xref: /linux/drivers/net/ethernet/sun/cassini.c (revision 69bfec7548f4c1595bac0e3ddfc0458a5af31f4c)
1 // SPDX-License-Identifier: GPL-2.0+
2 /* cassini.c: Sun Microsystems Cassini(+) ethernet driver.
3  *
4  * Copyright (C) 2004 Sun Microsystems Inc.
5  * Copyright (C) 2003 Adrian Sun (asun@darksunrising.com)
6  *
7  * This driver uses the sungem driver (c) David Miller
8  * (davem@redhat.com) as its basis.
9  *
10  * The cassini chip has a number of features that distinguish it from
11  * the gem chip:
12  *  4 transmit descriptor rings that are used for either QoS (VLAN) or
13  *      load balancing (non-VLAN mode)
14  *  batching of multiple packets
15  *  multiple CPU dispatching
16  *  page-based RX descriptor engine with separate completion rings
17  *  Gigabit support (GMII and PCS interface)
18  *  MIF link up/down detection works
19  *
20  * RX is handled by page sized buffers that are attached as fragments to
21  * the skb. here's what's done:
22  *  -- driver allocates pages at a time and keeps reference counts
23  *     on them.
24  *  -- the upper protocol layers assume that the header is in the skb
25  *     itself. as a result, cassini will copy a small amount (64 bytes)
26  *     to make them happy.
27  *  -- driver appends the rest of the data pages as frags to skbuffs
28  *     and increments the reference count
29  *  -- on page reclamation, the driver swaps the page with a spare page.
30  *     if that page is still in use, it frees its reference to that page,
31  *     and allocates a new page for use. otherwise, it just recycles the
32  *     page.
33  *
34  * NOTE: cassini can parse the header. however, it's not worth it
35  *       as long as the network stack requires a header copy.
36  *
37  * TX has 4 queues. currently these queues are used in a round-robin
38  * fashion for load balancing. They can also be used for QoS. for that
39  * to work, however, QoS information needs to be exposed down to the driver
40  * level so that subqueues get targeted to particular transmit rings.
41  * alternatively, the queues can be configured via use of the all-purpose
42  * ioctl.
43  *
44  * RX DATA: the rx completion ring has all the info, but the rx desc
45  * ring has all of the data. RX can conceivably come in under multiple
46  * interrupts, but the INT# assignment needs to be set up properly by
47  * the BIOS and conveyed to the driver. PCI BIOSes don't know how to do
48  * that. also, the two descriptor rings are designed to distinguish between
49  * encrypted and non-encrypted packets, but we use them for buffering
50  * instead.
51  *
52  * by default, the selective clear mask is set up to process rx packets.
53  */
54 
55 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
56 
57 #include <linux/module.h>
58 #include <linux/kernel.h>
59 #include <linux/types.h>
60 #include <linux/compiler.h>
61 #include <linux/slab.h>
62 #include <linux/delay.h>
63 #include <linux/init.h>
64 #include <linux/interrupt.h>
65 #include <linux/vmalloc.h>
66 #include <linux/ioport.h>
67 #include <linux/pci.h>
68 #include <linux/mm.h>
69 #include <linux/highmem.h>
70 #include <linux/list.h>
71 #include <linux/dma-mapping.h>
72 
73 #include <linux/netdevice.h>
74 #include <linux/etherdevice.h>
75 #include <linux/skbuff.h>
76 #include <linux/ethtool.h>
77 #include <linux/crc32.h>
78 #include <linux/random.h>
79 #include <linux/mii.h>
80 #include <linux/ip.h>
81 #include <linux/tcp.h>
82 #include <linux/mutex.h>
83 #include <linux/firmware.h>
84 
85 #include <net/checksum.h>
86 
87 #include <linux/atomic.h>
88 #include <asm/io.h>
89 #include <asm/byteorder.h>
90 #include <linux/uaccess.h>
91 #include <linux/jiffies.h>
92 
93 #define CAS_NCPUS            num_online_cpus()
94 
95 #define cas_skb_release(x)  netif_rx(x)
96 
97 /* select which firmware to use */
98 #define USE_HP_WORKAROUND
99 #define HP_WORKAROUND_DEFAULT /* select which firmware to use as default */
100 #define CAS_HP_ALT_FIRMWARE   cas_prog_null /* alternate firmware */
101 
102 #include "cassini.h"
103 
104 #define USE_TX_COMPWB      /* use completion writeback registers */
105 #define USE_CSMA_CD_PROTO  /* standard CSMA/CD */
106 #define USE_RX_BLANK       /* hw interrupt mitigation */
107 #undef USE_ENTROPY_DEV     /* don't test for entropy device */
108 
109 /* NOTE: these aren't useable unless PCI interrupts can be assigned.
110  * also, we need to make cp->lock finer-grained.
111  */
112 #undef  USE_PCI_INTB
113 #undef  USE_PCI_INTC
114 #undef  USE_PCI_INTD
115 #undef  USE_QOS
116 
117 #undef  USE_VPD_DEBUG       /* debug vpd information if defined */
118 
119 /* rx processing options */
120 #define USE_PAGE_ORDER      /* specify to allocate large rx pages */
121 #define RX_DONT_BATCH  0    /* if 1, don't batch flows */
122 #define RX_COPY_ALWAYS 0    /* if 0, use frags */
123 #define RX_COPY_MIN    64   /* copy a little to make upper layers happy */
124 #undef  RX_COUNT_BUFFERS    /* define to calculate RX buffer stats */
125 
126 #define DRV_MODULE_NAME		"cassini"
127 #define DRV_MODULE_VERSION	"1.6"
128 #define DRV_MODULE_RELDATE	"21 May 2008"
129 
130 #define CAS_DEF_MSG_ENABLE	  \
131 	(NETIF_MSG_DRV		| \
132 	 NETIF_MSG_PROBE	| \
133 	 NETIF_MSG_LINK		| \
134 	 NETIF_MSG_TIMER	| \
135 	 NETIF_MSG_IFDOWN	| \
136 	 NETIF_MSG_IFUP		| \
137 	 NETIF_MSG_RX_ERR	| \
138 	 NETIF_MSG_TX_ERR)
139 
140 /* length of time before we decide the hardware is borked,
141  * and dev->tx_timeout() should be called to fix the problem
142  */
143 #define CAS_TX_TIMEOUT			(HZ)
144 #define CAS_LINK_TIMEOUT                (22*HZ/10)
145 #define CAS_LINK_FAST_TIMEOUT           (1)
146 
147 /* timeout values for state changing. these specify the number
148  * of 10us delays to be used before giving up.
149  */
150 #define STOP_TRIES_PHY 1000
151 #define STOP_TRIES     5000
152 
153 /* specify a minimum frame size to deal with some fifo issues
154  * max mtu == 2 * page size - ethernet header - 64 - swivel =
155  *            2 * page_size - 0x50
156  */
157 #define CAS_MIN_FRAME			97
158 #define CAS_1000MB_MIN_FRAME            255
159 #define CAS_MIN_MTU                     60
160 #define CAS_MAX_MTU                     min(((cp->page_size << 1) - 0x50), 9000)
161 
162 #if 1
163 /*
164  * Eliminate these and use separate atomic counters for each, to
165  * avoid a race condition.
166  */
167 #else
168 #define CAS_RESET_MTU                   1
169 #define CAS_RESET_ALL                   2
170 #define CAS_RESET_SPARE                 3
171 #endif
172 
173 static char version[] =
174 	DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
175 
176 static int cassini_debug = -1;	/* -1 == use CAS_DEF_MSG_ENABLE as value */
177 static int link_mode;
178 
179 MODULE_AUTHOR("Adrian Sun (asun@darksunrising.com)");
180 MODULE_DESCRIPTION("Sun Cassini(+) ethernet driver");
181 MODULE_LICENSE("GPL");
182 MODULE_FIRMWARE("sun/cassini.bin");
183 module_param(cassini_debug, int, 0);
184 MODULE_PARM_DESC(cassini_debug, "Cassini bitmapped debugging message enable value");
185 module_param(link_mode, int, 0);
186 MODULE_PARM_DESC(link_mode, "default link mode");
187 
188 /*
189  * Work around for a PCS bug in which the link goes down due to the chip
190  * being confused and never showing a link status of "up."
191  */
192 #define DEFAULT_LINKDOWN_TIMEOUT 5
193 /*
194  * Value in seconds, for user input.
195  */
196 static int linkdown_timeout = DEFAULT_LINKDOWN_TIMEOUT;
197 module_param(linkdown_timeout, int, 0);
198 MODULE_PARM_DESC(linkdown_timeout,
199 "min reset interval in sec. for PCS linkdown issue; disabled if not positive");
200 
201 /*
202  * value in 'ticks' (units used by jiffies). Set when we init the
203  * module because 'HZ' in actually a function call on some flavors of
204  * Linux.  This will default to DEFAULT_LINKDOWN_TIMEOUT * HZ.
205  */
206 static int link_transition_timeout;
207 
208 
209 
210 static u16 link_modes[] = {
211 	BMCR_ANENABLE,			 /* 0 : autoneg */
212 	0,				 /* 1 : 10bt half duplex */
213 	BMCR_SPEED100,			 /* 2 : 100bt half duplex */
214 	BMCR_FULLDPLX,			 /* 3 : 10bt full duplex */
215 	BMCR_SPEED100|BMCR_FULLDPLX,	 /* 4 : 100bt full duplex */
216 	CAS_BMCR_SPEED1000|BMCR_FULLDPLX /* 5 : 1000bt full duplex */
217 };
218 
219 static const struct pci_device_id cas_pci_tbl[] = {
220 	{ PCI_VENDOR_ID_SUN, PCI_DEVICE_ID_SUN_CASSINI,
221 	  PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0UL },
222 	{ PCI_VENDOR_ID_NS, PCI_DEVICE_ID_NS_SATURN,
223 	  PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0UL },
224 	{ 0, }
225 };
226 
227 MODULE_DEVICE_TABLE(pci, cas_pci_tbl);
228 
229 static void cas_set_link_modes(struct cas *cp);
230 
231 static inline void cas_lock_tx(struct cas *cp)
232 {
233 	int i;
234 
235 	for (i = 0; i < N_TX_RINGS; i++)
236 		spin_lock_nested(&cp->tx_lock[i], i);
237 }
238 
239 /* WTZ: QA was finding deadlock problems with the previous
240  * versions after long test runs with multiple cards per machine.
241  * See if replacing cas_lock_all with safer versions helps. The
242  * symptoms QA is reporting match those we'd expect if interrupts
243  * aren't being properly restored, and we fixed a previous deadlock
244  * with similar symptoms by using save/restore versions in other
245  * places.
246  */
247 #define cas_lock_all_save(cp, flags) \
248 do { \
249 	struct cas *xxxcp = (cp); \
250 	spin_lock_irqsave(&xxxcp->lock, flags); \
251 	cas_lock_tx(xxxcp); \
252 } while (0)
253 
254 static inline void cas_unlock_tx(struct cas *cp)
255 {
256 	int i;
257 
258 	for (i = N_TX_RINGS; i > 0; i--)
259 		spin_unlock(&cp->tx_lock[i - 1]);
260 }
261 
262 #define cas_unlock_all_restore(cp, flags) \
263 do { \
264 	struct cas *xxxcp = (cp); \
265 	cas_unlock_tx(xxxcp); \
266 	spin_unlock_irqrestore(&xxxcp->lock, flags); \
267 } while (0)
268 
269 static void cas_disable_irq(struct cas *cp, const int ring)
270 {
271 	/* Make sure we won't get any more interrupts */
272 	if (ring == 0) {
273 		writel(0xFFFFFFFF, cp->regs + REG_INTR_MASK);
274 		return;
275 	}
276 
277 	/* disable completion interrupts and selectively mask */
278 	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
279 		switch (ring) {
280 #if defined (USE_PCI_INTB) || defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
281 #ifdef USE_PCI_INTB
282 		case 1:
283 #endif
284 #ifdef USE_PCI_INTC
285 		case 2:
286 #endif
287 #ifdef USE_PCI_INTD
288 		case 3:
289 #endif
290 			writel(INTRN_MASK_CLEAR_ALL | INTRN_MASK_RX_EN,
291 			       cp->regs + REG_PLUS_INTRN_MASK(ring));
292 			break;
293 #endif
294 		default:
295 			writel(INTRN_MASK_CLEAR_ALL, cp->regs +
296 			       REG_PLUS_INTRN_MASK(ring));
297 			break;
298 		}
299 	}
300 }
301 
302 static inline void cas_mask_intr(struct cas *cp)
303 {
304 	int i;
305 
306 	for (i = 0; i < N_RX_COMP_RINGS; i++)
307 		cas_disable_irq(cp, i);
308 }
309 
310 static void cas_enable_irq(struct cas *cp, const int ring)
311 {
312 	if (ring == 0) { /* all but TX_DONE */
313 		writel(INTR_TX_DONE, cp->regs + REG_INTR_MASK);
314 		return;
315 	}
316 
317 	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
318 		switch (ring) {
319 #if defined (USE_PCI_INTB) || defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
320 #ifdef USE_PCI_INTB
321 		case 1:
322 #endif
323 #ifdef USE_PCI_INTC
324 		case 2:
325 #endif
326 #ifdef USE_PCI_INTD
327 		case 3:
328 #endif
329 			writel(INTRN_MASK_RX_EN, cp->regs +
330 			       REG_PLUS_INTRN_MASK(ring));
331 			break;
332 #endif
333 		default:
334 			break;
335 		}
336 	}
337 }
338 
339 static inline void cas_unmask_intr(struct cas *cp)
340 {
341 	int i;
342 
343 	for (i = 0; i < N_RX_COMP_RINGS; i++)
344 		cas_enable_irq(cp, i);
345 }
346 
347 static inline void cas_entropy_gather(struct cas *cp)
348 {
349 #ifdef USE_ENTROPY_DEV
350 	if ((cp->cas_flags & CAS_FLAG_ENTROPY_DEV) == 0)
351 		return;
352 
353 	batch_entropy_store(readl(cp->regs + REG_ENTROPY_IV),
354 			    readl(cp->regs + REG_ENTROPY_IV),
355 			    sizeof(uint64_t)*8);
356 #endif
357 }
358 
359 static inline void cas_entropy_reset(struct cas *cp)
360 {
361 #ifdef USE_ENTROPY_DEV
362 	if ((cp->cas_flags & CAS_FLAG_ENTROPY_DEV) == 0)
363 		return;
364 
365 	writel(BIM_LOCAL_DEV_PAD | BIM_LOCAL_DEV_PROM | BIM_LOCAL_DEV_EXT,
366 	       cp->regs + REG_BIM_LOCAL_DEV_EN);
367 	writeb(ENTROPY_RESET_STC_MODE, cp->regs + REG_ENTROPY_RESET);
368 	writeb(0x55, cp->regs + REG_ENTROPY_RAND_REG);
369 
370 	/* if we read back 0x0, we don't have an entropy device */
371 	if (readb(cp->regs + REG_ENTROPY_RAND_REG) == 0)
372 		cp->cas_flags &= ~CAS_FLAG_ENTROPY_DEV;
373 #endif
374 }
375 
376 /* access to the phy. the following assumes that we've initialized the MIF to
377  * be in frame rather than bit-bang mode
378  */
379 static u16 cas_phy_read(struct cas *cp, int reg)
380 {
381 	u32 cmd;
382 	int limit = STOP_TRIES_PHY;
383 
384 	cmd = MIF_FRAME_ST | MIF_FRAME_OP_READ;
385 	cmd |= CAS_BASE(MIF_FRAME_PHY_ADDR, cp->phy_addr);
386 	cmd |= CAS_BASE(MIF_FRAME_REG_ADDR, reg);
387 	cmd |= MIF_FRAME_TURN_AROUND_MSB;
388 	writel(cmd, cp->regs + REG_MIF_FRAME);
389 
390 	/* poll for completion */
391 	while (limit-- > 0) {
392 		udelay(10);
393 		cmd = readl(cp->regs + REG_MIF_FRAME);
394 		if (cmd & MIF_FRAME_TURN_AROUND_LSB)
395 			return cmd & MIF_FRAME_DATA_MASK;
396 	}
397 	return 0xFFFF; /* -1 */
398 }
399 
400 static int cas_phy_write(struct cas *cp, int reg, u16 val)
401 {
402 	int limit = STOP_TRIES_PHY;
403 	u32 cmd;
404 
405 	cmd = MIF_FRAME_ST | MIF_FRAME_OP_WRITE;
406 	cmd |= CAS_BASE(MIF_FRAME_PHY_ADDR, cp->phy_addr);
407 	cmd |= CAS_BASE(MIF_FRAME_REG_ADDR, reg);
408 	cmd |= MIF_FRAME_TURN_AROUND_MSB;
409 	cmd |= val & MIF_FRAME_DATA_MASK;
410 	writel(cmd, cp->regs + REG_MIF_FRAME);
411 
412 	/* poll for completion */
413 	while (limit-- > 0) {
414 		udelay(10);
415 		cmd = readl(cp->regs + REG_MIF_FRAME);
416 		if (cmd & MIF_FRAME_TURN_AROUND_LSB)
417 			return 0;
418 	}
419 	return -1;
420 }
421 
422 static void cas_phy_powerup(struct cas *cp)
423 {
424 	u16 ctl = cas_phy_read(cp, MII_BMCR);
425 
426 	if ((ctl & BMCR_PDOWN) == 0)
427 		return;
428 	ctl &= ~BMCR_PDOWN;
429 	cas_phy_write(cp, MII_BMCR, ctl);
430 }
431 
432 static void cas_phy_powerdown(struct cas *cp)
433 {
434 	u16 ctl = cas_phy_read(cp, MII_BMCR);
435 
436 	if (ctl & BMCR_PDOWN)
437 		return;
438 	ctl |= BMCR_PDOWN;
439 	cas_phy_write(cp, MII_BMCR, ctl);
440 }
441 
442 /* cp->lock held. note: the last put_page will free the buffer */
443 static int cas_page_free(struct cas *cp, cas_page_t *page)
444 {
445 	dma_unmap_page(&cp->pdev->dev, page->dma_addr, cp->page_size,
446 		       DMA_FROM_DEVICE);
447 	__free_pages(page->buffer, cp->page_order);
448 	kfree(page);
449 	return 0;
450 }
451 
452 #ifdef RX_COUNT_BUFFERS
453 #define RX_USED_ADD(x, y)       ((x)->used += (y))
454 #define RX_USED_SET(x, y)       ((x)->used  = (y))
455 #else
456 #define RX_USED_ADD(x, y) do { } while(0)
457 #define RX_USED_SET(x, y) do { } while(0)
458 #endif
459 
460 /* local page allocation routines for the receive buffers. jumbo pages
461  * require at least 8K contiguous and 8K aligned buffers.
462  */
463 static cas_page_t *cas_page_alloc(struct cas *cp, const gfp_t flags)
464 {
465 	cas_page_t *page;
466 
467 	page = kmalloc(sizeof(cas_page_t), flags);
468 	if (!page)
469 		return NULL;
470 
471 	INIT_LIST_HEAD(&page->list);
472 	RX_USED_SET(page, 0);
473 	page->buffer = alloc_pages(flags, cp->page_order);
474 	if (!page->buffer)
475 		goto page_err;
476 	page->dma_addr = dma_map_page(&cp->pdev->dev, page->buffer, 0,
477 				      cp->page_size, DMA_FROM_DEVICE);
478 	return page;
479 
480 page_err:
481 	kfree(page);
482 	return NULL;
483 }
484 
485 /* initialize spare pool of rx buffers, but allocate during the open */
486 static void cas_spare_init(struct cas *cp)
487 {
488 	spin_lock(&cp->rx_inuse_lock);
489 	INIT_LIST_HEAD(&cp->rx_inuse_list);
490 	spin_unlock(&cp->rx_inuse_lock);
491 
492 	spin_lock(&cp->rx_spare_lock);
493 	INIT_LIST_HEAD(&cp->rx_spare_list);
494 	cp->rx_spares_needed = RX_SPARE_COUNT;
495 	spin_unlock(&cp->rx_spare_lock);
496 }
497 
498 /* used on close. free all the spare buffers. */
499 static void cas_spare_free(struct cas *cp)
500 {
501 	struct list_head list, *elem, *tmp;
502 
503 	/* free spare buffers */
504 	INIT_LIST_HEAD(&list);
505 	spin_lock(&cp->rx_spare_lock);
506 	list_splice_init(&cp->rx_spare_list, &list);
507 	spin_unlock(&cp->rx_spare_lock);
508 	list_for_each_safe(elem, tmp, &list) {
509 		cas_page_free(cp, list_entry(elem, cas_page_t, list));
510 	}
511 
512 	INIT_LIST_HEAD(&list);
513 #if 1
514 	/*
515 	 * Looks like Adrian had protected this with a different
516 	 * lock than used everywhere else to manipulate this list.
517 	 */
518 	spin_lock(&cp->rx_inuse_lock);
519 	list_splice_init(&cp->rx_inuse_list, &list);
520 	spin_unlock(&cp->rx_inuse_lock);
521 #else
522 	spin_lock(&cp->rx_spare_lock);
523 	list_splice_init(&cp->rx_inuse_list, &list);
524 	spin_unlock(&cp->rx_spare_lock);
525 #endif
526 	list_for_each_safe(elem, tmp, &list) {
527 		cas_page_free(cp, list_entry(elem, cas_page_t, list));
528 	}
529 }
530 
531 /* replenish spares if needed */
532 static void cas_spare_recover(struct cas *cp, const gfp_t flags)
533 {
534 	struct list_head list, *elem, *tmp;
535 	int needed, i;
536 
537 	/* check inuse list. if we don't need any more free buffers,
538 	 * just free it
539 	 */
540 
541 	/* make a local copy of the list */
542 	INIT_LIST_HEAD(&list);
543 	spin_lock(&cp->rx_inuse_lock);
544 	list_splice_init(&cp->rx_inuse_list, &list);
545 	spin_unlock(&cp->rx_inuse_lock);
546 
547 	list_for_each_safe(elem, tmp, &list) {
548 		cas_page_t *page = list_entry(elem, cas_page_t, list);
549 
550 		/*
551 		 * With the lockless pagecache, cassini buffering scheme gets
552 		 * slightly less accurate: we might find that a page has an
553 		 * elevated reference count here, due to a speculative ref,
554 		 * and skip it as in-use. Ideally we would be able to reclaim
555 		 * it. However this would be such a rare case, it doesn't
556 		 * matter too much as we should pick it up the next time round.
557 		 *
558 		 * Importantly, if we find that the page has a refcount of 1
559 		 * here (our refcount), then we know it is definitely not inuse
560 		 * so we can reuse it.
561 		 */
562 		if (page_count(page->buffer) > 1)
563 			continue;
564 
565 		list_del(elem);
566 		spin_lock(&cp->rx_spare_lock);
567 		if (cp->rx_spares_needed > 0) {
568 			list_add(elem, &cp->rx_spare_list);
569 			cp->rx_spares_needed--;
570 			spin_unlock(&cp->rx_spare_lock);
571 		} else {
572 			spin_unlock(&cp->rx_spare_lock);
573 			cas_page_free(cp, page);
574 		}
575 	}
576 
577 	/* put any inuse buffers back on the list */
578 	if (!list_empty(&list)) {
579 		spin_lock(&cp->rx_inuse_lock);
580 		list_splice(&list, &cp->rx_inuse_list);
581 		spin_unlock(&cp->rx_inuse_lock);
582 	}
583 
584 	spin_lock(&cp->rx_spare_lock);
585 	needed = cp->rx_spares_needed;
586 	spin_unlock(&cp->rx_spare_lock);
587 	if (!needed)
588 		return;
589 
590 	/* we still need spares, so try to allocate some */
591 	INIT_LIST_HEAD(&list);
592 	i = 0;
593 	while (i < needed) {
594 		cas_page_t *spare = cas_page_alloc(cp, flags);
595 		if (!spare)
596 			break;
597 		list_add(&spare->list, &list);
598 		i++;
599 	}
600 
601 	spin_lock(&cp->rx_spare_lock);
602 	list_splice(&list, &cp->rx_spare_list);
603 	cp->rx_spares_needed -= i;
604 	spin_unlock(&cp->rx_spare_lock);
605 }
606 
607 /* pull a page from the list. */
608 static cas_page_t *cas_page_dequeue(struct cas *cp)
609 {
610 	struct list_head *entry;
611 	int recover;
612 
613 	spin_lock(&cp->rx_spare_lock);
614 	if (list_empty(&cp->rx_spare_list)) {
615 		/* try to do a quick recovery */
616 		spin_unlock(&cp->rx_spare_lock);
617 		cas_spare_recover(cp, GFP_ATOMIC);
618 		spin_lock(&cp->rx_spare_lock);
619 		if (list_empty(&cp->rx_spare_list)) {
620 			netif_err(cp, rx_err, cp->dev,
621 				  "no spare buffers available\n");
622 			spin_unlock(&cp->rx_spare_lock);
623 			return NULL;
624 		}
625 	}
626 
627 	entry = cp->rx_spare_list.next;
628 	list_del(entry);
629 	recover = ++cp->rx_spares_needed;
630 	spin_unlock(&cp->rx_spare_lock);
631 
632 	/* trigger the timer to do the recovery */
633 	if ((recover & (RX_SPARE_RECOVER_VAL - 1)) == 0) {
634 #if 1
635 		atomic_inc(&cp->reset_task_pending);
636 		atomic_inc(&cp->reset_task_pending_spare);
637 		schedule_work(&cp->reset_task);
638 #else
639 		atomic_set(&cp->reset_task_pending, CAS_RESET_SPARE);
640 		schedule_work(&cp->reset_task);
641 #endif
642 	}
643 	return list_entry(entry, cas_page_t, list);
644 }
645 
646 
647 static void cas_mif_poll(struct cas *cp, const int enable)
648 {
649 	u32 cfg;
650 
651 	cfg  = readl(cp->regs + REG_MIF_CFG);
652 	cfg &= (MIF_CFG_MDIO_0 | MIF_CFG_MDIO_1);
653 
654 	if (cp->phy_type & CAS_PHY_MII_MDIO1)
655 		cfg |= MIF_CFG_PHY_SELECT;
656 
657 	/* poll and interrupt on link status change. */
658 	if (enable) {
659 		cfg |= MIF_CFG_POLL_EN;
660 		cfg |= CAS_BASE(MIF_CFG_POLL_REG, MII_BMSR);
661 		cfg |= CAS_BASE(MIF_CFG_POLL_PHY, cp->phy_addr);
662 	}
663 	writel((enable) ? ~(BMSR_LSTATUS | BMSR_ANEGCOMPLETE) : 0xFFFF,
664 	       cp->regs + REG_MIF_MASK);
665 	writel(cfg, cp->regs + REG_MIF_CFG);
666 }
667 
668 /* Must be invoked under cp->lock */
669 static void cas_begin_auto_negotiation(struct cas *cp,
670 				       const struct ethtool_link_ksettings *ep)
671 {
672 	u16 ctl;
673 #if 1
674 	int lcntl;
675 	int changed = 0;
676 	int oldstate = cp->lstate;
677 	int link_was_not_down = !(oldstate == link_down);
678 #endif
679 	/* Setup link parameters */
680 	if (!ep)
681 		goto start_aneg;
682 	lcntl = cp->link_cntl;
683 	if (ep->base.autoneg == AUTONEG_ENABLE) {
684 		cp->link_cntl = BMCR_ANENABLE;
685 	} else {
686 		u32 speed = ep->base.speed;
687 		cp->link_cntl = 0;
688 		if (speed == SPEED_100)
689 			cp->link_cntl |= BMCR_SPEED100;
690 		else if (speed == SPEED_1000)
691 			cp->link_cntl |= CAS_BMCR_SPEED1000;
692 		if (ep->base.duplex == DUPLEX_FULL)
693 			cp->link_cntl |= BMCR_FULLDPLX;
694 	}
695 #if 1
696 	changed = (lcntl != cp->link_cntl);
697 #endif
698 start_aneg:
699 	if (cp->lstate == link_up) {
700 		netdev_info(cp->dev, "PCS link down\n");
701 	} else {
702 		if (changed) {
703 			netdev_info(cp->dev, "link configuration changed\n");
704 		}
705 	}
706 	cp->lstate = link_down;
707 	cp->link_transition = LINK_TRANSITION_LINK_DOWN;
708 	if (!cp->hw_running)
709 		return;
710 #if 1
711 	/*
712 	 * WTZ: If the old state was link_up, we turn off the carrier
713 	 * to replicate everything we do elsewhere on a link-down
714 	 * event when we were already in a link-up state..
