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 = dma_alloc_coherent(&pdev->dev, TX_TOTAL_SIZE, &ring_dma,
438 					GFP_KERNEL);
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 = dma_alloc_coherent(&pdev->dev, RX_TOTAL_SIZE, &ring_dma,
445 					GFP_KERNEL);
446 	if (!ring_space)
447 		goto err_out_unmap_tx;
448 	np->rx_ring = ring_space;
449 	np->rx_ring_dma = ring_dma;
450 
451 	ring_space = dma_alloc_coherent(&pdev->dev, STATUS_TOTAL_SIZE,
452 					&ring_dma, GFP_KERNEL);
453 	if (!ring_space)
454 		goto err_out_unmap_rx;
455 	np->tx_status = ring_space;
456 	np->tx_status_dma = ring_dma;
457 
458 	if (dev->mem_start)
459 		option = dev->mem_start;
460 
461 	/* The lower four bits are the media type. */
462 	if (option > 0) {
463 		if (option & 0x200)
464 			np->full_duplex = 1;
465 		np->default_port = option & 15;
466 		if (np->default_port)
467 			np->medialock = 1;
468 	}
469 	if (find_cnt < MAX_UNITS  &&  full_duplex[find_cnt] > 0)
470 		np->full_duplex = 1;
471 
472 	if (np->full_duplex)
473 		np->duplex_lock = 1;
474 
475 	/* The Yellowfin-specific entries in the device structure. */
476 	dev->netdev_ops = &netdev_ops;
477 	dev->ethtool_ops = &ethtool_ops;
478 	dev->watchdog_timeo = TX_TIMEOUT;
479 
480 	if (mtu)
481 		dev->mtu = mtu;
482 
483 	i = register_netdev(dev);
484 	if (i)
485 		goto err_out_unmap_status;
486 
487 	netdev_info(dev, "%s type %8x at %p, %pM, IRQ %d\n",
488 		    pci_id_tbl[chip_idx].name,
489 		    ioread32(ioaddr + ChipRev), ioaddr,
490 		    dev->dev_addr, irq);
491 
492 	if (np->drv_flags & HasMII) {
493 		int phy, phy_idx = 0;
494 		for (phy = 0; phy < 32 && phy_idx < MII_CNT; phy++) {
495 			int mii_status = mdio_read(ioaddr, phy, 1);
496 			if (mii_status != 0xffff  &&  mii_status != 0x0000) {
497 				np->phys[phy_idx++] = phy;
498 				np->advertising = mdio_read(ioaddr, phy, 4);
499 				netdev_info(dev, "MII PHY found at address %d, status 0x%04x advertising %04x\n",
500 					    phy, mii_status, np->advertising);
501 			}
502 		}
503 		np->mii_cnt = phy_idx;
504 	}
505 
506 	find_cnt++;
507 
508 	return 0;
509 
510 err_out_unmap_status:
511 	dma_free_coherent(&pdev->dev, STATUS_TOTAL_SIZE, np->tx_status,
512 			  np->tx_status_dma);
513 err_out_unmap_rx:
514 	dma_free_coherent(&pdev->dev, RX_TOTAL_SIZE, np->rx_ring,
515 			  np->rx_ring_dma);
516 err_out_unmap_tx:
517 	dma_free_coherent(&pdev->dev, TX_TOTAL_SIZE, np->tx_ring,
518 			  np->tx_ring_dma);
519 err_out_cleardev:
520 	pci_iounmap(pdev, ioaddr);
521 err_out_free_res:
522 	pci_release_regions(pdev);
523 err_out_free_netdev:
524 	free_netdev (dev);
525 	return -ENODEV;
526 }
527 
528 static int read_eeprom(void __iomem *ioaddr, int location)
529 {
530 	int bogus_cnt = 10000;		/* Typical 33Mhz: 1050 ticks */
531 
532 	iowrite8(location, ioaddr + EEAddr);
533 	iowrite8(0x30 | ((location >> 8) & 7), ioaddr + EECtrl);
534 	while ((ioread8(ioaddr + EEStatus) & 0x80)  &&  --bogus_cnt > 0)
535 		;
536 	return ioread8(ioaddr + EERead);
537 }
538 
539 /* MII Managemen Data I/O accesses.
540    These routines assume the MDIO controller is idle, and do not exit until
541    the command is finished. */
542 
543 static int mdio_read(void __iomem *ioaddr, int phy_id, int location)
544 {
545 	int i;
546 
547 	iowrite16((phy_id<<8) + location, ioaddr + MII_Addr);
548 	iowrite16(1, ioaddr + MII_Cmd);
549 	for (i = 10000; i >= 0; i--)
550 		if ((ioread16(ioaddr + MII_Status) & 1) == 0)
551 			break;
552 	return ioread16(ioaddr + MII_Rd_Data);
553 }
554 
555 static void mdio_write(void __iomem *ioaddr, int phy_id, int location, int value)
556 {
557 	int i;
558 
559 	iowrite16((phy_id<<8) + location, ioaddr + MII_Addr);
560 	iowrite16(value, ioaddr + MII_Wr_Data);
561 
562 	/* Wait for the command to finish. */
563 	for (i = 10000; i >= 0; i--)
564 		if ((ioread16(ioaddr + MII_Status) & 1) == 0)
565 			break;
566 }
567 
568 
569 static int yellowfin_open(struct net_device *dev)
570 {
571 	struct yellowfin_private *yp = netdev_priv(dev);
572 	const int irq = yp->pci_dev->irq;
573 	void __iomem *ioaddr = yp->base;
574 	int i, rc;
575 
576 	/* Reset the chip. */
577 	iowrite32(0x80000000, ioaddr + DMACtrl);
578 
579 	rc = request_irq(irq, yellowfin_interrupt, IRQF_SHARED, dev->name, dev);
580 	if (rc)
581 		return rc;
582 
583 	rc = yellowfin_init_ring(dev);
584 	if (rc < 0)
585 		goto err_free_irq;
586 
587 	iowrite32(yp->rx_ring_dma, ioaddr + RxPtr);
588 	iowrite32(yp->tx_ring_dma, ioaddr + TxPtr);
589 
590 	for (i = 0; i < 6; i++)
591 		iowrite8(dev->dev_addr[i], ioaddr + StnAddr + i);
592 
593 	/* Set up various condition 'select' registers.
