xref: /linux/drivers/net/ethernet/sun/cassini.c (revision fc6dfd5547794b0bf10790576a9d97443d975439)
1 // SPDX-License-Identifier: GPL-2.0+
2 /* cassini.c: Sun Microsystems Cassini(+) ethernet driver.
3  *
4  * Copyright (C) 2004 Sun Microsystems Inc.
5  * Copyright (C) 2003 Adrian Sun (asun@darksunrising.com)
6  *
7  * This driver uses the sungem driver (c) David Miller
8  * (davem@redhat.com) as its basis.
9  *
10  * The cassini chip has a number of features that distinguish it from
11  * the gem chip:
12  *  4 transmit descriptor rings that are used for either QoS (VLAN) or
13  *      load balancing (non-VLAN mode)
14  *  batching of multiple packets
15  *  multiple CPU dispatching
16  *  page-based RX descriptor engine with separate completion rings
17  *  Gigabit support (GMII and PCS interface)
18  *  MIF link up/down detection works
19  *
20  * RX is handled by page sized buffers that are attached as fragments to
21  * the skb. here's what's done:
22  *  -- driver allocates pages at a time and keeps reference counts
23  *     on them.
24  *  -- the upper protocol layers assume that the header is in the skb
25  *     itself. as a result, cassini will copy a small amount (64 bytes)
26  *     to make them happy.
27  *  -- driver appends the rest of the data pages as frags to skbuffs
28  *     and increments the reference count
29  *  -- on page reclamation, the driver swaps the page with a spare page.
30  *     if that page is still in use, it frees its reference to that page,
31  *     and allocates a new page for use. otherwise, it just recycles the
32  *     page.
33  *
34  * NOTE: cassini can parse the header. however, it's not worth it
35  *       as long as the network stack requires a header copy.
36  *
37  * TX has 4 queues. currently these queues are used in a round-robin
38  * fashion for load balancing. They can also be used for QoS. for that
39  * to work, however, QoS information needs to be exposed down to the driver
40  * level so that subqueues get targeted to particular transmit rings.
41  * alternatively, the queues can be configured via use of the all-purpose
42  * ioctl.
43  *
44  * RX DATA: the rx completion ring has all the info, but the rx desc
45  * ring has all of the data. RX can conceivably come in under multiple
46  * interrupts, but the INT# assignment needs to be set up properly by
47  * the BIOS and conveyed to the driver. PCI BIOSes don't know how to do
48  * that. also, the two descriptor rings are designed to distinguish between
49  * encrypted and non-encrypted packets, but we use them for buffering
50  * instead.
51  *
52  * by default, the selective clear mask is set up to process rx packets.
53  */
54 
55 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
56 
57 #include <linux/module.h>
58 #include <linux/kernel.h>
59 #include <linux/types.h>
60 #include <linux/compiler.h>
61 #include <linux/slab.h>
62 #include <linux/delay.h>
63 #include <linux/init.h>
64 #include <linux/interrupt.h>
65 #include <linux/vmalloc.h>
66 #include <linux/ioport.h>
67 #include <linux/pci.h>
68 #include <linux/mm.h>
69 #include <linux/highmem.h>
70 #include <linux/list.h>
71 #include <linux/dma-mapping.h>
72 
73 #include <linux/netdevice.h>
74 #include <linux/etherdevice.h>
75 #include <linux/skbuff.h>
76 #include <linux/ethtool.h>
77 #include <linux/crc32.h>
78 #include <linux/random.h>
79 #include <linux/mii.h>
80 #include <linux/ip.h>
81 #include <linux/tcp.h>
82 #include <linux/mutex.h>
83 #include <linux/firmware.h>
84 
85 #include <net/checksum.h>
86 
87 #include <linux/atomic.h>
88 #include <asm/io.h>
89 #include <asm/byteorder.h>
90 #include <linux/uaccess.h>
91 #include <linux/jiffies.h>
92 
93 #define cas_page_map(x)      kmap_atomic((x))
94 #define cas_page_unmap(x)    kunmap_atomic((x))
95 #define CAS_NCPUS            num_online_cpus()
96 
97 #define cas_skb_release(x)  netif_rx(x)
98 
99 /* select which firmware to use */
100 #define USE_HP_WORKAROUND
101 #define HP_WORKAROUND_DEFAULT /* select which firmware to use as default */
102 #define CAS_HP_ALT_FIRMWARE   cas_prog_null /* alternate firmware */
103 
104 #include "cassini.h"
105 
106 #define USE_TX_COMPWB      /* use completion writeback registers */
107 #define USE_CSMA_CD_PROTO  /* standard CSMA/CD */
108 #define USE_RX_BLANK       /* hw interrupt mitigation */
109 #undef USE_ENTROPY_DEV     /* don't test for entropy device */
110 
111 /* NOTE: these aren't useable unless PCI interrupts can be assigned.
112  * also, we need to make cp->lock finer-grained.
113  */
114 #undef  USE_PCI_INTB
115 #undef  USE_PCI_INTC
116 #undef  USE_PCI_INTD
117 #undef  USE_QOS
118 
119 #undef  USE_VPD_DEBUG       /* debug vpd information if defined */
120 
121 /* rx processing options */
122 #define USE_PAGE_ORDER      /* specify to allocate large rx pages */
123 #define RX_DONT_BATCH  0    /* if 1, don't batch flows */
124 #define RX_COPY_ALWAYS 0    /* if 0, use frags */
125 #define RX_COPY_MIN    64   /* copy a little to make upper layers happy */
126 #undef  RX_COUNT_BUFFERS    /* define to calculate RX buffer stats */
127 
128 #define DRV_MODULE_NAME		"cassini"
129 #define DRV_MODULE_VERSION	"1.6"
130 #define DRV_MODULE_RELDATE	"21 May 2008"
131 
132 #define CAS_DEF_MSG_ENABLE	  \
133 	(NETIF_MSG_DRV		| \
134 	 NETIF_MSG_PROBE	| \
135 	 NETIF_MSG_LINK		| \
136 	 NETIF_MSG_TIMER	| \
137 	 NETIF_MSG_IFDOWN	| \
138 	 NETIF_MSG_IFUP		| \
139 	 NETIF_MSG_RX_ERR	| \
140 	 NETIF_MSG_TX_ERR)
141 
142 /* length of time before we decide the hardware is borked,
143  * and dev->tx_timeout() should be called to fix the problem
144  */
145 #define CAS_TX_TIMEOUT			(HZ)
146 #define CAS_LINK_TIMEOUT                (22*HZ/10)
147 #define CAS_LINK_FAST_TIMEOUT           (1)
148 
149 /* timeout values for state changing. these specify the number
150  * of 10us delays to be used before giving up.
151  */
152 #define STOP_TRIES_PHY 1000
153 #define STOP_TRIES     5000
154 
155 /* specify a minimum frame size to deal with some fifo issues
156  * max mtu == 2 * page size - ethernet header - 64 - swivel =
157  *            2 * page_size - 0x50
158  */
159 #define CAS_MIN_FRAME			97
160 #define CAS_1000MB_MIN_FRAME            255
161 #define CAS_MIN_MTU                     60
162 #define CAS_MAX_MTU                     min(((cp->page_size << 1) - 0x50), 9000)
163 
164 #if 1
165 /*
166  * Eliminate these and use separate atomic counters for each, to
167  * avoid a race condition.
168  */
169 #else
170 #define CAS_RESET_MTU                   1
171 #define CAS_RESET_ALL                   2
172 #define CAS_RESET_SPARE                 3
173 #endif
174 
175 static char version[] =
176 	DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
177 
178 static int cassini_debug = -1;	/* -1 == use CAS_DEF_MSG_ENABLE as value */
179 static int link_mode;
180 
181 MODULE_AUTHOR("Adrian Sun (asun@darksunrising.com)");
182 MODULE_DESCRIPTION("Sun Cassini(+) ethernet driver");
183 MODULE_LICENSE("GPL");
184 MODULE_FIRMWARE("sun/cassini.bin");
185 module_param(cassini_debug, int, 0);
186 MODULE_PARM_DESC(cassini_debug, "Cassini bitmapped debugging message enable value");
187 module_param(link_mode, int, 0);
188 MODULE_PARM_DESC(link_mode, "default link mode");
189 
190 /*
191  * Work around for a PCS bug in which the link goes down due to the chip
192  * being confused and never showing a link status of "up."
193  */
194 #define DEFAULT_LINKDOWN_TIMEOUT 5
195 /*
196  * Value in seconds, for user input.
197  */
198 static int linkdown_timeout = DEFAULT_LINKDOWN_TIMEOUT;
199 module_param(linkdown_timeout, int, 0);
200 MODULE_PARM_DESC(linkdown_timeout,
201 "min reset interval in sec. for PCS linkdown issue; disabled if not positive");
202 
203 /*
204  * value in 'ticks' (units used by jiffies). Set when we init the
205  * module because 'HZ' in actually a function call on some flavors of
206  * Linux.  This will default to DEFAULT_LINKDOWN_TIMEOUT * HZ.
207  */
208 static int link_transition_timeout;
209 
210 
211 
212 static u16 link_modes[] = {
213 	BMCR_ANENABLE,			 /* 0 : autoneg */
214 	0,				 /* 1 : 10bt half duplex */
215 	BMCR_SPEED100,			 /* 2 : 100bt half duplex */
216 	BMCR_FULLDPLX,			 /* 3 : 10bt full duplex */
217 	BMCR_SPEED100|BMCR_FULLDPLX,	 /* 4 : 100bt full duplex */
218 	CAS_BMCR_SPEED1000|BMCR_FULLDPLX /* 5 : 1000bt full duplex */
219 };
220 
221 static const struct pci_device_id cas_pci_tbl[] = {
222 	{ PCI_VENDOR_ID_SUN, PCI_DEVICE_ID_SUN_CASSINI,
223 	  PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0UL },
224 	{ PCI_VENDOR_ID_NS, PCI_DEVICE_ID_NS_SATURN,
225 	  PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0UL },
226 	{ 0, }
227 };
228 
229 MODULE_DEVICE_TABLE(pci, cas_pci_tbl);
230 
231 static void cas_set_link_modes(struct cas *cp);
232 
233 static inline void cas_lock_tx(struct cas *cp)
234 {
235 	int i;
236 
237 	for (i = 0; i < N_TX_RINGS; i++)
238 		spin_lock_nested(&cp->tx_lock[i], i);
239 }
240 
241 /* WTZ: QA was finding deadlock problems with the previous
242  * versions after long test runs with multiple cards per machine.
243  * See if replacing cas_lock_all with safer versions helps. The
244  * symptoms QA is reporting match those we'd expect if interrupts
245  * aren't being properly restored, and we fixed a previous deadlock
246  * with similar symptoms by using save/restore versions in other
247  * places.
248  */
249 #define cas_lock_all_save(cp, flags) \
250 do { \
251 	struct cas *xxxcp = (cp); \
252 	spin_lock_irqsave(&xxxcp->lock, flags); \
253 	cas_lock_tx(xxxcp); \
254 } while (0)
255 
256 static inline void cas_unlock_tx(struct cas *cp)
257 {
258 	int i;
259 
260 	for (i = N_TX_RINGS; i > 0; i--)
261 		spin_unlock(&cp->tx_lock[i - 1]);
262 }
263 
264 #define cas_unlock_all_restore(cp, flags) \
265 do { \
266 	struct cas *xxxcp = (cp); \
267 	cas_unlock_tx(xxxcp); \
268 	spin_unlock_irqrestore(&xxxcp->lock, flags); \
269 } while (0)
270 
271 static void cas_disable_irq(struct cas *cp, const int ring)
272 {
273 	/* Make sure we won't get any more interrupts */
274 	if (ring == 0) {
275 		writel(0xFFFFFFFF, cp->regs + REG_INTR_MASK);
276 		return;
277 	}
278 
279 	/* disable completion interrupts and selectively mask */
280 	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
281 		switch (ring) {
282 #if defined (USE_PCI_INTB) || defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
283 #ifdef USE_PCI_INTB
284 		case 1:
285 #endif
286 #ifdef USE_PCI_INTC
287 		case 2:
288 #endif
289 #ifdef USE_PCI_INTD
290 		case 3:
291 #endif
292 			writel(INTRN_MASK_CLEAR_ALL | INTRN_MASK_RX_EN,
293 			       cp->regs + REG_PLUS_INTRN_MASK(ring));
294 			break;
295 #endif
296 		default:
297 			writel(INTRN_MASK_CLEAR_ALL, cp->regs +
298 			       REG_PLUS_INTRN_MASK(ring));
299 			break;
300 		}
301 	}
302 }
303 
304 static inline void cas_mask_intr(struct cas *cp)
305 {
306 	int i;
307 
308 	for (i = 0; i < N_RX_COMP_RINGS; i++)
309 		cas_disable_irq(cp, i);
310 }
311 
312 static void cas_enable_irq(struct cas *cp, const int ring)
313 {
314 	if (ring == 0) { /* all but TX_DONE */
315 		writel(INTR_TX_DONE, cp->regs + REG_INTR_MASK);
316 		return;
317 	}
318 
319 	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
320 		switch (ring) {
321 #if defined (USE_PCI_INTB) || defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
322 #ifdef USE_PCI_INTB
323 		case 1:
324 #endif
325 #ifdef USE_PCI_INTC
326 		case 2:
327 #endif
328 #ifdef USE_PCI_INTD
329 		case 3:
330 #endif
331 			writel(INTRN_MASK_RX_EN, cp->regs +
332 			       REG_PLUS_INTRN_MASK(ring));
333 			break;
334 #endif
335 		default:
336 			break;
337 		}
338 	}
339 }
340 
341 static inline void cas_unmask_intr(struct cas *cp)
342 {
343 	int i;
344 
345 	for (i = 0; i < N_RX_COMP_RINGS; i++)
346 		cas_enable_irq(cp, i);
347 }
348 
349 static inline void cas_entropy_gather(struct cas *cp)
350 {
351 #ifdef USE_ENTROPY_DEV
352 	if ((cp->cas_flags & CAS_FLAG_ENTROPY_DEV) == 0)
353 		return;
354 
355 	batch_entropy_store(readl(cp->regs + REG_ENTROPY_IV),
356 			    readl(cp->regs + REG_ENTROPY_IV),
357 			    sizeof(uint64_t)*8);
358 #endif
359 }
360 
361 static inline void cas_entropy_reset(struct cas *cp)
362 {
363 #ifdef USE_ENTROPY_DEV
364 	if ((cp->cas_flags & CAS_FLAG_ENTROPY_DEV) == 0)
365 		return;
366 
367 	writel(BIM_LOCAL_DEV_PAD | BIM_LOCAL_DEV_PROM | BIM_LOCAL_DEV_EXT,
368 	       cp->regs + REG_BIM_LOCAL_DEV_EN);
369 	writeb(ENTROPY_RESET_STC_MODE, cp->regs + REG_ENTROPY_RESET);
370 	writeb(0x55, cp->regs + REG_ENTROPY_RAND_REG);
371 
372 	/* if we read back 0x0, we don't have an entropy device */
373 	if (readb(cp->regs + REG_ENTROPY_RAND_REG) == 0)
374 		cp->cas_flags &= ~CAS_FLAG_ENTROPY_DEV;
375 #endif
376 }
377 
378 /* access to the phy. the following assumes that we've initialized the MIF to
379  * be in frame rather than bit-bang mode
380  */
381 static u16 cas_phy_read(struct cas *cp, int reg)
382 {
383 	u32 cmd;
384 	int limit = STOP_TRIES_PHY;
385 
386 	cmd = MIF_FRAME_ST | MIF_FRAME_OP_READ;
387 	cmd |= CAS_BASE(MIF_FRAME_PHY_ADDR, cp->phy_addr);
388 	cmd |= CAS_BASE(MIF_FRAME_REG_ADDR, reg);
389 	cmd |= MIF_FRAME_TURN_AROUND_MSB;
390 	writel(cmd, cp->regs + REG_MIF_FRAME);
391 
392 	/* poll for completion */
393 	while (limit-- > 0) {
394 		udelay(10);
395 		cmd = readl(cp->regs + REG_MIF_FRAME);
396 		if (cmd & MIF_FRAME_TURN_AROUND_LSB)
397 			return cmd & MIF_FRAME_DATA_MASK;
398 	}
399 	return 0xFFFF; /* -1 */
400 }
401 
402 static int cas_phy_write(struct cas *cp, int reg, u16 val)
403 {
404 	int limit = STOP_TRIES_PHY;
405 	u32 cmd;
406 
407 	cmd = MIF_FRAME_ST | MIF_FRAME_OP_WRITE;
408 	cmd |= CAS_BASE(MIF_FRAME_PHY_ADDR, cp->phy_addr);
409 	cmd |= CAS_BASE(MIF_FRAME_REG_ADDR, reg);
410 	cmd |= MIF_FRAME_TURN_AROUND_MSB;
411 	cmd |= val & MIF_FRAME_DATA_MASK;
412 	writel(cmd, cp->regs + REG_MIF_FRAME);
413 
414 	/* poll for completion */
415 	while (limit-- > 0) {
416 		udelay(10);
417 		cmd = readl(cp->regs + REG_MIF_FRAME);
418 		if (cmd & MIF_FRAME_TURN_AROUND_LSB)
419 			return 0;
420 	}
421 	return -1;
422 }
423 
424 static void cas_phy_powerup(struct cas *cp)
425 {
426 	u16 ctl = cas_phy_read(cp, MII_BMCR);
427 
428 	if ((ctl & BMCR_PDOWN) == 0)
429 		return;
430 	ctl &= ~BMCR_PDOWN;
431 	cas_phy_write(cp, MII_BMCR, ctl);
432 }
433 
434 static void cas_phy_powerdown(struct cas *cp)
435 {
436 	u16 ctl = cas_phy_read(cp, MII_BMCR);
437 
438 	if (ctl & BMCR_PDOWN)
439 		return;
440 	ctl |= BMCR_PDOWN;
441 	cas_phy_write(cp, MII_BMCR, ctl);
442 }
443 
444 /* cp->lock held. note: the last put_page will free the buffer */
445 static int cas_page_free(struct cas *cp, cas_page_t *page)
446 {
447 	dma_unmap_page(&cp->pdev->dev, page->dma_addr, cp->page_size,
448 		       DMA_FROM_DEVICE);
449 	__free_pages(page->buffer, cp->page_order);
450 	kfree(page);
451 	return 0;
452 }
453 
454 #ifdef RX_COUNT_BUFFERS
455 #define RX_USED_ADD(x, y)       ((x)->used += (y))
456 #define RX_USED_SET(x, y)       ((x)->used  = (y))
457 #else
458 #define RX_USED_ADD(x, y) do { } while(0)
459 #define RX_USED_SET(x, y) do { } while(0)
460 #endif
461 
462 /* local page allocation routines for the receive buffers. jumbo pages
463  * require at least 8K contiguous and 8K aligned buffers.