715 	 */
716 	if (oldstate == link_up)
717 		netif_carrier_off(cp->dev);
718 	if (changed  && link_was_not_down) {
719 		/*
720 		 * WTZ: This branch will simply schedule a full reset after
721 		 * we explicitly changed link modes in an ioctl. See if this
722 		 * fixes the link-problems we were having for forced mode.
723 		 */
724 		atomic_inc(&cp->reset_task_pending);
725 		atomic_inc(&cp->reset_task_pending_all);
726 		schedule_work(&cp->reset_task);
727 		cp->timer_ticks = 0;
728 		mod_timer(&cp->link_timer, jiffies + CAS_LINK_TIMEOUT);
729 		return;
730 	}
731 #endif
732 	if (cp->phy_type & CAS_PHY_SERDES) {
733 		u32 val = readl(cp->regs + REG_PCS_MII_CTRL);
734 
735 		if (cp->link_cntl & BMCR_ANENABLE) {
736 			val |= (PCS_MII_RESTART_AUTONEG | PCS_MII_AUTONEG_EN);
737 			cp->lstate = link_aneg;
738 		} else {
739 			if (cp->link_cntl & BMCR_FULLDPLX)
740 				val |= PCS_MII_CTRL_DUPLEX;
741 			val &= ~PCS_MII_AUTONEG_EN;
742 			cp->lstate = link_force_ok;
743 		}
744 		cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
745 		writel(val, cp->regs + REG_PCS_MII_CTRL);
746 
747 	} else {
748 		cas_mif_poll(cp, 0);
749 		ctl = cas_phy_read(cp, MII_BMCR);
750 		ctl &= ~(BMCR_FULLDPLX | BMCR_SPEED100 |
751 			 CAS_BMCR_SPEED1000 | BMCR_ANENABLE);
752 		ctl |= cp->link_cntl;
753 		if (ctl & BMCR_ANENABLE) {
754 			ctl |= BMCR_ANRESTART;
755 			cp->lstate = link_aneg;
756 		} else {
757 			cp->lstate = link_force_ok;
758 		}
759 		cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
760 		cas_phy_write(cp, MII_BMCR, ctl);
761 		cas_mif_poll(cp, 1);
762 	}
763 
764 	cp->timer_ticks = 0;
765 	mod_timer(&cp->link_timer, jiffies + CAS_LINK_TIMEOUT);
766 }
767 
768 /* Must be invoked under cp->lock. */
769 static int cas_reset_mii_phy(struct cas *cp)
770 {
771 	int limit = STOP_TRIES_PHY;
772 	u16 val;
773 
774 	cas_phy_write(cp, MII_BMCR, BMCR_RESET);
775 	udelay(100);
776 	while (--limit) {
777 		val = cas_phy_read(cp, MII_BMCR);
778 		if ((val & BMCR_RESET) == 0)
779 			break;
780 		udelay(10);
781 	}
782 	return limit <= 0;
783 }
784 
785 static void cas_saturn_firmware_init(struct cas *cp)
786 {
787 	const struct firmware *fw;
788 	const char fw_name[] = "sun/cassini.bin";
789 	int err;
790 
791 	if (PHY_NS_DP83065 != cp->phy_id)
792 		return;
793 
794 	err = request_firmware(&fw, fw_name, &cp->pdev->dev);
795 	if (err) {
796 		pr_err("Failed to load firmware \"%s\"\n",
797 		       fw_name);
798 		return;
799 	}
800 	if (fw->size < 2) {
801 		pr_err("bogus length %zu in \"%s\"\n",
802 		       fw->size, fw_name);
803 		goto out;
804 	}
805 	cp->fw_load_addr= fw->data[1] << 8 | fw->data[0];
806 	cp->fw_size = fw->size - 2;
807 	cp->fw_data = vmalloc(cp->fw_size);
808 	if (!cp->fw_data)
809 		goto out;
810 	memcpy(cp->fw_data, &fw->data[2], cp->fw_size);
811 out:
812 	release_firmware(fw);
813 }
814 
815 static void cas_saturn_firmware_load(struct cas *cp)
816 {
817 	int i;
818 
819 	if (!cp->fw_data)
820 		return;
821 
822 	cas_phy_powerdown(cp);
823 
824 	/* expanded memory access mode */
825 	cas_phy_write(cp, DP83065_MII_MEM, 0x0);
826 
827 	/* pointer configuration for new firmware */
828 	cas_phy_write(cp, DP83065_MII_REGE, 0x8ff9);
829 	cas_phy_write(cp, DP83065_MII_REGD, 0xbd);
830 	cas_phy_write(cp, DP83065_MII_REGE, 0x8ffa);
831 	cas_phy_write(cp, DP83065_MII_REGD, 0x82);
832 	cas_phy_write(cp, DP83065_MII_REGE, 0x8ffb);
833 	cas_phy_write(cp, DP83065_MII_REGD, 0x0);
834 	cas_phy_write(cp, DP83065_MII_REGE, 0x8ffc);
835 	cas_phy_write(cp, DP83065_MII_REGD, 0x39);
836 
837 	/* download new firmware */
838 	cas_phy_write(cp, DP83065_MII_MEM, 0x1);
839 	cas_phy_write(cp, DP83065_MII_REGE, cp->fw_load_addr);
840 	for (i = 0; i < cp->fw_size; i++)
841 		cas_phy_write(cp, DP83065_MII_REGD, cp->fw_data[i]);
842 
843 	/* enable firmware */
844 	cas_phy_write(cp, DP83065_MII_REGE, 0x8ff8);
845 	cas_phy_write(cp, DP83065_MII_REGD, 0x1);
846 }
847 
848 
849 /* phy initialization */
850 static void cas_phy_init(struct cas *cp)
851 {
852 	u16 val;
853 
854 	/* if we're in MII/GMII mode, set up phy */
855 	if (CAS_PHY_MII(cp->phy_type)) {
856 		writel(PCS_DATAPATH_MODE_MII,
857 		       cp->regs + REG_PCS_DATAPATH_MODE);
858 
859 		cas_mif_poll(cp, 0);
860 		cas_reset_mii_phy(cp); /* take out of isolate mode */
861 
862 		if (PHY_LUCENT_B0 == cp->phy_id) {
863 			/* workaround link up/down issue with lucent */
864 			cas_phy_write(cp, LUCENT_MII_REG, 0x8000);
865 			cas_phy_write(cp, MII_BMCR, 0x00f1);
866 			cas_phy_write(cp, LUCENT_MII_REG, 0x0);
867 
868 		} else if (PHY_BROADCOM_B0 == (cp->phy_id & 0xFFFFFFFC)) {
869 			/* workarounds for broadcom phy */
870 			cas_phy_write(cp, BROADCOM_MII_REG8, 0x0C20);
871 			cas_phy_write(cp, BROADCOM_MII_REG7, 0x0012);
872 			cas_phy_write(cp, BROADCOM_MII_REG5, 0x1804);
873 			cas_phy_write(cp, BROADCOM_MII_REG7, 0x0013);
874 			cas_phy_write(cp, BROADCOM_MII_REG5, 0x1204);
875 			cas_phy_write(cp, BROADCOM_MII_REG7, 0x8006);
876 			cas_phy_write(cp, BROADCOM_MII_REG5, 0x0132);
877 			cas_phy_write(cp, BROADCOM_MII_REG7, 0x8006);
878 			cas_phy_write(cp, BROADCOM_MII_REG5, 0x0232);
879 			cas_phy_write(cp, BROADCOM_MII_REG7, 0x201F);
880 			cas_phy_write(cp, BROADCOM_MII_REG5, 0x0A20);
881 
882 		} else if (PHY_BROADCOM_5411 == cp->phy_id) {
883 			val = cas_phy_read(cp, BROADCOM_MII_REG4);
884 			val = cas_phy_read(cp, BROADCOM_MII_REG4);
885 			if (val & 0x0080) {
886 				/* link workaround */
887 				cas_phy_write(cp, BROADCOM_MII_REG4,
888 					      val & ~0x0080);
889 			}
890 
891 		} else if (cp->cas_flags & CAS_FLAG_SATURN) {
892 			writel((cp->phy_type & CAS_PHY_MII_MDIO0) ?
893 			       SATURN_PCFG_FSI : 0x0,
894 			       cp->regs + REG_SATURN_PCFG);
895 
896 			/* load firmware to address 10Mbps auto-negotiation
897 			 * issue. NOTE: this will need to be changed if the
898 			 * default firmware gets fixed.
899 			 */
900 			if (PHY_NS_DP83065 == cp->phy_id) {
901 				cas_saturn_firmware_load(cp);
902 			}
903 			cas_phy_powerup(cp);
904 		}
905 
906 		/* advertise capabilities */
907 		val = cas_phy_read(cp, MII_BMCR);
908 		val &= ~BMCR_ANENABLE;
909 		cas_phy_write(cp, MII_BMCR, val);
910 		udelay(10);
911 
912 		cas_phy_write(cp, MII_ADVERTISE,
913 			      cas_phy_read(cp, MII_ADVERTISE) |
914 			      (ADVERTISE_10HALF | ADVERTISE_10FULL |
915 			       ADVERTISE_100HALF | ADVERTISE_100FULL |
916 			       CAS_ADVERTISE_PAUSE |
917 			       CAS_ADVERTISE_ASYM_PAUSE));
918 
919 		if (cp->cas_flags & CAS_FLAG_1000MB_CAP) {
920 			/* make sure that we don't advertise half
921 			 * duplex to avoid a chip issue
922 			 */
923 			val  = cas_phy_read(cp, CAS_MII_1000_CTRL);
924 			val &= ~CAS_ADVERTISE_1000HALF;
925 			val |= CAS_ADVERTISE_1000FULL;
926 			cas_phy_write(cp, CAS_MII_1000_CTRL, val);
927 		}
928 
929 	} else {
930 		/* reset pcs for serdes */
931 		u32 val;
932 		int limit;
933 
934 		writel(PCS_DATAPATH_MODE_SERDES,
935 		       cp->regs + REG_PCS_DATAPATH_MODE);
936 
937 		/* enable serdes pins on saturn */
938 		if (cp->cas_flags & CAS_FLAG_SATURN)
939 			writel(0, cp->regs + REG_SATURN_PCFG);
940 
941 		/* Reset PCS unit. */
942 		val = readl(cp->regs + REG_PCS_MII_CTRL);
943 		val |= PCS_MII_RESET;
944 		writel(val, cp->regs + REG_PCS_MII_CTRL);
945 
946 		limit = STOP_TRIES;
947 		while (--limit > 0) {
948 			udelay(10);
949 			if ((readl(cp->regs + REG_PCS_MII_CTRL) &
950 			     PCS_MII_RESET) == 0)
951 				break;
952 		}
953 		if (limit <= 0)
954 			netdev_warn(cp->dev, "PCS reset bit would not clear [%08x]\n",
955 				    readl(cp->regs + REG_PCS_STATE_MACHINE));
956 
957 		/* Make sure PCS is disabled while changing advertisement
958 		 * configuration.
959 		 */
960 		writel(0x0, cp->regs + REG_PCS_CFG);
961 
962 		/* Advertise all capabilities except half-duplex. */
963 		val  = readl(cp->regs + REG_PCS_MII_ADVERT);
964 		val &= ~PCS_MII_ADVERT_HD;
965 		val |= (PCS_MII_ADVERT_FD | PCS_MII_ADVERT_SYM_PAUSE |
966 			PCS_MII_ADVERT_ASYM_PAUSE);
967 		writel(val, cp->regs + REG_PCS_MII_ADVERT);
968 
969 		/* enable PCS */
970 		writel(PCS_CFG_EN, cp->regs + REG_PCS_CFG);
971 
972 		/* pcs workaround: enable sync detect */
973 		writel(PCS_SERDES_CTRL_SYNCD_EN,
974 		       cp->regs + REG_PCS_SERDES_CTRL);
975 	}
976 }
977 
978 
979 static int cas_pcs_link_check(struct cas *cp)
980 {
981 	u32 stat, state_machine;
982 	int retval = 0;
983 
984 	/* The link status bit latches on zero, so you must
985 	 * read it twice in such a case to see a transition
986 	 * to the link being up.
987 	 */
988 	stat = readl(cp->regs + REG_PCS_MII_STATUS);
989 	if ((stat & PCS_MII_STATUS_LINK_STATUS) == 0)
990 		stat = readl(cp->regs + REG_PCS_MII_STATUS);
991 
992 	/* The remote-fault indication is only valid
993 	 * when autoneg has completed.
994 	 */
995 	if ((stat & (PCS_MII_STATUS_AUTONEG_COMP |
996 		     PCS_MII_STATUS_REMOTE_FAULT)) ==
997 	    (PCS_MII_STATUS_AUTONEG_COMP | PCS_MII_STATUS_REMOTE_FAULT))
998 		netif_info(cp, link, cp->dev, "PCS RemoteFault\n");
999 
1000 	/* work around link detection issue by querying the PCS state
1001 	 * machine directly.
1002 	 */
1003 	state_machine = readl(cp->regs + REG_PCS_STATE_MACHINE);
1004 	if ((state_machine & PCS_SM_LINK_STATE_MASK) != SM_LINK_STATE_UP) {
1005 		stat &= ~PCS_MII_STATUS_LINK_STATUS;
1006 	} else if (state_machine & PCS_SM_WORD_SYNC_STATE_MASK) {
1007 		stat |= PCS_MII_STATUS_LINK_STATUS;
1008 	}
1009 
1010 	if (stat & PCS_MII_STATUS_LINK_STATUS) {
1011 		if (cp->lstate != link_up) {
1012 			if (cp->opened) {
1013 				cp->lstate = link_up;
1014 				cp->link_transition = LINK_TRANSITION_LINK_UP;
1015 
1016 				cas_set_link_modes(cp);
1017 				netif_carrier_on(cp->dev);
1018 			}
1019 		}
1020 	} else if (cp->lstate == link_up) {
1021 		cp->lstate = link_down;
1022 		if (link_transition_timeout != 0 &&
1023 		    cp->link_transition != LINK_TRANSITION_REQUESTED_RESET &&
1024 		    !cp->link_transition_jiffies_valid) {
1025 			/*
1026 			 * force a reset, as a workaround for the
1027 			 * link-failure problem. May want to move this to a
1028 			 * point a bit earlier in the sequence. If we had
1029 			 * generated a reset a short time ago, we'll wait for
1030 			 * the link timer to check the status until a
1031 			 * timer expires (link_transistion_jiffies_valid is
1032 			 * true when the timer is running.)  Instead of using
1033 			 * a system timer, we just do a check whenever the
1034 			 * link timer is running - this clears the flag after
1035 			 * a suitable delay.
1036 			 */
1037 			retval = 1;
1038 			cp->link_transition = LINK_TRANSITION_REQUESTED_RESET;
1039 			cp->link_transition_jiffies = jiffies;
1040 			cp->link_transition_jiffies_valid = 1;
1041 		} else {
1042 			cp->link_transition = LINK_TRANSITION_ON_FAILURE;
1043 		}
1044 		netif_carrier_off(cp->dev);
1045 		if (cp->opened)
1046 			netif_info(cp, link, cp->dev, "PCS link down\n");
1047 
1048 		/* Cassini only: if you force a mode, there can be
1049 		 * sync problems on link down. to fix that, the following
1050 		 * things need to be checked:
1051 		 * 1) read serialink state register
1052 		 * 2) read pcs status register to verify link down.
1053 		 * 3) if link down and serial link == 0x03, then you need
1054 		 *    to global reset the chip.
1055 		 */
1056 		if ((cp->cas_flags & CAS_FLAG_REG_PLUS) == 0) {
1057 			/* should check to see if we're in a forced mode */
1058 			stat = readl(cp->regs + REG_PCS_SERDES_STATE);
1059 			if (stat == 0x03)
1060 				return 1;
1061 		}
1062 	} else if (cp->lstate == link_down) {
1063 		if (link_transition_timeout != 0 &&
1064 		    cp->link_transition != LINK_TRANSITION_REQUESTED_RESET &&
1065 		    !cp->link_transition_jiffies_valid) {
1066 			/* force a reset, as a workaround for the
1067 			 * link-failure problem.  May want to move
1068 			 * this to a point a bit earlier in the
1069 			 * sequence.
1070 			 */
1071 			retval = 1;
1072 			cp->link_transition = LINK_TRANSITION_REQUESTED_RESET;
1073 			cp->link_transition_jiffies = jiffies;
1074 			cp->link_transition_jiffies_valid = 1;
1075 		} else {
1076 			cp->link_transition = LINK_TRANSITION_STILL_FAILED;
1077 		}
1078 	}
1079 
1080 	return retval;
1081 }
1082 
1083 static int cas_pcs_interrupt(struct net_device *dev,
1084 			     struct cas *cp, u32 status)
1085 {
1086 	u32 stat = readl(cp->regs + REG_PCS_INTR_STATUS);
1087 
1088 	if ((stat & PCS_INTR_STATUS_LINK_CHANGE) == 0)
1089 		return 0;
1090 	return cas_pcs_link_check(cp);
1091 }
1092 
1093 static int cas_txmac_interrupt(struct net_device *dev,
1094 			       struct cas *cp, u32 status)
1095 {
1096 	u32 txmac_stat = readl(cp->regs + REG_MAC_TX_STATUS);
1097 
1098 	if (!txmac_stat)
1099 		return 0;
1100 
1101 	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
1102 		     "txmac interrupt, txmac_stat: 0x%x\n", txmac_stat);
1103 
1104 	/* Defer timer expiration is quite normal,
1105 	 * don't even log the event.
1106 	 */
1107 	if ((txmac_stat & MAC_TX_DEFER_TIMER) &&
1108 	    !(txmac_stat & ~MAC_TX_DEFER_TIMER))
1109 		return 0;
1110 
1111 	spin_lock(&cp->stat_lock[0]);
1112 	if (txmac_stat & MAC_TX_UNDERRUN) {
1113 		netdev_err(dev, "TX MAC xmit underrun\n");
1114 		cp->net_stats[0].tx_fifo_errors++;
1115 	}
1116 
1117 	if (txmac_stat & MAC_TX_MAX_PACKET_ERR) {
1118 		netdev_err(dev, "TX MAC max packet size error\n");
1119 		cp->net_stats[0].tx_errors++;
1120 	}
1121 
1122 	/* The rest are all cases of one of the 16-bit TX
1123 	 * counters expiring.
1124 	 */
1125 	if (txmac_stat & MAC_TX_COLL_NORMAL)
1126 		cp->net_stats[0].collisions += 0x10000;
1127 
1128 	if (txmac_stat & MAC_TX_COLL_EXCESS) {
1129 		cp->net_stats[0].tx_aborted_errors += 0x10000;
1130 		cp->net_stats[0].collisions += 0x10000;
1131 	}
1132 
1133 	if (txmac_stat & MAC_TX_COLL_LATE) {
1134 		cp->net_stats[0].tx_aborted_errors += 0x10000;
1135 		cp->net_stats[0].collisions += 0x10000;
1136 	}
1137 	spin_unlock(&cp->stat_lock[0]);
1138 
1139 	/* We do not keep track of MAC_TX_COLL_FIRST and
1140 	 * MAC_TX_PEAK_ATTEMPTS events.
1141 	 */
1142 	return 0;
1143 }
1144 
1145 static void cas_load_firmware(struct cas *cp, cas_hp_inst_t *firmware)
1146 {
1147 	cas_hp_inst_t *inst;
1148 	u32 val;
1149 	int i;
1150 
1151 	i = 0;
1152 	while ((inst = firmware) && inst->note) {
1153 		writel(i, cp->regs + REG_HP_INSTR_RAM_ADDR);
1154 
1155 		val = CAS_BASE(HP_INSTR_RAM_HI_VAL, inst->val);
1156 		val |= CAS_BASE(HP_INSTR_RAM_HI_MASK, inst->mask);
1157 		writel(val, cp->regs + REG_HP_INSTR_RAM_DATA_HI);
1158 
1159 		val = CAS_BASE(HP_INSTR_RAM_MID_OUTARG, inst->outarg >> 10);
1160 		val |= CAS_BASE(HP_INSTR_RAM_MID_OUTOP, inst->outop);
1161 		val |= CAS_BASE(HP_INSTR_RAM_MID_FNEXT, inst->fnext);
1162 		val |= CAS_BASE(HP_INSTR_RAM_MID_FOFF, inst->foff);
1163 		val |= CAS_BASE(HP_INSTR_RAM_MID_SNEXT, inst->snext);
1164 		val |= CAS_BASE(HP_INSTR_RAM_MID_SOFF, inst->soff);
1165 		val |= CAS_BASE(HP_INSTR_RAM_MID_OP, inst->op);
1166 		writel(val, cp->regs + REG_HP_INSTR_RAM_DATA_MID);
1167 
1168 		val = CAS_BASE(HP_INSTR_RAM_LOW_OUTMASK, inst->outmask);
1169 		val |= CAS_BASE(HP_INSTR_RAM_LOW_OUTSHIFT, inst->outshift);
1170 		val |= CAS_BASE(HP_INSTR_RAM_LOW_OUTEN, inst->outenab);
1171 		val |= CAS_BASE(HP_INSTR_RAM_LOW_OUTARG, inst->outarg);
1172 		writel(val, cp->regs + REG_HP_INSTR_RAM_DATA_LOW);
1173 		++firmware;
1174 		++i;
1175 	}
1176 }
1177 
1178 static void cas_init_rx_dma(struct cas *cp)
1179 {
1180 	u64 desc_dma = cp->block_dvma;
1181 	u32 val;
1182 	int i, size;
1183 
1184 	/* rx free descriptors */
1185 	val = CAS_BASE(RX_CFG_SWIVEL, RX_SWIVEL_OFF_VAL);
1186 	val |= CAS_BASE(RX_CFG_DESC_RING, RX_DESC_RINGN_INDEX(0));
1187 	val |= CAS_BASE(RX_CFG_COMP_RING, RX_COMP_RINGN_INDEX(0));
1188 	if ((N_RX_DESC_RINGS > 1) &&
1189 	    (cp->cas_flags & CAS_FLAG_REG_PLUS))  /* do desc 2 */
1190 		val |= CAS_BASE(RX_CFG_DESC_RING1, RX_DESC_RINGN_INDEX(1));
1191 	writel(val, cp->regs + REG_RX_CFG);
1192 
1193 	val = (unsigned long) cp->init_rxds[0] -
1194 		(unsigned long) cp->init_block;
1195 	writel((desc_dma + val) >> 32, cp->regs + REG_RX_DB_HI);
1196 	writel((desc_dma + val) & 0xffffffff, cp->regs + REG_RX_DB_LOW);
1197 	writel(RX_DESC_RINGN_SIZE(0) - 4, cp->regs + REG_RX_KICK);
1198 
1199 	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
1200 		/* rx desc 2 is for IPSEC packets. however,
1201 		 * we don't it that for that purpose.
1202 		 */
1203 		val = (unsigned long) cp->init_rxds[1] -
1204 			(unsigned long) cp->init_block;
1205 		writel((desc_dma + val) >> 32, cp->regs + REG_PLUS_RX_DB1_HI);
1206 		writel((desc_dma + val) & 0xffffffff, cp->regs +
1207 		       REG_PLUS_RX_DB1_LOW);
1208 		writel(RX_DESC_RINGN_SIZE(1) - 4, cp->regs +
1209 		       REG_PLUS_RX_KICK1);
1210 	}
1211 
1212 	/* rx completion registers */
1213 	val = (unsigned long) cp->init_rxcs[0] -
1214 		(unsigned long) cp->init_block;
1215 	writel((desc_dma + val) >> 32, cp->regs + REG_RX_CB_HI);
1216 	writel((desc_dma + val) & 0xffffffff, cp->regs + REG_RX_CB_LOW);
1217 
1218 	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
1219 		/* rx comp 2-4 */
1220 		for (i = 1; i < MAX_RX_COMP_RINGS; i++) {
1221 			val = (unsigned long) cp->init_rxcs[i] -
1222 				(unsigned long) cp->init_block;
1223 			writel((desc_dma + val) >> 32, cp->regs +
1224 			       REG_PLUS_RX_CBN_HI(i));
1225 			writel((desc_dma + val) & 0xffffffff, cp->regs +
1226 			       REG_PLUS_RX_CBN_LOW(i));
1227 		}
1228 	}
1229 
1230 	/* read selective clear regs to prevent spurious interrupts
1231 	 * on reset because complete == kick.
1232 	 * selective clear set up to prevent interrupts on resets
1233 	 */
1234 	readl(cp->regs + REG_INTR_STATUS_ALIAS);
1235 	writel(INTR_RX_DONE | INTR_RX_BUF_UNAVAIL, cp->regs + REG_ALIAS_CLEAR);
1236 
1237 	/* set up pause thresholds */
1238 	val  = CAS_BASE(RX_PAUSE_THRESH_OFF,
1239 			cp->rx_pause_off / RX_PAUSE_THRESH_QUANTUM);
1240 	val |= CAS_BASE(RX_PAUSE_THRESH_ON,
1241 			cp->rx_pause_on / RX_PAUSE_THRESH_QUANTUM);
1242 	writel(val, cp->regs + REG_RX_PAUSE_THRESH);
1243 
1244 	/* zero out dma reassembly buffers */
1245 	for (i = 0; i < 64; i++) {
1246 		writel(i, cp->regs + REG_RX_TABLE_ADDR);
1247 		writel(0x0, cp->regs + REG_RX_TABLE_DATA_LOW);
1248 		writel(0x0, cp->regs + REG_RX_TABLE_DATA_MID);
1249 		writel(0x0, cp->regs + REG_RX_TABLE_DATA_HI);
1250 	}
1251 
1252 	/* make sure address register is 0 for normal operation */
1253 	writel(0x0, cp->regs + REG_RX_CTRL_FIFO_ADDR);
1254 	writel(0x0, cp->regs + REG_RX_IPP_FIFO_ADDR);
1255 
1256 	/* interrupt mitigation */
1257 #ifdef USE_RX_BLANK
1258 	val = CAS_BASE(RX_BLANK_INTR_TIME, RX_BLANK_INTR_TIME_VAL);
1259 	val |= CAS_BASE(RX_BLANK_INTR_PKT, RX_BLANK_INTR_PKT_VAL);
1260 	writel(val, cp->regs + REG_RX_BLANK);
1261 #else
1262 	writel(0x0, cp->regs + REG_RX_BLANK);
1263 #endif
1264 
1265 	/* interrupt generation as a function of low water marks for
1266 	 * free desc and completion entries. these are used to trigger
1267 	 * housekeeping for rx descs. we don't use the free interrupt
1268 	 * as it's not very useful
1269 	 */
1270 	/* val = CAS_BASE(RX_AE_THRESH_FREE, RX_AE_FREEN_VAL(0)); */
1271 	val = CAS_BASE(RX_AE_THRESH_COMP, RX_AE_COMP_VAL);
1272 	writel(val, cp->regs + REG_RX_AE_THRESH);
1273 	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
1274 		val = CAS_BASE(RX_AE1_THRESH_FREE, RX_AE_FREEN_VAL(1));
1275 		writel(val, cp->regs + REG_PLUS_RX_AE1_THRESH);
1276 	}
1277 
1278 	/* Random early detect registers. useful for congestion avoidance.
1279 	 * this should be tunable.