594 	   There are no options here. */
595 	iowrite32(0x00800080, ioaddr + TxIntrSel); 	/* Interrupt on Tx abort */
596 	iowrite32(0x00800080, ioaddr + TxBranchSel);	/* Branch on Tx abort */
597 	iowrite32(0x00400040, ioaddr + TxWaitSel); 	/* Wait on Tx status */
598 	iowrite32(0x00400040, ioaddr + RxIntrSel);	/* Interrupt on Rx done */
599 	iowrite32(0x00400040, ioaddr + RxBranchSel);	/* Branch on Rx error */
600 	iowrite32(0x00400040, ioaddr + RxWaitSel);	/* Wait on Rx done */
601 
602 	/* Initialize other registers: with so many this eventually this will
603 	   converted to an offset/value list. */
604 	iowrite32(dma_ctrl, ioaddr + DMACtrl);
605 	iowrite16(fifo_cfg, ioaddr + FIFOcfg);
606 	/* Enable automatic generation of flow control frames, period 0xffff. */
607 	iowrite32(0x0030FFFF, ioaddr + FlowCtrl);
608 
609 	yp->tx_threshold = 32;
610 	iowrite32(yp->tx_threshold, ioaddr + TxThreshold);
611 
612 	if (dev->if_port == 0)
613 		dev->if_port = yp->default_port;
614 
615 	netif_start_queue(dev);
616 
617 	/* Setting the Rx mode will start the Rx process. */
618 	if (yp->drv_flags & IsGigabit) {
619 		/* We are always in full-duplex mode with gigabit! */
620 		yp->full_duplex = 1;
621 		iowrite16(0x01CF, ioaddr + Cnfg);
622 	} else {
623 		iowrite16(0x0018, ioaddr + FrameGap0); /* 0060/4060 for non-MII 10baseT */
624 		iowrite16(0x1018, ioaddr + FrameGap1);
625 		iowrite16(0x101C | (yp->full_duplex ? 2 : 0), ioaddr + Cnfg);
626 	}
627 	set_rx_mode(dev);
628 
629 	/* Enable interrupts by setting the interrupt mask. */
630 	iowrite16(0x81ff, ioaddr + IntrEnb);			/* See enum intr_status_bits */
631 	iowrite16(0x0000, ioaddr + EventStatus);		/* Clear non-interrupting events */
632 	iowrite32(0x80008000, ioaddr + RxCtrl);		/* Start Rx and Tx channels. */
633 	iowrite32(0x80008000, ioaddr + TxCtrl);
634 
635 	if (yellowfin_debug > 2) {
636 		netdev_printk(KERN_DEBUG, dev, "Done %s()\n", __func__);
637 	}
638 
639 	/* Set the timer to check for link beat. */
640 	timer_setup(&yp->timer, yellowfin_timer, 0);
641 	yp->timer.expires = jiffies + 3*HZ;
642 	add_timer(&yp->timer);
643 out:
644 	return rc;
645 
646 err_free_irq:
647 	free_irq(irq, dev);
648 	goto out;
649 }
650 
651 static void yellowfin_timer(struct timer_list *t)
652 {
653 	struct yellowfin_private *yp = from_timer(yp, t, timer);
654 	struct net_device *dev = pci_get_drvdata(yp->pci_dev);
655 	void __iomem *ioaddr = yp->base;
656 	int next_tick = 60*HZ;
657 
658 	if (yellowfin_debug > 3) {
659 		netdev_printk(KERN_DEBUG, dev, "Yellowfin timer tick, status %08x\n",
660 			      ioread16(ioaddr + IntrStatus));
661 	}
662 
663 	if (yp->mii_cnt) {
664 		int bmsr = mdio_read(ioaddr, yp->phys[0], MII_BMSR);
665 		int lpa = mdio_read(ioaddr, yp->phys[0], MII_LPA);
666 		int negotiated = lpa & yp->advertising;
667 		if (yellowfin_debug > 1)
668 			netdev_printk(KERN_DEBUG, dev, "MII #%d status register is %04x, link partner capability %04x\n",
669 				      yp->phys[0], bmsr, lpa);
670 
671 		yp->full_duplex = mii_duplex(yp->duplex_lock, negotiated);
672 
673 		iowrite16(0x101C | (yp->full_duplex ? 2 : 0), ioaddr + Cnfg);
674 
675 		if (bmsr & BMSR_LSTATUS)
676 			next_tick = 60*HZ;
677 		else
678 			next_tick = 3*HZ;
679 	}
680 
681 	yp->timer.expires = jiffies + next_tick;
682 	add_timer(&yp->timer);
683 }
684 
685 static void yellowfin_tx_timeout(struct net_device *dev, unsigned int txqueue)
686 {
687 	struct yellowfin_private *yp = netdev_priv(dev);
688 	void __iomem *ioaddr = yp->base;
689 
690 	netdev_warn(dev, "Yellowfin transmit timed out at %d/%d Tx status %04x, Rx status %04x, resetting...\n",
691 		    yp->cur_tx, yp->dirty_tx,
692 		    ioread32(ioaddr + TxStatus),
693 		    ioread32(ioaddr + RxStatus));
694 
695 	/* Note: these should be KERN_DEBUG. */
696 	if (yellowfin_debug) {
697 		int i;
698 		pr_warn("  Rx ring %p: ", yp->rx_ring);
699 		for (i = 0; i < RX_RING_SIZE; i++)
700 			pr_cont(" %08x", yp->rx_ring[i].result_status);
701 		pr_cont("\n");
702 		pr_warn("  Tx ring %p: ", yp->tx_ring);
703 		for (i = 0; i < TX_RING_SIZE; i++)
704 			pr_cont(" %04x /%08x",
705 			       yp->tx_status[i].tx_errs,
706 			       yp->tx_ring[i].result_status);
707 		pr_cont("\n");
708 	}
709 
710 	/* If the hardware is found to hang regularly, we will update the code