464  */
465 static cas_page_t *cas_page_alloc(struct cas *cp, const gfp_t flags)
466 {
467 	cas_page_t *page;
468 
469 	page = kmalloc(sizeof(cas_page_t), flags);
470 	if (!page)
471 		return NULL;
472 
473 	INIT_LIST_HEAD(&page->list);
474 	RX_USED_SET(page, 0);
475 	page->buffer = alloc_pages(flags, cp->page_order);
476 	if (!page->buffer)
477 		goto page_err;
478 	page->dma_addr = dma_map_page(&cp->pdev->dev, page->buffer, 0,
479 				      cp->page_size, DMA_FROM_DEVICE);
480 	return page;
481 
482 page_err:
483 	kfree(page);
484 	return NULL;
485 }
486 
487 /* initialize spare pool of rx buffers, but allocate during the open */
488 static void cas_spare_init(struct cas *cp)
489 {
490 	spin_lock(&cp->rx_inuse_lock);
491 	INIT_LIST_HEAD(&cp->rx_inuse_list);
492 	spin_unlock(&cp->rx_inuse_lock);
493 
494 	spin_lock(&cp->rx_spare_lock);
495 	INIT_LIST_HEAD(&cp->rx_spare_list);
496 	cp->rx_spares_needed = RX_SPARE_COUNT;
497 	spin_unlock(&cp->rx_spare_lock);
498 }
499 
500 /* used on close. free all the spare buffers. */
501 static void cas_spare_free(struct cas *cp)
502 {
503 	struct list_head list, *elem, *tmp;
504 
505 	/* free spare buffers */
506 	INIT_LIST_HEAD(&list);
507 	spin_lock(&cp->rx_spare_lock);
508 	list_splice_init(&cp->rx_spare_list, &list);
509 	spin_unlock(&cp->rx_spare_lock);
510 	list_for_each_safe(elem, tmp, &list) {
511 		cas_page_free(cp, list_entry(elem, cas_page_t, list));
512 	}
513 
514 	INIT_LIST_HEAD(&list);
515 #if 1
516 	/*
517 	 * Looks like Adrian had protected this with a different
518 	 * lock than used everywhere else to manipulate this list.
519 	 */
520 	spin_lock(&cp->rx_inuse_lock);
521 	list_splice_init(&cp->rx_inuse_list, &list);
522 	spin_unlock(&cp->rx_inuse_lock);
523 #else
524 	spin_lock(&cp->rx_spare_lock);
525 	list_splice_init(&cp->rx_inuse_list, &list);
526 	spin_unlock(&cp->rx_spare_lock);
527 #endif
528 	list_for_each_safe(elem, tmp, &list) {
529 		cas_page_free(cp, list_entry(elem, cas_page_t, list));
530 	}
531 }
532 
533 /* replenish spares if needed */
534 static void cas_spare_recover(struct cas *cp, const gfp_t flags)
535 {
536 	struct list_head list, *elem, *tmp;
537 	int needed, i;
538 
539 	/* check inuse list. if we don't need any more free buffers,
540 	 * just free it
541 	 */
542 
543 	/* make a local copy of the list */
544 	INIT_LIST_HEAD(&list);
545 	spin_lock(&cp->rx_inuse_lock);
546 	list_splice_init(&cp->rx_inuse_list, &list);
547 	spin_unlock(&cp->rx_inuse_lock);
548 
549 	list_for_each_safe(elem, tmp, &list) {
550 		cas_page_t *page = list_entry(elem, cas_page_t, list);
551 
552 		/*
553 		 * With the lockless pagecache, cassini buffering scheme gets
554 		 * slightly less accurate: we might find that a page has an
555 		 * elevated reference count here, due to a speculative ref,
556 		 * and skip it as in-use. Ideally we would be able to reclaim
557 		 * it. However this would be such a rare case, it doesn't
558 		 * matter too much as we should pick it up the next time round.
559 		 *
560 		 * Importantly, if we find that the page has a refcount of 1
561 		 * here (our refcount), then we know it is definitely not inuse
562 		 * so we can reuse it.
563 		 */
564 		if (page_count(page->buffer) > 1)
565 			continue;
566 
567 		list_del(elem);
568 		spin_lock(&cp->rx_spare_lock);
569 		if (cp->rx_spares_needed > 0) {
570 			list_add(elem, &cp->rx_spare_list);
571 			cp->rx_spares_needed--;
572 			spin_unlock(&cp->rx_spare_lock);
573 		} else {
574 			spin_unlock(&cp->rx_spare_lock);
575 			cas_page_free(cp, page);
576 		}
577 	}
578 
579 	/* put any inuse buffers back on the list */
580 	if (!list_empty(&list)) {
581 		spin_lock(&cp->rx_inuse_lock);
582 		list_splice(&list, &cp->rx_inuse_list);
583 		spin_unlock(&cp->rx_inuse_lock);
584 	}
585 
586 	spin_lock(&cp->rx_spare_lock);
587 	needed = cp->rx_spares_needed;
588 	spin_unlock(&cp->rx_spare_lock);
589 	if (!needed)
590 		return;
591 
592 	/* we still need spares, so try to allocate some */
593 	INIT_LIST_HEAD(&list);
594 	i = 0;
595 	while (i < needed) {
596 		cas_page_t *spare = cas_page_alloc(cp, flags);
597 		if (!spare)
598 			break;
599 		list_add(&spare->list, &list);
600 		i++;
601 	}
602 
603 	spin_lock(&cp->rx_spare_lock);
604 	list_splice(&list, &cp->rx_spare_list);
605 	cp->rx_spares_needed -= i;
606 	spin_unlock(&cp->rx_spare_lock);
607 }
608 
609 /* pull a page from the list. */
610 static cas_page_t *cas_page_dequeue(struct cas *cp)
611 {
612 	struct list_head *entry;
613 	int recover;
614 
615 	spin_lock(&cp->rx_spare_lock);
616 	if (list_empty(&cp->rx_spare_list)) {
617 		/* try to do a quick recovery */
618 		spin_unlock(&cp->rx_spare_lock);
619 		cas_spare_recover(cp, GFP_ATOMIC);
620 		spin_lock(&cp->rx_spare_lock);
621 		if (list_empty(&cp->rx_spare_list)) {
622 			netif_err(cp, rx_err, cp->dev,
623 				  "no spare buffers available\n");
624 			spin_unlock(&cp->rx_spare_lock);
625 			return NULL;
626 		}
627 	}
628 
629 	entry = cp->rx_spare_list.next;
630 	list_del(entry);
631 	recover = ++cp->rx_spares_needed;
632 	spin_unlock(&cp->rx_spare_lock);
633 
634 	/* trigger the timer to do the recovery */
635 	if ((recover & (RX_SPARE_RECOVER_VAL - 1)) == 0) {
636 #if 1
637 		atomic_inc(&cp->reset_task_pending);
638 		atomic_inc(&cp->reset_task_pending_spare);
639 		schedule_work(&cp->reset_task);
640 #else
641 		atomic_set(&cp->reset_task_pending, CAS_RESET_SPARE);
642 		schedule_work(&cp->reset_task);
643 #endif
644 	}
645 	return list_entry(entry, cas_page_t, list);
646 }
647 
648 
649 static void cas_mif_poll(struct cas *cp, const int enable)
650 {
651 	u32 cfg;
652 
653 	cfg  = readl(cp->regs + REG_MIF_CFG);
654 	cfg &= (MIF_CFG_MDIO_0 | MIF_CFG_MDIO_1);
655 
656 	if (cp->phy_type & CAS_PHY_MII_MDIO1)
657 		cfg |= MIF_CFG_PHY_SELECT;
658 
659 	/* poll and interrupt on link status change. */
660 	if (enable) {
661 		cfg |= MIF_CFG_POLL_EN;
662 		cfg |= CAS_BASE(MIF_CFG_POLL_REG, MII_BMSR);
663 		cfg |= CAS_BASE(MIF_CFG_POLL_PHY, cp->phy_addr);
664 	}
665 	writel((enable) ? ~(BMSR_LSTATUS | BMSR_ANEGCOMPLETE) : 0xFFFF,
666 	       cp->regs + REG_MIF_MASK);
667 	writel(cfg, cp->regs + REG_MIF_CFG);
668 }
669 
670 /* Must be invoked under cp->lock */
671 static void cas_begin_auto_negotiation(struct cas *cp,
672 				       const struct ethtool_link_ksettings *ep)
673 {
674 	u16 ctl;
675 #if 1
676 	int lcntl;
677 	int changed = 0;
678 	int oldstate = cp->lstate;
679 	int link_was_not_down = !(oldstate == link_down);
680 #endif
681 	/* Setup link parameters */
682 	if (!ep)
683 		goto start_aneg;
684 	lcntl = cp->link_cntl;
685 	if (ep->base.autoneg == AUTONEG_ENABLE) {
686 		cp->link_cntl = BMCR_ANENABLE;
687 	} else {
688 		u32 speed = ep->base.speed;
689 		cp->link_cntl = 0;
690 		if (speed == SPEED_100)
691 			cp->link_cntl |= BMCR_SPEED100;
692 		else if (speed == SPEED_1000)
693 			cp->link_cntl |= CAS_BMCR_SPEED1000;
694 		if (ep->base.duplex == DUPLEX_FULL)
695 			cp->link_cntl |= BMCR_FULLDPLX;
696 	}
697 #if 1
698 	changed = (lcntl != cp->link_cntl);
699 #endif
700 start_aneg:
701 	if (cp->lstate == link_up) {
702 		netdev_info(cp->dev, "PCS link down\n");
703 	} else {
704 		if (changed) {
705 			netdev_info(cp->dev, "link configuration changed\n");
706 		}
707 	}
708 	cp->lstate = link_down;
709 	cp->link_transition = LINK_TRANSITION_LINK_DOWN;
710 	if (!cp->hw_running)
711 		return;
712 #if 1
713 	/*
714 	 * WTZ: If the old state was link_up, we turn off the carrier
715 	 * to replicate everything we do elsewhere on a link-down
716 	 * event when we were already in a link-up state..
717 	 */
718 	if (oldstate == link_up)
719 		netif_carrier_off(cp->dev);
720 	if (changed  && link_was_not_down) {
721 		/*
722 		 * WTZ: This branch will simply schedule a full reset after
723 		 * we explicitly changed link modes in an ioctl. See if this
724 		 * fixes the link-problems we were having for forced mode.
725 		 */
726 		atomic_inc(&cp->reset_task_pending);
727 		atomic_inc(&cp->reset_task_pending_all);
728 		schedule_work(&cp->reset_task);
729 		cp->timer_ticks = 0;
730 		mod_timer(&cp->link_timer, jiffies + CAS_LINK_TIMEOUT);
731 		return;
732 	}
733 #endif
734 	if (cp->phy_type & CAS_PHY_SERDES) {
735 		u32 val = readl(cp->regs + REG_PCS_MII_CTRL);
736 
737 		if (cp->link_cntl & BMCR_ANENABLE) {
738 			val |= (PCS_MII_RESTART_AUTONEG | PCS_MII_AUTONEG_EN);
739 			cp->lstate = link_aneg;
740 		} else {
741 			if (cp->link_cntl & BMCR_FULLDPLX)
742 				val |= PCS_MII_CTRL_DUPLEX;
743 			val &= ~PCS_MII_AUTONEG_EN;
744 			cp->lstate = link_force_ok;
745 		}
746 		cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
747 		writel(val, cp->regs + REG_PCS_MII_CTRL);
748 
749 	} else {
750 		cas_mif_poll(cp, 0);
751 		ctl = cas_phy_read(cp, MII_BMCR);
752 		ctl &= ~(BMCR_FULLDPLX | BMCR_SPEED100 |
753 			 CAS_BMCR_SPEED1000 | BMCR_ANENABLE);
754 		ctl |= cp->link_cntl;
755 		if (ctl & BMCR_ANENABLE) {
756 			ctl |= BMCR_ANRESTART;
757 			cp->lstate = link_aneg;
758 		} else {
759 			cp->lstate = link_force_ok;
760 		}
761 		cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
762 		cas_phy_write(cp, MII_BMCR, ctl);
763 		cas_mif_poll(cp, 1);
764 	}
765 
766 	cp->timer_ticks = 0;
767 	mod_timer(&cp->link_timer, jiffies + CAS_LINK_TIMEOUT);
768 }
769 
770 /* Must be invoked under cp->lock. */
771 static int cas_reset_mii_phy(struct cas *cp)
772 {
773 	int limit = STOP_TRIES_PHY;
774 	u16 val;
775 
776 	cas_phy_write(cp, MII_BMCR, BMCR_RESET);
777 	udelay(100);
778 	while (--limit) {
779 		val = cas_phy_read(cp, MII_BMCR);
780 		if ((val & BMCR_RESET) == 0)
781 			break;
782 		udelay(10);
783 	}
784 	return limit <= 0;
785 }
786 
787 static void cas_saturn_firmware_init(struct cas *cp)
788 {
789 	const struct firmware *fw;
790 	const char fw_name[] = "sun/cassini.bin";
791 	int err;
792 
793 	if (PHY_NS_DP83065 != cp->phy_id)
794 		return;
795 
796 	err = request_firmware(&fw, fw_name, &cp->pdev->dev);
797 	if (err) {
798 		pr_err("Failed to load firmware \"%s\"\n",
799 		       fw_name);
800 		return;
801 	}
802 	if (fw->size < 2) {
803 		pr_err("bogus length %zu in \"%s\"\n",
804 		       fw->size, fw_name);
805 		goto out;
806 	}
807 	cp->fw_load_addr= fw->data[1] << 8 | fw->data[0];
808 	cp->fw_size = fw->size - 2;
809 	cp->fw_data = vmalloc(cp->fw_size);
810 	if (!cp->fw_data)
811 		goto out;
812 	memcpy(cp->fw_data, &fw->data[2], cp->fw_size);
813 out:
814 	release_firmware(fw);
815 }
816 
817 static void cas_saturn_firmware_load(struct cas *cp)
818 {
819 	int i;
820 
821 	if (!cp->fw_data)
822 		return;
823 
824 	cas_phy_powerdown(cp);
825 
826 	/* expanded memory access mode */
827 	cas_phy_write(cp, DP83065_MII_MEM, 0x0);
828 
829 	/* pointer configuration for new firmware */
830 	cas_phy_write(cp, DP83065_MII_REGE, 0x8ff9);
831 	cas_phy_write(cp, DP83065_MII_REGD, 0xbd);
832 	cas_phy_write(cp, DP83065_MII_REGE, 0x8ffa);
833 	cas_phy_write(cp, DP83065_MII_REGD, 0x82);
834 	cas_phy_write(cp, DP83065_MII_REGE, 0x8ffb);
835 	cas_phy_write(cp, DP83065_MII_REGD, 0x0);
836 	cas_phy_write(cp, DP83065_MII_REGE, 0x8ffc);
837 	cas_phy_write(cp, DP83065_MII_REGD, 0x39);
838 
839 	/* download new firmware */
840 	cas_phy_write(cp, DP83065_MII_MEM, 0x1);
841 	cas_phy_write(cp, DP83065_MII_REGE, cp->fw_load_addr);
842 	for (i = 0; i < cp->fw_size; i++)
843 		cas_phy_write(cp, DP83065_MII_REGD, cp->fw_data[i]);
844 
845 	/* enable firmware */
846 	cas_phy_write(cp, DP83065_MII_REGE, 0x8ff8);
847 	cas_phy_write(cp, DP83065_MII_REGD, 0x1);
848 }
849 
850 
851 /* phy initialization */
852 static void cas_phy_init(struct cas *cp)
853 {
854 	u16 val;
855 
856 	/* if we're in MII/GMII mode, set up phy */
857 	if (CAS_PHY_MII(cp->phy_type)) {
858 		writel(PCS_DATAPATH_MODE_MII,
859 		       cp->regs + REG_PCS_DATAPATH_MODE);
860 
861 		cas_mif_poll(cp, 0);
862 		cas_reset_mii_phy(cp); /* take out of isolate mode */
863 
864 		if (PHY_LUCENT_B0 == cp->phy_id) {
865 			/* workaround link up/down issue with lucent */
866 			cas_phy_write(cp, LUCENT_MII_REG, 0x8000);
867 			cas_phy_write(cp, MII_BMCR, 0x00f1);
868 			cas_phy_write(cp, LUCENT_MII_REG, 0x0);
869 
870 		} else if (PHY_BROADCOM_B0 == (cp->phy_id & 0xFFFFFFFC)) {
871 			/* workarounds for broadcom phy */
872 			cas_phy_write(cp, BROADCOM_MII_REG8, 0x0C20);
873 			cas_phy_write(cp, BROADCOM_MII_REG7, 0x0012);
874 			cas_phy_write(cp, BROADCOM_MII_REG5, 0x1804);
875 			cas_phy_write(cp, BROADCOM_MII_REG7, 0x0013);
876 			cas_phy_write(cp, BROADCOM_MII_REG5, 0x1204);
877 			cas_phy_write(cp, BROADCOM_MII_REG7, 0x8006);
878 			cas_phy_write(cp, BROADCOM_MII_REG5, 0x0132);
879 			cas_phy_write(cp, BROADCOM_MII_REG7, 0x8006);
880 			cas_phy_write(cp, BROADCOM_MII_REG5, 0x0232);
881 			cas_phy_write(cp, BROADCOM_MII_REG7, 0x201F);
882 			cas_phy_write(cp, BROADCOM_MII_REG5, 0x0A20);
883 
884 		} else if (PHY_BROADCOM_5411 == cp->phy_id) {
885 			val = cas_phy_read(cp, BROADCOM_MII_REG4);
886 			val = cas_phy_read(cp, BROADCOM_MII_REG4);
887 			if (val & 0x0080) {
888 				/* link workaround */
889 				cas_phy_write(cp, BROADCOM_MII_REG4,
890 					      val & ~0x0080);
891 			}
892 
893 		} else if (cp->cas_flags & CAS_FLAG_SATURN) {
894 			writel((cp->phy_type & CAS_PHY_MII_MDIO0) ?
895 			       SATURN_PCFG_FSI : 0x0,
896 			       cp->regs + REG_SATURN_PCFG);
897 
898 			/* load firmware to address 10Mbps auto-negotiation
899 			 * issue. NOTE: this will need to be changed if the
900 			 * default firmware gets fixed.
901 			 */
902 			if (PHY_NS_DP83065 == cp->phy_id) {
903 				cas_saturn_firmware_load(cp);
904 			}
905 			cas_phy_powerup(cp);
906 		}
907 
908 		/* advertise capabilities */
909 		val = cas_phy_read(cp, MII_BMCR);
910 		val &= ~BMCR_ANENABLE;
911 		cas_phy_write(cp, MII_BMCR, val);
912 		udelay(10);
913 
914 		cas_phy_write(cp, MII_ADVERTISE,
915 			      cas_phy_read(cp, MII_ADVERTISE) |
916 			      (ADVERTISE_10HALF | ADVERTISE_10FULL |
917 			       ADVERTISE_100HALF | ADVERTISE_100FULL |
918 			       CAS_ADVERTISE_PAUSE |
919 			       CAS_ADVERTISE_ASYM_PAUSE));
920 
921 		if (cp->cas_flags & CAS_FLAG_1000MB_CAP) {
922 			/* make sure that we don't advertise half
923 			 * duplex to avoid a chip issue
924 			 */
925 			val  = cas_phy_read(cp, CAS_MII_1000_CTRL);
926 			val &= ~CAS_ADVERTISE_1000HALF;
927 			val |= CAS_ADVERTISE_1000FULL;
928 			cas_phy_write(cp, CAS_MII_1000_CTRL, val);
929 		}
930 
931 	} else {
932 		/* reset pcs for serdes */
933 		u32 val;
934 		int limit;
935 
936 		writel(PCS_DATAPATH_MODE_SERDES,
937 		       cp->regs + REG_PCS_DATAPATH_MODE);
938 
939 		/* enable serdes pins on saturn */
940 		if (cp->cas_flags & CAS_FLAG_SATURN)
941 			writel(0, cp->regs + REG_SATURN_PCFG);
942 
943 		/* Reset PCS unit. */
944 		val = readl(cp->regs + REG_PCS_MII_CTRL);
945 		val |= PCS_MII_RESET;
946 		writel(val, cp->regs + REG_PCS_MII_CTRL);
947 
948 		limit = STOP_TRIES;
949 		while (--limit > 0) {
950 			udelay(10);
951 			if ((readl(cp->regs + REG_PCS_MII_CTRL) &
952 			     PCS_MII_RESET) == 0)
953 				break;
954 		}
955 		if (limit <= 0)
956 			netdev_warn(cp->dev, "PCS reset bit would not clear [%08x]\n",
957 				    readl(cp->regs + REG_PCS_STATE_MACHINE));
958 
959 		/* Make sure PCS is disabled while changing advertisement
960 		 * configuration.