1280 	 */
1281 	writel(0x0, cp->regs + REG_RX_RED);
1282 
1283 	/* receive page sizes. default == 2K (0x800) */
1284 	val = 0;
1285 	if (cp->page_size == 0x1000)
1286 		val = 0x1;
1287 	else if (cp->page_size == 0x2000)
1288 		val = 0x2;
1289 	else if (cp->page_size == 0x4000)
1290 		val = 0x3;
1291 
1292 	/* round mtu + offset. constrain to page size. */
1293 	size = cp->dev->mtu + 64;
1294 	if (size > cp->page_size)
1295 		size = cp->page_size;
1296 
1297 	if (size <= 0x400)
1298 		i = 0x0;
1299 	else if (size <= 0x800)
1300 		i = 0x1;
1301 	else if (size <= 0x1000)
1302 		i = 0x2;
1303 	else
1304 		i = 0x3;
1305 
1306 	cp->mtu_stride = 1 << (i + 10);
1307 	val  = CAS_BASE(RX_PAGE_SIZE, val);
1308 	val |= CAS_BASE(RX_PAGE_SIZE_MTU_STRIDE, i);
1309 	val |= CAS_BASE(RX_PAGE_SIZE_MTU_COUNT, cp->page_size >> (i + 10));
1310 	val |= CAS_BASE(RX_PAGE_SIZE_MTU_OFF, 0x1);
1311 	writel(val, cp->regs + REG_RX_PAGE_SIZE);
1312 
1313 	/* enable the header parser if desired */
1314 	if (&CAS_HP_FIRMWARE[0] == &cas_prog_null[0])
1315 		return;
1316 
1317 	val = CAS_BASE(HP_CFG_NUM_CPU, CAS_NCPUS > 63 ? 0 : CAS_NCPUS);
1318 	val |= HP_CFG_PARSE_EN | HP_CFG_SYN_INC_MASK;
1319 	val |= CAS_BASE(HP_CFG_TCP_THRESH, HP_TCP_THRESH_VAL);
1320 	writel(val, cp->regs + REG_HP_CFG);
1321 }
1322 
1323 static inline void cas_rxc_init(struct cas_rx_comp *rxc)
1324 {
1325 	memset(rxc, 0, sizeof(*rxc));
1326 	rxc->word4 = cpu_to_le64(RX_COMP4_ZERO);
1327 }
1328 
1329 /* NOTE: we use the ENC RX DESC ring for spares. the rx_page[0,1]
1330  * flipping is protected by the fact that the chip will not
1331  * hand back the same page index while it's being processed.
1332  */
1333 static inline cas_page_t *cas_page_spare(struct cas *cp, const int index)
1334 {
1335 	cas_page_t *page = cp->rx_pages[1][index];
1336 	cas_page_t *new;
1337 
1338 	if (page_count(page->buffer) == 1)
1339 		return page;
1340 
1341 	new = cas_page_dequeue(cp);
1342 	if (new) {
1343 		spin_lock(&cp->rx_inuse_lock);
1344 		list_add(&page->list, &cp->rx_inuse_list);
1345 		spin_unlock(&cp->rx_inuse_lock);
1346 	}
1347 	return new;
1348 }
1349 
1350 /* this needs to be changed if we actually use the ENC RX DESC ring */
1351 static cas_page_t *cas_page_swap(struct cas *cp, const int ring,
1352 				 const int index)
1353 {
1354 	cas_page_t **page0 = cp->rx_pages[0];
1355 	cas_page_t **page1 = cp->rx_pages[1];
1356 
1357 	/* swap if buffer is in use */
1358 	if (page_count(page0[index]->buffer) > 1) {
1359 		cas_page_t *new = cas_page_spare(cp, index);
1360 		if (new) {
1361 			page1[index] = page0[index];
1362 			page0[index] = new;
1363 		}
1364 	}
1365 	RX_USED_SET(page0[index], 0);
1366 	return page0[index];
1367 }
1368 
1369 static void cas_clean_rxds(struct cas *cp)
1370 {
1371 	/* only clean ring 0 as ring 1 is used for spare buffers */
1372         struct cas_rx_desc *rxd = cp->init_rxds[0];
1373 	int i, size;
1374 
1375 	/* release all rx flows */
1376 	for (i = 0; i < N_RX_FLOWS; i++) {
1377 		struct sk_buff *skb;
1378 		while ((skb = __skb_dequeue(&cp->rx_flows[i]))) {
1379 			cas_skb_release(skb);
1380 		}
1381 	}
1382 
1383 	/* initialize descriptors */
1384 	size = RX_DESC_RINGN_SIZE(0);
1385 	for (i = 0; i < size; i++) {
1386 		cas_page_t *page = cas_page_swap(cp, 0, i);
1387 		rxd[i].buffer = cpu_to_le64(page->dma_addr);
1388 		rxd[i].index  = cpu_to_le64(CAS_BASE(RX_INDEX_NUM, i) |
1389 					    CAS_BASE(RX_INDEX_RING, 0));
1390 	}
1391 
1392 	cp->rx_old[0]  = RX_DESC_RINGN_SIZE(0) - 4;
1393 	cp->rx_last[0] = 0;
1394 	cp->cas_flags &= ~CAS_FLAG_RXD_POST(0);
1395 }
1396 
1397 static void cas_clean_rxcs(struct cas *cp)
1398 {
1399 	int i, j;
1400 
1401 	/* take ownership of rx comp descriptors */
1402 	memset(cp->rx_cur, 0, sizeof(*cp->rx_cur)*N_RX_COMP_RINGS);
1403 	memset(cp->rx_new, 0, sizeof(*cp->rx_new)*N_RX_COMP_RINGS);
1404 	for (i = 0; i < N_RX_COMP_RINGS; i++) {
1405 		struct cas_rx_comp *rxc = cp->init_rxcs[i];
1406 		for (j = 0; j < RX_COMP_RINGN_SIZE(i); j++) {
1407 			cas_rxc_init(rxc + j);
1408 		}
1409 	}
1410 }
1411 
1412 #if 0
1413 /* When we get a RX fifo overflow, the RX unit is probably hung
1414  * so we do the following.
1415  *
1416  * If any part of the reset goes wrong, we return 1 and that causes the
1417  * whole chip to be reset.
1418  */
1419 static int cas_rxmac_reset(struct cas *cp)
1420 {
1421 	struct net_device *dev = cp->dev;
1422 	int limit;
1423 	u32 val;
1424 
1425 	/* First, reset MAC RX. */
1426 	writel(cp->mac_rx_cfg & ~MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
1427 	for (limit = 0; limit < STOP_TRIES; limit++) {
1428 		if (!(readl(cp->regs + REG_MAC_RX_CFG) & MAC_RX_CFG_EN))
1429 			break;
1430 		udelay(10);
1431 	}
1432 	if (limit == STOP_TRIES) {
1433 		netdev_err(dev, "RX MAC will not disable, resetting whole chip\n");
1434 		return 1;
1435 	}
1436 
1437 	/* Second, disable RX DMA. */
1438 	writel(0, cp->regs + REG_RX_CFG);
1439 	for (limit = 0; limit < STOP_TRIES; limit++) {
1440 		if (!(readl(cp->regs + REG_RX_CFG) & RX_CFG_DMA_EN))
1441 			break;
1442 		udelay(10);
1443 	}
1444 	if (limit == STOP_TRIES) {
1445 		netdev_err(dev, "RX DMA will not disable, resetting whole chip\n");
1446 		return 1;
1447 	}
1448 
1449 	mdelay(5);
1450 
1451 	/* Execute RX reset command. */
1452 	writel(SW_RESET_RX, cp->regs + REG_SW_RESET);
1453 	for (limit = 0; limit < STOP_TRIES; limit++) {
1454 		if (!(readl(cp->regs + REG_SW_RESET) & SW_RESET_RX))
1455 			break;
1456 		udelay(10);
1457 	}
1458 	if (limit == STOP_TRIES) {
1459 		netdev_err(dev, "RX reset command will not execute, resetting whole chip\n");
1460 		return 1;
1461 	}
1462 
1463 	/* reset driver rx state */
1464 	cas_clean_rxds(cp);
1465 	cas_clean_rxcs(cp);
1466 
1467 	/* Now, reprogram the rest of RX unit. */
1468 	cas_init_rx_dma(cp);
1469 
1470 	/* re-enable */
1471 	val = readl(cp->regs + REG_RX_CFG);
1472 	writel(val | RX_CFG_DMA_EN, cp->regs + REG_RX_CFG);
1473 	writel(MAC_RX_FRAME_RECV, cp->regs + REG_MAC_RX_MASK);
1474 	val = readl(cp->regs + REG_MAC_RX_CFG);
1475 	writel(val | MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
1476 	return 0;
1477 }
1478 #endif
1479 
1480 static int cas_rxmac_interrupt(struct net_device *dev, struct cas *cp,
1481 			       u32 status)
1482 {
1483 	u32 stat = readl(cp->regs + REG_MAC_RX_STATUS);
1484 
1485 	if (!stat)
1486 		return 0;
1487 
1488 	netif_dbg(cp, intr, cp->dev, "rxmac interrupt, stat: 0x%x\n", stat);
1489 
1490 	/* these are all rollovers */
1491 	spin_lock(&cp->stat_lock[0]);
1492 	if (stat & MAC_RX_ALIGN_ERR)
1493 		cp->net_stats[0].rx_frame_errors += 0x10000;
1494 
1495 	if (stat & MAC_RX_CRC_ERR)
1496 		cp->net_stats[0].rx_crc_errors += 0x10000;
1497 
1498 	if (stat & MAC_RX_LEN_ERR)
1499 		cp->net_stats[0].rx_length_errors += 0x10000;
1500 
1501 	if (stat & MAC_RX_OVERFLOW) {
1502 		cp->net_stats[0].rx_over_errors++;
1503 		cp->net_stats[0].rx_fifo_errors++;
1504 	}
1505 
1506 	/* We do not track MAC_RX_FRAME_COUNT and MAC_RX_VIOL_ERR
1507 	 * events.
1508 	 */
1509 	spin_unlock(&cp->stat_lock[0]);
1510 	return 0;
1511 }
1512 
1513 static int cas_mac_interrupt(struct net_device *dev, struct cas *cp,
1514 			     u32 status)
1515 {
1516 	u32 stat = readl(cp->regs + REG_MAC_CTRL_STATUS);
1517 
1518 	if (!stat)
1519 		return 0;
1520 
1521 	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
1522 		     "mac interrupt, stat: 0x%x\n", stat);
1523 
1524 	/* This interrupt is just for pause frame and pause
1525 	 * tracking.  It is useful for diagnostics and debug
1526 	 * but probably by default we will mask these events.
1527 	 */
1528 	if (stat & MAC_CTRL_PAUSE_STATE)
1529 		cp->pause_entered++;
1530 
1531 	if (stat & MAC_CTRL_PAUSE_RECEIVED)
1532 		cp->pause_last_time_recvd = (stat >> 16);
1533 
1534 	return 0;
1535 }
1536 
1537 
1538 /* Must be invoked under cp->lock. */
1539 static inline int cas_mdio_link_not_up(struct cas *cp)
1540 {
1541 	u16 val;
1542 
1543 	switch (cp->lstate) {
1544 	case link_force_ret:
1545 		netif_info(cp, link, cp->dev, "Autoneg failed again, keeping forced mode\n");
1546 		cas_phy_write(cp, MII_BMCR, cp->link_fcntl);
1547 		cp->timer_ticks = 5;
1548 		cp->lstate = link_force_ok;
1549 		cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
1550 		break;
1551 
1552 	case link_aneg:
1553 		val = cas_phy_read(cp, MII_BMCR);
1554 
1555 		/* Try forced modes. we try things in the following order:
1556 		 * 1000 full -> 100 full/half -> 10 half
1557 		 */
1558 		val &= ~(BMCR_ANRESTART | BMCR_ANENABLE);
1559 		val |= BMCR_FULLDPLX;
1560 		val |= (cp->cas_flags & CAS_FLAG_1000MB_CAP) ?
1561 			CAS_BMCR_SPEED1000 : BMCR_SPEED100;
1562 		cas_phy_write(cp, MII_BMCR, val);
1563 		cp->timer_ticks = 5;
1564 		cp->lstate = link_force_try;
1565 		cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
1566 		break;
1567 
1568 	case link_force_try:
1569 		/* Downgrade from 1000 to 100 to 10 Mbps if necessary. */
1570 		val = cas_phy_read(cp, MII_BMCR);
1571 		cp->timer_ticks = 5;
1572 		if (val & CAS_BMCR_SPEED1000) { /* gigabit */
1573 			val &= ~CAS_BMCR_SPEED1000;
1574 			val |= (BMCR_SPEED100 | BMCR_FULLDPLX);
1575 			cas_phy_write(cp, MII_BMCR, val);
1576 			break;
1577 		}
1578 
1579 		if (val & BMCR_SPEED100) {
1580 			if (val & BMCR_FULLDPLX) /* fd failed */
1581 				val &= ~BMCR_FULLDPLX;
1582 			else { /* 100Mbps failed */
1583 				val &= ~BMCR_SPEED100;
1584 			}
1585 			cas_phy_write(cp, MII_BMCR, val);
1586 			break;
1587 		}
1588 		break;
1589 	default:
1590 		break;
1591 	}
1592 	return 0;
1593 }
1594 
1595 
1596 /* must be invoked with cp->lock held */
1597 static int cas_mii_link_check(struct cas *cp, const u16 bmsr)
1598 {
1599 	int restart;
1600 
1601 	if (bmsr & BMSR_LSTATUS) {
1602 		/* Ok, here we got a link. If we had it due to a forced
1603 		 * fallback, and we were configured for autoneg, we
1604 		 * retry a short autoneg pass. If you know your hub is
1605 		 * broken, use ethtool ;)
1606 		 */
1607 		if ((cp->lstate == link_force_try) &&
1608 		    (cp->link_cntl & BMCR_ANENABLE)) {
1609 			cp->lstate = link_force_ret;
1610 			cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
1611 			cas_mif_poll(cp, 0);
1612 			cp->link_fcntl = cas_phy_read(cp, MII_BMCR);
1613 			cp->timer_ticks = 5;
1614 			if (cp->opened)
1615 				netif_info(cp, link, cp->dev,
1616 					   "Got link after fallback, retrying autoneg once...\n");
1617 			cas_phy_write(cp, MII_BMCR,
1618 				      cp->link_fcntl | BMCR_ANENABLE |
1619 				      BMCR_ANRESTART);
1620 			cas_mif_poll(cp, 1);
1621 
1622 		} else if (cp->lstate != link_up) {
1623 			cp->lstate = link_up;
1624 			cp->link_transition = LINK_TRANSITION_LINK_UP;
1625 
1626 			if (cp->opened) {
1627 				cas_set_link_modes(cp);
1628 				netif_carrier_on(cp->dev);
1629 			}
1630 		}
1631 		return 0;
1632 	}
1633 
1634 	/* link not up. if the link was previously up, we restart the
1635 	 * whole process
1636 	 */
1637 	restart = 0;
1638 	if (cp->lstate == link_up) {
1639 		cp->lstate = link_down;
1640 		cp->link_transition = LINK_TRANSITION_LINK_DOWN;
1641 
1642 		netif_carrier_off(cp->dev);
1643 		if (cp->opened)
1644 			netif_info(cp, link, cp->dev, "Link down\n");
1645 		restart = 1;
1646 
1647 	} else if (++cp->timer_ticks > 10)
1648 		cas_mdio_link_not_up(cp);
1649 
1650 	return restart;
1651 }
1652 
1653 static int cas_mif_interrupt(struct net_device *dev, struct cas *cp,
1654 			     u32 status)
1655 {
1656 	u32 stat = readl(cp->regs + REG_MIF_STATUS);
1657 	u16 bmsr;
1658 
1659 	/* check for a link change */
1660 	if (CAS_VAL(MIF_STATUS_POLL_STATUS, stat) == 0)
1661 		return 0;
1662 
1663 	bmsr = CAS_VAL(MIF_STATUS_POLL_DATA, stat);
1664 	return cas_mii_link_check(cp, bmsr);
1665 }
1666 
1667 static int cas_pci_interrupt(struct net_device *dev, struct cas *cp,
1668 			     u32 status)
1669 {
1670 	u32 stat = readl(cp->regs + REG_PCI_ERR_STATUS);
1671 
1672 	if (!stat)
1673 		return 0;
1674 
1675 	netdev_err(dev, "PCI error [%04x:%04x]",
1676 		   stat, readl(cp->regs + REG_BIM_DIAG));
1677 
1678 	/* cassini+ has this reserved */
1679 	if ((stat & PCI_ERR_BADACK) &&
1680 	    ((cp->cas_flags & CAS_FLAG_REG_PLUS) == 0))
1681 		pr_cont(" <No ACK64# during ABS64 cycle>");
1682 
1683 	if (stat & PCI_ERR_DTRTO)
1684 		pr_cont(" <Delayed transaction timeout>");
1685 	if (stat & PCI_ERR_OTHER)
1686 		pr_cont(" <other>");
1687 	if (stat & PCI_ERR_BIM_DMA_WRITE)
1688 		pr_cont(" <BIM DMA 0 write req>");
1689 	if (stat & PCI_ERR_BIM_DMA_READ)
1690 		pr_cont(" <BIM DMA 0 read req>");
1691 	pr_cont("\n");
1692 
1693 	if (stat & PCI_ERR_OTHER) {
1694 		int pci_errs;
1695 
1696 		/* Interrogate PCI config space for the
1697 		 * true cause.
1698 		 */
1699 		pci_errs = pci_status_get_and_clear_errors(cp->pdev);
1700 
1701 		netdev_err(dev, "PCI status errors[%04x]\n", pci_errs);
1702 		if (pci_errs & PCI_STATUS_PARITY)
1703 			netdev_err(dev, "PCI parity error detected\n");
1704 		if (pci_errs & PCI_STATUS_SIG_TARGET_ABORT)
1705 			netdev_err(dev, "PCI target abort\n");
1706 		if (pci_errs & PCI_STATUS_REC_TARGET_ABORT)
1707 			netdev_err(dev, "PCI master acks target abort\n");
1708 		if (pci_errs & PCI_STATUS_REC_MASTER_ABORT)
1709 			netdev_err(dev, "PCI master abort\n");
1710 		if (pci_errs & PCI_STATUS_SIG_SYSTEM_ERROR)
1711 			netdev_err(dev, "PCI system error SERR#\n");
1712 		if (pci_errs & PCI_STATUS_DETECTED_PARITY)
1713 			netdev_err(dev, "PCI parity error\n");
1714 	}
1715 
1716 	/* For all PCI errors, we should reset the chip. */
1717 	return 1;
1718 }
1719 
1720 /* All non-normal interrupt conditions get serviced here.
1721  * Returns non-zero if we should just exit the interrupt
1722  * handler right now (ie. if we reset the card which invalidates
1723  * all of the other original irq status bits).
1724  */
1725 static int cas_abnormal_irq(struct net_device *dev, struct cas *cp,
1726 			    u32 status)
1727 {
1728 	if (status & INTR_RX_TAG_ERROR) {
1729 		/* corrupt RX tag framing */
1730 		netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
1731 			     "corrupt rx tag framing\n");
1732 		spin_lock(&cp->stat_lock[0]);
1733 		cp->net_stats[0].rx_errors++;
1734 		spin_unlock(&cp->stat_lock[0]);
1735 		goto do_reset;
1736 	}
1737 
1738 	if (status & INTR_RX_LEN_MISMATCH) {
1739 		/* length mismatch. */
1740 		netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
1741 			     "length mismatch for rx frame\n");
1742 		spin_lock(&cp->stat_lock[0]);
1743 		cp->net_stats[0].rx_errors++;
1744 		spin_unlock(&cp->stat_lock[0]);
1745 		goto do_reset;
1746 	}
1747 
1748 	if (status & INTR_PCS_STATUS) {
1749 		if (cas_pcs_interrupt(dev, cp, status))
1750 			goto do_reset;
1751 	}
1752 
1753 	if (status & INTR_TX_MAC_STATUS) {
1754 		if (cas_txmac_interrupt(dev, cp, status))
1755 			goto do_reset;
1756 	}
1757 
1758 	if (status & INTR_RX_MAC_STATUS) {
1759 		if (cas_rxmac_interrupt(dev, cp, status))
1760 			goto do_reset;
1761 	}
1762 
1763 	if (status & INTR_MAC_CTRL_STATUS) {
1764 		if (cas_mac_interrupt(dev, cp, status))
1765 			goto do_reset;
1766 	}
1767 
1768 	if (status & INTR_MIF_STATUS) {
1769 		if (cas_mif_interrupt(dev, cp, status))
1770 			goto do_reset;
1771 	}
1772 
1773 	if (status & INTR_PCI_ERROR_STATUS) {
1774 		if (cas_pci_interrupt(dev, cp, status))
1775 			goto do_reset;
1776 	}
1777 	return 0;
1778 
1779 do_reset:
1780 #if 1
1781 	atomic_inc(&cp->reset_task_pending);
1782 	atomic_inc(&cp->reset_task_pending_all);
1783 	netdev_err(dev, "reset called in cas_abnormal_irq [0x%x]\n", status);
1784 	schedule_work(&cp->reset_task);
1785 #else
1786 	atomic_set(&cp->reset_task_pending, CAS_RESET_ALL);
1787 	netdev_err(dev, "reset called in cas_abnormal_irq\n");
1788 	schedule_work(&cp->reset_task);
1789 #endif
1790 	return 1;
1791 }
1792 
1793 /* NOTE: CAS_TABORT returns 1 or 2 so that it can be used when
1794  *       determining whether to do a netif_stop/wakeup
1795  */
1796 #define CAS_TABORT(x)      (((x)->cas_flags & CAS_FLAG_TARGET_ABORT) ? 2 : 1)
1797 #define CAS_ROUND_PAGE(x)  (((x) + PAGE_SIZE - 1) & PAGE_MASK)
1798 static inline int cas_calc_tabort(struct cas *cp, const unsigned long addr,
1799 				  const int len)
1800 {
1801 	unsigned long off = addr + len;
1802 
1803 	if (CAS_TABORT(cp) == 1)
1804 		return 0;
1805 	if ((CAS_ROUND_PAGE(off) - off) > TX_TARGET_ABORT_LEN)
1806 		return 0;
1807 	return TX_TARGET_ABORT_LEN;
1808 }
1809 
1810 static inline void cas_tx_ringN(struct cas *cp, int ring, int limit)
1811 {
1812 	struct cas_tx_desc *txds;
1813 	struct sk_buff **skbs;
1814 	struct net_device *dev = cp->dev;
1815 	int entry, count;
1816 
1817 	spin_lock(&cp->tx_lock[ring]);
1818 	txds = cp->init_txds[ring];
1819 	skbs = cp->tx_skbs[ring];
1820 	entry = cp->tx_old[ring];
1821 
1822 	count = TX_BUFF_COUNT(ring, entry, limit);
1823 	while (entry != limit) {
1824 		struct sk_buff *skb = skbs[entry];
1825 		dma_addr_t daddr;
1826 		u32 dlen;
1827 		int frag;
1828 
1829 		if (!skb) {
1830 			/* this should never occur */
1831 			entry = TX_DESC_NEXT(ring, entry);
1832 			continue;
1833 		}
1834 
1835 		/* however, we might get only a partial skb release. */
1836 		count -= skb_shinfo(skb)->nr_frags +
1837 			+ cp->tx_tiny_use[ring][entry].nbufs + 1;
1838 		if (count < 0)
1839 			break;
1840 
1841 		netif_printk(cp, tx_done, KERN_DEBUG, cp->dev,
1842 			     "tx[%d] done, slot %d\n", ring, entry);
1843 
1844 		skbs[entry] = NULL;
1845 		cp->tx_tiny_use[ring][entry].nbufs = 0;
1846 
1847 		for (frag = 0; frag <= skb_shinfo(skb)->nr_frags; frag++) {
1848 			struct cas_tx_desc *txd = txds + entry;
1849 
1850 			daddr = le64_to_cpu(txd->buffer);
1851 			dlen = CAS_VAL(TX_DESC_BUFLEN,
1852 				       le64_to_cpu(txd->control));
1853 			dma_unmap_page(&cp->pdev->dev, daddr, dlen,
1854 				       DMA_TO_DEVICE);
1855 			entry = TX_DESC_NEXT(ring, entry);
1856 
1857 			/* tiny buffer may follow */
1858 			if (cp->tx_tiny_use[ring][entry].used) {
1859 				cp->tx_tiny_use[ring][entry].used = 0;
1860 				entry = TX_DESC_NEXT(ring, entry);
1861 			}
1862 		}
1863 
1864 		spin_lock(&cp->stat_lock[ring]);
1865 		cp->net_stats[ring].tx_packets++;
1866 		cp->net_stats[ring].tx_bytes += skb->len;
1867 		spin_unlock(&cp->stat_lock[ring]);
1868 		dev_consume_skb_irq(skb);
1869 	}
1870 	cp->tx_old[ring] = entry;
1871 
1872 	/* this is wrong for multiple tx rings. the net device needs
1873 	 * multiple queues for this to do the right thing.  we wait
1874 	 * for 2*packets to be available when using tiny buffers
1875 	 */
1876 	if (netif_queue_stopped(dev) &&
1877 	    (TX_BUFFS_AVAIL(cp, ring) > CAS_TABORT(cp)*(MAX_SKB_FRAGS + 1)))
1878 		netif_wake_queue(dev);
1879 	spin_unlock(&cp->tx_lock[ring]);
1880 }
1881 
1882 static void cas_tx(struct net_device *dev, struct cas *cp,
1883 		   u32 status)
1884 {
1885         int limit, ring;
1886 #ifdef USE_TX_COMPWB
1887 	u64 compwb = le64_to_cpu(cp->init_block->tx_compwb);
1888 #endif
1889 	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
1890 		     "tx interrupt, status: 0x%x, %llx\n",
1891 		     status, (unsigned long long)compwb);
1892 	/* process all the rings */
1893 	for (ring = 0; ring < N_TX_RINGS; ring++) {
1894 #ifdef USE_TX_COMPWB
1895 		/* use the completion writeback registers */
1896 		limit = (CAS_VAL(TX_COMPWB_MSB, compwb) << 8) |
1897 			CAS_VAL(TX_COMPWB_LSB, compwb);
1898 		compwb = TX_COMPWB_NEXT(compwb);
1899 #else
1900 		limit = readl(cp->regs + REG_TX_COMPN(ring));
1901 #endif
1902 		if (cp->tx_old[ring] != limit)
1903 			cas_tx_ringN(cp, ring, limit);
1904 	}
1905 }
1906 
1907 
1908 static int cas_rx_process_pkt(struct cas *cp, struct cas_rx_comp *rxc,
1909 			      int entry, const u64 *words,
1910 			      struct sk_buff **skbref)
1911 {
1912 	int dlen, hlen, len, i, alloclen;
1913 	int off, swivel = RX_SWIVEL_OFF_VAL;
1914 	struct cas_page *page;
1915 	struct sk_buff *skb;
1916 	void *crcaddr;
1917 	__sum16 csum;
1918 	char *p;
1919 
1920 	hlen = CAS_VAL(RX_COMP2_HDR_SIZE, words[1]);
1921 	dlen = CAS_VAL(RX_COMP1_DATA_SIZE, words[0]);
1922 	len  = hlen + dlen;
1923 
1924 	if (RX_COPY_ALWAYS || (words[2] & RX_COMP3_SMALL_PKT))
1925 		alloclen = len;
1926 	else
1927 		alloclen = max(hlen, RX_COPY_MIN);
1928 
1929 	skb = netdev_alloc_skb(cp->dev, alloclen + swivel + cp->crc_size);
1930 	if (skb == NULL)
1931 		return -1;
1932 
1933 	*skbref = skb;
1934 	skb_reserve(skb, swivel);
1935 
1936 	p = skb->data;
1937 	crcaddr = NULL;
1938 	if (hlen) { /* always copy header pages */
1939 		i = CAS_VAL(RX_COMP2_HDR_INDEX, words[1]);
1940 		page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
1941 		off = CAS_VAL(RX_COMP2_HDR_OFF, words[1]) * 0x100 +
1942 			swivel;
1943 
1944 		i = hlen;
1945 		if (!dlen) /* attach FCS */
1946 			i += cp->crc_size;
1947 		dma_sync_single_for_cpu(&cp->pdev->dev, page->dma_addr + off,
1948 					i, DMA_FROM_DEVICE);
1949 		memcpy(p, page_address(page->buffer) + off, i);
1950 		dma_sync_single_for_device(&cp->pdev->dev,
1951 					   page->dma_addr + off, i,
1952 					   DMA_FROM_DEVICE);
1953 		RX_USED_ADD(page, 0x100);
1954 		p += hlen;
1955 		swivel = 0;
1956 	}
1957 
1958 
1959 	if (alloclen < (hlen + dlen)) {
1960 		skb_frag_t *frag = skb_shinfo(skb)->frags;
1961 
1962 		/* normal or jumbo packets. we use frags */
1963 		i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
1964 		page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
1965 		off = CAS_VAL(RX_COMP1_DATA_OFF, words[0]) + swivel;
1966 
1967 		hlen = min(cp->page_size - off, dlen);
1968 		if (hlen < 0) {
1969 			netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
1970 				     "rx page overflow: %d\n", hlen);
1971 			dev_kfree_skb_irq(skb);
1972 			return -1;
1973 		}
1974 		i = hlen;
1975 		if (i == dlen)  /* attach FCS */
1976 			i += cp->crc_size;
1977 		dma_sync_single_for_cpu(&cp->pdev->dev, page->dma_addr + off,
1978 					i, DMA_FROM_DEVICE);
1979 
1980 		/* make sure we always copy a header */
1981 		swivel = 0;
1982 		if (p == (char *) skb->data) { /* not split */
1983 			memcpy(p, page_address(page->buffer) + off,
1984 			       RX_COPY_MIN);
1985 			dma_sync_single_for_device(&cp->pdev->dev,
1986 						   page->dma_addr + off, i,
1987 						   DMA_FROM_DEVICE);
1988 			off += RX_COPY_MIN;
1989 			swivel = RX_COPY_MIN;
1990 			RX_USED_ADD(page, cp->mtu_stride);
1991 		} else {
1992 			RX_USED_ADD(page, hlen);
1993 		}
1994 		skb_put(skb, alloclen);
1995 
1996 		skb_shinfo(skb)->nr_frags++;
1997 		skb->data_len += hlen - swivel;
1998 		skb->truesize += hlen - swivel;
1999 		skb->len      += hlen - swivel;
2000 
2001 		__skb_frag_set_page(frag, page->buffer);
2002 		__skb_frag_ref(frag);
2003 		skb_frag_off_set(frag, off);
2004 		skb_frag_size_set(frag, hlen - swivel);
2005 
2006 		/* any more data? */
2007 		if ((words[0] & RX_COMP1_SPLIT_PKT) && ((dlen -= hlen) > 0)) {
2008 			hlen = dlen;
2009 			off = 0;
2010 
2011 			i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
2012 			page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
2013 			dma_sync_single_for_cpu(&cp->pdev->dev,
2014 						page->dma_addr,
2015 						hlen + cp->crc_size,
2016 						DMA_FROM_DEVICE);
2017 			dma_sync_single_for_device(&cp->pdev->dev,
2018 						   page->dma_addr,
2019 						   hlen + cp->crc_size,
2020 						   DMA_FROM_DEVICE);
2021 
2022 			skb_shinfo(skb)->nr_frags++;
2023 			skb->data_len += hlen;
2024 			skb->len      += hlen;
2025 			frag++;
2026 
2027 			__skb_frag_set_page(frag, page->buffer);
2028 			__skb_frag_ref(frag);
2029 			skb_frag_off_set(frag, 0);
2030 			skb_frag_size_set(frag, hlen);
2031 			RX_USED_ADD(page, hlen + cp->crc_size);
2032 		}
2033 
2034 		if (cp->crc_size)
2035 			crcaddr = page_address(page->buffer) + off + hlen;
2036 
2037 	} else {
2038 		/* copying packet */
2039 		if (!dlen)
2040 			goto end_copy_pkt;
2041 
2042 		i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
2043 		page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
2044 		off = CAS_VAL(RX_COMP1_DATA_OFF, words[0]) + swivel;
2045 		hlen = min(cp->page_size - off, dlen);
2046 		if (hlen < 0) {
2047 			netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
2048 				     "rx page overflow: %d\n", hlen);
2049 			dev_kfree_skb_irq(skb);
2050 			return -1;
2051 		}
2052 		i = hlen;
2053 		if (i == dlen) /* attach FCS */
2054 			i += cp->crc_size;
2055 		dma_sync_single_for_cpu(&cp->pdev->dev, page->dma_addr + off,
2056 					i, DMA_FROM_DEVICE);
2057 		memcpy(p, page_address(page->buffer) + off, i);
2058 		dma_sync_single_for_device(&cp->pdev->dev,
2059 					   page->dma_addr + off, i,
2060 					   DMA_FROM_DEVICE);
2061 		if (p == (char *) skb->data) /* not split */
2062 			RX_USED_ADD(page, cp->mtu_stride);
2063 		else
2064 			RX_USED_ADD(page, i);
2065 
2066 		/* any more data? */
2067 		if ((words[0] & RX_COMP1_SPLIT_PKT) && ((dlen -= hlen) > 0)) {
2068 			p += hlen;
2069 			i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
2070 			page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
2071 			dma_sync_single_for_cpu(&cp->pdev->dev,
2072 						page->dma_addr,
2073 						dlen + cp->crc_size,
2074 						DMA_FROM_DEVICE);
2075 			memcpy(p, page_address(page->buffer), dlen + cp->crc_size);
2076 			dma_sync_single_for_device(&cp->pdev->dev,
2077 						   page->dma_addr,
2078 						   dlen + cp->crc_size,
2079 						   DMA_FROM_DEVICE);
2080 			RX_USED_ADD(page, dlen + cp->crc_size);
2081 		}
2082 end_copy_pkt:
2083 		if (cp->crc_size)
2084 			crcaddr = skb->data + alloclen;
2085 
2086 		skb_put(skb, alloclen);
2087 	}
2088 
2089 	csum = (__force __sum16)htons(CAS_VAL(RX_COMP4_TCP_CSUM, words[3]));
2090 	if (cp->crc_size) {
2091 		/* checksum includes FCS. strip it out. */
2092 		csum = csum_fold(csum_partial(crcaddr, cp->crc_size,
2093 					      csum_unfold(csum)));
2094 	}
2095 	skb->protocol = eth_type_trans(skb, cp->dev);
2096 	if (skb->protocol == htons(ETH_P_IP)) {
2097 		skb->csum = csum_unfold(~csum);
2098 		skb->ip_summed = CHECKSUM_COMPLETE;
2099 	} else
2100 		skb_checksum_none_assert(skb);
2101 	return len;
2102 }
2103 
2104 
2105 /* we can handle up to 64 rx flows at a time. we do the same thing
2106  * as nonreassm except that we batch up the buffers.
