xref: /openbmc/linux/drivers/net/ethernet/sun/cassini.c (revision e7bae9bb)
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  *     the 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 
92 #define cas_page_map(x)      kmap_atomic((x))
93 #define cas_page_unmap(x)    kunmap_atomic((x))
94 #define CAS_NCPUS            num_online_cpus()
95 
96 #define cas_skb_release(x)  netif_rx(x)
97 
98 /* select which firmware to use */
99 #define USE_HP_WORKAROUND
100 #define HP_WORKAROUND_DEFAULT /* select which firmware to use as default */
101 #define CAS_HP_ALT_FIRMWARE   cas_prog_null /* alternate firmware */
102 
103 #include "cassini.h"
104 
105 #define USE_TX_COMPWB      /* use completion writeback registers */
106 #define USE_CSMA_CD_PROTO  /* standard CSMA/CD */
107 #define USE_RX_BLANK       /* hw interrupt mitigation */
108 #undef USE_ENTROPY_DEV     /* don't test for entropy device */
109 
110 /* NOTE: these aren't useable unless PCI interrupts can be assigned.
111  * also, we need to make cp->lock finer-grained.
112  */
113 #undef  USE_PCI_INTB
114 #undef  USE_PCI_INTC
115 #undef  USE_PCI_INTD
116 #undef  USE_QOS
117 
118 #undef  USE_VPD_DEBUG       /* debug vpd information if defined */
119 
120 /* rx processing options */
121 #define USE_PAGE_ORDER      /* specify to allocate large rx pages */
122 #define RX_DONT_BATCH  0    /* if 1, don't batch flows */
123 #define RX_COPY_ALWAYS 0    /* if 0, use frags */
124 #define RX_COPY_MIN    64   /* copy a little to make upper layers happy */
125 #undef  RX_COUNT_BUFFERS    /* define to calculate RX buffer stats */
126 
127 #define DRV_MODULE_NAME		"cassini"
128 #define DRV_MODULE_VERSION	"1.6"
129 #define DRV_MODULE_RELDATE	"21 May 2008"
130 
131 #define CAS_DEF_MSG_ENABLE	  \
132 	(NETIF_MSG_DRV		| \
133 	 NETIF_MSG_PROBE	| \
134 	 NETIF_MSG_LINK		| \
135 	 NETIF_MSG_TIMER	| \
136 	 NETIF_MSG_IFDOWN	| \
137 	 NETIF_MSG_IFUP		| \
138 	 NETIF_MSG_RX_ERR	| \
139 	 NETIF_MSG_TX_ERR)
140 
141 /* length of time before we decide the hardware is borked,
142  * and dev->tx_timeout() should be called to fix the problem
143  */
144 #define CAS_TX_TIMEOUT			(HZ)
145 #define CAS_LINK_TIMEOUT                (22*HZ/10)
146 #define CAS_LINK_FAST_TIMEOUT           (1)
147 
148 /* timeout values for state changing. these specify the number
149  * of 10us delays to be used before giving up.
150  */
151 #define STOP_TRIES_PHY 1000
152 #define STOP_TRIES     5000
153 
154 /* specify a minimum frame size to deal with some fifo issues
155  * max mtu == 2 * page size - ethernet header - 64 - swivel =
156  *            2 * page_size - 0x50
157  */
158 #define CAS_MIN_FRAME			97
159 #define CAS_1000MB_MIN_FRAME            255
160 #define CAS_MIN_MTU                     60
161 #define CAS_MAX_MTU                     min(((cp->page_size << 1) - 0x50), 9000)
162 
163 #if 1
164 /*
165  * Eliminate these and use separate atomic counters for each, to
166  * avoid a race condition.
167  */
168 #else
169 #define CAS_RESET_MTU                   1
170 #define CAS_RESET_ALL                   2
171 #define CAS_RESET_SPARE                 3
172 #endif
173 
174 static char version[] =
175 	DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
176 
177 static int cassini_debug = -1;	/* -1 == use CAS_DEF_MSG_ENABLE as value */
178 static int link_mode;
179 
180 MODULE_AUTHOR("Adrian Sun (asun@darksunrising.com)");
181 MODULE_DESCRIPTION("Sun Cassini(+) ethernet driver");
182 MODULE_LICENSE("GPL");
183 MODULE_FIRMWARE("sun/cassini.bin");
184 module_param(cassini_debug, int, 0);
185 MODULE_PARM_DESC(cassini_debug, "Cassini bitmapped debugging message enable value");
186 module_param(link_mode, int, 0);
187 MODULE_PARM_DESC(link_mode, "default link mode");
188 
189 /*
190  * Work around for a PCS bug in which the link goes down due to the chip
191  * being confused and never showing a link status of "up."
192  */
193 #define DEFAULT_LINKDOWN_TIMEOUT 5
194 /*
195  * Value in seconds, for user input.
196  */
197 static int linkdown_timeout = DEFAULT_LINKDOWN_TIMEOUT;
198 module_param(linkdown_timeout, int, 0);
199 MODULE_PARM_DESC(linkdown_timeout,
200 "min reset interval in sec. for PCS linkdown issue; disabled if not positive");
201 
202 /*
203  * value in 'ticks' (units used by jiffies). Set when we init the
204  * module because 'HZ' in actually a function call on some flavors of
205  * Linux.  This will default to DEFAULT_LINKDOWN_TIMEOUT * HZ.
206  */
207 static int link_transition_timeout;
208 
209 
210 
211 static u16 link_modes[] = {
212 	BMCR_ANENABLE,			 /* 0 : autoneg */
213 	0,				 /* 1 : 10bt half duplex */
214 	BMCR_SPEED100,			 /* 2 : 100bt half duplex */
215 	BMCR_FULLDPLX,			 /* 3 : 10bt full duplex */
216 	BMCR_SPEED100|BMCR_FULLDPLX,	 /* 4 : 100bt full duplex */
217 	CAS_BMCR_SPEED1000|BMCR_FULLDPLX /* 5 : 1000bt full duplex */
218 };
219 
220 static const struct pci_device_id cas_pci_tbl[] = {
221 	{ PCI_VENDOR_ID_SUN, PCI_DEVICE_ID_SUN_CASSINI,
222 	  PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0UL },
223 	{ PCI_VENDOR_ID_NS, PCI_DEVICE_ID_NS_SATURN,
224 	  PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0UL },
225 	{ 0, }
226 };
227 
228 MODULE_DEVICE_TABLE(pci, cas_pci_tbl);
229 
230 static void cas_set_link_modes(struct cas *cp);
231 
232 static inline void cas_lock_tx(struct cas *cp)
233 {
234 	int i;
235 
236 	for (i = 0; i < N_TX_RINGS; i++)
237 		spin_lock_nested(&cp->tx_lock[i], i);
238 }
239 
240 /* WTZ: QA was finding deadlock problems with the previous
241  * versions after long test runs with multiple cards per machine.
242  * See if replacing cas_lock_all with safer versions helps. The
243  * symptoms QA is reporting match those we'd expect if interrupts
244  * aren't being properly restored, and we fixed a previous deadlock
245  * with similar symptoms by using save/restore versions in other
246  * places.
247  */
248 #define cas_lock_all_save(cp, flags) \
249 do { \
250 	struct cas *xxxcp = (cp); \
251 	spin_lock_irqsave(&xxxcp->lock, flags); \
252 	cas_lock_tx(xxxcp); \
253 } while (0)
254 
255 static inline void cas_unlock_tx(struct cas *cp)
256 {
257 	int i;
258 
259 	for (i = N_TX_RINGS; i > 0; i--)
260 		spin_unlock(&cp->tx_lock[i - 1]);
261 }
262 
263 #define cas_unlock_all_restore(cp, flags) \
264 do { \
265 	struct cas *xxxcp = (cp); \
266 	cas_unlock_tx(xxxcp); \
267 	spin_unlock_irqrestore(&xxxcp->lock, flags); \
268 } while (0)
269 
270 static void cas_disable_irq(struct cas *cp, const int ring)
271 {
272 	/* Make sure we won't get any more interrupts */
273 	if (ring == 0) {
274 		writel(0xFFFFFFFF, cp->regs + REG_INTR_MASK);
275 		return;
276 	}
277 
278 	/* disable completion interrupts and selectively mask */
279 	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
280 		switch (ring) {
281 #if defined (USE_PCI_INTB) || defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
282 #ifdef USE_PCI_INTB
283 		case 1:
284 #endif
285 #ifdef USE_PCI_INTC
286 		case 2:
287 #endif
288 #ifdef USE_PCI_INTD
289 		case 3:
290 #endif
291 			writel(INTRN_MASK_CLEAR_ALL | INTRN_MASK_RX_EN,
292 			       cp->regs + REG_PLUS_INTRN_MASK(ring));
293 			break;
294 #endif
295 		default:
296 			writel(INTRN_MASK_CLEAR_ALL, cp->regs +
297 			       REG_PLUS_INTRN_MASK(ring));
298 			break;
299 		}
300 	}
301 }
302 
303 static inline void cas_mask_intr(struct cas *cp)
304 {
305 	int i;
306 
307 	for (i = 0; i < N_RX_COMP_RINGS; i++)
308 		cas_disable_irq(cp, i);
309 }
310 
311 static void cas_enable_irq(struct cas *cp, const int ring)
312 {
313 	if (ring == 0) { /* all but TX_DONE */
314 		writel(INTR_TX_DONE, cp->regs + REG_INTR_MASK);
315 		return;
316 	}
317 
318 	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
319 		switch (ring) {
320 #if defined (USE_PCI_INTB) || defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
321 #ifdef USE_PCI_INTB
322 		case 1:
323 #endif
324 #ifdef USE_PCI_INTC
325 		case 2:
326 #endif
327 #ifdef USE_PCI_INTD
328 		case 3:
329 #endif
330 			writel(INTRN_MASK_RX_EN, cp->regs +
331 			       REG_PLUS_INTRN_MASK(ring));
332 			break;
333 #endif
334 		default:
335 			break;
336 		}
337 	}
338 }
339 
340 static inline void cas_unmask_intr(struct cas *cp)
341 {
342 	int i;
343 
344 	for (i = 0; i < N_RX_COMP_RINGS; i++)
345 		cas_enable_irq(cp, i);
346 }
347 
348 static inline void cas_entropy_gather(struct cas *cp)
349 {
350 #ifdef USE_ENTROPY_DEV
351 	if ((cp->cas_flags & CAS_FLAG_ENTROPY_DEV) == 0)
352 		return;
353 
354 	batch_entropy_store(readl(cp->regs + REG_ENTROPY_IV),
355 			    readl(cp->regs + REG_ENTROPY_IV),
356 			    sizeof(uint64_t)*8);
357 #endif
358 }
359 
360 static inline void cas_entropy_reset(struct cas *cp)
361 {
362 #ifdef USE_ENTROPY_DEV
363 	if ((cp->cas_flags & CAS_FLAG_ENTROPY_DEV) == 0)
364 		return;
365 
366 	writel(BIM_LOCAL_DEV_PAD | BIM_LOCAL_DEV_PROM | BIM_LOCAL_DEV_EXT,
367 	       cp->regs + REG_BIM_LOCAL_DEV_EN);
368 	writeb(ENTROPY_RESET_STC_MODE, cp->regs + REG_ENTROPY_RESET);
369 	writeb(0x55, cp->regs + REG_ENTROPY_RAND_REG);
370 
371 	/* if we read back 0x0, we don't have an entropy device */
372 	if (readb(cp->regs + REG_ENTROPY_RAND_REG) == 0)
373 		cp->cas_flags &= ~CAS_FLAG_ENTROPY_DEV;
374 #endif
375 }
376 
377 /* access to the phy. the following assumes that we've initialized the MIF to
378  * be in frame rather than bit-bang mode
379  */
380 static u16 cas_phy_read(struct cas *cp, int reg)
381 {
382 	u32 cmd;
383 	int limit = STOP_TRIES_PHY;
384 
385 	cmd = MIF_FRAME_ST | MIF_FRAME_OP_READ;
386 	cmd |= CAS_BASE(MIF_FRAME_PHY_ADDR, cp->phy_addr);
387 	cmd |= CAS_BASE(MIF_FRAME_REG_ADDR, reg);
388 	cmd |= MIF_FRAME_TURN_AROUND_MSB;
389 	writel(cmd, cp->regs + REG_MIF_FRAME);
390 
391 	/* poll for completion */
392 	while (limit-- > 0) {
393 		udelay(10);
394 		cmd = readl(cp->regs + REG_MIF_FRAME);
395 		if (cmd & MIF_FRAME_TURN_AROUND_LSB)
396 			return cmd & MIF_FRAME_DATA_MASK;
397 	}
398 	return 0xFFFF; /* -1 */
399 }
400 
401 static int cas_phy_write(struct cas *cp, int reg, u16 val)
402 {
403 	int limit = STOP_TRIES_PHY;
404 	u32 cmd;
405 
406 	cmd = MIF_FRAME_ST | MIF_FRAME_OP_WRITE;
407 	cmd |= CAS_BASE(MIF_FRAME_PHY_ADDR, cp->phy_addr);
408 	cmd |= CAS_BASE(MIF_FRAME_REG_ADDR, reg);
409 	cmd |= MIF_FRAME_TURN_AROUND_MSB;
410 	cmd |= val & MIF_FRAME_DATA_MASK;
411 	writel(cmd, cp->regs + REG_MIF_FRAME);
412 
413 	/* poll for completion */
414 	while (limit-- > 0) {
415 		udelay(10);
416 		cmd = readl(cp->regs + REG_MIF_FRAME);
417 		if (cmd & MIF_FRAME_TURN_AROUND_LSB)
418 			return 0;
419 	}
420 	return -1;
421 }
422 
423 static void cas_phy_powerup(struct cas *cp)
424 {
425 	u16 ctl = cas_phy_read(cp, MII_BMCR);
426 
427 	if ((ctl & BMCR_PDOWN) == 0)
428 		return;
429 	ctl &= ~BMCR_PDOWN;
430 	cas_phy_write(cp, MII_BMCR, ctl);
431 }
432 
433 static void cas_phy_powerdown(struct cas *cp)
434 {
435 	u16 ctl = cas_phy_read(cp, MII_BMCR);
436 
437 	if (ctl & BMCR_PDOWN)
438 		return;
439 	ctl |= BMCR_PDOWN;
440 	cas_phy_write(cp, MII_BMCR, ctl);
441 }
442 
443 /* cp->lock held. note: the last put_page will free the buffer */
444 static int cas_page_free(struct cas *cp, cas_page_t *page)
445 {
446 	dma_unmap_page(&cp->pdev->dev, page->dma_addr, cp->page_size,
447 		       DMA_FROM_DEVICE);
448 	__free_pages(page->buffer, cp->page_order);
449 	kfree(page);
450 	return 0;
451 }
452 
453 #ifdef RX_COUNT_BUFFERS
454 #define RX_USED_ADD(x, y)       ((x)->used += (y))
455 #define RX_USED_SET(x, y)       ((x)->used  = (y))
456 #else
457 #define RX_USED_ADD(x, y)
458 #define RX_USED_SET(x, y)
459 #endif
460 
461 /* local page allocation routines for the receive buffers. jumbo pages
462  * require at least 8K contiguous and 8K aligned buffers.
463  */
464 static cas_page_t *cas_page_alloc(struct cas *cp, const gfp_t flags)
465 {
466 	cas_page_t *page;
467 
468 	page = kmalloc(sizeof(cas_page_t), flags);
469 	if (!page)
470 		return NULL;
471 
472 	INIT_LIST_HEAD(&page->list);
473 	RX_USED_SET(page, 0);
474 	page->buffer = alloc_pages(flags, cp->page_order);
475 	if (!page->buffer)
476 		goto page_err;
477 	page->dma_addr = dma_map_page(&cp->pdev->dev, page->buffer, 0,
478 				      cp->page_size, DMA_FROM_DEVICE);
479 	return page;
480 
481 page_err:
482 	kfree(page);
483 	return NULL;
484 }
485 
486 /* initialize spare pool of rx buffers, but allocate during the open */
487 static void cas_spare_init(struct cas *cp)
488 {
489   	spin_lock(&cp->rx_inuse_lock);
490 	INIT_LIST_HEAD(&cp->rx_inuse_list);
491 	spin_unlock(&cp->rx_inuse_lock);
492 
493 	spin_lock(&cp->rx_spare_lock);
494 	INIT_LIST_HEAD(&cp->rx_spare_list);
495 	cp->rx_spares_needed = RX_SPARE_COUNT;
496 	spin_unlock(&cp->rx_spare_lock);
497 }
498 
499 /* used on close. free all the spare buffers. */
500 static void cas_spare_free(struct cas *cp)
501 {
502 	struct list_head list, *elem, *tmp;
503 
504 	/* free spare buffers */
505 	INIT_LIST_HEAD(&list);
506 	spin_lock(&cp->rx_spare_lock);
507 	list_splice_init(&cp->rx_spare_list, &list);
508 	spin_unlock(&cp->rx_spare_lock);
509 	list_for_each_safe(elem, tmp, &list) {
510 		cas_page_free(cp, list_entry(elem, cas_page_t, list));
511 	}
512 
513 	INIT_LIST_HEAD(&list);
514 #if 1
515 	/*
516 	 * Looks like Adrian had protected this with a different
517 	 * lock than used everywhere else to manipulate this list.
518 	 */
519 	spin_lock(&cp->rx_inuse_lock);
520 	list_splice_init(&cp->rx_inuse_list, &list);
521 	spin_unlock(&cp->rx_inuse_lock);
522 #else
523 	spin_lock(&cp->rx_spare_lock);
524 	list_splice_init(&cp->rx_inuse_list, &list);
525 	spin_unlock(&cp->rx_spare_lock);
526 #endif
527 	list_for_each_safe(elem, tmp, &list) {
528 		cas_page_free(cp, list_entry(elem, cas_page_t, list));
529 	}
530 }
531 
532 /* replenish spares if needed */
533 static void cas_spare_recover(struct cas *cp, const gfp_t flags)
534 {
535 	struct list_head list, *elem, *tmp;
536 	int needed, i;
537 
538 	/* check inuse list. if we don't need any more free buffers,
539 	 * just free it
540 	 */
541 
542 	/* make a local copy of the list */
543 	INIT_LIST_HEAD(&list);
544 	spin_lock(&cp->rx_inuse_lock);
545 	list_splice_init(&cp->rx_inuse_list, &list);
546 	spin_unlock(&cp->rx_inuse_lock);
547 
548 	list_for_each_safe(elem, tmp, &list) {
549 		cas_page_t *page = list_entry(elem, cas_page_t, list);
550 
551 		/*
552 		 * With the lockless pagecache, cassini buffering scheme gets
553 		 * slightly less accurate: we might find that a page has an
554 		 * elevated reference count here, due to a speculative ref,
555 		 * and skip it as in-use. Ideally we would be able to reclaim
556 		 * it. However this would be such a rare case, it doesn't
557 		 * matter too much as we should pick it up the next time round.
558 		 *
559 		 * Importantly, if we find that the page has a refcount of 1
560 		 * here (our refcount), then we know it is definitely not inuse
561 		 * so we can reuse it.
562 		 */
563 		if (page_count(page->buffer) > 1)
564 			continue;
565 
566 		list_del(elem);
567 		spin_lock(&cp->rx_spare_lock);
568 		if (cp->rx_spares_needed > 0) {
569 			list_add(elem, &cp->rx_spare_list);
570 			cp->rx_spares_needed--;
571 			spin_unlock(&cp->rx_spare_lock);
572 		} else {
573 			spin_unlock(&cp->rx_spare_lock);
574 			cas_page_free(cp, page);
575 		}
576 	}
577 
578 	/* put any inuse buffers back on the list */
579 	if (!list_empty(&list)) {
580 		spin_lock(&cp->rx_inuse_lock);
581 		list_splice(&list, &cp->rx_inuse_list);
582 		spin_unlock(&cp->rx_inuse_lock);
583 	}
584 
585 	spin_lock(&cp->rx_spare_lock);
586 	needed = cp->rx_spares_needed;
587 	spin_unlock(&cp->rx_spare_lock);
588 	if (!needed)
589 		return;
590 
591 	/* we still need spares, so try to allocate some */
592 	INIT_LIST_HEAD(&list);
593 	i = 0;
594 	while (i < needed) {
595 		cas_page_t *spare = cas_page_alloc(cp, flags);
596 		if (!spare)
597 			break;
598 		list_add(&spare->list, &list);
599 		i++;
600 	}
601 
602 	spin_lock(&cp->rx_spare_lock);
603 	list_splice(&list, &cp->rx_spare_list);
604 	cp->rx_spares_needed -= i;
605 	spin_unlock(&cp->rx_spare_lock);
606 }
607 
608 /* pull a page from the list. */
609 static cas_page_t *cas_page_dequeue(struct cas *cp)
610 {
611 	struct list_head *entry;
612 	int recover;
613 
614 	spin_lock(&cp->rx_spare_lock);
615 	if (list_empty(&cp->rx_spare_list)) {
616 		/* try to do a quick recovery */
617 		spin_unlock(&cp->rx_spare_lock);
618 		cas_spare_recover(cp, GFP_ATOMIC);
619 		spin_lock(&cp->rx_spare_lock);
620 		if (list_empty(&cp->rx_spare_list)) {
621 			netif_err(cp, rx_err, cp->dev,
622 				  "no spare buffers available\n");
623 			spin_unlock(&cp->rx_spare_lock);
624 			return NULL;
625 		}
626 	}
627 
628 	entry = cp->rx_spare_list.next;
629 	list_del(entry);
630 	recover = ++cp->rx_spares_needed;
631 	spin_unlock(&cp->rx_spare_lock);
632 
633 	/* trigger the timer to do the recovery */
634 	if ((recover & (RX_SPARE_RECOVER_VAL - 1)) == 0) {
635 #if 1
636 		atomic_inc(&cp->reset_task_pending);
637 		atomic_inc(&cp->reset_task_pending_spare);
638 		schedule_work(&cp->reset_task);
639 #else
640 		atomic_set(&cp->reset_task_pending, CAS_RESET_SPARE);
641 		schedule_work(&cp->reset_task);
642 #endif
643 	}
644 	return list_entry(entry, cas_page_t, list);
645 }
646 
647 
648 static void cas_mif_poll(struct cas *cp, const int enable)
649 {
650 	u32 cfg;
651 
652 	cfg  = readl(cp->regs + REG_MIF_CFG);
653 	cfg &= (MIF_CFG_MDIO_0 | MIF_CFG_MDIO_1);
654 
655 	if (cp->phy_type & CAS_PHY_MII_MDIO1)
656 		cfg |= MIF_CFG_PHY_SELECT;
657 
658 	/* poll and interrupt on link status change. */
659 	if (enable) {
660 		cfg |= MIF_CFG_POLL_EN;
661 		cfg |= CAS_BASE(MIF_CFG_POLL_REG, MII_BMSR);
662 		cfg |= CAS_BASE(MIF_CFG_POLL_PHY, cp->phy_addr);
663 	}
664 	writel((enable) ? ~(BMSR_LSTATUS | BMSR_ANEGCOMPLETE) : 0xFFFF,
665 	       cp->regs + REG_MIF_MASK);
666 	writel(cfg, cp->regs + REG_MIF_CFG);
667 }
668 
669 /* Must be invoked under cp->lock */
670 static void cas_begin_auto_negotiation(struct cas *cp,
671 				       const struct ethtool_link_ksettings *ep)
672 {
673 	u16 ctl;
674 #if 1
675 	int lcntl;
676 	int changed = 0;
677 	int oldstate = cp->lstate;
678 	int link_was_not_down = !(oldstate == link_down);
679 #endif
680 	/* Setup link parameters */
681 	if (!ep)
682 		goto start_aneg;
683 	lcntl = cp->link_cntl;
684 	if (ep->base.autoneg == AUTONEG_ENABLE) {
685 		cp->link_cntl = BMCR_ANENABLE;
686 	} else {
687 		u32 speed = ep->base.speed;
688 		cp->link_cntl = 0;
689 		if (speed == SPEED_100)
690 			cp->link_cntl |= BMCR_SPEED100;
691 		else if (speed == SPEED_1000)
692 			cp->link_cntl |= CAS_BMCR_SPEED1000;
693 		if (ep->base.duplex == DUPLEX_FULL)
694 			cp->link_cntl |= BMCR_FULLDPLX;
695 	}
696 #if 1
697 	changed = (lcntl != cp->link_cntl);
698 #endif
699 start_aneg:
700 	if (cp->lstate == link_up) {
701 		netdev_info(cp->dev, "PCS link down\n");
702 	} else {
703 		if (changed) {
704 			netdev_info(cp->dev, "link configuration changed\n");
705 		}
706 	}
707 	cp->lstate = link_down;
708 	cp->link_transition = LINK_TRANSITION_LINK_DOWN;
709 	if (!cp->hw_running)
710 		return;
711 #if 1
712 	/*
713 	 * WTZ: If the old state was link_up, we turn off the carrier
714 	 * to replicate everything we do elsewhere on a link-down
715 	 * event when we were already in a link-up state..
716 	 */
717 	if (oldstate == link_up)
718 		netif_carrier_off(cp->dev);
719 	if (changed  && link_was_not_down) {
720 		/*
721 		 * WTZ: This branch will simply schedule a full reset after
722 		 * we explicitly changed link modes in an ioctl. See if this
723 		 * fixes the link-problems we were having for forced mode.
