722 	int i, j;
723 
724 	yp->tx_full = 0;
725 	yp->cur_rx = yp->cur_tx = 0;
726 	yp->dirty_tx = 0;
727 
728 	yp->rx_buf_sz = (dev->mtu <= 1500 ? PKT_BUF_SZ : dev->mtu + 32);
729 
730 	for (i = 0; i < RX_RING_SIZE; i++) {
731 		yp->rx_ring[i].dbdma_cmd =
732 			cpu_to_le32(CMD_RX_BUF | INTR_ALWAYS | yp->rx_buf_sz);
733 		yp->rx_ring[i].branch_addr = cpu_to_le32(yp->rx_ring_dma +
734 			((i+1)%RX_RING_SIZE)*sizeof(struct yellowfin_desc));
735 	}
736 
737 	for (i = 0; i < RX_RING_SIZE; i++) {
738 		struct sk_buff *skb = netdev_alloc_skb(dev, yp->rx_buf_sz + 2);
739 		yp->rx_skbuff[i] = skb;
740 		if (skb == NULL)
741 			break;
742 		skb_reserve(skb, 2);	/* 16 byte align the IP header. */
743 		yp->rx_ring[i].addr = cpu_to_le32(pci_map_single(yp->pci_dev,
744 			skb->data, yp->rx_buf_sz, PCI_DMA_FROMDEVICE));
745 	}
746 	if (i != RX_RING_SIZE) {
747 		for (j = 0; j < i; j++)
748 			dev_kfree_skb(yp->rx_skbuff[j]);
749 		return -ENOMEM;
750 	}
751 	yp->rx_ring[i-1].dbdma_cmd = cpu_to_le32(CMD_STOP);
752 	yp->dirty_rx = (unsigned int)(i - RX_RING_SIZE);
753 
754 #define NO_TXSTATS
755 #ifdef NO_TXSTATS
756 	/* In this mode the Tx ring needs only a single descriptor. */
757 	for (i = 0; i < TX_RING_SIZE; i++) {
758 		yp->tx_skbuff[i] = NULL;
759 		yp->tx_ring[i].dbdma_cmd = cpu_to_le32(CMD_STOP);
760 		yp->tx_ring[i].branch_addr = cpu_to_le32(yp->tx_ring_dma +
761 			((i+1)%TX_RING_SIZE)*sizeof(struct yellowfin_desc));
762 	}
763 	/* Wrap ring */
764 	yp->tx_ring[--i].dbdma_cmd = cpu_to_le32(CMD_STOP | BRANCH_ALWAYS);
765 #else
766 {
767 	/* Tx ring needs a pair of descriptors, the second for the status. */
768 	for (i = 0; i < TX_RING_SIZE; i++) {
769 		j = 2*i;
770 		yp->tx_skbuff[i] = 0;
771 		/* Branch on Tx error. */
772 		yp->tx_ring[j].dbdma_cmd = cpu_to_le32(CMD_STOP);
773 		yp->tx_ring[j].branch_addr = cpu_to_le32(yp->tx_ring_dma +
774 			(j+1)*sizeof(struct yellowfin_desc));
775 		j++;
776 		if (yp->flags & FullTxStatus) {
777 			yp->tx_ring[j].dbdma_cmd =
778 				cpu_to_le32(CMD_TXSTATUS | sizeof(*yp->tx_status));
779 			yp->tx_ring[j].request_cnt = sizeof(*yp->tx_status);
780 			yp->tx_ring[j].addr = cpu_to_le32(yp->tx_status_dma +
781 				i*sizeof(struct tx_status_words));
782 		} else {
783 			/* Symbios chips write only tx_errs word. */
784 			yp->tx_ring[j].dbdma_cmd =
785 				cpu_to_le32(CMD_TXSTATUS | INTR_ALWAYS | 2);
786 			yp->tx_ring[j].request_cnt = 2;
787 			/* Om pade ummmmm... */
788 			yp->tx_ring[j].addr = cpu_to_le32(yp->tx_status_dma +
789 				i*sizeof(struct tx_status_words) +
790 				&(yp->tx_status[0].tx_errs) -
791 				&(yp->tx_status[0]));
792 		}
793 		yp->tx_ring[j].branch_addr = cpu_to_le32(yp->tx_ring_dma +
794 			((j+1)%(2*TX_RING_SIZE))*sizeof(struct yellowfin_desc));
795 	}
796 	/* Wrap ring */
797 	yp->tx_ring[++j].dbdma_cmd |= cpu_to_le32(BRANCH_ALWAYS | INTR_ALWAYS);
798 }
799 #endif
800 	yp->tx_tail_desc = &yp->tx_status[0];
801 	return 0;
802 }
803 
804 static netdev_tx_t yellowfin_start_xmit(struct sk_buff *skb,
805 					struct net_device *dev)
806 {
807 	struct yellowfin_private *yp = netdev_priv(dev);
808 	unsigned entry;
809 	int len = skb->len;