2107  * NOTE: we currently just treat each flow as a bunch of packets that
2108  *       we pass up. a better way would be to coalesce the packets
2109  *       into a jumbo packet. to do that, we need to do the following:
2110  *       1) the first packet will have a clean split between header and
2111  *          data. save both.
2112  *       2) each time the next flow packet comes in, extend the
2113  *          data length and merge the checksums.
2114  *       3) on flow release, fix up the header.
2115  *       4) make sure the higher layer doesn't care.
2116  * because packets get coalesced, we shouldn't run into fragment count
2117  * issues.
2118  */
2119 static inline void cas_rx_flow_pkt(struct cas *cp, const u64 *words,
2120 				   struct sk_buff *skb)
2121 {
2122 	int flowid = CAS_VAL(RX_COMP3_FLOWID, words[2]) & (N_RX_FLOWS - 1);
2123 	struct sk_buff_head *flow = &cp->rx_flows[flowid];
2124 
2125 	/* this is protected at a higher layer, so no need to
2126 	 * do any additional locking here. stick the buffer
2127 	 * at the end.
2128 	 */
2129 	__skb_queue_tail(flow, skb);
2130 	if (words[0] & RX_COMP1_RELEASE_FLOW) {
2131 		while ((skb = __skb_dequeue(flow))) {
2132 			cas_skb_release(skb);
2133 		}
2134 	}
2135 }
2136 
2137 /* put rx descriptor back on ring. if a buffer is in use by a higher
2138  * layer, this will need to put in a replacement.
2139  */
2140 static void cas_post_page(struct cas *cp, const int ring, const int index)
2141 {
2142 	cas_page_t *new;
2143 	int entry;
2144 
2145 	entry = cp->rx_old[ring];
2146 
2147 	new = cas_page_swap(cp, ring, index);
2148 	cp->init_rxds[ring][entry].buffer = cpu_to_le64(new->dma_addr);
2149 	cp->init_rxds[ring][entry].index  =
2150 		cpu_to_le64(CAS_BASE(RX_INDEX_NUM, index) |
2151 			    CAS_BASE(RX_INDEX_RING, ring));
2152 
2153 	entry = RX_DESC_ENTRY(ring, entry + 1);
2154 	cp->rx_old[ring] = entry;
2155 
2156 	if (entry % 4)
2157 		return;
2158 
2159 	if (ring == 0)
2160 		writel(entry, cp->regs + REG_RX_KICK);
2161 	else if ((N_RX_DESC_RINGS > 1) &&
2162 		 (cp->cas_flags & CAS_FLAG_REG_PLUS))
2163 		writel(entry, cp->regs + REG_PLUS_RX_KICK1);
2164 }
2165 
2166 
2167 /* only when things are bad */
2168 static int cas_post_rxds_ringN(struct cas *cp, int ring, int num)
2169 {
2170 	unsigned int entry, last, count, released;
2171 	int cluster;
2172 	cas_page_t **page = cp->rx_pages[ring];
2173 
2174 	entry = cp->rx_old[ring];
2175 
2176 	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
2177 		     "rxd[%d] interrupt, done: %d\n", ring, entry);
2178 
2179 	cluster = -1;
2180 	count = entry & 0x3;
2181 	last = RX_DESC_ENTRY(ring, num ? entry + num - 4: entry - 4);
2182 	released = 0;
2183 	while (entry != last) {
2184 		/* make a new buffer if it's still in use */
2185 		if (page_count(page[entry]->buffer) > 1) {
2186 			cas_page_t *new = cas_page_dequeue(cp);
2187 			if (!new) {
2188 				/* let the timer know that we need to
2189 				 * do this again
2190 				 */
2191 				cp->cas_flags |= CAS_FLAG_RXD_POST(ring);
2192 				if (!timer_pending(&cp->link_timer))
2193 					mod_timer(&cp->link_timer, jiffies +
2194 						  CAS_LINK_FAST_TIMEOUT);
2195 				cp->rx_old[ring]  = entry;
2196 				cp->rx_last[ring] = num ? num - released : 0;
2197 				return -ENOMEM;
2198 			}
2199 			spin_lock(&cp->rx_inuse_lock);
2200 			list_add(&page[entry]->list, &cp->rx_inuse_list);
2201 			spin_unlock(&cp->rx_inuse_lock);
2202 			cp->init_rxds[ring][entry].buffer =
2203 				cpu_to_le64(new->dma_addr);
2204 			page[entry] = new;
2205 
2206 		}
2207 
2208 		if (++count == 4) {
2209 			cluster = entry;
2210 			count = 0;
2211 		}
2212 		released++;
2213 		entry = RX_DESC_ENTRY(ring, entry + 1);
2214 	}
2215 	cp->rx_old[ring] = entry;
2216 
2217 	if (cluster < 0)
2218 		return 0;
2219 
2220 	if (ring == 0)
2221 		writel(cluster, cp->regs + REG_RX_KICK);
2222 	else if ((N_RX_DESC_RINGS > 1) &&
2223 		 (cp->cas_flags & CAS_FLAG_REG_PLUS))
2224 		writel(cluster, cp->regs + REG_PLUS_RX_KICK1);
2225 	return 0;
2226 }
2227 
2228 
2229 /* process a completion ring. packets are set up in three basic ways:
2230  * small packets: should be copied header + data in single buffer.
2231  * large packets: header and data in a single buffer.
2232  * split packets: header in a separate buffer from data.
2233  *                data may be in multiple pages. data may be > 256
2234  *                bytes but in a single page.
2235  *
2236  * NOTE: RX page posting is done in this routine as well. while there's
2237  *       the capability of using multiple RX completion rings, it isn't
2238  *       really worthwhile due to the fact that the page posting will
2239  *       force serialization on the single descriptor ring.
2240  */
2241 static int cas_rx_ringN(struct cas *cp, int ring, int budget)
2242 {
2243 	struct cas_rx_comp *rxcs = cp->init_rxcs[ring];
2244 	int entry, drops;
2245 	int npackets = 0;
2246 
2247 	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
2248 		     "rx[%d] interrupt, done: %d/%d\n",
2249 		     ring,
2250 		     readl(cp->regs + REG_RX_COMP_HEAD), cp->rx_new[ring]);
2251 
2252 	entry = cp->rx_new[ring];
2253 	drops = 0;
2254 	while (1) {
2255 		struct cas_rx_comp *rxc = rxcs + entry;
2256 		struct sk_buff *skb;
2257 		int type, len;
2258 		u64 words[4];
2259 		int i, dring;
2260 
2261 		words[0] = le64_to_cpu(rxc->word1);
2262 		words[1] = le64_to_cpu(rxc->word2);
2263 		words[2] = le64_to_cpu(rxc->word3);
2264 		words[3] = le64_to_cpu(rxc->word4);
2265 
2266 		/* don't touch if still owned by hw */
2267 		type = CAS_VAL(RX_COMP1_TYPE, words[0]);
2268 		if (type == 0)
2269 			break;
2270 
2271 		/* hw hasn't cleared the zero bit yet */
2272 		if (words[3] & RX_COMP4_ZERO) {
2273 			break;
2274 		}
2275 
2276 		/* get info on the packet */
2277 		if (words[3] & (RX_COMP4_LEN_MISMATCH | RX_COMP4_BAD)) {
2278 			spin_lock(&cp->stat_lock[ring]);
2279 			cp->net_stats[ring].rx_errors++;
2280 			if (words[3] & RX_COMP4_LEN_MISMATCH)
2281 				cp->net_stats[ring].rx_length_errors++;
2282 			if (words[3] & RX_COMP4_BAD)
2283 				cp->net_stats[ring].rx_crc_errors++;
2284 			spin_unlock(&cp->stat_lock[ring]);
2285 
2286 			/* We'll just return it to Cassini. */
2287 		drop_it:
2288 			spin_lock(&cp->stat_lock[ring]);
2289 			++cp->net_stats[ring].rx_dropped;
2290 			spin_unlock(&cp->stat_lock[ring]);
2291 			goto next;
2292 		}
2293 
2294 		len = cas_rx_process_pkt(cp, rxc, entry, words, &skb);
2295 		if (len < 0) {
2296 			++drops;
2297 			goto drop_it;
2298 		}
2299 
2300 		/* see if it's a flow re-assembly or not. the driver
2301 		 * itself handles release back up.
2302 		 */
2303 		if (RX_DONT_BATCH || (type == 0x2)) {
2304 			/* non-reassm: these always get released */
2305 			cas_skb_release(skb);
2306 		} else {
2307 			cas_rx_flow_pkt(cp, words, skb);
2308 		}
2309 
2310 		spin_lock(&cp->stat_lock[ring]);
2311 		cp->net_stats[ring].rx_packets++;
2312 		cp->net_stats[ring].rx_bytes += len;
2313 		spin_unlock(&cp->stat_lock[ring]);
2314 
2315 	next:
2316 		npackets++;
2317 
2318 		/* should it be released? */
2319 		if (words[0] & RX_COMP1_RELEASE_HDR) {
2320 			i = CAS_VAL(RX_COMP2_HDR_INDEX, words[1]);
2321 			dring = CAS_VAL(RX_INDEX_RING, i);
2322 			i = CAS_VAL(RX_INDEX_NUM, i);
2323 			cas_post_page(cp, dring, i);
2324 		}
2325 
2326 		if (words[0] & RX_COMP1_RELEASE_DATA) {
2327 			i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
2328 			dring = CAS_VAL(RX_INDEX_RING, i);
2329 			i = CAS_VAL(RX_INDEX_NUM, i);
2330 			cas_post_page(cp, dring, i);
2331 		}
2332 
2333 		if (words[0] & RX_COMP1_RELEASE_NEXT) {
2334 			i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
2335 			dring = CAS_VAL(RX_INDEX_RING, i);
2336 			i = CAS_VAL(RX_INDEX_NUM, i);
2337 			cas_post_page(cp, dring, i);
2338 		}
2339 
2340 		/* skip to the next entry */
2341 		entry = RX_COMP_ENTRY(ring, entry + 1 +
2342 				      CAS_VAL(RX_COMP1_SKIP, words[0]));
2343 #ifdef USE_NAPI
2344 		if (budget && (npackets >= budget))
2345 			break;
2346 #endif
2347 	}
2348 	cp->rx_new[ring] = entry;
2349 
2350 	if (drops)
2351 		netdev_info(cp->dev, "Memory squeeze, deferring packet\n");
2352 	return npackets;
2353 }
2354 
2355 
2356 /* put completion entries back on the ring */
2357 static void cas_post_rxcs_ringN(struct net_device *dev,
2358 				struct cas *cp, int ring)
2359 {
2360 	struct cas_rx_comp *rxc = cp->init_rxcs[ring];
2361 	int last, entry;
2362 
2363 	last = cp->rx_cur[ring];
2364 	entry = cp->rx_new[ring];
2365 	netif_printk(cp, intr, KERN_DEBUG, dev,
2366 		     "rxc[%d] interrupt, done: %d/%d\n",
2367 		     ring, readl(cp->regs + REG_RX_COMP_HEAD), entry);
2368 
2369 	/* zero and re-mark descriptors */
2370 	while (last != entry) {
2371 		cas_rxc_init(rxc + last);
2372 		last = RX_COMP_ENTRY(ring, last + 1);
2373 	}
2374 	cp->rx_cur[ring] = last;
2375 
2376 	if (ring == 0)
2377 		writel(last, cp->regs + REG_RX_COMP_TAIL);
2378 	else if (cp->cas_flags & CAS_FLAG_REG_PLUS)
2379 		writel(last, cp->regs + REG_PLUS_RX_COMPN_TAIL(ring));
2380 }
2381 
2382 
2383 
2384 /* cassini can use all four PCI interrupts for the completion ring.
2385  * rings 3 and 4 are identical
2386  */
2387 #if defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
2388 static inline void cas_handle_irqN(struct net_device *dev,
2389 				   struct cas *cp, const u32 status,
2390 				   const int ring)
2391 {
2392 	if (status & (INTR_RX_COMP_FULL_ALT | INTR_RX_COMP_AF_ALT))
2393 		cas_post_rxcs_ringN(dev, cp, ring);
2394 }
2395 
2396 static irqreturn_t cas_interruptN(int irq, void *dev_id)
2397 {
2398 	struct net_device *dev = dev_id;
2399 	struct cas *cp = netdev_priv(dev);
2400 	unsigned long flags;
2401 	int ring = (irq == cp->pci_irq_INTC) ? 2 : 3;
2402 	u32 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(ring));
2403 
2404 	/* check for shared irq */
2405 	if (status == 0)
2406 		return IRQ_NONE;
2407 
2408 	spin_lock_irqsave(&cp->lock, flags);
2409 	if (status & INTR_RX_DONE_ALT) { /* handle rx separately */
2410 #ifdef USE_NAPI
2411 		cas_mask_intr(cp);
2412 		napi_schedule(&cp->napi);
2413 #else
2414 		cas_rx_ringN(cp, ring, 0);
2415 #endif
2416 		status &= ~INTR_RX_DONE_ALT;
2417 	}
2418 
2419 	if (status)
2420 		cas_handle_irqN(dev, cp, status, ring);
2421 	spin_unlock_irqrestore(&cp->lock, flags);
2422 	return IRQ_HANDLED;
2423 }
2424 #endif
2425 
2426 #ifdef USE_PCI_INTB
2427 /* everything but rx packets */
2428 static inline void cas_handle_irq1(struct cas *cp, const u32 status)
2429 {
2430 	if (status & INTR_RX_BUF_UNAVAIL_1) {
2431 		/* Frame arrived, no free RX buffers available.
2432 		 * NOTE: we can get this on a link transition. */
2433 		cas_post_rxds_ringN(cp, 1, 0);
2434 		spin_lock(&cp->stat_lock[1]);
2435 		cp->net_stats[1].rx_dropped++;
2436 		spin_unlock(&cp->stat_lock[1]);
2437 	}
2438 
2439 	if (status & INTR_RX_BUF_AE_1)
2440 		cas_post_rxds_ringN(cp, 1, RX_DESC_RINGN_SIZE(1) -
2441 				    RX_AE_FREEN_VAL(1));
2442 
2443 	if (status & (INTR_RX_COMP_AF | INTR_RX_COMP_FULL))
2444 		cas_post_rxcs_ringN(cp, 1);
2445 }
2446 
2447 /* ring 2 handles a few more events than 3 and 4 */
2448 static irqreturn_t cas_interrupt1(int irq, void *dev_id)
2449 {
2450 	struct net_device *dev = dev_id;
2451 	struct cas *cp = netdev_priv(dev);
2452 	unsigned long flags;
2453 	u32 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(1));
2454 
2455 	/* check for shared interrupt */
2456 	if (status == 0)
2457 		return IRQ_NONE;
2458 
2459 	spin_lock_irqsave(&cp->lock, flags);
2460 	if (status & INTR_RX_DONE_ALT) { /* handle rx separately */
2461 #ifdef USE_NAPI
2462 		cas_mask_intr(cp);
2463 		napi_schedule(&cp->napi);
2464 #else
2465 		cas_rx_ringN(cp, 1, 0);
2466 #endif
2467 		status &= ~INTR_RX_DONE_ALT;
2468 	}
2469 	if (status)
2470 		cas_handle_irq1(cp, status);
2471 	spin_unlock_irqrestore(&cp->lock, flags);
2472 	return IRQ_HANDLED;
2473 }
2474 #endif
2475 
2476 static inline void cas_handle_irq(struct net_device *dev,
2477 				  struct cas *cp, const u32 status)
2478 {
2479 	/* housekeeping interrupts */
2480 	if (status & INTR_ERROR_MASK)
2481 		cas_abnormal_irq(dev, cp, status);
2482 
2483 	if (status & INTR_RX_BUF_UNAVAIL) {
2484 		/* Frame arrived, no free RX buffers available.
2485 		 * NOTE: we can get this on a link transition.