724 		 */
725 		atomic_inc(&cp->reset_task_pending);
726 		atomic_inc(&cp->reset_task_pending_all);
727 		schedule_work(&cp->reset_task);
728 		cp->timer_ticks = 0;
729 		mod_timer(&cp->link_timer, jiffies + CAS_LINK_TIMEOUT);
730 		return;
731 	}
732 #endif
733 	if (cp->phy_type & CAS_PHY_SERDES) {
734 		u32 val = readl(cp->regs + REG_PCS_MII_CTRL);
735 
736 		if (cp->link_cntl & BMCR_ANENABLE) {
737 			val |= (PCS_MII_RESTART_AUTONEG | PCS_MII_AUTONEG_EN);
738 			cp->lstate = link_aneg;
739 		} else {
740 			if (cp->link_cntl & BMCR_FULLDPLX)
741 				val |= PCS_MII_CTRL_DUPLEX;
742 			val &= ~PCS_MII_AUTONEG_EN;
743 			cp->lstate = link_force_ok;
744 		}
745 		cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
746 		writel(val, cp->regs + REG_PCS_MII_CTRL);
747 
748 	} else {
749 		cas_mif_poll(cp, 0);
750 		ctl = cas_phy_read(cp, MII_BMCR);
751 		ctl &= ~(BMCR_FULLDPLX | BMCR_SPEED100 |
752 			 CAS_BMCR_SPEED1000 | BMCR_ANENABLE);
753 		ctl |= cp->link_cntl;
754 		if (ctl & BMCR_ANENABLE) {
755 			ctl |= BMCR_ANRESTART;
756 			cp->lstate = link_aneg;
757 		} else {
758 			cp->lstate = link_force_ok;
759 		}
760 		cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
761 		cas_phy_write(cp, MII_BMCR, ctl);
762 		cas_mif_poll(cp, 1);
763 	}
764 
765 	cp->timer_ticks = 0;
766 	mod_timer(&cp->link_timer, jiffies + CAS_LINK_TIMEOUT);
767 }
768 
769 /* Must be invoked under cp->lock. */
770 static int cas_reset_mii_phy(struct cas *cp)
771 {
772 	int limit = STOP_TRIES_PHY;
773 	u16 val;
774 
775 	cas_phy_write(cp, MII_BMCR, BMCR_RESET);
776 	udelay(100);
777 	while (--limit) {
778 		val = cas_phy_read(cp, MII_BMCR);
779 		if ((val & BMCR_RESET) == 0)
780 			break;
781 		udelay(10);
782 	}
783 	return limit <= 0;
784 }
785 
786 static void cas_saturn_firmware_init(struct cas *cp)
787 {
788 	const struct firmware *fw;
789 	const char fw_name[] = "sun/cassini.bin";
790 	int err;
791 
792 	if (PHY_NS_DP83065 != cp->phy_id)
793 		return;
794 
795 	err = request_firmware(&fw, fw_name, &cp->pdev->dev);
796 	if (err) {
797 		pr_err("Failed to load firmware \"%s\"\n",
798 		       fw_name);
799 		return;
800 	}
801 	if (fw->size < 2) {
802 		pr_err("bogus length %zu in \"%s\"\n",
803 		       fw->size, fw_name);
804 		goto out;
805 	}
806 	cp->fw_load_addr= fw->data[1] << 8 | fw->data[0];
807 	cp->fw_size = fw->size - 2;
808 	cp->fw_data = vmalloc(cp->fw_size);
809 	if (!cp->fw_data)
810 		goto out;
811 	memcpy(cp->fw_data, &fw->data[2], cp->fw_size);
812 out:
813 	release_firmware(fw);
814 }
815 
816 static void cas_saturn_firmware_load(struct cas *cp)
817 {
818 	int i;
819 
820 	if (!cp->fw_data)
821 		return;
822 
823 	cas_phy_powerdown(cp);
824 
825 	/* expanded memory access mode */
826 	cas_phy_write(cp, DP83065_MII_MEM, 0x0);
827 
828 	/* pointer configuration for new firmware */
829 	cas_phy_write(cp, DP83065_MII_REGE, 0x8ff9);
830 	cas_phy_write(cp, DP83065_MII_REGD, 0xbd);
831 	cas_phy_write(cp, DP83065_MII_REGE, 0x8ffa);
832 	cas_phy_write(cp, DP83065_MII_REGD, 0x82);
833 	cas_phy_write(cp, DP83065_MII_REGE, 0x8ffb);
834 	cas_phy_write(cp, DP83065_MII_REGD, 0x0);
835 	cas_phy_write(cp, DP83065_MII_REGE, 0x8ffc);
836 	cas_phy_write(cp, DP83065_MII_REGD, 0x39);
837 
838 	/* download new firmware */
839 	cas_phy_write(cp, DP83065_MII_MEM, 0x1);
840 	cas_phy_write(cp, DP83065_MII_REGE, cp->fw_load_addr);
841 	for (i = 0; i < cp->fw_size; i++)
842 		cas_phy_write(cp, DP83065_MII_REGD, cp->fw_data[i]);
843 
844 	/* enable firmware */
845 	cas_phy_write(cp, DP83065_MII_REGE, 0x8ff8);
846 	cas_phy_write(cp, DP83065_MII_REGD, 0x1);
847 }
848 
849 
850 /* phy initialization */
851 static void cas_phy_init(struct cas *cp)
852 {
853 	u16 val;
854 
855 	/* if we're in MII/GMII mode, set up phy */
856 	if (CAS_PHY_MII(cp->phy_type)) {
857 		writel(PCS_DATAPATH_MODE_MII,
858 		       cp->regs + REG_PCS_DATAPATH_MODE);
859 
860 		cas_mif_poll(cp, 0);
861 		cas_reset_mii_phy(cp); /* take out of isolate mode */
862 
863 		if (PHY_LUCENT_B0 == cp->phy_id) {
864 			/* workaround link up/down issue with lucent */
865 			cas_phy_write(cp, LUCENT_MII_REG, 0x8000);
866 			cas_phy_write(cp, MII_BMCR, 0x00f1);
867 			cas_phy_write(cp, LUCENT_MII_REG, 0x0);
868 
869 		} else if (PHY_BROADCOM_B0 == (cp->phy_id & 0xFFFFFFFC)) {
870 			/* workarounds for broadcom phy */
871 			cas_phy_write(cp, BROADCOM_MII_REG8, 0x0C20);
872 			cas_phy_write(cp, BROADCOM_MII_REG7, 0x0012);
873 			cas_phy_write(cp, BROADCOM_MII_REG5, 0x1804);
874 			cas_phy_write(cp, BROADCOM_MII_REG7, 0x0013);
875 			cas_phy_write(cp, BROADCOM_MII_REG5, 0x1204);
876 			cas_phy_write(cp, BROADCOM_MII_REG7, 0x8006);
877 			cas_phy_write(cp, BROADCOM_MII_REG5, 0x0132);
878 			cas_phy_write(cp, BROADCOM_MII_REG7, 0x8006);
879 			cas_phy_write(cp, BROADCOM_MII_REG5, 0x0232);
880 			cas_phy_write(cp, BROADCOM_MII_REG7, 0x201F);
881 			cas_phy_write(cp, BROADCOM_MII_REG5, 0x0A20);
882 
883 		} else if (PHY_BROADCOM_5411 == cp->phy_id) {
884 			val = cas_phy_read(cp, BROADCOM_MII_REG4);
885 			val = cas_phy_read(cp, BROADCOM_MII_REG4);
886 			if (val & 0x0080) {
887 				/* link workaround */
888 				cas_phy_write(cp, BROADCOM_MII_REG4,
889 					      val & ~0x0080);
890 			}
891 
892 		} else if (cp->cas_flags & CAS_FLAG_SATURN) {
893 			writel((cp->phy_type & CAS_PHY_MII_MDIO0) ?
894 			       SATURN_PCFG_FSI : 0x0,
895 			       cp->regs + REG_SATURN_PCFG);
896 
897 			/* load firmware to address 10Mbps auto-negotiation
898 			 * issue. NOTE: this will need to be changed if the
899 			 * default firmware gets fixed.
900 			 */
901 			if (PHY_NS_DP83065 == cp->phy_id) {
902 				cas_saturn_firmware_load(cp);
903 			}
904 			cas_phy_powerup(cp);
905 		}
906 
907 		/* advertise capabilities */
908 		val = cas_phy_read(cp, MII_BMCR);
909 		val &= ~BMCR_ANENABLE;
910 		cas_phy_write(cp, MII_BMCR, val);
911 		udelay(10);
912 
913 		cas_phy_write(cp, MII_ADVERTISE,
914 			      cas_phy_read(cp, MII_ADVERTISE) |
915 			      (ADVERTISE_10HALF | ADVERTISE_10FULL |
916 			       ADVERTISE_100HALF | ADVERTISE_100FULL |
917 			       CAS_ADVERTISE_PAUSE |
918 			       CAS_ADVERTISE_ASYM_PAUSE));
919 
920 		if (cp->cas_flags & CAS_FLAG_1000MB_CAP) {
921 			/* make sure that we don't advertise half
922 			 * duplex to avoid a chip issue
923 			 */
924 			val  = cas_phy_read(cp, CAS_MII_1000_CTRL);
925 			val &= ~CAS_ADVERTISE_1000HALF;
926 			val |= CAS_ADVERTISE_1000FULL;
927 			cas_phy_write(cp, CAS_MII_1000_CTRL, val);
928 		}
929 
930 	} else {
931 		/* reset pcs for serdes */
932 		u32 val;
933 		int limit;
934 
935 		writel(PCS_DATAPATH_MODE_SERDES,
936 		       cp->regs + REG_PCS_DATAPATH_MODE);
937 
938 		/* enable serdes pins on saturn */
939 		if (cp->cas_flags & CAS_FLAG_SATURN)
940 			writel(0, cp->regs + REG_SATURN_PCFG);
941 
942 		/* Reset PCS unit. */
943 		val = readl(cp->regs + REG_PCS_MII_CTRL);
944 		val |= PCS_MII_RESET;
945 		writel(val, cp->regs + REG_PCS_MII_CTRL);
946 
947 		limit = STOP_TRIES;
948 		while (--limit > 0) {
949 			udelay(10);
950 			if ((readl(cp->regs + REG_PCS_MII_CTRL) &
951 			     PCS_MII_RESET) == 0)
952 				break;
953 		}
954 		if (limit <= 0)
955 			netdev_warn(cp->dev, "PCS reset bit would not clear [%08x]\n",
956 				    readl(cp->regs + REG_PCS_STATE_MACHINE));
957 
958 		/* Make sure PCS is disabled while changing advertisement
959 		 * configuration.
960 		 */
961 		writel(0x0, cp->regs + REG_PCS_CFG);
962 
963 		/* Advertise all capabilities except half-duplex. */
964 		val  = readl(cp->regs + REG_PCS_MII_ADVERT);
965 		val &= ~PCS_MII_ADVERT_HD;
966 		val |= (PCS_MII_ADVERT_FD | PCS_MII_ADVERT_SYM_PAUSE |
967 			PCS_MII_ADVERT_ASYM_PAUSE);
968 		writel(val, cp->regs + REG_PCS_MII_ADVERT);
969 
970 		/* enable PCS */
971 		writel(PCS_CFG_EN, cp->regs + REG_PCS_CFG);
972 
973 		/* pcs workaround: enable sync detect */
974 		writel(PCS_SERDES_CTRL_SYNCD_EN,
975 		       cp->regs + REG_PCS_SERDES_CTRL);
976 	}
977 }
978 
979 
980 static int cas_pcs_link_check(struct cas *cp)
981 {
982 	u32 stat, state_machine;
983 	int retval = 0;
984 
985 	/* The link status bit latches on zero, so you must
986 	 * read it twice in such a case to see a transition
987 	 * to the link being up.
988 	 */
989 	stat = readl(cp->regs + REG_PCS_MII_STATUS);
990 	if ((stat & PCS_MII_STATUS_LINK_STATUS) == 0)
991 		stat = readl(cp->regs + REG_PCS_MII_STATUS);
992 
993 	/* The remote-fault indication is only valid
994 	 * when autoneg has completed.
995 	 */
996 	if ((stat & (PCS_MII_STATUS_AUTONEG_COMP |
997 		     PCS_MII_STATUS_REMOTE_FAULT)) ==
998 	    (PCS_MII_STATUS_AUTONEG_COMP | PCS_MII_STATUS_REMOTE_FAULT))
999 		netif_info(cp, link, cp->dev, "PCS RemoteFault\n");
1000 
1001 	/* work around link detection issue by querying the PCS state
1002 	 * machine directly.
1003 	 */
1004 	state_machine = readl(cp->regs + REG_PCS_STATE_MACHINE);
1005 	if ((state_machine & PCS_SM_LINK_STATE_MASK) != SM_LINK_STATE_UP) {
1006 		stat &= ~PCS_MII_STATUS_LINK_STATUS;
1007 	} else if (state_machine & PCS_SM_WORD_SYNC_STATE_MASK) {
1008 		stat |= PCS_MII_STATUS_LINK_STATUS;
1009 	}
1010 
1011 	if (stat & PCS_MII_STATUS_LINK_STATUS) {
1012 		if (cp->lstate != link_up) {
1013 			if (cp->opened) {
1014 				cp->lstate = link_up;
1015 				cp->link_transition = LINK_TRANSITION_LINK_UP;
1016 
1017 				cas_set_link_modes(cp);
1018 				netif_carrier_on(cp->dev);
1019 			}
1020 		}
1021 	} else if (cp->lstate == link_up) {
1022 		cp->lstate = link_down;
1023 		if (link_transition_timeout != 0 &&
1024 		    cp->link_transition != LINK_TRANSITION_REQUESTED_RESET &&
1025 		    !cp->link_transition_jiffies_valid) {
1026 			/*
1027 			 * force a reset, as a workaround for the
1028 			 * link-failure problem. May want to move this to a
1029 			 * point a bit earlier in the sequence. If we had
1030 			 * generated a reset a short time ago, we'll wait for
1031 			 * the link timer to check the status until a
1032 			 * timer expires (link_transistion_jiffies_valid is
1033 			 * true when the timer is running.)  Instead of using
1034 			 * a system timer, we just do a check whenever the
1035 			 * link timer is running - this clears the flag after
1036 			 * a suitable delay.
1037 			 */
1038 			retval = 1;
1039 			cp->link_transition = LINK_TRANSITION_REQUESTED_RESET;
1040 			cp->link_transition_jiffies = jiffies;
1041 			cp->link_transition_jiffies_valid = 1;
1042 		} else {
1043 			cp->link_transition = LINK_TRANSITION_ON_FAILURE;
1044 		}
1045 		netif_carrier_off(cp->dev);
1046 		if (cp->opened)
1047 			netif_info(cp, link, cp->dev, "PCS link down\n");
1048 
1049 		/* Cassini only: if you force a mode, there can be
1050 		 * sync problems on link down. to fix that, the following
1051 		 * things need to be checked:
1052 		 * 1) read serialink state register
1053 		 * 2) read pcs status register to verify link down.
1054 		 * 3) if link down and serial link == 0x03, then you need
1055 		 *    to global reset the chip.
1056 		 */
1057 		if ((cp->cas_flags & CAS_FLAG_REG_PLUS) == 0) {
1058 			/* should check to see if we're in a forced mode */
1059 			stat = readl(cp->regs + REG_PCS_SERDES_STATE);
1060 			if (stat == 0x03)
1061 				return 1;
1062 		}
1063 	} else if (cp->lstate == link_down) {
1064 		if (link_transition_timeout != 0 &&
1065 		    cp->link_transition != LINK_TRANSITION_REQUESTED_RESET &&
1066 		    !cp->link_transition_jiffies_valid) {
1067 			/* force a reset, as a workaround for the
1068 			 * link-failure problem.  May want to move
1069 			 * this to a point a bit earlier in the
1070 			 * sequence.
1071 			 */
1072 			retval = 1;
1073 			cp->link_transition = LINK_TRANSITION_REQUESTED_RESET;
1074 			cp->link_transition_jiffies = jiffies;
1075 			cp->link_transition_jiffies_valid = 1;
1076 		} else {
1077 			cp->link_transition = LINK_TRANSITION_STILL_FAILED;
1078 		}
1079 	}
1080 
1081 	return retval;
1082 }
1083 
1084 static int cas_pcs_interrupt(struct net_device *dev,
1085 			     struct cas *cp, u32 status)
1086 {
1087 	u32 stat = readl(cp->regs + REG_PCS_INTR_STATUS);
1088 
1089 	if ((stat & PCS_INTR_STATUS_LINK_CHANGE) == 0)
1090 		return 0;
1091 	return cas_pcs_link_check(cp);
1092 }
1093 
1094 static int cas_txmac_interrupt(struct net_device *dev,
1095 			       struct cas *cp, u32 status)
1096 {
1097 	u32 txmac_stat = readl(cp->regs + REG_MAC_TX_STATUS);
1098 
1099 	if (!txmac_stat)
1100 		return 0;
1101 
1102 	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
1103 		     "txmac interrupt, txmac_stat: 0x%x\n", txmac_stat);
1104 
1105 	/* Defer timer expiration is quite normal,
1106 	 * don't even log the event.
1107 	 */
1108 	if ((txmac_stat & MAC_TX_DEFER_TIMER) &&
1109 	    !(txmac_stat & ~MAC_TX_DEFER_TIMER))
1110 		return 0;
1111 
1112 	spin_lock(&cp->stat_lock[0]);
1113 	if (txmac_stat & MAC_TX_UNDERRUN) {
1114 		netdev_err(dev, "TX MAC xmit underrun\n");
1115 		cp->net_stats[0].tx_fifo_errors++;
1116 	}
1117 
1118 	if (txmac_stat & MAC_TX_MAX_PACKET_ERR) {
1119 		netdev_err(dev, "TX MAC max packet size error\n");
1120 		cp->net_stats[0].tx_errors++;
1121 	}
1122 
1123 	/* The rest are all cases of one of the 16-bit TX
1124 	 * counters expiring.
1125 	 */
1126 	if (txmac_stat & MAC_TX_COLL_NORMAL)
1127 		cp->net_stats[0].collisions += 0x10000;
1128 
1129 	if (txmac_stat & MAC_TX_COLL_EXCESS) {
1130 		cp->net_stats[0].tx_aborted_errors += 0x10000;
1131 		cp->net_stats[0].collisions += 0x10000;
1132 	}
1133 
1134 	if (txmac_stat & MAC_TX_COLL_LATE) {
1135 		cp->net_stats[0].tx_aborted_errors += 0x10000;
1136 		cp->net_stats[0].collisions += 0x10000;
1137 	}
1138 	spin_unlock(&cp->stat_lock[0]);
1139 
1140 	/* We do not keep track of MAC_TX_COLL_FIRST and
1141 	 * MAC_TX_PEAK_ATTEMPTS events.
1142 	 */
1143 	return 0;
1144 }
1145 
1146 static void cas_load_firmware(struct cas *cp, cas_hp_inst_t *firmware)
1147 {
1148 	cas_hp_inst_t *inst;
1149 	u32 val;
1150 	int i;
1151 
1152 	i = 0;
1153 	while ((inst = firmware) && inst->note) {
1154 		writel(i, cp->regs + REG_HP_INSTR_RAM_ADDR);
1155 
1156 		val = CAS_BASE(HP_INSTR_RAM_HI_VAL, inst->val);
1157 		val |= CAS_BASE(HP_INSTR_RAM_HI_MASK, inst->mask);
1158 		writel(val, cp->regs + REG_HP_INSTR_RAM_DATA_HI);
1159 
1160 		val = CAS_BASE(HP_INSTR_RAM_MID_OUTARG, inst->outarg >> 10);
1161 		val |= CAS_BASE(HP_INSTR_RAM_MID_OUTOP, inst->outop);
1162 		val |= CAS_BASE(HP_INSTR_RAM_MID_FNEXT, inst->fnext);
1163 		val |= CAS_BASE(HP_INSTR_RAM_MID_FOFF, inst->foff);
1164 		val |= CAS_BASE(HP_INSTR_RAM_MID_SNEXT, inst->snext);
1165 		val |= CAS_BASE(HP_INSTR_RAM_MID_SOFF, inst->soff);
1166 		val |= CAS_BASE(HP_INSTR_RAM_MID_OP, inst->op);
1167 		writel(val, cp->regs + REG_HP_INSTR_RAM_DATA_MID);
1168 
1169 		val = CAS_BASE(HP_INSTR_RAM_LOW_OUTMASK, inst->outmask);
1170 		val |= CAS_BASE(HP_INSTR_RAM_LOW_OUTSHIFT, inst->outshift);
1171 		val |= CAS_BASE(HP_INSTR_RAM_LOW_OUTEN, inst->outenab);
1172 		val |= CAS_BASE(HP_INSTR_RAM_LOW_OUTARG, inst->outarg);
1173 		writel(val, cp->regs + REG_HP_INSTR_RAM_DATA_LOW);
1174 		++firmware;
1175 		++i;
1176 	}
1177 }
1178 
1179 static void cas_init_rx_dma(struct cas *cp)
1180 {
1181 	u64 desc_dma = cp->block_dvma;
1182 	u32 val;
1183 	int i, size;
1184 
1185 	/* rx free descriptors */
1186 	val = CAS_BASE(RX_CFG_SWIVEL, RX_SWIVEL_OFF_VAL);
1187 	val |= CAS_BASE(RX_CFG_DESC_RING, RX_DESC_RINGN_INDEX(0));
1188 	val |= CAS_BASE(RX_CFG_COMP_RING, RX_COMP_RINGN_INDEX(0));
1189 	if ((N_RX_DESC_RINGS > 1) &&
1190 	    (cp->cas_flags & CAS_FLAG_REG_PLUS))  /* do desc 2 */
1191 		val |= CAS_BASE(RX_CFG_DESC_RING1, RX_DESC_RINGN_INDEX(1));
1192 	writel(val, cp->regs + REG_RX_CFG);
1193 
1194 	val = (unsigned long) cp->init_rxds[0] -
1195 		(unsigned long) cp->init_block;
1196 	writel((desc_dma + val) >> 32, cp->regs + REG_RX_DB_HI);
1197 	writel((desc_dma + val) & 0xffffffff, cp->regs + REG_RX_DB_LOW);
1198 	writel(RX_DESC_RINGN_SIZE(0) - 4, cp->regs + REG_RX_KICK);
1199 
1200 	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
1201 		/* rx desc 2 is for IPSEC packets. however,
1202 		 * we don't it that for that purpose.
1203 		 */
1204 		val = (unsigned long) cp->init_rxds[1] -
1205 			(unsigned long) cp->init_block;
1206 		writel((desc_dma + val) >> 32, cp->regs + REG_PLUS_RX_DB1_HI);
1207 		writel((desc_dma + val) & 0xffffffff, cp->regs +
1208 		       REG_PLUS_RX_DB1_LOW);
1209 		writel(RX_DESC_RINGN_SIZE(1) - 4, cp->regs +
1210 		       REG_PLUS_RX_KICK1);
1211 	}
1212 
1213 	/* rx completion registers */
1214 	val = (unsigned long) cp->init_rxcs[0] -
1215 		(unsigned long) cp->init_block;
1216 	writel((desc_dma + val) >> 32, cp->regs + REG_RX_CB_HI);
1217 	writel((desc_dma + val) & 0xffffffff, cp->regs + REG_RX_CB_LOW);
1218 
1219 	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
1220 		/* rx comp 2-4 */
1221 		for (i = 1; i < MAX_RX_COMP_RINGS; i++) {
1222 			val = (unsigned long) cp->init_rxcs[i] -
1223 				(unsigned long) cp->init_block;
1224 			writel((desc_dma + val) >> 32, cp->regs +
1225 			       REG_PLUS_RX_CBN_HI(i));
1226 			writel((desc_dma + val) & 0xffffffff, cp->regs +
1227 			       REG_PLUS_RX_CBN_LOW(i));
1228 		}
1229 	}
1230 
1231 	/* read selective clear regs to prevent spurious interrupts
1232 	 * on reset because complete == kick.
1233 	 * selective clear set up to prevent interrupts on resets
1234 	 */
1235 	readl(cp->regs + REG_INTR_STATUS_ALIAS);
1236 	writel(INTR_RX_DONE | INTR_RX_BUF_UNAVAIL, cp->regs + REG_ALIAS_CLEAR);
1237 	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
1238 		for (i = 1; i < N_RX_COMP_RINGS; i++)
1239 			readl(cp->regs + REG_PLUS_INTRN_STATUS_ALIAS(i));
1240 
1241 		/* 2 is different from 3 and 4 */
1242 		if (N_RX_COMP_RINGS > 1)
1243 			writel(INTR_RX_DONE_ALT | INTR_RX_BUF_UNAVAIL_1,
1244 			       cp->regs + REG_PLUS_ALIASN_CLEAR(1));
1245 
1246 		for (i = 2; i < N_RX_COMP_RINGS; i++)
1247 			writel(INTR_RX_DONE_ALT,
1248 			       cp->regs + REG_PLUS_ALIASN_CLEAR(i));
1249 	}
1250 
1251 	/* set up pause thresholds */
1252 	val  = CAS_BASE(RX_PAUSE_THRESH_OFF,
1253 			cp->rx_pause_off / RX_PAUSE_THRESH_QUANTUM);
1254 	val |= CAS_BASE(RX_PAUSE_THRESH_ON,
1255 			cp->rx_pause_on / RX_PAUSE_THRESH_QUANTUM);
1256 	writel(val, cp->regs + REG_RX_PAUSE_THRESH);
1257 
1258 	/* zero out dma reassembly buffers */
1259 	for (i = 0; i < 64; i++) {
1260 		writel(i, cp->regs + REG_RX_TABLE_ADDR);
1261 		writel(0x0, cp->regs + REG_RX_TABLE_DATA_LOW);
1262 		writel(0x0, cp->regs + REG_RX_TABLE_DATA_MID);
1263 		writel(0x0, cp->regs + REG_RX_TABLE_DATA_HI);
1264 	}
1265 
1266 	/* make sure address register is 0 for normal operation */
1267 	writel(0x0, cp->regs + REG_RX_CTRL_FIFO_ADDR);
1268 	writel(0x0, cp->regs + REG_RX_IPP_FIFO_ADDR);
1269 
1270 	/* interrupt mitigation */
1271 #ifdef USE_RX_BLANK
1272 	val = CAS_BASE(RX_BLANK_INTR_TIME, RX_BLANK_INTR_TIME_VAL);
1273 	val |= CAS_BASE(RX_BLANK_INTR_PKT, RX_BLANK_INTR_PKT_VAL);
1274 	writel(val, cp->regs + REG_RX_BLANK);
1275 #else
1276 	writel(0x0, cp->regs + REG_RX_BLANK);
1277 #endif
1278 
1279 	/* interrupt generation as a function of low water marks for
1280 	 * free desc and completion entries. these are used to trigger
1281 	 * housekeeping for rx descs. we don't use the free interrupt
1282 	 * as it's not very useful
1283 	 */
1284 	/* val = CAS_BASE(RX_AE_THRESH_FREE, RX_AE_FREEN_VAL(0)); */
1285 	val = CAS_BASE(RX_AE_THRESH_COMP, RX_AE_COMP_VAL);
1286 	writel(val, cp->regs + REG_RX_AE_THRESH);
1287 	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
1288 		val = CAS_BASE(RX_AE1_THRESH_FREE, RX_AE_FREEN_VAL(1));
1289 		writel(val, cp->regs + REG_PLUS_RX_AE1_THRESH);
1290 	}
1291 
1292 	/* Random early detect registers. useful for congestion avoidance.
1293 	 * this should be tunable.
1294 	 */
1295 	writel(0x0, cp->regs + REG_RX_RED);
1296 
1297 	/* receive page sizes. default == 2K (0x800) */
1298 	val = 0;
1299 	if (cp->page_size == 0x1000)
1300 		val = 0x1;
1301 	else if (cp->page_size == 0x2000)
1302 		val = 0x2;
1303 	else if (cp->page_size == 0x4000)
1304 		val = 0x3;
1305 
1306 	/* round mtu + offset. constrain to page size. */
1307 	size = cp->dev->mtu + 64;
1308 	if (size > cp->page_size)
1309 		size = cp->page_size;
1310 
1311 	if (size <= 0x400)
1312 		i = 0x0;
1313 	else if (size <= 0x800)
1314 		i = 0x1;
1315 	else if (size <= 0x1000)
1316 		i = 0x2;
1317 	else
1318 		i = 0x3;
1319 
1320 	cp->mtu_stride = 1 << (i + 10);
1321 	val  = CAS_BASE(RX_PAGE_SIZE, val);
1322 	val |= CAS_BASE(RX_PAGE_SIZE_MTU_STRIDE, i);
1323 	val |= CAS_BASE(RX_PAGE_SIZE_MTU_COUNT, cp->page_size >> (i + 10));
1324 	val |= CAS_BASE(RX_PAGE_SIZE_MTU_OFF, 0x1);
1325 	writel(val, cp->regs + REG_RX_PAGE_SIZE);
1326 
1327 	/* enable the header parser if desired */
1328 	if (CAS_HP_FIRMWARE == cas_prog_null)
1329 		return;
1330 
1331 	val = CAS_BASE(HP_CFG_NUM_CPU, CAS_NCPUS > 63 ? 0 : CAS_NCPUS);
1332 	val |= HP_CFG_PARSE_EN | HP_CFG_SYN_INC_MASK;
1333 	val |= CAS_BASE(HP_CFG_TCP_THRESH, HP_TCP_THRESH_VAL);
1334 	writel(val, cp->regs + REG_HP_CFG);
1335 }
1336 
1337 static inline void cas_rxc_init(struct cas_rx_comp *rxc)
1338 {
1339 	memset(rxc, 0, sizeof(*rxc));
1340 	rxc->word4 = cpu_to_le64(RX_COMP4_ZERO);
1341 }
1342 
1343 /* NOTE: we use the ENC RX DESC ring for spares. the rx_page[0,1]
1344  * flipping is protected by the fact that the chip will not
1345  * hand back the same page index while it's being processed.