961 		 */
962 		writel(0x0, cp->regs + REG_PCS_CFG);
963 
964 		/* Advertise all capabilities except half-duplex. */
965 		val  = readl(cp->regs + REG_PCS_MII_ADVERT);
966 		val &= ~PCS_MII_ADVERT_HD;
967 		val |= (PCS_MII_ADVERT_FD | PCS_MII_ADVERT_SYM_PAUSE |
968 			PCS_MII_ADVERT_ASYM_PAUSE);
969 		writel(val, cp->regs + REG_PCS_MII_ADVERT);
970 
971 		/* enable PCS */
972 		writel(PCS_CFG_EN, cp->regs + REG_PCS_CFG);
973 
974 		/* pcs workaround: enable sync detect */
975 		writel(PCS_SERDES_CTRL_SYNCD_EN,
976 		       cp->regs + REG_PCS_SERDES_CTRL);
977 	}
978 }
979 
980 
981 static int cas_pcs_link_check(struct cas *cp)
982 {
983 	u32 stat, state_machine;
984 	int retval = 0;
985 
986 	/* The link status bit latches on zero, so you must
987 	 * read it twice in such a case to see a transition
988 	 * to the link being up.
989 	 */
990 	stat = readl(cp->regs + REG_PCS_MII_STATUS);
991 	if ((stat & PCS_MII_STATUS_LINK_STATUS) == 0)
992 		stat = readl(cp->regs + REG_PCS_MII_STATUS);
993 
994 	/* The remote-fault indication is only valid
995 	 * when autoneg has completed.
996 	 */
997 	if ((stat & (PCS_MII_STATUS_AUTONEG_COMP |
998 		     PCS_MII_STATUS_REMOTE_FAULT)) ==
999 	    (PCS_MII_STATUS_AUTONEG_COMP | PCS_MII_STATUS_REMOTE_FAULT))
1000 		netif_info(cp, link, cp->dev, "PCS RemoteFault\n");
1001 
1002 	/* work around link detection issue by querying the PCS state
1003 	 * machine directly.
1004 	 */
1005 	state_machine = readl(cp->regs + REG_PCS_STATE_MACHINE);
1006 	if ((state_machine & PCS_SM_LINK_STATE_MASK) != SM_LINK_STATE_UP) {
1007 		stat &= ~PCS_MII_STATUS_LINK_STATUS;
1008 	} else if (state_machine & PCS_SM_WORD_SYNC_STATE_MASK) {
1009 		stat |= PCS_MII_STATUS_LINK_STATUS;
1010 	}
1011 
1012 	if (stat & PCS_MII_STATUS_LINK_STATUS) {
1013 		if (cp->lstate != link_up) {
1014 			if (cp->opened) {
1015 				cp->lstate = link_up;
1016 				cp->link_transition = LINK_TRANSITION_LINK_UP;
1017 
1018 				cas_set_link_modes(cp);
1019 				netif_carrier_on(cp->dev);
1020 			}
1021 		}
1022 	} else if (cp->lstate == link_up) {
1023 		cp->lstate = link_down;
1024 		if (link_transition_timeout != 0 &&
1025 		    cp->link_transition != LINK_TRANSITION_REQUESTED_RESET &&
1026 		    !cp->link_transition_jiffies_valid) {
1027 			/*
1028 			 * force a reset, as a workaround for the
1029 			 * link-failure problem. May want to move this to a
1030 			 * point a bit earlier in the sequence. If we had
1031 			 * generated a reset a short time ago, we'll wait for
1032 			 * the link timer to check the status until a
1033 			 * timer expires (link_transistion_jiffies_valid is
1034 			 * true when the timer is running.)  Instead of using
1035 			 * a system timer, we just do a check whenever the
1036 			 * link timer is running - this clears the flag after
1037 			 * a suitable delay.
1038 			 */
1039 			retval = 1;
1040 			cp->link_transition = LINK_TRANSITION_REQUESTED_RESET;
1041 			cp->link_transition_jiffies = jiffies;
1042 			cp->link_transition_jiffies_valid = 1;
1043 		} else {
1044 			cp->link_transition = LINK_TRANSITION_ON_FAILURE;
1045 		}
1046 		netif_carrier_off(cp->dev);
1047 		if (cp->opened)
1048 			netif_info(cp, link, cp->dev, "PCS link down\n");
1049 
1050 		/* Cassini only: if you force a mode, there can be
1051 		 * sync problems on link down. to fix that, the following
1052 		 * things need to be checked:
1053 		 * 1) read serialink state register
1054 		 * 2) read pcs status register to verify link down.
1055 		 * 3) if link down and serial link == 0x03, then you need
1056 		 *    to global reset the chip.
1057 		 */
1058 		if ((cp->cas_flags & CAS_FLAG_REG_PLUS) == 0) {
1059 			/* should check to see if we're in a forced mode */
1060 			stat = readl(cp->regs + REG_PCS_SERDES_STATE);
1061 			if (stat == 0x03)
1062 				return 1;
1063 		}
1064 	} else if (cp->lstate == link_down) {
1065 		if (link_transition_timeout != 0 &&
1066 		    cp->link_transition != LINK_TRANSITION_REQUESTED_RESET &&
1067 		    !cp->link_transition_jiffies_valid) {
1068 			/* force a reset, as a workaround for the
1069 			 * link-failure problem.  May want to move
1070 			 * this to a point a bit earlier in the
1071 			 * sequence.
1072 			 */
1073 			retval = 1;
1074 			cp->link_transition = LINK_TRANSITION_REQUESTED_RESET;
1075 			cp->link_transition_jiffies = jiffies;
1076 			cp->link_transition_jiffies_valid = 1;
1077 		} else {
1078 			cp->link_transition = LINK_TRANSITION_STILL_FAILED;
1079 		}
1080 	}
1081 
1082 	return retval;
1083 }
1084 
1085 static int cas_pcs_interrupt(struct net_device *dev,
1086 			     struct cas *cp, u32 status)
1087 {
1088 	u32 stat = readl(cp->regs + REG_PCS_INTR_STATUS);
1089 
1090 	if ((stat & PCS_INTR_STATUS_LINK_CHANGE) == 0)
1091 		return 0;
1092 	return cas_pcs_link_check(cp);
1093 }
1094 
1095 static int cas_txmac_interrupt(struct net_device *dev,
1096 			       struct cas *cp, u32 status)
1097 {
1098 	u32 txmac_stat = readl(cp->regs + REG_MAC_TX_STATUS);
1099 
1100 	if (!txmac_stat)
1101 		return 0;
1102 
1103 	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
1104 		     "txmac interrupt, txmac_stat: 0x%x\n", txmac_stat);
1105 
1106 	/* Defer timer expiration is quite normal,
1107 	 * don't even log the event.
1108 	 */
1109 	if ((txmac_stat & MAC_TX_DEFER_TIMER) &&
1110 	    !(txmac_stat & ~MAC_TX_DEFER_TIMER))
1111 		return 0;
1112 
1113 	spin_lock(&cp->stat_lock[0]);
1114 	if (txmac_stat & MAC_TX_UNDERRUN) {
1115 		netdev_err(dev, "TX MAC xmit underrun\n");
1116 		cp->net_stats[0].tx_fifo_errors++;
1117 	}
1118 
1119 	if (txmac_stat & MAC_TX_MAX_PACKET_ERR) {
1120 		netdev_err(dev, "TX MAC max packet size error\n");
1121 		cp->net_stats[0].tx_errors++;
1122 	}
1123 
1124 	/* The rest are all cases of one of the 16-bit TX
1125 	 * counters expiring.
1126 	 */
1127 	if (txmac_stat & MAC_TX_COLL_NORMAL)
1128 		cp->net_stats[0].collisions += 0x10000;
1129 
1130 	if (txmac_stat & MAC_TX_COLL_EXCESS) {
1131 		cp->net_stats[0].tx_aborted_errors += 0x10000;
1132 		cp->net_stats[0].collisions += 0x10000;
1133 	}
1134 
1135 	if (txmac_stat & MAC_TX_COLL_LATE) {
1136 		cp->net_stats[0].tx_aborted_errors += 0x10000;
1137 		cp->net_stats[0].collisions += 0x10000;
1138 	}
1139 	spin_unlock(&cp->stat_lock[0]);
1140 
1141 	/* We do not keep track of MAC_TX_COLL_FIRST and
1142 	 * MAC_TX_PEAK_ATTEMPTS events.
1143 	 */
1144 	return 0;
1145 }
1146 
1147 static void cas_load_firmware(struct cas *cp, cas_hp_inst_t *firmware)
1148 {
1149 	cas_hp_inst_t *inst;
1150 	u32 val;
1151 	int i;
1152 
1153 	i = 0;
1154 	while ((inst = firmware) && inst->note) {
1155 		writel(i, cp->regs + REG_HP_INSTR_RAM_ADDR);
1156 
1157 		val = CAS_BASE(HP_INSTR_RAM_HI_VAL, inst->val);
1158 		val |= CAS_BASE(HP_INSTR_RAM_HI_MASK, inst->mask);
1159 		writel(val, cp->regs + REG_HP_INSTR_RAM_DATA_HI);
1160 
1161 		val = CAS_BASE(HP_INSTR_RAM_MID_OUTARG, inst->outarg >> 10);
1162 		val |= CAS_BASE(HP_INSTR_RAM_MID_OUTOP, inst->outop);
1163 		val |= CAS_BASE(HP_INSTR_RAM_MID_FNEXT, inst->fnext);
1164 		val |= CAS_BASE(HP_INSTR_RAM_MID_FOFF, inst->foff);
1165 		val |= CAS_BASE(HP_INSTR_RAM_MID_SNEXT, inst->snext);
1166 		val |= CAS_BASE(HP_INSTR_RAM_MID_SOFF, inst->soff);
1167 		val |= CAS_BASE(HP_INSTR_RAM_MID_OP, inst->op);
1168 		writel(val, cp->regs + REG_HP_INSTR_RAM_DATA_MID);
1169 
1170 		val = CAS_BASE(HP_INSTR_RAM_LOW_OUTMASK, inst->outmask);
1171 		val |= CAS_BASE(HP_INSTR_RAM_LOW_OUTSHIFT, inst->outshift);
1172 		val |= CAS_BASE(HP_INSTR_RAM_LOW_OUTEN, inst->outenab);
1173 		val |= CAS_BASE(HP_INSTR_RAM_LOW_OUTARG, inst->outarg);
1174 		writel(val, cp->regs + REG_HP_INSTR_RAM_DATA_LOW);
1175 		++firmware;
1176 		++i;
1177 	}
1178 }
1179 
1180 static void cas_init_rx_dma(struct cas *cp)
1181 {
1182 	u64 desc_dma = cp->block_dvma;
1183 	u32 val;
1184 	int i, size;
1185 
1186 	/* rx free descriptors */
1187 	val = CAS_BASE(RX_CFG_SWIVEL, RX_SWIVEL_OFF_VAL);
1188 	val |= CAS_BASE(RX_CFG_DESC_RING, RX_DESC_RINGN_INDEX(0));
1189 	val |= CAS_BASE(RX_CFG_COMP_RING, RX_COMP_RINGN_INDEX(0));
1190 	if ((N_RX_DESC_RINGS > 1) &&
1191 	    (cp->cas_flags & CAS_FLAG_REG_PLUS))  /* do desc 2 */
1192 		val |= CAS_BASE(RX_CFG_DESC_RING1, RX_DESC_RINGN_INDEX(1));
1193 	writel(val, cp->regs + REG_RX_CFG);
1194 
1195 	val = (unsigned long) cp->init_rxds[0] -
1196 		(unsigned long) cp->init_block;
1197 	writel((desc_dma + val) >> 32, cp->regs + REG_RX_DB_HI);
1198 	writel((desc_dma + val) & 0xffffffff, cp->regs + REG_RX_DB_LOW);
1199 	writel(RX_DESC_RINGN_SIZE(0) - 4, cp->regs + REG_RX_KICK);
1200 
1201 	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
1202 		/* rx desc 2 is for IPSEC packets. however,
1203 		 * we don't it that for that purpose.
1204 		 */
1205 		val = (unsigned long) cp->init_rxds[1] -
1206 			(unsigned long) cp->init_block;
1207 		writel((desc_dma + val) >> 32, cp->regs + REG_PLUS_RX_DB1_HI);
1208 		writel((desc_dma + val) & 0xffffffff, cp->regs +
1209 		       REG_PLUS_RX_DB1_LOW);
1210 		writel(RX_DESC_RINGN_SIZE(1) - 4, cp->regs +
1211 		       REG_PLUS_RX_KICK1);
1212 	}
1213 
1214 	/* rx completion registers */
1215 	val = (unsigned long) cp->init_rxcs[0] -
1216 		(unsigned long) cp->init_block;
1217 	writel((desc_dma + val) >> 32, cp->regs + REG_RX_CB_HI);
1218 	writel((desc_dma + val) & 0xffffffff, cp->regs + REG_RX_CB_LOW);
1219 
1220 	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
1221 		/* rx comp 2-4 */
1222 		for (i = 1; i < MAX_RX_COMP_RINGS; i++) {
1223 			val = (unsigned long) cp->init_rxcs[i] -
1224 				(unsigned long) cp->init_block;
1225 			writel((desc_dma + val) >> 32, cp->regs +
1226 			       REG_PLUS_RX_CBN_HI(i));
1227 			writel((desc_dma + val) & 0xffffffff, cp->regs +
1228 			       REG_PLUS_RX_CBN_LOW(i));
1229 		}
1230 	}
1231 
1232 	/* read selective clear regs to prevent spurious interrupts
1233 	 * on reset because complete == kick.
1234 	 * selective clear set up to prevent interrupts on resets
1235 	 */
1236 	readl(cp->regs + REG_INTR_STATUS_ALIAS);
1237 	writel(INTR_RX_DONE | INTR_RX_BUF_UNAVAIL, cp->regs + REG_ALIAS_CLEAR);
1238 
1239 	/* set up pause thresholds */
1240 	val  = CAS_BASE(RX_PAUSE_THRESH_OFF,
1241 			cp->rx_pause_off / RX_PAUSE_THRESH_QUANTUM);
1242 	val |= CAS_BASE(RX_PAUSE_THRESH_ON,
1243 			cp->rx_pause_on / RX_PAUSE_THRESH_QUANTUM);
1244 	writel(val, cp->regs + REG_RX_PAUSE_THRESH);
1245 
1246 	/* zero out dma reassembly buffers */
1247 	for (i = 0; i < 64; i++) {
1248 		writel(i, cp->regs + REG_RX_TABLE_ADDR);
1249 		writel(0x0, cp->regs + REG_RX_TABLE_DATA_LOW);
1250 		writel(0x0, cp->regs + REG_RX_TABLE_DATA_MID);
1251 		writel(0x0, cp->regs + REG_RX_TABLE_DATA_HI);
1252 	}
1253 
1254 	/* make sure address register is 0 for normal operation */
1255 	writel(0x0, cp->regs + REG_RX_CTRL_FIFO_ADDR);
1256 	writel(0x0, cp->regs + REG_RX_IPP_FIFO_ADDR);
1257 
1258 	/* interrupt mitigation */
1259 #ifdef USE_RX_BLANK
1260 	val = CAS_BASE(RX_BLANK_INTR_TIME, RX_BLANK_INTR_TIME_VAL);
1261 	val |= CAS_BASE(RX_BLANK_INTR_PKT, RX_BLANK_INTR_PKT_VAL);
1262 	writel(val, cp->regs + REG_RX_BLANK);
1263 #else
1264 	writel(0x0, cp->regs + REG_RX_BLANK);
1265 #endif
1266 
1267 	/* interrupt generation as a function of low water marks for
1268 	 * free desc and completion entries. these are used to trigger
1269 	 * housekeeping for rx descs. we don't use the free interrupt
1270 	 * as it's not very useful
1271 	 */
1272 	/* val = CAS_BASE(RX_AE_THRESH_FREE, RX_AE_FREEN_VAL(0)); */
1273 	val = CAS_BASE(RX_AE_THRESH_COMP, RX_AE_COMP_VAL);
1274 	writel(val, cp->regs + REG_RX_AE_THRESH);
1275 	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
1276 		val = CAS_BASE(RX_AE1_THRESH_FREE, RX_AE_FREEN_VAL(1));
1277 		writel(val, cp->regs + REG_PLUS_RX_AE1_THRESH);
1278 	}
1279 
1280 	/* Random early detect registers. useful for congestion avoidance.
1281 	 * this should be tunable.
1282 	 */
1283 	writel(0x0, cp->regs + REG_RX_RED);
1284 
1285 	/* receive page sizes. default == 2K (0x800) */
1286 	val = 0;
1287 	if (cp->page_size == 0x1000)
1288 		val = 0x1;
1289 	else if (cp->page_size == 0x2000)
1290 		val = 0x2;
1291 	else if (cp->page_size == 0x4000)
1292 		val = 0x3;
1293 
1294 	/* round mtu + offset. constrain to page size. */
1295 	size = cp->dev->mtu + 64;
1296 	if (size > cp->page_size)
1297 		size = cp->page_size;
1298 
1299 	if (size <= 0x400)
1300 		i = 0x0;
1301 	else if (size <= 0x800)
1302 		i = 0x1;
1303 	else if (size <= 0x1000)
1304 		i = 0x2;
1305 	else
1306 		i = 0x3;
1307 
1308 	cp->mtu_stride = 1 << (i + 10);
1309 	val  = CAS_BASE(RX_PAGE_SIZE, val);
1310 	val |= CAS_BASE(RX_PAGE_SIZE_MTU_STRIDE, i);
1311 	val |= CAS_BASE(RX_PAGE_SIZE_MTU_COUNT, cp->page_size >> (i + 10));
1312 	val |= CAS_BASE(RX_PAGE_SIZE_MTU_OFF, 0x1);
1313 	writel(val, cp->regs + REG_RX_PAGE_SIZE);
1314 
1315 	/* enable the header parser if desired */
1316 	if (&CAS_HP_FIRMWARE[0] == &cas_prog_null[0])
1317 		return;
1318 
1319 	val = CAS_BASE(HP_CFG_NUM_CPU, CAS_NCPUS > 63 ? 0 : CAS_NCPUS);
1320 	val |= HP_CFG_PARSE_EN | HP_CFG_SYN_INC_MASK;
1321 	val |= CAS_BASE(HP_CFG_TCP_THRESH, HP_TCP_THRESH_VAL);
1322 	writel(val, cp->regs + REG_HP_CFG);
1323 }
1324 
1325 static inline void cas_rxc_init(struct cas_rx_comp *rxc)
1326 {
1327 	memset(rxc, 0, sizeof(*rxc));
1328 	rxc->word4 = cpu_to_le64(RX_COMP4_ZERO);
1329 }
1330 
1331 /* NOTE: we use the ENC RX DESC ring for spares. the rx_page[0,1]
1332  * flipping is protected by the fact that the chip will not
1333  * hand back the same page index while it's being processed.