810 
811 	netif_stop_queue (dev);
812 
813 	/* Note: Ordering is important here, set the field with the
814 	   "ownership" bit last, and only then increment cur_tx. */
815 
816 	/* Calculate the next Tx descriptor entry. */
817 	entry = yp->cur_tx % TX_RING_SIZE;
818 
819 	if (gx_fix) {	/* Note: only works for paddable protocols e.g.  IP. */
820 		int cacheline_end = ((unsigned long)skb->data + skb->len) % 32;
821 		/* Fix GX chipset errata. */
822 		if (cacheline_end > 24  || cacheline_end == 0) {
823 			len = skb->len + 32 - cacheline_end + 1;
824 			if (skb_padto(skb, len)) {
825 				yp->tx_skbuff[entry] = NULL;
826 				netif_wake_queue(dev);
827 				return NETDEV_TX_OK;
828 			}
829 		}
830 	}
831 	yp->tx_skbuff[entry] = skb;
832 
833 #ifdef NO_TXSTATS
834 	yp->tx_ring[entry].addr = cpu_to_le32(pci_map_single(yp->pci_dev,
835 		skb->data, len, PCI_DMA_TODEVICE));
836 	yp->tx_ring[entry].result_status = 0;
837 	if (entry >= TX_RING_SIZE-1) {
838 		/* New stop command. */
839 		yp->tx_ring[0].dbdma_cmd = cpu_to_le32(CMD_STOP);
840 		yp->tx_ring[TX_RING_SIZE-1].dbdma_cmd =
841 			cpu_to_le32(CMD_TX_PKT|BRANCH_ALWAYS | len);
842 	} else {
843 		yp->tx_ring[entry+1].dbdma_cmd = cpu_to_le32(CMD_STOP);
844 		yp->tx_ring[entry].dbdma_cmd =
845 			cpu_to_le32(CMD_TX_PKT | BRANCH_IFTRUE | len);
846 	}
847 	yp->cur_tx++;
848 #else
849 	yp->tx_ring[entry<<1].request_cnt = len;
850 	yp->tx_ring[entry<<1].addr = cpu_to_le32(pci_map_single(yp->pci_dev,
851 		skb->data, len, PCI_DMA_TODEVICE));
852 	/* The input_last (status-write) command is constant, but we must
853 	   rewrite the subsequent 'stop' command. */
854 
855 	yp->cur_tx++;
856 	{
857 		unsigned next_entry = yp->cur_tx % TX_RING_SIZE;
858 		yp->tx_ring[next_entry<<1].dbdma_cmd = cpu_to_le32(CMD_STOP);
859 	}
860 	/* Final step -- overwrite the old 'stop' command. */
861 
862 	yp->tx_ring[entry<<1].dbdma_cmd =
863 		cpu_to_le32( ((entry % 6) == 0 ? CMD_TX_PKT|INTR_ALWAYS|BRANCH_IFTRUE :
864 					  CMD_TX_PKT | BRANCH_IFTRUE) | len);
865 #endif
866 
867 	/* Non-x86 Todo: explicitly flush cache lines here. */
868 
869 	/* Wake the potentially-idle transmit channel. */
870 	iowrite32(0x10001000, yp->base + TxCtrl);
871 
872 	if (yp->cur_tx - yp->dirty_tx < TX_QUEUE_SIZE)
873 		netif_start_queue (dev);		/* Typical path */
874 	else
875 		yp->tx_full = 1;
876 
877 	if (yellowfin_debug > 4) {
878 		netdev_printk(KERN_DEBUG, dev, "Yellowfin transmit frame #%d queued in slot %d\n",
879 			      yp->cur_tx, entry);
880 	}
881 	return NETDEV_TX_OK;
882 }
883 
884 /* The interrupt handler does all of the Rx thread work and cleans up
885    after the Tx thread. */
886 static irqreturn_t yellowfin_interrupt(int irq, void *dev_instance)
887 {
888 	struct net_device *dev = dev_instance;
889 	struct yellowfin_private *yp;
890 	void __iomem *ioaddr;
891 	int boguscnt = max_interrupt_work;
892 	unsigned int handled = 0;
893 
894 	yp = netdev_priv(dev);
895 	ioaddr = yp->base;
896 
897 	spin_lock (&yp->lock);
898 
899 	do {
900 		u16 intr_status = ioread16(ioaddr + IntrClear);
901 
902 		if (yellowfin_debug > 4)
903 			netdev_printk(KERN_DEBUG, dev, "Yellowfin interrupt, status %04x\n",