1346  */
1347 static inline cas_page_t *cas_page_spare(struct cas *cp, const int index)
1348 {
1349 	cas_page_t *page = cp->rx_pages[1][index];
1350 	cas_page_t *new;
1351 
1352 	if (page_count(page->buffer) == 1)
1353 		return page;
1354 
1355 	new = cas_page_dequeue(cp);
1356 	if (new) {
1357 		spin_lock(&cp->rx_inuse_lock);
1358 		list_add(&page->list, &cp->rx_inuse_list);
1359 		spin_unlock(&cp->rx_inuse_lock);
1360 	}
1361 	return new;
1362 }
1363 
1364 /* this needs to be changed if we actually use the ENC RX DESC ring */
1365 static cas_page_t *cas_page_swap(struct cas *cp, const int ring,
1366 				 const int index)
1367 {
1368 	cas_page_t **page0 = cp->rx_pages[0];
1369 	cas_page_t **page1 = cp->rx_pages[1];
1370 
1371 	/* swap if buffer is in use */
1372 	if (page_count(page0[index]->buffer) > 1) {
1373 		cas_page_t *new = cas_page_spare(cp, index);
1374 		if (new) {
1375 			page1[index] = page0[index];
1376 			page0[index] = new;
1377 		}
1378 	}
1379 	RX_USED_SET(page0[index], 0);
1380 	return page0[index];
1381 }
1382 
1383 static void cas_clean_rxds(struct cas *cp)
1384 {
1385 	/* only clean ring 0 as ring 1 is used for spare buffers */
1386         struct cas_rx_desc *rxd = cp->init_rxds[0];
1387 	int i, size;
1388 
1389 	/* release all rx flows */
1390 	for (i = 0; i < N_RX_FLOWS; i++) {
1391 		struct sk_buff *skb;
1392 		while ((skb = __skb_dequeue(&cp->rx_flows[i]))) {
1393 			cas_skb_release(skb);
1394 		}
1395 	}
1396 
1397 	/* initialize descriptors */
1398 	size = RX_DESC_RINGN_SIZE(0);
1399 	for (i = 0; i < size; i++) {
1400 		cas_page_t *page = cas_page_swap(cp, 0, i);
1401 		rxd[i].buffer = cpu_to_le64(page->dma_addr);
1402 		rxd[i].index  = cpu_to_le64(CAS_BASE(RX_INDEX_NUM, i) |
1403 					    CAS_BASE(RX_INDEX_RING, 0));
1404 	}
1405 
1406 	cp->rx_old[0]  = RX_DESC_RINGN_SIZE(0) - 4;
1407 	cp->rx_last[0] = 0;
1408 	cp->cas_flags &= ~CAS_FLAG_RXD_POST(0);
1409 }
1410 
1411 static void cas_clean_rxcs(struct cas *cp)
1412 {
1413 	int i, j;
1414 
1415 	/* take ownership of rx comp descriptors */
1416 	memset(cp->rx_cur, 0, sizeof(*cp->rx_cur)*N_RX_COMP_RINGS);
1417 	memset(cp->rx_new, 0, sizeof(*cp->rx_new)*N_RX_COMP_RINGS);
1418 	for (i = 0; i < N_RX_COMP_RINGS; i++) {
1419 		struct cas_rx_comp *rxc = cp->init_rxcs[i];
1420 		for (j = 0; j < RX_COMP_RINGN_SIZE(i); j++) {
1421 			cas_rxc_init(rxc + j);
1422 		}
1423 	}
1424 }
1425 
1426 #if 0
1427 /* When we get a RX fifo overflow, the RX unit is probably hung
1428  * so we do the following.
1429  *
1430  * If any part of the reset goes wrong, we return 1 and that causes the
1431  * whole chip to be reset.
1432  */
1433 static int cas_rxmac_reset(struct cas *cp)
1434 {
1435 	struct net_device *dev = cp->dev;
1436 	int limit;
1437 	u32 val;
1438 
1439 	/* First, reset MAC RX. */
1440 	writel(cp->mac_rx_cfg & ~MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
1441 	for (limit = 0; limit < STOP_TRIES; limit++) {
1442 		if (!(readl(cp->regs + REG_MAC_RX_CFG) & MAC_RX_CFG_EN))
1443 			break;
1444 		udelay(10);
1445 	}
1446 	if (limit == STOP_TRIES) {
1447 		netdev_err(dev, "RX MAC will not disable, resetting whole chip\n");
1448 		return 1;
1449 	}
1450 
1451 	/* Second, disable RX DMA. */
1452 	writel(0, cp->regs + REG_RX_CFG);
1453 	for (limit = 0; limit < STOP_TRIES; limit++) {
1454 		if (!(readl(cp->regs + REG_RX_CFG) & RX_CFG_DMA_EN))
1455 			break;
1456 		udelay(10);
1457 	}
1458 	if (limit == STOP_TRIES) {
1459 		netdev_err(dev, "RX DMA will not disable, resetting whole chip\n");
1460 		return 1;
1461 	}
1462 
1463 	mdelay(5);
1464 
1465 	/* Execute RX reset command. */
1466 	writel(SW_RESET_RX, cp->regs + REG_SW_RESET);
1467 	for (limit = 0; limit < STOP_TRIES; limit++) {
1468 		if (!(readl(cp->regs + REG_SW_RESET) & SW_RESET_RX))
1469 			break;
1470 		udelay(10);
1471 	}
1472 	if (limit == STOP_TRIES) {
1473 		netdev_err(dev, "RX reset command will not execute, resetting whole chip\n");
1474 		return 1;
1475 	}
1476 
1477 	/* reset driver rx state */
1478 	cas_clean_rxds(cp);
1479 	cas_clean_rxcs(cp);
1480 
1481 	/* Now, reprogram the rest of RX unit. */
1482 	cas_init_rx_dma(cp);
1483 
1484 	/* re-enable */
1485 	val = readl(cp->regs + REG_RX_CFG);
1486 	writel(val | RX_CFG_DMA_EN, cp->regs + REG_RX_CFG);
1487 	writel(MAC_RX_FRAME_RECV, cp->regs + REG_MAC_RX_MASK);
1488 	val = readl(cp->regs + REG_MAC_RX_CFG);
1489 	writel(val | MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
1490 	return 0;
1491 }
1492 #endif
1493 
1494 static int cas_rxmac_interrupt(struct net_device *dev, struct cas *cp,
1495 			       u32 status)
1496 {
1497 	u32 stat = readl(cp->regs + REG_MAC_RX_STATUS);
1498 
1499 	if (!stat)
1500 		return 0;
1501 
1502 	netif_dbg(cp, intr, cp->dev, "rxmac interrupt, stat: 0x%x\n", stat);
1503 
1504 	/* these are all rollovers */
1505 	spin_lock(&cp->stat_lock[0]);
1506 	if (stat & MAC_RX_ALIGN_ERR)
1507 		cp->net_stats[0].rx_frame_errors += 0x10000;
1508 
1509 	if (stat & MAC_RX_CRC_ERR)
1510 		cp->net_stats[0].rx_crc_errors += 0x10000;
1511 
1512 	if (stat & MAC_RX_LEN_ERR)
1513 		cp->net_stats[0].rx_length_errors += 0x10000;
1514 
1515 	if (stat & MAC_RX_OVERFLOW) {
1516 		cp->net_stats[0].rx_over_errors++;
1517 		cp->net_stats[0].rx_fifo_errors++;
1518 	}
1519 
1520 	/* We do not track MAC_RX_FRAME_COUNT and MAC_RX_VIOL_ERR
1521 	 * events.
1522 	 */
1523 	spin_unlock(&cp->stat_lock[0]);
1524 	return 0;
1525 }
1526 
1527 static int cas_mac_interrupt(struct net_device *dev, struct cas *cp,
1528 			     u32 status)
1529 {
1530 	u32 stat = readl(cp->regs + REG_MAC_CTRL_STATUS);
1531 
1532 	if (!stat)
1533 		return 0;
1534 
1535 	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
1536 		     "mac interrupt, stat: 0x%x\n", stat);
1537 
1538 	/* This interrupt is just for pause frame and pause
1539 	 * tracking.  It is useful for diagnostics and debug
1540 	 * but probably by default we will mask these events.
1541 	 */
1542 	if (stat & MAC_CTRL_PAUSE_STATE)
1543 		cp->pause_entered++;
1544 
1545 	if (stat & MAC_CTRL_PAUSE_RECEIVED)
1546 		cp->pause_last_time_recvd = (stat >> 16);
1547 
1548 	return 0;
1549 }
1550 
1551 
1552 /* Must be invoked under cp->lock. */
1553 static inline int cas_mdio_link_not_up(struct cas *cp)
1554 {
1555 	u16 val;
1556 
1557 	switch (cp->lstate) {
1558 	case link_force_ret:
1559 		netif_info(cp, link, cp->dev, "Autoneg failed again, keeping forced mode\n");
1560 		cas_phy_write(cp, MII_BMCR, cp->link_fcntl);
1561 		cp->timer_ticks = 5;
1562 		cp->lstate = link_force_ok;
1563 		cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
1564 		break;
1565 
1566 	case link_aneg:
1567 		val = cas_phy_read(cp, MII_BMCR);
1568 
1569 		/* Try forced modes. we try things in the following order:
1570 		 * 1000 full -> 100 full/half -> 10 half
1571 		 */
1572 		val &= ~(BMCR_ANRESTART | BMCR_ANENABLE);
1573 		val |= BMCR_FULLDPLX;
1574 		val |= (cp->cas_flags & CAS_FLAG_1000MB_CAP) ?
1575 			CAS_BMCR_SPEED1000 : BMCR_SPEED100;
1576 		cas_phy_write(cp, MII_BMCR, val);
1577 		cp->timer_ticks = 5;
1578 		cp->lstate = link_force_try;
1579 		cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
1580 		break;
1581 
1582 	case link_force_try:
1583 		/* Downgrade from 1000 to 100 to 10 Mbps if necessary. */
1584 		val = cas_phy_read(cp, MII_BMCR);
1585 		cp->timer_ticks = 5;
1586 		if (val & CAS_BMCR_SPEED1000) { /* gigabit */
1587 			val &= ~CAS_BMCR_SPEED1000;
1588 			val |= (BMCR_SPEED100 | BMCR_FULLDPLX);
1589 			cas_phy_write(cp, MII_BMCR, val);
1590 			break;
1591 		}
1592 
1593 		if (val & BMCR_SPEED100) {
1594 			if (val & BMCR_FULLDPLX) /* fd failed */
1595 				val &= ~BMCR_FULLDPLX;
1596 			else { /* 100Mbps failed */
1597 				val &= ~BMCR_SPEED100;
1598 			}
1599 			cas_phy_write(cp, MII_BMCR, val);
1600 			break;
1601 		}
1602 	default:
1603 		break;
1604 	}
1605 	return 0;
1606 }
1607 
1608 
1609 /* must be invoked with cp->lock held */
1610 static int cas_mii_link_check(struct cas *cp, const u16 bmsr)
1611 {
1612 	int restart;
1613 
1614 	if (bmsr & BMSR_LSTATUS) {
1615 		/* Ok, here we got a link. If we had it due to a forced
1616 		 * fallback, and we were configured for autoneg, we
1617 		 * retry a short autoneg pass. If you know your hub is
1618 		 * broken, use ethtool ;)
1619 		 */
1620 		if ((cp->lstate == link_force_try) &&
1621 		    (cp->link_cntl & BMCR_ANENABLE)) {
1622 			cp->lstate = link_force_ret;
1623 			cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
1624 			cas_mif_poll(cp, 0);
1625 			cp->link_fcntl = cas_phy_read(cp, MII_BMCR);
1626 			cp->timer_ticks = 5;
1627 			if (cp->opened)
1628 				netif_info(cp, link, cp->dev,
1629 					   "Got link after fallback, retrying autoneg once...\n");
1630 			cas_phy_write(cp, MII_BMCR,
1631 				      cp->link_fcntl | BMCR_ANENABLE |
1632 				      BMCR_ANRESTART);
1633 			cas_mif_poll(cp, 1);
1634 
1635 		} else if (cp->lstate != link_up) {
1636 			cp->lstate = link_up;
1637 			cp->link_transition = LINK_TRANSITION_LINK_UP;
1638 
1639 			if (cp->opened) {
1640 				cas_set_link_modes(cp);
1641 				netif_carrier_on(cp->dev);
1642 			}
1643 		}
1644 		return 0;
1645 	}
1646 
1647 	/* link not up. if the link was previously up, we restart the
1648 	 * whole process
1649 	 */
1650 	restart = 0;
1651 	if (cp->lstate == link_up) {
1652 		cp->lstate = link_down;
1653 		cp->link_transition = LINK_TRANSITION_LINK_DOWN;
1654 
1655 		netif_carrier_off(cp->dev);
1656 		if (cp->opened)
1657 			netif_info(cp, link, cp->dev, "Link down\n");
1658 		restart = 1;
1659 
1660 	} else if (++cp->timer_ticks > 10)
1661 		cas_mdio_link_not_up(cp);
1662 
1663 	return restart;
1664 }
1665 
1666 static int cas_mif_interrupt(struct net_device *dev, struct cas *cp,
1667 			     u32 status)
1668 {
1669 	u32 stat = readl(cp->regs + REG_MIF_STATUS);
1670 	u16 bmsr;
1671 
1672 	/* check for a link change */
1673 	if (CAS_VAL(MIF_STATUS_POLL_STATUS, stat) == 0)
1674 		return 0;
1675 
1676 	bmsr = CAS_VAL(MIF_STATUS_POLL_DATA, stat);
1677 	return cas_mii_link_check(cp, bmsr);
1678 }
1679 
1680 static int cas_pci_interrupt(struct net_device *dev, struct cas *cp,
1681 			     u32 status)
1682 {
1683 	u32 stat = readl(cp->regs + REG_PCI_ERR_STATUS);
1684 
1685 	if (!stat)
1686 		return 0;
1687 
1688 	netdev_err(dev, "PCI error [%04x:%04x]",
1689 		   stat, readl(cp->regs + REG_BIM_DIAG));
1690 
1691 	/* cassini+ has this reserved */
1692 	if ((stat & PCI_ERR_BADACK) &&
1693 	    ((cp->cas_flags & CAS_FLAG_REG_PLUS) == 0))
1694 		pr_cont(" <No ACK64# during ABS64 cycle>");
1695 
1696 	if (stat & PCI_ERR_DTRTO)
1697 		pr_cont(" <Delayed transaction timeout>");
1698 	if (stat & PCI_ERR_OTHER)
1699 		pr_cont(" <other>");
1700 	if (stat & PCI_ERR_BIM_DMA_WRITE)
1701 		pr_cont(" <BIM DMA 0 write req>");
1702 	if (stat & PCI_ERR_BIM_DMA_READ)
1703 		pr_cont(" <BIM DMA 0 read req>");
1704 	pr_cont("\n");
1705 
1706 	if (stat & PCI_ERR_OTHER) {
1707 		int pci_errs;
1708 
1709 		/* Interrogate PCI config space for the
1710 		 * true cause.
1711 		 */
1712 		pci_errs = pci_status_get_and_clear_errors(cp->pdev);
1713 
1714 		netdev_err(dev, "PCI status errors[%04x]\n", pci_errs);
1715 		if (pci_errs & PCI_STATUS_PARITY)
1716 			netdev_err(dev, "PCI parity error detected\n");
1717 		if (pci_errs & PCI_STATUS_SIG_TARGET_ABORT)
1718 			netdev_err(dev, "PCI target abort\n");
1719 		if (pci_errs & PCI_STATUS_REC_TARGET_ABORT)
1720 			netdev_err(dev, "PCI master acks target abort\n");
1721 		if (pci_errs & PCI_STATUS_REC_MASTER_ABORT)
1722 			netdev_err(dev, "PCI master abort\n");
1723 		if (pci_errs & PCI_STATUS_SIG_SYSTEM_ERROR)
1724 			netdev_err(dev, "PCI system error SERR#\n");
1725 		if (pci_errs & PCI_STATUS_DETECTED_PARITY)
1726 			netdev_err(dev, "PCI parity error\n");
1727 	}
1728 
1729 	/* For all PCI errors, we should reset the chip. */
1730 	return 1;
1731 }
1732 
1733 /* All non-normal interrupt conditions get serviced here.
1734  * Returns non-zero if we should just exit the interrupt
1735  * handler right now (ie. if we reset the card which invalidates
1736  * all of the other original irq status bits).
1737  */
1738 static int cas_abnormal_irq(struct net_device *dev, struct cas *cp,
1739 			    u32 status)
1740 {
1741 	if (status & INTR_RX_TAG_ERROR) {
1742 		/* corrupt RX tag framing */
1743 		netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
1744 			     "corrupt rx tag framing\n");
1745 		spin_lock(&cp->stat_lock[0]);
1746 		cp->net_stats[0].rx_errors++;
1747 		spin_unlock(&cp->stat_lock[0]);
1748 		goto do_reset;
1749 	}
1750 
1751 	if (status & INTR_RX_LEN_MISMATCH) {
1752 		/* length mismatch. */
1753 		netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
1754 			     "length mismatch for rx frame\n");
1755 		spin_lock(&cp->stat_lock[0]);
1756 		cp->net_stats[0].rx_errors++;
1757 		spin_unlock(&cp->stat_lock[0]);
1758 		goto do_reset;
1759 	}
1760 
1761 	if (status & INTR_PCS_STATUS) {
1762 		if (cas_pcs_interrupt(dev, cp, status))
1763 			goto do_reset;
1764 	}
1765 
1766 	if (status & INTR_TX_MAC_STATUS) {
1767 		if (cas_txmac_interrupt(dev, cp, status))
1768 			goto do_reset;
1769 	}
1770 
1771 	if (status & INTR_RX_MAC_STATUS) {
1772 		if (cas_rxmac_interrupt(dev, cp, status))
1773 			goto do_reset;
1774 	}
1775 
1776 	if (status & INTR_MAC_CTRL_STATUS) {
1777 		if (cas_mac_interrupt(dev, cp, status))
1778 			goto do_reset;
1779 	}
1780 
1781 	if (status & INTR_MIF_STATUS) {
1782 		if (cas_mif_interrupt(dev, cp, status))
1783 			goto do_reset;
1784 	}
1785 
1786 	if (status & INTR_PCI_ERROR_STATUS) {
1787 		if (cas_pci_interrupt(dev, cp, status))
1788 			goto do_reset;
1789 	}
1790 	return 0;
1791 
1792 do_reset:
1793 #if 1
1794 	atomic_inc(&cp->reset_task_pending);
1795 	atomic_inc(&cp->reset_task_pending_all);
1796 	netdev_err(dev, "reset called in cas_abnormal_irq [0x%x]\n", status);
1797 	schedule_work(&cp->reset_task);
1798 #else
1799 	atomic_set(&cp->reset_task_pending, CAS_RESET_ALL);
1800 	netdev_err(dev, "reset called in cas_abnormal_irq\n");
1801 	schedule_work(&cp->reset_task);
1802 #endif
1803 	return 1;
1804 }
1805 
1806 /* NOTE: CAS_TABORT returns 1 or 2 so that it can be used when
1807  *       determining whether to do a netif_stop/wakeup
1808  */
1809 #define CAS_TABORT(x)      (((x)->cas_flags & CAS_FLAG_TARGET_ABORT) ? 2 : 1)
1810 #define CAS_ROUND_PAGE(x)  (((x) + PAGE_SIZE - 1) & PAGE_MASK)
1811 static inline int cas_calc_tabort(struct cas *cp, const unsigned long addr,
1812 				  const int len)
1813 {
1814 	unsigned long off = addr + len;
1815 
1816 	if (CAS_TABORT(cp) == 1)
1817 		return 0;
1818 	if ((CAS_ROUND_PAGE(off) - off) > TX_TARGET_ABORT_LEN)
1819 		return 0;
1820 	return TX_TARGET_ABORT_LEN;
1821 }
1822 
1823 static inline void cas_tx_ringN(struct cas *cp, int ring, int limit)
1824 {
1825 	struct cas_tx_desc *txds;
1826 	struct sk_buff **skbs;
1827 	struct net_device *dev = cp->dev;
1828 	int entry, count;
1829 
1830 	spin_lock(&cp->tx_lock[ring]);
1831 	txds = cp->init_txds[ring];
1832 	skbs = cp->tx_skbs[ring];
1833 	entry = cp->tx_old[ring];
1834 
1835 	count = TX_BUFF_COUNT(ring, entry, limit);
1836 	while (entry != limit) {
1837 		struct sk_buff *skb = skbs[entry];
1838 		dma_addr_t daddr;
1839 		u32 dlen;
1840 		int frag;
1841 
1842 		if (!skb) {
1843 			/* this should never occur */
1844 			entry = TX_DESC_NEXT(ring, entry);
1845 			continue;
1846 		}
1847 
1848 		/* however, we might get only a partial skb release. */
1849 		count -= skb_shinfo(skb)->nr_frags +
1850 			+ cp->tx_tiny_use[ring][entry].nbufs + 1;
1851 		if (count < 0)
1852 			break;
1853 
1854 		netif_printk(cp, tx_done, KERN_DEBUG, cp->dev,
1855 			     "tx[%d] done, slot %d\n", ring, entry);
1856 
1857 		skbs[entry] = NULL;
1858 		cp->tx_tiny_use[ring][entry].nbufs = 0;
1859 
1860 		for (frag = 0; frag <= skb_shinfo(skb)->nr_frags; frag++) {
1861 			struct cas_tx_desc *txd = txds + entry;
1862 
1863 			daddr = le64_to_cpu(txd->buffer);
1864 			dlen = CAS_VAL(TX_DESC_BUFLEN,
1865 				       le64_to_cpu(txd->control));
1866 			dma_unmap_page(&cp->pdev->dev, daddr, dlen,
1867 				       DMA_TO_DEVICE);
1868 			entry = TX_DESC_NEXT(ring, entry);
1869 
1870 			/* tiny buffer may follow */
1871 			if (cp->tx_tiny_use[ring][entry].used) {
1872 				cp->tx_tiny_use[ring][entry].used = 0;
1873 				entry = TX_DESC_NEXT(ring, entry);
1874 			}
1875 		}
1876 
1877 		spin_lock(&cp->stat_lock[ring]);
1878 		cp->net_stats[ring].tx_packets++;
1879 		cp->net_stats[ring].tx_bytes += skb->len;
1880 		spin_unlock(&cp->stat_lock[ring]);
1881 		dev_consume_skb_irq(skb);
1882 	}
1883 	cp->tx_old[ring] = entry;
1884 
1885 	/* this is wrong for multiple tx rings. the net device needs
1886 	 * multiple queues for this to do the right thing.  we wait
1887 	 * for 2*packets to be available when using tiny buffers
1888 	 */
1889 	if (netif_queue_stopped(dev) &&
1890 	    (TX_BUFFS_AVAIL(cp, ring) > CAS_TABORT(cp)*(MAX_SKB_FRAGS + 1)))
1891 		netif_wake_queue(dev);
1892 	spin_unlock(&cp->tx_lock[ring]);
1893 }
1894 
1895 static void cas_tx(struct net_device *dev, struct cas *cp,
1896 		   u32 status)
1897 {
1898         int limit, ring;
1899 #ifdef USE_TX_COMPWB
1900 	u64 compwb = le64_to_cpu(cp->init_block->tx_compwb);
1901 #endif
1902 	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
1903 		     "tx interrupt, status: 0x%x, %llx\n",
1904 		     status, (unsigned long long)compwb);
1905 	/* process all the rings */
1906 	for (ring = 0; ring < N_TX_RINGS; ring++) {
1907 #ifdef USE_TX_COMPWB
1908 		/* use the completion writeback registers */
1909 		limit = (CAS_VAL(TX_COMPWB_MSB, compwb) << 8) |
1910 			CAS_VAL(TX_COMPWB_LSB, compwb);
1911 		compwb = TX_COMPWB_NEXT(compwb);
1912 #else
1913 		limit = readl(cp->regs + REG_TX_COMPN(ring));
1914 #endif
1915 		if (cp->tx_old[ring] != limit)
1916 			cas_tx_ringN(cp, ring, limit);
1917 	}
1918 }
1919 
1920 
1921 static int cas_rx_process_pkt(struct cas *cp, struct cas_rx_comp *rxc,
1922 			      int entry, const u64 *words,
1923 			      struct sk_buff **skbref)
1924 {
1925 	int dlen, hlen, len, i, alloclen;
1926 	int off, swivel = RX_SWIVEL_OFF_VAL;
1927 	struct cas_page *page;
1928 	struct sk_buff *skb;
1929 	void *addr, *crcaddr;
1930 	__sum16 csum;
1931 	char *p;
1932 
1933 	hlen = CAS_VAL(RX_COMP2_HDR_SIZE, words[1]);
1934 	dlen = CAS_VAL(RX_COMP1_DATA_SIZE, words[0]);
1935 	len  = hlen + dlen;
1936 
1937 	if (RX_COPY_ALWAYS || (words[2] & RX_COMP3_SMALL_PKT))
1938 		alloclen = len;
1939 	else
1940 		alloclen = max(hlen, RX_COPY_MIN);
1941 
1942 	skb = netdev_alloc_skb(cp->dev, alloclen + swivel + cp->crc_size);
1943 	if (skb == NULL)
1944 		return -1;
1945 
1946 	*skbref = skb;
1947 	skb_reserve(skb, swivel);
1948 
1949 	p = skb->data;
1950 	addr = crcaddr = NULL;
1951 	if (hlen) { /* always copy header pages */
1952 		i = CAS_VAL(RX_COMP2_HDR_INDEX, words[1]);
1953 		page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
1954 		off = CAS_VAL(RX_COMP2_HDR_OFF, words[1]) * 0x100 +
1955 			swivel;
1956 
1957 		i = hlen;
1958 		if (!dlen) /* attach FCS */
1959 			i += cp->crc_size;
1960 		dma_sync_single_for_cpu(&cp->pdev->dev, page->dma_addr + off,
1961 					i, DMA_FROM_DEVICE);
1962 		addr = cas_page_map(page->buffer);
1963 		memcpy(p, addr + off, i);
1964 		dma_sync_single_for_device(&cp->pdev->dev,
1965 					   page->dma_addr + off, i,
1966 					   DMA_FROM_DEVICE);
1967 		cas_page_unmap(addr);
1968 		RX_USED_ADD(page, 0x100);
1969 		p += hlen;
1970 		swivel = 0;
1971 	}
1972 
1973 
1974 	if (alloclen < (hlen + dlen)) {
1975 		skb_frag_t *frag = skb_shinfo(skb)->frags;
1976 
1977 		/* normal or jumbo packets. we use frags */
1978 		i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
1979 		page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
1980 		off = CAS_VAL(RX_COMP1_DATA_OFF, words[0]) + swivel;
1981 
1982 		hlen = min(cp->page_size - off, dlen);
1983 		if (hlen < 0) {
1984 			netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
1985 				     "rx page overflow: %d\n", hlen);
1986 			dev_kfree_skb_irq(skb);
1987 			return -1;
1988 		}
1989 		i = hlen;
1990 		if (i == dlen)  /* attach FCS */
1991 			i += cp->crc_size;
1992 		dma_sync_single_for_cpu(&cp->pdev->dev, page->dma_addr + off,
1993 					i, DMA_FROM_DEVICE);
1994 
1995 		/* make sure we always copy a header */
1996 		swivel = 0;
1997 		if (p == (char *) skb->data) { /* not split */
1998 			addr = cas_page_map(page->buffer);
1999 			memcpy(p, addr + off, RX_COPY_MIN);
2000 			dma_sync_single_for_device(&cp->pdev->dev,
2001 						   page->dma_addr + off, i,
2002 						   DMA_FROM_DEVICE);
2003 			cas_page_unmap(addr);
2004 			off += RX_COPY_MIN;
2005 			swivel = RX_COPY_MIN;
2006 			RX_USED_ADD(page, cp->mtu_stride);
2007 		} else {
2008 			RX_USED_ADD(page, hlen);
2009 		}
2010 		skb_put(skb, alloclen);
2011 
2012 		skb_shinfo(skb)->nr_frags++;
2013 		skb->data_len += hlen - swivel;
2014 		skb->truesize += hlen - swivel;
2015 		skb->len      += hlen - swivel;
2016 
2017 		__skb_frag_set_page(frag, page->buffer);
2018 		__skb_frag_ref(frag);
2019 		skb_frag_off_set(frag, off);
2020 		skb_frag_size_set(frag, hlen - swivel);
2021 
2022 		/* any more data? */
2023 		if ((words[0] & RX_COMP1_SPLIT_PKT) && ((dlen -= hlen) > 0)) {
2024 			hlen = dlen;
2025 			off = 0;
2026 
2027 			i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
2028 			page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
2029 			dma_sync_single_for_cpu(&cp->pdev->dev,
2030 						page->dma_addr,
2031 						hlen + cp->crc_size,
2032 						DMA_FROM_DEVICE);
2033 			dma_sync_single_for_device(&cp->pdev->dev,
2034 						   page->dma_addr,
2035 						   hlen + cp->crc_size,
2036 						   DMA_FROM_DEVICE);
2037 
2038 			skb_shinfo(skb)->nr_frags++;
2039 			skb->data_len += hlen;
2040 			skb->len      += hlen;
2041 			frag++;
2042 
2043 			__skb_frag_set_page(frag, page->buffer);
2044 			__skb_frag_ref(frag);
2045 			skb_frag_off_set(frag, 0);
2046 			skb_frag_size_set(frag, hlen);
2047 			RX_USED_ADD(page, hlen + cp->crc_size);
2048 		}
2049 
2050 		if (cp->crc_size) {
2051 			addr = cas_page_map(page->buffer);
2052 			crcaddr  = addr + off + hlen;
2053 		}
2054 
2055 	} else {
2056 		/* copying packet */
2057 		if (!dlen)
2058 			goto end_copy_pkt;
2059 
2060 		i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
2061 		page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
2062 		off = CAS_VAL(RX_COMP1_DATA_OFF, words[0]) + swivel;
2063 		hlen = min(cp->page_size - off, dlen);
2064 		if (hlen < 0) {
2065 			netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
2066 				     "rx page overflow: %d\n", hlen);
2067 			dev_kfree_skb_irq(skb);
2068 			return -1;
2069 		}
2070 		i = hlen;
2071 		if (i == dlen) /* attach FCS */
2072 			i += cp->crc_size;
2073 		dma_sync_single_for_cpu(&cp->pdev->dev, page->dma_addr + off,
2074 					i, DMA_FROM_DEVICE);
2075 		addr = cas_page_map(page->buffer);
2076 		memcpy(p, addr + off, i);
2077 		dma_sync_single_for_device(&cp->pdev->dev,
2078 					   page->dma_addr + off, i,
2079 					   DMA_FROM_DEVICE);
2080 		cas_page_unmap(addr);
2081 		if (p == (char *) skb->data) /* not split */
2082 			RX_USED_ADD(page, cp->mtu_stride);
2083 		else
2084 			RX_USED_ADD(page, i);
2085 
2086 		/* any more data? */
2087 		if ((words[0] & RX_COMP1_SPLIT_PKT) && ((dlen -= hlen) > 0)) {
2088 			p += hlen;
2089 			i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
2090 			page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
2091 			dma_sync_single_for_cpu(&cp->pdev->dev,
2092 						page->dma_addr,
2093 						dlen + cp->crc_size,
2094 						DMA_FROM_DEVICE);
2095 			addr = cas_page_map(page->buffer);
2096 			memcpy(p, addr, dlen + cp->crc_size);
2097 			dma_sync_single_for_device(&cp->pdev->dev,
2098 						   page->dma_addr,
2099 						   dlen + cp->crc_size,
2100 						   DMA_FROM_DEVICE);
2101 			cas_page_unmap(addr);
2102 			RX_USED_ADD(page, dlen + cp->crc_size);
2103 		}
2104 end_copy_pkt:
2105 		if (cp->crc_size) {
2106 			addr    = NULL;
2107 			crcaddr = skb->data + alloclen;
2108 		}
2109 		skb_put(skb, alloclen);
2110 	}
2111 
2112 	csum = (__force __sum16)htons(CAS_VAL(RX_COMP4_TCP_CSUM, words[3]));
2113 	if (cp->crc_size) {
2114 		/* checksum includes FCS. strip it out. */
2115 		csum = csum_fold(csum_partial(crcaddr, cp->crc_size,
2116 					      csum_unfold(csum)));
2117 		if (addr)
2118 			cas_page_unmap(addr);
2119 	}
2120 	skb->protocol = eth_type_trans(skb, cp->dev);
2121 	if (skb->protocol == htons(ETH_P_IP)) {
2122 		skb->csum = csum_unfold(~csum);
2123 		skb->ip_summed = CHECKSUM_COMPLETE;
2124 	} else
2125 		skb_checksum_none_assert(skb);
2126 	return len;
2127 }
2128 
2129 
2130 /* we can handle up to 64 rx flows at a time. we do the same thing
2131  * as nonreassm except that we batch up the buffers.