1334  */
1335 static inline cas_page_t *cas_page_spare(struct cas *cp, const int index)
1336 {
1337 	cas_page_t *page = cp->rx_pages[1][index];
1338 	cas_page_t *new;
1339 
1340 	if (page_count(page->buffer) == 1)
1341 		return page;
1342 
1343 	new = cas_page_dequeue(cp);
1344 	if (new) {
1345 		spin_lock(&cp->rx_inuse_lock);
1346 		list_add(&page->list, &cp->rx_inuse_list);
1347 		spin_unlock(&cp->rx_inuse_lock);
1348 	}
1349 	return new;
1350 }
1351 
1352 /* this needs to be changed if we actually use the ENC RX DESC ring */
1353 static cas_page_t *cas_page_swap(struct cas *cp, const int ring,
1354 				 const int index)
1355 {
1356 	cas_page_t **page0 = cp->rx_pages[0];
1357 	cas_page_t **page1 = cp->rx_pages[1];
1358 
1359 	/* swap if buffer is in use */
1360 	if (page_count(page0[index]->buffer) > 1) {
1361 		cas_page_t *new = cas_page_spare(cp, index);
1362 		if (new) {
1363 			page1[index] = page0[index];
1364 			page0[index] = new;
1365 		}
1366 	}
1367 	RX_USED_SET(page0[index], 0);
1368 	return page0[index];
1369 }
1370 
1371 static void cas_clean_rxds(struct cas *cp)
1372 {
1373 	/* only clean ring 0 as ring 1 is used for spare buffers */
1374         struct cas_rx_desc *rxd = cp->init_rxds[0];
1375 	int i, size;
1376 
1377 	/* release all rx flows */
1378 	for (i = 0; i < N_RX_FLOWS; i++) {
1379 		struct sk_buff *skb;
1380 		while ((skb = __skb_dequeue(&cp->rx_flows[i]))) {
1381 			cas_skb_release(skb);
1382 		}
1383 	}
1384 
1385 	/* initialize descriptors */
1386 	size = RX_DESC_RINGN_SIZE(0);
1387 	for (i = 0; i < size; i++) {
1388 		cas_page_t *page = cas_page_swap(cp, 0, i);
1389 		rxd[i].buffer = cpu_to_le64(page->dma_addr);
1390 		rxd[i].index  = cpu_to_le64(CAS_BASE(RX_INDEX_NUM, i) |
1391 					    CAS_BASE(RX_INDEX_RING, 0));
1392 	}
1393 
1394 	cp->rx_old[0]  = RX_DESC_RINGN_SIZE(0) - 4;
1395 	cp->rx_last[0] = 0;
1396 	cp->cas_flags &= ~CAS_FLAG_RXD_POST(0);
1397 }
1398 
1399 static void cas_clean_rxcs(struct cas *cp)
1400 {
1401 	int i, j;
1402 
1403 	/* take ownership of rx comp descriptors */
1404 	memset(cp->rx_cur, 0, sizeof(*cp->rx_cur)*N_RX_COMP_RINGS);
1405 	memset(cp->rx_new, 0, sizeof(*cp->rx_new)*N_RX_COMP_RINGS);
1406 	for (i = 0; i < N_RX_COMP_RINGS; i++) {
1407 		struct cas_rx_comp *rxc = cp->init_rxcs[i];
1408 		for (j = 0; j < RX_COMP_RINGN_SIZE(i); j++) {
1409 			cas_rxc_init(rxc + j);
1410 		}
1411 	}
1412 }
1413 
1414 #if 0
1415 /* When we get a RX fifo overflow, the RX unit is probably hung
1416  * so we do the following.
1417  *
1418  * If any part of the reset goes wrong, we return 1 and that causes the
1419  * whole chip to be reset.
1420  */
1421 static int cas_rxmac_reset(struct cas *cp)
1422 {
1423 	struct net_device *dev = cp->dev;
1424 	int limit;
1425 	u32 val;
1426 
1427 	/* First, reset MAC RX. */
1428 	writel(cp->mac_rx_cfg & ~MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
1429 	for (limit = 0; limit < STOP_TRIES; limit++) {
1430 		if (!(readl(cp->regs + REG_MAC_RX_CFG) & MAC_RX_CFG_EN))
1431 			break;
1432 		udelay(10);
1433 	}
1434 	if (limit == STOP_TRIES) {
1435 		netdev_err(dev, "RX MAC will not disable, resetting whole chip\n");
1436 		return 1;
1437 	}
1438 
1439 	/* Second, disable RX DMA. */
1440 	writel(0, cp->regs + REG_RX_CFG);
1441 	for (limit = 0; limit < STOP_TRIES; limit++) {
1442 		if (!(readl(cp->regs + REG_RX_CFG) & RX_CFG_DMA_EN))
1443 			break;
1444 		udelay(10);
1445 	}
1446 	if (limit == STOP_TRIES) {
1447 		netdev_err(dev, "RX DMA will not disable, resetting whole chip\n");
1448 		return 1;
1449 	}
1450 
1451 	mdelay(5);
1452 
1453 	/* Execute RX reset command. */
1454 	writel(SW_RESET_RX, cp->regs + REG_SW_RESET);
1455 	for (limit = 0; limit < STOP_TRIES; limit++) {
1456 		if (!(readl(cp->regs + REG_SW_RESET) & SW_RESET_RX))
1457 			break;
1458 		udelay(10);
1459 	}
1460 	if (limit == STOP_TRIES) {
1461 		netdev_err(dev, "RX reset command will not execute, resetting whole chip\n");
1462 		return 1;
1463 	}
1464 
1465 	/* reset driver rx state */
1466 	cas_clean_rxds(cp);
1467 	cas_clean_rxcs(cp);
1468 
1469 	/* Now, reprogram the rest of RX unit. */
1470 	cas_init_rx_dma(cp);
1471 
1472 	/* re-enable */
1473 	val = readl(cp->regs + REG_RX_CFG);
1474 	writel(val | RX_CFG_DMA_EN, cp->regs + REG_RX_CFG);
1475 	writel(MAC_RX_FRAME_RECV, cp->regs + REG_MAC_RX_MASK);
1476 	val = readl(cp->regs + REG_MAC_RX_CFG);
1477 	writel(val | MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
1478 	return 0;
1479 }
1480 #endif
1481 
1482 static int cas_rxmac_interrupt(struct net_device *dev, struct cas *cp,
1483 			       u32 status)
1484 {
1485 	u32 stat = readl(cp->regs + REG_MAC_RX_STATUS);
1486 
1487 	if (!stat)
1488 		return 0;
1489 
1490 	netif_dbg(cp, intr, cp->dev, "rxmac interrupt, stat: 0x%x\n", stat);
1491 
1492 	/* these are all rollovers */
1493 	spin_lock(&cp->stat_lock[0]);
1494 	if (stat & MAC_RX_ALIGN_ERR)
1495 		cp->net_stats[0].rx_frame_errors += 0x10000;
1496 
1497 	if (stat & MAC_RX_CRC_ERR)
1498 		cp->net_stats[0].rx_crc_errors += 0x10000;
1499 
1500 	if (stat & MAC_RX_LEN_ERR)
1501 		cp->net_stats[0].rx_length_errors += 0x10000;
1502 
1503 	if (stat & MAC_RX_OVERFLOW) {
1504 		cp->net_stats[0].rx_over_errors++;
1505 		cp->net_stats[0].rx_fifo_errors++;
1506 	}
1507 
1508 	/* We do not track MAC_RX_FRAME_COUNT and MAC_RX_VIOL_ERR
1509 	 * events.
1510 	 */
1511 	spin_unlock(&cp->stat_lock[0]);
1512 	return 0;
1513 }
1514 
1515 static int cas_mac_interrupt(struct net_device *dev, struct cas *cp,
1516 			     u32 status)
1517 {
1518 	u32 stat = readl(cp->regs + REG_MAC_CTRL_STATUS);
1519 
1520 	if (!stat)
1521 		return 0;
1522 
1523 	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
1524 		     "mac interrupt, stat: 0x%x\n", stat);
1525 
1526 	/* This interrupt is just for pause frame and pause
1527 	 * tracking.  It is useful for diagnostics and debug
1528 	 * but probably by default we will mask these events.
1529 	 */
1530 	if (stat & MAC_CTRL_PAUSE_STATE)
1531 		cp->pause_entered++;
1532 
1533 	if (stat & MAC_CTRL_PAUSE_RECEIVED)
1534 		cp->pause_last_time_recvd = (stat >> 16);
1535 
1536 	return 0;
1537 }
1538 
1539 
1540 /* Must be invoked under cp->lock. */
1541 static inline int cas_mdio_link_not_up(struct cas *cp)
1542 {
1543 	u16 val;
1544 
1545 	switch (cp->lstate) {
1546 	case link_force_ret:
1547 		netif_info(cp, link, cp->dev, "Autoneg failed again, keeping forced mode\n");
1548 		cas_phy_write(cp, MII_BMCR, cp->link_fcntl);
1549 		cp->timer_ticks = 5;
1550 		cp->lstate = link_force_ok;
1551 		cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
1552 		break;
1553 
1554 	case link_aneg:
1555 		val = cas_phy_read(cp, MII_BMCR);
1556 
1557 		/* Try forced modes. we try things in the following order:
1558 		 * 1000 full -> 100 full/half -> 10 half
1559 		 */
1560 		val &= ~(BMCR_ANRESTART | BMCR_ANENABLE);
1561 		val |= BMCR_FULLDPLX;
1562 		val |= (cp->cas_flags & CAS_FLAG_1000MB_CAP) ?
1563 			CAS_BMCR_SPEED1000 : BMCR_SPEED100;
1564 		cas_phy_write(cp, MII_BMCR, val);
1565 		cp->timer_ticks = 5;
1566 		cp->lstate = link_force_try;
1567 		cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
1568 		break;
1569 
1570 	case link_force_try:
1571 		/* Downgrade from 1000 to 100 to 10 Mbps if necessary. */
1572 		val = cas_phy_read(cp, MII_BMCR);
1573 		cp->timer_ticks = 5;
1574 		if (val & CAS_BMCR_SPEED1000) { /* gigabit */
1575 			val &= ~CAS_BMCR_SPEED1000;
1576 			val |= (BMCR_SPEED100 | BMCR_FULLDPLX);
1577 			cas_phy_write(cp, MII_BMCR, val);
1578 			break;
1579 		}
1580 
1581 		if (val & BMCR_SPEED100) {
1582 			if (val & BMCR_FULLDPLX) /* fd failed */
1583 				val &= ~BMCR_FULLDPLX;
1584 			else { /* 100Mbps failed */
1585 				val &= ~BMCR_SPEED100;
1586 			}
1587 			cas_phy_write(cp, MII_BMCR, val);
1588 			break;
1589 		}
1590 		break;
1591 	default:
1592 		break;
1593 	}
1594 	return 0;
1595 }
1596 
1597 
1598 /* must be invoked with cp->lock held */
1599 static int cas_mii_link_check(struct cas *cp, const u16 bmsr)
1600 {
1601 	int restart;
1602 
1603 	if (bmsr & BMSR_LSTATUS) {
1604 		/* Ok, here we got a link. If we had it due to a forced
1605 		 * fallback, and we were configured for autoneg, we
1606 		 * retry a short autoneg pass. If you know your hub is
1607 		 * broken, use ethtool ;)
1608 		 */
1609 		if ((cp->lstate == link_force_try) &&
1610 		    (cp->link_cntl & BMCR_ANENABLE)) {
1611 			cp->lstate = link_force_ret;
1612 			cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
1613 			cas_mif_poll(cp, 0);
1614 			cp->link_fcntl = cas_phy_read(cp, MII_BMCR);
1615 			cp->timer_ticks = 5;
1616 			if (cp->opened)
1617 				netif_info(cp, link, cp->dev,
1618 					   "Got link after fallback, retrying autoneg once...\n");
1619 			cas_phy_write(cp, MII_BMCR,
1620 				      cp->link_fcntl | BMCR_ANENABLE |
1621 				      BMCR_ANRESTART);
1622 			cas_mif_poll(cp, 1);
1623 
1624 		} else if (cp->lstate != link_up) {
1625 			cp->lstate = link_up;
1626 			cp->link_transition = LINK_TRANSITION_LINK_UP;
1627 
1628 			if (cp->opened) {
1629 				cas_set_link_modes(cp);
1630 				netif_carrier_on(cp->dev);
1631 			}
1632 		}
1633 		return 0;
1634 	}
1635 
1636 	/* link not up. if the link was previously up, we restart the
1637 	 * whole process
1638 	 */
1639 	restart = 0;
1640 	if (cp->lstate == link_up) {
1641 		cp->lstate = link_down;
1642 		cp->link_transition = LINK_TRANSITION_LINK_DOWN;
1643 
1644 		netif_carrier_off(cp->dev);
1645 		if (cp->opened)
1646 			netif_info(cp, link, cp->dev, "Link down\n");
1647 		restart = 1;
1648 
1649 	} else if (++cp->timer_ticks > 10)
1650 		cas_mdio_link_not_up(cp);
1651 
1652 	return restart;
1653 }
1654 
1655 static int cas_mif_interrupt(struct net_device *dev, struct cas *cp,
1656 			     u32 status)
1657 {
1658 	u32 stat = readl(cp->regs + REG_MIF_STATUS);
1659 	u16 bmsr;
1660 
1661 	/* check for a link change */
1662 	if (CAS_VAL(MIF_STATUS_POLL_STATUS, stat) == 0)
1663 		return 0;
1664 
1665 	bmsr = CAS_VAL(MIF_STATUS_POLL_DATA, stat);
1666 	return cas_mii_link_check(cp, bmsr);
1667 }
1668 
1669 static int cas_pci_interrupt(struct net_device *dev, struct cas *cp,
1670 			     u32 status)
1671 {
1672 	u32 stat = readl(cp->regs + REG_PCI_ERR_STATUS);
1673 
1674 	if (!stat)
1675 		return 0;
1676 
1677 	netdev_err(dev, "PCI error [%04x:%04x]",
1678 		   stat, readl(cp->regs + REG_BIM_DIAG));
1679 
1680 	/* cassini+ has this reserved */
1681 	if ((stat & PCI_ERR_BADACK) &&
1682 	    ((cp->cas_flags & CAS_FLAG_REG_PLUS) == 0))
1683 		pr_cont(" <No ACK64# during ABS64 cycle>");
1684 
1685 	if (stat & PCI_ERR_DTRTO)
1686 		pr_cont(" <Delayed transaction timeout>");
1687 	if (stat & PCI_ERR_OTHER)
1688 		pr_cont(" <other>");
1689 	if (stat & PCI_ERR_BIM_DMA_WRITE)
1690 		pr_cont(" <BIM DMA 0 write req>");
1691 	if (stat & PCI_ERR_BIM_DMA_READ)
1692 		pr_cont(" <BIM DMA 0 read req>");
1693 	pr_cont("\n");
1694 
1695 	if (stat & PCI_ERR_OTHER) {
1696 		int pci_errs;
1697 
1698 		/* Interrogate PCI config space for the
1699 		 * true cause.
1700 		 */
1701 		pci_errs = pci_status_get_and_clear_errors(cp->pdev);
1702 
1703 		netdev_err(dev, "PCI status errors[%04x]\n", pci_errs);
1704 		if (pci_errs & PCI_STATUS_PARITY)
1705 			netdev_err(dev, "PCI parity error detected\n");
1706 		if (pci_errs & PCI_STATUS_SIG_TARGET_ABORT)
1707 			netdev_err(dev, "PCI target abort\n");
1708 		if (pci_errs & PCI_STATUS_REC_TARGET_ABORT)
1709 			netdev_err(dev, "PCI master acks target abort\n");
1710 		if (pci_errs & PCI_STATUS_REC_MASTER_ABORT)
1711 			netdev_err(dev, "PCI master abort\n");
1712 		if (pci_errs & PCI_STATUS_SIG_SYSTEM_ERROR)
1713 			netdev_err(dev, "PCI system error SERR#\n");
1714 		if (pci_errs & PCI_STATUS_DETECTED_PARITY)
1715 			netdev_err(dev, "PCI parity error\n");
1716 	}
1717 
1718 	/* For all PCI errors, we should reset the chip. */
1719 	return 1;
1720 }
1721 
1722 /* All non-normal interrupt conditions get serviced here.
1723  * Returns non-zero if we should just exit the interrupt
1724  * handler right now (ie. if we reset the card which invalidates
1725  * all of the other original irq status bits).