2486 		 */
2487 		cas_post_rxds_ringN(cp, 0, 0);
2488 		spin_lock(&cp->stat_lock[0]);
2489 		cp->net_stats[0].rx_dropped++;
2490 		spin_unlock(&cp->stat_lock[0]);
2491 	} else if (status & INTR_RX_BUF_AE) {
2492 		cas_post_rxds_ringN(cp, 0, RX_DESC_RINGN_SIZE(0) -
2493 				    RX_AE_FREEN_VAL(0));
2494 	}
2495 
2496 	if (status & (INTR_RX_COMP_AF | INTR_RX_COMP_FULL))
2497 		cas_post_rxcs_ringN(dev, cp, 0);
2498 }
2499 
2500 static irqreturn_t cas_interrupt(int irq, void *dev_id)
2501 {
2502 	struct net_device *dev = dev_id;
2503 	struct cas *cp = netdev_priv(dev);
2504 	unsigned long flags;
2505 	u32 status = readl(cp->regs + REG_INTR_STATUS);
2506 
2507 	if (status == 0)
2508 		return IRQ_NONE;
2509 
2510 	spin_lock_irqsave(&cp->lock, flags);
2511 	if (status & (INTR_TX_ALL | INTR_TX_INTME)) {
2512 		cas_tx(dev, cp, status);
2513 		status &= ~(INTR_TX_ALL | INTR_TX_INTME);
2514 	}
2515 
2516 	if (status & INTR_RX_DONE) {
2517 #ifdef USE_NAPI
2518 		cas_mask_intr(cp);
2519 		napi_schedule(&cp->napi);
2520 #else
2521 		cas_rx_ringN(cp, 0, 0);
2522 #endif
2523 		status &= ~INTR_RX_DONE;
2524 	}
2525 
2526 	if (status)
2527 		cas_handle_irq(dev, cp, status);
2528 	spin_unlock_irqrestore(&cp->lock, flags);
2529 	return IRQ_HANDLED;
2530 }
2531 
2532 
2533 #ifdef USE_NAPI
2534 static int cas_poll(struct napi_struct *napi, int budget)
2535 {
2536 	struct cas *cp = container_of(napi, struct cas, napi);
2537 	struct net_device *dev = cp->dev;
2538 	int i, enable_intr, credits;
2539 	u32 status = readl(cp->regs + REG_INTR_STATUS);
2540 	unsigned long flags;
2541 
2542 	spin_lock_irqsave(&cp->lock, flags);
2543 	cas_tx(dev, cp, status);
2544 	spin_unlock_irqrestore(&cp->lock, flags);
2545 
2546 	/* NAPI rx packets. we spread the credits across all of the
2547 	 * rxc rings
2548 	 *
2549 	 * to make sure we're fair with the work we loop through each
2550 	 * ring N_RX_COMP_RING times with a request of
2551 	 * budget / N_RX_COMP_RINGS
2552 	 */
2553 	enable_intr = 1;
2554 	credits = 0;
2555 	for (i = 0; i < N_RX_COMP_RINGS; i++) {
2556 		int j;
2557 		for (j = 0; j < N_RX_COMP_RINGS; j++) {
2558 			credits += cas_rx_ringN(cp, j, budget / N_RX_COMP_RINGS);
2559 			if (credits >= budget) {
2560 				enable_intr = 0;
2561 				goto rx_comp;
2562 			}
2563 		}
2564 	}
2565 
2566 rx_comp:
2567 	/* final rx completion */
2568 	spin_lock_irqsave(&cp->lock, flags);
2569 	if (status)
2570 		cas_handle_irq(dev, cp, status);
2571 
2572 #ifdef USE_PCI_INTB
2573 	if (N_RX_COMP_RINGS > 1) {
2574 		status = readl(cp->regs + REG_PLUS_INTRN_STATUS(1));
2575 		if (status)
2576 			cas_handle_irq1(dev, cp, status);
2577 	}
2578 #endif
2579 
2580 #ifdef USE_PCI_INTC
2581 	if (N_RX_COMP_RINGS > 2) {
2582 		status = readl(cp->regs + REG_PLUS_INTRN_STATUS(2));
2583 		if (status)
2584 			cas_handle_irqN(dev, cp, status, 2);
2585 	}
2586 #endif
2587 
2588 #ifdef USE_PCI_INTD
2589 	if (N_RX_COMP_RINGS > 3) {
2590 		status = readl(cp->regs + REG_PLUS_INTRN_STATUS(3));
2591 		if (status)
2592 			cas_handle_irqN(dev, cp, status, 3);
2593 	}
2594 #endif
2595 	spin_unlock_irqrestore(&cp->lock, flags);
2596 	if (enable_intr) {
2597 		napi_complete(napi);
2598 		cas_unmask_intr(cp);
2599 	}
2600 	return credits;
2601 }
2602 #endif
2603 
2604 #ifdef CONFIG_NET_POLL_CONTROLLER
2605 static void cas_netpoll(struct net_device *dev)
2606 {
2607 	struct cas *cp = netdev_priv(dev);
2608 
2609 	cas_disable_irq(cp, 0);
2610 	cas_interrupt(cp->pdev->irq, dev);
2611 	cas_enable_irq(cp, 0);
2612 
2613 #ifdef USE_PCI_INTB
2614 	if (N_RX_COMP_RINGS > 1) {
2615 		/* cas_interrupt1(); */
2616 	}
2617 #endif
2618 #ifdef USE_PCI_INTC
2619 	if (N_RX_COMP_RINGS > 2) {
2620 		/* cas_interruptN(); */
2621 	}
2622 #endif
2623 #ifdef USE_PCI_INTD
2624 	if (N_RX_COMP_RINGS > 3) {
2625 		/* cas_interruptN(); */
2626 	}
2627 #endif
2628 }
2629 #endif
2630 
2631 static void cas_tx_timeout(struct net_device *dev, unsigned int txqueue)
2632 {
2633 	struct cas *cp = netdev_priv(dev);
2634 
2635 	netdev_err(dev, "transmit timed out, resetting\n");
2636 	if (!cp->hw_running) {
2637 		netdev_err(dev, "hrm.. hw not running!\n");
2638 		return;
2639 	}
2640 
2641 	netdev_err(dev, "MIF_STATE[%08x]\n",
2642 		   readl(cp->regs + REG_MIF_STATE_MACHINE));
2643 
2644 	netdev_err(dev, "MAC_STATE[%08x]\n",
2645 		   readl(cp->regs + REG_MAC_STATE_MACHINE));
2646 
2647 	netdev_err(dev, "TX_STATE[%08x:%08x:%08x] FIFO[%08x:%08x:%08x] SM1[%08x] SM2[%08x]\n",
2648 		   readl(cp->regs + REG_TX_CFG),
2649 		   readl(cp->regs + REG_MAC_TX_STATUS),
2650 		   readl(cp->regs + REG_MAC_TX_CFG),
2651 		   readl(cp->regs + REG_TX_FIFO_PKT_CNT),
2652 		   readl(cp->regs + REG_TX_FIFO_WRITE_PTR),
2653 		   readl(cp->regs + REG_TX_FIFO_READ_PTR),
2654 		   readl(cp->regs + REG_TX_SM_1),
2655 		   readl(cp->regs + REG_TX_SM_2));
2656 
2657 	netdev_err(dev, "RX_STATE[%08x:%08x:%08x]\n",
2658 		   readl(cp->regs + REG_RX_CFG),
2659 		   readl(cp->regs + REG_MAC_RX_STATUS),
2660 		   readl(cp->regs + REG_MAC_RX_CFG));
2661 
2662 	netdev_err(dev, "HP_STATE[%08x:%08x:%08x:%08x]\n",
2663 		   readl(cp->regs + REG_HP_STATE_MACHINE),
2664 		   readl(cp->regs + REG_HP_STATUS0),
2665 		   readl(cp->regs + REG_HP_STATUS1),
2666 		   readl(cp->regs + REG_HP_STATUS2));
2667 
2668 #if 1
2669 	atomic_inc(&cp->reset_task_pending);
2670 	atomic_inc(&cp->reset_task_pending_all);
2671 	schedule_work(&cp->reset_task);
2672 #else
2673 	atomic_set(&cp->reset_task_pending, CAS_RESET_ALL);
2674 	schedule_work(&cp->reset_task);
2675 #endif
2676 }
2677 
2678 static inline int cas_intme(int ring, int entry)
2679 {
2680 	/* Algorithm: IRQ every 1/2 of descriptors. */
2681 	if (!(entry & ((TX_DESC_RINGN_SIZE(ring) >> 1) - 1)))
2682 		return 1;
2683 	return 0;
2684 }
2685 
2686 
2687 static void cas_write_txd(struct cas *cp, int ring, int entry,
2688 			  dma_addr_t mapping, int len, u64 ctrl, int last)
2689 {
2690 	struct cas_tx_desc *txd = cp->init_txds[ring] + entry;
2691 
2692 	ctrl |= CAS_BASE(TX_DESC_BUFLEN, len);
2693 	if (cas_intme(ring, entry))
2694 		ctrl |= TX_DESC_INTME;
2695 	if (last)
2696 		ctrl |= TX_DESC_EOF;
2697 	txd->control = cpu_to_le64(ctrl);
2698 	txd->buffer = cpu_to_le64(mapping);
2699 }
2700 
2701 static inline void *tx_tiny_buf(struct cas *cp, const int ring,
2702 				const int entry)
2703 {
2704 	return cp->tx_tiny_bufs[ring] + TX_TINY_BUF_LEN*entry;
2705 }
2706 
2707 static inline dma_addr_t tx_tiny_map(struct cas *cp, const int ring,
2708 				     const int entry, const int tentry)
2709 {
2710 	cp->tx_tiny_use[ring][tentry].nbufs++;
2711 	cp->tx_tiny_use[ring][entry].used = 1;
2712 	return cp->tx_tiny_dvma[ring] + TX_TINY_BUF_LEN*entry;
2713 }
2714 
2715 static inline int cas_xmit_tx_ringN(struct cas *cp, int ring,
2716 				    struct sk_buff *skb)
2717 {
2718 	struct net_device *dev = cp->dev;
2719 	int entry, nr_frags, frag, tabort, tentry;
2720 	dma_addr_t mapping;
2721 	unsigned long flags;
2722 	u64 ctrl;
2723 	u32 len;
2724 
2725 	spin_lock_irqsave(&cp->tx_lock[ring], flags);
2726 
2727 	/* This is a hard error, log it. */
2728 	if (TX_BUFFS_AVAIL(cp, ring) <=
2729 	    CAS_TABORT(cp)*(skb_shinfo(skb)->nr_frags + 1)) {
2730 		netif_stop_queue(dev);
2731 		spin_unlock_irqrestore(&cp->tx_lock[ring], flags);
2732 		netdev_err(dev, "BUG! Tx Ring full when queue awake!\n");
2733 		return 1;
2734 	}
2735 
2736 	ctrl = 0;
2737 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
2738 		const u64 csum_start_off = skb_checksum_start_offset(skb);
2739 		const u64 csum_stuff_off = csum_start_off + skb->csum_offset;
2740 
2741 		ctrl =  TX_DESC_CSUM_EN |
2742 			CAS_BASE(TX_DESC_CSUM_START, csum_start_off) |
2743 			CAS_BASE(TX_DESC_CSUM_STUFF, csum_stuff_off);
2744 	}
2745 
2746 	entry = cp->tx_new[ring];
2747 	cp->tx_skbs[ring][entry] = skb;
2748 
2749 	nr_frags = skb_shinfo(skb)->nr_frags;
2750 	len = skb_headlen(skb);
2751 	mapping = dma_map_page(&cp->pdev->dev, virt_to_page(skb->data),
2752 			       offset_in_page(skb->data), len, DMA_TO_DEVICE);
2753 
2754 	tentry = entry;
2755 	tabort = cas_calc_tabort(cp, (unsigned long) skb->data, len);
2756 	if (unlikely(tabort)) {
2757 		/* NOTE: len is always >  tabort */
2758 		cas_write_txd(cp, ring, entry, mapping, len - tabort,
2759 			      ctrl | TX_DESC_SOF, 0);
2760 		entry = TX_DESC_NEXT(ring, entry);
2761 
2762 		skb_copy_from_linear_data_offset(skb, len - tabort,
2763 			      tx_tiny_buf(cp, ring, entry), tabort);
2764 		mapping = tx_tiny_map(cp, ring, entry, tentry);
2765 		cas_write_txd(cp, ring, entry, mapping, tabort, ctrl,
2766 			      (nr_frags == 0));
2767 	} else {
2768 		cas_write_txd(cp, ring, entry, mapping, len, ctrl |
2769 			      TX_DESC_SOF, (nr_frags == 0));
2770 	}
2771 	entry = TX_DESC_NEXT(ring, entry);
2772 
2773 	for (frag = 0; frag < nr_frags; frag++) {
2774 		const skb_frag_t *fragp = &skb_shinfo(skb)->frags[frag];
2775 
2776 		len = skb_frag_size(fragp);
2777 		mapping = skb_frag_dma_map(&cp->pdev->dev, fragp, 0, len,
2778 					   DMA_TO_DEVICE);
2779 
2780 		tabort = cas_calc_tabort(cp, skb_frag_off(fragp), len);
2781 		if (unlikely(tabort)) {
2782 			/* NOTE: len is always > tabort */
2783 			cas_write_txd(cp, ring, entry, mapping, len - tabort,
2784 				      ctrl, 0);
2785 			entry = TX_DESC_NEXT(ring, entry);
2786 			memcpy_from_page(tx_tiny_buf(cp, ring, entry),
2787 					 skb_frag_page(fragp),
2788 					 skb_frag_off(fragp) + len - tabort,
2789 					 tabort);
2790 			mapping = tx_tiny_map(cp, ring, entry, tentry);
2791 			len     = tabort;
2792 		}
2793 
2794 		cas_write_txd(cp, ring, entry, mapping, len, ctrl,
2795 			      (frag + 1 == nr_frags));
2796 		entry = TX_DESC_NEXT(ring, entry);
2797 	}
2798 
2799 	cp->tx_new[ring] = entry;
2800 	if (TX_BUFFS_AVAIL(cp, ring) <= CAS_TABORT(cp)*(MAX_SKB_FRAGS + 1))
2801 		netif_stop_queue(dev);
2802 
2803 	netif_printk(cp, tx_queued, KERN_DEBUG, dev,
2804 		     "tx[%d] queued, slot %d, skblen %d, avail %d\n",
2805 		     ring, entry, skb->len, TX_BUFFS_AVAIL(cp, ring));
2806 	writel(entry, cp->regs + REG_TX_KICKN(ring));
2807 	spin_unlock_irqrestore(&cp->tx_lock[ring], flags);
2808 	return 0;
2809 }
2810 
2811 static netdev_tx_t cas_start_xmit(struct sk_buff *skb, struct net_device *dev)
2812 {
2813 	struct cas *cp = netdev_priv(dev);
2814 
2815 	/* this is only used as a load-balancing hint, so it doesn't
2816 	 * need to be SMP safe
2817 	 */
2818 	static int ring;
2819 
2820 	if (skb_padto(skb, cp->min_frame_size))
2821 		return NETDEV_TX_OK;
2822 
2823 	/* XXX: we need some higher-level QoS hooks to steer packets to
2824 	 *      individual queues.
2825 	 */
2826 	if (cas_xmit_tx_ringN(cp, ring++ & N_TX_RINGS_MASK, skb))
2827 		return NETDEV_TX_BUSY;
2828 	return NETDEV_TX_OK;
2829 }
2830 
2831 static void cas_init_tx_dma(struct cas *cp)
2832 {
2833 	u64 desc_dma = cp->block_dvma;
2834 	unsigned long off;
2835 	u32 val;
2836 	int i;
2837 
2838 	/* set up tx completion writeback registers. must be 8-byte aligned */
2839 #ifdef USE_TX_COMPWB
2840 	off = offsetof(struct cas_init_block, tx_compwb);
2841 	writel((desc_dma + off) >> 32, cp->regs + REG_TX_COMPWB_DB_HI);
2842 	writel((desc_dma + off) & 0xffffffff, cp->regs + REG_TX_COMPWB_DB_LOW);
2843 #endif
2844 
2845 	/* enable completion writebacks, enable paced mode,
2846 	 * disable read pipe, and disable pre-interrupt compwbs
2847 	 */
2848 	val =   TX_CFG_COMPWB_Q1 | TX_CFG_COMPWB_Q2 |
2849 		TX_CFG_COMPWB_Q3 | TX_CFG_COMPWB_Q4 |
2850 		TX_CFG_DMA_RDPIPE_DIS | TX_CFG_PACED_MODE |
2851 		TX_CFG_INTR_COMPWB_DIS;
2852 
2853 	/* write out tx ring info and tx desc bases */
2854 	for (i = 0; i < MAX_TX_RINGS; i++) {
2855 		off = (unsigned long) cp->init_txds[i] -
2856 			(unsigned long) cp->init_block;
2857 
2858 		val |= CAS_TX_RINGN_BASE(i);
2859 		writel((desc_dma + off) >> 32, cp->regs + REG_TX_DBN_HI(i));
2860 		writel((desc_dma + off) & 0xffffffff, cp->regs +
2861 		       REG_TX_DBN_LOW(i));
2862 		/* don't zero out the kick register here as the system
2863 		 * will wedge
2864 		 */
2865 	}
2866 	writel(val, cp->regs + REG_TX_CFG);
2867 
2868 	/* program max burst sizes. these numbers should be different
2869 	 * if doing QoS.
2870 	 */
2871 #ifdef USE_QOS
2872 	writel(0x800, cp->regs + REG_TX_MAXBURST_0);
2873 	writel(0x1600, cp->regs + REG_TX_MAXBURST_1);
2874 	writel(0x2400, cp->regs + REG_TX_MAXBURST_2);
2875 	writel(0x4800, cp->regs + REG_TX_MAXBURST_3);
2876 #else
2877 	writel(0x800, cp->regs + REG_TX_MAXBURST_0);
2878 	writel(0x800, cp->regs + REG_TX_MAXBURST_1);
2879 	writel(0x800, cp->regs + REG_TX_MAXBURST_2);
2880 	writel(0x800, cp->regs + REG_TX_MAXBURST_3);
2881 #endif
2882 }
2883 
2884 /* Must be invoked under cp->lock. */
2885 static inline void cas_init_dma(struct cas *cp)
2886 {
2887 	cas_init_tx_dma(cp);
2888 	cas_init_rx_dma(cp);
2889 }
2890 
2891 static void cas_process_mc_list(struct cas *cp)
2892 {
2893 	u16 hash_table[16];
2894 	u32 crc;
2895 	struct netdev_hw_addr *ha;
2896 	int i = 1;
2897 
2898 	memset(hash_table, 0, sizeof(hash_table));
2899 	netdev_for_each_mc_addr(ha, cp->dev) {
2900 		if (i <= CAS_MC_EXACT_MATCH_SIZE) {
2901 			/* use the alternate mac address registers for the
2902 			 * first 15 multicast addresses
2903 			 */
2904 			writel((ha->addr[4] << 8) | ha->addr[5],
2905 			       cp->regs + REG_MAC_ADDRN(i*3 + 0));
2906 			writel((ha->addr[2] << 8) | ha->addr[3],
2907 			       cp->regs + REG_MAC_ADDRN(i*3 + 1));
2908 			writel((ha->addr[0] << 8) | ha->addr[1],
2909 			       cp->regs + REG_MAC_ADDRN(i*3 + 2));
2910 			i++;
2911 		}
2912 		else {
2913 			/* use hw hash table for the next series of
2914 			 * multicast addresses
2915 			 */
2916 			crc = ether_crc_le(ETH_ALEN, ha->addr);
2917 			crc >>= 24;
2918 			hash_table[crc >> 4] |= 1 << (15 - (crc & 0xf));
2919 		}
2920 	}
2921 	for (i = 0; i < 16; i++)
2922 		writel(hash_table[i], cp->regs + REG_MAC_HASH_TABLEN(i));
2923 }
2924 
2925 /* Must be invoked under cp->lock. */
2926 static u32 cas_setup_multicast(struct cas *cp)
2927 {
2928 	u32 rxcfg = 0;
2929 	int i;
2930 
2931 	if (cp->dev->flags & IFF_PROMISC) {
2932 		rxcfg |= MAC_RX_CFG_PROMISC_EN;
2933 
2934 	} else if (cp->dev->flags & IFF_ALLMULTI) {
2935 	    	for (i=0; i < 16; i++)
2936 			writel(0xFFFF, cp->regs + REG_MAC_HASH_TABLEN(i));
2937 		rxcfg |= MAC_RX_CFG_HASH_FILTER_EN;
2938 
2939 	} else {
2940 		cas_process_mc_list(cp);
2941 		rxcfg |= MAC_RX_CFG_HASH_FILTER_EN;
2942 	}
2943 
2944 	return rxcfg;
2945 }
2946 
2947 /* must be invoked under cp->stat_lock[N_TX_RINGS] */
2948 static void cas_clear_mac_err(struct cas *cp)
2949 {
2950 	writel(0, cp->regs + REG_MAC_COLL_NORMAL);
2951 	writel(0, cp->regs + REG_MAC_COLL_FIRST);
2952 	writel(0, cp->regs + REG_MAC_COLL_EXCESS);
2953 	writel(0, cp->regs + REG_MAC_COLL_LATE);
2954 	writel(0, cp->regs + REG_MAC_TIMER_DEFER);
2955 	writel(0, cp->regs + REG_MAC_ATTEMPTS_PEAK);
2956 	writel(0, cp->regs + REG_MAC_RECV_FRAME);
2957 	writel(0, cp->regs + REG_MAC_LEN_ERR);
2958 	writel(0, cp->regs + REG_MAC_ALIGN_ERR);
2959 	writel(0, cp->regs + REG_MAC_FCS_ERR);
2960 	writel(0, cp->regs + REG_MAC_RX_CODE_ERR);
2961 }
2962 
2963 
2964 static void cas_mac_reset(struct cas *cp)
2965 {
2966 	int i;
2967 
2968 	/* do both TX and RX reset */
2969 	writel(0x1, cp->regs + REG_MAC_TX_RESET);
2970 	writel(0x1, cp->regs + REG_MAC_RX_RESET);
2971 
2972 	/* wait for TX */
2973 	i = STOP_TRIES;
2974 	while (i-- > 0) {
2975 		if (readl(cp->regs + REG_MAC_TX_RESET) == 0)
2976 			break;
2977 		udelay(10);
2978 	}
2979 
2980 	/* wait for RX */
2981 	i = STOP_TRIES;
2982 	while (i-- > 0) {
2983 		if (readl(cp->regs + REG_MAC_RX_RESET) == 0)
2984 			break;
2985 		udelay(10);
2986 	}
2987 
2988 	if (readl(cp->regs + REG_MAC_TX_RESET) |
2989 	    readl(cp->regs + REG_MAC_RX_RESET))
2990 		netdev_err(cp->dev, "mac tx[%d]/rx[%d] reset failed [%08x]\n",
2991 			   readl(cp->regs + REG_MAC_TX_RESET),
2992 			   readl(cp->regs + REG_MAC_RX_RESET),
2993 			   readl(cp->regs + REG_MAC_STATE_MACHINE));
2994 }
2995 
2996 
2997 /* Must be invoked under cp->lock. */
2998 static void cas_init_mac(struct cas *cp)
2999 {
3000 	const unsigned char *e = &cp->dev->dev_addr[0];
3001 	int i;
3002 	cas_mac_reset(cp);
3003 
3004 	/* setup core arbitration weight register */
3005 	writel(CAWR_RR_DIS, cp->regs + REG_CAWR);
3006 
3007 #if !defined(CONFIG_SPARC64) && !defined(CONFIG_ALPHA)
3008 	/* set the infinite burst register for chips that don't have
3009 	 * pci issues.
3010 	 */
3011 	if ((cp->cas_flags & CAS_FLAG_TARGET_ABORT) == 0)
3012 		writel(INF_BURST_EN, cp->regs + REG_INF_BURST);
3013 #endif
3014 
3015 	writel(0x1BF0, cp->regs + REG_MAC_SEND_PAUSE);
3016 
3017 	writel(0x00, cp->regs + REG_MAC_IPG0);
3018 	writel(0x08, cp->regs + REG_MAC_IPG1);
3019 	writel(0x04, cp->regs + REG_MAC_IPG2);
3020 
3021 	/* change later for 802.3z */
3022 	writel(0x40, cp->regs + REG_MAC_SLOT_TIME);
3023 
3024 	/* min frame + FCS */
3025 	writel(ETH_ZLEN + 4, cp->regs + REG_MAC_FRAMESIZE_MIN);
3026 
3027 	/* Ethernet payload + header + FCS + optional VLAN tag. NOTE: we
3028 	 * specify the maximum frame size to prevent RX tag errors on
3029 	 * oversized frames.
3030 	 */
3031 	writel(CAS_BASE(MAC_FRAMESIZE_MAX_BURST, 0x2000) |
3032 	       CAS_BASE(MAC_FRAMESIZE_MAX_FRAME,
3033 			(CAS_MAX_MTU + ETH_HLEN + 4 + 4)),
3034 	       cp->regs + REG_MAC_FRAMESIZE_MAX);
3035 
3036 	/* NOTE: crc_size is used as a surrogate for half-duplex.
3037 	 * workaround saturn half-duplex issue by increasing preamble
3038 	 * size to 65 bytes.
3039 	 */
3040 	if ((cp->cas_flags & CAS_FLAG_SATURN) && cp->crc_size)
3041 		writel(0x41, cp->regs + REG_MAC_PA_SIZE);
3042 	else
3043 		writel(0x07, cp->regs + REG_MAC_PA_SIZE);
3044 	writel(0x04, cp->regs + REG_MAC_JAM_SIZE);
3045 	writel(0x10, cp->regs + REG_MAC_ATTEMPT_LIMIT);
3046 	writel(0x8808, cp->regs + REG_MAC_CTRL_TYPE);
3047 
3048 	writel((e[5] | (e[4] << 8)) & 0x3ff, cp->regs + REG_MAC_RANDOM_SEED);
3049 
3050 	writel(0, cp->regs + REG_MAC_ADDR_FILTER0);
3051 	writel(0, cp->regs + REG_MAC_ADDR_FILTER1);
3052 	writel(0, cp->regs + REG_MAC_ADDR_FILTER2);
3053 	writel(0, cp->regs + REG_MAC_ADDR_FILTER2_1_MASK);
3054 	writel(0, cp->regs + REG_MAC_ADDR_FILTER0_MASK);
3055 
3056 	/* setup mac address in perfect filter array */
3057 	for (i = 0; i < 45; i++)
3058 		writel(0x0, cp->regs + REG_MAC_ADDRN(i));
3059 
3060 	writel((e[4] << 8) | e[5], cp->regs + REG_MAC_ADDRN(0));
3061 	writel((e[2] << 8) | e[3], cp->regs + REG_MAC_ADDRN(1));
3062 	writel((e[0] << 8) | e[1], cp->regs + REG_MAC_ADDRN(2));
3063 
3064 	writel(0x0001, cp->regs + REG_MAC_ADDRN(42));
3065 	writel(0xc200, cp->regs + REG_MAC_ADDRN(43));
3066 	writel(0x0180, cp->regs + REG_MAC_ADDRN(44));
3067 
3068 	cp->mac_rx_cfg = cas_setup_multicast(cp);
3069 
3070 	spin_lock(&cp->stat_lock[N_TX_RINGS]);
3071 	cas_clear_mac_err(cp);
3072 	spin_unlock(&cp->stat_lock[N_TX_RINGS]);
3073 
3074 	/* Setup MAC interrupts.  We want to get all of the interesting
3075 	 * counter expiration events, but we do not want to hear about
3076 	 * normal rx/tx as the DMA engine tells us that.
3077 	 */
3078 	writel(MAC_TX_FRAME_XMIT, cp->regs + REG_MAC_TX_MASK);
3079 	writel(MAC_RX_FRAME_RECV, cp->regs + REG_MAC_RX_MASK);
3080 
3081 	/* Don't enable even the PAUSE interrupts for now, we
3082 	 * make no use of those events other than to record them.
3083 	 */
3084 	writel(0xffffffff, cp->regs + REG_MAC_CTRL_MASK);
3085 }
3086 
3087 /* Must be invoked under cp->lock. */
3088 static void cas_init_pause_thresholds(struct cas *cp)
3089 {
3090 	/* Calculate pause thresholds.  Setting the OFF threshold to the
3091 	 * full RX fifo size effectively disables PAUSE generation
3092 	 */
3093 	if (cp->rx_fifo_size <= (2 * 1024)) {
3094 		cp->rx_pause_off = cp->rx_pause_on = cp->rx_fifo_size;
3095 	} else {
3096 		int max_frame = (cp->dev->mtu + ETH_HLEN + 4 + 4 + 64) & ~63;
3097 		if (max_frame * 3 > cp->rx_fifo_size) {
3098 			cp->rx_pause_off = 7104;
3099 			cp->rx_pause_on  = 960;
3100 		} else {
3101 			int off = (cp->rx_fifo_size - (max_frame * 2));
3102 			int on = off - max_frame;
3103 			cp->rx_pause_off = off;
3104 			cp->rx_pause_on = on;
3105 		}
3106 	}
3107 }
3108 
3109 static int cas_vpd_match(const void __iomem *p, const char *str)
3110 {
3111 	int len = strlen(str) + 1;
3112 	int i;
3113 
3114 	for (i = 0; i < len; i++) {
3115 		if (readb(p + i) != str[i])
3116 			return 0;
3117 	}
3118 	return 1;
3119 }
3120 
3121 
3122 /* get the mac address by reading the vpd information in the rom.
3123  * also get the phy type and determine if there's an entropy generator.
3124  * NOTE: this is a bit convoluted for the following reasons:
3125  *  1) vpd info has order-dependent mac addresses for multinic cards
3126  *  2) the only way to determine the nic order is to use the slot
3127  *     number.
3128  *  3) fiber cards don't have bridges, so their slot numbers don't
3129  *     mean anything.
3130  *  4) we don't actually know we have a fiber card until after
3131  *     the mac addresses are parsed.
3132  */
3133 static int cas_get_vpd_info(struct cas *cp, unsigned char *dev_addr,
3134 			    const int offset)
3135 {
3136 	void __iomem *p = cp->regs + REG_EXPANSION_ROM_RUN_START;
3137 	void __iomem *base, *kstart;
3138 	int i, len;
3139 	int found = 0;
3140 #define VPD_FOUND_MAC        0x01
3141 #define VPD_FOUND_PHY        0x02
3142 
3143 	int phy_type = CAS_PHY_MII_MDIO0; /* default phy type */
3144 	int mac_off  = 0;
3145 
3146 #if defined(CONFIG_SPARC)
3147 	const unsigned char *addr;
3148 #endif
3149 
3150 	/* give us access to the PROM */
3151 	writel(BIM_LOCAL_DEV_PROM | BIM_LOCAL_DEV_PAD,
3152 	       cp->regs + REG_BIM_LOCAL_DEV_EN);
3153 
3154 	/* check for an expansion rom */
3155 	if (readb(p) != 0x55 || readb(p + 1) != 0xaa)
3156 		goto use_random_mac_addr;
3157 
3158 	/* search for beginning of vpd */
3159 	base = NULL;
3160 	for (i = 2; i < EXPANSION_ROM_SIZE; i++) {
3161 		/* check for PCIR */
3162 		if ((readb(p + i + 0) == 0x50) &&
3163 		    (readb(p + i + 1) == 0x43) &&
3164 		    (readb(p + i + 2) == 0x49) &&
3165 		    (readb(p + i + 3) == 0x52)) {
3166 			base = p + (readb(p + i + 8) |
3167 				    (readb(p + i + 9) << 8));
3168 			break;
3169 		}
3170 	}
3171 
3172 	if (!base || (readb(base) != 0x82))
3173 		goto use_random_mac_addr;
3174 
3175 	i = (readb(base + 1) | (readb(base + 2) << 8)) + 3;
3176 	while (i < EXPANSION_ROM_SIZE) {
3177 		if (readb(base + i) != 0x90) /* no vpd found */
3178 			goto use_random_mac_addr;
3179 
3180 		/* found a vpd field */
3181 		len = readb(base + i + 1) | (readb(base + i + 2) << 8);
3182 
3183 		/* extract keywords */
3184 		kstart = base + i + 3;
3185 		p = kstart;
3186 		while ((p - kstart) < len) {
3187 			int klen = readb(p + 2);
3188 			int j;
3189 			char type;
3190 
3191 			p += 3;
3192 
3193 			/* look for the following things:
3194 			 * -- correct length == 29
3195 			 * 3 (type) + 2 (size) +
3196 			 * 18 (strlen("local-mac-address") + 1) +
3197 			 * 6 (mac addr)
3198 			 * -- VPD Instance 'I'
3199 			 * -- VPD Type Bytes 'B'
3200 			 * -- VPD data length == 6
3201 			 * -- property string == local-mac-address
3202 			 *
3203 			 * -- correct length == 24
3204 			 * 3 (type) + 2 (size) +
3205 			 * 12 (strlen("entropy-dev") + 1) +
3206 			 * 7 (strlen("vms110") + 1)
3207 			 * -- VPD Instance 'I'
3208 			 * -- VPD Type String 'B'
3209 			 * -- VPD data length == 7
3210 			 * -- property string == entropy-dev
3211 			 *
3212 			 * -- correct length == 18
3213 			 * 3 (type) + 2 (size) +
3214 			 * 9 (strlen("phy-type") + 1) +
3215 			 * 4 (strlen("pcs") + 1)
3216 			 * -- VPD Instance 'I'
3217 			 * -- VPD Type String 'S'
3218 			 * -- VPD data length == 4
3219 			 * -- property string == phy-type
3220 			 *
3221 			 * -- correct length == 23
3222 			 * 3 (type) + 2 (size) +
3223 			 * 14 (strlen("phy-interface") + 1) +
3224 			 * 4 (strlen("pcs") + 1)
3225 			 * -- VPD Instance 'I'
3226 			 * -- VPD Type String 'S'
3227 			 * -- VPD data length == 4
3228 			 * -- property string == phy-interface
3229 			 */
3230 			if (readb(p) != 'I')
3231 				goto next;
3232 
3233 			/* finally, check string and length */
3234 			type = readb(p + 3);
3235 			if (type == 'B') {
3236 				if ((klen == 29) && readb(p + 4) == 6 &&
3237 				    cas_vpd_match(p + 5,
3238 						  "local-mac-address")) {
3239 					if (mac_off++ > offset)
3240 						goto next;
3241 
3242 					/* set mac address */
3243 					for (j = 0; j < 6; j++)
3244 						dev_addr[j] =
3245 							readb(p + 23 + j);
3246 					goto found_mac;
3247 				}
3248 			}
3249 
3250 			if (type != 'S')
3251 				goto next;
3252 
3253 #ifdef USE_ENTROPY_DEV
3254 			if ((klen == 24) &&
3255 			    cas_vpd_match(p + 5, "entropy-dev") &&
3256 			    cas_vpd_match(p + 17, "vms110")) {
3257 				cp->cas_flags |= CAS_FLAG_ENTROPY_DEV;
3258 				goto next;
3259 			}
3260 #endif
3261 
3262 			if (found & VPD_FOUND_PHY)
3263 				goto next;
3264 
3265 			if ((klen == 18) && readb(p + 4) == 4 &&
3266 			    cas_vpd_match(p + 5, "phy-type")) {
3267 				if (cas_vpd_match(p + 14, "pcs")) {
3268 					phy_type = CAS_PHY_SERDES;
3269 					goto found_phy;
3270 				}
3271 			}
3272 
3273 			if ((klen == 23) && readb(p + 4) == 4 &&
3274 			    cas_vpd_match(p + 5, "phy-interface")) {
3275 				if (cas_vpd_match(p + 19, "pcs")) {
3276 					phy_type = CAS_PHY_SERDES;
3277 					goto found_phy;
3278 				}
3279 			}
3280 found_mac:
3281 			found |= VPD_FOUND_MAC;
3282 			goto next;
3283 
3284 found_phy:
3285 			found |= VPD_FOUND_PHY;
3286 
3287 next:
3288 			p += klen;
3289 		}
3290 		i += len + 3;
3291 	}
3292 
3293 use_random_mac_addr:
3294 	if (found & VPD_FOUND_MAC)
3295 		goto done;
3296 
3297 #if defined(CONFIG_SPARC)
3298 	addr = of_get_property(cp->of_node, "local-mac-address", NULL);
3299 	if (addr != NULL) {
3300 		memcpy(dev_addr, addr, ETH_ALEN);
3301 		goto done;
3302 	}
3303 #endif
3304 
3305 	/* Sun MAC prefix then 3 random bytes. */
3306 	pr_info("MAC address not found in ROM VPD\n");
3307 	dev_addr[0] = 0x08;
3308 	dev_addr[1] = 0x00;
3309 	dev_addr[2] = 0x20;
3310 	get_random_bytes(dev_addr + 3, 3);
3311 
3312 done:
3313 	writel(0, cp->regs + REG_BIM_LOCAL_DEV_EN);
3314 	return phy_type;
3315 }
3316 
3317 /* check pci invariants */
3318 static void cas_check_pci_invariants(struct cas *cp)
3319 {
3320 	struct pci_dev *pdev = cp->pdev;
3321 
3322 	cp->cas_flags = 0;
3323 	if ((pdev->vendor == PCI_VENDOR_ID_SUN) &&
3324 	    (pdev->device == PCI_DEVICE_ID_SUN_CASSINI)) {
3325 		if (pdev->revision >= CAS_ID_REVPLUS)
3326 			cp->cas_flags |= CAS_FLAG_REG_PLUS;
3327 		if (pdev->revision < CAS_ID_REVPLUS02u)
3328 			cp->cas_flags |= CAS_FLAG_TARGET_ABORT;
3329 
3330 		/* Original Cassini supports HW CSUM, but it's not
3331 		 * enabled by default as it can trigger TX hangs.