2132  * NOTE: we currently just treat each flow as a bunch of packets that
2133  *       we pass up. a better way would be to coalesce the packets
2134  *       into a jumbo packet. to do that, we need to do the following:
2135  *       1) the first packet will have a clean split between header and
2136  *          data. save both.
2137  *       2) each time the next flow packet comes in, extend the
2138  *          data length and merge the checksums.
2139  *       3) on flow release, fix up the header.
2140  *       4) make sure the higher layer doesn't care.
2141  * because packets get coalesced, we shouldn't run into fragment count
2142  * issues.
2143  */
2144 static inline void cas_rx_flow_pkt(struct cas *cp, const u64 *words,
2145 				   struct sk_buff *skb)
2146 {
2147 	int flowid = CAS_VAL(RX_COMP3_FLOWID, words[2]) & (N_RX_FLOWS - 1);
2148 	struct sk_buff_head *flow = &cp->rx_flows[flowid];
2149 
2150 	/* this is protected at a higher layer, so no need to
2151 	 * do any additional locking here. stick the buffer
2152 	 * at the end.
2153 	 */
2154 	__skb_queue_tail(flow, skb);
2155 	if (words[0] & RX_COMP1_RELEASE_FLOW) {
2156 		while ((skb = __skb_dequeue(flow))) {
2157 			cas_skb_release(skb);
2158 		}
2159 	}
2160 }
2161 
2162 /* put rx descriptor back on ring. if a buffer is in use by a higher
2163  * layer, this will need to put in a replacement.
2164  */
2165 static void cas_post_page(struct cas *cp, const int ring, const int index)
2166 {
2167 	cas_page_t *new;
2168 	int entry;
2169 
2170 	entry = cp->rx_old[ring];
2171 
2172 	new = cas_page_swap(cp, ring, index);
2173 	cp->init_rxds[ring][entry].buffer = cpu_to_le64(new->dma_addr);
2174 	cp->init_rxds[ring][entry].index  =
2175 		cpu_to_le64(CAS_BASE(RX_INDEX_NUM, index) |
2176 			    CAS_BASE(RX_INDEX_RING, ring));
2177 
2178 	entry = RX_DESC_ENTRY(ring, entry + 1);
2179 	cp->rx_old[ring] = entry;
2180 
2181 	if (entry % 4)
2182 		return;
2183 
2184 	if (ring == 0)
2185 		writel(entry, cp->regs + REG_RX_KICK);
2186 	else if ((N_RX_DESC_RINGS > 1) &&
2187 		 (cp->cas_flags & CAS_FLAG_REG_PLUS))
2188 		writel(entry, cp->regs + REG_PLUS_RX_KICK1);
2189 }
2190 
2191 
2192 /* only when things are bad */
2193 static int cas_post_rxds_ringN(struct cas *cp, int ring, int num)
2194 {
2195 	unsigned int entry, last, count, released;
2196 	int cluster;
2197 	cas_page_t **page = cp->rx_pages[ring];
2198 
2199 	entry = cp->rx_old[ring];
2200 
2201 	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
2202 		     "rxd[%d] interrupt, done: %d\n", ring, entry);
2203 
2204 	cluster = -1;
2205 	count = entry & 0x3;
2206 	last = RX_DESC_ENTRY(ring, num ? entry + num - 4: entry - 4);
2207 	released = 0;
2208 	while (entry != last) {
2209 		/* make a new buffer if it's still in use */
2210 		if (page_count(page[entry]->buffer) > 1) {
2211 			cas_page_t *new = cas_page_dequeue(cp);
2212 			if (!new) {
2213 				/* let the timer know that we need to
2214 				 * do this again
2215 				 */
2216 				cp->cas_flags |= CAS_FLAG_RXD_POST(ring);
2217 				if (!timer_pending(&cp->link_timer))
2218 					mod_timer(&cp->link_timer, jiffies +
2219 						  CAS_LINK_FAST_TIMEOUT);
2220 				cp->rx_old[ring]  = entry;
2221 				cp->rx_last[ring] = num ? num - released : 0;
2222 				return -ENOMEM;
2223 			}
2224 			spin_lock(&cp->rx_inuse_lock);
2225 			list_add(&page[entry]->list, &cp->rx_inuse_list);
2226 			spin_unlock(&cp->rx_inuse_lock);
2227 			cp->init_rxds[ring][entry].buffer =
2228 				cpu_to_le64(new->dma_addr);
2229 			page[entry] = new;
2230 
2231 		}
2232 
2233 		if (++count == 4) {
2234 			cluster = entry;
2235 			count = 0;
2236 		}
2237 		released++;
2238 		entry = RX_DESC_ENTRY(ring, entry + 1);
2239 	}
2240 	cp->rx_old[ring] = entry;
2241 
2242 	if (cluster < 0)
2243 		return 0;
2244 
2245 	if (ring == 0)
2246 		writel(cluster, cp->regs + REG_RX_KICK);
2247 	else if ((N_RX_DESC_RINGS > 1) &&
2248 		 (cp->cas_flags & CAS_FLAG_REG_PLUS))
2249 		writel(cluster, cp->regs + REG_PLUS_RX_KICK1);
2250 	return 0;
2251 }
2252 
2253 
2254 /* process a completion ring. packets are set up in three basic ways:
2255  * small packets: should be copied header + data in single buffer.
2256  * large packets: header and data in a single buffer.
2257  * split packets: header in a separate buffer from data.
2258  *                data may be in multiple pages. data may be > 256
2259  *                bytes but in a single page.
2260  *
2261  * NOTE: RX page posting is done in this routine as well. while there's
2262  *       the capability of using multiple RX completion rings, it isn't
2263  *       really worthwhile due to the fact that the page posting will
2264  *       force serialization on the single descriptor ring.
2265  */
2266 static int cas_rx_ringN(struct cas *cp, int ring, int budget)
2267 {
2268 	struct cas_rx_comp *rxcs = cp->init_rxcs[ring];
2269 	int entry, drops;
2270 	int npackets = 0;
2271 
2272 	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
2273 		     "rx[%d] interrupt, done: %d/%d\n",
2274 		     ring,
2275 		     readl(cp->regs + REG_RX_COMP_HEAD), cp->rx_new[ring]);
2276 
2277 	entry = cp->rx_new[ring];
2278 	drops = 0;
2279 	while (1) {
2280 		struct cas_rx_comp *rxc = rxcs + entry;
2281 		struct sk_buff *skb;
2282 		int type, len;
2283 		u64 words[4];
2284 		int i, dring;
2285 
2286 		words[0] = le64_to_cpu(rxc->word1);
2287 		words[1] = le64_to_cpu(rxc->word2);
2288 		words[2] = le64_to_cpu(rxc->word3);
2289 		words[3] = le64_to_cpu(rxc->word4);
2290 
2291 		/* don't touch if still owned by hw */
2292 		type = CAS_VAL(RX_COMP1_TYPE, words[0]);
2293 		if (type == 0)
2294 			break;
2295 
2296 		/* hw hasn't cleared the zero bit yet */
2297 		if (words[3] & RX_COMP4_ZERO) {
2298 			break;
2299 		}
2300 
2301 		/* get info on the packet */
2302 		if (words[3] & (RX_COMP4_LEN_MISMATCH | RX_COMP4_BAD)) {
2303 			spin_lock(&cp->stat_lock[ring]);
2304 			cp->net_stats[ring].rx_errors++;
2305 			if (words[3] & RX_COMP4_LEN_MISMATCH)
2306 				cp->net_stats[ring].rx_length_errors++;
2307 			if (words[3] & RX_COMP4_BAD)
2308 				cp->net_stats[ring].rx_crc_errors++;
2309 			spin_unlock(&cp->stat_lock[ring]);
2310 
2311 			/* We'll just return it to Cassini. */
2312 		drop_it:
2313 			spin_lock(&cp->stat_lock[ring]);
2314 			++cp->net_stats[ring].rx_dropped;
2315 			spin_unlock(&cp->stat_lock[ring]);
2316 			goto next;
2317 		}
2318 
2319 		len = cas_rx_process_pkt(cp, rxc, entry, words, &skb);
2320 		if (len < 0) {
2321 			++drops;
2322 			goto drop_it;
2323 		}
2324 
2325 		/* see if it's a flow re-assembly or not. the driver
2326 		 * itself handles release back up.
2327 		 */
2328 		if (RX_DONT_BATCH || (type == 0x2)) {
2329 			/* non-reassm: these always get released */
2330 			cas_skb_release(skb);
2331 		} else {
2332 			cas_rx_flow_pkt(cp, words, skb);
2333 		}
2334 
2335 		spin_lock(&cp->stat_lock[ring]);
2336 		cp->net_stats[ring].rx_packets++;
2337 		cp->net_stats[ring].rx_bytes += len;
2338 		spin_unlock(&cp->stat_lock[ring]);
2339 
2340 	next:
2341 		npackets++;
2342 
2343 		/* should it be released? */
2344 		if (words[0] & RX_COMP1_RELEASE_HDR) {
2345 			i = CAS_VAL(RX_COMP2_HDR_INDEX, words[1]);
2346 			dring = CAS_VAL(RX_INDEX_RING, i);
2347 			i = CAS_VAL(RX_INDEX_NUM, i);
2348 			cas_post_page(cp, dring, i);
2349 		}
2350 
2351 		if (words[0] & RX_COMP1_RELEASE_DATA) {
2352 			i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
2353 			dring = CAS_VAL(RX_INDEX_RING, i);
2354 			i = CAS_VAL(RX_INDEX_NUM, i);
2355 			cas_post_page(cp, dring, i);
2356 		}
2357 
2358 		if (words[0] & RX_COMP1_RELEASE_NEXT) {
2359 			i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
2360 			dring = CAS_VAL(RX_INDEX_RING, i);
2361 			i = CAS_VAL(RX_INDEX_NUM, i);
2362 			cas_post_page(cp, dring, i);
2363 		}
2364 
2365 		/* skip to the next entry */
2366 		entry = RX_COMP_ENTRY(ring, entry + 1 +
2367 				      CAS_VAL(RX_COMP1_SKIP, words[0]));
2368 #ifdef USE_NAPI
2369 		if (budget && (npackets >= budget))
2370 			break;
2371 #endif
2372 	}
2373 	cp->rx_new[ring] = entry;
2374 
2375 	if (drops)
2376 		netdev_info(cp->dev, "Memory squeeze, deferring packet\n");
2377 	return npackets;
2378 }
2379 
2380 
2381 /* put completion entries back on the ring */
2382 static void cas_post_rxcs_ringN(struct net_device *dev,
2383 				struct cas *cp, int ring)
2384 {
2385 	struct cas_rx_comp *rxc = cp->init_rxcs[ring];
2386 	int last, entry;
2387 
2388 	last = cp->rx_cur[ring];
2389 	entry = cp->rx_new[ring];
2390 	netif_printk(cp, intr, KERN_DEBUG, dev,
2391 		     "rxc[%d] interrupt, done: %d/%d\n",
2392 		     ring, readl(cp->regs + REG_RX_COMP_HEAD), entry);
2393 
2394 	/* zero and re-mark descriptors */
2395 	while (last != entry) {
2396 		cas_rxc_init(rxc + last);
2397 		last = RX_COMP_ENTRY(ring, last + 1);
2398 	}
2399 	cp->rx_cur[ring] = last;
2400 
2401 	if (ring == 0)
2402 		writel(last, cp->regs + REG_RX_COMP_TAIL);
2403 	else if (cp->cas_flags & CAS_FLAG_REG_PLUS)
2404 		writel(last, cp->regs + REG_PLUS_RX_COMPN_TAIL(ring));
2405 }
2406 
2407 
2408 
2409 /* cassini can use all four PCI interrupts for the completion ring.
2410  * rings 3 and 4 are identical
2411  */
2412 #if defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
2413 static inline void cas_handle_irqN(struct net_device *dev,
2414 				   struct cas *cp, const u32 status,
2415 				   const int ring)
2416 {
2417 	if (status & (INTR_RX_COMP_FULL_ALT | INTR_RX_COMP_AF_ALT))
2418 		cas_post_rxcs_ringN(dev, cp, ring);
2419 }
2420 
2421 static irqreturn_t cas_interruptN(int irq, void *dev_id)
2422 {
2423 	struct net_device *dev = dev_id;
2424 	struct cas *cp = netdev_priv(dev);
2425 	unsigned long flags;
2426 	int ring = (irq == cp->pci_irq_INTC) ? 2 : 3;
2427 	u32 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(ring));
2428 
2429 	/* check for shared irq */
2430 	if (status == 0)
2431 		return IRQ_NONE;
2432 
2433 	spin_lock_irqsave(&cp->lock, flags);
2434 	if (status & INTR_RX_DONE_ALT) { /* handle rx separately */
2435 #ifdef USE_NAPI
2436 		cas_mask_intr(cp);
2437 		napi_schedule(&cp->napi);
2438 #else
2439 		cas_rx_ringN(cp, ring, 0);
2440 #endif
2441 		status &= ~INTR_RX_DONE_ALT;
2442 	}
2443 
2444 	if (status)
2445 		cas_handle_irqN(dev, cp, status, ring);
2446 	spin_unlock_irqrestore(&cp->lock, flags);
2447 	return IRQ_HANDLED;
2448 }
2449 #endif
2450 
2451 #ifdef USE_PCI_INTB
2452 /* everything but rx packets */
2453 static inline void cas_handle_irq1(struct cas *cp, const u32 status)
2454 {
2455 	if (status & INTR_RX_BUF_UNAVAIL_1) {
2456 		/* Frame arrived, no free RX buffers available.
2457 		 * NOTE: we can get this on a link transition. */
2458 		cas_post_rxds_ringN(cp, 1, 0);
2459 		spin_lock(&cp->stat_lock[1]);
2460 		cp->net_stats[1].rx_dropped++;
2461 		spin_unlock(&cp->stat_lock[1]);
2462 	}
2463 
2464 	if (status & INTR_RX_BUF_AE_1)
2465 		cas_post_rxds_ringN(cp, 1, RX_DESC_RINGN_SIZE(1) -
2466 				    RX_AE_FREEN_VAL(1));
2467 
2468 	if (status & (INTR_RX_COMP_AF | INTR_RX_COMP_FULL))
2469 		cas_post_rxcs_ringN(cp, 1);
2470 }
2471 
2472 /* ring 2 handles a few more events than 3 and 4 */
2473 static irqreturn_t cas_interrupt1(int irq, void *dev_id)
2474 {
2475 	struct net_device *dev = dev_id;
2476 	struct cas *cp = netdev_priv(dev);
2477 	unsigned long flags;
2478 	u32 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(1));
2479 
2480 	/* check for shared interrupt */
2481 	if (status == 0)
2482 		return IRQ_NONE;
2483 
2484 	spin_lock_irqsave(&cp->lock, flags);
2485 	if (status & INTR_RX_DONE_ALT) { /* handle rx separately */
2486 #ifdef USE_NAPI
2487 		cas_mask_intr(cp);
2488 		napi_schedule(&cp->napi);
2489 #else
2490 		cas_rx_ringN(cp, 1, 0);
2491 #endif
2492 		status &= ~INTR_RX_DONE_ALT;
2493 	}
2494 	if (status)
2495 		cas_handle_irq1(cp, status);
2496 	spin_unlock_irqrestore(&cp->lock, flags);
2497 	return IRQ_HANDLED;
2498 }
2499 #endif
2500 
2501 static inline void cas_handle_irq(struct net_device *dev,
2502 				  struct cas *cp, const u32 status)
2503 {
2504 	/* housekeeping interrupts */
2505 	if (status & INTR_ERROR_MASK)
2506 		cas_abnormal_irq(dev, cp, status);
2507 
2508 	if (status & INTR_RX_BUF_UNAVAIL) {
2509 		/* Frame arrived, no free RX buffers available.
2510 		 * NOTE: we can get this on a link transition.
2511 		 */
2512 		cas_post_rxds_ringN(cp, 0, 0);
2513 		spin_lock(&cp->stat_lock[0]);
2514 		cp->net_stats[0].rx_dropped++;
2515 		spin_unlock(&cp->stat_lock[0]);
2516 	} else if (status & INTR_RX_BUF_AE) {
2517 		cas_post_rxds_ringN(cp, 0, RX_DESC_RINGN_SIZE(0) -
2518 				    RX_AE_FREEN_VAL(0));
2519 	}
2520 
2521 	if (status & (INTR_RX_COMP_AF | INTR_RX_COMP_FULL))
2522 		cas_post_rxcs_ringN(dev, cp, 0);
2523 }
2524 
2525 static irqreturn_t cas_interrupt(int irq, void *dev_id)
2526 {
2527 	struct net_device *dev = dev_id;
2528 	struct cas *cp = netdev_priv(dev);
2529 	unsigned long flags;
2530 	u32 status = readl(cp->regs + REG_INTR_STATUS);
2531 
2532 	if (status == 0)
2533 		return IRQ_NONE;
2534 
2535 	spin_lock_irqsave(&cp->lock, flags);
2536 	if (status & (INTR_TX_ALL | INTR_TX_INTME)) {
2537 		cas_tx(dev, cp, status);
2538 		status &= ~(INTR_TX_ALL | INTR_TX_INTME);
2539 	}
2540 
2541 	if (status & INTR_RX_DONE) {
2542 #ifdef USE_NAPI
2543 		cas_mask_intr(cp);
2544 		napi_schedule(&cp->napi);
2545 #else
2546 		cas_rx_ringN(cp, 0, 0);
2547 #endif
2548 		status &= ~INTR_RX_DONE;
2549 	}
2550 
2551 	if (status)
2552 		cas_handle_irq(dev, cp, status);
2553 	spin_unlock_irqrestore(&cp->lock, flags);
2554 	return IRQ_HANDLED;
2555 }
2556 
2557 
2558 #ifdef USE_NAPI
2559 static int cas_poll(struct napi_struct *napi, int budget)
2560 {
2561 	struct cas *cp = container_of(napi, struct cas, napi);
2562 	struct net_device *dev = cp->dev;
2563 	int i, enable_intr, credits;
2564 	u32 status = readl(cp->regs + REG_INTR_STATUS);
2565 	unsigned long flags;
2566 
2567 	spin_lock_irqsave(&cp->lock, flags);
2568 	cas_tx(dev, cp, status);
2569 	spin_unlock_irqrestore(&cp->lock, flags);
2570 
2571 	/* NAPI rx packets. we spread the credits across all of the
2572 	 * rxc rings
2573 	 *
2574 	 * to make sure we're fair with the work we loop through each
2575 	 * ring N_RX_COMP_RING times with a request of
2576 	 * budget / N_RX_COMP_RINGS
2577 	 */
2578 	enable_intr = 1;
2579 	credits = 0;
2580 	for (i = 0; i < N_RX_COMP_RINGS; i++) {
2581 		int j;
2582 		for (j = 0; j < N_RX_COMP_RINGS; j++) {
2583 			credits += cas_rx_ringN(cp, j, budget / N_RX_COMP_RINGS);
2584 			if (credits >= budget) {
2585 				enable_intr = 0;
2586 				goto rx_comp;
2587 			}
2588 		}
2589 	}
2590 
2591 rx_comp:
2592 	/* final rx completion */
2593 	spin_lock_irqsave(&cp->lock, flags);
2594 	if (status)
2595 		cas_handle_irq(dev, cp, status);
2596 
2597 #ifdef USE_PCI_INTB
2598 	if (N_RX_COMP_RINGS > 1) {
2599 		status = readl(cp->regs + REG_PLUS_INTRN_STATUS(1));
2600 		if (status)
2601 			cas_handle_irq1(dev, cp, status);
2602 	}
2603 #endif
2604 
2605 #ifdef USE_PCI_INTC
2606 	if (N_RX_COMP_RINGS > 2) {
2607 		status = readl(cp->regs + REG_PLUS_INTRN_STATUS(2));
2608 		if (status)
2609 			cas_handle_irqN(dev, cp, status, 2);
2610 	}
2611 #endif
2612 
2613 #ifdef USE_PCI_INTD
2614 	if (N_RX_COMP_RINGS > 3) {
2615 		status = readl(cp->regs + REG_PLUS_INTRN_STATUS(3));
2616 		if (status)
2617 			cas_handle_irqN(dev, cp, status, 3);
2618 	}
2619 #endif
2620 	spin_unlock_irqrestore(&cp->lock, flags);
2621 	if (enable_intr) {
2622 		napi_complete(napi);
2623 		cas_unmask_intr(cp);
2624 	}
2625 	return credits;
2626 }
2627 #endif
2628 
2629 #ifdef CONFIG_NET_POLL_CONTROLLER
2630 static void cas_netpoll(struct net_device *dev)
2631 {
2632 	struct cas *cp = netdev_priv(dev);
2633 
2634 	cas_disable_irq(cp, 0);
2635 	cas_interrupt(cp->pdev->irq, dev);
2636 	cas_enable_irq(cp, 0);
2637 
2638 #ifdef USE_PCI_INTB
2639 	if (N_RX_COMP_RINGS > 1) {
2640 		/* cas_interrupt1(); */
2641 	}
2642 #endif
2643 #ifdef USE_PCI_INTC
2644 	if (N_RX_COMP_RINGS > 2) {
2645 		/* cas_interruptN(); */
2646 	}
2647 #endif
2648 #ifdef USE_PCI_INTD
2649 	if (N_RX_COMP_RINGS > 3) {
2650 		/* cas_interruptN(); */
2651 	}
2652 #endif
2653 }
2654 #endif
2655 
2656 static void cas_tx_timeout(struct net_device *dev, unsigned int txqueue)
2657 {
2658 	struct cas *cp = netdev_priv(dev);
2659 
2660 	netdev_err(dev, "transmit timed out, resetting\n");
2661 	if (!cp->hw_running) {
2662 		netdev_err(dev, "hrm.. hw not running!\n");
2663 		return;
2664 	}
2665 
2666 	netdev_err(dev, "MIF_STATE[%08x]\n",
2667 		   readl(cp->regs + REG_MIF_STATE_MACHINE));
2668 
2669 	netdev_err(dev, "MAC_STATE[%08x]\n",
2670 		   readl(cp->regs + REG_MAC_STATE_MACHINE));
2671 
2672 	netdev_err(dev, "TX_STATE[%08x:%08x:%08x] FIFO[%08x:%08x:%08x] SM1[%08x] SM2[%08x]\n",
2673 		   readl(cp->regs + REG_TX_CFG),
2674 		   readl(cp->regs + REG_MAC_TX_STATUS),
2675 		   readl(cp->regs + REG_MAC_TX_CFG),
2676 		   readl(cp->regs + REG_TX_FIFO_PKT_CNT),
2677 		   readl(cp->regs + REG_TX_FIFO_WRITE_PTR),
2678 		   readl(cp->regs + REG_TX_FIFO_READ_PTR),
2679 		   readl(cp->regs + REG_TX_SM_1),
2680 		   readl(cp->regs + REG_TX_SM_2));
2681 
2682 	netdev_err(dev, "RX_STATE[%08x:%08x:%08x]\n",
2683 		   readl(cp->regs + REG_RX_CFG),
2684 		   readl(cp->regs + REG_MAC_RX_STATUS),
2685 		   readl(cp->regs + REG_MAC_RX_CFG));
2686 
2687 	netdev_err(dev, "HP_STATE[%08x:%08x:%08x:%08x]\n",
2688 		   readl(cp->regs + REG_HP_STATE_MACHINE),
2689 		   readl(cp->regs + REG_HP_STATUS0),
2690 		   readl(cp->regs + REG_HP_STATUS1),
2691 		   readl(cp->regs + REG_HP_STATUS2));
2692 
2693 #if 1
2694 	atomic_inc(&cp->reset_task_pending);
2695 	atomic_inc(&cp->reset_task_pending_all);
2696 	schedule_work(&cp->reset_task);
2697 #else
2698 	atomic_set(&cp->reset_task_pending, CAS_RESET_ALL);
2699 	schedule_work(&cp->reset_task);
2700 #endif
2701 }
2702 
2703 static inline int cas_intme(int ring, int entry)
2704 {
2705 	/* Algorithm: IRQ every 1/2 of descriptors. */
2706 	if (!(entry & ((TX_DESC_RINGN_SIZE(ring) >> 1) - 1)))
2707 		return 1;
2708 	return 0;
2709 }
2710 
2711 
2712 static void cas_write_txd(struct cas *cp, int ring, int entry,
2713 			  dma_addr_t mapping, int len, u64 ctrl, int last)
2714 {
2715 	struct cas_tx_desc *txd = cp->init_txds[ring] + entry;
2716 
2717 	ctrl |= CAS_BASE(TX_DESC_BUFLEN, len);
2718 	if (cas_intme(ring, entry))
2719 		ctrl |= TX_DESC_INTME;
2720 	if (last)
2721 		ctrl |= TX_DESC_EOF;
2722 	txd->control = cpu_to_le64(ctrl);
2723 	txd->buffer = cpu_to_le64(mapping);
2724 }
2725 
2726 static inline void *tx_tiny_buf(struct cas *cp, const int ring,
2727 				const int entry)
2728 {
2729 	return cp->tx_tiny_bufs[ring] + TX_TINY_BUF_LEN*entry;
2730 }
2731 
2732 static inline dma_addr_t tx_tiny_map(struct cas *cp, const int ring,
2733 				     const int entry, const int tentry)
2734 {
2735 	cp->tx_tiny_use[ring][tentry].nbufs++;
2736 	cp->tx_tiny_use[ring][entry].used = 1;
2737 	return cp->tx_tiny_dvma[ring] + TX_TINY_BUF_LEN*entry;
2738 }
2739 
2740 static inline int cas_xmit_tx_ringN(struct cas *cp, int ring,
2741 				    struct sk_buff *skb)
2742 {
2743 	struct net_device *dev = cp->dev;
2744 	int entry, nr_frags, frag, tabort, tentry;
2745 	dma_addr_t mapping;
2746 	unsigned long flags;
2747 	u64 ctrl;
2748 	u32 len;
2749 
2750 	spin_lock_irqsave(&cp->tx_lock[ring], flags);
2751 
2752 	/* This is a hard error, log it. */
2753 	if (TX_BUFFS_AVAIL(cp, ring) <=
2754 	    CAS_TABORT(cp)*(skb_shinfo(skb)->nr_frags + 1)) {
2755 		netif_stop_queue(dev);
2756 		spin_unlock_irqrestore(&cp->tx_lock[ring], flags);
2757 		netdev_err(dev, "BUG! Tx Ring full when queue awake!\n");
2758 		return 1;
2759 	}
2760 
2761 	ctrl = 0;
2762 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
2763 		const u64 csum_start_off = skb_checksum_start_offset(skb);
2764 		const u64 csum_stuff_off = csum_start_off + skb->csum_offset;
2765 
2766 		ctrl =  TX_DESC_CSUM_EN |
2767 			CAS_BASE(TX_DESC_CSUM_START, csum_start_off) |
2768 			CAS_BASE(TX_DESC_CSUM_STUFF, csum_stuff_off);
2769 	}
2770 
2771 	entry = cp->tx_new[ring];
2772 	cp->tx_skbs[ring][entry] = skb;
2773 
2774 	nr_frags = skb_shinfo(skb)->nr_frags;
2775 	len = skb_headlen(skb);
2776 	mapping = dma_map_page(&cp->pdev->dev, virt_to_page(skb->data),
2777 			       offset_in_page(skb->data), len, DMA_TO_DEVICE);
2778 
2779 	tentry = entry;
2780 	tabort = cas_calc_tabort(cp, (unsigned long) skb->data, len);
2781 	if (unlikely(tabort)) {
2782 		/* NOTE: len is always >  tabort */
2783 		cas_write_txd(cp, ring, entry, mapping, len - tabort,
2784 			      ctrl | TX_DESC_SOF, 0);
2785 		entry = TX_DESC_NEXT(ring, entry);
2786 
2787 		skb_copy_from_linear_data_offset(skb, len - tabort,
2788 			      tx_tiny_buf(cp, ring, entry), tabort);
2789 		mapping = tx_tiny_map(cp, ring, entry, tentry);
2790 		cas_write_txd(cp, ring, entry, mapping, tabort, ctrl,
2791 			      (nr_frags == 0));
2792 	} else {
2793 		cas_write_txd(cp, ring, entry, mapping, len, ctrl |
2794 			      TX_DESC_SOF, (nr_frags == 0));
2795 	}
2796 	entry = TX_DESC_NEXT(ring, entry);
2797 
2798 	for (frag = 0; frag < nr_frags; frag++) {
2799 		const skb_frag_t *fragp = &skb_shinfo(skb)->frags[frag];
2800 
2801 		len = skb_frag_size(fragp);
2802 		mapping = skb_frag_dma_map(&cp->pdev->dev, fragp, 0, len,
2803 					   DMA_TO_DEVICE);
2804 
2805 		tabort = cas_calc_tabort(cp, skb_frag_off(fragp), len);
2806 		if (unlikely(tabort)) {
2807 			void *addr;
2808 
2809 			/* NOTE: len is always > tabort */
2810 			cas_write_txd(cp, ring, entry, mapping, len - tabort,
2811 				      ctrl, 0);
2812 			entry = TX_DESC_NEXT(ring, entry);
2813 
2814 			addr = cas_page_map(skb_frag_page(fragp));
2815 			memcpy(tx_tiny_buf(cp, ring, entry),
2816 			       addr + skb_frag_off(fragp) + len - tabort,
2817 			       tabort);
2818 			cas_page_unmap(addr);
2819 			mapping = tx_tiny_map(cp, ring, entry, tentry);
2820 			len     = tabort;
2821 		}
2822 
2823 		cas_write_txd(cp, ring, entry, mapping, len, ctrl,
2824 			      (frag + 1 == nr_frags));
2825 		entry = TX_DESC_NEXT(ring, entry);
2826 	}
2827 
2828 	cp->tx_new[ring] = entry;
2829 	if (TX_BUFFS_AVAIL(cp, ring) <= CAS_TABORT(cp)*(MAX_SKB_FRAGS + 1))
2830 		netif_stop_queue(dev);
2831 
2832 	netif_printk(cp, tx_queued, KERN_DEBUG, dev,
2833 		     "tx[%d] queued, slot %d, skblen %d, avail %d\n",
2834 		     ring, entry, skb->len, TX_BUFFS_AVAIL(cp, ring));
2835 	writel(entry, cp->regs + REG_TX_KICKN(ring));
2836 	spin_unlock_irqrestore(&cp->tx_lock[ring], flags);
2837 	return 0;
2838 }
2839 
2840 static netdev_tx_t cas_start_xmit(struct sk_buff *skb, struct net_device *dev)
2841 {
2842 	struct cas *cp = netdev_priv(dev);
2843 
2844 	/* this is only used as a load-balancing hint, so it doesn't
2845 	 * need to be SMP safe
2846 	 */
2847 	static int ring;
2848 
2849 	if (skb_padto(skb, cp->min_frame_size))
2850 		return NETDEV_TX_OK;
2851 
2852 	/* XXX: we need some higher-level QoS hooks to steer packets to
2853 	 *      individual queues.