711 	   to reinitialize the chip here. */
712 	dev->if_port = 0;
713 
714 	/* Wake the potentially-idle transmit channel. */
715 	iowrite32(0x10001000, yp->base + TxCtrl);
716 	if (yp->cur_tx - yp->dirty_tx < TX_QUEUE_SIZE)
717 		netif_wake_queue (dev);		/* Typical path */
718 
719 	netif_trans_update(dev); /* prevent tx timeout */
720 	dev->stats.tx_errors++;
721 }
722 
723 /* Initialize the Rx and Tx rings, along with various 'dev' bits. */
724 static int yellowfin_init_ring(struct net_device *dev)
725 {
726 	struct yellowfin_private *yp = netdev_priv(dev);
727 	int i, j;
728 
729 	yp->tx_full = 0;
730 	yp->cur_rx = yp->cur_tx = 0;
731 	yp->dirty_tx = 0;
732 
733 	yp->rx_buf_sz = (dev->mtu <= 1500 ? PKT_BUF_SZ : dev->mtu + 32);
734 
735 	for (i = 0; i < RX_RING_SIZE; i++) {
736 		yp->rx_ring[i].dbdma_cmd =
737 			cpu_to_le32(CMD_RX_BUF | INTR_ALWAYS | yp->rx_buf_sz);
738 		yp->rx_ring[i].branch_addr = cpu_to_le32(yp->rx_ring_dma +
739 			((i+1)%RX_RING_SIZE)*sizeof(struct yellowfin_desc));
740 	}
741 
742 	for (i = 0; i < RX_RING_SIZE; i++) {
743 		struct sk_buff *skb = netdev_alloc_skb(dev, yp->rx_buf_sz + 2);
744 		yp->rx_skbuff[i] = skb;
745 		if (skb == NULL)
746 			break;
747 		skb_reserve(skb, 2);	/* 16 byte align the IP header. */
748 		yp->rx_ring[i].addr = cpu_to_le32(dma_map_single(&yp->pci_dev->dev,
749 								 skb->data,
750 								 yp->rx_buf_sz,
751 								 DMA_FROM_DEVICE));
752 	}
753 	if (i != RX_RING_SIZE) {
754 		for (j = 0; j < i; j++)
755 			dev_kfree_skb(yp->rx_skbuff[j]);
756 		return -ENOMEM;
757 	}
758 	yp->rx_ring[i-1].dbdma_cmd = cpu_to_le32(CMD_STOP);
759 	yp->dirty_rx = (unsigned int)(i - RX_RING_SIZE);
760 
761 #define NO_TXSTATS
762 #ifdef NO_TXSTATS
763 	/* In this mode the Tx ring needs only a single descriptor. */
764 	for (i = 0; i < TX_RING_SIZE; i++) {
765 		yp->tx_skbuff[i] = NULL;
766 		yp->tx_ring[i].dbdma_cmd = cpu_to_le32(CMD_STOP);
767 		yp->tx_ring[i].branch_addr = cpu_to_le32(yp->tx_ring_dma +
768 			((i+1)%TX_RING_SIZE)*sizeof(struct yellowfin_desc));
769 	}
770 	/* Wrap ring */
771 	yp->tx_ring[--i].dbdma_cmd = cpu_to_le32(CMD_STOP | BRANCH_ALWAYS);
772 #else
773 {
774 	/* Tx ring needs a pair of descriptors, the second for the status. */
775 	for (i = 0; i < TX_RING_SIZE; i++) {
776 		j = 2*i;
777 		yp->tx_skbuff[i] = 0;
778 		/* Branch on Tx error. */
779 		yp->tx_ring[j].dbdma_cmd = cpu_to_le32(CMD_STOP);
780 		yp->tx_ring[j].branch_addr = cpu_to_le32(yp->tx_ring_dma +
781 			(j+1)*sizeof(struct yellowfin_desc));
782 		j++;
783 		if (yp->flags & FullTxStatus) {
784 			yp->tx_ring[j].dbdma_cmd =
785 				cpu_to_le32(CMD_TXSTATUS | sizeof(*yp->tx_status));
786 			yp->tx_ring[j].request_cnt = sizeof(*yp->tx_status);
787 			yp->tx_ring[j].addr = cpu_to_le32(yp->tx_status_dma +
788 				i*sizeof(struct tx_status_words));
789 		} else {
790 			/* Symbios chips write only tx_errs word. */
791 			yp->tx_ring[j].dbdma_cmd =
792 				cpu_to_le32(CMD_TXSTATUS | INTR_ALWAYS | 2);
793 			yp->tx_ring[j].request_cnt = 2;
794 			/* Om pade ummmmm... */
795 			yp->tx_ring[j].addr = cpu_to_le32(yp->tx_status_dma +
796 				i*sizeof(struct tx_status_words) +
797 				&(yp->tx_status[0].tx_errs) -
798 				&(yp->tx_status[0]));
799 		}
800 		yp->tx_ring[j].branch_addr = cpu_to_le32(yp->tx_ring_dma +
801 			((j+1)%(2*TX_RING_SIZE))*sizeof(struct yellowfin_desc));
802 	}
803 	/* Wrap ring */
804 	yp->tx_ring[++j].dbdma_cmd |= cpu_to_le32(BRANCH_ALWAYS | INTR_ALWAYS);
805 }
806 #endif
807 	yp->tx_tail_desc = &yp->tx_status[0];
808 	return 0;
809 }
810 
811 static netdev_tx_t yellowfin_start_xmit(struct sk_buff *skb,
812 					struct net_device *dev)
813 {
814 	struct yellowfin_private *yp = netdev_priv(dev);
815 	unsigned entry;
816 	int len = skb->len;
817 
818 	netif_stop_queue (dev);
819 
820 	/* Note: Ordering is important here, set the field with the
821 	   "ownership" bit last, and only then increment cur_tx. */
822 
823 	/* Calculate the next Tx descriptor entry. */
824 	entry = yp->cur_tx % TX_RING_SIZE;
825 
826 	if (gx_fix) {	/* Note: only works for paddable protocols e.g.  IP. */
827 		int cacheline_end = ((unsigned long)skb->data + skb->len) % 32;
828 		/* Fix GX chipset errata. */
829 		if (cacheline_end > 24  || cacheline_end == 0) {
830 			len = skb->len + 32 - cacheline_end + 1;
831 			if (skb_padto(skb, len)) {