904 				      intr_status);
905 
906 		if (intr_status == 0)
907 			break;
908 		handled = 1;
909 
910 		if (intr_status & (IntrRxDone | IntrEarlyRx)) {
911 			yellowfin_rx(dev);
912 			iowrite32(0x10001000, ioaddr + RxCtrl);		/* Wake Rx engine. */
913 		}
914 
915 #ifdef NO_TXSTATS
916 		for (; yp->cur_tx - yp->dirty_tx > 0; yp->dirty_tx++) {
917 			int entry = yp->dirty_tx % TX_RING_SIZE;
918 			struct sk_buff *skb;
919 
920 			if (yp->tx_ring[entry].result_status == 0)
921 				break;
922 			skb = yp->tx_skbuff[entry];
923 			dev->stats.tx_packets++;
924 			dev->stats.tx_bytes += skb->len;
925 			/* Free the original skb. */
926 			pci_unmap_single(yp->pci_dev, le32_to_cpu(yp->tx_ring[entry].addr),
927 				skb->len, PCI_DMA_TODEVICE);
928 			dev_consume_skb_irq(skb);
929 			yp->tx_skbuff[entry] = NULL;
930 		}
931 		if (yp->tx_full &&
932 		    yp->cur_tx - yp->dirty_tx < TX_QUEUE_SIZE - 4) {
933 			/* The ring is no longer full, clear tbusy. */
934 			yp->tx_full = 0;
935 			netif_wake_queue(dev);
936 		}
937 #else
938 		if ((intr_status & IntrTxDone) || (yp->tx_tail_desc->tx_errs)) {
939 			unsigned dirty_tx = yp->dirty_tx;
940 
941 			for (dirty_tx = yp->dirty_tx; yp->cur_tx - dirty_tx > 0;
942 				 dirty_tx++) {
943 				/* Todo: optimize this. */
944 				int entry = dirty_tx % TX_RING_SIZE;
945 				u16 tx_errs = yp->tx_status[entry].tx_errs;
946 				struct sk_buff *skb;
947 
948 #ifndef final_version
949 				if (yellowfin_debug > 5)
950 					netdev_printk(KERN_DEBUG, dev, "Tx queue %d check, Tx status %04x %04x %04x %04x\n",
951 						      entry,
952 						      yp->tx_status[entry].tx_cnt,
953 						      yp->tx_status[entry].tx_errs,
954 						      yp->tx_status[entry].total_tx_cnt,
955 						      yp->tx_status[entry].paused);
956 #endif
957 				if (tx_errs == 0)
958 					break;	/* It still hasn't been Txed */
959 				skb = yp->tx_skbuff[entry];
960 				if (tx_errs & 0xF810) {
961 					/* There was an major error, log it. */
962 #ifndef final_version
963 					if (yellowfin_debug > 1)
964 						netdev_printk(KERN_DEBUG, dev, "Transmit error, Tx status %04x\n",
965 							      tx_errs);
966 #endif
967 					dev->stats.tx_errors++;
968 					if (tx_errs & 0xF800) dev->stats.tx_aborted_errors++;
969 					if (tx_errs & 0x0800) dev->stats.tx_carrier_errors++;
970 					if (tx_errs & 0x2000) dev->stats.tx_window_errors++;
971 					if (tx_errs & 0x8000) dev->stats.tx_fifo_errors++;
972 				} else {
973 #ifndef final_version
974 					if (yellowfin_debug > 4)
975 						netdev_printk(KERN_DEBUG, dev, "Normal transmit, Tx status %04x\n",
976 							      tx_errs);
977 #endif
978 					dev->stats.tx_bytes += skb->len;
979 					dev->stats.collisions += tx_errs & 15;
980 					dev->stats.tx_packets++;
981 				}
982 				/* Free the original skb. */
983 				pci_unmap_single(yp->pci_dev,
984 					yp->tx_ring[entry<<1].addr, skb->len,
985 					PCI_DMA_TODEVICE);
986 				dev_consume_skb_irq(skb);
987 				yp->tx_skbuff[entry] = 0;
988 				/* Mark status as empty. */
989 				yp->tx_status[entry].tx_errs = 0;
990 			}
991 
992 #ifndef final_version