2854 	 */
2855 	if (cas_xmit_tx_ringN(cp, ring++ & N_TX_RINGS_MASK, skb))
2856 		return NETDEV_TX_BUSY;
2857 	return NETDEV_TX_OK;
2858 }
2859 
2860 static void cas_init_tx_dma(struct cas *cp)
2861 {
2862 	u64 desc_dma = cp->block_dvma;
2863 	unsigned long off;
2864 	u32 val;
2865 	int i;
2866 
2867 	/* set up tx completion writeback registers. must be 8-byte aligned */
2868 #ifdef USE_TX_COMPWB
2869 	off = offsetof(struct cas_init_block, tx_compwb);
2870 	writel((desc_dma + off) >> 32, cp->regs + REG_TX_COMPWB_DB_HI);
2871 	writel((desc_dma + off) & 0xffffffff, cp->regs + REG_TX_COMPWB_DB_LOW);
2872 #endif
2873 
2874 	/* enable completion writebacks, enable paced mode,
2875 	 * disable read pipe, and disable pre-interrupt compwbs
2876 	 */
2877 	val =   TX_CFG_COMPWB_Q1 | TX_CFG_COMPWB_Q2 |
2878 		TX_CFG_COMPWB_Q3 | TX_CFG_COMPWB_Q4 |
2879 		TX_CFG_DMA_RDPIPE_DIS | TX_CFG_PACED_MODE |
2880 		TX_CFG_INTR_COMPWB_DIS;
2881 
2882 	/* write out tx ring info and tx desc bases */
2883 	for (i = 0; i < MAX_TX_RINGS; i++) {
2884 		off = (unsigned long) cp->init_txds[i] -
2885 			(unsigned long) cp->init_block;
2886 
2887 		val |= CAS_TX_RINGN_BASE(i);
2888 		writel((desc_dma + off) >> 32, cp->regs + REG_TX_DBN_HI(i));
2889 		writel((desc_dma + off) & 0xffffffff, cp->regs +
2890 		       REG_TX_DBN_LOW(i));
2891 		/* don't zero out the kick register here as the system
2892 		 * will wedge
2893 		 */
2894 	}
2895 	writel(val, cp->regs + REG_TX_CFG);
2896 
2897 	/* program max burst sizes. these numbers should be different
2898 	 * if doing QoS.
2899 	 */
2900 #ifdef USE_QOS
2901 	writel(0x800, cp->regs + REG_TX_MAXBURST_0);
2902 	writel(0x1600, cp->regs + REG_TX_MAXBURST_1);
2903 	writel(0x2400, cp->regs + REG_TX_MAXBURST_2);
2904 	writel(0x4800, cp->regs + REG_TX_MAXBURST_3);
2905 #else
2906 	writel(0x800, cp->regs + REG_TX_MAXBURST_0);
2907 	writel(0x800, cp->regs + REG_TX_MAXBURST_1);
2908 	writel(0x800, cp->regs + REG_TX_MAXBURST_2);
2909 	writel(0x800, cp->regs + REG_TX_MAXBURST_3);
2910 #endif
2911 }
2912 
2913 /* Must be invoked under cp->lock. */
2914 static inline void cas_init_dma(struct cas *cp)
2915 {
2916 	cas_init_tx_dma(cp);
2917 	cas_init_rx_dma(cp);
2918 }
2919 
2920 static void cas_process_mc_list(struct cas *cp)
2921 {
2922 	u16 hash_table[16];
2923 	u32 crc;
2924 	struct netdev_hw_addr *ha;
2925 	int i = 1;
2926 
2927 	memset(hash_table, 0, sizeof(hash_table));
2928 	netdev_for_each_mc_addr(ha, cp->dev) {
2929 		if (i <= CAS_MC_EXACT_MATCH_SIZE) {
2930 			/* use the alternate mac address registers for the
2931 			 * first 15 multicast addresses
2932 			 */
2933 			writel((ha->addr[4] << 8) | ha->addr[5],
2934 			       cp->regs + REG_MAC_ADDRN(i*3 + 0));
2935 			writel((ha->addr[2] << 8) | ha->addr[3],
2936 			       cp->regs + REG_MAC_ADDRN(i*3 + 1));
2937 			writel((ha->addr[0] << 8) | ha->addr[1],
2938 			       cp->regs + REG_MAC_ADDRN(i*3 + 2));
2939 			i++;
2940 		}
2941 		else {
2942 			/* use hw hash table for the next series of
2943 			 * multicast addresses
2944 			 */
2945 			crc = ether_crc_le(ETH_ALEN, ha->addr);
2946 			crc >>= 24;
2947 			hash_table[crc >> 4] |= 1 << (15 - (crc & 0xf));
2948 		}
2949 	}
2950 	for (i = 0; i < 16; i++)
2951 		writel(hash_table[i], cp->regs + REG_MAC_HASH_TABLEN(i));
2952 }
2953 
2954 /* Must be invoked under cp->lock. */
2955 static u32 cas_setup_multicast(struct cas *cp)
2956 {
2957 	u32 rxcfg = 0;
2958 	int i;
2959 
2960 	if (cp->dev->flags & IFF_PROMISC) {
2961 		rxcfg |= MAC_RX_CFG_PROMISC_EN;
2962 
2963 	} else if (cp->dev->flags & IFF_ALLMULTI) {
2964 	    	for (i=0; i < 16; i++)
2965 			writel(0xFFFF, cp->regs + REG_MAC_HASH_TABLEN(i));
2966 		rxcfg |= MAC_RX_CFG_HASH_FILTER_EN;
2967 
2968 	} else {
2969 		cas_process_mc_list(cp);
2970 		rxcfg |= MAC_RX_CFG_HASH_FILTER_EN;
2971 	}
2972 
2973 	return rxcfg;
2974 }
2975 
2976 /* must be invoked under cp->stat_lock[N_TX_RINGS] */
2977 static void cas_clear_mac_err(struct cas *cp)
2978 {
2979 	writel(0, cp->regs + REG_MAC_COLL_NORMAL);
2980 	writel(0, cp->regs + REG_MAC_COLL_FIRST);
2981 	writel(0, cp->regs + REG_MAC_COLL_EXCESS);
2982 	writel(0, cp->regs + REG_MAC_COLL_LATE);
2983 	writel(0, cp->regs + REG_MAC_TIMER_DEFER);
2984 	writel(0, cp->regs + REG_MAC_ATTEMPTS_PEAK);
2985 	writel(0, cp->regs + REG_MAC_RECV_FRAME);
2986 	writel(0, cp->regs + REG_MAC_LEN_ERR);
2987 	writel(0, cp->regs + REG_MAC_ALIGN_ERR);
2988 	writel(0, cp->regs + REG_MAC_FCS_ERR);
2989 	writel(0, cp->regs + REG_MAC_RX_CODE_ERR);
2990 }
2991 
2992 
2993 static void cas_mac_reset(struct cas *cp)
2994 {
2995 	int i;
2996 
2997 	/* do both TX and RX reset */
2998 	writel(0x1, cp->regs + REG_MAC_TX_RESET);
2999 	writel(0x1, cp->regs + REG_MAC_RX_RESET);
3000 
3001 	/* wait for TX */
3002 	i = STOP_TRIES;
3003 	while (i-- > 0) {
3004 		if (readl(cp->regs + REG_MAC_TX_RESET) == 0)
3005 			break;
3006 		udelay(10);
3007 	}
3008 
3009 	/* wait for RX */
3010 	i = STOP_TRIES;
3011 	while (i-- > 0) {
3012 		if (readl(cp->regs + REG_MAC_RX_RESET) == 0)
3013 			break;
3014 		udelay(10);
3015 	}
3016 
3017 	if (readl(cp->regs + REG_MAC_TX_RESET) |
3018 	    readl(cp->regs + REG_MAC_RX_RESET))
3019 		netdev_err(cp->dev, "mac tx[%d]/rx[%d] reset failed [%08x]\n",
3020 			   readl(cp->regs + REG_MAC_TX_RESET),
3021 			   readl(cp->regs + REG_MAC_RX_RESET),
3022 			   readl(cp->regs + REG_MAC_STATE_MACHINE));
3023 }
3024 
3025 
3026 /* Must be invoked under cp->lock. */
3027 static void cas_init_mac(struct cas *cp)
3028 {
3029 	unsigned char *e = &cp->dev->dev_addr[0];
3030 	int i;
3031 	cas_mac_reset(cp);
3032 
3033 	/* setup core arbitration weight register */
3034 	writel(CAWR_RR_DIS, cp->regs + REG_CAWR);
3035 
3036 #if !defined(CONFIG_SPARC64) && !defined(CONFIG_ALPHA)
3037 	/* set the infinite burst register for chips that don't have
3038 	 * pci issues.
3039 	 */
3040 	if ((cp->cas_flags & CAS_FLAG_TARGET_ABORT) == 0)
3041 		writel(INF_BURST_EN, cp->regs + REG_INF_BURST);
3042 #endif
3043 
3044 	writel(0x1BF0, cp->regs + REG_MAC_SEND_PAUSE);
3045 
3046 	writel(0x00, cp->regs + REG_MAC_IPG0);
3047 	writel(0x08, cp->regs + REG_MAC_IPG1);
3048 	writel(0x04, cp->regs + REG_MAC_IPG2);
3049 
3050 	/* change later for 802.3z */
3051 	writel(0x40, cp->regs + REG_MAC_SLOT_TIME);
3052 
3053 	/* min frame + FCS */
3054 	writel(ETH_ZLEN + 4, cp->regs + REG_MAC_FRAMESIZE_MIN);
3055 
3056 	/* Ethernet payload + header + FCS + optional VLAN tag. NOTE: we
3057 	 * specify the maximum frame size to prevent RX tag errors on
3058 	 * oversized frames.
3059 	 */
3060 	writel(CAS_BASE(MAC_FRAMESIZE_MAX_BURST, 0x2000) |
3061 	       CAS_BASE(MAC_FRAMESIZE_MAX_FRAME,
3062 			(CAS_MAX_MTU + ETH_HLEN + 4 + 4)),
3063 	       cp->regs + REG_MAC_FRAMESIZE_MAX);
3064 
3065 	/* NOTE: crc_size is used as a surrogate for half-duplex.
3066 	 * workaround saturn half-duplex issue by increasing preamble
3067 	 * size to 65 bytes.
3068 	 */
3069 	if ((cp->cas_flags & CAS_FLAG_SATURN) && cp->crc_size)
3070 		writel(0x41, cp->regs + REG_MAC_PA_SIZE);
3071 	else
3072 		writel(0x07, cp->regs + REG_MAC_PA_SIZE);
3073 	writel(0x04, cp->regs + REG_MAC_JAM_SIZE);
3074 	writel(0x10, cp->regs + REG_MAC_ATTEMPT_LIMIT);
3075 	writel(0x8808, cp->regs + REG_MAC_CTRL_TYPE);
3076 
3077 	writel((e[5] | (e[4] << 8)) & 0x3ff, cp->regs + REG_MAC_RANDOM_SEED);
3078 
3079 	writel(0, cp->regs + REG_MAC_ADDR_FILTER0);
3080 	writel(0, cp->regs + REG_MAC_ADDR_FILTER1);
3081 	writel(0, cp->regs + REG_MAC_ADDR_FILTER2);
3082 	writel(0, cp->regs + REG_MAC_ADDR_FILTER2_1_MASK);
3083 	writel(0, cp->regs + REG_MAC_ADDR_FILTER0_MASK);
3084 
3085 	/* setup mac address in perfect filter array */
3086 	for (i = 0; i < 45; i++)
3087 		writel(0x0, cp->regs + REG_MAC_ADDRN(i));
3088 
3089 	writel((e[4] << 8) | e[5], cp->regs + REG_MAC_ADDRN(0));
3090 	writel((e[2] << 8) | e[3], cp->regs + REG_MAC_ADDRN(1));
3091 	writel((e[0] << 8) | e[1], cp->regs + REG_MAC_ADDRN(2));
3092 
3093 	writel(0x0001, cp->regs + REG_MAC_ADDRN(42));
3094 	writel(0xc200, cp->regs + REG_MAC_ADDRN(43));
3095 	writel(0x0180, cp->regs + REG_MAC_ADDRN(44));
3096 
3097 	cp->mac_rx_cfg = cas_setup_multicast(cp);
3098 
3099 	spin_lock(&cp->stat_lock[N_TX_RINGS]);
3100 	cas_clear_mac_err(cp);
3101 	spin_unlock(&cp->stat_lock[N_TX_RINGS]);
3102 
3103 	/* Setup MAC interrupts.  We want to get all of the interesting
3104 	 * counter expiration events, but we do not want to hear about
3105 	 * normal rx/tx as the DMA engine tells us that.
3106 	 */
3107 	writel(MAC_TX_FRAME_XMIT, cp->regs + REG_MAC_TX_MASK);
3108 	writel(MAC_RX_FRAME_RECV, cp->regs + REG_MAC_RX_MASK);
3109 
3110 	/* Don't enable even the PAUSE interrupts for now, we
3111 	 * make no use of those events other than to record them.
3112 	 */
3113 	writel(0xffffffff, cp->regs + REG_MAC_CTRL_MASK);
3114 }
3115 
3116 /* Must be invoked under cp->lock. */
3117 static void cas_init_pause_thresholds(struct cas *cp)
3118 {
3119 	/* Calculate pause thresholds.  Setting the OFF threshold to the
3120 	 * full RX fifo size effectively disables PAUSE generation
3121 	 */
3122 	if (cp->rx_fifo_size <= (2 * 1024)) {
3123 		cp->rx_pause_off = cp->rx_pause_on = cp->rx_fifo_size;
3124 	} else {
3125 		int max_frame = (cp->dev->mtu + ETH_HLEN + 4 + 4 + 64) & ~63;
3126 		if (max_frame * 3 > cp->rx_fifo_size) {
3127 			cp->rx_pause_off = 7104;
3128 			cp->rx_pause_on  = 960;
3129 		} else {
3130 			int off = (cp->rx_fifo_size - (max_frame * 2));
3131 			int on = off - max_frame;
3132 			cp->rx_pause_off = off;
3133 			cp->rx_pause_on = on;
3134 		}
3135 	}
3136 }
3137 
3138 static int cas_vpd_match(const void __iomem *p, const char *str)
3139 {
3140 	int len = strlen(str) + 1;
3141 	int i;
3142 
3143 	for (i = 0; i < len; i++) {
3144 		if (readb(p + i) != str[i])
3145 			return 0;
3146 	}
3147 	return 1;
3148 }
3149 
3150 
3151 /* get the mac address by reading the vpd information in the rom.
3152  * also get the phy type and determine if there's an entropy generator.
3153  * NOTE: this is a bit convoluted for the following reasons:
3154  *  1) vpd info has order-dependent mac addresses for multinic cards
3155  *  2) the only way to determine the nic order is to use the slot
3156  *     number.
3157  *  3) fiber cards don't have bridges, so their slot numbers don't
3158  *     mean anything.
3159  *  4) we don't actually know we have a fiber card until after
3160  *     the mac addresses are parsed.
3161  */
3162 static int cas_get_vpd_info(struct cas *cp, unsigned char *dev_addr,
3163 			    const int offset)
3164 {
3165 	void __iomem *p = cp->regs + REG_EXPANSION_ROM_RUN_START;
3166 	void __iomem *base, *kstart;
3167 	int i, len;
3168 	int found = 0;
3169 #define VPD_FOUND_MAC        0x01
3170 #define VPD_FOUND_PHY        0x02
3171 
3172 	int phy_type = CAS_PHY_MII_MDIO0; /* default phy type */
3173 	int mac_off  = 0;
3174 
3175 #if defined(CONFIG_SPARC)
3176 	const unsigned char *addr;
3177 #endif
3178 
3179 	/* give us access to the PROM */
3180 	writel(BIM_LOCAL_DEV_PROM | BIM_LOCAL_DEV_PAD,
3181 	       cp->regs + REG_BIM_LOCAL_DEV_EN);
3182 
3183 	/* check for an expansion rom */
3184 	if (readb(p) != 0x55 || readb(p + 1) != 0xaa)
3185 		goto use_random_mac_addr;
3186 
3187 	/* search for beginning of vpd */
3188 	base = NULL;
3189 	for (i = 2; i < EXPANSION_ROM_SIZE; i++) {
3190 		/* check for PCIR */
3191 		if ((readb(p + i + 0) == 0x50) &&
3192 		    (readb(p + i + 1) == 0x43) &&
3193 		    (readb(p + i + 2) == 0x49) &&
3194 		    (readb(p + i + 3) == 0x52)) {
3195 			base = p + (readb(p + i + 8) |
3196 				    (readb(p + i + 9) << 8));
3197 			break;
3198 		}
3199 	}
3200 
3201 	if (!base || (readb(base) != 0x82))
3202 		goto use_random_mac_addr;
3203 
3204 	i = (readb(base + 1) | (readb(base + 2) << 8)) + 3;
3205 	while (i < EXPANSION_ROM_SIZE) {
3206 		if (readb(base + i) != 0x90) /* no vpd found */
3207 			goto use_random_mac_addr;
3208 
3209 		/* found a vpd field */
3210 		len = readb(base + i + 1) | (readb(base + i + 2) << 8);
3211 
3212 		/* extract keywords */
3213 		kstart = base + i + 3;
3214 		p = kstart;
3215 		while ((p - kstart) < len) {
3216 			int klen = readb(p + 2);
3217 			int j;
3218 			char type;
3219 
3220 			p += 3;
3221 
3222 			/* look for the following things:
3223 			 * -- correct length == 29
3224 			 * 3 (type) + 2 (size) +
3225 			 * 18 (strlen("local-mac-address") + 1) +
3226 			 * 6 (mac addr)
3227 			 * -- VPD Instance 'I'
3228 			 * -- VPD Type Bytes 'B'
3229 			 * -- VPD data length == 6
3230 			 * -- property string == local-mac-address
3231 			 *
3232 			 * -- correct length == 24
3233 			 * 3 (type) + 2 (size) +
3234 			 * 12 (strlen("entropy-dev") + 1) +
3235 			 * 7 (strlen("vms110") + 1)
3236 			 * -- VPD Instance 'I'
3237 			 * -- VPD Type String 'B'
3238 			 * -- VPD data length == 7
3239 			 * -- property string == entropy-dev
3240 			 *
3241 			 * -- correct length == 18
3242 			 * 3 (type) + 2 (size) +
3243 			 * 9 (strlen("phy-type") + 1) +
3244 			 * 4 (strlen("pcs") + 1)
3245 			 * -- VPD Instance 'I'
3246 			 * -- VPD Type String 'S'
3247 			 * -- VPD data length == 4
3248 			 * -- property string == phy-type
3249 			 *
3250 			 * -- correct length == 23
3251 			 * 3 (type) + 2 (size) +
3252 			 * 14 (strlen("phy-interface") + 1) +
3253 			 * 4 (strlen("pcs") + 1)
3254 			 * -- VPD Instance 'I'
3255 			 * -- VPD Type String 'S'
3256 			 * -- VPD data length == 4
3257 			 * -- property string == phy-interface
3258 			 */
3259 			if (readb(p) != 'I')
3260 				goto next;
3261 
3262 			/* finally, check string and length */
3263 			type = readb(p + 3);
3264 			if (type == 'B') {
3265 				if ((klen == 29) && readb(p + 4) == 6 &&
3266 				    cas_vpd_match(p + 5,
3267 						  "local-mac-address")) {
3268 					if (mac_off++ > offset)
3269 						goto next;
3270 
3271 					/* set mac address */
3272 					for (j = 0; j < 6; j++)
3273 						dev_addr[j] =
3274 							readb(p + 23 + j);
3275 					goto found_mac;
3276 				}
3277 			}
3278 
3279 			if (type != 'S')
3280 				goto next;
3281 
3282 #ifdef USE_ENTROPY_DEV
3283 			if ((klen == 24) &&
3284 			    cas_vpd_match(p + 5, "entropy-dev") &&
3285 			    cas_vpd_match(p + 17, "vms110")) {
3286 				cp->cas_flags |= CAS_FLAG_ENTROPY_DEV;
3287 				goto next;
3288 			}
3289 #endif
3290 
3291 			if (found & VPD_FOUND_PHY)
3292 				goto next;
3293 
3294 			if ((klen == 18) && readb(p + 4) == 4 &&
3295 			    cas_vpd_match(p + 5, "phy-type")) {
3296 				if (cas_vpd_match(p + 14, "pcs")) {
3297 					phy_type = CAS_PHY_SERDES;
3298 					goto found_phy;
3299 				}
3300 			}
3301 
3302 			if ((klen == 23) && readb(p + 4) == 4 &&
3303 			    cas_vpd_match(p + 5, "phy-interface")) {
3304 				if (cas_vpd_match(p + 19, "pcs")) {
3305 					phy_type = CAS_PHY_SERDES;
3306 					goto found_phy;
3307 				}
3308 			}
3309 found_mac:
3310 			found |= VPD_FOUND_MAC;
3311 			goto next;
3312 
3313 found_phy:
3314 			found |= VPD_FOUND_PHY;
3315 
3316 next:
3317 			p += klen;
3318 		}
3319 		i += len + 3;
3320 	}
3321 
3322 use_random_mac_addr:
3323 	if (found & VPD_FOUND_MAC)
3324 		goto done;
3325 
3326 #if defined(CONFIG_SPARC)
3327 	addr = of_get_property(cp->of_node, "local-mac-address", NULL);
3328 	if (addr != NULL) {
3329 		memcpy(dev_addr, addr, ETH_ALEN);
3330 		goto done;
3331 	}
3332 #endif
3333 
3334 	/* Sun MAC prefix then 3 random bytes. */
3335 	pr_info("MAC address not found in ROM VPD\n");
3336 	dev_addr[0] = 0x08;
3337 	dev_addr[1] = 0x00;
3338 	dev_addr[2] = 0x20;
3339 	get_random_bytes(dev_addr + 3, 3);
3340 
3341 done:
3342 	writel(0, cp->regs + REG_BIM_LOCAL_DEV_EN);
3343 	return phy_type;
3344 }
3345 
3346 /* check pci invariants */
3347 static void cas_check_pci_invariants(struct cas *cp)
3348 {
3349 	struct pci_dev *pdev = cp->pdev;
3350 
3351 	cp->cas_flags = 0;
3352 	if ((pdev->vendor == PCI_VENDOR_ID_SUN) &&
3353 	    (pdev->device == PCI_DEVICE_ID_SUN_CASSINI)) {
3354 		if (pdev->revision >= CAS_ID_REVPLUS)
3355 			cp->cas_flags |= CAS_FLAG_REG_PLUS;
3356 		if (pdev->revision < CAS_ID_REVPLUS02u)
3357 			cp->cas_flags |= CAS_FLAG_TARGET_ABORT;
3358 
3359 		/* Original Cassini supports HW CSUM, but it's not
3360 		 * enabled by default as it can trigger TX hangs.