832 				yp->tx_skbuff[entry] = NULL;
833 				netif_wake_queue(dev);
834 				return NETDEV_TX_OK;
835 			}
836 		}
837 	}
838 	yp->tx_skbuff[entry] = skb;
839 
840 #ifdef NO_TXSTATS
841 	yp->tx_ring[entry].addr = cpu_to_le32(dma_map_single(&yp->pci_dev->dev,
842 							     skb->data,
843 							     len, DMA_TO_DEVICE));
844 	yp->tx_ring[entry].result_status = 0;
845 	if (entry >= TX_RING_SIZE-1) {
846 		/* New stop command. */
847 		yp->tx_ring[0].dbdma_cmd = cpu_to_le32(CMD_STOP);
848 		yp->tx_ring[TX_RING_SIZE-1].dbdma_cmd =
849 			cpu_to_le32(CMD_TX_PKT|BRANCH_ALWAYS | len);
850 	} else {
851 		yp->tx_ring[entry+1].dbdma_cmd = cpu_to_le32(CMD_STOP);
852 		yp->tx_ring[entry].dbdma_cmd =
853 			cpu_to_le32(CMD_TX_PKT | BRANCH_IFTRUE | len);
854 	}
855 	yp->cur_tx++;
856 #else
857 	yp->tx_ring[entry<<1].request_cnt = len;
858 	yp->tx_ring[entry<<1].addr = cpu_to_le32(dma_map_single(&yp->pci_dev->dev,
859 								skb->data,
860 								len, DMA_TO_DEVICE));
861 	/* The input_last (status-write) command is constant, but we must
862 	   rewrite the subsequent 'stop' command. */
863 
864 	yp->cur_tx++;
865 	{
866 		unsigned next_entry = yp->cur_tx % TX_RING_SIZE;
867 		yp->tx_ring[next_entry<<1].dbdma_cmd = cpu_to_le32(CMD_STOP);
868 	}
869 	/* Final step -- overwrite the old 'stop' command. */
870 
871 	yp->tx_ring[entry<<1].dbdma_cmd =
872 		cpu_to_le32( ((entry % 6) == 0 ? CMD_TX_PKT|INTR_ALWAYS|BRANCH_IFTRUE :
873 					  CMD_TX_PKT | BRANCH_IFTRUE) | len);
874 #endif
875 
876 	/* Non-x86 Todo: explicitly flush cache lines here. */
877 
878 	/* Wake the potentially-idle transmit channel. */
879 	iowrite32(0x10001000, yp->base + TxCtrl);
880 
881 	if (yp->cur_tx - yp->dirty_tx < TX_QUEUE_SIZE)
882 		netif_start_queue (dev);		/* Typical path */
883 	else
884 		yp->tx_full = 1;
885 
886 	if (yellowfin_debug > 4) {
887 		netdev_printk(KERN_DEBUG, dev, "Yellowfin transmit frame #%d queued in slot %d\n",
888 			      yp->cur_tx, entry);
889 	}
890 	return NETDEV_TX_OK;
891 }
892 
893 /* The interrupt handler does all of the Rx thread work and cleans up
894    after the Tx thread. */
895 static irqreturn_t yellowfin_interrupt(int irq, void *dev_instance)
896 {
897 	struct net_device *dev = dev_instance;
898 	struct yellowfin_private *yp;
899 	void __iomem *ioaddr;
900 	int boguscnt = max_interrupt_work;
901 	unsigned int handled = 0;
902 
903 	yp = netdev_priv(dev);
904 	ioaddr = yp->base;
905 
906 	spin_lock (&yp->lock);
907 
908 	do {
909 		u16 intr_status = ioread16(ioaddr + IntrClear);
910 
911 		if (yellowfin_debug > 4)
912 			netdev_printk(KERN_DEBUG, dev, "Yellowfin interrupt, status %04x\n",
913 				      intr_status);
914 
915 		if (intr_status == 0)
916 			break;
917 		handled = 1;
918 
919 		if (intr_status & (IntrRxDone | IntrEarlyRx)) {
920 			yellowfin_rx(dev);
921 			iowrite32(0x10001000, ioaddr + RxCtrl);		/* Wake Rx engine. */
922 		}
923 
924 #ifdef NO_TXSTATS
925 		for (; yp->cur_tx - yp->dirty_tx > 0; yp->dirty_tx++) {
926 			int entry = yp->dirty_tx % TX_RING_SIZE;
927 			struct sk_buff *skb;
928 
929 			if (yp->tx_ring[entry].result_status == 0)
930 				break;
931 			skb = yp->tx_skbuff[entry];
932 			dev->stats.tx_packets++;
933 			dev->stats.tx_bytes += skb->len;
934 			/* Free the original skb. */
935 			dma_unmap_single(&yp->pci_dev->dev,
936 					 le32_to_cpu(yp->tx_ring[entry].addr),
937 					 skb->len, DMA_TO_DEVICE);
938 			dev_consume_skb_irq(skb);
939 			yp->tx_skbuff[entry] = NULL;
940 		}
941 		if (yp->tx_full &&
942 		    yp->cur_tx - yp->dirty_tx < TX_QUEUE_SIZE - 4) {
943 			/* The ring is no longer full, clear tbusy. */
944 			yp->tx_full = 0;
945 			netif_wake_queue(dev);
946 		}
947 #else
948 		if ((intr_status & IntrTxDone) || (yp->tx_tail_desc->tx_errs)) {
949 			unsigned dirty_tx = yp->dirty_tx;
950 
951 			for (dirty_tx = yp->dirty_tx; yp->cur_tx - dirty_tx > 0;
952 				 dirty_tx++) {
953 				/* Todo: optimize this. */
954 				int entry = dirty_tx % TX_RING_SIZE;
955 				u16 tx_errs = yp->tx_status[entry].tx_errs;
956 				struct sk_buff *skb;
957 
958 #ifndef final_version
959 				if (yellowfin_debug > 5)
960 					netdev_printk(KERN_DEBUG, dev, "Tx queue %d check, Tx status %04x %04x %04x %04x\n",
961 						      entry,
962 						      yp->tx_status[entry].tx_cnt,
963 						      yp->tx_status[entry].tx_errs,