993 			if (yp->cur_tx - dirty_tx > TX_RING_SIZE) {
994 				netdev_err(dev, "Out-of-sync dirty pointer, %d vs. %d, full=%d\n",
995 					   dirty_tx, yp->cur_tx, yp->tx_full);
996 				dirty_tx += TX_RING_SIZE;
997 			}
998 #endif
999 
1000 			if (yp->tx_full &&
1001 			    yp->cur_tx - dirty_tx < TX_QUEUE_SIZE - 2) {
1002 				/* The ring is no longer full, clear tbusy. */
1003 				yp->tx_full = 0;
1004 				netif_wake_queue(dev);
1005 			}
1006 
1007 			yp->dirty_tx = dirty_tx;
1008 			yp->tx_tail_desc = &yp->tx_status[dirty_tx % TX_RING_SIZE];
1009 		}
1010 #endif
1011 
1012 		/* Log errors and other uncommon events. */
1013 		if (intr_status & 0x2ee)	/* Abnormal error summary. */
1014 			yellowfin_error(dev, intr_status);
1015 
1016 		if (--boguscnt < 0) {
1017 			netdev_warn(dev, "Too much work at interrupt, status=%#04x\n",
1018 				    intr_status);
1019 			break;
1020 		}
1021 	} while (1);
1022 
1023 	if (yellowfin_debug > 3)
1024 		netdev_printk(KERN_DEBUG, dev, "exiting interrupt, status=%#04x\n",
1025 			      ioread16(ioaddr + IntrStatus));
1026 
1027 	spin_unlock (&yp->lock);
1028 	return IRQ_RETVAL(handled);
1029 }
1030 
1031 /* This routine is logically part of the interrupt handler, but separated
1032    for clarity and better register allocation. */
1033 static int yellowfin_rx(struct net_device *dev)
1034 {
1035 	struct yellowfin_private *yp = netdev_priv(dev);
1036 	int entry = yp->cur_rx % RX_RING_SIZE;
1037 	int boguscnt = yp->dirty_rx + RX_RING_SIZE - yp->cur_rx;
1038 
1039 	if (yellowfin_debug > 4) {
1040 		printk(KERN_DEBUG " In yellowfin_rx(), entry %d status %08x\n",
1041 			   entry, yp->rx_ring[entry].result_status);
1042 		printk(KERN_DEBUG "   #%d desc. %08x %08x %08x\n",
1043 			   entry, yp->rx_ring[entry].dbdma_cmd, yp->rx_ring[entry].addr,
1044 			   yp->rx_ring[entry].result_status);
1045 	}
1046 
1047 	/* If EOP is set on the next entry, it's a new packet. Send it up. */
1048 	while (1) {
1049 		struct yellowfin_desc *desc = &yp->rx_ring[entry];
1050 		struct sk_buff *rx_skb = yp->rx_skbuff[entry];
1051 		s16 frame_status;
1052 		u16 desc_status;
1053 		int data_size, yf_size;
1054 		u8 *buf_addr;
1055 
1056 		if(!desc->result_status)
1057 			break;
1058 		pci_dma_sync_single_for_cpu(yp->pci_dev, le32_to_cpu(desc->addr),
1059 			yp->rx_buf_sz, PCI_DMA_FROMDEVICE);
1060 		desc_status = le32_to_cpu(desc->result_status) >> 16;
1061 		buf_addr = rx_skb->data;
1062 		data_size = (le32_to_cpu(desc->dbdma_cmd) -
1063 			le32_to_cpu(desc->result_status)) & 0xffff;
1064 		frame_status = get_unaligned_le16(&(buf_addr[data_size - 2]));
1065 		if (yellowfin_debug > 4)
1066 			printk(KERN_DEBUG "  %s() status was %04x\n",
1067 			       __func__, frame_status);
1068 		if (--boguscnt < 0)
1069 			break;
1070 
1071 		yf_size = sizeof(struct yellowfin_desc);
1072 
1073 		if ( ! (desc_status & RX_EOP)) {
1074 			if (data_size != 0)
1075 				netdev_warn(dev, "Oversized Ethernet frame spanned multiple buffers, status %04x, data_size %d!\n",
1076 					    desc_status, data_size);
1077 			dev->stats.rx_length_errors++;
1078 		} else if ((yp->drv_flags & IsGigabit)  &&  (frame_status & 0x0038)) {