3361 		 */
3362 		if (pdev->revision < CAS_ID_REV2)
3363 			cp->cas_flags |= CAS_FLAG_NO_HW_CSUM;
3364 	} else {
3365 		/* Only sun has original cassini chips.  */
3366 		cp->cas_flags |= CAS_FLAG_REG_PLUS;
3367 
3368 		/* We use a flag because the same phy might be externally
3369 		 * connected.
3370 		 */
3371 		if ((pdev->vendor == PCI_VENDOR_ID_NS) &&
3372 		    (pdev->device == PCI_DEVICE_ID_NS_SATURN))
3373 			cp->cas_flags |= CAS_FLAG_SATURN;
3374 	}
3375 }
3376 
3377 
3378 static int cas_check_invariants(struct cas *cp)
3379 {
3380 	struct pci_dev *pdev = cp->pdev;
3381 	u32 cfg;
3382 	int i;
3383 
3384 	/* get page size for rx buffers. */
3385 	cp->page_order = 0;
3386 #ifdef USE_PAGE_ORDER
3387 	if (PAGE_SHIFT < CAS_JUMBO_PAGE_SHIFT) {
3388 		/* see if we can allocate larger pages */
3389 		struct page *page = alloc_pages(GFP_ATOMIC,
3390 						CAS_JUMBO_PAGE_SHIFT -
3391 						PAGE_SHIFT);
3392 		if (page) {
3393 			__free_pages(page, CAS_JUMBO_PAGE_SHIFT - PAGE_SHIFT);
3394 			cp->page_order = CAS_JUMBO_PAGE_SHIFT - PAGE_SHIFT;
3395 		} else {
3396 			printk("MTU limited to %d bytes\n", CAS_MAX_MTU);
3397 		}
3398 	}
3399 #endif
3400 	cp->page_size = (PAGE_SIZE << cp->page_order);
3401 
3402 	/* Fetch the FIFO configurations. */
3403 	cp->tx_fifo_size = readl(cp->regs + REG_TX_FIFO_SIZE) * 64;
3404 	cp->rx_fifo_size = RX_FIFO_SIZE;
3405 
3406 	/* finish phy determination. MDIO1 takes precedence over MDIO0 if
3407 	 * they're both connected.
3408 	 */
3409 	cp->phy_type = cas_get_vpd_info(cp, cp->dev->dev_addr,
3410 					PCI_SLOT(pdev->devfn));
3411 	if (cp->phy_type & CAS_PHY_SERDES) {
3412 		cp->cas_flags |= CAS_FLAG_1000MB_CAP;
3413 		return 0; /* no more checking needed */
3414 	}
3415 
3416 	/* MII */
3417 	cfg = readl(cp->regs + REG_MIF_CFG);
3418 	if (cfg & MIF_CFG_MDIO_1) {
3419 		cp->phy_type = CAS_PHY_MII_MDIO1;
3420 	} else if (cfg & MIF_CFG_MDIO_0) {
3421 		cp->phy_type = CAS_PHY_MII_MDIO0;
3422 	}
3423 
3424 	cas_mif_poll(cp, 0);
3425 	writel(PCS_DATAPATH_MODE_MII, cp->regs + REG_PCS_DATAPATH_MODE);
3426 
3427 	for (i = 0; i < 32; i++) {
3428 		u32 phy_id;
3429 		int j;
3430 
3431 		for (j = 0; j < 3; j++) {
3432 			cp->phy_addr = i;
3433 			phy_id = cas_phy_read(cp, MII_PHYSID1) << 16;
3434 			phy_id |= cas_phy_read(cp, MII_PHYSID2);
3435 			if (phy_id && (phy_id != 0xFFFFFFFF)) {
3436 				cp->phy_id = phy_id;
3437 				goto done;
3438 			}
3439 		}
3440 	}
3441 	pr_err("MII phy did not respond [%08x]\n",
3442 	       readl(cp->regs + REG_MIF_STATE_MACHINE));
3443 	return -1;
3444 
3445 done:
3446 	/* see if we can do gigabit */
3447 	cfg = cas_phy_read(cp, MII_BMSR);
3448 	if ((cfg & CAS_BMSR_1000_EXTEND) &&
3449 	    cas_phy_read(cp, CAS_MII_1000_EXTEND))
3450 		cp->cas_flags |= CAS_FLAG_1000MB_CAP;
3451 	return 0;
3452 }
3453 
3454 /* Must be invoked under cp->lock. */
3455 static inline void cas_start_dma(struct cas *cp)
3456 {
3457 	int i;
3458 	u32 val;
3459 	int txfailed = 0;
3460 
3461 	/* enable dma */
3462 	val = readl(cp->regs + REG_TX_CFG) | TX_CFG_DMA_EN;
3463 	writel(val, cp->regs + REG_TX_CFG);
3464 	val = readl(cp->regs + REG_RX_CFG) | RX_CFG_DMA_EN;
3465 	writel(val, cp->regs + REG_RX_CFG);
3466 
3467 	/* enable the mac */
3468 	val = readl(cp->regs + REG_MAC_TX_CFG) | MAC_TX_CFG_EN;
3469 	writel(val, cp->regs + REG_MAC_TX_CFG);
3470 	val = readl(cp->regs + REG_MAC_RX_CFG) | MAC_RX_CFG_EN;
3471 	writel(val, cp->regs + REG_MAC_RX_CFG);
3472 
3473 	i = STOP_TRIES;
3474 	while (i-- > 0) {
3475 		val = readl(cp->regs + REG_MAC_TX_CFG);
3476 		if ((val & MAC_TX_CFG_EN))
3477 			break;
3478 		udelay(10);
3479 	}
3480 	if (i < 0) txfailed = 1;
3481 	i = STOP_TRIES;
3482 	while (i-- > 0) {
3483 		val = readl(cp->regs + REG_MAC_RX_CFG);
3484 		if ((val & MAC_RX_CFG_EN)) {
3485 			if (txfailed) {
3486 				netdev_err(cp->dev,
3487 					   "enabling mac failed [tx:%08x:%08x]\n",
3488 					   readl(cp->regs + REG_MIF_STATE_MACHINE),
3489 					   readl(cp->regs + REG_MAC_STATE_MACHINE));
3490 			}
3491 			goto enable_rx_done;
3492 		}
3493 		udelay(10);
3494 	}
3495 	netdev_err(cp->dev, "enabling mac failed [%s:%08x:%08x]\n",
3496 		   (txfailed ? "tx,rx" : "rx"),
3497 		   readl(cp->regs + REG_MIF_STATE_MACHINE),
3498 		   readl(cp->regs + REG_MAC_STATE_MACHINE));
3499 
3500 enable_rx_done:
3501 	cas_unmask_intr(cp); /* enable interrupts */
3502 	writel(RX_DESC_RINGN_SIZE(0) - 4, cp->regs + REG_RX_KICK);
3503 	writel(0, cp->regs + REG_RX_COMP_TAIL);
3504 
3505 	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
3506 		if (N_RX_DESC_RINGS > 1)
3507 			writel(RX_DESC_RINGN_SIZE(1) - 4,
3508 			       cp->regs + REG_PLUS_RX_KICK1);
3509 
3510 		for (i = 1; i < N_RX_COMP_RINGS; i++)
3511 			writel(0, cp->regs + REG_PLUS_RX_COMPN_TAIL(i));
3512 	}
3513 }
3514 
3515 /* Must be invoked under cp->lock. */
3516 static void cas_read_pcs_link_mode(struct cas *cp, int *fd, int *spd,
3517 				   int *pause)
3518 {
3519 	u32 val = readl(cp->regs + REG_PCS_MII_LPA);
3520 	*fd     = (val & PCS_MII_LPA_FD) ? 1 : 0;
3521 	*pause  = (val & PCS_MII_LPA_SYM_PAUSE) ? 0x01 : 0x00;
3522 	if (val & PCS_MII_LPA_ASYM_PAUSE)
3523 		*pause |= 0x10;
3524 	*spd = 1000;
3525 }
3526 
3527 /* Must be invoked under cp->lock. */
3528 static void cas_read_mii_link_mode(struct cas *cp, int *fd, int *spd,
3529 				   int *pause)
3530 {
3531 	u32 val;
3532 
3533 	*fd = 0;
3534 	*spd = 10;
3535 	*pause = 0;
3536 
3537 	/* use GMII registers */
3538 	val = cas_phy_read(cp, MII_LPA);
3539 	if (val & CAS_LPA_PAUSE)
3540 		*pause = 0x01;
3541 
3542 	if (val & CAS_LPA_ASYM_PAUSE)
3543 		*pause |= 0x10;
3544 
3545 	if (val & LPA_DUPLEX)
3546 		*fd = 1;
3547 	if (val & LPA_100)
3548 		*spd = 100;
3549 
3550 	if (cp->cas_flags & CAS_FLAG_1000MB_CAP) {
3551 		val = cas_phy_read(cp, CAS_MII_1000_STATUS);
3552 		if (val & (CAS_LPA_1000FULL | CAS_LPA_1000HALF))
3553 			*spd = 1000;
3554 		if (val & CAS_LPA_1000FULL)
3555 			*fd = 1;
3556 	}
3557 }
3558 
3559 /* A link-up condition has occurred, initialize and enable the
3560  * rest of the chip.
3561  *
3562  * Must be invoked under cp->lock.
3563  */
3564 static void cas_set_link_modes(struct cas *cp)
3565 {
3566 	u32 val;
3567 	int full_duplex, speed, pause;
3568 
3569 	full_duplex = 0;
3570 	speed = 10;
3571 	pause = 0;
3572 
3573 	if (CAS_PHY_MII(cp->phy_type)) {
3574 		cas_mif_poll(cp, 0);
3575 		val = cas_phy_read(cp, MII_BMCR);
3576 		if (val & BMCR_ANENABLE) {
3577 			cas_read_mii_link_mode(cp, &full_duplex, &speed,
3578 					       &pause);
3579 		} else {
3580 			if (val & BMCR_FULLDPLX)
3581 				full_duplex = 1;
3582 
3583 			if (val & BMCR_SPEED100)
3584 				speed = 100;
3585 			else if (val & CAS_BMCR_SPEED1000)
3586 				speed = (cp->cas_flags & CAS_FLAG_1000MB_CAP) ?
3587 					1000 : 100;
3588 		}
3589 		cas_mif_poll(cp, 1);
3590 
3591 	} else {
3592 		val = readl(cp->regs + REG_PCS_MII_CTRL);
3593 		cas_read_pcs_link_mode(cp, &full_duplex, &speed, &pause);
3594 		if ((val & PCS_MII_AUTONEG_EN) == 0) {
3595 			if (val & PCS_MII_CTRL_DUPLEX)
3596 				full_duplex = 1;
3597 		}
3598 	}
3599 
3600 	netif_info(cp, link, cp->dev, "Link up at %d Mbps, %s-duplex\n",
3601 		   speed, full_duplex ? "full" : "half");
3602 
3603 	val = MAC_XIF_TX_MII_OUTPUT_EN | MAC_XIF_LINK_LED;
3604 	if (CAS_PHY_MII(cp->phy_type)) {
3605 		val |= MAC_XIF_MII_BUFFER_OUTPUT_EN;
3606 		if (!full_duplex)
3607 			val |= MAC_XIF_DISABLE_ECHO;
3608 	}
3609 	if (full_duplex)
3610 		val |= MAC_XIF_FDPLX_LED;
3611 	if (speed == 1000)
3612 		val |= MAC_XIF_GMII_MODE;
3613 	writel(val, cp->regs + REG_MAC_XIF_CFG);
3614 
3615 	/* deal with carrier and collision detect. */
3616 	val = MAC_TX_CFG_IPG_EN;
3617 	if (full_duplex) {
3618 		val |= MAC_TX_CFG_IGNORE_CARRIER;
3619 		val |= MAC_TX_CFG_IGNORE_COLL;
3620 	} else {
3621 #ifndef USE_CSMA_CD_PROTO
3622 		val |= MAC_TX_CFG_NEVER_GIVE_UP_EN;
3623 		val |= MAC_TX_CFG_NEVER_GIVE_UP_LIM;
3624 #endif
3625 	}
3626 	/* val now set up for REG_MAC_TX_CFG */
3627 
3628 	/* If gigabit and half-duplex, enable carrier extension
3629 	 * mode.  increase slot time to 512 bytes as well.
3630 	 * else, disable it and make sure slot time is 64 bytes.
3631 	 * also activate checksum bug workaround
3632 	 */
3633 	if ((speed == 1000) && !full_duplex) {
3634 		writel(val | MAC_TX_CFG_CARRIER_EXTEND,
3635 		       cp->regs + REG_MAC_TX_CFG);
3636 
3637 		val = readl(cp->regs + REG_MAC_RX_CFG);
3638 		val &= ~MAC_RX_CFG_STRIP_FCS; /* checksum workaround */
3639 		writel(val | MAC_RX_CFG_CARRIER_EXTEND,
3640 		       cp->regs + REG_MAC_RX_CFG);
3641 
3642 		writel(0x200, cp->regs + REG_MAC_SLOT_TIME);
3643 
3644 		cp->crc_size = 4;
3645 		/* minimum size gigabit frame at half duplex */
3646 		cp->min_frame_size = CAS_1000MB_MIN_FRAME;
3647 
3648 	} else {
3649 		writel(val, cp->regs + REG_MAC_TX_CFG);
3650 
3651 		/* checksum bug workaround. don't strip FCS when in
3652 		 * half-duplex mode
3653 		 */
3654 		val = readl(cp->regs + REG_MAC_RX_CFG);
3655 		if (full_duplex) {
3656 			val |= MAC_RX_CFG_STRIP_FCS;
3657 			cp->crc_size = 0;
3658 			cp->min_frame_size = CAS_MIN_MTU;
3659 		} else {
3660 			val &= ~MAC_RX_CFG_STRIP_FCS;
3661 			cp->crc_size = 4;
3662 			cp->min_frame_size = CAS_MIN_FRAME;
3663 		}
3664 		writel(val & ~MAC_RX_CFG_CARRIER_EXTEND,
3665 		       cp->regs + REG_MAC_RX_CFG);
3666 		writel(0x40, cp->regs + REG_MAC_SLOT_TIME);
3667 	}
3668 
3669 	if (netif_msg_link(cp)) {
3670 		if (pause & 0x01) {
3671 			netdev_info(cp->dev, "Pause is enabled (rxfifo: %d off: %d on: %d)\n",
3672 				    cp->rx_fifo_size,
3673 				    cp->rx_pause_off,
3674 				    cp->rx_pause_on);
3675 		} else if (pause & 0x10) {
3676 			netdev_info(cp->dev, "TX pause enabled\n");
3677 		} else {
3678 			netdev_info(cp->dev, "Pause is disabled\n");
3679 		}
3680 	}
3681 
3682 	val = readl(cp->regs + REG_MAC_CTRL_CFG);
3683 	val &= ~(MAC_CTRL_CFG_SEND_PAUSE_EN | MAC_CTRL_CFG_RECV_PAUSE_EN);
3684 	if (pause) { /* symmetric or asymmetric pause */
3685 		val |= MAC_CTRL_CFG_SEND_PAUSE_EN;
3686 		if (pause & 0x01) { /* symmetric pause */
3687 			val |= MAC_CTRL_CFG_RECV_PAUSE_EN;
3688 		}
3689 	}
3690 	writel(val, cp->regs + REG_MAC_CTRL_CFG);
3691 	cas_start_dma(cp);
3692 }
3693 
3694 /* Must be invoked under cp->lock. */
3695 static void cas_init_hw(struct cas *cp, int restart_link)
3696 {
3697 	if (restart_link)
3698 		cas_phy_init(cp);
3699 
3700 	cas_init_pause_thresholds(cp);
3701 	cas_init_mac(cp);
3702 	cas_init_dma(cp);
3703 
3704 	if (restart_link) {
3705 		/* Default aneg parameters */
3706 		cp->timer_ticks = 0;
3707 		cas_begin_auto_negotiation(cp, NULL);
3708 	} else if (cp->lstate == link_up) {
3709 		cas_set_link_modes(cp);
3710 		netif_carrier_on(cp->dev);
3711 	}
3712 }
3713 
3714 /* Must be invoked under cp->lock. on earlier cassini boards,
3715  * SOFT_0 is tied to PCI reset. we use this to force a pci reset,
3716  * let it settle out, and then restore pci state.
3717  */
3718 static void cas_hard_reset(struct cas *cp)
3719 {
3720 	writel(BIM_LOCAL_DEV_SOFT_0, cp->regs + REG_BIM_LOCAL_DEV_EN);
3721 	udelay(20);
3722 	pci_restore_state(cp->pdev);
3723 }
3724 
3725 
3726 static void cas_global_reset(struct cas *cp, int blkflag)
3727 {
3728 	int limit;
3729 
3730 	/* issue a global reset. don't use RSTOUT. */
3731 	if (blkflag && !CAS_PHY_MII(cp->phy_type)) {
3732 		/* For PCS, when the blkflag is set, we should set the
3733 		 * SW_REST_BLOCK_PCS_SLINK bit to prevent the results of
3734 		 * the last autonegotiation from being cleared.  We'll
3735 		 * need some special handling if the chip is set into a
3736 		 * loopback mode.
3737 		 */
3738 		writel((SW_RESET_TX | SW_RESET_RX | SW_RESET_BLOCK_PCS_SLINK),
3739 		       cp->regs + REG_SW_RESET);
3740 	} else {
3741 		writel(SW_RESET_TX | SW_RESET_RX, cp->regs + REG_SW_RESET);
3742 	}
3743 
3744 	/* need to wait at least 3ms before polling register */
3745 	mdelay(3);
3746 
3747 	limit = STOP_TRIES;
3748 	while (limit-- > 0) {
3749 		u32 val = readl(cp->regs + REG_SW_RESET);
3750 		if ((val & (SW_RESET_TX | SW_RESET_RX)) == 0)
3751 			goto done;
3752 		udelay(10);
3753 	}
3754 	netdev_err(cp->dev, "sw reset failed\n");
3755 
3756 done:
3757 	/* enable various BIM interrupts */
3758 	writel(BIM_CFG_DPAR_INTR_ENABLE | BIM_CFG_RMA_INTR_ENABLE |
3759 	       BIM_CFG_RTA_INTR_ENABLE, cp->regs + REG_BIM_CFG);
3760 
3761 	/* clear out pci error status mask for handled errors.
3762 	 * we don't deal with DMA counter overflows as they happen
3763 	 * all the time.
3764 	 */
3765 	writel(0xFFFFFFFFU & ~(PCI_ERR_BADACK | PCI_ERR_DTRTO |
3766 			       PCI_ERR_OTHER | PCI_ERR_BIM_DMA_WRITE |
3767 			       PCI_ERR_BIM_DMA_READ), cp->regs +
3768 	       REG_PCI_ERR_STATUS_MASK);
3769 
3770 	/* set up for MII by default to address mac rx reset timeout
3771 	 * issue
3772 	 */
3773 	writel(PCS_DATAPATH_MODE_MII, cp->regs + REG_PCS_DATAPATH_MODE);
3774 }
3775 
3776 static void cas_reset(struct cas *cp, int blkflag)
3777 {
3778 	u32 val;
3779 
3780 	cas_mask_intr(cp);
3781 	cas_global_reset(cp, blkflag);
3782 	cas_mac_reset(cp);
3783 	cas_entropy_reset(cp);
3784 
3785 	/* disable dma engines. */
3786 	val = readl(cp->regs + REG_TX_CFG);
3787 	val &= ~TX_CFG_DMA_EN;
3788 	writel(val, cp->regs + REG_TX_CFG);
3789 
3790 	val = readl(cp->regs + REG_RX_CFG);
3791 	val &= ~RX_CFG_DMA_EN;
3792 	writel(val, cp->regs + REG_RX_CFG);
3793 
3794 	/* program header parser */
3795 	if ((cp->cas_flags & CAS_FLAG_TARGET_ABORT) ||
3796 	    (CAS_HP_ALT_FIRMWARE == cas_prog_null)) {
3797 		cas_load_firmware(cp, CAS_HP_FIRMWARE);
3798 	} else {
3799 		cas_load_firmware(cp, CAS_HP_ALT_FIRMWARE);
3800 	}
3801 
3802 	/* clear out error registers */
3803 	spin_lock(&cp->stat_lock[N_TX_RINGS]);
3804 	cas_clear_mac_err(cp);
3805 	spin_unlock(&cp->stat_lock[N_TX_RINGS]);
3806 }
3807 
3808 /* Shut down the chip, must be called with pm_mutex held.  */
3809 static void cas_shutdown(struct cas *cp)
3810 {
3811 	unsigned long flags;
3812 
3813 	/* Make us not-running to avoid timers respawning */
3814 	cp->hw_running = 0;
3815 
3816 	del_timer_sync(&cp->link_timer);
3817 
3818 	/* Stop the reset task */
3819 #if 0
3820 	while (atomic_read(&cp->reset_task_pending_mtu) ||
3821 	       atomic_read(&cp->reset_task_pending_spare) ||
3822 	       atomic_read(&cp->reset_task_pending_all))
3823 		schedule();
3824 
3825 #else
3826 	while (atomic_read(&cp->reset_task_pending))
3827 		schedule();
3828 #endif
3829 	/* Actually stop the chip */
3830 	cas_lock_all_save(cp, flags);
3831 	cas_reset(cp, 0);
3832 	if (cp->cas_flags & CAS_FLAG_SATURN)
3833 		cas_phy_powerdown(cp);
3834 	cas_unlock_all_restore(cp, flags);
3835 }
3836 
3837 static int cas_change_mtu(struct net_device *dev, int new_mtu)
3838 {
3839 	struct cas *cp = netdev_priv(dev);
3840 
3841 	dev->mtu = new_mtu;
3842 	if (!netif_running(dev) || !netif_device_present(dev))
3843 		return 0;
3844 
3845 	/* let the reset task handle it */
3846 #if 1
3847 	atomic_inc(&cp->reset_task_pending);
3848 	if ((cp->phy_type & CAS_PHY_SERDES)) {
3849 		atomic_inc(&cp->reset_task_pending_all);
3850 	} else {
3851 		atomic_inc(&cp->reset_task_pending_mtu);
3852 	}
3853 	schedule_work(&cp->reset_task);
3854 #else
3855 	atomic_set(&cp->reset_task_pending, (cp->phy_type & CAS_PHY_SERDES) ?
3856 		   CAS_RESET_ALL : CAS_RESET_MTU);
3857 	pr_err("reset called in cas_change_mtu\n");
3858 	schedule_work(&cp->reset_task);
3859 #endif
3860 
3861 	flush_work(&cp->reset_task);
3862 	return 0;
3863 }
3864 
3865 static void cas_clean_txd(struct cas *cp, int ring)
3866 {
3867 	struct cas_tx_desc *txd = cp->init_txds[ring];
3868 	struct sk_buff *skb, **skbs = cp->tx_skbs[ring];
3869 	u64 daddr, dlen;
3870 	int i, size;
3871 
3872 	size = TX_DESC_RINGN_SIZE(ring);
3873 	for (i = 0; i < size; i++) {
3874 		int frag;
3875 
3876 		if (skbs[i] == NULL)
3877 			continue;
3878 
3879 		skb = skbs[i];
3880 		skbs[i] = NULL;
3881 
3882 		for (frag = 0; frag <= skb_shinfo(skb)->nr_frags;  frag++) {
3883 			int ent = i & (size - 1);
3884 
3885 			/* first buffer is never a tiny buffer and so
3886 			 * needs to be unmapped.
3887 			 */
3888 			daddr = le64_to_cpu(txd[ent].buffer);
3889 			dlen  =  CAS_VAL(TX_DESC_BUFLEN,
3890 					 le64_to_cpu(txd[ent].control));
3891 			dma_unmap_page(&cp->pdev->dev, daddr, dlen,
3892 				       DMA_TO_DEVICE);
3893 
3894 			if (frag != skb_shinfo(skb)->nr_frags) {
3895 				i++;
3896 
3897 				/* next buffer might by a tiny buffer.
3898 				 * skip past it.