964 						      yp->tx_status[entry].total_tx_cnt,
965 						      yp->tx_status[entry].paused);
966 #endif
967 				if (tx_errs == 0)
968 					break;	/* It still hasn't been Txed */
969 				skb = yp->tx_skbuff[entry];
970 				if (tx_errs & 0xF810) {
971 					/* There was an major error, log it. */
972 #ifndef final_version
973 					if (yellowfin_debug > 1)
974 						netdev_printk(KERN_DEBUG, dev, "Transmit error, Tx status %04x\n",
975 							      tx_errs);
976 #endif
977 					dev->stats.tx_errors++;
978 					if (tx_errs & 0xF800) dev->stats.tx_aborted_errors++;
979 					if (tx_errs & 0x0800) dev->stats.tx_carrier_errors++;
980 					if (tx_errs & 0x2000) dev->stats.tx_window_errors++;
981 					if (tx_errs & 0x8000) dev->stats.tx_fifo_errors++;
982 				} else {
983 #ifndef final_version
984 					if (yellowfin_debug > 4)
985 						netdev_printk(KERN_DEBUG, dev, "Normal transmit, Tx status %04x\n",
986 							      tx_errs);
987 #endif
988 					dev->stats.tx_bytes += skb->len;
989 					dev->stats.collisions += tx_errs & 15;
990 					dev->stats.tx_packets++;
991 				}
992 				/* Free the original skb. */
993 				dma_unmap_single(&yp->pci_dev->dev,
994 						 yp->tx_ring[entry << 1].addr,
995 						 skb->len, DMA_TO_DEVICE);
996 				dev_consume_skb_irq(skb);
997 				yp->tx_skbuff[entry] = 0;
998 				/* Mark status as empty. */
999 				yp->tx_status[entry].tx_errs = 0;
1000 			}
1001 
1002 #ifndef final_version
1003 			if (yp->cur_tx - dirty_tx > TX_RING_SIZE) {
1004 				netdev_err(dev, "Out-of-sync dirty pointer, %d vs. %d, full=%d\n",
1005 					   dirty_tx, yp->cur_tx, yp->tx_full);
1006 				dirty_tx += TX_RING_SIZE;
1007 			}
1008 #endif
1009 
1010 			if (yp->tx_full &&
1011 			    yp->cur_tx - dirty_tx < TX_QUEUE_SIZE - 2) {
1012 				/* The ring is no longer full, clear tbusy. */
1013 				yp->tx_full = 0;
1014 				netif_wake_queue(dev);
1015 			}
1016 
1017 			yp->dirty_tx = dirty_tx;
1018 			yp->tx_tail_desc = &yp->tx_status[dirty_tx % TX_RING_SIZE];
1019 		}
1020 #endif
1021 
1022 		/* Log errors and other uncommon events. */
1023 		if (intr_status & 0x2ee)	/* Abnormal error summary. */
1024 			yellowfin_error(dev, intr_status);
1025 
1026 		if (--boguscnt < 0) {
1027 			netdev_warn(dev, "Too much work at interrupt, status=%#04x\n",
1028 				    intr_status);
1029 			break;
1030 		}
1031 	} while (1);
1032 
1033 	if (yellowfin_debug > 3)
1034 		netdev_printk(KERN_DEBUG, dev, "exiting interrupt, status=%#04x\n",
1035 			      ioread16(ioaddr + IntrStatus));
1036 
1037 	spin_unlock (&yp->lock);
1038 	return IRQ_RETVAL(handled);
1039 }
1040 
1041 /* This routine is logically part of the interrupt handler, but separated
1042    for clarity and better register allocation. */
1043 static int yellowfin_rx(struct net_device *dev)
1044 {
1045 	struct yellowfin_private *yp = netdev_priv(dev);
1046 	int entry = yp->cur_rx % RX_RING_SIZE;
1047 	int boguscnt = yp->dirty_rx + RX_RING_SIZE - yp->cur_rx;
1048 
1049 	if (yellowfin_debug > 4) {
1050 		printk(KERN_DEBUG " In yellowfin_rx(), entry %d status %08x\n",
1051 			   entry, yp->rx_ring[entry].result_status);
1052 		printk(KERN_DEBUG "   #%d desc. %08x %08x %08x\n",
1053 			   entry, yp->rx_ring[entry].dbdma_cmd, yp->rx_ring[entry].addr,
1054 			   yp->rx_ring[entry].result_status);
1055 	}
1056 
1057 	/* If EOP is set on the next entry, it's a new packet. Send it up. */
1058 	while (1) {
1059 		struct yellowfin_desc *desc = &yp->rx_ring[entry];
1060 		struct sk_buff *rx_skb = yp->rx_skbuff[entry];
1061 		s16 frame_status;
1062 		u16 desc_status;
1063 		int data_size, __maybe_unused yf_size;
1064 		u8 *buf_addr;
1065 
1066 		if(!desc->result_status)
1067 			break;
1068 		dma_sync_single_for_cpu(&yp->pci_dev->dev,
1069 					le32_to_cpu(desc->addr),
1070 					yp->rx_buf_sz, DMA_FROM_DEVICE);
1071 		desc_status = le32_to_cpu(desc->result_status) >> 16;
1072 		buf_addr = rx_skb->data;
1073 		data_size = (le32_to_cpu(desc->dbdma_cmd) -
1074 			le32_to_cpu(desc->result_status)) & 0xffff;
1075 		frame_status = get_unaligned_le16(&(buf_addr[data_size - 2]));
1076 		if (yellowfin_debug > 4)
1077 			printk(KERN_DEBUG "  %s() status was %04x\n",
1078 			       __func__, frame_status);
1079 		if (--boguscnt < 0)
1080 			break;
1081 
1082 		yf_size = sizeof(struct yellowfin_desc);
1083 
1084 		if ( ! (desc_status & RX_EOP)) {