1079 			/* There was a error. */
1080 			if (yellowfin_debug > 3)
1081 				printk(KERN_DEBUG "  %s() Rx error was %04x\n",
1082 				       __func__, frame_status);
1083 			dev->stats.rx_errors++;
1084 			if (frame_status & 0x0060) dev->stats.rx_length_errors++;
1085 			if (frame_status & 0x0008) dev->stats.rx_frame_errors++;
1086 			if (frame_status & 0x0010) dev->stats.rx_crc_errors++;
1087 			if (frame_status < 0) dev->stats.rx_dropped++;
1088 		} else if ( !(yp->drv_flags & IsGigabit)  &&
1089 				   ((buf_addr[data_size-1] & 0x85) || buf_addr[data_size-2] & 0xC0)) {
1090 			u8 status1 = buf_addr[data_size-2];
1091 			u8 status2 = buf_addr[data_size-1];
1092 			dev->stats.rx_errors++;
1093 			if (status1 & 0xC0) dev->stats.rx_length_errors++;
1094 			if (status2 & 0x03) dev->stats.rx_frame_errors++;
1095 			if (status2 & 0x04) dev->stats.rx_crc_errors++;
1096 			if (status2 & 0x80) dev->stats.rx_dropped++;
1097 #ifdef YF_PROTOTYPE		/* Support for prototype hardware errata. */
1098 		} else if ((yp->flags & HasMACAddrBug)  &&
1099 			!ether_addr_equal(le32_to_cpu(yp->rx_ring_dma +
1100 						      entry * yf_size),
1101 					  dev->dev_addr) &&
1102 			!ether_addr_equal(le32_to_cpu(yp->rx_ring_dma +
1103 						      entry * yf_size),
1104 					  "\377\377\377\377\377\377")) {
1105 			if (bogus_rx++ == 0)
1106 				netdev_warn(dev, "Bad frame to %pM\n",
1107 					    buf_addr);
1108 #endif
1109 		} else {
1110 			struct sk_buff *skb;
1111 			int pkt_len = data_size -
1112 				(yp->chip_id ? 7 : 8 + buf_addr[data_size - 8]);
1113 			/* To verify: Yellowfin Length should omit the CRC! */
1114 
1115 #ifndef final_version
1116 			if (yellowfin_debug > 4)
1117 				printk(KERN_DEBUG "  %s() normal Rx pkt length %d of %d, bogus_cnt %d\n",
1118 				       __func__, pkt_len, data_size, boguscnt);
1119 #endif
1120 			/* Check if the packet is long enough to just pass up the skbuff
1121 			   without copying to a properly sized skbuff. */
1122 			if (pkt_len > rx_copybreak) {
1123 				skb_put(skb = rx_skb, pkt_len);
1124 				pci_unmap_single(yp->pci_dev,
1125 					le32_to_cpu(yp->rx_ring[entry].addr),
1126 					yp->rx_buf_sz,
1127 					PCI_DMA_FROMDEVICE);
1128 				yp->rx_skbuff[entry] = NULL;
1129 			} else {
1130 				skb = netdev_alloc_skb(dev, pkt_len + 2);
1131 				if (skb == NULL)
1132 					break;
1133 				skb_reserve(skb, 2);	/* 16 byte align the IP header */
1134 				skb_copy_to_linear_data(skb, rx_skb->data, pkt_len);
1135 				skb_put(skb, pkt_len);
1136 				pci_dma_sync_single_for_device(yp->pci_dev,
1137 								le32_to_cpu(desc->addr),
1138 								yp->rx_buf_sz,
1139 								PCI_DMA_FROMDEVICE);
1140 			}
1141 			skb->protocol = eth_type_trans(skb, dev);
1142 			netif_rx(skb);
1143 			dev->stats.rx_packets++;
1144 			dev->stats.rx_bytes += pkt_len;
1145 		}
1146 		entry = (++yp->cur_rx) % RX_RING_SIZE;
1147 	}
1148 
1149 	/* Refill the Rx ring buffers. */
1150 	for (; yp->cur_rx - yp->dirty_rx > 0; yp->dirty_rx++) {
1151 		entry = yp->dirty_rx % RX_RING_SIZE;
1152 		if (yp->rx_skbuff[entry] == NULL) {
1153 			struct sk_buff *skb = netdev_alloc_skb(dev, yp->rx_buf_sz + 2);