3899 				 */
3900 				ent = i & (size - 1);
3901 				if (cp->tx_tiny_use[ring][ent].used)
3902 					i++;
3903 			}
3904 		}
3905 		dev_kfree_skb_any(skb);
3906 	}
3907 
3908 	/* zero out tiny buf usage */
3909 	memset(cp->tx_tiny_use[ring], 0, size*sizeof(*cp->tx_tiny_use[ring]));
3910 }
3911 
3912 /* freed on close */
3913 static inline void cas_free_rx_desc(struct cas *cp, int ring)
3914 {
3915 	cas_page_t **page = cp->rx_pages[ring];
3916 	int i, size;
3917 
3918 	size = RX_DESC_RINGN_SIZE(ring);
3919 	for (i = 0; i < size; i++) {
3920 		if (page[i]) {
3921 			cas_page_free(cp, page[i]);
3922 			page[i] = NULL;
3923 		}
3924 	}
3925 }
3926 
3927 static void cas_free_rxds(struct cas *cp)
3928 {
3929 	int i;
3930 
3931 	for (i = 0; i < N_RX_DESC_RINGS; i++)
3932 		cas_free_rx_desc(cp, i);
3933 }
3934 
3935 /* Must be invoked under cp->lock. */
3936 static void cas_clean_rings(struct cas *cp)
3937 {
3938 	int i;
3939 
3940 	/* need to clean all tx rings */
3941 	memset(cp->tx_old, 0, sizeof(*cp->tx_old)*N_TX_RINGS);
3942 	memset(cp->tx_new, 0, sizeof(*cp->tx_new)*N_TX_RINGS);
3943 	for (i = 0; i < N_TX_RINGS; i++)
3944 		cas_clean_txd(cp, i);
3945 
3946 	/* zero out init block */
3947 	memset(cp->init_block, 0, sizeof(struct cas_init_block));
3948 	cas_clean_rxds(cp);
3949 	cas_clean_rxcs(cp);
3950 }
3951 
3952 /* allocated on open */
3953 static inline int cas_alloc_rx_desc(struct cas *cp, int ring)
3954 {
3955 	cas_page_t **page = cp->rx_pages[ring];
3956 	int size, i = 0;
3957 
3958 	size = RX_DESC_RINGN_SIZE(ring);
3959 	for (i = 0; i < size; i++) {
3960 		if ((page[i] = cas_page_alloc(cp, GFP_KERNEL)) == NULL)
3961 			return -1;
3962 	}
3963 	return 0;
3964 }
3965 
3966 static int cas_alloc_rxds(struct cas *cp)
3967 {
3968 	int i;
3969 
3970 	for (i = 0; i < N_RX_DESC_RINGS; i++) {
3971 		if (cas_alloc_rx_desc(cp, i) < 0) {
3972 			cas_free_rxds(cp);
3973 			return -1;
3974 		}
3975 	}
3976 	return 0;
3977 }
3978 
3979 static void cas_reset_task(struct work_struct *work)
3980 {
3981 	struct cas *cp = container_of(work, struct cas, reset_task);
3982 #if 0
3983 	int pending = atomic_read(&cp->reset_task_pending);
3984 #else
3985 	int pending_all = atomic_read(&cp->reset_task_pending_all);
3986 	int pending_spare = atomic_read(&cp->reset_task_pending_spare);
3987 	int pending_mtu = atomic_read(&cp->reset_task_pending_mtu);
3988 
3989 	if (pending_all == 0 && pending_spare == 0 && pending_mtu == 0) {
3990 		/* We can have more tasks scheduled than actually
3991 		 * needed.
3992 		 */
3993 		atomic_dec(&cp->reset_task_pending);
3994 		return;
3995 	}
3996 #endif
3997 	/* The link went down, we reset the ring, but keep
3998 	 * DMA stopped. Use this function for reset
3999 	 * on error as well.
4000 	 */
4001 	if (cp->hw_running) {
4002 		unsigned long flags;
4003 
4004 		/* Make sure we don't get interrupts or tx packets */
4005 		netif_device_detach(cp->dev);
4006 		cas_lock_all_save(cp, flags);
4007 
4008 		if (cp->opened) {
4009 			/* We call cas_spare_recover when we call cas_open.
4010 			 * but we do not initialize the lists cas_spare_recover
4011 			 * uses until cas_open is called.
4012 			 */
4013 			cas_spare_recover(cp, GFP_ATOMIC);
4014 		}
4015 #if 1
4016 		/* test => only pending_spare set */
4017 		if (!pending_all && !pending_mtu)
4018 			goto done;
4019 #else
4020 		if (pending == CAS_RESET_SPARE)
4021 			goto done;
4022 #endif
4023 		/* when pending == CAS_RESET_ALL, the following
4024 		 * call to cas_init_hw will restart auto negotiation.
4025 		 * Setting the second argument of cas_reset to
4026 		 * !(pending == CAS_RESET_ALL) will set this argument
4027 		 * to 1 (avoiding reinitializing the PHY for the normal
4028 		 * PCS case) when auto negotiation is not restarted.
4029 		 */
4030 #if 1
4031 		cas_reset(cp, !(pending_all > 0));
4032 		if (cp->opened)
4033 			cas_clean_rings(cp);
4034 		cas_init_hw(cp, (pending_all > 0));
4035 #else
4036 		cas_reset(cp, !(pending == CAS_RESET_ALL));
4037 		if (cp->opened)
4038 			cas_clean_rings(cp);
4039 		cas_init_hw(cp, pending == CAS_RESET_ALL);
4040 #endif
4041 
4042 done:
4043 		cas_unlock_all_restore(cp, flags);
4044 		netif_device_attach(cp->dev);
4045 	}
4046 #if 1
4047 	atomic_sub(pending_all, &cp->reset_task_pending_all);
4048 	atomic_sub(pending_spare, &cp->reset_task_pending_spare);
4049 	atomic_sub(pending_mtu, &cp->reset_task_pending_mtu);
4050 	atomic_dec(&cp->reset_task_pending);
4051 #else
4052 	atomic_set(&cp->reset_task_pending, 0);
4053 #endif
4054 }
4055 
4056 static void cas_link_timer(struct timer_list *t)
4057 {
4058 	struct cas *cp = from_timer(cp, t, link_timer);
4059 	int mask, pending = 0, reset = 0;
4060 	unsigned long flags;
4061 
4062 	if (link_transition_timeout != 0 &&
4063 	    cp->link_transition_jiffies_valid &&
4064 	    ((jiffies - cp->link_transition_jiffies) >
4065 	      (link_transition_timeout))) {
4066 		/* One-second counter so link-down workaround doesn't
4067 		 * cause resets to occur so fast as to fool the switch
4068 		 * into thinking the link is down.
4069 		 */
4070 		cp->link_transition_jiffies_valid = 0;
4071 	}
4072 
4073 	if (!cp->hw_running)
4074 		return;
4075 
4076 	spin_lock_irqsave(&cp->lock, flags);
4077 	cas_lock_tx(cp);
4078 	cas_entropy_gather(cp);
4079 
4080 	/* If the link task is still pending, we just
4081 	 * reschedule the link timer
4082 	 */
4083 #if 1
4084 	if (atomic_read(&cp->reset_task_pending_all) ||
4085 	    atomic_read(&cp->reset_task_pending_spare) ||
4086 	    atomic_read(&cp->reset_task_pending_mtu))
4087 		goto done;
4088 #else
4089 	if (atomic_read(&cp->reset_task_pending))
4090 		goto done;
4091 #endif
4092 
4093 	/* check for rx cleaning */
4094 	if ((mask = (cp->cas_flags & CAS_FLAG_RXD_POST_MASK))) {
4095 		int i, rmask;
4096 
4097 		for (i = 0; i < MAX_RX_DESC_RINGS; i++) {
4098 			rmask = CAS_FLAG_RXD_POST(i);
4099 			if ((mask & rmask) == 0)
4100 				continue;
4101 
4102 			/* post_rxds will do a mod_timer */
4103 			if (cas_post_rxds_ringN(cp, i, cp->rx_last[i]) < 0) {
4104 				pending = 1;
4105 				continue;
4106 			}
4107 			cp->cas_flags &= ~rmask;
4108 		}
4109 	}
4110 
4111 	if (CAS_PHY_MII(cp->phy_type)) {
4112 		u16 bmsr;
4113 		cas_mif_poll(cp, 0);
4114 		bmsr = cas_phy_read(cp, MII_BMSR);
4115 		/* WTZ: Solaris driver reads this twice, but that
4116 		 * may be due to the PCS case and the use of a
4117 		 * common implementation. Read it twice here to be
4118 		 * safe.
4119 		 */
4120 		bmsr = cas_phy_read(cp, MII_BMSR);
4121 		cas_mif_poll(cp, 1);
4122 		readl(cp->regs + REG_MIF_STATUS); /* avoid dups */
4123 		reset = cas_mii_link_check(cp, bmsr);
4124 	} else {
4125 		reset = cas_pcs_link_check(cp);
4126 	}
4127 
4128 	if (reset)
4129 		goto done;
4130 
4131 	/* check for tx state machine confusion */
4132 	if ((readl(cp->regs + REG_MAC_TX_STATUS) & MAC_TX_FRAME_XMIT) == 0) {
4133 		u32 val = readl(cp->regs + REG_MAC_STATE_MACHINE);
4134 		u32 wptr, rptr;
4135 		int tlm  = CAS_VAL(MAC_SM_TLM, val);
4136 
4137 		if (((tlm == 0x5) || (tlm == 0x3)) &&
4138 		    (CAS_VAL(MAC_SM_ENCAP_SM, val) == 0)) {
4139 			netif_printk(cp, tx_err, KERN_DEBUG, cp->dev,
4140 				     "tx err: MAC_STATE[%08x]\n", val);
4141 			reset = 1;
4142 			goto done;
4143 		}
4144 
4145 		val  = readl(cp->regs + REG_TX_FIFO_PKT_CNT);
4146 		wptr = readl(cp->regs + REG_TX_FIFO_WRITE_PTR);
4147 		rptr = readl(cp->regs + REG_TX_FIFO_READ_PTR);
4148 		if ((val == 0) && (wptr != rptr)) {
4149 			netif_printk(cp, tx_err, KERN_DEBUG, cp->dev,
4150 				     "tx err: TX_FIFO[%08x:%08x:%08x]\n",
4151 				     val, wptr, rptr);
4152 			reset = 1;
4153 		}
4154 
4155 		if (reset)
4156 			cas_hard_reset(cp);
4157 	}
4158 
4159 done:
4160 	if (reset) {
4161 #if 1
4162 		atomic_inc(&cp->reset_task_pending);
4163 		atomic_inc(&cp->reset_task_pending_all);
4164 		schedule_work(&cp->reset_task);
4165 #else
4166 		atomic_set(&cp->reset_task_pending, CAS_RESET_ALL);
4167 		pr_err("reset called in cas_link_timer\n");
4168 		schedule_work(&cp->reset_task);
4169 #endif
4170 	}
4171 
4172 	if (!pending)
4173 		mod_timer(&cp->link_timer, jiffies + CAS_LINK_TIMEOUT);
4174 	cas_unlock_tx(cp);
4175 	spin_unlock_irqrestore(&cp->lock, flags);
4176 }
4177 
4178 /* tiny buffers are used to avoid target abort issues with
4179  * older cassini's
4180  */
4181 static void cas_tx_tiny_free(struct cas *cp)
4182 {
4183 	struct pci_dev *pdev = cp->pdev;
4184 	int i;
4185 
4186 	for (i = 0; i < N_TX_RINGS; i++) {
4187 		if (!cp->tx_tiny_bufs[i])
4188 			continue;
4189 
4190 		dma_free_coherent(&pdev->dev, TX_TINY_BUF_BLOCK,
4191 				  cp->tx_tiny_bufs[i], cp->tx_tiny_dvma[i]);
4192 		cp->tx_tiny_bufs[i] = NULL;
4193 	}
4194 }
4195 
4196 static int cas_tx_tiny_alloc(struct cas *cp)
4197 {
4198 	struct pci_dev *pdev = cp->pdev;
4199 	int i;
4200 
4201 	for (i = 0; i < N_TX_RINGS; i++) {
4202 		cp->tx_tiny_bufs[i] =
4203 			dma_alloc_coherent(&pdev->dev, TX_TINY_BUF_BLOCK,
4204 					   &cp->tx_tiny_dvma[i], GFP_KERNEL);
4205 		if (!cp->tx_tiny_bufs[i]) {
4206 			cas_tx_tiny_free(cp);
4207 			return -1;
4208 		}
4209 	}
4210 	return 0;
4211 }
4212 
4213 
4214 static int cas_open(struct net_device *dev)
4215 {
4216 	struct cas *cp = netdev_priv(dev);
4217 	int hw_was_up, err;
4218 	unsigned long flags;
4219 
4220 	mutex_lock(&cp->pm_mutex);
4221 
4222 	hw_was_up = cp->hw_running;
4223 
4224 	/* The power-management mutex protects the hw_running
4225 	 * etc. state so it is safe to do this bit without cp->lock
4226 	 */
4227 	if (!cp->hw_running) {
4228 		/* Reset the chip */
4229 		cas_lock_all_save(cp, flags);
4230 		/* We set the second arg to cas_reset to zero
4231 		 * because cas_init_hw below will have its second
4232 		 * argument set to non-zero, which will force
4233 		 * autonegotiation to start.
4234 		 */
4235 		cas_reset(cp, 0);
4236 		cp->hw_running = 1;
4237 		cas_unlock_all_restore(cp, flags);
4238 	}
4239 
4240 	err = -ENOMEM;
4241 	if (cas_tx_tiny_alloc(cp) < 0)
4242 		goto err_unlock;
4243 
4244 	/* alloc rx descriptors */
4245 	if (cas_alloc_rxds(cp) < 0)
4246 		goto err_tx_tiny;
4247 
4248 	/* allocate spares */
4249 	cas_spare_init(cp);
4250 	cas_spare_recover(cp, GFP_KERNEL);
4251 
4252 	/* We can now request the interrupt as we know it's masked
4253 	 * on the controller. cassini+ has up to 4 interrupts
4254 	 * that can be used, but you need to do explicit pci interrupt
4255 	 * mapping to expose them
4256 	 */
4257 	if (request_irq(cp->pdev->irq, cas_interrupt,
4258 			IRQF_SHARED, dev->name, (void *) dev)) {
4259 		netdev_err(cp->dev, "failed to request irq !\n");
4260 		err = -EAGAIN;
4261 		goto err_spare;
4262 	}
4263 
4264 #ifdef USE_NAPI
4265 	napi_enable(&cp->napi);
4266 #endif
4267 	/* init hw */
4268 	cas_lock_all_save(cp, flags);
4269 	cas_clean_rings(cp);
4270 	cas_init_hw(cp, !hw_was_up);
4271 	cp->opened = 1;
4272 	cas_unlock_all_restore(cp, flags);
4273 
4274 	netif_start_queue(dev);
4275 	mutex_unlock(&cp->pm_mutex);
4276 	return 0;
4277 
4278 err_spare:
4279 	cas_spare_free(cp);
4280 	cas_free_rxds(cp);
4281 err_tx_tiny:
4282 	cas_tx_tiny_free(cp);
4283 err_unlock:
4284 	mutex_unlock(&cp->pm_mutex);
4285 	return err;
4286 }
4287 
4288 static int cas_close(struct net_device *dev)
4289 {
4290 	unsigned long flags;
4291 	struct cas *cp = netdev_priv(dev);
4292 
4293 #ifdef USE_NAPI
4294 	napi_disable(&cp->napi);
4295 #endif
4296 	/* Make sure we don't get distracted by suspend/resume */
4297 	mutex_lock(&cp->pm_mutex);
4298 
4299 	netif_stop_queue(dev);
4300 
4301 	/* Stop traffic, mark us closed */
4302 	cas_lock_all_save(cp, flags);
4303 	cp->opened = 0;
4304 	cas_reset(cp, 0);
4305 	cas_phy_init(cp);
4306 	cas_begin_auto_negotiation(cp, NULL);
4307 	cas_clean_rings(cp);
4308 	cas_unlock_all_restore(cp, flags);
4309 
4310 	free_irq(cp->pdev->irq, (void *) dev);
4311 	cas_spare_free(cp);
4312 	cas_free_rxds(cp);
4313 	cas_tx_tiny_free(cp);
4314 	mutex_unlock(&cp->pm_mutex);
4315 	return 0;
4316 }
4317 
4318 static struct {
4319 	const char name[ETH_GSTRING_LEN];
4320 } ethtool_cassini_statnames[] = {
4321 	{"collisions"},
4322 	{"rx_bytes"},
4323 	{"rx_crc_errors"},
4324 	{"rx_dropped"},
4325 	{"rx_errors"},
4326 	{"rx_fifo_errors"},
4327 	{"rx_frame_errors"},
4328 	{"rx_length_errors"},
4329 	{"rx_over_errors"},
4330 	{"rx_packets"},
4331 	{"tx_aborted_errors"},
4332 	{"tx_bytes"},
4333 	{"tx_dropped"},
4334 	{"tx_errors"},
4335 	{"tx_fifo_errors"},
4336 	{"tx_packets"}
4337 };
4338 #define CAS_NUM_STAT_KEYS ARRAY_SIZE(ethtool_cassini_statnames)
4339 
4340 static struct {
4341 	const int offsets;	/* neg. values for 2nd arg to cas_read_phy */
4342 } ethtool_register_table[] = {
4343 	{-MII_BMSR},
4344 	{-MII_BMCR},
4345 	{REG_CAWR},
4346 	{REG_INF_BURST},
4347 	{REG_BIM_CFG},
4348 	{REG_RX_CFG},
4349 	{REG_HP_CFG},
4350 	{REG_MAC_TX_CFG},
4351 	{REG_MAC_RX_CFG},
4352 	{REG_MAC_CTRL_CFG},
4353 	{REG_MAC_XIF_CFG},
4354 	{REG_MIF_CFG},
4355 	{REG_PCS_CFG},
4356 	{REG_SATURN_PCFG},
4357 	{REG_PCS_MII_STATUS},
4358 	{REG_PCS_STATE_MACHINE},
4359 	{REG_MAC_COLL_EXCESS},
4360 	{REG_MAC_COLL_LATE}
4361 };
4362 #define CAS_REG_LEN 	ARRAY_SIZE(ethtool_register_table)
4363 #define CAS_MAX_REGS 	(sizeof (u32)*CAS_REG_LEN)
4364 
4365 static void cas_read_regs(struct cas *cp, u8 *ptr, int len)
4366 {
4367 	u8 *p;
4368 	int i;
4369 	unsigned long flags;
4370 
4371 	spin_lock_irqsave(&cp->lock, flags);
4372 	for (i = 0, p = ptr; i < len ; i ++, p += sizeof(u32)) {
4373 		u16 hval;
4374 		u32 val;
4375 		if (ethtool_register_table[i].offsets < 0) {
4376 			hval = cas_phy_read(cp,
4377 				    -ethtool_register_table[i].offsets);
4378 			val = hval;
4379 		} else {
4380 			val= readl(cp->regs+ethtool_register_table[i].offsets);
4381 		}
4382 		memcpy(p, (u8 *)&val, sizeof(u32));
4383 	}
4384 	spin_unlock_irqrestore(&cp->lock, flags);
4385 }
4386 
4387 static struct net_device_stats *cas_get_stats(struct net_device *dev)
4388 {
4389 	struct cas *cp = netdev_priv(dev);
4390 	struct net_device_stats *stats = cp->net_stats;
4391 	unsigned long flags;
4392 	int i;
4393 	unsigned long tmp;
4394 
4395 	/* we collate all of the stats into net_stats[N_TX_RING] */
4396 	if (!cp->hw_running)
4397 		return stats + N_TX_RINGS;
4398 
4399 	/* collect outstanding stats */
4400 	/* WTZ: the Cassini spec gives these as 16 bit counters but
4401 	 * stored in 32-bit words.  Added a mask of 0xffff to be safe,
4402 	 * in case the chip somehow puts any garbage in the other bits.
4403 	 * Also, counter usage didn't seem to mach what Adrian did
4404 	 * in the parts of the code that set these quantities. Made
4405 	 * that consistent.
4406 	 */
4407 	spin_lock_irqsave(&cp->stat_lock[N_TX_RINGS], flags);
4408 	stats[N_TX_RINGS].rx_crc_errors +=
4409 	  readl(cp->regs + REG_MAC_FCS_ERR) & 0xffff;
4410 	stats[N_TX_RINGS].rx_frame_errors +=
4411 		readl(cp->regs + REG_MAC_ALIGN_ERR) &0xffff;
4412 	stats[N_TX_RINGS].rx_length_errors +=
4413 		readl(cp->regs + REG_MAC_LEN_ERR) & 0xffff;
4414 #if 1
4415 	tmp = (readl(cp->regs + REG_MAC_COLL_EXCESS) & 0xffff) +
4416 		(readl(cp->regs + REG_MAC_COLL_LATE) & 0xffff);
4417 	stats[N_TX_RINGS].tx_aborted_errors += tmp;
4418 	stats[N_TX_RINGS].collisions +=
4419 	  tmp + (readl(cp->regs + REG_MAC_COLL_NORMAL) & 0xffff);
4420 #else
4421 	stats[N_TX_RINGS].tx_aborted_errors +=
4422 		readl(cp->regs + REG_MAC_COLL_EXCESS);
4423 	stats[N_TX_RINGS].collisions += readl(cp->regs + REG_MAC_COLL_EXCESS) +
4424 		readl(cp->regs + REG_MAC_COLL_LATE);
4425 #endif
4426 	cas_clear_mac_err(cp);
4427 
4428 	/* saved bits that are unique to ring 0 */
4429 	spin_lock(&cp->stat_lock[0]);
4430 	stats[N_TX_RINGS].collisions        += stats[0].collisions;
4431 	stats[N_TX_RINGS].rx_over_errors    += stats[0].rx_over_errors;
4432 	stats[N_TX_RINGS].rx_frame_errors   += stats[0].rx_frame_errors;
4433 	stats[N_TX_RINGS].rx_fifo_errors    += stats[0].rx_fifo_errors;
4434 	stats[N_TX_RINGS].tx_aborted_errors += stats[0].tx_aborted_errors;
4435 	stats[N_TX_RINGS].tx_fifo_errors    += stats[0].tx_fifo_errors;
4436 	spin_unlock(&cp->stat_lock[0]);
4437 
4438 	for (i = 0; i < N_TX_RINGS; i++) {
4439 		spin_lock(&cp->stat_lock[i]);
4440 		stats[N_TX_RINGS].rx_length_errors +=
4441 			stats[i].rx_length_errors;
4442 		stats[N_TX_RINGS].rx_crc_errors += stats[i].rx_crc_errors;
4443 		stats[N_TX_RINGS].rx_packets    += stats[i].rx_packets;
4444 		stats[N_TX_RINGS].tx_packets    += stats[i].tx_packets;
4445 		stats[N_TX_RINGS].rx_bytes      += stats[i].rx_bytes;
4446 		stats[N_TX_RINGS].tx_bytes      += stats[i].tx_bytes;
4447 		stats[N_TX_RINGS].rx_errors     += stats[i].rx_errors;
4448 		stats[N_TX_RINGS].tx_errors     += stats[i].tx_errors;
4449 		stats[N_TX_RINGS].rx_dropped    += stats[i].rx_dropped;
4450 		stats[N_TX_RINGS].tx_dropped    += stats[i].tx_dropped;
4451 		memset(stats + i, 0, sizeof(struct net_device_stats));
4452 		spin_unlock(&cp->stat_lock[i]);
4453 	}
4454 	spin_unlock_irqrestore(&cp->stat_lock[N_TX_RINGS], flags);
4455 	return stats + N_TX_RINGS;
4456 }
4457 
4458 
4459 static void cas_set_multicast(struct net_device *dev)
4460 {
4461 	struct cas *cp = netdev_priv(dev);
4462 	u32 rxcfg, rxcfg_new;
4463 	unsigned long flags;
4464 	int limit = STOP_TRIES;
4465 
4466 	if (!cp->hw_running)
4467 		return;
4468 
4469 	spin_lock_irqsave(&cp->lock, flags);
4470 	rxcfg = readl(cp->regs + REG_MAC_RX_CFG);
4471 
4472 	/* disable RX MAC and wait for completion */
4473 	writel(rxcfg & ~MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
4474 	while (readl(cp->regs + REG_MAC_RX_CFG) & MAC_RX_CFG_EN) {
4475 		if (!limit--)
4476 			break;
4477 		udelay(10);
4478 	}
4479 
4480 	/* disable hash filter and wait for completion */
4481 	limit = STOP_TRIES;
4482 	rxcfg &= ~(MAC_RX_CFG_PROMISC_EN | MAC_RX_CFG_HASH_FILTER_EN);
4483 	writel(rxcfg & ~MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
4484 	while (readl(cp->regs + REG_MAC_RX_CFG) & MAC_RX_CFG_HASH_FILTER_EN) {
4485 		if (!limit--)
4486 			break;
4487 		udelay(10);
4488 	}
4489 
4490 	/* program hash filters */
4491 	cp->mac_rx_cfg = rxcfg_new = cas_setup_multicast(cp);
4492 	rxcfg |= rxcfg_new;
4493 	writel(rxcfg, cp->regs + REG_MAC_RX_CFG);
4494 	spin_unlock_irqrestore(&cp->lock, flags);
4495 }
4496 
4497 static void cas_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
4498 {
4499 	struct cas *cp = netdev_priv(dev);
4500 	strlcpy(info->driver, DRV_MODULE_NAME, sizeof(info->driver));
4501 	strlcpy(info->version, DRV_MODULE_VERSION, sizeof(info->version));
4502 	strlcpy(info->bus_info, pci_name(cp->pdev), sizeof(info->bus_info));
4503 }
4504 
4505 static int cas_get_link_ksettings(struct net_device *dev,
4506 				  struct ethtool_link_ksettings *cmd)
4507 {
4508 	struct cas *cp = netdev_priv(dev);
4509 	u16 bmcr;
4510 	int full_duplex, speed, pause;
4511 	unsigned long flags;
4512 	enum link_state linkstate = link_up;
4513 	u32 supported, advertising;
4514 
4515 	advertising = 0;
4516 	supported = SUPPORTED_Autoneg;
4517 	if (cp->cas_flags & CAS_FLAG_1000MB_CAP) {
4518 		supported |= SUPPORTED_1000baseT_Full;
4519 		advertising |= ADVERTISED_1000baseT_Full;
4520 	}
4521 
4522 	/* Record PHY settings if HW is on. */
4523 	spin_lock_irqsave(&cp->lock, flags);
4524 	bmcr = 0;
4525 	linkstate = cp->lstate;
4526 	if (CAS_PHY_MII(cp->phy_type)) {
4527 		cmd->base.port = PORT_MII;
4528 		cmd->base.phy_address = cp->phy_addr;
4529 		advertising |= ADVERTISED_TP | ADVERTISED_MII |
4530 			ADVERTISED_10baseT_Half |
4531 			ADVERTISED_10baseT_Full |
4532 			ADVERTISED_100baseT_Half |
4533 			ADVERTISED_100baseT_Full;
4534 
4535 		supported |=
4536 			(SUPPORTED_10baseT_Half |
4537 			 SUPPORTED_10baseT_Full |
4538 			 SUPPORTED_100baseT_Half |
4539 			 SUPPORTED_100baseT_Full |
4540 			 SUPPORTED_TP | SUPPORTED_MII);
4541 
4542 		if (cp->hw_running) {
4543 			cas_mif_poll(cp, 0);
4544 			bmcr = cas_phy_read(cp, MII_BMCR);
4545 			cas_read_mii_link_mode(cp, &full_duplex,
4546 					       &speed, &pause);
4547 			cas_mif_poll(cp, 1);
4548 		}
4549 
4550 	} else {
4551 		cmd->base.port = PORT_FIBRE;
4552 		cmd->base.phy_address = 0;
4553 		supported   |= SUPPORTED_FIBRE;
4554 		advertising |= ADVERTISED_FIBRE;
4555 
4556 		if (cp->hw_running) {
4557 			/* pcs uses the same bits as mii */
4558 			bmcr = readl(cp->regs + REG_PCS_MII_CTRL);
4559 			cas_read_pcs_link_mode(cp, &full_duplex,
4560 					       &speed, &pause);
4561 		}
4562 	}
4563 	spin_unlock_irqrestore(&cp->lock, flags);
4564 
4565 	if (bmcr & BMCR_ANENABLE) {
4566 		advertising |= ADVERTISED_Autoneg;
4567 		cmd->base.autoneg = AUTONEG_ENABLE;
4568 		cmd->base.speed =  ((speed == 10) ?
4569 					    SPEED_10 :
4570 					    ((speed == 1000) ?
4571 					     SPEED_1000 : SPEED_100));
4572 		cmd->base.duplex = full_duplex ? DUPLEX_FULL : DUPLEX_HALF;
4573 	} else {
4574 		cmd->base.autoneg = AUTONEG_DISABLE;
4575 		cmd->base.speed = ((bmcr & CAS_BMCR_SPEED1000) ?
4576 					    SPEED_1000 :
4577 					    ((bmcr & BMCR_SPEED100) ?
4578 					     SPEED_100 : SPEED_10));
4579 		cmd->base.duplex = (bmcr & BMCR_FULLDPLX) ?