1085 			if (data_size != 0)
1086 				netdev_warn(dev, "Oversized Ethernet frame spanned multiple buffers, status %04x, data_size %d!\n",
1087 					    desc_status, data_size);
1088 			dev->stats.rx_length_errors++;
1089 		} else if ((yp->drv_flags & IsGigabit)  &&  (frame_status & 0x0038)) {
1090 			/* There was a error. */
1091 			if (yellowfin_debug > 3)
1092 				printk(KERN_DEBUG "  %s() Rx error was %04x\n",
1093 				       __func__, frame_status);
1094 			dev->stats.rx_errors++;
1095 			if (frame_status & 0x0060) dev->stats.rx_length_errors++;
1096 			if (frame_status & 0x0008) dev->stats.rx_frame_errors++;
1097 			if (frame_status & 0x0010) dev->stats.rx_crc_errors++;
1098 			if (frame_status < 0) dev->stats.rx_dropped++;
1099 		} else if ( !(yp->drv_flags & IsGigabit)  &&
1100 				   ((buf_addr[data_size-1] & 0x85) || buf_addr[data_size-2] & 0xC0)) {
1101 			u8 status1 = buf_addr[data_size-2];
1102 			u8 status2 = buf_addr[data_size-1];
1103 			dev->stats.rx_errors++;
1104 			if (status1 & 0xC0) dev->stats.rx_length_errors++;
1105 			if (status2 & 0x03) dev->stats.rx_frame_errors++;
1106 			if (status2 & 0x04) dev->stats.rx_crc_errors++;
1107 			if (status2 & 0x80) dev->stats.rx_dropped++;
1108 #ifdef YF_PROTOTYPE		/* Support for prototype hardware errata. */
1109 		} else if ((yp->flags & HasMACAddrBug)  &&
1110 			!ether_addr_equal(le32_to_cpu(yp->rx_ring_dma +
1111 						      entry * yf_size),
1112 					  dev->dev_addr) &&
1113 			!ether_addr_equal(le32_to_cpu(yp->rx_ring_dma +
1114 						      entry * yf_size),
1115 					  "\377\377\377\377\377\377")) {
1116 			if (bogus_rx++ == 0)
1117 				netdev_warn(dev, "Bad frame to %pM\n",
1118 					    buf_addr);
1119 #endif
1120 		} else {
1121 			struct sk_buff *skb;
1122 			int pkt_len = data_size -
1123 				(yp->chip_id ? 7 : 8 + buf_addr[data_size - 8]);
1124 			/* To verify: Yellowfin Length should omit the CRC! */
1125 
1126 #ifndef final_version
1127 			if (yellowfin_debug > 4)
1128 				printk(KERN_DEBUG "  %s() normal Rx pkt length %d of %d, bogus_cnt %d\n",
1129 				       __func__, pkt_len, data_size, boguscnt);
1130 #endif
1131 			/* Check if the packet is long enough to just pass up the skbuff
1132 			   without copying to a properly sized skbuff. */
1133 			if (pkt_len > rx_copybreak) {
1134 				skb_put(skb = rx_skb, pkt_len);
1135 				dma_unmap_single(&yp->pci_dev->dev,
1136 						 le32_to_cpu(yp->rx_ring[entry].addr),
1137 						 yp->rx_buf_sz,
1138 						 DMA_FROM_DEVICE);
1139 				yp->rx_skbuff[entry] = NULL;
1140 			} else {
1141 				skb = netdev_alloc_skb(dev, pkt_len + 2);
1142 				if (skb == NULL)
1143 					break;
1144 				skb_reserve(skb, 2);	/* 16 byte align the IP header */
1145 				skb_copy_to_linear_data(skb, rx_skb->data, pkt_len);
1146 				skb_put(skb, pkt_len);
1147 				dma_sync_single_for_device(&yp->pci_dev->dev,
1148 							   le32_to_cpu(desc->addr),
1149 							   yp->rx_buf_sz,
1150 							   DMA_FROM_DEVICE);
1151 			}
1152 			skb->protocol = eth_type_trans(skb, dev);
1153 			netif_rx(skb);
1154 			dev->stats.rx_packets++;
1155 			dev->stats.rx_bytes += pkt_len;
1156 		}
1157 		entry = (++yp->cur_rx) % RX_RING_SIZE;
1158 	}
1159 
1160 	/* Refill the Rx ring buffers. */
1161 	for (; yp->cur_rx - yp->dirty_rx > 0; yp->dirty_rx++) {
1162 		entry = yp->dirty_rx % RX_RING_SIZE;
1163 		if (yp->rx_skbuff[entry] == NULL) {
1164 			struct sk_buff *skb = netdev_alloc_skb(dev, yp->rx_buf_sz + 2);
1165 			if (skb == NULL)
1166 				break;				/* Better luck next round. */
1167 			yp->rx_skbuff[entry] = skb;
1168 			skb_reserve(skb, 2);	/* Align IP on 16 byte boundaries */
1169 			yp->rx_ring[entry].addr = cpu_to_le32(dma_map_single(&yp->pci_dev->dev,
1170 									     skb->data,
1171 									     yp->rx_buf_sz,
1172 									     DMA_FROM_DEVICE));
1173 		}
1174 		yp->rx_ring[entry].dbdma_cmd = cpu_to_le32(CMD_STOP);
1175 		yp->rx_ring[entry].result_status = 0;	/* Clear complete bit. */
1176 		if (entry != 0)
1177 			yp->rx_ring[entry - 1].dbdma_cmd =
1178 				cpu_to_le32(CMD_RX_BUF | INTR_ALWAYS | yp->rx_buf_sz);
1179 		else
1180 			yp->rx_ring[RX_RING_SIZE - 1].dbdma_cmd =
1181 				cpu_to_le32(CMD_RX_BUF | INTR_ALWAYS | BRANCH_ALWAYS
1182 							| yp->rx_buf_sz);
1183 	}
1184 
1185 	return 0;
1186 }
1187 
1188 static void yellowfin_error(struct net_device *dev, int intr_status)