1726  */
1727 static int cas_abnormal_irq(struct net_device *dev, struct cas *cp,
1728 			    u32 status)
1729 {
1730 	if (status & INTR_RX_TAG_ERROR) {
1731 		/* corrupt RX tag framing */
1732 		netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
1733 			     "corrupt rx tag framing\n");
1734 		spin_lock(&cp->stat_lock[0]);
1735 		cp->net_stats[0].rx_errors++;
1736 		spin_unlock(&cp->stat_lock[0]);
1737 		goto do_reset;
1738 	}
1739 
1740 	if (status & INTR_RX_LEN_MISMATCH) {
1741 		/* length mismatch. */
1742 		netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
1743 			     "length mismatch for rx frame\n");
1744 		spin_lock(&cp->stat_lock[0]);
1745 		cp->net_stats[0].rx_errors++;
1746 		spin_unlock(&cp->stat_lock[0]);
1747 		goto do_reset;
1748 	}
1749 
1750 	if (status & INTR_PCS_STATUS) {
1751 		if (cas_pcs_interrupt(dev, cp, status))
1752 			goto do_reset;
1753 	}
1754 
1755 	if (status & INTR_TX_MAC_STATUS) {
1756 		if (cas_txmac_interrupt(dev, cp, status))
1757 			goto do_reset;
1758 	}
1759 
1760 	if (status & INTR_RX_MAC_STATUS) {
1761 		if (cas_rxmac_interrupt(dev, cp, status))
1762 			goto do_reset;
1763 	}
1764 
1765 	if (status & INTR_MAC_CTRL_STATUS) {
1766 		if (cas_mac_interrupt(dev, cp, status))
1767 			goto do_reset;
1768 	}
1769 
1770 	if (status & INTR_MIF_STATUS) {
1771 		if (cas_mif_interrupt(dev, cp, status))
1772 			goto do_reset;
1773 	}
1774 
1775 	if (status & INTR_PCI_ERROR_STATUS) {
1776 		if (cas_pci_interrupt(dev, cp, status))
1777 			goto do_reset;
1778 	}
1779 	return 0;
1780 
1781 do_reset:
1782 #if 1
1783 	atomic_inc(&cp->reset_task_pending);
1784 	atomic_inc(&cp->reset_task_pending_all);
1785 	netdev_err(dev, "reset called in cas_abnormal_irq [0x%x]\n", status);
1786 	schedule_work(&cp->reset_task);
1787 #else
1788 	atomic_set(&cp->reset_task_pending, CAS_RESET_ALL);
1789 	netdev_err(dev, "reset called in cas_abnormal_irq\n");
1790 	schedule_work(&cp->reset_task);
1791 #endif
1792 	return 1;
1793 }
1794 
1795 /* NOTE: CAS_TABORT returns 1 or 2 so that it can be used when
1796  *       determining whether to do a netif_stop/wakeup
1797  */
1798 #define CAS_TABORT(x)      (((x)->cas_flags & CAS_FLAG_TARGET_ABORT) ? 2 : 1)
1799 #define CAS_ROUND_PAGE(x)  (((x) + PAGE_SIZE - 1) & PAGE_MASK)
1800 static inline int cas_calc_tabort(struct cas *cp, const unsigned long addr,
1801 				  const int len)
1802 {
1803 	unsigned long off = addr + len;
1804 
1805 	if (CAS_TABORT(cp) == 1)
1806 		return 0;
1807 	if ((CAS_ROUND_PAGE(off) - off) > TX_TARGET_ABORT_LEN)
1808 		return 0;
1809 	return TX_TARGET_ABORT_LEN;
1810 }
1811 
1812 static inline void cas_tx_ringN(struct cas *cp, int ring, int limit)
1813 {
1814 	struct cas_tx_desc *txds;
1815 	struct sk_buff **skbs;
1816 	struct net_device *dev = cp->dev;
1817 	int entry, count;
1818 
1819 	spin_lock(&cp->tx_lock[ring]);
1820 	txds = cp->init_txds[ring];
1821 	skbs = cp->tx_skbs[ring];
1822 	entry = cp->tx_old[ring];
1823 
1824 	count = TX_BUFF_COUNT(ring, entry, limit);
1825 	while (entry != limit) {
1826 		struct sk_buff *skb = skbs[entry];
1827 		dma_addr_t daddr;
1828 		u32 dlen;
1829 		int frag;
1830 
1831 		if (!skb) {
1832 			/* this should never occur */
1833 			entry = TX_DESC_NEXT(ring, entry);
1834 			continue;
1835 		}
1836 
1837 		/* however, we might get only a partial skb release. */
1838 		count -= skb_shinfo(skb)->nr_frags +
1839 			+ cp->tx_tiny_use[ring][entry].nbufs + 1;
1840 		if (count < 0)
1841 			break;
1842 
1843 		netif_printk(cp, tx_done, KERN_DEBUG, cp->dev,
1844 			     "tx[%d] done, slot %d\n", ring, entry);
1845 
1846 		skbs[entry] = NULL;
1847 		cp->tx_tiny_use[ring][entry].nbufs = 0;
1848 
1849 		for (frag = 0; frag <= skb_shinfo(skb)->nr_frags; frag++) {
1850 			struct cas_tx_desc *txd = txds + entry;
1851 
1852 			daddr = le64_to_cpu(txd->buffer);
1853 			dlen = CAS_VAL(TX_DESC_BUFLEN,
1854 				       le64_to_cpu(txd->control));
1855 			dma_unmap_page(&cp->pdev->dev, daddr, dlen,
1856 				       DMA_TO_DEVICE);
1857 			entry = TX_DESC_NEXT(ring, entry);
1858 
1859 			/* tiny buffer may follow */
1860 			if (cp->tx_tiny_use[ring][entry].used) {
1861 				cp->tx_tiny_use[ring][entry].used = 0;
1862 				entry = TX_DESC_NEXT(ring, entry);
1863 			}
1864 		}
1865 
1866 		spin_lock(&cp->stat_lock[ring]);
1867 		cp->net_stats[ring].tx_packets++;
1868 		cp->net_stats[ring].tx_bytes += skb->len;
1869 		spin_unlock(&cp->stat_lock[ring]);
1870 		dev_consume_skb_irq(skb);
1871 	}
1872 	cp->tx_old[ring] = entry;
1873 
1874 	/* this is wrong for multiple tx rings. the net device needs
1875 	 * multiple queues for this to do the right thing.  we wait
1876 	 * for 2*packets to be available when using tiny buffers
1877 	 */
1878 	if (netif_queue_stopped(dev) &&
1879 	    (TX_BUFFS_AVAIL(cp, ring) > CAS_TABORT(cp)*(MAX_SKB_FRAGS + 1)))
1880 		netif_wake_queue(dev);
1881 	spin_unlock(&cp->tx_lock[ring]);
1882 }
1883 
1884 static void cas_tx(struct net_device *dev, struct cas *cp,
1885 		   u32 status)
1886 {
1887         int limit, ring;
1888 #ifdef USE_TX_COMPWB
1889 	u64 compwb = le64_to_cpu(cp->init_block->tx_compwb);
1890 #endif
1891 	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
1892 		     "tx interrupt, status: 0x%x, %llx\n",
1893 		     status, (unsigned long long)compwb);
1894 	/* process all the rings */
1895 	for (ring = 0; ring < N_TX_RINGS; ring++) {
1896 #ifdef USE_TX_COMPWB
1897 		/* use the completion writeback registers */
1898 		limit = (CAS_VAL(TX_COMPWB_MSB, compwb) << 8) |
1899 			CAS_VAL(TX_COMPWB_LSB, compwb);
1900 		compwb = TX_COMPWB_NEXT(compwb);
1901 #else
1902 		limit = readl(cp->regs + REG_TX_COMPN(ring));
1903 #endif
1904 		if (cp->tx_old[ring] != limit)
1905 			cas_tx_ringN(cp, ring, limit);
1906 	}
1907 }
1908 
1909 
1910 static int cas_rx_process_pkt(struct cas *cp, struct cas_rx_comp *rxc,
1911 			      int entry, const u64 *words,
1912 			      struct sk_buff **skbref)
1913 {
1914 	int dlen, hlen, len, i, alloclen;
1915 	int off, swivel = RX_SWIVEL_OFF_VAL;
1916 	struct cas_page *page;
1917 	struct sk_buff *skb;
1918 	void *addr, *crcaddr;
1919 	__sum16 csum;
1920 	char *p;
1921 
1922 	hlen = CAS_VAL(RX_COMP2_HDR_SIZE, words[1]);
1923 	dlen = CAS_VAL(RX_COMP1_DATA_SIZE, words[0]);
1924 	len  = hlen + dlen;
1925 
1926 	if (RX_COPY_ALWAYS || (words[2] & RX_COMP3_SMALL_PKT))
1927 		alloclen = len;
1928 	else
1929 		alloclen = max(hlen, RX_COPY_MIN);
1930 
1931 	skb = netdev_alloc_skb(cp->dev, alloclen + swivel + cp->crc_size);
1932 	if (skb == NULL)
1933 		return -1;
1934 
1935 	*skbref = skb;
1936 	skb_reserve(skb, swivel);
1937 
1938 	p = skb->data;
1939 	addr = crcaddr = NULL;
1940 	if (hlen) { /* always copy header pages */
1941 		i = CAS_VAL(RX_COMP2_HDR_INDEX, words[1]);
1942 		page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
1943 		off = CAS_VAL(RX_COMP2_HDR_OFF, words[1]) * 0x100 +
1944 			swivel;
1945 
1946 		i = hlen;
1947 		if (!dlen) /* attach FCS */
1948 			i += cp->crc_size;
1949 		dma_sync_single_for_cpu(&cp->pdev->dev, page->dma_addr + off,
1950 					i, DMA_FROM_DEVICE);
1951 		addr = cas_page_map(page->buffer);
1952 		memcpy(p, addr + off, i);
1953 		dma_sync_single_for_device(&cp->pdev->dev,
1954 					   page->dma_addr + off, i,
1955 					   DMA_FROM_DEVICE);
1956 		cas_page_unmap(addr);
1957 		RX_USED_ADD(page, 0x100);
1958 		p += hlen;
1959 		swivel = 0;
1960 	}
1961 
1962 
1963 	if (alloclen < (hlen + dlen)) {
1964 		skb_frag_t *frag = skb_shinfo(skb)->frags;
1965 
1966 		/* normal or jumbo packets. we use frags */
1967 		i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
1968 		page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
1969 		off = CAS_VAL(RX_COMP1_DATA_OFF, words[0]) + swivel;
1970 
1971 		hlen = min(cp->page_size - off, dlen);
1972 		if (hlen < 0) {
1973 			netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
1974 				     "rx page overflow: %d\n", hlen);
1975 			dev_kfree_skb_irq(skb);
1976 			return -1;
1977 		}
1978 		i = hlen;
1979 		if (i == dlen)  /* attach FCS */
1980 			i += cp->crc_size;
1981 		dma_sync_single_for_cpu(&cp->pdev->dev, page->dma_addr + off,
1982 					i, DMA_FROM_DEVICE);
1983 
1984 		/* make sure we always copy a header */
1985 		swivel = 0;
1986 		if (p == (char *) skb->data) { /* not split */
1987 			addr = cas_page_map(page->buffer);
1988 			memcpy(p, addr + off, RX_COPY_MIN);
1989 			dma_sync_single_for_device(&cp->pdev->dev,
1990 						   page->dma_addr + off, i,
1991 						   DMA_FROM_DEVICE);
1992 			cas_page_unmap(addr);
1993 			off += RX_COPY_MIN;
1994 			swivel = RX_COPY_MIN;
1995 			RX_USED_ADD(page, cp->mtu_stride);
1996 		} else {
1997 			RX_USED_ADD(page, hlen);
1998 		}
1999 		skb_put(skb, alloclen);
2000 
2001 		skb_shinfo(skb)->nr_frags++;
2002 		skb->data_len += hlen - swivel;
2003 		skb->truesize += hlen - swivel;
2004 		skb->len      += hlen - swivel;
2005 
2006 		__skb_frag_set_page(frag, page->buffer);
2007 		__skb_frag_ref(frag);
2008 		skb_frag_off_set(frag, off);
2009 		skb_frag_size_set(frag, hlen - swivel);
2010 
2011 		/* any more data? */
2012 		if ((words[0] & RX_COMP1_SPLIT_PKT) && ((dlen -= hlen) > 0)) {
2013 			hlen = dlen;
2014 			off = 0;
2015 
2016 			i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
2017 			page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
2018 			dma_sync_single_for_cpu(&cp->pdev->dev,
2019 						page->dma_addr,
2020 						hlen + cp->crc_size,
2021 						DMA_FROM_DEVICE);
2022 			dma_sync_single_for_device(&cp->pdev->dev,
2023 						   page->dma_addr,
2024 						   hlen + cp->crc_size,
2025 						   DMA_FROM_DEVICE);
2026 
2027 			skb_shinfo(skb)->nr_frags++;
2028 			skb->data_len += hlen;
2029 			skb->len      += hlen;
2030 			frag++;
2031 
2032 			__skb_frag_set_page(frag, page->buffer);
2033 			__skb_frag_ref(frag);
2034 			skb_frag_off_set(frag, 0);
2035 			skb_frag_size_set(frag, hlen);
2036 			RX_USED_ADD(page, hlen + cp->crc_size);
2037 		}
2038 
2039 		if (cp->crc_size) {
2040 			addr = cas_page_map(page->buffer);
2041 			crcaddr  = addr + off + hlen;
2042 		}
2043 
2044 	} else {
2045 		/* copying packet */
2046 		if (!dlen)
2047 			goto end_copy_pkt;
2048 
2049 		i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
2050 		page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
2051 		off = CAS_VAL(RX_COMP1_DATA_OFF, words[0]) + swivel;
2052 		hlen = min(cp->page_size - off, dlen);
2053 		if (hlen < 0) {
2054 			netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
2055 				     "rx page overflow: %d\n", hlen);
2056 			dev_kfree_skb_irq(skb);
2057 			return -1;
2058 		}
2059 		i = hlen;
2060 		if (i == dlen) /* attach FCS */
2061 			i += cp->crc_size;
2062 		dma_sync_single_for_cpu(&cp->pdev->dev, page->dma_addr + off,
2063 					i, DMA_FROM_DEVICE);
2064 		addr = cas_page_map(page->buffer);
2065 		memcpy(p, addr + off, i);
2066 		dma_sync_single_for_device(&cp->pdev->dev,
2067 					   page->dma_addr + off, i,
2068 					   DMA_FROM_DEVICE);
2069 		cas_page_unmap(addr);
2070 		if (p == (char *) skb->data) /* not split */
2071 			RX_USED_ADD(page, cp->mtu_stride);
2072 		else
2073 			RX_USED_ADD(page, i);
2074 
2075 		/* any more data? */
2076 		if ((words[0] & RX_COMP1_SPLIT_PKT) && ((dlen -= hlen) > 0)) {
2077 			p += hlen;
2078 			i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
2079 			page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
2080 			dma_sync_single_for_cpu(&cp->pdev->dev,
2081 						page->dma_addr,
2082 						dlen + cp->crc_size,
2083 						DMA_FROM_DEVICE);
2084 			addr = cas_page_map(page->buffer);
2085 			memcpy(p, addr, dlen + cp->crc_size);
2086 			dma_sync_single_for_device(&cp->pdev->dev,
2087 						   page->dma_addr,
2088 						   dlen + cp->crc_size,
2089 						   DMA_FROM_DEVICE);
2090 			cas_page_unmap(addr);
2091 			RX_USED_ADD(page, dlen + cp->crc_size);
2092 		}
2093 end_copy_pkt:
2094 		if (cp->crc_size) {
2095 			addr    = NULL;
2096 			crcaddr = skb->data + alloclen;
2097 		}
2098 		skb_put(skb, alloclen);
2099 	}
2100 
2101 	csum = (__force __sum16)htons(CAS_VAL(RX_COMP4_TCP_CSUM, words[3]));
2102 	if (cp->crc_size) {
2103 		/* checksum includes FCS. strip it out. */
2104 		csum = csum_fold(csum_partial(crcaddr, cp->crc_size,
2105 					      csum_unfold(csum)));
2106 		if (addr)
2107 			cas_page_unmap(addr);
2108 	}
2109 	skb->protocol = eth_type_trans(skb, cp->dev);
2110 	if (skb->protocol == htons(ETH_P_IP)) {
2111 		skb->csum = csum_unfold(~csum);
2112 		skb->ip_summed = CHECKSUM_COMPLETE;
2113 	} else
2114 		skb_checksum_none_assert(skb);
2115 	return len;
2116 }
2117 
2118 
2119 /* we can handle up to 64 rx flows at a time. we do the same thing
2120  * as nonreassm except that we batch up the buffers.
2121  * NOTE: we currently just treat each flow as a bunch of packets that
2122  *       we pass up. a better way would be to coalesce the packets
2123  *       into a jumbo packet. to do that, we need to do the following:
2124  *       1) the first packet will have a clean split between header and
2125  *          data. save both.
2126  *       2) each time the next flow packet comes in, extend the
2127  *          data length and merge the checksums.
2128  *       3) on flow release, fix up the header.
2129  *       4) make sure the higher layer doesn't care.
2130  * because packets get coalesced, we shouldn't run into fragment count
2131  * issues.
2132  */
2133 static inline void cas_rx_flow_pkt(struct cas *cp, const u64 *words,
2134 				   struct sk_buff *skb)
2135 {
2136 	int flowid = CAS_VAL(RX_COMP3_FLOWID, words[2]) & (N_RX_FLOWS - 1);
2137 	struct sk_buff_head *flow = &cp->rx_flows[flowid];
2138 
2139 	/* this is protected at a higher layer, so no need to
2140 	 * do any additional locking here. stick the buffer
2141 	 * at the end.
2142 	 */
2143 	__skb_queue_tail(flow, skb);
2144 	if (words[0] & RX_COMP1_RELEASE_FLOW) {
2145 		while ((skb = __skb_dequeue(flow))) {
2146 			cas_skb_release(skb);
2147 		}
2148 	}
2149 }
2150 
2151 /* put rx descriptor back on ring. if a buffer is in use by a higher
2152  * layer, this will need to put in a replacement.
2153  */
2154 static void cas_post_page(struct cas *cp, const int ring, const int index)
2155 {
2156 	cas_page_t *new;
2157 	int entry;
2158 
2159 	entry = cp->rx_old[ring];
2160 
2161 	new = cas_page_swap(cp, ring, index);
2162 	cp->init_rxds[ring][entry].buffer = cpu_to_le64(new->dma_addr);
2163 	cp->init_rxds[ring][entry].index  =
2164 		cpu_to_le64(CAS_BASE(RX_INDEX_NUM, index) |
2165 			    CAS_BASE(RX_INDEX_RING, ring));
2166 
2167 	entry = RX_DESC_ENTRY(ring, entry + 1);
2168 	cp->rx_old[ring] = entry;
2169 
2170 	if (entry % 4)
2171 		return;
2172 
2173 	if (ring == 0)
2174 		writel(entry, cp->regs + REG_RX_KICK);
2175 	else if ((N_RX_DESC_RINGS > 1) &&
2176 		 (cp->cas_flags & CAS_FLAG_REG_PLUS))
2177 		writel(entry, cp->regs + REG_PLUS_RX_KICK1);
2178 }
2179 
2180 
2181 /* only when things are bad */
2182 static int cas_post_rxds_ringN(struct cas *cp, int ring, int num)
2183 {
2184 	unsigned int entry, last, count, released;
2185 	int cluster;
2186 	cas_page_t **page = cp->rx_pages[ring];
2187 
2188 	entry = cp->rx_old[ring];
2189 
2190 	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
2191 		     "rxd[%d] interrupt, done: %d\n", ring, entry);
2192 
2193 	cluster = -1;
2194 	count = entry & 0x3;
2195 	last = RX_DESC_ENTRY(ring, num ? entry + num - 4: entry - 4);
2196 	released = 0;
2197 	while (entry != last) {
2198 		/* make a new buffer if it's still in use */
2199 		if (page_count(page[entry]->buffer) > 1) {
2200 			cas_page_t *new = cas_page_dequeue(cp);
2201 			if (!new) {
2202 				/* let the timer know that we need to
2203 				 * do this again
2204 				 */
2205 				cp->cas_flags |= CAS_FLAG_RXD_POST(ring);
2206 				if (!timer_pending(&cp->link_timer))
2207 					mod_timer(&cp->link_timer, jiffies +
2208 						  CAS_LINK_FAST_TIMEOUT);
2209 				cp->rx_old[ring]  = entry;
2210 				cp->rx_last[ring] = num ? num - released : 0;
2211 				return -ENOMEM;
2212 			}
2213 			spin_lock(&cp->rx_inuse_lock);
2214 			list_add(&page[entry]->list, &cp->rx_inuse_list);
2215 			spin_unlock(&cp->rx_inuse_lock);
2216 			cp->init_rxds[ring][entry].buffer =
2217 				cpu_to_le64(new->dma_addr);
2218 			page[entry] = new;
2219 
2220 		}
2221 
2222 		if (++count == 4) {
2223 			cluster = entry;
2224 			count = 0;
2225 		}
2226 		released++;
2227 		entry = RX_DESC_ENTRY(ring, entry + 1);
2228 	}
2229 	cp->rx_old[ring] = entry;
2230 
2231 	if (cluster < 0)
2232 		return 0;
2233 
2234 	if (ring == 0)
2235 		writel(cluster, cp->regs + REG_RX_KICK);
2236 	else if ((N_RX_DESC_RINGS > 1) &&
2237 		 (cp->cas_flags & CAS_FLAG_REG_PLUS))
2238 		writel(cluster, cp->regs + REG_PLUS_RX_KICK1);
2239 	return 0;
2240 }
2241 
2242 
2243 /* process a completion ring. packets are set up in three basic ways:
2244  * small packets: should be copied header + data in single buffer.
2245  * large packets: header and data in a single buffer.
2246  * split packets: header in a separate buffer from data.
2247  *                data may be in multiple pages. data may be > 256
2248  *                bytes but in a single page.
2249  *
2250  * NOTE: RX page posting is done in this routine as well. while there's
2251  *       the capability of using multiple RX completion rings, it isn't
2252  *       really worthwhile due to the fact that the page posting will
2253  *       force serialization on the single descriptor ring.