4580 			DUPLEX_FULL : DUPLEX_HALF;
4581 	}
4582 	if (linkstate != link_up) {
4583 		/* Force these to "unknown" if the link is not up and
4584 		 * autonogotiation in enabled. We can set the link
4585 		 * speed to 0, but not cmd->duplex,
4586 		 * because its legal values are 0 and 1.  Ethtool will
4587 		 * print the value reported in parentheses after the
4588 		 * word "Unknown" for unrecognized values.
4589 		 *
4590 		 * If in forced mode, we report the speed and duplex
4591 		 * settings that we configured.
4592 		 */
4593 		if (cp->link_cntl & BMCR_ANENABLE) {
4594 			cmd->base.speed = 0;
4595 			cmd->base.duplex = 0xff;
4596 		} else {
4597 			cmd->base.speed = SPEED_10;
4598 			if (cp->link_cntl & BMCR_SPEED100) {
4599 				cmd->base.speed = SPEED_100;
4600 			} else if (cp->link_cntl & CAS_BMCR_SPEED1000) {
4601 				cmd->base.speed = SPEED_1000;
4602 			}
4603 			cmd->base.duplex = (cp->link_cntl & BMCR_FULLDPLX) ?
4604 				DUPLEX_FULL : DUPLEX_HALF;
4605 		}
4606 	}
4607 
4608 	ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.supported,
4609 						supported);
4610 	ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.advertising,
4611 						advertising);
4612 
4613 	return 0;
4614 }
4615 
4616 static int cas_set_link_ksettings(struct net_device *dev,
4617 				  const struct ethtool_link_ksettings *cmd)
4618 {
4619 	struct cas *cp = netdev_priv(dev);
4620 	unsigned long flags;
4621 	u32 speed = cmd->base.speed;
4622 
4623 	/* Verify the settings we care about. */
4624 	if (cmd->base.autoneg != AUTONEG_ENABLE &&
4625 	    cmd->base.autoneg != AUTONEG_DISABLE)
4626 		return -EINVAL;
4627 
4628 	if (cmd->base.autoneg == AUTONEG_DISABLE &&
4629 	    ((speed != SPEED_1000 &&
4630 	      speed != SPEED_100 &&
4631 	      speed != SPEED_10) ||
4632 	     (cmd->base.duplex != DUPLEX_HALF &&
4633 	      cmd->base.duplex != DUPLEX_FULL)))
4634 		return -EINVAL;
4635 
4636 	/* Apply settings and restart link process. */
4637 	spin_lock_irqsave(&cp->lock, flags);
4638 	cas_begin_auto_negotiation(cp, cmd);
4639 	spin_unlock_irqrestore(&cp->lock, flags);
4640 	return 0;
4641 }
4642 
4643 static int cas_nway_reset(struct net_device *dev)
4644 {
4645 	struct cas *cp = netdev_priv(dev);
4646 	unsigned long flags;
4647 
4648 	if ((cp->link_cntl & BMCR_ANENABLE) == 0)
4649 		return -EINVAL;
4650 
4651 	/* Restart link process. */
4652 	spin_lock_irqsave(&cp->lock, flags);
4653 	cas_begin_auto_negotiation(cp, NULL);
4654 	spin_unlock_irqrestore(&cp->lock, flags);
4655 
4656 	return 0;
4657 }
4658 
4659 static u32 cas_get_link(struct net_device *dev)
4660 {
4661 	struct cas *cp = netdev_priv(dev);
4662 	return cp->lstate == link_up;
4663 }
4664 
4665 static u32 cas_get_msglevel(struct net_device *dev)
4666 {
4667 	struct cas *cp = netdev_priv(dev);
4668 	return cp->msg_enable;
4669 }
4670 
4671 static void cas_set_msglevel(struct net_device *dev, u32 value)
4672 {
4673 	struct cas *cp = netdev_priv(dev);
4674 	cp->msg_enable = value;
4675 }
4676 
4677 static int cas_get_regs_len(struct net_device *dev)
4678 {
4679 	struct cas *cp = netdev_priv(dev);
4680 	return cp->casreg_len < CAS_MAX_REGS ? cp->casreg_len: CAS_MAX_REGS;
4681 }
4682 
4683 static void cas_get_regs(struct net_device *dev, struct ethtool_regs *regs,
4684 			     void *p)
4685 {
4686 	struct cas *cp = netdev_priv(dev);
4687 	regs->version = 0;
4688 	/* cas_read_regs handles locks (cp->lock).  */
4689 	cas_read_regs(cp, p, regs->len / sizeof(u32));
4690 }
4691 
4692 static int cas_get_sset_count(struct net_device *dev, int sset)
4693 {
4694 	switch (sset) {
4695 	case ETH_SS_STATS:
4696 		return CAS_NUM_STAT_KEYS;
4697 	default:
4698 		return -EOPNOTSUPP;
4699 	}
4700 }
4701 
4702 static void cas_get_strings(struct net_device *dev, u32 stringset, u8 *data)
4703 {
4704 	 memcpy(data, &ethtool_cassini_statnames,
4705 					 CAS_NUM_STAT_KEYS * ETH_GSTRING_LEN);
4706 }
4707 
4708 static void cas_get_ethtool_stats(struct net_device *dev,
4709 				      struct ethtool_stats *estats, u64 *data)
4710 {
4711 	struct cas *cp = netdev_priv(dev);
4712 	struct net_device_stats *stats = cas_get_stats(cp->dev);
4713 	int i = 0;
4714 	data[i++] = stats->collisions;
4715 	data[i++] = stats->rx_bytes;
4716 	data[i++] = stats->rx_crc_errors;
4717 	data[i++] = stats->rx_dropped;
4718 	data[i++] = stats->rx_errors;
4719 	data[i++] = stats->rx_fifo_errors;
4720 	data[i++] = stats->rx_frame_errors;
4721 	data[i++] = stats->rx_length_errors;
4722 	data[i++] = stats->rx_over_errors;
4723 	data[i++] = stats->rx_packets;
4724 	data[i++] = stats->tx_aborted_errors;
4725 	data[i++] = stats->tx_bytes;
4726 	data[i++] = stats->tx_dropped;
4727 	data[i++] = stats->tx_errors;
4728 	data[i++] = stats->tx_fifo_errors;
4729 	data[i++] = stats->tx_packets;
4730 	BUG_ON(i != CAS_NUM_STAT_KEYS);
4731 }
4732 
4733 static const struct ethtool_ops cas_ethtool_ops = {
4734 	.get_drvinfo		= cas_get_drvinfo,
4735 	.nway_reset		= cas_nway_reset,
4736 	.get_link		= cas_get_link,
4737 	.get_msglevel		= cas_get_msglevel,
4738 	.set_msglevel		= cas_set_msglevel,
4739 	.get_regs_len		= cas_get_regs_len,
4740 	.get_regs		= cas_get_regs,
4741 	.get_sset_count		= cas_get_sset_count,
4742 	.get_strings		= cas_get_strings,
4743 	.get_ethtool_stats	= cas_get_ethtool_stats,
4744 	.get_link_ksettings	= cas_get_link_ksettings,
4745 	.set_link_ksettings	= cas_set_link_ksettings,
4746 };
4747 
4748 static int cas_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
4749 {
4750 	struct cas *cp = netdev_priv(dev);
4751 	struct mii_ioctl_data *data = if_mii(ifr);
4752 	unsigned long flags;
4753 	int rc = -EOPNOTSUPP;
4754 
4755 	/* Hold the PM mutex while doing ioctl's or we may collide
4756 	 * with open/close and power management and oops.
4757 	 */
4758 	mutex_lock(&cp->pm_mutex);
4759 	switch (cmd) {
4760 	case SIOCGMIIPHY:		/* Get address of MII PHY in use. */
4761 		data->phy_id = cp->phy_addr;
4762 		fallthrough;
4763 
4764 	case SIOCGMIIREG:		/* Read MII PHY register. */
4765 		spin_lock_irqsave(&cp->lock, flags);
4766 		cas_mif_poll(cp, 0);
4767 		data->val_out = cas_phy_read(cp, data->reg_num & 0x1f);
4768 		cas_mif_poll(cp, 1);
4769 		spin_unlock_irqrestore(&cp->lock, flags);
4770 		rc = 0;
4771 		break;
4772 
4773 	case SIOCSMIIREG:		/* Write MII PHY register. */
4774 		spin_lock_irqsave(&cp->lock, flags);
4775 		cas_mif_poll(cp, 0);
4776 		rc = cas_phy_write(cp, data->reg_num & 0x1f, data->val_in);
4777 		cas_mif_poll(cp, 1);
4778 		spin_unlock_irqrestore(&cp->lock, flags);
4779 		break;
4780 	default:
4781 		break;
4782 	}
4783 
4784 	mutex_unlock(&cp->pm_mutex);
4785 	return rc;
4786 }
4787 
4788 /* When this chip sits underneath an Intel 31154 bridge, it is the
4789  * only subordinate device and we can tweak the bridge settings to
4790  * reflect that fact.
4791  */
4792 static void cas_program_bridge(struct pci_dev *cas_pdev)
4793 {
4794 	struct pci_dev *pdev = cas_pdev->bus->self;
4795 	u32 val;
4796 
4797 	if (!pdev)
4798 		return;
4799 
4800 	if (pdev->vendor != 0x8086 || pdev->device != 0x537c)
4801 		return;
4802 
4803 	/* Clear bit 10 (Bus Parking Control) in the Secondary
4804 	 * Arbiter Control/Status Register which lives at offset
4805 	 * 0x41.  Using a 32-bit word read/modify/write at 0x40
4806 	 * is much simpler so that's how we do this.
4807 	 */
4808 	pci_read_config_dword(pdev, 0x40, &val);
4809 	val &= ~0x00040000;
4810 	pci_write_config_dword(pdev, 0x40, val);
4811 
4812 	/* Max out the Multi-Transaction Timer settings since
4813 	 * Cassini is the only device present.
4814 	 *
4815 	 * The register is 16-bit and lives at 0x50.  When the
4816 	 * settings are enabled, it extends the GRANT# signal
4817 	 * for a requestor after a transaction is complete.  This
4818 	 * allows the next request to run without first needing
4819 	 * to negotiate the GRANT# signal back.
4820 	 *
4821 	 * Bits 12:10 define the grant duration:
4822 	 *
4823 	 *	1	--	16 clocks
4824 	 *	2	--	32 clocks
4825 	 *	3	--	64 clocks
4826 	 *	4	--	128 clocks
4827 	 *	5	--	256 clocks
4828 	 *
4829 	 * All other values are illegal.
4830 	 *
4831 	 * Bits 09:00 define which REQ/GNT signal pairs get the
4832 	 * GRANT# signal treatment.  We set them all.
4833 	 */
4834 	pci_write_config_word(pdev, 0x50, (5 << 10) | 0x3ff);
4835 
4836 	/* The Read Prefecth Policy register is 16-bit and sits at
4837 	 * offset 0x52.  It enables a "smart" pre-fetch policy.  We
4838 	 * enable it and max out all of the settings since only one
4839 	 * device is sitting underneath and thus bandwidth sharing is
4840 	 * not an issue.
4841 	 *
4842 	 * The register has several 3 bit fields, which indicates a
4843 	 * multiplier applied to the base amount of prefetching the
4844 	 * chip would do.  These fields are at:
4845 	 *
4846 	 *	15:13	---	ReRead Primary Bus
4847 	 *	12:10	---	FirstRead Primary Bus
4848 	 *	09:07	---	ReRead Secondary Bus
4849 	 *	06:04	---	FirstRead Secondary Bus
4850 	 *
4851 	 * Bits 03:00 control which REQ/GNT pairs the prefetch settings
4852 	 * get enabled on.  Bit 3 is a grouped enabler which controls
4853 	 * all of the REQ/GNT pairs from [8:3].  Bits 2 to 0 control
4854 	 * the individual REQ/GNT pairs [2:0].
4855 	 */
4856 	pci_write_config_word(pdev, 0x52,
4857 			      (0x7 << 13) |
4858 			      (0x7 << 10) |
4859 			      (0x7 <<  7) |
4860 			      (0x7 <<  4) |
4861 			      (0xf <<  0));
4862 
4863 	/* Force cacheline size to 0x8 */
4864 	pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE, 0x08);
4865 
4866 	/* Force latency timer to maximum setting so Cassini can
4867 	 * sit on the bus as long as it likes.
4868 	 */
4869 	pci_write_config_byte(pdev, PCI_LATENCY_TIMER, 0xff);
4870 }
4871 
4872 static const struct net_device_ops cas_netdev_ops = {
4873 	.ndo_open		= cas_open,
4874 	.ndo_stop		= cas_close,
4875 	.ndo_start_xmit		= cas_start_xmit,
4876 	.ndo_get_stats 		= cas_get_stats,
4877 	.ndo_set_rx_mode	= cas_set_multicast,
4878 	.ndo_do_ioctl		= cas_ioctl,
4879 	.ndo_tx_timeout		= cas_tx_timeout,
4880 	.ndo_change_mtu		= cas_change_mtu,
4881 	.ndo_set_mac_address	= eth_mac_addr,
4882 	.ndo_validate_addr	= eth_validate_addr,
4883 #ifdef CONFIG_NET_POLL_CONTROLLER
4884 	.ndo_poll_controller	= cas_netpoll,
4885 #endif
4886 };
4887 
4888 static int cas_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
4889 {
4890 	static int cas_version_printed = 0;
4891 	unsigned long casreg_len;
4892 	struct net_device *dev;
4893 	struct cas *cp;
4894 	int i, err, pci_using_dac;
4895 	u16 pci_cmd;
4896 	u8 orig_cacheline_size = 0, cas_cacheline_size = 0;
4897 
4898 	if (cas_version_printed++ == 0)
4899 		pr_info("%s", version);
4900 
4901 	err = pci_enable_device(pdev);
4902 	if (err) {
4903 		dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
4904 		return err;
4905 	}
4906 
4907 	if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
4908 		dev_err(&pdev->dev, "Cannot find proper PCI device "
4909 		       "base address, aborting\n");
4910 		err = -ENODEV;
4911 		goto err_out_disable_pdev;
4912 	}
4913 
4914 	dev = alloc_etherdev(sizeof(*cp));
4915 	if (!dev) {
4916 		err = -ENOMEM;
4917 		goto err_out_disable_pdev;
4918 	}
4919 	SET_NETDEV_DEV(dev, &pdev->dev);
4920 
4921 	err = pci_request_regions(pdev, dev->name);
4922 	if (err) {
4923 		dev_err(&pdev->dev, "Cannot obtain PCI resources, aborting\n");
4924 		goto err_out_free_netdev;
4925 	}
4926 	pci_set_master(pdev);
4927 
4928 	/* we must always turn on parity response or else parity
4929 	 * doesn't get generated properly. disable SERR/PERR as well.
4930 	 * in addition, we want to turn MWI on.
4931 	 */
4932 	pci_read_config_word(pdev, PCI_COMMAND, &pci_cmd);
4933 	pci_cmd &= ~PCI_COMMAND_SERR;
4934 	pci_cmd |= PCI_COMMAND_PARITY;
4935 	pci_write_config_word(pdev, PCI_COMMAND, pci_cmd);
4936 	if (pci_try_set_mwi(pdev))
4937 		pr_warn("Could not enable MWI for %s\n", pci_name(pdev));
4938 
4939 	cas_program_bridge(pdev);
4940 
4941 	/*
4942 	 * On some architectures, the default cache line size set
4943 	 * by pci_try_set_mwi reduces perforamnce.  We have to increase
4944 	 * it for this case.  To start, we'll print some configuration
4945 	 * data.
4946 	 */
4947 #if 1
4948 	pci_read_config_byte(pdev, PCI_CACHE_LINE_SIZE,
4949 			     &orig_cacheline_size);
4950 	if (orig_cacheline_size < CAS_PREF_CACHELINE_SIZE) {
4951 		cas_cacheline_size =
4952 			(CAS_PREF_CACHELINE_SIZE < SMP_CACHE_BYTES) ?
4953 			CAS_PREF_CACHELINE_SIZE : SMP_CACHE_BYTES;
4954 		if (pci_write_config_byte(pdev,
4955 					  PCI_CACHE_LINE_SIZE,
4956 					  cas_cacheline_size)) {
4957 			dev_err(&pdev->dev, "Could not set PCI cache "
4958 			       "line size\n");
4959 			goto err_out_free_res;
4960 		}
4961 	}
4962 #endif
4963 
4964 
4965 	/* Configure DMA attributes. */
4966 	if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64))) {
4967 		pci_using_dac = 1;
4968 		err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
4969 		if (err < 0) {
4970 			dev_err(&pdev->dev, "Unable to obtain 64-bit DMA "
4971 			       "for consistent allocations\n");
4972 			goto err_out_free_res;
4973 		}
4974 
4975 	} else {
4976 		err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
4977 		if (err) {
4978 			dev_err(&pdev->dev, "No usable DMA configuration, "
4979 			       "aborting\n");
4980 			goto err_out_free_res;
4981 		}
4982 		pci_using_dac = 0;
4983 	}
4984 
4985 	casreg_len = pci_resource_len(pdev, 0);
4986 
4987 	cp = netdev_priv(dev);
4988 	cp->pdev = pdev;
4989 #if 1
4990 	/* A value of 0 indicates we never explicitly set it */
4991 	cp->orig_cacheline_size = cas_cacheline_size ? orig_cacheline_size: 0;
4992 #endif
4993 	cp->dev = dev;
4994 	cp->msg_enable = (cassini_debug < 0) ? CAS_DEF_MSG_ENABLE :
4995 	  cassini_debug;
4996 
4997 #if defined(CONFIG_SPARC)
4998 	cp->of_node = pci_device_to_OF_node(pdev);
4999 #endif
5000 
5001 	cp->link_transition = LINK_TRANSITION_UNKNOWN;
5002 	cp->link_transition_jiffies_valid = 0;
5003 
5004 	spin_lock_init(&cp->lock);
5005 	spin_lock_init(&cp->rx_inuse_lock);
5006 	spin_lock_init(&cp->rx_spare_lock);
5007 	for (i = 0; i < N_TX_RINGS; i++) {
5008 		spin_lock_init(&cp->stat_lock[i]);
5009 		spin_lock_init(&cp->tx_lock[i]);
5010 	}
5011 	spin_lock_init(&cp->stat_lock[N_TX_RINGS]);
5012 	mutex_init(&cp->pm_mutex);
5013 
5014 	timer_setup(&cp->link_timer, cas_link_timer, 0);
5015 
5016 #if 1
5017 	/* Just in case the implementation of atomic operations
5018 	 * change so that an explicit initialization is necessary.
5019 	 */
5020 	atomic_set(&cp->reset_task_pending, 0);
5021 	atomic_set(&cp->reset_task_pending_all, 0);
5022 	atomic_set(&cp->reset_task_pending_spare, 0);
5023 	atomic_set(&cp->reset_task_pending_mtu, 0);
5024 #endif
5025 	INIT_WORK(&cp->reset_task, cas_reset_task);
5026 
5027 	/* Default link parameters */
5028 	if (link_mode >= 0 && link_mode < 6)
5029 		cp->link_cntl = link_modes[link_mode];
5030 	else
5031 		cp->link_cntl = BMCR_ANENABLE;
5032 	cp->lstate = link_down;
5033 	cp->link_transition = LINK_TRANSITION_LINK_DOWN;
5034 	netif_carrier_off(cp->dev);
5035 	cp->timer_ticks = 0;
5036 
5037 	/* give us access to cassini registers */
5038 	cp->regs = pci_iomap(pdev, 0, casreg_len);
5039 	if (!cp->regs) {
5040 		dev_err(&pdev->dev, "Cannot map device registers, aborting\n");
5041 		goto err_out_free_res;
5042 	}
5043 	cp->casreg_len = casreg_len;
5044 
5045 	pci_save_state(pdev);
5046 	cas_check_pci_invariants(cp);
5047 	cas_hard_reset(cp);
5048 	cas_reset(cp, 0);
5049 	if (cas_check_invariants(cp))
5050 		goto err_out_iounmap;
5051 	if (cp->cas_flags & CAS_FLAG_SATURN)
5052 		cas_saturn_firmware_init(cp);
5053 
5054 	cp->init_block =
5055 		dma_alloc_coherent(&pdev->dev, sizeof(struct cas_init_block),
5056 				   &cp->block_dvma, GFP_KERNEL);
5057 	if (!cp->init_block) {
5058 		dev_err(&pdev->dev, "Cannot allocate init block, aborting\n");
5059 		goto err_out_iounmap;
5060 	}
5061 
5062 	for (i = 0; i < N_TX_RINGS; i++)
5063 		cp->init_txds[i] = cp->init_block->txds[i];
5064 
5065 	for (i = 0; i < N_RX_DESC_RINGS; i++)
5066 		cp->init_rxds[i] = cp->init_block->rxds[i];
5067 
5068 	for (i = 0; i < N_RX_COMP_RINGS; i++)
5069 		cp->init_rxcs[i] = cp->init_block->rxcs[i];
5070 
5071 	for (i = 0; i < N_RX_FLOWS; i++)
5072 		skb_queue_head_init(&cp->rx_flows[i]);
5073 
5074 	dev->netdev_ops = &cas_netdev_ops;
5075 	dev->ethtool_ops = &cas_ethtool_ops;
5076 	dev->watchdog_timeo = CAS_TX_TIMEOUT;
5077 
5078 #ifdef USE_NAPI
5079 	netif_napi_add(dev, &cp->napi, cas_poll, 64);
5080 #endif
5081 	dev->irq = pdev->irq;
5082 	dev->dma = 0;
5083 
5084 	/* Cassini features. */
5085 	if ((cp->cas_flags & CAS_FLAG_NO_HW_CSUM) == 0)
5086 		dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG;
5087 
5088 	if (pci_using_dac)
5089 		dev->features |= NETIF_F_HIGHDMA;
5090 
5091 	/* MTU range: 60 - varies or 9000 */
5092 	dev->min_mtu = CAS_MIN_MTU;
5093 	dev->max_mtu = CAS_MAX_MTU;
5094 
5095 	if (register_netdev(dev)) {
5096 		dev_err(&pdev->dev, "Cannot register net device, aborting\n");
5097 		goto err_out_free_consistent;
5098 	}
5099 
5100 	i = readl(cp->regs + REG_BIM_CFG);
5101 	netdev_info(dev, "Sun Cassini%s (%sbit/%sMHz PCI/%s) Ethernet[%d] %pM\n",
5102 		    (cp->cas_flags & CAS_FLAG_REG_PLUS) ? "+" : "",
5103 		    (i & BIM_CFG_32BIT) ? "32" : "64",
5104 		    (i & BIM_CFG_66MHZ) ? "66" : "33",
5105 		    (cp->phy_type == CAS_PHY_SERDES) ? "Fi" : "Cu", pdev->irq,
5106 		    dev->dev_addr);
5107 
5108 	pci_set_drvdata(pdev, dev);
5109 	cp->hw_running = 1;
5110 	cas_entropy_reset(cp);
5111 	cas_phy_init(cp);
5112 	cas_begin_auto_negotiation(cp, NULL);
5113 	return 0;
5114 
5115 err_out_free_consistent:
5116 	dma_free_coherent(&pdev->dev, sizeof(struct cas_init_block),
5117 			  cp->init_block, cp->block_dvma);
5118 
5119 err_out_iounmap:
5120 	mutex_lock(&cp->pm_mutex);
5121 	if (cp->hw_running)
5122 		cas_shutdown(cp);
5123 	mutex_unlock(&cp->pm_mutex);
5124 
5125 	pci_iounmap(pdev, cp->regs);
5126 
5127 
5128 err_out_free_res:
5129 	pci_release_regions(pdev);
5130 
5131 	/* Try to restore it in case the error occurred after we
5132 	 * set it.
5133 	 */
5134 	pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE, orig_cacheline_size);
5135 
5136 err_out_free_netdev:
5137 	free_netdev(dev);
5138 
5139 err_out_disable_pdev:
5140 	pci_disable_device(pdev);
5141 	return -ENODEV;
5142 }
5143 
5144 static void cas_remove_one(struct pci_dev *pdev)
5145 {
5146 	struct net_device *dev = pci_get_drvdata(pdev);
5147 	struct cas *cp;
5148 	if (!dev)
5149 		return;
5150 
5151 	cp = netdev_priv(dev);
5152 	unregister_netdev(dev);
5153 
5154 	vfree(cp->fw_data);
5155 
5156 	mutex_lock(&cp->pm_mutex);
5157 	cancel_work_sync(&cp->reset_task);
5158 	if (cp->hw_running)
5159 		cas_shutdown(cp);
5160 	mutex_unlock(&cp->pm_mutex);
5161 
5162 #if 1
5163 	if (cp->orig_cacheline_size) {
5164 		/* Restore the cache line size if we had modified
5165 		 * it.
5166 		 */
5167 		pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE,
5168 				      cp->orig_cacheline_size);
5169 	}
5170 #endif
5171 	dma_free_coherent(&pdev->dev, sizeof(struct cas_init_block),
5172 			  cp->init_block, cp->block_dvma);
5173 	pci_iounmap(pdev, cp->regs);
5174 	free_netdev(dev);
5175 	pci_release_regions(pdev);
5176 	pci_disable_device(pdev);
5177 }
5178 
5179 static int __maybe_unused cas_suspend(struct device *dev_d)
5180 {
5181 	struct net_device *dev = dev_get_drvdata(dev_d);
5182 	struct cas *cp = netdev_priv(dev);
5183 	unsigned long flags;
5184 
5185 	mutex_lock(&cp->pm_mutex);
5186 
5187 	/* If the driver is opened, we stop the DMA */
5188 	if (cp->opened) {
5189 		netif_device_detach(dev);
5190 
5191 		cas_lock_all_save(cp, flags);
5192 
5193 		/* We can set the second arg of cas_reset to 0
5194 		 * because on resume, we'll call cas_init_hw with
5195 		 * its second arg set so that autonegotiation is
5196 		 * restarted.
5197 		 */
5198 		cas_reset(cp, 0);
5199 		cas_clean_rings(cp);
5200 		cas_unlock_all_restore(cp, flags);
5201 	}
5202 
5203 	if (cp->hw_running)
5204 		cas_shutdown(cp);
5205 	mutex_unlock(&cp->pm_mutex);
5206 
5207 	return 0;
5208 }
5209 
5210 static int __maybe_unused cas_resume(struct device *dev_d)
5211 {
5212 	struct net_device *dev = dev_get_drvdata(dev_d);
5213 	struct cas *cp = netdev_priv(dev);
5214 
5215 	netdev_info(dev, "resuming\n");
5216 
5217 	mutex_lock(&cp->pm_mutex);
5218 	cas_hard_reset(cp);
5219 	if (cp->opened) {
5220 		unsigned long flags;
5221 		cas_lock_all_save(cp, flags);
5222 		cas_reset(cp, 0);
5223 		cp->hw_running = 1;
5224 		cas_clean_rings(cp);
5225 		cas_init_hw(cp, 1);
5226 		cas_unlock_all_restore(cp, flags);
5227 
5228 		netif_device_attach(dev);
5229 	}
5230 	mutex_unlock(&cp->pm_mutex);
5231 	return 0;
5232 }
5233 
5234 static SIMPLE_DEV_PM_OPS(cas_pm_ops, cas_suspend, cas_resume);
5235 
5236 static struct pci_driver cas_driver = {
5237 	.name		= DRV_MODULE_NAME,
5238 	.id_table	= cas_pci_tbl,
5239 	.probe		= cas_init_one,
5240 	.remove		= cas_remove_one,
5241 	.driver.pm	= &cas_pm_ops,
5242 };
5243 
5244 static int __init cas_init(void)
5245 {
5246 	if (linkdown_timeout > 0)
5247 		link_transition_timeout = linkdown_timeout * HZ;
5248 	else
5249 		link_transition_timeout = 0;
5250 
5251 	return pci_register_driver(&cas_driver);
5252 }
5253 
5254 static void __exit cas_cleanup(void)
5255 {
5256 	pci_unregister_driver(&cas_driver);
5257 }
5258 
5259 module_init(cas_init);
5260 module_exit(cas_cleanup);
5261