3332 		 */
3333 		if (pdev->revision < CAS_ID_REV2)
3334 			cp->cas_flags |= CAS_FLAG_NO_HW_CSUM;
3335 	} else {
3336 		/* Only sun has original cassini chips.  */
3337 		cp->cas_flags |= CAS_FLAG_REG_PLUS;
3338 
3339 		/* We use a flag because the same phy might be externally
3340 		 * connected.
3341 		 */
3342 		if ((pdev->vendor == PCI_VENDOR_ID_NS) &&
3343 		    (pdev->device == PCI_DEVICE_ID_NS_SATURN))
3344 			cp->cas_flags |= CAS_FLAG_SATURN;
3345 	}
3346 }
3347 
3348 
3349 static int cas_check_invariants(struct cas *cp)
3350 {
3351 	struct pci_dev *pdev = cp->pdev;
3352 	u8 addr[ETH_ALEN];
3353 	u32 cfg;
3354 	int i;
3355 
3356 	/* get page size for rx buffers. */
3357 	cp->page_order = 0;
3358 #ifdef USE_PAGE_ORDER
3359 	if (PAGE_SHIFT < CAS_JUMBO_PAGE_SHIFT) {
3360 		/* see if we can allocate larger pages */
3361 		struct page *page = alloc_pages(GFP_ATOMIC,
3362 						CAS_JUMBO_PAGE_SHIFT -
3363 						PAGE_SHIFT);
3364 		if (page) {
3365 			__free_pages(page, CAS_JUMBO_PAGE_SHIFT - PAGE_SHIFT);
3366 			cp->page_order = CAS_JUMBO_PAGE_SHIFT - PAGE_SHIFT;
3367 		} else {
3368 			printk("MTU limited to %d bytes\n", CAS_MAX_MTU);
3369 		}
3370 	}
3371 #endif
3372 	cp->page_size = (PAGE_SIZE << cp->page_order);
3373 
3374 	/* Fetch the FIFO configurations. */
3375 	cp->tx_fifo_size = readl(cp->regs + REG_TX_FIFO_SIZE) * 64;
3376 	cp->rx_fifo_size = RX_FIFO_SIZE;
3377 
3378 	/* finish phy determination. MDIO1 takes precedence over MDIO0 if
3379 	 * they're both connected.
3380 	 */
3381 	cp->phy_type = cas_get_vpd_info(cp, addr, PCI_SLOT(pdev->devfn));
3382 	eth_hw_addr_set(cp->dev, addr);
3383 	if (cp->phy_type & CAS_PHY_SERDES) {
3384 		cp->cas_flags |= CAS_FLAG_1000MB_CAP;
3385 		return 0; /* no more checking needed */
3386 	}
3387 
3388 	/* MII */
3389 	cfg = readl(cp->regs + REG_MIF_CFG);
3390 	if (cfg & MIF_CFG_MDIO_1) {
3391 		cp->phy_type = CAS_PHY_MII_MDIO1;
3392 	} else if (cfg & MIF_CFG_MDIO_0) {
3393 		cp->phy_type = CAS_PHY_MII_MDIO0;
3394 	}
3395 
3396 	cas_mif_poll(cp, 0);
3397 	writel(PCS_DATAPATH_MODE_MII, cp->regs + REG_PCS_DATAPATH_MODE);
3398 
3399 	for (i = 0; i < 32; i++) {
3400 		u32 phy_id;
3401 		int j;
3402 
3403 		for (j = 0; j < 3; j++) {
3404 			cp->phy_addr = i;
3405 			phy_id = cas_phy_read(cp, MII_PHYSID1) << 16;
3406 			phy_id |= cas_phy_read(cp, MII_PHYSID2);
3407 			if (phy_id && (phy_id != 0xFFFFFFFF)) {
3408 				cp->phy_id = phy_id;
3409 				goto done;
3410 			}
3411 		}
3412 	}
3413 	pr_err("MII phy did not respond [%08x]\n",
3414 	       readl(cp->regs + REG_MIF_STATE_MACHINE));
3415 	return -1;
3416 
3417 done:
3418 	/* see if we can do gigabit */
3419 	cfg = cas_phy_read(cp, MII_BMSR);
3420 	if ((cfg & CAS_BMSR_1000_EXTEND) &&
3421 	    cas_phy_read(cp, CAS_MII_1000_EXTEND))
3422 		cp->cas_flags |= CAS_FLAG_1000MB_CAP;
3423 	return 0;
3424 }
3425 
3426 /* Must be invoked under cp->lock. */
3427 static inline void cas_start_dma(struct cas *cp)
3428 {
3429 	int i;
3430 	u32 val;
3431 	int txfailed = 0;
3432 
3433 	/* enable dma */
3434 	val = readl(cp->regs + REG_TX_CFG) | TX_CFG_DMA_EN;
3435 	writel(val, cp->regs + REG_TX_CFG);
3436 	val = readl(cp->regs + REG_RX_CFG) | RX_CFG_DMA_EN;
3437 	writel(val, cp->regs + REG_RX_CFG);
3438 
3439 	/* enable the mac */
3440 	val = readl(cp->regs + REG_MAC_TX_CFG) | MAC_TX_CFG_EN;
3441 	writel(val, cp->regs + REG_MAC_TX_CFG);
3442 	val = readl(cp->regs + REG_MAC_RX_CFG) | MAC_RX_CFG_EN;
3443 	writel(val, cp->regs + REG_MAC_RX_CFG);
3444 
3445 	i = STOP_TRIES;
3446 	while (i-- > 0) {
3447 		val = readl(cp->regs + REG_MAC_TX_CFG);
3448 		if ((val & MAC_TX_CFG_EN))
3449 			break;
3450 		udelay(10);
3451 	}
3452 	if (i < 0) txfailed = 1;
3453 	i = STOP_TRIES;
3454 	while (i-- > 0) {
3455 		val = readl(cp->regs + REG_MAC_RX_CFG);
3456 		if ((val & MAC_RX_CFG_EN)) {
3457 			if (txfailed) {
3458 				netdev_err(cp->dev,
3459 					   "enabling mac failed [tx:%08x:%08x]\n",
3460 					   readl(cp->regs + REG_MIF_STATE_MACHINE),
3461 					   readl(cp->regs + REG_MAC_STATE_MACHINE));
3462 			}
3463 			goto enable_rx_done;
3464 		}
3465 		udelay(10);
3466 	}
3467 	netdev_err(cp->dev, "enabling mac failed [%s:%08x:%08x]\n",
3468 		   (txfailed ? "tx,rx" : "rx"),
3469 		   readl(cp->regs + REG_MIF_STATE_MACHINE),
3470 		   readl(cp->regs + REG_MAC_STATE_MACHINE));
3471 
3472 enable_rx_done:
3473 	cas_unmask_intr(cp); /* enable interrupts */
3474 	writel(RX_DESC_RINGN_SIZE(0) - 4, cp->regs + REG_RX_KICK);
3475 	writel(0, cp->regs + REG_RX_COMP_TAIL);
3476 
3477 	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
3478 		if (N_RX_DESC_RINGS > 1)
3479 			writel(RX_DESC_RINGN_SIZE(1) - 4,
3480 			       cp->regs + REG_PLUS_RX_KICK1);
3481 	}
3482 }
3483 
3484 /* Must be invoked under cp->lock. */
3485 static void cas_read_pcs_link_mode(struct cas *cp, int *fd, int *spd,
3486 				   int *pause)
3487 {
3488 	u32 val = readl(cp->regs + REG_PCS_MII_LPA);
3489 	*fd     = (val & PCS_MII_LPA_FD) ? 1 : 0;
3490 	*pause  = (val & PCS_MII_LPA_SYM_PAUSE) ? 0x01 : 0x00;
3491 	if (val & PCS_MII_LPA_ASYM_PAUSE)
3492 		*pause |= 0x10;
3493 	*spd = 1000;
3494 }
3495 
3496 /* Must be invoked under cp->lock. */
3497 static void cas_read_mii_link_mode(struct cas *cp, int *fd, int *spd,
3498 				   int *pause)
3499 {
3500 	u32 val;
3501 
3502 	*fd = 0;
3503 	*spd = 10;
3504 	*pause = 0;
3505 
3506 	/* use GMII registers */
3507 	val = cas_phy_read(cp, MII_LPA);
3508 	if (val & CAS_LPA_PAUSE)
3509 		*pause = 0x01;
3510 
3511 	if (val & CAS_LPA_ASYM_PAUSE)
3512 		*pause |= 0x10;
3513 
3514 	if (val & LPA_DUPLEX)
3515 		*fd = 1;
3516 	if (val & LPA_100)
3517 		*spd = 100;
3518 
3519 	if (cp->cas_flags & CAS_FLAG_1000MB_CAP) {
3520 		val = cas_phy_read(cp, CAS_MII_1000_STATUS);
3521 		if (val & (CAS_LPA_1000FULL | CAS_LPA_1000HALF))
3522 			*spd = 1000;
3523 		if (val & CAS_LPA_1000FULL)
3524 			*fd = 1;
3525 	}
3526 }
3527 
3528 /* A link-up condition has occurred, initialize and enable the
3529  * rest of the chip.
3530  *
3531  * Must be invoked under cp->lock.
3532  */
3533 static void cas_set_link_modes(struct cas *cp)
3534 {
3535 	u32 val;
3536 	int full_duplex, speed, pause;
3537 
3538 	full_duplex = 0;
3539 	speed = 10;
3540 	pause = 0;
3541 
3542 	if (CAS_PHY_MII(cp->phy_type)) {
3543 		cas_mif_poll(cp, 0);
3544 		val = cas_phy_read(cp, MII_BMCR);
3545 		if (val & BMCR_ANENABLE) {
3546 			cas_read_mii_link_mode(cp, &full_duplex, &speed,
3547 					       &pause);
3548 		} else {
3549 			if (val & BMCR_FULLDPLX)
3550 				full_duplex = 1;
3551 
3552 			if (val & BMCR_SPEED100)
3553 				speed = 100;
3554 			else if (val & CAS_BMCR_SPEED1000)
3555 				speed = (cp->cas_flags & CAS_FLAG_1000MB_CAP) ?
3556 					1000 : 100;
3557 		}
3558 		cas_mif_poll(cp, 1);
3559 
3560 	} else {
3561 		val = readl(cp->regs + REG_PCS_MII_CTRL);
3562 		cas_read_pcs_link_mode(cp, &full_duplex, &speed, &pause);
3563 		if ((val & PCS_MII_AUTONEG_EN) == 0) {
3564 			if (val & PCS_MII_CTRL_DUPLEX)
3565 				full_duplex = 1;
3566 		}
3567 	}
3568 
3569 	netif_info(cp, link, cp->dev, "Link up at %d Mbps, %s-duplex\n",
3570 		   speed, full_duplex ? "full" : "half");
3571 
3572 	val = MAC_XIF_TX_MII_OUTPUT_EN | MAC_XIF_LINK_LED;
3573 	if (CAS_PHY_MII(cp->phy_type)) {
3574 		val |= MAC_XIF_MII_BUFFER_OUTPUT_EN;
3575 		if (!full_duplex)
3576 			val |= MAC_XIF_DISABLE_ECHO;
3577 	}
3578 	if (full_duplex)
3579 		val |= MAC_XIF_FDPLX_LED;
3580 	if (speed == 1000)
3581 		val |= MAC_XIF_GMII_MODE;
3582 	writel(val, cp->regs + REG_MAC_XIF_CFG);
3583 
3584 	/* deal with carrier and collision detect. */
3585 	val = MAC_TX_CFG_IPG_EN;
3586 	if (full_duplex) {
3587 		val |= MAC_TX_CFG_IGNORE_CARRIER;
3588 		val |= MAC_TX_CFG_IGNORE_COLL;
3589 	} else {
3590 #ifndef USE_CSMA_CD_PROTO
3591 		val |= MAC_TX_CFG_NEVER_GIVE_UP_EN;
3592 		val |= MAC_TX_CFG_NEVER_GIVE_UP_LIM;
3593 #endif
3594 	}
3595 	/* val now set up for REG_MAC_TX_CFG */
3596 
3597 	/* If gigabit and half-duplex, enable carrier extension
3598 	 * mode.  increase slot time to 512 bytes as well.
3599 	 * else, disable it and make sure slot time is 64 bytes.
3600 	 * also activate checksum bug workaround
3601 	 */
3602 	if ((speed == 1000) && !full_duplex) {
3603 		writel(val | MAC_TX_CFG_CARRIER_EXTEND,
3604 		       cp->regs + REG_MAC_TX_CFG);
3605 
3606 		val = readl(cp->regs + REG_MAC_RX_CFG);
3607 		val &= ~MAC_RX_CFG_STRIP_FCS; /* checksum workaround */
3608 		writel(val | MAC_RX_CFG_CARRIER_EXTEND,
3609 		       cp->regs + REG_MAC_RX_CFG);
3610 
3611 		writel(0x200, cp->regs + REG_MAC_SLOT_TIME);
3612 
3613 		cp->crc_size = 4;
3614 		/* minimum size gigabit frame at half duplex */
3615 		cp->min_frame_size = CAS_1000MB_MIN_FRAME;
3616 
3617 	} else {
3618 		writel(val, cp->regs + REG_MAC_TX_CFG);
3619 
3620 		/* checksum bug workaround. don't strip FCS when in
3621 		 * half-duplex mode
3622 		 */
3623 		val = readl(cp->regs + REG_MAC_RX_CFG);
3624 		if (full_duplex) {
3625 			val |= MAC_RX_CFG_STRIP_FCS;
3626 			cp->crc_size = 0;
3627 			cp->min_frame_size = CAS_MIN_MTU;
3628 		} else {
3629 			val &= ~MAC_RX_CFG_STRIP_FCS;
3630 			cp->crc_size = 4;
3631 			cp->min_frame_size = CAS_MIN_FRAME;
3632 		}
3633 		writel(val & ~MAC_RX_CFG_CARRIER_EXTEND,
3634 		       cp->regs + REG_MAC_RX_CFG);
3635 		writel(0x40, cp->regs + REG_MAC_SLOT_TIME);
3636 	}
3637 
3638 	if (netif_msg_link(cp)) {
3639 		if (pause & 0x01) {
3640 			netdev_info(cp->dev, "Pause is enabled (rxfifo: %d off: %d on: %d)\n",
3641 				    cp->rx_fifo_size,
3642 				    cp->rx_pause_off,
3643 				    cp->rx_pause_on);
3644 		} else if (pause & 0x10) {
3645 			netdev_info(cp->dev, "TX pause enabled\n");
3646 		} else {
3647 			netdev_info(cp->dev, "Pause is disabled\n");
3648 		}
3649 	}
3650 
3651 	val = readl(cp->regs + REG_MAC_CTRL_CFG);
3652 	val &= ~(MAC_CTRL_CFG_SEND_PAUSE_EN | MAC_CTRL_CFG_RECV_PAUSE_EN);
3653 	if (pause) { /* symmetric or asymmetric pause */
3654 		val |= MAC_CTRL_CFG_SEND_PAUSE_EN;
3655 		if (pause & 0x01) { /* symmetric pause */
3656 			val |= MAC_CTRL_CFG_RECV_PAUSE_EN;
3657 		}
3658 	}
3659 	writel(val, cp->regs + REG_MAC_CTRL_CFG);
3660 	cas_start_dma(cp);
3661 }
3662 
3663 /* Must be invoked under cp->lock. */
3664 static void cas_init_hw(struct cas *cp, int restart_link)
3665 {
3666 	if (restart_link)
3667 		cas_phy_init(cp);
3668 
3669 	cas_init_pause_thresholds(cp);
3670 	cas_init_mac(cp);
3671 	cas_init_dma(cp);
3672 
3673 	if (restart_link) {
3674 		/* Default aneg parameters */
3675 		cp->timer_ticks = 0;
3676 		cas_begin_auto_negotiation(cp, NULL);
3677 	} else if (cp->lstate == link_up) {
3678 		cas_set_link_modes(cp);
3679 		netif_carrier_on(cp->dev);
3680 	}
3681 }
3682 
3683 /* Must be invoked under cp->lock. on earlier cassini boards,
3684  * SOFT_0 is tied to PCI reset. we use this to force a pci reset,
3685  * let it settle out, and then restore pci state.
3686  */
3687 static void cas_hard_reset(struct cas *cp)
3688 {
3689 	writel(BIM_LOCAL_DEV_SOFT_0, cp->regs + REG_BIM_LOCAL_DEV_EN);
3690 	udelay(20);
3691 	pci_restore_state(cp->pdev);
3692 }
3693 
3694 
3695 static void cas_global_reset(struct cas *cp, int blkflag)
3696 {
3697 	int limit;
3698 
3699 	/* issue a global reset. don't use RSTOUT. */
3700 	if (blkflag && !CAS_PHY_MII(cp->phy_type)) {
3701 		/* For PCS, when the blkflag is set, we should set the
3702 		 * SW_REST_BLOCK_PCS_SLINK bit to prevent the results of
3703 		 * the last autonegotiation from being cleared.  We'll
3704 		 * need some special handling if the chip is set into a
3705 		 * loopback mode.
3706 		 */
3707 		writel((SW_RESET_TX | SW_RESET_RX | SW_RESET_BLOCK_PCS_SLINK),
3708 		       cp->regs + REG_SW_RESET);
3709 	} else {
3710 		writel(SW_RESET_TX | SW_RESET_RX, cp->regs + REG_SW_RESET);
3711 	}
3712 
3713 	/* need to wait at least 3ms before polling register */
3714 	mdelay(3);
3715 
3716 	limit = STOP_TRIES;
3717 	while (limit-- > 0) {
3718 		u32 val = readl(cp->regs + REG_SW_RESET);
3719 		if ((val & (SW_RESET_TX | SW_RESET_RX)) == 0)
3720 			goto done;
3721 		udelay(10);
3722 	}
3723 	netdev_err(cp->dev, "sw reset failed\n");
3724 
3725 done:
3726 	/* enable various BIM interrupts */
3727 	writel(BIM_CFG_DPAR_INTR_ENABLE | BIM_CFG_RMA_INTR_ENABLE |
3728 	       BIM_CFG_RTA_INTR_ENABLE, cp->regs + REG_BIM_CFG);
3729 
3730 	/* clear out pci error status mask for handled errors.
3731 	 * we don't deal with DMA counter overflows as they happen
3732 	 * all the time.
3733 	 */
3734 	writel(0xFFFFFFFFU & ~(PCI_ERR_BADACK | PCI_ERR_DTRTO |
3735 			       PCI_ERR_OTHER | PCI_ERR_BIM_DMA_WRITE |
3736 			       PCI_ERR_BIM_DMA_READ), cp->regs +
3737 	       REG_PCI_ERR_STATUS_MASK);
3738 
3739 	/* set up for MII by default to address mac rx reset timeout
3740 	 * issue
3741 	 */
3742 	writel(PCS_DATAPATH_MODE_MII, cp->regs + REG_PCS_DATAPATH_MODE);
3743 }
3744 
3745 static void cas_reset(struct cas *cp, int blkflag)
3746 {
3747 	u32 val;
3748 
3749 	cas_mask_intr(cp);
3750 	cas_global_reset(cp, blkflag);
3751 	cas_mac_reset(cp);
3752 	cas_entropy_reset(cp);
3753 
3754 	/* disable dma engines. */
3755 	val = readl(cp->regs + REG_TX_CFG);
3756 	val &= ~TX_CFG_DMA_EN;
3757 	writel(val, cp->regs + REG_TX_CFG);
3758 
3759 	val = readl(cp->regs + REG_RX_CFG);
3760 	val &= ~RX_CFG_DMA_EN;
3761 	writel(val, cp->regs + REG_RX_CFG);
3762 
3763 	/* program header parser */
3764 	if ((cp->cas_flags & CAS_FLAG_TARGET_ABORT) ||
3765 	    (&CAS_HP_ALT_FIRMWARE[0] == &cas_prog_null[0])) {
3766 		cas_load_firmware(cp, CAS_HP_FIRMWARE);
3767 	} else {
3768 		cas_load_firmware(cp, CAS_HP_ALT_FIRMWARE);
3769 	}
3770 
3771 	/* clear out error registers */
3772 	spin_lock(&cp->stat_lock[N_TX_RINGS]);
3773 	cas_clear_mac_err(cp);
3774 	spin_unlock(&cp->stat_lock[N_TX_RINGS]);
3775 }
3776 
3777 /* Shut down the chip, must be called with pm_mutex held.  */
3778 static void cas_shutdown(struct cas *cp)
3779 {
3780 	unsigned long flags;
3781 
3782 	/* Make us not-running to avoid timers respawning */
3783 	cp->hw_running = 0;
3784 
3785 	del_timer_sync(&cp->link_timer);
3786 
3787 	/* Stop the reset task */
3788 #if 0
3789 	while (atomic_read(&cp->reset_task_pending_mtu) ||
3790 	       atomic_read(&cp->reset_task_pending_spare) ||
3791 	       atomic_read(&cp->reset_task_pending_all))
3792 		schedule();
3793 
3794 #else
3795 	while (atomic_read(&cp->reset_task_pending))
3796 		schedule();
3797 #endif
3798 	/* Actually stop the chip */
3799 	cas_lock_all_save(cp, flags);
3800 	cas_reset(cp, 0);
3801 	if (cp->cas_flags & CAS_FLAG_SATURN)
3802 		cas_phy_powerdown(cp);
3803 	cas_unlock_all_restore(cp, flags);
3804 }
3805 
3806 static int cas_change_mtu(struct net_device *dev, int new_mtu)
3807 {
3808 	struct cas *cp = netdev_priv(dev);
3809 
3810 	dev->mtu = new_mtu;
3811 	if (!netif_running(dev) || !netif_device_present(dev))
3812 		return 0;
3813 
3814 	/* let the reset task handle it */
3815 #if 1
3816 	atomic_inc(&cp->reset_task_pending);
3817 	if ((cp->phy_type & CAS_PHY_SERDES)) {
3818 		atomic_inc(&cp->reset_task_pending_all);
3819 	} else {
3820 		atomic_inc(&cp->reset_task_pending_mtu);
3821 	}
3822 	schedule_work(&cp->reset_task);
3823 #else
3824 	atomic_set(&cp->reset_task_pending, (cp->phy_type & CAS_PHY_SERDES) ?
3825 		   CAS_RESET_ALL : CAS_RESET_MTU);
3826 	pr_err("reset called in cas_change_mtu\n");
3827 	schedule_work(&cp->reset_task);
3828 #endif
3829 
3830 	flush_work(&cp->reset_task);
3831 	return 0;
3832 }
3833 
3834 static void cas_clean_txd(struct cas *cp, int ring)
3835 {
3836 	struct cas_tx_desc *txd = cp->init_txds[ring];
3837 	struct sk_buff *skb, **skbs = cp->tx_skbs[ring];
3838 	u64 daddr, dlen;
3839 	int i, size;
3840 
3841 	size = TX_DESC_RINGN_SIZE(ring);
3842 	for (i = 0; i < size; i++) {
3843 		int frag;
3844 
3845 		if (skbs[i] == NULL)
3846 			continue;
3847 
3848 		skb = skbs[i];
3849 		skbs[i] = NULL;
3850 
3851 		for (frag = 0; frag <= skb_shinfo(skb)->nr_frags;  frag++) {
3852 			int ent = i & (size - 1);
3853 
3854 			/* first buffer is never a tiny buffer and so
3855 			 * needs to be unmapped.
3856 			 */
3857 			daddr = le64_to_cpu(txd[ent].buffer);
3858 			dlen  =  CAS_VAL(TX_DESC_BUFLEN,
3859 					 le64_to_cpu(txd[ent].control));
3860 			dma_unmap_page(&cp->pdev->dev, daddr, dlen,
3861 				       DMA_TO_DEVICE);
3862 
3863 			if (frag != skb_shinfo(skb)->nr_frags) {
3864 				i++;
3865 
3866 				/* next buffer might by a tiny buffer.
3867 				 * skip past it.