2254  */
2255 static int cas_rx_ringN(struct cas *cp, int ring, int budget)
2256 {
2257 	struct cas_rx_comp *rxcs = cp->init_rxcs[ring];
2258 	int entry, drops;
2259 	int npackets = 0;
2260 
2261 	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
2262 		     "rx[%d] interrupt, done: %d/%d\n",
2263 		     ring,
2264 		     readl(cp->regs + REG_RX_COMP_HEAD), cp->rx_new[ring]);
2265 
2266 	entry = cp->rx_new[ring];
2267 	drops = 0;
2268 	while (1) {
2269 		struct cas_rx_comp *rxc = rxcs + entry;
2270 		struct sk_buff *skb;
2271 		int type, len;
2272 		u64 words[4];
2273 		int i, dring;
2274 
2275 		words[0] = le64_to_cpu(rxc->word1);
2276 		words[1] = le64_to_cpu(rxc->word2);
2277 		words[2] = le64_to_cpu(rxc->word3);
2278 		words[3] = le64_to_cpu(rxc->word4);
2279 
2280 		/* don't touch if still owned by hw */
2281 		type = CAS_VAL(RX_COMP1_TYPE, words[0]);
2282 		if (type == 0)
2283 			break;
2284 
2285 		/* hw hasn't cleared the zero bit yet */
2286 		if (words[3] & RX_COMP4_ZERO) {
2287 			break;
2288 		}
2289 
2290 		/* get info on the packet */
2291 		if (words[3] & (RX_COMP4_LEN_MISMATCH | RX_COMP4_BAD)) {
2292 			spin_lock(&cp->stat_lock[ring]);
2293 			cp->net_stats[ring].rx_errors++;
2294 			if (words[3] & RX_COMP4_LEN_MISMATCH)
2295 				cp->net_stats[ring].rx_length_errors++;
2296 			if (words[3] & RX_COMP4_BAD)
2297 				cp->net_stats[ring].rx_crc_errors++;
2298 			spin_unlock(&cp->stat_lock[ring]);
2299 
2300 			/* We'll just return it to Cassini. */
2301 		drop_it:
2302 			spin_lock(&cp->stat_lock[ring]);
2303 			++cp->net_stats[ring].rx_dropped;
2304 			spin_unlock(&cp->stat_lock[ring]);
2305 			goto next;
2306 		}
2307 
2308 		len = cas_rx_process_pkt(cp, rxc, entry, words, &skb);
2309 		if (len < 0) {
2310 			++drops;
2311 			goto drop_it;
2312 		}
2313 
2314 		/* see if it's a flow re-assembly or not. the driver
2315 		 * itself handles release back up.
2316 		 */
2317 		if (RX_DONT_BATCH || (type == 0x2)) {
2318 			/* non-reassm: these always get released */
2319 			cas_skb_release(skb);
2320 		} else {
2321 			cas_rx_flow_pkt(cp, words, skb);
2322 		}
2323 
2324 		spin_lock(&cp->stat_lock[ring]);
2325 		cp->net_stats[ring].rx_packets++;
2326 		cp->net_stats[ring].rx_bytes += len;
2327 		spin_unlock(&cp->stat_lock[ring]);
2328 
2329 	next:
2330 		npackets++;
2331 
2332 		/* should it be released? */
2333 		if (words[0] & RX_COMP1_RELEASE_HDR) {
2334 			i = CAS_VAL(RX_COMP2_HDR_INDEX, words[1]);
2335 			dring = CAS_VAL(RX_INDEX_RING, i);
2336 			i = CAS_VAL(RX_INDEX_NUM, i);
2337 			cas_post_page(cp, dring, i);
2338 		}
2339 
2340 		if (words[0] & RX_COMP1_RELEASE_DATA) {
2341 			i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
2342 			dring = CAS_VAL(RX_INDEX_RING, i);
2343 			i = CAS_VAL(RX_INDEX_NUM, i);
2344 			cas_post_page(cp, dring, i);
2345 		}
2346 
2347 		if (words[0] & RX_COMP1_RELEASE_NEXT) {
2348 			i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
2349 			dring = CAS_VAL(RX_INDEX_RING, i);
2350 			i = CAS_VAL(RX_INDEX_NUM, i);
2351 			cas_post_page(cp, dring, i);
2352 		}
2353 
2354 		/* skip to the next entry */
2355 		entry = RX_COMP_ENTRY(ring, entry + 1 +
2356 				      CAS_VAL(RX_COMP1_SKIP, words[0]));
2357 #ifdef USE_NAPI
2358 		if (budget && (npackets >= budget))
2359 			break;
2360 #endif
2361 	}
2362 	cp->rx_new[ring] = entry;
2363 
2364 	if (drops)
2365 		netdev_info(cp->dev, "Memory squeeze, deferring packet\n");
2366 	return npackets;
2367 }
2368 
2369 
2370 /* put completion entries back on the ring */
2371 static void cas_post_rxcs_ringN(struct net_device *dev,
2372 				struct cas *cp, int ring)
2373 {
2374 	struct cas_rx_comp *rxc = cp->init_rxcs[ring];
2375 	int last, entry;
2376 
2377 	last = cp->rx_cur[ring];
2378 	entry = cp->rx_new[ring];
2379 	netif_printk(cp, intr, KERN_DEBUG, dev,
2380 		     "rxc[%d] interrupt, done: %d/%d\n",
2381 		     ring, readl(cp->regs + REG_RX_COMP_HEAD), entry);
2382 
2383 	/* zero and re-mark descriptors */
2384 	while (last != entry) {
2385 		cas_rxc_init(rxc + last);
2386 		last = RX_COMP_ENTRY(ring, last + 1);
2387 	}
2388 	cp->rx_cur[ring] = last;
2389 
2390 	if (ring == 0)
2391 		writel(last, cp->regs + REG_RX_COMP_TAIL);
2392 	else if (cp->cas_flags & CAS_FLAG_REG_PLUS)
2393 		writel(last, cp->regs + REG_PLUS_RX_COMPN_TAIL(ring));
2394 }
2395 
2396 
2397 
2398 /* cassini can use all four PCI interrupts for the completion ring.
2399  * rings 3 and 4 are identical
2400  */
2401 #if defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
2402 static inline void cas_handle_irqN(struct net_device *dev,
2403 				   struct cas *cp, const u32 status,
2404 				   const int ring)
2405 {
2406 	if (status & (INTR_RX_COMP_FULL_ALT | INTR_RX_COMP_AF_ALT))
2407 		cas_post_rxcs_ringN(dev, cp, ring);
2408 }
2409 
2410 static irqreturn_t cas_interruptN(int irq, void *dev_id)
2411 {
2412 	struct net_device *dev = dev_id;
2413 	struct cas *cp = netdev_priv(dev);
2414 	unsigned long flags;
2415 	int ring = (irq == cp->pci_irq_INTC) ? 2 : 3;
2416 	u32 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(ring));
2417 
2418 	/* check for shared irq */
2419 	if (status == 0)
2420 		return IRQ_NONE;
2421 
2422 	spin_lock_irqsave(&cp->lock, flags);
2423 	if (status & INTR_RX_DONE_ALT) { /* handle rx separately */
2424 #ifdef USE_NAPI
2425 		cas_mask_intr(cp);
2426 		napi_schedule(&cp->napi);
2427 #else
2428 		cas_rx_ringN(cp, ring, 0);
2429 #endif
2430 		status &= ~INTR_RX_DONE_ALT;
2431 	}
2432 
2433 	if (status)
2434 		cas_handle_irqN(dev, cp, status, ring);
2435 	spin_unlock_irqrestore(&cp->lock, flags);
2436 	return IRQ_HANDLED;
2437 }
2438 #endif
2439 
2440 #ifdef USE_PCI_INTB
2441 /* everything but rx packets */
2442 static inline void cas_handle_irq1(struct cas *cp, const u32 status)
2443 {
2444 	if (status & INTR_RX_BUF_UNAVAIL_1) {
2445 		/* Frame arrived, no free RX buffers available.
2446 		 * NOTE: we can get this on a link transition. */
2447 		cas_post_rxds_ringN(cp, 1, 0);
2448 		spin_lock(&cp->stat_lock[1]);
2449 		cp->net_stats[1].rx_dropped++;
2450 		spin_unlock(&cp->stat_lock[1]);
2451 	}
2452 
2453 	if (status & INTR_RX_BUF_AE_1)
2454 		cas_post_rxds_ringN(cp, 1, RX_DESC_RINGN_SIZE(1) -
2455 				    RX_AE_FREEN_VAL(1));
2456 
2457 	if (status & (INTR_RX_COMP_AF | INTR_RX_COMP_FULL))
2458 		cas_post_rxcs_ringN(cp, 1);
2459 }
2460 
2461 /* ring 2 handles a few more events than 3 and 4 */
2462 static irqreturn_t cas_interrupt1(int irq, void *dev_id)
2463 {
2464 	struct net_device *dev = dev_id;
2465 	struct cas *cp = netdev_priv(dev);
2466 	unsigned long flags;
2467 	u32 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(1));
2468 
2469 	/* check for shared interrupt */
2470 	if (status == 0)
2471 		return IRQ_NONE;
2472 
2473 	spin_lock_irqsave(&cp->lock, flags);
2474 	if (status & INTR_RX_DONE_ALT) { /* handle rx separately */
2475 #ifdef USE_NAPI
2476 		cas_mask_intr(cp);
2477 		napi_schedule(&cp->napi);
2478 #else
2479 		cas_rx_ringN(cp, 1, 0);
2480 #endif
2481 		status &= ~INTR_RX_DONE_ALT;
2482 	}
2483 	if (status)
2484 		cas_handle_irq1(cp, status);
2485 	spin_unlock_irqrestore(&cp->lock, flags);
2486 	return IRQ_HANDLED;
2487 }
2488 #endif
2489 
2490 static inline void cas_handle_irq(struct net_device *dev,
2491 				  struct cas *cp, const u32 status)
2492 {
2493 	/* housekeeping interrupts */
2494 	if (status & INTR_ERROR_MASK)
2495 		cas_abnormal_irq(dev, cp, status);
2496 
2497 	if (status & INTR_RX_BUF_UNAVAIL) {
2498 		/* Frame arrived, no free RX buffers available.
2499 		 * NOTE: we can get this on a link transition.
2500 		 */
2501 		cas_post_rxds_ringN(cp, 0, 0);
2502 		spin_lock(&cp->stat_lock[0]);
2503 		cp->net_stats[0].rx_dropped++;
2504 		spin_unlock(&cp->stat_lock[0]);
2505 	} else if (status & INTR_RX_BUF_AE) {
2506 		cas_post_rxds_ringN(cp, 0, RX_DESC_RINGN_SIZE(0) -
2507 				    RX_AE_FREEN_VAL(0));
2508 	}
2509 
2510 	if (status & (INTR_RX_COMP_AF | INTR_RX_COMP_FULL))
2511 		cas_post_rxcs_ringN(dev, cp, 0);
2512 }
2513 
2514 static irqreturn_t cas_interrupt(int irq, void *dev_id)
2515 {
2516 	struct net_device *dev = dev_id;
2517 	struct cas *cp = netdev_priv(dev);
2518 	unsigned long flags;
2519 	u32 status = readl(cp->regs + REG_INTR_STATUS);
2520 
2521 	if (status == 0)
2522 		return IRQ_NONE;
2523 
2524 	spin_lock_irqsave(&cp->lock, flags);
2525 	if (status & (INTR_TX_ALL | INTR_TX_INTME)) {
2526 		cas_tx(dev, cp, status);
2527 		status &= ~(INTR_TX_ALL | INTR_TX_INTME);
2528 	}
2529 
2530 	if (status & INTR_RX_DONE) {
2531 #ifdef USE_NAPI
2532 		cas_mask_intr(cp);
2533 		napi_schedule(&cp->napi);
2534 #else
2535 		cas_rx_ringN(cp, 0, 0);
2536 #endif
2537 		status &= ~INTR_RX_DONE;
2538 	}
2539 
2540 	if (status)
2541 		cas_handle_irq(dev, cp, status);
2542 	spin_unlock_irqrestore(&cp->lock, flags);
2543 	return IRQ_HANDLED;
2544 }
2545 
2546 
2547 #ifdef USE_NAPI
2548 static int cas_poll(struct napi_struct *napi, int budget)
2549 {
2550 	struct cas *cp = container_of(napi, struct cas, napi);
2551 	struct net_device *dev = cp->dev;
2552 	int i, enable_intr, credits;
2553 	u32 status = readl(cp->regs + REG_INTR_STATUS);
2554 	unsigned long flags;
2555 
2556 	spin_lock_irqsave(&cp->lock, flags);
2557 	cas_tx(dev, cp, status);
2558 	spin_unlock_irqrestore(&cp->lock, flags);
2559 
2560 	/* NAPI rx packets. we spread the credits across all of the
2561 	 * rxc rings
2562 	 *
2563 	 * to make sure we're fair with the work we loop through each
2564 	 * ring N_RX_COMP_RING times with a request of
2565 	 * budget / N_RX_COMP_RINGS
2566 	 */
2567 	enable_intr = 1;
2568 	credits = 0;
2569 	for (i = 0; i < N_RX_COMP_RINGS; i++) {
2570 		int j;
2571 		for (j = 0; j < N_RX_COMP_RINGS; j++) {
2572 			credits += cas_rx_ringN(cp, j, budget / N_RX_COMP_RINGS);
2573 			if (credits >= budget) {
2574 				enable_intr = 0;
2575 				goto rx_comp;
2576 			}
2577 		}
2578 	}
2579 
2580 rx_comp:
2581 	/* final rx completion */
2582 	spin_lock_irqsave(&cp->lock, flags);
2583 	if (status)
2584 		cas_handle_irq(dev, cp, status);
2585 
2586 #ifdef USE_PCI_INTB
2587 	if (N_RX_COMP_RINGS > 1) {
2588 		status = readl(cp->regs + REG_PLUS_INTRN_STATUS(1));
2589 		if (status)
2590 			cas_handle_irq1(dev, cp, status);
2591 	}
2592 #endif
2593 
2594 #ifdef USE_PCI_INTC
2595 	if (N_RX_COMP_RINGS > 2) {
2596 		status = readl(cp->regs + REG_PLUS_INTRN_STATUS(2));
2597 		if (status)
2598 			cas_handle_irqN(dev, cp, status, 2);
2599 	}
2600 #endif
2601 
2602 #ifdef USE_PCI_INTD
2603 	if (N_RX_COMP_RINGS > 3) {
2604 		status = readl(cp->regs + REG_PLUS_INTRN_STATUS(3));
2605 		if (status)
2606 			cas_handle_irqN(dev, cp, status, 3);
2607 	}
2608 #endif
2609 	spin_unlock_irqrestore(&cp->lock, flags);
2610 	if (enable_intr) {
2611 		napi_complete(napi);
2612 		cas_unmask_intr(cp);
2613 	}
2614 	return credits;
2615 }
2616 #endif
2617 
2618 #ifdef CONFIG_NET_POLL_CONTROLLER
2619 static void cas_netpoll(struct net_device *dev)
2620 {
2621 	struct cas *cp = netdev_priv(dev);
2622 
2623 	cas_disable_irq(cp, 0);
2624 	cas_interrupt(cp->pdev->irq, dev);
2625 	cas_enable_irq(cp, 0);
2626 
2627 #ifdef USE_PCI_INTB
2628 	if (N_RX_COMP_RINGS > 1) {
2629 		/* cas_interrupt1(); */
2630 	}
2631 #endif
2632 #ifdef USE_PCI_INTC
2633 	if (N_RX_COMP_RINGS > 2) {
2634 		/* cas_interruptN(); */
2635 	}
2636 #endif
2637 #ifdef USE_PCI_INTD
2638 	if (N_RX_COMP_RINGS > 3) {
2639 		/* cas_interruptN(); */
2640 	}
2641 #endif
2642 }
2643 #endif
2644 
2645 static void cas_tx_timeout(struct net_device *dev, unsigned int txqueue)
2646 {
2647 	struct cas *cp = netdev_priv(dev);
2648 
2649 	netdev_err(dev, "transmit timed out, resetting\n");
2650 	if (!cp->hw_running) {
2651 		netdev_err(dev, "hrm.. hw not running!\n");
2652 		return;
2653 	}
2654 
2655 	netdev_err(dev, "MIF_STATE[%08x]\n",
2656 		   readl(cp->regs + REG_MIF_STATE_MACHINE));
2657 
2658 	netdev_err(dev, "MAC_STATE[%08x]\n",
2659 		   readl(cp->regs + REG_MAC_STATE_MACHINE));
2660 
2661 	netdev_err(dev, "TX_STATE[%08x:%08x:%08x] FIFO[%08x:%08x:%08x] SM1[%08x] SM2[%08x]\n",
2662 		   readl(cp->regs + REG_TX_CFG),
2663 		   readl(cp->regs + REG_MAC_TX_STATUS),
2664 		   readl(cp->regs + REG_MAC_TX_CFG),
2665 		   readl(cp->regs + REG_TX_FIFO_PKT_CNT),
2666 		   readl(cp->regs + REG_TX_FIFO_WRITE_PTR),
2667 		   readl(cp->regs + REG_TX_FIFO_READ_PTR),
2668 		   readl(cp->regs + REG_TX_SM_1),
2669 		   readl(cp->regs + REG_TX_SM_2));
2670 
2671 	netdev_err(dev, "RX_STATE[%08x:%08x:%08x]\n",
2672 		   readl(cp->regs + REG_RX_CFG),
2673 		   readl(cp->regs + REG_MAC_RX_STATUS),
2674 		   readl(cp->regs + REG_MAC_RX_CFG));
2675 
2676 	netdev_err(dev, "HP_STATE[%08x:%08x:%08x:%08x]\n",
2677 		   readl(cp->regs + REG_HP_STATE_MACHINE),
2678 		   readl(cp->regs + REG_HP_STATUS0),
2679 		   readl(cp->regs + REG_HP_STATUS1),
2680 		   readl(cp->regs + REG_HP_STATUS2));
2681 
2682 #if 1
2683 	atomic_inc(&cp->reset_task_pending);
2684 	atomic_inc(&cp->reset_task_pending_all);
2685 	schedule_work(&cp->reset_task);
2686 #else
2687 	atomic_set(&cp->reset_task_pending, CAS_RESET_ALL);
2688 	schedule_work(&cp->reset_task);
2689 #endif
2690 }
2691 
2692 static inline int cas_intme(int ring, int entry)
2693 {
2694 	/* Algorithm: IRQ every 1/2 of descriptors. */
2695 	if (!(entry & ((TX_DESC_RINGN_SIZE(ring) >> 1) - 1)))
2696 		return 1;
2697 	return 0;
2698 }
2699 
2700 
2701 static void cas_write_txd(struct cas *cp, int ring, int entry,
2702 			  dma_addr_t mapping, int len, u64 ctrl, int last)
2703 {
2704 	struct cas_tx_desc *txd = cp->init_txds[ring] + entry;
2705 
2706 	ctrl |= CAS_BASE(TX_DESC_BUFLEN, len);
2707 	if (cas_intme(ring, entry))
2708 		ctrl |= TX_DESC_INTME;
2709 	if (last)
2710 		ctrl |= TX_DESC_EOF;
2711 	txd->control = cpu_to_le64(ctrl);
2712 	txd->buffer = cpu_to_le64(mapping);
2713 }
2714 
2715 static inline void *tx_tiny_buf(struct cas *cp, const int ring,
2716 				const int entry)
2717 {
2718 	return cp->tx_tiny_bufs[ring] + TX_TINY_BUF_LEN*entry;
2719 }
2720 
2721 static inline dma_addr_t tx_tiny_map(struct cas *cp, const int ring,
2722 				     const int entry, const int tentry)
2723 {
2724 	cp->tx_tiny_use[ring][tentry].nbufs++;
2725 	cp->tx_tiny_use[ring][entry].used = 1;
2726 	return cp->tx_tiny_dvma[ring] + TX_TINY_BUF_LEN*entry;
2727 }
2728 
2729 static inline int cas_xmit_tx_ringN(struct cas *cp, int ring,
2730 				    struct sk_buff *skb)
2731 {
2732 	struct net_device *dev = cp->dev;
2733 	int entry, nr_frags, frag, tabort, tentry;
2734 	dma_addr_t mapping;
2735 	unsigned long flags;
2736 	u64 ctrl;
2737 	u32 len;
2738 
2739 	spin_lock_irqsave(&cp->tx_lock[ring], flags);
2740 
2741 	/* This is a hard error, log it. */
2742 	if (TX_BUFFS_AVAIL(cp, ring) <=
2743 	    CAS_TABORT(cp)*(skb_shinfo(skb)->nr_frags + 1)) {
2744 		netif_stop_queue(dev);
2745 		spin_unlock_irqrestore(&cp->tx_lock[ring], flags);
2746 		netdev_err(dev, "BUG! Tx Ring full when queue awake!\n");
2747 		return 1;
2748 	}
2749 
2750 	ctrl = 0;
2751 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
2752 		const u64 csum_start_off = skb_checksum_start_offset(skb);
2753 		const u64 csum_stuff_off = csum_start_off + skb->csum_offset;
2754 
2755 		ctrl =  TX_DESC_CSUM_EN |
2756 			CAS_BASE(TX_DESC_CSUM_START, csum_start_off) |
2757 			CAS_BASE(TX_DESC_CSUM_STUFF, csum_stuff_off);
2758 	}
2759 
2760 	entry = cp->tx_new[ring];
2761 	cp->tx_skbs[ring][entry] = skb;
2762 
2763 	nr_frags = skb_shinfo(skb)->nr_frags;
2764 	len = skb_headlen(skb);
2765 	mapping = dma_map_page(&cp->pdev->dev, virt_to_page(skb->data),
2766 			       offset_in_page(skb->data), len, DMA_TO_DEVICE);
2767 
2768 	tentry = entry;
2769 	tabort = cas_calc_tabort(cp, (unsigned long) skb->data, len);
2770 	if (unlikely(tabort)) {
2771 		/* NOTE: len is always >  tabort */
2772 		cas_write_txd(cp, ring, entry, mapping, len - tabort,
2773 			      ctrl | TX_DESC_SOF, 0);
2774 		entry = TX_DESC_NEXT(ring, entry);
2775 
2776 		skb_copy_from_linear_data_offset(skb, len - tabort,
2777 			      tx_tiny_buf(cp, ring, entry), tabort);
2778 		mapping = tx_tiny_map(cp, ring, entry, tentry);
2779 		cas_write_txd(cp, ring, entry, mapping, tabort, ctrl,
2780 			      (nr_frags == 0));
2781 	} else {
2782 		cas_write_txd(cp, ring, entry, mapping, len, ctrl |
2783 			      TX_DESC_SOF, (nr_frags == 0));
2784 	}
2785 	entry = TX_DESC_NEXT(ring, entry);
2786 
2787 	for (frag = 0; frag < nr_frags; frag++) {
2788 		const skb_frag_t *fragp = &skb_shinfo(skb)->frags[frag];
2789 
2790 		len = skb_frag_size(fragp);
2791 		mapping = skb_frag_dma_map(&cp->pdev->dev, fragp, 0, len,
2792 					   DMA_TO_DEVICE);
2793 
2794 		tabort = cas_calc_tabort(cp, skb_frag_off(fragp), len);
2795 		if (unlikely(tabort)) {
2796 			void *addr;
2797 
2798 			/* NOTE: len is always > tabort */
2799 			cas_write_txd(cp, ring, entry, mapping, len - tabort,
2800 				      ctrl, 0);
2801 			entry = TX_DESC_NEXT(ring, entry);
2802 
2803 			addr = cas_page_map(skb_frag_page(fragp));
2804 			memcpy(tx_tiny_buf(cp, ring, entry),
2805 			       addr + skb_frag_off(fragp) + len - tabort,
2806 			       tabort);
2807 			cas_page_unmap(addr);
2808 			mapping = tx_tiny_map(cp, ring, entry, tentry);
2809 			len     = tabort;
2810 		}
2811 
2812 		cas_write_txd(cp, ring, entry, mapping, len, ctrl,
2813 			      (frag + 1 == nr_frags));
2814 		entry = TX_DESC_NEXT(ring, entry);
2815 	}
2816 
2817 	cp->tx_new[ring] = entry;
2818 	if (TX_BUFFS_AVAIL(cp, ring) <= CAS_TABORT(cp)*(MAX_SKB_FRAGS + 1))
2819 		netif_stop_queue(dev);
2820 
2821 	netif_printk(cp, tx_queued, KERN_DEBUG, dev,
2822 		     "tx[%d] queued, slot %d, skblen %d, avail %d\n",
2823 		     ring, entry, skb->len, TX_BUFFS_AVAIL(cp, ring));
2824 	writel(entry, cp->regs + REG_TX_KICKN(ring));
2825 	spin_unlock_irqrestore(&cp->tx_lock[ring], flags);
2826 	return 0;
2827 }
2828 
2829 static netdev_tx_t cas_start_xmit(struct sk_buff *skb, struct net_device *dev)
2830 {
2831 	struct cas *cp = netdev_priv(dev);
2832 
2833 	/* this is only used as a load-balancing hint, so it doesn't
2834 	 * need to be SMP safe
2835 	 */
2836 	static int ring;
2837 
2838 	if (skb_padto(skb, cp->min_frame_size))
2839 		return NETDEV_TX_OK;
2840 
2841 	/* XXX: we need some higher-level QoS hooks to steer packets to
2842 	 *      individual queues.