1189 {
1190 	netdev_err(dev, "Something Wicked happened! %04x\n", intr_status);
1191 	/* Hmmmmm, it's not clear what to do here. */
1192 	if (intr_status & (IntrTxPCIErr | IntrTxPCIFault))
1193 		dev->stats.tx_errors++;
1194 	if (intr_status & (IntrRxPCIErr | IntrRxPCIFault))
1195 		dev->stats.rx_errors++;
1196 }
1197 
1198 static int yellowfin_close(struct net_device *dev)
1199 {
1200 	struct yellowfin_private *yp = netdev_priv(dev);
1201 	void __iomem *ioaddr = yp->base;
1202 	int i;
1203 
1204 	netif_stop_queue (dev);
1205 
1206 	if (yellowfin_debug > 1) {
1207 		netdev_printk(KERN_DEBUG, dev, "Shutting down ethercard, status was Tx %04x Rx %04x Int %02x\n",
1208 			      ioread16(ioaddr + TxStatus),
1209 			      ioread16(ioaddr + RxStatus),
1210 			      ioread16(ioaddr + IntrStatus));
1211 		netdev_printk(KERN_DEBUG, dev, "Queue pointers were Tx %d / %d,  Rx %d / %d\n",
1212 			      yp->cur_tx, yp->dirty_tx,
1213 			      yp->cur_rx, yp->dirty_rx);
1214 	}
1215 
1216 	/* Disable interrupts by clearing the interrupt mask. */
1217 	iowrite16(0x0000, ioaddr + IntrEnb);
1218 
1219 	/* Stop the chip's Tx and Rx processes. */
1220 	iowrite32(0x80000000, ioaddr + RxCtrl);
1221 	iowrite32(0x80000000, ioaddr + TxCtrl);
1222 
1223 	del_timer(&yp->timer);
1224 
1225 #if defined(__i386__)
1226 	if (yellowfin_debug > 2) {
1227 		printk(KERN_DEBUG "  Tx ring at %08llx:\n",
1228 				(unsigned long long)yp->tx_ring_dma);
1229 		for (i = 0; i < TX_RING_SIZE*2; i++)
1230 			printk(KERN_DEBUG " %c #%d desc. %08x %08x %08x %08x\n",
1231 				   ioread32(ioaddr + TxPtr) == (long)&yp->tx_ring[i] ? '>' : ' ',
1232 				   i, yp->tx_ring[i].dbdma_cmd, yp->tx_ring[i].addr,
1233 				   yp->tx_ring[i].branch_addr, yp->tx_ring[i].result_status);
1234 		printk(KERN_DEBUG "  Tx status %p:\n", yp->tx_status);
1235 		for (i = 0; i < TX_RING_SIZE; i++)
1236 			printk(KERN_DEBUG "   #%d status %04x %04x %04x %04x\n",
1237 				   i, yp->tx_status[i].tx_cnt, yp->tx_status[i].tx_errs,
1238 				   yp->tx_status[i].total_tx_cnt, yp->tx_status[i].paused);
1239 
1240 		printk(KERN_DEBUG "  Rx ring %08llx:\n",
1241 				(unsigned long long)yp->rx_ring_dma);
1242 		for (i = 0; i < RX_RING_SIZE; i++) {
1243 			printk(KERN_DEBUG " %c #%d desc. %08x %08x %08x\n",
1244 				   ioread32(ioaddr + RxPtr) == (long)&yp->rx_ring[i] ? '>' : ' ',
1245 				   i, yp->rx_ring[i].dbdma_cmd, yp->rx_ring[i].addr,
1246 				   yp->rx_ring[i].result_status);
1247 			if (yellowfin_debug > 6) {
1248 				if (get_unaligned((u8*)yp->rx_ring[i].addr) != 0x69) {
1249 					int j;
1250 
1251 					printk(KERN_DEBUG);
1252 					for (j = 0; j < 0x50; j++)
1253 						pr_cont(" %04x",
1254 							get_unaligned(((u16*)yp->rx_ring[i].addr) + j));
1255 					pr_cont("\n");
1256 				}
1257 			}
1258 		}
1259 	}
1260 #endif /* __i386__ debugging only */
1261 
1262 	free_irq(yp->pci_dev->irq, dev);
1263 
1264 	/* Free all the skbuffs in the Rx queue. */
1265 	for (i = 0; i < RX_RING_SIZE; i++) {
1266 		yp->rx_ring[i].dbdma_cmd = cpu_to_le32(CMD_STOP);
1267 		yp->rx_ring[i].addr = cpu_to_le32(0xBADF00D0); /* An invalid address. */
1268 		if (yp->rx_skbuff[i]) {
1269 			dev_kfree_skb(yp->rx_skbuff[i]);
1270 		}
1271 		yp->rx_skbuff[i] = NULL;
1272 	}
1273 	for (i = 0; i < TX_RING_SIZE; i++) {
1274 		dev_kfree_skb(yp->tx_skbuff[i]);
1275 		yp->tx_skbuff[i] = NULL;
1276 	}
1277 
1278 #ifdef YF_PROTOTYPE			/* Support for prototype hardware errata. */
1279 	if (yellowfin_debug > 0) {
1280 		netdev_printk(KERN_DEBUG, dev, "Received %d frames that we should not have\n",
1281 			      bogus_rx);
1282 	}
1283 #endif
1284 
1285 	return 0;
1286 }
1287 
1288 /* Set or clear the multicast filter for this adaptor. */
1289 
1290 static void set_rx_mode(struct net_device *dev)
1291 {
1292 	struct yellowfin_private *yp = netdev_priv(dev);
1293 	void __iomem *ioaddr = yp->base;
1294 	u16 cfg_value = ioread16(ioaddr + Cnfg);
1295 
1296 	/* Stop the Rx process to change any value. */
1297 	iowrite16(cfg_value & ~0x1000, ioaddr + Cnfg);
1298 	if (dev->flags & IFF_PROMISC) {			/* Set promiscuous. */
1299 		iowrite16(0x000F, ioaddr + AddrMode);
1300 	} else if ((netdev_mc_count(dev) > 64) ||
1301 		   (dev->flags & IFF_ALLMULTI)) {
1302 		/* Too many to filter well, or accept all multicasts. */
1303 		iowrite16(0x000B, ioaddr + AddrMode);