1154 			if (skb == NULL)
1155 				break;				/* Better luck next round. */
1156 			yp->rx_skbuff[entry] = skb;
1157 			skb_reserve(skb, 2);	/* Align IP on 16 byte boundaries */
1158 			yp->rx_ring[entry].addr = cpu_to_le32(pci_map_single(yp->pci_dev,
1159 				skb->data, yp->rx_buf_sz, PCI_DMA_FROMDEVICE));
1160 		}
1161 		yp->rx_ring[entry].dbdma_cmd = cpu_to_le32(CMD_STOP);
1162 		yp->rx_ring[entry].result_status = 0;	/* Clear complete bit. */
1163 		if (entry != 0)
1164 			yp->rx_ring[entry - 1].dbdma_cmd =
1165 				cpu_to_le32(CMD_RX_BUF | INTR_ALWAYS | yp->rx_buf_sz);
1166 		else
1167 			yp->rx_ring[RX_RING_SIZE - 1].dbdma_cmd =
1168 				cpu_to_le32(CMD_RX_BUF | INTR_ALWAYS | BRANCH_ALWAYS
1169 							| yp->rx_buf_sz);
1170 	}
1171 
1172 	return 0;
1173 }
1174 
1175 static void yellowfin_error(struct net_device *dev, int intr_status)
1176 {
1177 	netdev_err(dev, "Something Wicked happened! %04x\n", intr_status);
1178 	/* Hmmmmm, it's not clear what to do here. */
1179 	if (intr_status & (IntrTxPCIErr | IntrTxPCIFault))
1180 		dev->stats.tx_errors++;
1181 	if (intr_status & (IntrRxPCIErr | IntrRxPCIFault))
1182 		dev->stats.rx_errors++;
1183 }
1184 
1185 static int yellowfin_close(struct net_device *dev)
1186 {
1187 	struct yellowfin_private *yp = netdev_priv(dev);
1188 	void __iomem *ioaddr = yp->base;
1189 	int i;
1190 
1191 	netif_stop_queue (dev);
1192 
1193 	if (yellowfin_debug > 1) {
1194 		netdev_printk(KERN_DEBUG, dev, "Shutting down ethercard, status was Tx %04x Rx %04x Int %02x\n",
1195 			      ioread16(ioaddr + TxStatus),
1196 			      ioread16(ioaddr + RxStatus),
1197 			      ioread16(ioaddr + IntrStatus));
1198 		netdev_printk(KERN_DEBUG, dev, "Queue pointers were Tx %d / %d,  Rx %d / %d\n",
1199 			      yp->cur_tx, yp->dirty_tx,
1200 			      yp->cur_rx, yp->dirty_rx);
1201 	}
1202 
1203 	/* Disable interrupts by clearing the interrupt mask. */
1204 	iowrite16(0x0000, ioaddr + IntrEnb);
1205 
1206 	/* Stop the chip's Tx and Rx processes. */
1207 	iowrite32(0x80000000, ioaddr + RxCtrl);
1208 	iowrite32(0x80000000, ioaddr + TxCtrl);
1209 
1210 	del_timer(&yp->timer);
1211 
1212 #if defined(__i386__)
1213 	if (yellowfin_debug > 2) {
1214 		printk(KERN_DEBUG "  Tx ring at %08llx:\n",
1215 				(unsigned long long)yp->tx_ring_dma);
1216 		for (i = 0; i < TX_RING_SIZE*2; i++)
1217 			printk(KERN_DEBUG " %c #%d desc. %08x %08x %08x %08x\n",
1218 				   ioread32(ioaddr + TxPtr) == (long)&yp->tx_ring[i] ? '>' : ' ',
1219 				   i, yp->tx_ring[i].dbdma_cmd, yp->tx_ring[i].addr,
1220 				   yp->tx_ring[i].branch_addr, yp->tx_ring[i].result_status);
1221 		printk(KERN_DEBUG "  Tx status %p:\n", yp->tx_status);
1222 		for (i = 0; i < TX_RING_SIZE; i++)
1223 			printk(KERN_DEBUG "   #%d status %04x %04x %04x %04x\n",
1224 				   i, yp->tx_status[i].tx_cnt, yp->tx_status[i].tx_errs,
1225 				   yp->tx_status[i].total_tx_cnt, yp->tx_status[i].paused);
1226 
1227 		printk(KERN_DEBUG "  Rx ring %08llx:\n",
1228 				(unsigned long long)yp->rx_ring_dma);
1229 		for (i = 0; i < RX_RING_SIZE; i++) {
1230 			printk(KERN_DEBUG " %c #%d desc. %08x %08x %08x\n",
1231 				   ioread32(ioaddr + RxPtr) == (long)&yp->rx_ring[i] ? '>' : ' ',