2843 	 */
2844 	if (cas_xmit_tx_ringN(cp, ring++ & N_TX_RINGS_MASK, skb))
2845 		return NETDEV_TX_BUSY;
2846 	return NETDEV_TX_OK;
2847 }
2848 
2849 static void cas_init_tx_dma(struct cas *cp)
2850 {
2851 	u64 desc_dma = cp->block_dvma;
2852 	unsigned long off;
2853 	u32 val;
2854 	int i;
2855 
2856 	/* set up tx completion writeback registers. must be 8-byte aligned */
2857 #ifdef USE_TX_COMPWB
2858 	off = offsetof(struct cas_init_block, tx_compwb);
2859 	writel((desc_dma + off) >> 32, cp->regs + REG_TX_COMPWB_DB_HI);
2860 	writel((desc_dma + off) & 0xffffffff, cp->regs + REG_TX_COMPWB_DB_LOW);
2861 #endif
2862 
2863 	/* enable completion writebacks, enable paced mode,
2864 	 * disable read pipe, and disable pre-interrupt compwbs
2865 	 */
2866 	val =   TX_CFG_COMPWB_Q1 | TX_CFG_COMPWB_Q2 |
2867 		TX_CFG_COMPWB_Q3 | TX_CFG_COMPWB_Q4 |
2868 		TX_CFG_DMA_RDPIPE_DIS | TX_CFG_PACED_MODE |
2869 		TX_CFG_INTR_COMPWB_DIS;
2870 
2871 	/* write out tx ring info and tx desc bases */
2872 	for (i = 0; i < MAX_TX_RINGS; i++) {
2873 		off = (unsigned long) cp->init_txds[i] -
2874 			(unsigned long) cp->init_block;
2875 
2876 		val |= CAS_TX_RINGN_BASE(i);
2877 		writel((desc_dma + off) >> 32, cp->regs + REG_TX_DBN_HI(i));
2878 		writel((desc_dma + off) & 0xffffffff, cp->regs +
2879 		       REG_TX_DBN_LOW(i));
2880 		/* don't zero out the kick register here as the system
2881 		 * will wedge
2882 		 */
2883 	}
2884 	writel(val, cp->regs + REG_TX_CFG);
2885 
2886 	/* program max burst sizes. these numbers should be different
2887 	 * if doing QoS.
2888 	 */
2889 #ifdef USE_QOS
2890 	writel(0x800, cp->regs + REG_TX_MAXBURST_0);
2891 	writel(0x1600, cp->regs + REG_TX_MAXBURST_1);
2892 	writel(0x2400, cp->regs + REG_TX_MAXBURST_2);
2893 	writel(0x4800, cp->regs + REG_TX_MAXBURST_3);
2894 #else
2895 	writel(0x800, cp->regs + REG_TX_MAXBURST_0);
2896 	writel(0x800, cp->regs + REG_TX_MAXBURST_1);
2897 	writel(0x800, cp->regs + REG_TX_MAXBURST_2);
2898 	writel(0x800, cp->regs + REG_TX_MAXBURST_3);
2899 #endif
2900 }
2901 
2902 /* Must be invoked under cp->lock. */
2903 static inline void cas_init_dma(struct cas *cp)
2904 {
2905 	cas_init_tx_dma(cp);
2906 	cas_init_rx_dma(cp);
2907 }
2908 
2909 static void cas_process_mc_list(struct cas *cp)
2910 {
2911 	u16 hash_table[16];
2912 	u32 crc;
2913 	struct netdev_hw_addr *ha;
2914 	int i = 1;
2915 
2916 	memset(hash_table, 0, sizeof(hash_table));
2917 	netdev_for_each_mc_addr(ha, cp->dev) {
2918 		if (i <= CAS_MC_EXACT_MATCH_SIZE) {
2919 			/* use the alternate mac address registers for the
2920 			 * first 15 multicast addresses
2921 			 */
2922 			writel((ha->addr[4] << 8) | ha->addr[5],
2923 			       cp->regs + REG_MAC_ADDRN(i*3 + 0));
2924 			writel((ha->addr[2] << 8) | ha->addr[3],
2925 			       cp->regs + REG_MAC_ADDRN(i*3 + 1));
2926 			writel((ha->addr[0] << 8) | ha->addr[1],
2927 			       cp->regs + REG_MAC_ADDRN(i*3 + 2));
2928 			i++;
2929 		}
2930 		else {
2931 			/* use hw hash table for the next series of
2932 			 * multicast addresses
2933 			 */
2934 			crc = ether_crc_le(ETH_ALEN, ha->addr);
2935 			crc >>= 24;
2936 			hash_table[crc >> 4] |= 1 << (15 - (crc & 0xf));
2937 		}
2938 	}
2939 	for (i = 0; i < 16; i++)
2940 		writel(hash_table[i], cp->regs + REG_MAC_HASH_TABLEN(i));
2941 }
2942 
2943 /* Must be invoked under cp->lock. */
2944 static u32 cas_setup_multicast(struct cas *cp)
2945 {
2946 	u32 rxcfg = 0;
2947 	int i;
2948 
2949 	if (cp->dev->flags & IFF_PROMISC) {
2950 		rxcfg |= MAC_RX_CFG_PROMISC_EN;
2951 
2952 	} else if (cp->dev->flags & IFF_ALLMULTI) {
2953 	    	for (i=0; i < 16; i++)
2954 			writel(0xFFFF, cp->regs + REG_MAC_HASH_TABLEN(i));
2955 		rxcfg |= MAC_RX_CFG_HASH_FILTER_EN;
2956 
2957 	} else {
2958 		cas_process_mc_list(cp);
2959 		rxcfg |= MAC_RX_CFG_HASH_FILTER_EN;
2960 	}
2961 
2962 	return rxcfg;
2963 }
2964 
2965 /* must be invoked under cp->stat_lock[N_TX_RINGS] */
2966 static void cas_clear_mac_err(struct cas *cp)
2967 {
2968 	writel(0, cp->regs + REG_MAC_COLL_NORMAL);
2969 	writel(0, cp->regs + REG_MAC_COLL_FIRST);
2970 	writel(0, cp->regs + REG_MAC_COLL_EXCESS);
2971 	writel(0, cp->regs + REG_MAC_COLL_LATE);
2972 	writel(0, cp->regs + REG_MAC_TIMER_DEFER);
2973 	writel(0, cp->regs + REG_MAC_ATTEMPTS_PEAK);
2974 	writel(0, cp->regs + REG_MAC_RECV_FRAME);
2975 	writel(0, cp->regs + REG_MAC_LEN_ERR);
2976 	writel(0, cp->regs + REG_MAC_ALIGN_ERR);
2977 	writel(0, cp->regs + REG_MAC_FCS_ERR);
2978 	writel(0, cp->regs + REG_MAC_RX_CODE_ERR);
2979 }
2980 
2981 
2982 static void cas_mac_reset(struct cas *cp)
2983 {
2984 	int i;
2985 
2986 	/* do both TX and RX reset */
2987 	writel(0x1, cp->regs + REG_MAC_TX_RESET);
2988 	writel(0x1, cp->regs + REG_MAC_RX_RESET);
2989 
2990 	/* wait for TX */
2991 	i = STOP_TRIES;
2992 	while (i-- > 0) {
2993 		if (readl(cp->regs + REG_MAC_TX_RESET) == 0)
2994 			break;
2995 		udelay(10);
2996 	}
2997 
2998 	/* wait for RX */
2999 	i = STOP_TRIES;
3000 	while (i-- > 0) {
3001 		if (readl(cp->regs + REG_MAC_RX_RESET) == 0)
3002 			break;
3003 		udelay(10);
3004 	}
3005 
3006 	if (readl(cp->regs + REG_MAC_TX_RESET) |
3007 	    readl(cp->regs + REG_MAC_RX_RESET))
3008 		netdev_err(cp->dev, "mac tx[%d]/rx[%d] reset failed [%08x]\n",
3009 			   readl(cp->regs + REG_MAC_TX_RESET),
3010 			   readl(cp->regs + REG_MAC_RX_RESET),
3011 			   readl(cp->regs + REG_MAC_STATE_MACHINE));
3012 }
3013 
3014 
3015 /* Must be invoked under cp->lock. */
3016 static void cas_init_mac(struct cas *cp)
3017 {
3018 	const unsigned char *e = &cp->dev->dev_addr[0];
3019 	int i;
3020 	cas_mac_reset(cp);
3021 
3022 	/* setup core arbitration weight register */
3023 	writel(CAWR_RR_DIS, cp->regs + REG_CAWR);
3024 
3025 #if !defined(CONFIG_SPARC64) && !defined(CONFIG_ALPHA)
3026 	/* set the infinite burst register for chips that don't have
3027 	 * pci issues.
3028 	 */
3029 	if ((cp->cas_flags & CAS_FLAG_TARGET_ABORT) == 0)
3030 		writel(INF_BURST_EN, cp->regs + REG_INF_BURST);
3031 #endif
3032 
3033 	writel(0x1BF0, cp->regs + REG_MAC_SEND_PAUSE);
3034 
3035 	writel(0x00, cp->regs + REG_MAC_IPG0);
3036 	writel(0x08, cp->regs + REG_MAC_IPG1);
3037 	writel(0x04, cp->regs + REG_MAC_IPG2);
3038 
3039 	/* change later for 802.3z */
3040 	writel(0x40, cp->regs + REG_MAC_SLOT_TIME);
3041 
3042 	/* min frame + FCS */
3043 	writel(ETH_ZLEN + 4, cp->regs + REG_MAC_FRAMESIZE_MIN);
3044 
3045 	/* Ethernet payload + header + FCS + optional VLAN tag. NOTE: we
3046 	 * specify the maximum frame size to prevent RX tag errors on
3047 	 * oversized frames.
3048 	 */
3049 	writel(CAS_BASE(MAC_FRAMESIZE_MAX_BURST, 0x2000) |
3050 	       CAS_BASE(MAC_FRAMESIZE_MAX_FRAME,
3051 			(CAS_MAX_MTU + ETH_HLEN + 4 + 4)),
3052 	       cp->regs + REG_MAC_FRAMESIZE_MAX);
3053 
3054 	/* NOTE: crc_size is used as a surrogate for half-duplex.
3055 	 * workaround saturn half-duplex issue by increasing preamble
3056 	 * size to 65 bytes.
3057 	 */
3058 	if ((cp->cas_flags & CAS_FLAG_SATURN) && cp->crc_size)
3059 		writel(0x41, cp->regs + REG_MAC_PA_SIZE);
3060 	else
3061 		writel(0x07, cp->regs + REG_MAC_PA_SIZE);
3062 	writel(0x04, cp->regs + REG_MAC_JAM_SIZE);
3063 	writel(0x10, cp->regs + REG_MAC_ATTEMPT_LIMIT);
3064 	writel(0x8808, cp->regs + REG_MAC_CTRL_TYPE);
3065 
3066 	writel((e[5] | (e[4] << 8)) & 0x3ff, cp->regs + REG_MAC_RANDOM_SEED);
3067 
3068 	writel(0, cp->regs + REG_MAC_ADDR_FILTER0);
3069 	writel(0, cp->regs + REG_MAC_ADDR_FILTER1);
3070 	writel(0, cp->regs + REG_MAC_ADDR_FILTER2);
3071 	writel(0, cp->regs + REG_MAC_ADDR_FILTER2_1_MASK);
3072 	writel(0, cp->regs + REG_MAC_ADDR_FILTER0_MASK);
3073 
3074 	/* setup mac address in perfect filter array */
3075 	for (i = 0; i < 45; i++)
3076 		writel(0x0, cp->regs + REG_MAC_ADDRN(i));
3077 
3078 	writel((e[4] << 8) | e[5], cp->regs + REG_MAC_ADDRN(0));
3079 	writel((e[2] << 8) | e[3], cp->regs + REG_MAC_ADDRN(1));
3080 	writel((e[0] << 8) | e[1], cp->regs + REG_MAC_ADDRN(2));
3081 
3082 	writel(0x0001, cp->regs + REG_MAC_ADDRN(42));
3083 	writel(0xc200, cp->regs + REG_MAC_ADDRN(43));
3084 	writel(0x0180, cp->regs + REG_MAC_ADDRN(44));
3085 
3086 	cp->mac_rx_cfg = cas_setup_multicast(cp);
3087 
3088 	spin_lock(&cp->stat_lock[N_TX_RINGS]);
3089 	cas_clear_mac_err(cp);
3090 	spin_unlock(&cp->stat_lock[N_TX_RINGS]);
3091 
3092 	/* Setup MAC interrupts.  We want to get all of the interesting
3093 	 * counter expiration events, but we do not want to hear about
3094 	 * normal rx/tx as the DMA engine tells us that.
3095 	 */
3096 	writel(MAC_TX_FRAME_XMIT, cp->regs + REG_MAC_TX_MASK);
3097 	writel(MAC_RX_FRAME_RECV, cp->regs + REG_MAC_RX_MASK);
3098 
3099 	/* Don't enable even the PAUSE interrupts for now, we
3100 	 * make no use of those events other than to record them.