3868 				 */
3869 				ent = i & (size - 1);
3870 				if (cp->tx_tiny_use[ring][ent].used)
3871 					i++;
3872 			}
3873 		}
3874 		dev_kfree_skb_any(skb);
3875 	}
3876 
3877 	/* zero out tiny buf usage */
3878 	memset(cp->tx_tiny_use[ring], 0, size*sizeof(*cp->tx_tiny_use[ring]));
3879 }
3880 
3881 /* freed on close */
3882 static inline void cas_free_rx_desc(struct cas *cp, int ring)
3883 {
3884 	cas_page_t **page = cp->rx_pages[ring];
3885 	int i, size;
3886 
3887 	size = RX_DESC_RINGN_SIZE(ring);
3888 	for (i = 0; i < size; i++) {
3889 		if (page[i]) {
3890 			cas_page_free(cp, page[i]);
3891 			page[i] = NULL;
3892 		}
3893 	}
3894 }
3895 
3896 static void cas_free_rxds(struct cas *cp)
3897 {
3898 	int i;
3899 
3900 	for (i = 0; i < N_RX_DESC_RINGS; i++)
3901 		cas_free_rx_desc(cp, i);
3902 }
3903 
3904 /* Must be invoked under cp->lock. */
3905 static void cas_clean_rings(struct cas *cp)
3906 {
3907 	int i;
3908 
3909 	/* need to clean all tx rings */
3910 	memset(cp->tx_old, 0, sizeof(*cp->tx_old)*N_TX_RINGS);
3911 	memset(cp->tx_new, 0, sizeof(*cp->tx_new)*N_TX_RINGS);
3912 	for (i = 0; i < N_TX_RINGS; i++)
3913 		cas_clean_txd(cp, i);
3914 
3915 	/* zero out init block */
3916 	memset(cp->init_block, 0, sizeof(struct cas_init_block));
3917 	cas_clean_rxds(cp);
3918 	cas_clean_rxcs(cp);
3919 }
3920 
3921 /* allocated on open */
3922 static inline int cas_alloc_rx_desc(struct cas *cp, int ring)
3923 {
3924 	cas_page_t **page = cp->rx_pages[ring];
3925 	int size, i = 0;
3926 
3927 	size = RX_DESC_RINGN_SIZE(ring);
3928 	for (i = 0; i < size; i++) {
3929 		if ((page[i] = cas_page_alloc(cp, GFP_KERNEL)) == NULL)
3930 			return -1;
3931 	}
3932 	return 0;
3933 }
3934 
3935 static int cas_alloc_rxds(struct cas *cp)
3936 {
3937 	int i;
3938 
3939 	for (i = 0; i < N_RX_DESC_RINGS; i++) {
3940 		if (cas_alloc_rx_desc(cp, i) < 0) {
3941 			cas_free_rxds(cp);
3942 			return -1;
3943 		}
3944 	}
3945 	return 0;
3946 }
3947 
3948 static void cas_reset_task(struct work_struct *work)
3949 {
3950 	struct cas *cp = container_of(work, struct cas, reset_task);
3951 #if 0
3952 	int pending = atomic_read(&cp->reset_task_pending);
3953 #else
3954 	int pending_all = atomic_read(&cp->reset_task_pending_all);
3955 	int pending_spare = atomic_read(&cp->reset_task_pending_spare);
3956 	int pending_mtu = atomic_read(&cp->reset_task_pending_mtu);
3957 
3958 	if (pending_all == 0 && pending_spare == 0 && pending_mtu == 0) {
3959 		/* We can have more tasks scheduled than actually
3960 		 * needed.
3961 		 */
3962 		atomic_dec(&cp->reset_task_pending);
3963 		return;
3964 	}
3965 #endif
3966 	/* The link went down, we reset the ring, but keep
3967 	 * DMA stopped. Use this function for reset
3968 	 * on error as well.
3969 	 */
3970 	if (cp->hw_running) {
3971 		unsigned long flags;
3972 
3973 		/* Make sure we don't get interrupts or tx packets */
3974 		netif_device_detach(cp->dev);
3975 		cas_lock_all_save(cp, flags);
3976 
3977 		if (cp->opened) {
3978 			/* We call cas_spare_recover when we call cas_open.
3979 			 * but we do not initialize the lists cas_spare_recover
3980 			 * uses until cas_open is called.
3981 			 */
3982 			cas_spare_recover(cp, GFP_ATOMIC);
3983 		}
3984 #if 1
3985 		/* test => only pending_spare set */
3986 		if (!pending_all && !pending_mtu)
3987 			goto done;
3988 #else
3989 		if (pending == CAS_RESET_SPARE)
3990 			goto done;
3991 #endif
3992 		/* when pending == CAS_RESET_ALL, the following
3993 		 * call to cas_init_hw will restart auto negotiation.
3994 		 * Setting the second argument of cas_reset to
3995 		 * !(pending == CAS_RESET_ALL) will set this argument
3996 		 * to 1 (avoiding reinitializing the PHY for the normal
3997 		 * PCS case) when auto negotiation is not restarted.
3998 		 */
3999 #if 1
4000 		cas_reset(cp, !(pending_all > 0));
4001 		if (cp->opened)
4002 			cas_clean_rings(cp);
4003 		cas_init_hw(cp, (pending_all > 0));
4004 #else
4005 		cas_reset(cp, !(pending == CAS_RESET_ALL));
4006 		if (cp->opened)
4007 			cas_clean_rings(cp);
4008 		cas_init_hw(cp, pending == CAS_RESET_ALL);
4009 #endif
4010 
4011 done:
4012 		cas_unlock_all_restore(cp, flags);
4013 		netif_device_attach(cp->dev);
4014 	}
4015 #if 1
4016 	atomic_sub(pending_all, &cp->reset_task_pending_all);
4017 	atomic_sub(pending_spare, &cp->reset_task_pending_spare);
4018 	atomic_sub(pending_mtu, &cp->reset_task_pending_mtu);
4019 	atomic_dec(&cp->reset_task_pending);
4020 #else
4021 	atomic_set(&cp->reset_task_pending, 0);
4022 #endif
4023 }
4024 
4025 static void cas_link_timer(struct timer_list *t)
4026 {
4027 	struct cas *cp = from_timer(cp, t, link_timer);
4028 	int mask, pending = 0, reset = 0;
4029 	unsigned long flags;
4030 
4031 	if (link_transition_timeout != 0 &&
4032 	    cp->link_transition_jiffies_valid &&
4033 	    time_is_before_jiffies(cp->link_transition_jiffies +
4034 	      link_transition_timeout)) {
4035 		/* One-second counter so link-down workaround doesn't
4036 		 * cause resets to occur so fast as to fool the switch
4037 		 * into thinking the link is down.
4038 		 */
4039 		cp->link_transition_jiffies_valid = 0;
4040 	}
4041 
4042 	if (!cp->hw_running)
4043 		return;
4044 
4045 	spin_lock_irqsave(&cp->lock, flags);
4046 	cas_lock_tx(cp);
4047 	cas_entropy_gather(cp);
4048 
4049 	/* If the link task is still pending, we just
4050 	 * reschedule the link timer
4051 	 */
4052 #if 1
4053 	if (atomic_read(&cp->reset_task_pending_all) ||
4054 	    atomic_read(&cp->reset_task_pending_spare) ||
4055 	    atomic_read(&cp->reset_task_pending_mtu))
4056 		goto done;
4057 #else
4058 	if (atomic_read(&cp->reset_task_pending))
4059 		goto done;
4060 #endif
4061 
4062 	/* check for rx cleaning */
4063 	if ((mask = (cp->cas_flags & CAS_FLAG_RXD_POST_MASK))) {
4064 		int i, rmask;
4065 
4066 		for (i = 0; i < MAX_RX_DESC_RINGS; i++) {
4067 			rmask = CAS_FLAG_RXD_POST(i);
4068 			if ((mask & rmask) == 0)
4069 				continue;
4070 
4071 			/* post_rxds will do a mod_timer */
4072 			if (cas_post_rxds_ringN(cp, i, cp->rx_last[i]) < 0) {
4073 				pending = 1;
4074 				continue;
4075 			}
4076 			cp->cas_flags &= ~rmask;
4077 		}
4078 	}
4079 
4080 	if (CAS_PHY_MII(cp->phy_type)) {
4081 		u16 bmsr;
4082 		cas_mif_poll(cp, 0);
4083 		bmsr = cas_phy_read(cp, MII_BMSR);
4084 		/* WTZ: Solaris driver reads this twice, but that
4085 		 * may be due to the PCS case and the use of a
4086 		 * common implementation. Read it twice here to be
4087 		 * safe.
4088 		 */
4089 		bmsr = cas_phy_read(cp, MII_BMSR);
4090 		cas_mif_poll(cp, 1);
4091 		readl(cp->regs + REG_MIF_STATUS); /* avoid dups */
4092 		reset = cas_mii_link_check(cp, bmsr);
4093 	} else {
4094 		reset = cas_pcs_link_check(cp);
4095 	}
4096 
4097 	if (reset)
4098 		goto done;
4099 
4100 	/* check for tx state machine confusion */
4101 	if ((readl(cp->regs + REG_MAC_TX_STATUS) & MAC_TX_FRAME_XMIT) == 0) {
4102 		u32 val = readl(cp->regs + REG_MAC_STATE_MACHINE);
4103 		u32 wptr, rptr;
4104 		int tlm  = CAS_VAL(MAC_SM_TLM, val);
4105 
4106 		if (((tlm == 0x5) || (tlm == 0x3)) &&
4107 		    (CAS_VAL(MAC_SM_ENCAP_SM, val) == 0)) {
4108 			netif_printk(cp, tx_err, KERN_DEBUG, cp->dev,
4109 				     "tx err: MAC_STATE[%08x]\n", val);
4110 			reset = 1;
4111 			goto done;
4112 		}
4113 
4114 		val  = readl(cp->regs + REG_TX_FIFO_PKT_CNT);
4115 		wptr = readl(cp->regs + REG_TX_FIFO_WRITE_PTR);
4116 		rptr = readl(cp->regs + REG_TX_FIFO_READ_PTR);
4117 		if ((val == 0) && (wptr != rptr)) {
4118 			netif_printk(cp, tx_err, KERN_DEBUG, cp->dev,
4119 				     "tx err: TX_FIFO[%08x:%08x:%08x]\n",
4120 				     val, wptr, rptr);
4121 			reset = 1;
4122 		}
4123 
4124 		if (reset)
4125 			cas_hard_reset(cp);
4126 	}
4127 
4128 done:
4129 	if (reset) {
4130 #if 1
4131 		atomic_inc(&cp->reset_task_pending);
4132 		atomic_inc(&cp->reset_task_pending_all);
4133 		schedule_work(&cp->reset_task);
4134 #else
4135 		atomic_set(&cp->reset_task_pending, CAS_RESET_ALL);
4136 		pr_err("reset called in cas_link_timer\n");
4137 		schedule_work(&cp->reset_task);
4138 #endif
4139 	}
4140 
4141 	if (!pending)
4142 		mod_timer(&cp->link_timer, jiffies + CAS_LINK_TIMEOUT);
4143 	cas_unlock_tx(cp);
4144 	spin_unlock_irqrestore(&cp->lock, flags);
4145 }
4146 
4147 /* tiny buffers are used to avoid target abort issues with
4148  * older cassini's
4149  */
4150 static void cas_tx_tiny_free(struct cas *cp)
4151 {
4152 	struct pci_dev *pdev = cp->pdev;
4153 	int i;
4154 
4155 	for (i = 0; i < N_TX_RINGS; i++) {
4156 		if (!cp->tx_tiny_bufs[i])
4157 			continue;
4158 
4159 		dma_free_coherent(&pdev->dev, TX_TINY_BUF_BLOCK,
4160 				  cp->tx_tiny_bufs[i], cp->tx_tiny_dvma[i]);
4161 		cp->tx_tiny_bufs[i] = NULL;
4162 	}
4163 }
4164 
4165 static int cas_tx_tiny_alloc(struct cas *cp)
4166 {
4167 	struct pci_dev *pdev = cp->pdev;
4168 	int i;
4169 
4170 	for (i = 0; i < N_TX_RINGS; i++) {
4171 		cp->tx_tiny_bufs[i] =
4172 			dma_alloc_coherent(&pdev->dev, TX_TINY_BUF_BLOCK,
4173 					   &cp->tx_tiny_dvma[i], GFP_KERNEL);
4174 		if (!cp->tx_tiny_bufs[i]) {
4175 			cas_tx_tiny_free(cp);
4176 			return -1;
4177 		}
4178 	}
4179 	return 0;
4180 }
4181 
4182 
4183 static int cas_open(struct net_device *dev)
4184 {
4185 	struct cas *cp = netdev_priv(dev);
4186 	int hw_was_up, err;
4187 	unsigned long flags;
4188 
4189 	mutex_lock(&cp->pm_mutex);
4190 
4191 	hw_was_up = cp->hw_running;
4192 
4193 	/* The power-management mutex protects the hw_running
4194 	 * etc. state so it is safe to do this bit without cp->lock
4195 	 */
4196 	if (!cp->hw_running) {
4197 		/* Reset the chip */
4198 		cas_lock_all_save(cp, flags);
4199 		/* We set the second arg to cas_reset to zero
4200 		 * because cas_init_hw below will have its second
4201 		 * argument set to non-zero, which will force
4202 		 * autonegotiation to start.
4203 		 */
4204 		cas_reset(cp, 0);
4205 		cp->hw_running = 1;
4206 		cas_unlock_all_restore(cp, flags);
4207 	}
4208 
4209 	err = -ENOMEM;
4210 	if (cas_tx_tiny_alloc(cp) < 0)
4211 		goto err_unlock;
4212 
4213 	/* alloc rx descriptors */
4214 	if (cas_alloc_rxds(cp) < 0)
4215 		goto err_tx_tiny;
4216 
4217 	/* allocate spares */
4218 	cas_spare_init(cp);
4219 	cas_spare_recover(cp, GFP_KERNEL);
4220 
4221 	/* We can now request the interrupt as we know it's masked
4222 	 * on the controller. cassini+ has up to 4 interrupts
4223 	 * that can be used, but you need to do explicit pci interrupt
4224 	 * mapping to expose them
4225 	 */
4226 	if (request_irq(cp->pdev->irq, cas_interrupt,
4227 			IRQF_SHARED, dev->name, (void *) dev)) {
4228 		netdev_err(cp->dev, "failed to request irq !\n");
4229 		err = -EAGAIN;
4230 		goto err_spare;
4231 	}
4232 
4233 #ifdef USE_NAPI
4234 	napi_enable(&cp->napi);
4235 #endif
4236 	/* init hw */
4237 	cas_lock_all_save(cp, flags);
4238 	cas_clean_rings(cp);
4239 	cas_init_hw(cp, !hw_was_up);
4240 	cp->opened = 1;
4241 	cas_unlock_all_restore(cp, flags);
4242 
4243 	netif_start_queue(dev);
4244 	mutex_unlock(&cp->pm_mutex);
4245 	return 0;
4246 
4247 err_spare:
4248 	cas_spare_free(cp);
4249 	cas_free_rxds(cp);
4250 err_tx_tiny:
4251 	cas_tx_tiny_free(cp);
4252 err_unlock:
4253 	mutex_unlock(&cp->pm_mutex);
4254 	return err;
4255 }
4256 
4257 static int cas_close(struct net_device *dev)
4258 {
4259 	unsigned long flags;
4260 	struct cas *cp = netdev_priv(dev);
4261 
4262 #ifdef USE_NAPI
4263 	napi_disable(&cp->napi);
4264 #endif
4265 	/* Make sure we don't get distracted by suspend/resume */
4266 	mutex_lock(&cp->pm_mutex);
4267 
4268 	netif_stop_queue(dev);
4269 
4270 	/* Stop traffic, mark us closed */
4271 	cas_lock_all_save(cp, flags);
4272 	cp->opened = 0;
4273 	cas_reset(cp, 0);
4274 	cas_phy_init(cp);
4275 	cas_begin_auto_negotiation(cp, NULL);
4276 	cas_clean_rings(cp);
4277 	cas_unlock_all_restore(cp, flags);
4278 
4279 	free_irq(cp->pdev->irq, (void *) dev);
4280 	cas_spare_free(cp);
4281 	cas_free_rxds(cp);
4282 	cas_tx_tiny_free(cp);
4283 	mutex_unlock(&cp->pm_mutex);
4284 	return 0;
4285 }
4286 
4287 static struct {
4288 	const char name[ETH_GSTRING_LEN];
4289 } ethtool_cassini_statnames[] = {
4290 	{"collisions"},
4291 	{"rx_bytes"},
4292 	{"rx_crc_errors"},
4293 	{"rx_dropped"},
4294 	{"rx_errors"},
4295 	{"rx_fifo_errors"},
4296 	{"rx_frame_errors"},
4297 	{"rx_length_errors"},
4298 	{"rx_over_errors"},
4299 	{"rx_packets"},
4300 	{"tx_aborted_errors"},
4301 	{"tx_bytes"},
4302 	{"tx_dropped"},
4303 	{"tx_errors"},
4304 	{"tx_fifo_errors"},
4305 	{"tx_packets"}
4306 };
4307 #define CAS_NUM_STAT_KEYS ARRAY_SIZE(ethtool_cassini_statnames)
4308 
4309 static struct {
4310 	const int offsets;	/* neg. values for 2nd arg to cas_read_phy */
4311 } ethtool_register_table[] = {
4312 	{-MII_BMSR},
4313 	{-MII_BMCR},
4314 	{REG_CAWR},
4315 	{REG_INF_BURST},
4316 	{REG_BIM_CFG},
4317 	{REG_RX_CFG},
4318 	{REG_HP_CFG},
4319 	{REG_MAC_TX_CFG},
4320 	{REG_MAC_RX_CFG},
4321 	{REG_MAC_CTRL_CFG},
4322 	{REG_MAC_XIF_CFG},
4323 	{REG_MIF_CFG},
4324 	{REG_PCS_CFG},
4325 	{REG_SATURN_PCFG},
4326 	{REG_PCS_MII_STATUS},
4327 	{REG_PCS_STATE_MACHINE},
4328 	{REG_MAC_COLL_EXCESS},
4329 	{REG_MAC_COLL_LATE}
4330 };
4331 #define CAS_REG_LEN 	ARRAY_SIZE(ethtool_register_table)
4332 #define CAS_MAX_REGS 	(sizeof (u32)*CAS_REG_LEN)
4333 
4334 static void cas_read_regs(struct cas *cp, u8 *ptr, int len)
4335 {
4336 	u8 *p;
4337 	int i;
4338 	unsigned long flags;
4339 
4340 	spin_lock_irqsave(&cp->lock, flags);
4341 	for (i = 0, p = ptr; i < len ; i ++, p += sizeof(u32)) {
4342 		u16 hval;
4343 		u32 val;
4344 		if (ethtool_register_table[i].offsets < 0) {
4345 			hval = cas_phy_read(cp,
4346 				    -ethtool_register_table[i].offsets);
4347 			val = hval;
4348 		} else {
4349 			val= readl(cp->regs+ethtool_register_table[i].offsets);
4350 		}
4351 		memcpy(p, (u8 *)&val, sizeof(u32));
4352 	}
4353 	spin_unlock_irqrestore(&cp->lock, flags);
4354 }
4355 
4356 static struct net_device_stats *cas_get_stats(struct net_device *dev)
4357 {
4358 	struct cas *cp = netdev_priv(dev);
4359 	struct net_device_stats *stats = cp->net_stats;
4360 	unsigned long flags;
4361 	int i;
4362 	unsigned long tmp;
4363 
4364 	/* we collate all of the stats into net_stats[N_TX_RING] */
4365 	if (!cp->hw_running)
4366 		return stats + N_TX_RINGS;
4367 
4368 	/* collect outstanding stats */
4369 	/* WTZ: the Cassini spec gives these as 16 bit counters but
4370 	 * stored in 32-bit words.  Added a mask of 0xffff to be safe,
4371 	 * in case the chip somehow puts any garbage in the other bits.
4372 	 * Also, counter usage didn't seem to mach what Adrian did
4373 	 * in the parts of the code that set these quantities. Made
4374 	 * that consistent.
4375 	 */
4376 	spin_lock_irqsave(&cp->stat_lock[N_TX_RINGS], flags);
4377 	stats[N_TX_RINGS].rx_crc_errors +=
4378 	  readl(cp->regs + REG_MAC_FCS_ERR) & 0xffff;
4379 	stats[N_TX_RINGS].rx_frame_errors +=
4380 		readl(cp->regs + REG_MAC_ALIGN_ERR) &0xffff;
4381 	stats[N_TX_RINGS].rx_length_errors +=
4382 		readl(cp->regs + REG_MAC_LEN_ERR) & 0xffff;
4383 #if 1
4384 	tmp = (readl(cp->regs + REG_MAC_COLL_EXCESS) & 0xffff) +
4385 		(readl(cp->regs + REG_MAC_COLL_LATE) & 0xffff);
4386 	stats[N_TX_RINGS].tx_aborted_errors += tmp;
4387 	stats[N_TX_RINGS].collisions +=
4388 	  tmp + (readl(cp->regs + REG_MAC_COLL_NORMAL) & 0xffff);
4389 #else
4390 	stats[N_TX_RINGS].tx_aborted_errors +=
4391 		readl(cp->regs + REG_MAC_COLL_EXCESS);
4392 	stats[N_TX_RINGS].collisions += readl(cp->regs + REG_MAC_COLL_EXCESS) +
4393 		readl(cp->regs + REG_MAC_COLL_LATE);
4394 #endif
4395 	cas_clear_mac_err(cp);
4396 
4397 	/* saved bits that are unique to ring 0 */
4398 	spin_lock(&cp->stat_lock[0]);
4399 	stats[N_TX_RINGS].collisions        += stats[0].collisions;
4400 	stats[N_TX_RINGS].rx_over_errors    += stats[0].rx_over_errors;
4401 	stats[N_TX_RINGS].rx_frame_errors   += stats[0].rx_frame_errors;
4402 	stats[N_TX_RINGS].rx_fifo_errors    += stats[0].rx_fifo_errors;
4403 	stats[N_TX_RINGS].tx_aborted_errors += stats[0].tx_aborted_errors;
4404 	stats[N_TX_RINGS].tx_fifo_errors    += stats[0].tx_fifo_errors;
4405 	spin_unlock(&cp->stat_lock[0]);
4406 
4407 	for (i = 0; i < N_TX_RINGS; i++) {
4408 		spin_lock(&cp->stat_lock[i]);
4409 		stats[N_TX_RINGS].rx_length_errors +=
4410 			stats[i].rx_length_errors;
4411 		stats[N_TX_RINGS].rx_crc_errors += stats[i].rx_crc_errors;
4412 		stats[N_TX_RINGS].rx_packets    += stats[i].rx_packets;
4413 		stats[N_TX_RINGS].tx_packets    += stats[i].tx_packets;
4414 		stats[N_TX_RINGS].rx_bytes      += stats[i].rx_bytes;
4415 		stats[N_TX_RINGS].tx_bytes      += stats[i].tx_bytes;
4416 		stats[N_TX_RINGS].rx_errors     += stats[i].rx_errors;
4417 		stats[N_TX_RINGS].tx_errors     += stats[i].tx_errors;
4418 		stats[N_TX_RINGS].rx_dropped    += stats[i].rx_dropped;
4419 		stats[N_TX_RINGS].tx_dropped    += stats[i].tx_dropped;
4420 		memset(stats + i, 0, sizeof(struct net_device_stats));
4421 		spin_unlock(&cp->stat_lock[i]);
4422 	}
4423 	spin_unlock_irqrestore(&cp->stat_lock[N_TX_RINGS], flags);
4424 	return stats + N_TX_RINGS;
4425 }
4426 
4427 
4428 static void cas_set_multicast(struct net_device *dev)
4429 {
4430 	struct cas *cp = netdev_priv(dev);
4431 	u32 rxcfg, rxcfg_new;
4432 	unsigned long flags;
4433 	int limit = STOP_TRIES;
4434 
4435 	if (!cp->hw_running)
4436 		return;
4437 
4438 	spin_lock_irqsave(&cp->lock, flags);
4439 	rxcfg = readl(cp->regs + REG_MAC_RX_CFG);
4440 
4441 	/* disable RX MAC and wait for completion */
4442 	writel(rxcfg & ~MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
4443 	while (readl(cp->regs + REG_MAC_RX_CFG) & MAC_RX_CFG_EN) {
4444 		if (!limit--)
4445 			break;
4446 		udelay(10);
4447 	}
4448 
4449 	/* disable hash filter and wait for completion */
4450 	limit = STOP_TRIES;
4451 	rxcfg &= ~(MAC_RX_CFG_PROMISC_EN | MAC_RX_CFG_HASH_FILTER_EN);
4452 	writel(rxcfg & ~MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
4453 	while (readl(cp->regs + REG_MAC_RX_CFG) & MAC_RX_CFG_HASH_FILTER_EN) {
4454 		if (!limit--)
4455 			break;
4456 		udelay(10);
4457 	}
4458 
4459 	/* program hash filters */
4460 	cp->mac_rx_cfg = rxcfg_new = cas_setup_multicast(cp);
4461 	rxcfg |= rxcfg_new;
4462 	writel(rxcfg, cp->regs + REG_MAC_RX_CFG);
4463 	spin_unlock_irqrestore(&cp->lock, flags);
4464 }
4465 
4466 static void cas_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
4467 {
4468 	struct cas *cp = netdev_priv(dev);
4469 	strscpy(info->driver, DRV_MODULE_NAME, sizeof(info->driver));
4470 	strscpy(info->version, DRV_MODULE_VERSION, sizeof(info->version));
4471 	strscpy(info->bus_info, pci_name(cp->pdev), sizeof(info->bus_info));
4472 }
4473 
4474 static int cas_get_link_ksettings(struct net_device *dev,
4475 				  struct ethtool_link_ksettings *cmd)
4476 {
4477 	struct cas *cp = netdev_priv(dev);
4478 	u16 bmcr;
4479 	int full_duplex, speed, pause;
4480 	unsigned long flags;
4481 	enum link_state linkstate = link_up;
4482 	u32 supported, advertising;
4483 
4484 	advertising = 0;
4485 	supported = SUPPORTED_Autoneg;
4486 	if (cp->cas_flags & CAS_FLAG_1000MB_CAP) {
4487 		supported |= SUPPORTED_1000baseT_Full;
4488 		advertising |= ADVERTISED_1000baseT_Full;
4489 	}
4490 
4491 	/* Record PHY settings if HW is on. */
4492 	spin_lock_irqsave(&cp->lock, flags);
4493 	bmcr = 0;
4494 	linkstate = cp->lstate;
4495 	if (CAS_PHY_MII(cp->phy_type)) {
4496 		cmd->base.port = PORT_MII;
4497 		cmd->base.phy_address = cp->phy_addr;
4498 		advertising |= ADVERTISED_TP | ADVERTISED_MII |
4499 			ADVERTISED_10baseT_Half |
4500 			ADVERTISED_10baseT_Full |
4501 			ADVERTISED_100baseT_Half |
4502 			ADVERTISED_100baseT_Full;
4503 
4504 		supported |=
4505 			(SUPPORTED_10baseT_Half |
4506 			 SUPPORTED_10baseT_Full |
4507 			 SUPPORTED_100baseT_Half |
4508 			 SUPPORTED_100baseT_Full |
4509 			 SUPPORTED_TP | SUPPORTED_MII);
4510 
4511 		if (cp->hw_running) {
4512 			cas_mif_poll(cp, 0);
4513 			bmcr = cas_phy_read(cp, MII_BMCR);
4514 			cas_read_mii_link_mode(cp, &full_duplex,
4515 					       &speed, &pause);
4516 			cas_mif_poll(cp, 1);
4517 		}
4518 
4519 	} else {
4520 		cmd->base.port = PORT_FIBRE;
4521 		cmd->base.phy_address = 0;
4522 		supported   |= SUPPORTED_FIBRE;
4523 		advertising |= ADVERTISED_FIBRE;
4524 
4525 		if (cp->hw_running) {
4526 			/* pcs uses the same bits as mii */
4527 			bmcr = readl(cp->regs + REG_PCS_MII_CTRL);
4528 			cas_read_pcs_link_mode(cp, &full_duplex,
4529 					       &speed, &pause);
4530 		}
4531 	}
4532 	spin_unlock_irqrestore(&cp->lock, flags);
4533 
4534 	if (bmcr & BMCR_ANENABLE) {
4535 		advertising |= ADVERTISED_Autoneg;
4536 		cmd->base.autoneg = AUTONEG_ENABLE;
4537 		cmd->base.speed =  ((speed == 10) ?
4538 					    SPEED_10 :
4539 					    ((speed == 1000) ?