1232 				   i, yp->rx_ring[i].dbdma_cmd, yp->rx_ring[i].addr,
1233 				   yp->rx_ring[i].result_status);
1234 			if (yellowfin_debug > 6) {
1235 				if (get_unaligned((u8*)yp->rx_ring[i].addr) != 0x69) {
1236 					int j;
1237 
1238 					printk(KERN_DEBUG);
1239 					for (j = 0; j < 0x50; j++)
1240 						pr_cont(" %04x",
1241 							get_unaligned(((u16*)yp->rx_ring[i].addr) + j));
1242 					pr_cont("\n");
1243 				}
1244 			}
1245 		}
1246 	}
1247 #endif /* __i386__ debugging only */
1248 
1249 	free_irq(yp->pci_dev->irq, dev);
1250 
1251 	/* Free all the skbuffs in the Rx queue. */
1252 	for (i = 0; i < RX_RING_SIZE; i++) {
1253 		yp->rx_ring[i].dbdma_cmd = cpu_to_le32(CMD_STOP);
1254 		yp->rx_ring[i].addr = cpu_to_le32(0xBADF00D0); /* An invalid address. */
1255 		if (yp->rx_skbuff[i]) {
1256 			dev_kfree_skb(yp->rx_skbuff[i]);
1257 		}
1258 		yp->rx_skbuff[i] = NULL;
1259 	}
1260 	for (i = 0; i < TX_RING_SIZE; i++) {
1261 		dev_kfree_skb(yp->tx_skbuff[i]);
1262 		yp->tx_skbuff[i] = NULL;
1263 	}
1264 
1265 #ifdef YF_PROTOTYPE			/* Support for prototype hardware errata. */
1266 	if (yellowfin_debug > 0) {
1267 		netdev_printk(KERN_DEBUG, dev, "Received %d frames that we should not have\n",
1268 			      bogus_rx);
1269 	}
1270 #endif
1271 
1272 	return 0;
1273 }
1274 
1275 /* Set or clear the multicast filter for this adaptor. */
1276 
1277 static void set_rx_mode(struct net_device *dev)
1278 {
1279 	struct yellowfin_private *yp = netdev_priv(dev);
1280 	void __iomem *ioaddr = yp->base;
1281 	u16 cfg_value = ioread16(ioaddr + Cnfg);
1282 
1283 	/* Stop the Rx process to change any value. */
1284 	iowrite16(cfg_value & ~0x1000, ioaddr + Cnfg);
1285 	if (dev->flags & IFF_PROMISC) {			/* Set promiscuous. */
1286 		iowrite16(0x000F, ioaddr + AddrMode);
1287 	} else if ((netdev_mc_count(dev) > 64) ||
1288 		   (dev->flags & IFF_ALLMULTI)) {
1289 		/* Too many to filter well, or accept all multicasts. */
1290 		iowrite16(0x000B, ioaddr + AddrMode);
1291 	} else if (!netdev_mc_empty(dev)) { /* Must use the multicast hash table. */
1292 		struct netdev_hw_addr *ha;
1293 		u16 hash_table[4];
1294 		int i;
1295 
1296 		memset(hash_table, 0, sizeof(hash_table));
1297 		netdev_for_each_mc_addr(ha, dev) {
1298 			unsigned int bit;
1299 
1300 			/* Due to a bug in the early chip versions, multiple filter
1301 			   slots must be set for each address. */
1302 			if (yp->drv_flags & HasMulticastBug) {
1303 				bit = (ether_crc_le(3, ha->addr) >> 3) & 0x3f;
1304 				hash_table[bit >> 4] |= (1 << bit);
1305 				bit = (ether_crc_le(4, ha->addr) >> 3) & 0x3f;
1306 				hash_table[bit >> 4] |= (1 << bit);
1307 				bit = (ether_crc_le(5, ha->addr) >> 3) & 0x3f;
1308 				hash_table[bit >> 4] |= (1 << bit);
1309 			}
1310 			bit = (ether_crc_le(6, ha->addr) >> 3) & 0x3f;
1311 			hash_table[bit >> 4] |= (1 << bit);
1312 		}
1313 		/* Copy the hash table to the chip. */
1314 		for (i = 0; i < 4; i++)
1315 			iowrite16(hash_table[i], ioaddr + HashTbl + i*2);
1316 		iowrite16(0x0003, ioaddr + AddrMode);
1317 	} else {					/* Normal, unicast/broadcast-only mode. */