1304 	} else if (!netdev_mc_empty(dev)) { /* Must use the multicast hash table. */
1305 		struct netdev_hw_addr *ha;
1306 		u16 hash_table[4];
1307 		int i;
1308 
1309 		memset(hash_table, 0, sizeof(hash_table));
1310 		netdev_for_each_mc_addr(ha, dev) {
1311 			unsigned int bit;
1312 
1313 			/* Due to a bug in the early chip versions, multiple filter
1314 			   slots must be set for each address. */
1315 			if (yp->drv_flags & HasMulticastBug) {
1316 				bit = (ether_crc_le(3, ha->addr) >> 3) & 0x3f;
1317 				hash_table[bit >> 4] |= (1 << bit);
1318 				bit = (ether_crc_le(4, ha->addr) >> 3) & 0x3f;
1319 				hash_table[bit >> 4] |= (1 << bit);
1320 				bit = (ether_crc_le(5, ha->addr) >> 3) & 0x3f;
1321 				hash_table[bit >> 4] |= (1 << bit);
1322 			}
1323 			bit = (ether_crc_le(6, ha->addr) >> 3) & 0x3f;
1324 			hash_table[bit >> 4] |= (1 << bit);
1325 		}
1326 		/* Copy the hash table to the chip. */
1327 		for (i = 0; i < 4; i++)
1328 			iowrite16(hash_table[i], ioaddr + HashTbl + i*2);
1329 		iowrite16(0x0003, ioaddr + AddrMode);
1330 	} else {					/* Normal, unicast/broadcast-only mode. */
1331 		iowrite16(0x0001, ioaddr + AddrMode);
1332 	}
1333 	/* Restart the Rx process. */
1334 	iowrite16(cfg_value | 0x1000, ioaddr + Cnfg);
1335 }
1336 
1337 static void yellowfin_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1338 {
1339 	struct yellowfin_private *np = netdev_priv(dev);
1340 
1341 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
1342 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1343 	strlcpy(info->bus_info, pci_name(np->pci_dev), sizeof(info->bus_info));
1344 }
1345 
1346 static const struct ethtool_ops ethtool_ops = {
1347 	.get_drvinfo = yellowfin_get_drvinfo
1348 };
1349 
1350 static int netdev_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
1351 {
1352 	struct yellowfin_private *np = netdev_priv(dev);
1353 	void __iomem *ioaddr = np->base;
1354 	struct mii_ioctl_data *data = if_mii(rq);
1355 
1356 	switch(cmd) {
1357 	case SIOCGMIIPHY:		/* Get address of MII PHY in use. */
1358 		data->phy_id = np->phys[0] & 0x1f;
1359 		fallthrough;
1360 
1361 	case SIOCGMIIREG:		/* Read MII PHY register. */
1362 		data->val_out = mdio_read(ioaddr, data->phy_id & 0x1f, data->reg_num & 0x1f);
1363 		return 0;
1364 
1365 	case SIOCSMIIREG:		/* Write MII PHY register. */
1366 		if (data->phy_id == np->phys[0]) {
1367 			u16 value = data->val_in;
1368 			switch (data->reg_num) {
1369 			case 0:
1370 				/* Check for autonegotiation on or reset. */
1371 				np->medialock = (value & 0x9000) ? 0 : 1;
1372 				if (np->medialock)
1373 					np->full_duplex = (value & 0x0100) ? 1 : 0;
1374 				break;
1375 			case 4: np->advertising = value; break;
1376 			}
1377 			/* Perhaps check_duplex(dev), depending on chip semantics. */
1378 		}
1379 		mdio_write(ioaddr, data->phy_id & 0x1f, data->reg_num & 0x1f, data->val_in);
1380 		return 0;
1381 	default:
1382 		return -EOPNOTSUPP;
1383 	}
1384 }
1385 
1386 
1387 static void yellowfin_remove_one(struct pci_dev *pdev)
1388 {
1389 	struct net_device *dev = pci_get_drvdata(pdev);
1390 	struct yellowfin_private *np;
1391 
1392 	BUG_ON(!dev);
1393 	np = netdev_priv(dev);
1394 
1395 	dma_free_coherent(&pdev->dev, STATUS_TOTAL_SIZE, np->tx_status,
1396 			  np->tx_status_dma);
1397 	dma_free_coherent(&pdev->dev, RX_TOTAL_SIZE, np->rx_ring,
1398 			  np->rx_ring_dma);
1399 	dma_free_coherent(&pdev->dev, TX_TOTAL_SIZE, np->tx_ring,
1400 			  np->tx_ring_dma);
1401 	unregister_netdev (dev);
1402 
1403 	pci_iounmap(pdev, np->base);
1404 
1405 	pci_release_regions (pdev);
1406 
1407 	free_netdev (dev);
1408 }
1409 
1410 
1411 static struct pci_driver yellowfin_driver = {
1412 	.name		= DRV_NAME,
1413 	.id_table	= yellowfin_pci_tbl,
1414 	.probe		= yellowfin_init_one,
1415 	.remove		= yellowfin_remove_one,
1416 };
1417 
1418 
1419 static int __init yellowfin_init (void)
1420 {
1421 /* when a module, this is printed whether or not devices are found in probe */
1422 #ifdef MODULE
1423 	printk(version);
1424 #endif
1425 	return pci_register_driver(&yellowfin_driver);
1426 }
1427 
1428 
1429 static void __exit yellowfin_cleanup (void)
1430 {
1431 	pci_unregister_driver (&yellowfin_driver);
1432 }
1433 
1434 
1435 module_init(yellowfin_init);
1436 module_exit(yellowfin_cleanup);
1437