4540 					     SPEED_1000 : SPEED_100));
4541 		cmd->base.duplex = full_duplex ? DUPLEX_FULL : DUPLEX_HALF;
4542 	} else {
4543 		cmd->base.autoneg = AUTONEG_DISABLE;
4544 		cmd->base.speed = ((bmcr & CAS_BMCR_SPEED1000) ?
4545 					    SPEED_1000 :
4546 					    ((bmcr & BMCR_SPEED100) ?
4547 					     SPEED_100 : SPEED_10));
4548 		cmd->base.duplex = (bmcr & BMCR_FULLDPLX) ?
4549 			DUPLEX_FULL : DUPLEX_HALF;
4550 	}
4551 	if (linkstate != link_up) {
4552 		/* Force these to "unknown" if the link is not up and
4553 		 * autonogotiation in enabled. We can set the link
4554 		 * speed to 0, but not cmd->duplex,
4555 		 * because its legal values are 0 and 1.  Ethtool will
4556 		 * print the value reported in parentheses after the
4557 		 * word "Unknown" for unrecognized values.
4558 		 *
4559 		 * If in forced mode, we report the speed and duplex
4560 		 * settings that we configured.
4561 		 */
4562 		if (cp->link_cntl & BMCR_ANENABLE) {
4563 			cmd->base.speed = 0;
4564 			cmd->base.duplex = 0xff;
4565 		} else {
4566 			cmd->base.speed = SPEED_10;
4567 			if (cp->link_cntl & BMCR_SPEED100) {
4568 				cmd->base.speed = SPEED_100;
4569 			} else if (cp->link_cntl & CAS_BMCR_SPEED1000) {
4570 				cmd->base.speed = SPEED_1000;
4571 			}
4572 			cmd->base.duplex = (cp->link_cntl & BMCR_FULLDPLX) ?
4573 				DUPLEX_FULL : DUPLEX_HALF;
4574 		}
4575 	}
4576 
4577 	ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.supported,
4578 						supported);
4579 	ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.advertising,
4580 						advertising);
4581 
4582 	return 0;
4583 }
4584 
4585 static int cas_set_link_ksettings(struct net_device *dev,
4586 				  const struct ethtool_link_ksettings *cmd)
4587 {
4588 	struct cas *cp = netdev_priv(dev);
4589 	unsigned long flags;
4590 	u32 speed = cmd->base.speed;
4591 
4592 	/* Verify the settings we care about. */
4593 	if (cmd->base.autoneg != AUTONEG_ENABLE &&
4594 	    cmd->base.autoneg != AUTONEG_DISABLE)
4595 		return -EINVAL;
4596 
4597 	if (cmd->base.autoneg == AUTONEG_DISABLE &&
4598 	    ((speed != SPEED_1000 &&
4599 	      speed != SPEED_100 &&
4600 	      speed != SPEED_10) ||
4601 	     (cmd->base.duplex != DUPLEX_HALF &&
4602 	      cmd->base.duplex != DUPLEX_FULL)))
4603 		return -EINVAL;
4604 
4605 	/* Apply settings and restart link process. */
4606 	spin_lock_irqsave(&cp->lock, flags);
4607 	cas_begin_auto_negotiation(cp, cmd);
4608 	spin_unlock_irqrestore(&cp->lock, flags);
4609 	return 0;
4610 }
4611 
4612 static int cas_nway_reset(struct net_device *dev)
4613 {
4614 	struct cas *cp = netdev_priv(dev);
4615 	unsigned long flags;
4616 
4617 	if ((cp->link_cntl & BMCR_ANENABLE) == 0)
4618 		return -EINVAL;
4619 
4620 	/* Restart link process. */
4621 	spin_lock_irqsave(&cp->lock, flags);
4622 	cas_begin_auto_negotiation(cp, NULL);
4623 	spin_unlock_irqrestore(&cp->lock, flags);
4624 
4625 	return 0;
4626 }
4627 
4628 static u32 cas_get_link(struct net_device *dev)
4629 {
4630 	struct cas *cp = netdev_priv(dev);
4631 	return cp->lstate == link_up;
4632 }
4633 
4634 static u32 cas_get_msglevel(struct net_device *dev)
4635 {
4636 	struct cas *cp = netdev_priv(dev);
4637 	return cp->msg_enable;
4638 }
4639 
4640 static void cas_set_msglevel(struct net_device *dev, u32 value)
4641 {
4642 	struct cas *cp = netdev_priv(dev);
4643 	cp->msg_enable = value;
4644 }
4645 
4646 static int cas_get_regs_len(struct net_device *dev)
4647 {
4648 	struct cas *cp = netdev_priv(dev);
4649 	return min_t(int, cp->casreg_len, CAS_MAX_REGS);
4650 }
4651 
4652 static void cas_get_regs(struct net_device *dev, struct ethtool_regs *regs,
4653 			     void *p)
4654 {
4655 	struct cas *cp = netdev_priv(dev);
4656 	regs->version = 0;
4657 	/* cas_read_regs handles locks (cp->lock).  */
4658 	cas_read_regs(cp, p, regs->len / sizeof(u32));
4659 }
4660 
4661 static int cas_get_sset_count(struct net_device *dev, int sset)
4662 {
4663 	switch (sset) {
4664 	case ETH_SS_STATS:
4665 		return CAS_NUM_STAT_KEYS;
4666 	default:
4667 		return -EOPNOTSUPP;
4668 	}
4669 }
4670 
4671 static void cas_get_strings(struct net_device *dev, u32 stringset, u8 *data)
4672 {
4673 	 memcpy(data, &ethtool_cassini_statnames,
4674 					 CAS_NUM_STAT_KEYS * ETH_GSTRING_LEN);
4675 }
4676 
4677 static void cas_get_ethtool_stats(struct net_device *dev,
4678 				      struct ethtool_stats *estats, u64 *data)
4679 {
4680 	struct cas *cp = netdev_priv(dev);
4681 	struct net_device_stats *stats = cas_get_stats(cp->dev);
4682 	int i = 0;
4683 	data[i++] = stats->collisions;
4684 	data[i++] = stats->rx_bytes;
4685 	data[i++] = stats->rx_crc_errors;
4686 	data[i++] = stats->rx_dropped;
4687 	data[i++] = stats->rx_errors;
4688 	data[i++] = stats->rx_fifo_errors;
4689 	data[i++] = stats->rx_frame_errors;
4690 	data[i++] = stats->rx_length_errors;
4691 	data[i++] = stats->rx_over_errors;
4692 	data[i++] = stats->rx_packets;
4693 	data[i++] = stats->tx_aborted_errors;
4694 	data[i++] = stats->tx_bytes;
4695 	data[i++] = stats->tx_dropped;
4696 	data[i++] = stats->tx_errors;
4697 	data[i++] = stats->tx_fifo_errors;
4698 	data[i++] = stats->tx_packets;
4699 	BUG_ON(i != CAS_NUM_STAT_KEYS);
4700 }
4701 
4702 static const struct ethtool_ops cas_ethtool_ops = {
4703 	.get_drvinfo		= cas_get_drvinfo,
4704 	.nway_reset		= cas_nway_reset,
4705 	.get_link		= cas_get_link,
4706 	.get_msglevel		= cas_get_msglevel,
4707 	.set_msglevel		= cas_set_msglevel,
4708 	.get_regs_len		= cas_get_regs_len,
4709 	.get_regs		= cas_get_regs,
4710 	.get_sset_count		= cas_get_sset_count,
4711 	.get_strings		= cas_get_strings,
4712 	.get_ethtool_stats	= cas_get_ethtool_stats,
4713 	.get_link_ksettings	= cas_get_link_ksettings,
4714 	.set_link_ksettings	= cas_set_link_ksettings,
4715 };
4716 
4717 static int cas_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
4718 {
4719 	struct cas *cp = netdev_priv(dev);
4720 	struct mii_ioctl_data *data = if_mii(ifr);
4721 	unsigned long flags;
4722 	int rc = -EOPNOTSUPP;
4723 
4724 	/* Hold the PM mutex while doing ioctl's or we may collide
4725 	 * with open/close and power management and oops.
4726 	 */
4727 	mutex_lock(&cp->pm_mutex);
4728 	switch (cmd) {
4729 	case SIOCGMIIPHY:		/* Get address of MII PHY in use. */
4730 		data->phy_id = cp->phy_addr;
4731 		fallthrough;
4732 
4733 	case SIOCGMIIREG:		/* Read MII PHY register. */
4734 		spin_lock_irqsave(&cp->lock, flags);
4735 		cas_mif_poll(cp, 0);
4736 		data->val_out = cas_phy_read(cp, data->reg_num & 0x1f);
4737 		cas_mif_poll(cp, 1);
4738 		spin_unlock_irqrestore(&cp->lock, flags);
4739 		rc = 0;
4740 		break;
4741 
4742 	case SIOCSMIIREG:		/* Write MII PHY register. */
4743 		spin_lock_irqsave(&cp->lock, flags);
4744 		cas_mif_poll(cp, 0);
4745 		rc = cas_phy_write(cp, data->reg_num & 0x1f, data->val_in);
4746 		cas_mif_poll(cp, 1);
4747 		spin_unlock_irqrestore(&cp->lock, flags);
4748 		break;
4749 	default:
4750 		break;
4751 	}
4752 
4753 	mutex_unlock(&cp->pm_mutex);
4754 	return rc;
4755 }
4756 
4757 /* When this chip sits underneath an Intel 31154 bridge, it is the
4758  * only subordinate device and we can tweak the bridge settings to
4759  * reflect that fact.
4760  */
4761 static void cas_program_bridge(struct pci_dev *cas_pdev)
4762 {
4763 	struct pci_dev *pdev = cas_pdev->bus->self;
4764 	u32 val;
4765 
4766 	if (!pdev)
4767 		return;
4768 
4769 	if (pdev->vendor != 0x8086 || pdev->device != 0x537c)
4770 		return;
4771 
4772 	/* Clear bit 10 (Bus Parking Control) in the Secondary
4773 	 * Arbiter Control/Status Register which lives at offset
4774 	 * 0x41.  Using a 32-bit word read/modify/write at 0x40
4775 	 * is much simpler so that's how we do this.
4776 	 */
4777 	pci_read_config_dword(pdev, 0x40, &val);
4778 	val &= ~0x00040000;
4779 	pci_write_config_dword(pdev, 0x40, val);
4780 
4781 	/* Max out the Multi-Transaction Timer settings since
4782 	 * Cassini is the only device present.
4783 	 *
4784 	 * The register is 16-bit and lives at 0x50.  When the
4785 	 * settings are enabled, it extends the GRANT# signal
4786 	 * for a requestor after a transaction is complete.  This
4787 	 * allows the next request to run without first needing
4788 	 * to negotiate the GRANT# signal back.
4789 	 *
4790 	 * Bits 12:10 define the grant duration:
4791 	 *
4792 	 *	1	--	16 clocks
4793 	 *	2	--	32 clocks
4794 	 *	3	--	64 clocks
4795 	 *	4	--	128 clocks
4796 	 *	5	--	256 clocks
4797 	 *
4798 	 * All other values are illegal.
4799 	 *
4800 	 * Bits 09:00 define which REQ/GNT signal pairs get the
4801 	 * GRANT# signal treatment.  We set them all.
4802 	 */
4803 	pci_write_config_word(pdev, 0x50, (5 << 10) | 0x3ff);
4804 
4805 	/* The Read Prefecth Policy register is 16-bit and sits at
4806 	 * offset 0x52.  It enables a "smart" pre-fetch policy.  We
4807 	 * enable it and max out all of the settings since only one
4808 	 * device is sitting underneath and thus bandwidth sharing is
4809 	 * not an issue.
4810 	 *
4811 	 * The register has several 3 bit fields, which indicates a
4812 	 * multiplier applied to the base amount of prefetching the
4813 	 * chip would do.  These fields are at:
4814 	 *
4815 	 *	15:13	---	ReRead Primary Bus
4816 	 *	12:10	---	FirstRead Primary Bus
4817 	 *	09:07	---	ReRead Secondary Bus
4818 	 *	06:04	---	FirstRead Secondary Bus
4819 	 *
4820 	 * Bits 03:00 control which REQ/GNT pairs the prefetch settings
4821 	 * get enabled on.  Bit 3 is a grouped enabler which controls
4822 	 * all of the REQ/GNT pairs from [8:3].  Bits 2 to 0 control
4823 	 * the individual REQ/GNT pairs [2:0].
4824 	 */
4825 	pci_write_config_word(pdev, 0x52,
4826 			      (0x7 << 13) |
4827 			      (0x7 << 10) |
4828 			      (0x7 <<  7) |
4829 			      (0x7 <<  4) |
4830 			      (0xf <<  0));
4831 
4832 	/* Force cacheline size to 0x8 */
4833 	pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE, 0x08);
4834 
4835 	/* Force latency timer to maximum setting so Cassini can
4836 	 * sit on the bus as long as it likes.
4837 	 */
4838 	pci_write_config_byte(pdev, PCI_LATENCY_TIMER, 0xff);
4839 }
4840 
4841 static const struct net_device_ops cas_netdev_ops = {
4842 	.ndo_open		= cas_open,
4843 	.ndo_stop		= cas_close,
4844 	.ndo_start_xmit		= cas_start_xmit,
4845 	.ndo_get_stats 		= cas_get_stats,
4846 	.ndo_set_rx_mode	= cas_set_multicast,
4847 	.ndo_eth_ioctl		= cas_ioctl,
4848 	.ndo_tx_timeout		= cas_tx_timeout,
4849 	.ndo_change_mtu		= cas_change_mtu,
4850 	.ndo_set_mac_address	= eth_mac_addr,
4851 	.ndo_validate_addr	= eth_validate_addr,
4852 #ifdef CONFIG_NET_POLL_CONTROLLER
4853 	.ndo_poll_controller	= cas_netpoll,
4854 #endif
4855 };
4856 
4857 static int cas_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
4858 {
4859 	static int cas_version_printed = 0;
4860 	unsigned long casreg_len;
4861 	struct net_device *dev;
4862 	struct cas *cp;
4863 	u16 pci_cmd;
4864 	int i, err;
4865 	u8 orig_cacheline_size = 0, cas_cacheline_size = 0;
4866 
4867 	if (cas_version_printed++ == 0)
4868 		pr_info("%s", version);
4869 
4870 	err = pci_enable_device(pdev);
4871 	if (err) {
4872 		dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
4873 		return err;
4874 	}
4875 
4876 	if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
4877 		dev_err(&pdev->dev, "Cannot find proper PCI device "
4878 		       "base address, aborting\n");
4879 		err = -ENODEV;
4880 		goto err_out_disable_pdev;
4881 	}
4882 
4883 	dev = alloc_etherdev(sizeof(*cp));
4884 	if (!dev) {
4885 		err = -ENOMEM;
4886 		goto err_out_disable_pdev;
4887 	}
4888 	SET_NETDEV_DEV(dev, &pdev->dev);
4889 
4890 	err = pci_request_regions(pdev, dev->name);
4891 	if (err) {
4892 		dev_err(&pdev->dev, "Cannot obtain PCI resources, aborting\n");
4893 		goto err_out_free_netdev;
4894 	}
4895 	pci_set_master(pdev);
4896 
4897 	/* we must always turn on parity response or else parity
4898 	 * doesn't get generated properly. disable SERR/PERR as well.
4899 	 * in addition, we want to turn MWI on.
4900 	 */
4901 	pci_read_config_word(pdev, PCI_COMMAND, &pci_cmd);
4902 	pci_cmd &= ~PCI_COMMAND_SERR;
4903 	pci_cmd |= PCI_COMMAND_PARITY;
4904 	pci_write_config_word(pdev, PCI_COMMAND, pci_cmd);
4905 	if (pci_try_set_mwi(pdev))
4906 		pr_warn("Could not enable MWI for %s\n", pci_name(pdev));
4907 
4908 	cas_program_bridge(pdev);
4909 
4910 	/*
4911 	 * On some architectures, the default cache line size set
4912 	 * by pci_try_set_mwi reduces perforamnce.  We have to increase
4913 	 * it for this case.  To start, we'll print some configuration
4914 	 * data.
4915 	 */
4916 #if 1
4917 	pci_read_config_byte(pdev, PCI_CACHE_LINE_SIZE,
4918 			     &orig_cacheline_size);
4919 	if (orig_cacheline_size < CAS_PREF_CACHELINE_SIZE) {
4920 		cas_cacheline_size =
4921 			(CAS_PREF_CACHELINE_SIZE < SMP_CACHE_BYTES) ?
4922 			CAS_PREF_CACHELINE_SIZE : SMP_CACHE_BYTES;
4923 		if (pci_write_config_byte(pdev,
4924 					  PCI_CACHE_LINE_SIZE,
4925 					  cas_cacheline_size)) {
4926 			dev_err(&pdev->dev, "Could not set PCI cache "
4927 			       "line size\n");
4928 			goto err_out_free_res;
4929 		}
4930 	}
4931 #endif
4932 
4933 
4934 	/* Configure DMA attributes. */
4935 	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
4936 	if (err) {
4937 		dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
4938 		goto err_out_free_res;
4939 	}
4940 
4941 	casreg_len = pci_resource_len(pdev, 0);
4942 
4943 	cp = netdev_priv(dev);
4944 	cp->pdev = pdev;
4945 #if 1
4946 	/* A value of 0 indicates we never explicitly set it */
4947 	cp->orig_cacheline_size = cas_cacheline_size ? orig_cacheline_size: 0;
4948 #endif
4949 	cp->dev = dev;
4950 	cp->msg_enable = (cassini_debug < 0) ? CAS_DEF_MSG_ENABLE :
4951 	  cassini_debug;
4952 
4953 #if defined(CONFIG_SPARC)
4954 	cp->of_node = pci_device_to_OF_node(pdev);
4955 #endif
4956 
4957 	cp->link_transition = LINK_TRANSITION_UNKNOWN;
4958 	cp->link_transition_jiffies_valid = 0;
4959 
4960 	spin_lock_init(&cp->lock);
4961 	spin_lock_init(&cp->rx_inuse_lock);
4962 	spin_lock_init(&cp->rx_spare_lock);
4963 	for (i = 0; i < N_TX_RINGS; i++) {
4964 		spin_lock_init(&cp->stat_lock[i]);
4965 		spin_lock_init(&cp->tx_lock[i]);
4966 	}
4967 	spin_lock_init(&cp->stat_lock[N_TX_RINGS]);
4968 	mutex_init(&cp->pm_mutex);
4969 
4970 	timer_setup(&cp->link_timer, cas_link_timer, 0);
4971 
4972 #if 1
4973 	/* Just in case the implementation of atomic operations
4974 	 * change so that an explicit initialization is necessary.
4975 	 */
4976 	atomic_set(&cp->reset_task_pending, 0);
4977 	atomic_set(&cp->reset_task_pending_all, 0);
4978 	atomic_set(&cp->reset_task_pending_spare, 0);
4979 	atomic_set(&cp->reset_task_pending_mtu, 0);
4980 #endif
4981 	INIT_WORK(&cp->reset_task, cas_reset_task);
4982 
4983 	/* Default link parameters */
4984 	if (link_mode >= 0 && link_mode < 6)
4985 		cp->link_cntl = link_modes[link_mode];
4986 	else
4987 		cp->link_cntl = BMCR_ANENABLE;
4988 	cp->lstate = link_down;
4989 	cp->link_transition = LINK_TRANSITION_LINK_DOWN;
4990 	netif_carrier_off(cp->dev);
4991 	cp->timer_ticks = 0;
4992 
4993 	/* give us access to cassini registers */
4994 	cp->regs = pci_iomap(pdev, 0, casreg_len);
4995 	if (!cp->regs) {
4996 		dev_err(&pdev->dev, "Cannot map device registers, aborting\n");
4997 		goto err_out_free_res;
4998 	}
4999 	cp->casreg_len = casreg_len;
5000 
5001 	pci_save_state(pdev);
5002 	cas_check_pci_invariants(cp);
5003 	cas_hard_reset(cp);
5004 	cas_reset(cp, 0);
5005 	if (cas_check_invariants(cp))
5006 		goto err_out_iounmap;
5007 	if (cp->cas_flags & CAS_FLAG_SATURN)
5008 		cas_saturn_firmware_init(cp);
5009 
5010 	cp->init_block =
5011 		dma_alloc_coherent(&pdev->dev, sizeof(struct cas_init_block),
5012 				   &cp->block_dvma, GFP_KERNEL);
5013 	if (!cp->init_block) {
5014 		dev_err(&pdev->dev, "Cannot allocate init block, aborting\n");
5015 		goto err_out_iounmap;
5016 	}
5017 
5018 	for (i = 0; i < N_TX_RINGS; i++)
5019 		cp->init_txds[i] = cp->init_block->txds[i];
5020 
5021 	for (i = 0; i < N_RX_DESC_RINGS; i++)
5022 		cp->init_rxds[i] = cp->init_block->rxds[i];
5023 
5024 	for (i = 0; i < N_RX_COMP_RINGS; i++)
5025 		cp->init_rxcs[i] = cp->init_block->rxcs[i];
5026 
5027 	for (i = 0; i < N_RX_FLOWS; i++)
5028 		skb_queue_head_init(&cp->rx_flows[i]);
5029 
5030 	dev->netdev_ops = &cas_netdev_ops;
5031 	dev->ethtool_ops = &cas_ethtool_ops;
5032 	dev->watchdog_timeo = CAS_TX_TIMEOUT;
5033 
5034 #ifdef USE_NAPI
5035 	netif_napi_add(dev, &cp->napi, cas_poll);
5036 #endif
5037 	dev->irq = pdev->irq;
5038 	dev->dma = 0;
5039 
5040 	/* Cassini features. */
5041 	if ((cp->cas_flags & CAS_FLAG_NO_HW_CSUM) == 0)
5042 		dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG;
5043 
5044 	dev->features |= NETIF_F_HIGHDMA;
5045 
5046 	/* MTU range: 60 - varies or 9000 */
5047 	dev->min_mtu = CAS_MIN_MTU;
5048 	dev->max_mtu = CAS_MAX_MTU;
5049 
5050 	if (register_netdev(dev)) {
5051 		dev_err(&pdev->dev, "Cannot register net device, aborting\n");
5052 		goto err_out_free_consistent;
5053 	}
5054 
5055 	i = readl(cp->regs + REG_BIM_CFG);
5056 	netdev_info(dev, "Sun Cassini%s (%sbit/%sMHz PCI/%s) Ethernet[%d] %pM\n",
5057 		    (cp->cas_flags & CAS_FLAG_REG_PLUS) ? "+" : "",
5058 		    (i & BIM_CFG_32BIT) ? "32" : "64",
5059 		    (i & BIM_CFG_66MHZ) ? "66" : "33",
5060 		    (cp->phy_type == CAS_PHY_SERDES) ? "Fi" : "Cu", pdev->irq,
5061 		    dev->dev_addr);
5062 
5063 	pci_set_drvdata(pdev, dev);
5064 	cp->hw_running = 1;
5065 	cas_entropy_reset(cp);
5066 	cas_phy_init(cp);
5067 	cas_begin_auto_negotiation(cp, NULL);
5068 	return 0;
5069 
5070 err_out_free_consistent:
5071 	dma_free_coherent(&pdev->dev, sizeof(struct cas_init_block),
5072 			  cp->init_block, cp->block_dvma);
5073 
5074 err_out_iounmap:
5075 	mutex_lock(&cp->pm_mutex);
5076 	if (cp->hw_running)
5077 		cas_shutdown(cp);
5078 	mutex_unlock(&cp->pm_mutex);
5079 
5080 	pci_iounmap(pdev, cp->regs);
5081 
5082 
5083 err_out_free_res:
5084 	pci_release_regions(pdev);
5085 
5086 	/* Try to restore it in case the error occurred after we
5087 	 * set it.
5088 	 */
5089 	pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE, orig_cacheline_size);
5090 
5091 err_out_free_netdev:
5092 	free_netdev(dev);
5093 
5094 err_out_disable_pdev:
5095 	pci_disable_device(pdev);
5096 	return -ENODEV;
5097 }
5098 
5099 static void cas_remove_one(struct pci_dev *pdev)
5100 {
5101 	struct net_device *dev = pci_get_drvdata(pdev);
5102 	struct cas *cp;
5103 	if (!dev)
5104 		return;
5105 
5106 	cp = netdev_priv(dev);
5107 	unregister_netdev(dev);
5108 
5109 	vfree(cp->fw_data);
5110 
5111 	mutex_lock(&cp->pm_mutex);
5112 	cancel_work_sync(&cp->reset_task);
5113 	if (cp->hw_running)
5114 		cas_shutdown(cp);
5115 	mutex_unlock(&cp->pm_mutex);
5116 
5117 #if 1
5118 	if (cp->orig_cacheline_size) {
5119 		/* Restore the cache line size if we had modified
5120 		 * it.
5121 		 */
5122 		pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE,
5123 				      cp->orig_cacheline_size);
5124 	}
5125 #endif
5126 	dma_free_coherent(&pdev->dev, sizeof(struct cas_init_block),
5127 			  cp->init_block, cp->block_dvma);
5128 	pci_iounmap(pdev, cp->regs);
5129 	free_netdev(dev);
5130 	pci_release_regions(pdev);
5131 	pci_disable_device(pdev);
5132 }
5133 
5134 static int __maybe_unused cas_suspend(struct device *dev_d)
5135 {
5136 	struct net_device *dev = dev_get_drvdata(dev_d);
5137 	struct cas *cp = netdev_priv(dev);
5138 	unsigned long flags;
5139 
5140 	mutex_lock(&cp->pm_mutex);
5141 
5142 	/* If the driver is opened, we stop the DMA */
5143 	if (cp->opened) {
5144 		netif_device_detach(dev);
5145 
5146 		cas_lock_all_save(cp, flags);
5147 
5148 		/* We can set the second arg of cas_reset to 0
5149 		 * because on resume, we'll call cas_init_hw with
5150 		 * its second arg set so that autonegotiation is
5151 		 * restarted.
5152 		 */
5153 		cas_reset(cp, 0);
5154 		cas_clean_rings(cp);
5155 		cas_unlock_all_restore(cp, flags);
5156 	}
5157 
5158 	if (cp->hw_running)
5159 		cas_shutdown(cp);
5160 	mutex_unlock(&cp->pm_mutex);
5161 
5162 	return 0;
5163 }
5164 
5165 static int __maybe_unused cas_resume(struct device *dev_d)
5166 {
5167 	struct net_device *dev = dev_get_drvdata(dev_d);
5168 	struct cas *cp = netdev_priv(dev);
5169 
5170 	netdev_info(dev, "resuming\n");
5171 
5172 	mutex_lock(&cp->pm_mutex);
5173 	cas_hard_reset(cp);
5174 	if (cp->opened) {
5175 		unsigned long flags;
5176 		cas_lock_all_save(cp, flags);
5177 		cas_reset(cp, 0);
5178 		cp->hw_running = 1;
5179 		cas_clean_rings(cp);
5180 		cas_init_hw(cp, 1);
5181 		cas_unlock_all_restore(cp, flags);
5182 
5183 		netif_device_attach(dev);
5184 	}
5185 	mutex_unlock(&cp->pm_mutex);
5186 	return 0;
5187 }
5188 
5189 static SIMPLE_DEV_PM_OPS(cas_pm_ops, cas_suspend, cas_resume);
5190 
5191 static struct pci_driver cas_driver = {
5192 	.name		= DRV_MODULE_NAME,
5193 	.id_table	= cas_pci_tbl,
5194 	.probe		= cas_init_one,
5195 	.remove		= cas_remove_one,
5196 	.driver.pm	= &cas_pm_ops,
5197 };
5198 
5199 static int __init cas_init(void)
5200 {
5201 	if (linkdown_timeout > 0)
5202 		link_transition_timeout = linkdown_timeout * HZ;
5203 	else
5204 		link_transition_timeout = 0;
5205 
5206 	return pci_register_driver(&cas_driver);
5207 }
5208 
5209 static void __exit cas_cleanup(void)
5210 {
5211 	pci_unregister_driver(&cas_driver);
5212 }
5213 
5214 module_init(cas_init);
5215 module_exit(cas_cleanup);
5216