1318 		iowrite16(0x0001, ioaddr + AddrMode);
1319 	}
1320 	/* Restart the Rx process. */
1321 	iowrite16(cfg_value | 0x1000, ioaddr + Cnfg);
1322 }
1323 
1324 static void yellowfin_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1325 {
1326 	struct yellowfin_private *np = netdev_priv(dev);
1327 
1328 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1329 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1330 	strlcpy(info->bus_info, pci_name(np->pci_dev), sizeof(info->bus_info));
1331 }
1332 
1333 static const struct ethtool_ops ethtool_ops = {
1334 	.get_drvinfo = yellowfin_get_drvinfo
1335 };
1336 
1337 static int netdev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
1338 {
1339 	struct yellowfin_private *np = netdev_priv(dev);
1340 	void __iomem *ioaddr = np->base;
1341 	struct mii_ioctl_data *data = if_mii(rq);
1342 
1343 	switch(cmd) {
1344 	case SIOCGMIIPHY:		/* Get address of MII PHY in use. */
1345 		data->phy_id = np->phys[0] & 0x1f;
1346 		/* Fall Through */
1347 
1348 	case SIOCGMIIREG:		/* Read MII PHY register. */
1349 		data->val_out = mdio_read(ioaddr, data->phy_id & 0x1f, data->reg_num & 0x1f);
1350 		return 0;
1351 
1352 	case SIOCSMIIREG:		/* Write MII PHY register. */
1353 		if (data->phy_id == np->phys[0]) {
1354 			u16 value = data->val_in;
1355 			switch (data->reg_num) {
1356 			case 0:
1357 				/* Check for autonegotiation on or reset. */
1358 				np->medialock = (value & 0x9000) ? 0 : 1;
1359 				if (np->medialock)
1360 					np->full_duplex = (value & 0x0100) ? 1 : 0;
1361 				break;
1362 			case 4: np->advertising = value; break;
1363 			}
1364 			/* Perhaps check_duplex(dev), depending on chip semantics. */
1365 		}
1366 		mdio_write(ioaddr, data->phy_id & 0x1f, data->reg_num & 0x1f, data->val_in);
1367 		return 0;
1368 	default:
1369 		return -EOPNOTSUPP;
1370 	}
1371 }
1372 
1373 
1374 static void yellowfin_remove_one(struct pci_dev *pdev)
1375 {
1376 	struct net_device *dev = pci_get_drvdata(pdev);
1377 	struct yellowfin_private *np;
1378 
1379 	BUG_ON(!dev);
1380 	np = netdev_priv(dev);
1381 
1382         pci_free_consistent(pdev, STATUS_TOTAL_SIZE, np->tx_status,
1383 		np->tx_status_dma);
1384 	pci_free_consistent(pdev, RX_TOTAL_SIZE, np->rx_ring, np->rx_ring_dma);
1385 	pci_free_consistent(pdev, TX_TOTAL_SIZE, np->tx_ring, np->tx_ring_dma);
1386 	unregister_netdev (dev);
1387 
1388 	pci_iounmap(pdev, np->base);
1389 
1390 	pci_release_regions (pdev);
1391 
1392 	free_netdev (dev);
1393 }
1394 
1395 
1396 static struct pci_driver yellowfin_driver = {
1397 	.name		= DRV_NAME,
1398 	.id_table	= yellowfin_pci_tbl,
1399 	.probe		= yellowfin_init_one,
1400 	.remove		= yellowfin_remove_one,
1401 };
1402 
1403 
1404 static int __init yellowfin_init (void)
1405 {
1406 /* when a module, this is printed whether or not devices are found in probe */
1407 #ifdef MODULE
1408 	printk(version);
1409 #endif
1410 	return pci_register_driver(&yellowfin_driver);
1411 }
1412 
1413 
1414 static void __exit yellowfin_cleanup (void)
1415 {
1416 	pci_unregister_driver (&yellowfin_driver);
1417 }
1418 
1419 
1420 module_init(yellowfin_init);
1421 module_exit(yellowfin_cleanup);
1422