3101 	 */
3102 	writel(0xffffffff, cp->regs + REG_MAC_CTRL_MASK);
3103 }
3104 
3105 /* Must be invoked under cp->lock. */
3106 static void cas_init_pause_thresholds(struct cas *cp)
3107 {
3108 	/* Calculate pause thresholds.  Setting the OFF threshold to the
3109 	 * full RX fifo size effectively disables PAUSE generation
3110 	 */
3111 	if (cp->rx_fifo_size <= (2 * 1024)) {
3112 		cp->rx_pause_off = cp->rx_pause_on = cp->rx_fifo_size;
3113 	} else {
3114 		int max_frame = (cp->dev->mtu + ETH_HLEN + 4 + 4 + 64) & ~63;
3115 		if (max_frame * 3 > cp->rx_fifo_size) {
3116 			cp->rx_pause_off = 7104;
3117 			cp->rx_pause_on  = 960;
3118 		} else {
3119 			int off = (cp->rx_fifo_size - (max_frame * 2));
3120 			int on = off - max_frame;
3121 			cp->rx_pause_off = off;
3122 			cp->rx_pause_on = on;
3123 		}
3124 	}
3125 }
3126 
3127 static int cas_vpd_match(const void __iomem *p, const char *str)
3128 {
3129 	int len = strlen(str) + 1;
3130 	int i;
3131 
3132 	for (i = 0; i < len; i++) {
3133 		if (readb(p + i) != str[i])
3134 			return 0;
3135 	}
3136 	return 1;
3137 }
3138 
3139 
3140 /* get the mac address by reading the vpd information in the rom.
3141  * also get the phy type and determine if there's an entropy generator.
3142  * NOTE: this is a bit convoluted for the following reasons:
3143  *  1) vpd info has order-dependent mac addresses for multinic cards
3144  *  2) the only way to determine the nic order is to use the slot
3145  *     number.
3146  *  3) fiber cards don't have bridges, so their slot numbers don't
3147  *     mean anything.
3148  *  4) we don't actually know we have a fiber card until after
3149  *     the mac addresses are parsed.
3150  */
3151 static int cas_get_vpd_info(struct cas *cp, unsigned char *dev_addr,
3152 			    const int offset)
3153 {
3154 	void __iomem *p = cp->regs + REG_EXPANSION_ROM_RUN_START;
3155 	void __iomem *base, *kstart;
3156 	int i, len;
3157 	int found = 0;
3158 #define VPD_FOUND_MAC        0x01
3159 #define VPD_FOUND_PHY        0x02
3160 
3161 	int phy_type = CAS_PHY_MII_MDIO0; /* default phy type */
3162 	int mac_off  = 0;
3163 
3164 #if defined(CONFIG_SPARC)
3165 	const unsigned char *addr;
3166 #endif
3167 
3168 	/* give us access to the PROM */
3169 	writel(BIM_LOCAL_DEV_PROM | BIM_LOCAL_DEV_PAD,
3170 	       cp->regs + REG_BIM_LOCAL_DEV_EN);
3171 
3172 	/* check for an expansion rom */
3173 	if (readb(p) != 0x55 || readb(p + 1) != 0xaa)
3174 		goto use_random_mac_addr;
3175 
3176 	/* search for beginning of vpd */
3177 	base = NULL;
3178 	for (i = 2; i < EXPANSION_ROM_SIZE; i++) {
3179 		/* check for PCIR */
3180 		if ((readb(p + i + 0) == 0x50) &&
3181 		    (readb(p + i + 1) == 0x43) &&
3182 		    (readb(p + i + 2) == 0x49) &&
3183 		    (readb(p + i + 3) == 0x52)) {
3184 			base = p + (readb(p + i + 8) |
3185 				    (readb(p + i + 9) << 8));
3186 			break;
3187 		}
3188 	}
3189 
3190 	if (!base || (readb(base) != 0x82))
3191 		goto use_random_mac_addr;
3192 
3193 	i = (readb(base + 1) | (readb(base + 2) << 8)) + 3;
3194 	while (i < EXPANSION_ROM_SIZE) {
3195 		if (readb(base + i) != 0x90) /* no vpd found */
3196 			goto use_random_mac_addr;
3197 
3198 		/* found a vpd field */
3199 		len = readb(base + i + 1) | (readb(base + i + 2) << 8);
3200 
3201 		/* extract keywords */
3202 		kstart = base + i + 3;
3203 		p = kstart;
3204 		while ((p - kstart) < len) {
3205 			int klen = readb(p + 2);
3206 			int j;
3207 			char type;
3208 
3209 			p += 3;
3210 
3211 			/* look for the following things:
3212 			 * -- correct length == 29
3213 			 * 3 (type) + 2 (size) +
3214 			 * 18 (strlen("local-mac-address") + 1) +
3215 			 * 6 (mac addr)
3216 			 * -- VPD Instance 'I'
3217 			 * -- VPD Type Bytes 'B'
3218 			 * -- VPD data length == 6
3219 			 * -- property string == local-mac-address
3220 			 *
3221 			 * -- correct length == 24
3222 			 * 3 (type) + 2 (size) +
3223 			 * 12 (strlen("entropy-dev") + 1) +
3224 			 * 7 (strlen("vms110") + 1)
3225 			 * -- VPD Instance 'I'
3226 			 * -- VPD Type String 'B'
3227 			 * -- VPD data length == 7
3228 			 * -- property string == entropy-dev
3229 			 *
3230 			 * -- correct length == 18
3231 			 * 3 (type) + 2 (size) +
3232 			 * 9 (strlen("phy-type") + 1) +
3233 			 * 4 (strlen("pcs") + 1)
3234 			 * -- VPD Instance 'I'
3235 			 * -- VPD Type String 'S'
3236 			 * -- VPD data length == 4
3237 			 * -- property string == phy-type
3238 			 *
3239 			 * -- correct length == 23
3240 			 * 3 (type) + 2 (size) +
3241 			 * 14 (strlen("phy-interface") + 1) +
3242 			 * 4 (strlen("pcs") + 1)
3243 			 * -- VPD Instance 'I'
3244 			 * -- VPD Type String 'S'
3245 			 * -- VPD data length == 4
3246 			 * -- property string == phy-interface
3247 			 */
3248 			if (readb(p) != 'I')
3249 				goto next;
3250 
3251 			/* finally, check string and length */
3252 			type = readb(p + 3);
3253 			if (type == 'B') {
3254 				if ((klen == 29) && readb(p + 4) == 6 &&
3255 				    cas_vpd_match(p + 5,
3256 						  "local-mac-address")) {
3257 					if (mac_off++ > offset)
3258 						goto next;
3259 
3260 					/* set mac address */
3261 					for (j = 0; j < 6; j++)
3262 						dev_addr[j] =
3263 							readb(p + 23 + j);
3264 					goto found_mac;
3265 				}
3266 			}
3267 
3268 			if (type != 'S')
3269 				goto next;
3270 
3271 #ifdef USE_ENTROPY_DEV
3272 			if ((klen == 24) &&
3273 			    cas_vpd_match(p + 5, "entropy-dev") &&
3274 			    cas_vpd_match(p + 17, "vms110")) {
3275 				cp->cas_flags |= CAS_FLAG_ENTROPY_DEV;
3276 				goto next;
3277 			}
3278 #endif
3279 
3280 			if (found & VPD_FOUND_PHY)
3281 				goto next;
3282 
3283 			if ((klen == 18) && readb(p + 4) == 4 &&
3284 			    cas_vpd_match(p + 5, "phy-type")) {
3285 				if (cas_vpd_match(p + 14, "pcs")) {
3286 					phy_type = CAS_PHY_SERDES;
3287 					goto found_phy;
3288 				}
3289 			}
3290 
3291 			if ((klen == 23) && readb(p + 4) == 4 &&
3292 			    cas_vpd_match(p + 5, "phy-interface")) {
3293 				if (cas_vpd_match(p + 19, "pcs")) {
3294 					phy_type = CAS_PHY_SERDES;
3295 					goto found_phy;
3296 				}
3297 			}
3298 found_mac:
3299 			found |= VPD_FOUND_MAC;
3300 			goto next;
3301 
3302 found_phy:
3303 			found |= VPD_FOUND_PHY;
3304 
3305 next:
3306 			p += klen;
3307 		}
3308 		i += len + 3;
3309 	}
3310 
3311 use_random_mac_addr:
3312 	if (found & VPD_FOUND_MAC)
3313 		goto done;
3314 
3315 #if defined(CONFIG_SPARC)
3316 	addr = of_get_property(cp->of_node, "local-mac-address", NULL);
3317 	if (addr != NULL) {
3318 		memcpy(dev_addr, addr, ETH_ALEN);
3319 		goto done;
3320 	}
3321 #endif
3322 
3323 	/* Sun MAC prefix then 3 random bytes. */
3324 	pr_info("MAC address not found in ROM VPD\n");
3325 	dev_addr[0] = 0x08;
3326 	dev_addr[1] = 0x00;
3327 	dev_addr[2] = 0x20;
3328 	get_random_bytes(dev_addr + 3, 3);
3329 
3330 done:
3331 	writel(0, cp->regs + REG_BIM_LOCAL_DEV_EN);
3332 	return phy_type;
3333 }
3334 
3335 /* check pci invariants */
3336 static void cas_check_pci_invariants(struct cas *cp)
3337 {
3338 	struct pci_dev *pdev = cp->pdev;
3339 
3340 	cp->cas_flags = 0;
3341 	if ((pdev->vendor == PCI_VENDOR_ID_SUN) &&
3342 	    (pdev->device == PCI_DEVICE_ID_SUN_CASSINI)) {
3343 		if (pdev->revision >= CAS_ID_REVPLUS)
3344 			cp->cas_flags |= CAS_FLAG_REG_PLUS;
3345 		if (pdev->revision < CAS_ID_REVPLUS02u)
3346 			cp->cas_flags |= CAS_FLAG_TARGET_ABORT;
3347 
3348 		/* Original Cassini supports HW CSUM, but it's not
3349 		 * enabled by default as it can trigger TX hangs.
3350 		 */
3351 		if (pdev->revision < CAS_ID_REV2)
3352 			cp->cas_flags |= CAS_FLAG_NO_HW_CSUM;
3353 	} else {
3354 		/* Only sun has original cassini chips.  */
3355 		cp->cas_flags |= CAS_FLAG_REG_PLUS;
3356 
3357 		/* We use a flag because the same phy might be externally
3358 		 * connected.
3359 		 */
3360 		if ((pdev->vendor == PCI_VENDOR_ID_NS) &&
3361 		    (pdev->device == PCI_DEVICE_ID_NS_SATURN))
3362 			cp->cas_flags |= CAS_FLAG_SATURN;
3363 	}
3364 }
3365 
3366 
3367 static int cas_check_invariants(struct cas *cp)
3368 {
3369 	struct pci_dev *pdev = cp->pdev;
3370 	u8 addr[ETH_ALEN];
3371 	u32 cfg;
3372 	int i;
3373 
3374 	/* get page size for rx buffers. */
3375 	cp->page_order = 0;
3376 #ifdef USE_PAGE_ORDER
3377 	if (PAGE_SHIFT < CAS_JUMBO_PAGE_SHIFT) {
3378 		/* see if we can allocate larger pages */
3379 		struct page *page = alloc_pages(GFP_ATOMIC,
3380 						CAS_JUMBO_PAGE_SHIFT -
3381 						PAGE_SHIFT);
3382 		if (page) {
3383 			__free_pages(page, CAS_JUMBO_PAGE_SHIFT - PAGE_SHIFT);
3384 			cp->page_order = CAS_JUMBO_PAGE_SHIFT - PAGE_SHIFT;
3385 		} else {
3386 			printk("MTU limited to %d bytes\n", CAS_MAX_MTU);
3387 		}
3388 	}
3389 #endif
3390 	cp->page_size = (PAGE_SIZE << cp->page_order);
3391 
3392 	/* Fetch the FIFO configurations. */
3393 	cp->tx_fifo_size = readl(cp->regs + REG_TX_FIFO_SIZE) * 64;
3394 	cp->rx_fifo_size = RX_FIFO_SIZE;
3395 
3396 	/* finish phy determination. MDIO1 takes precedence over MDIO0 if
3397 	 * they're both connected.
3398 	 */
3399 	cp->phy_type = cas_get_vpd_info(cp, addr, PCI_SLOT(pdev->devfn));
3400 	eth_hw_addr_set(cp->dev, addr);
3401 	if (cp->phy_type & CAS_PHY_SERDES) {
3402 		cp->cas_flags |= CAS_FLAG_1000MB_CAP;
3403 		return 0; /* no more checking needed */
3404 	}
3405 
3406 	/* MII */
3407 	cfg = readl(cp->regs + REG_MIF_CFG);
3408 	if (cfg & MIF_CFG_MDIO_1) {
3409 		cp->phy_type = CAS_PHY_MII_MDIO1;
3410 	} else if (cfg & MIF_CFG_MDIO_0) {
3411 		cp->phy_type = CAS_PHY_MII_MDIO0;
3412 	}
3413 
3414 	cas_mif_poll(cp, 0);
3415 	writel(PCS_DATAPATH_MODE_MII, cp->regs + REG_PCS_DATAPATH_MODE);
3416 
3417 	for (i = 0; i < 32; i++) {
3418 		u32 phy_id;
3419 		int j;
3420 
3421 		for (j = 0; j < 3; j++) {
3422 			cp->phy_addr = i;
3423 			phy_id = cas_phy_read(cp, MII_PHYSID1) << 16;
3424 			phy_id |= cas_phy_read(cp, MII_PHYSID2);
3425 			if (phy_id && (phy_id != 0xFFFFFFFF)) {
3426 				cp->phy_id = phy_id;
3427 				goto done;
3428 			}
3429 		}
3430 	}
3431 	pr_err("MII phy did not respond [%08x]\n",
3432 	       readl(cp->regs + REG_MIF_STATE_MACHINE));
3433 	return -1;
3434 
3435 done:
3436 	/* see if we can do gigabit */
3437 	cfg = cas_phy_read(cp, MII_BMSR);
3438 	if ((cfg & CAS_BMSR_1000_EXTEND) &&
3439 	    cas_phy_read(cp, CAS_MII_1000_EXTEND))
3440 		cp->cas_flags |= CAS_FLAG_1000MB_CAP;
3441 	return 0;
3442 }
3443 
3444 /* Must be invoked under cp->lock. */
3445 static inline void cas_start_dma(struct cas *cp)
3446 {
3447 	int i;
3448 	u32 val;
3449 	int txfailed = 0;
3450 
3451 	/* enable dma */
3452 	val = readl(cp->regs + REG_TX_CFG) | TX_CFG_DMA_EN;
3453 	writel(val, cp->regs + REG_TX_CFG);
3454 	val = readl(cp->regs + REG_RX_CFG) | RX_CFG_DMA_EN;
3455 	writel(val, cp->regs + REG_RX_CFG);
3456 
3457 	/* enable the mac */
3458 	val = readl(cp->regs + REG_MAC_TX_CFG) | MAC_TX_CFG_EN;
3459 	writel(val, cp->regs + REG_MAC_TX_CFG);
3460 	val = readl(cp->regs + REG_MAC_RX_CFG) | MAC_RX_CFG_EN;
3461 	writel(val, cp->regs + REG_MAC_RX_CFG);
3462 
3463 	i = STOP_TRIES;
3464 	while (i-- > 0) {
3465 		val = readl(cp->regs + REG_MAC_TX_CFG);
3466 		if ((val & MAC_TX_CFG_EN))
3467 			break;
3468 		udelay(10);
3469 	}
3470 	if (i < 0) txfailed = 1;
3471 	i = STOP_TRIES;
3472 	while (i-- > 0) {
3473 		val = readl(cp->regs + REG_MAC_RX_CFG);
3474 		if ((val & MAC_RX_CFG_EN)) {
3475 			if (txfailed) {
3476 				netdev_err(cp->dev,
3477 					   "enabling mac failed [tx:%08x:%08x]\n",
3478 					   readl(cp->regs + REG_MIF_STATE_MACHINE),
3479 					   readl(cp->regs + REG_MAC_STATE_MACHINE));
3480 			}
3481 			goto enable_rx_done;
3482 		}
3483 		udelay(10);
3484 	}
3485 	netdev_err(cp->dev, "enabling mac failed [%s:%08x:%08x]\n",
3486 		   (txfailed ? "tx,rx" : "rx"),
3487 		   readl(cp->regs + REG_MIF_STATE_MACHINE),
3488 		   readl(cp->regs + REG_MAC_STATE_MACHINE));
3489 
3490 enable_rx_done:
3491 	cas_unmask_intr(cp); /* enable interrupts */
3492 	writel(RX_DESC_RINGN_SIZE(0) - 4, cp->regs + REG_RX_KICK);
3493 	writel(0, cp->regs + REG_RX_COMP_TAIL);
3494 
3495 	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
3496 		if (N_RX_DESC_RINGS > 1)
3497 			writel(RX_DESC_RINGN_SIZE(1) - 4,
3498 			       cp->regs + REG_PLUS_RX_KICK1);
3499 	}
3500 }
3501 
3502 /* Must be invoked under cp->lock. */
3503 static void cas_read_pcs_link_mode(struct cas *cp, int *fd, int *spd,
3504 				   int *pause)
3505 {
3506 	u32 val = readl(cp->regs + REG_PCS_MII_LPA);
3507 	*fd     = (val & PCS_MII_LPA_FD) ? 1 : 0;
3508 	*pause  = (val & PCS_MII_LPA_SYM_PAUSE) ? 0x01 : 0x00;
3509 	if (val & PCS_MII_LPA_ASYM_PAUSE)
3510 		*pause |= 0x10;
3511 	*spd = 1000;
3512 }
3513 
3514 /* Must be invoked under cp->lock. */
3515 static void cas_read_mii_link_mode(struct cas *cp, int *fd, int *spd,
3516 				   int *pause)
3517 {
3518 	u32 val;
3519 
3520 	*fd = 0;
3521 	*spd = 10;
3522 	*pause = 0;
3523 
3524 	/* use GMII registers */
3525 	val = cas_phy_read(cp, MII_LPA);
3526 	if (val & CAS_LPA_PAUSE)
3527 		*pause = 0x01;
3528 
3529 	if (val & CAS_LPA_ASYM_PAUSE)
3530 		*pause |= 0x10;
3531 
3532 	if (val & LPA_DUPLEX)
3533 		*fd = 1;
3534 	if (val & LPA_100)
3535 		*spd = 100;
3536 
3537 	if (cp->cas_flags & CAS_FLAG_1000MB_CAP) {
3538 		val = cas_phy_read(cp, CAS_MII_1000_STATUS);
3539 		if (val & (CAS_LPA_1000FULL | CAS_LPA_1000HALF))
3540 			*spd = 1000;
3541 		if (val & CAS_LPA_1000FULL)
3542 			*fd = 1;
3543 	}
3544 }
3545 
3546 /* A link-up condition has occurred, initialize and enable the
3547  * rest of the chip.
3548  *
3549  * Must be invoked under cp->lock.
3550  */
3551 static void cas_set_link_modes(struct cas *cp)
3552 {
3553 	u32 val;
3554 	int full_duplex, speed, pause;
3555 
3556 	full_duplex = 0;
3557 	speed = 10;
3558 	pause = 0;
3559 
3560 	if (CAS_PHY_MII(cp->phy_type)) {
3561 		cas_mif_poll(cp, 0);
3562 		val = cas_phy_read(cp, MII_BMCR);
3563 		if (val & BMCR_ANENABLE) {
3564 			cas_read_mii_link_mode(cp, &full_duplex, &speed,
3565 					       &pause);
3566 		} else {
3567 			if (val & BMCR_FULLDPLX)
3568 				full_duplex = 1;
3569 
3570 			if (val & BMCR_SPEED100)
3571 				speed = 100;
3572 			else if (val & CAS_BMCR_SPEED1000)
3573 				speed = (cp->cas_flags & CAS_FLAG_1000MB_CAP) ?
3574 					1000 : 100;
3575 		}
3576 		cas_mif_poll(cp, 1);
3577 
3578 	} else {
3579 		val = readl(cp->regs + REG_PCS_MII_CTRL);
3580 		cas_read_pcs_link_mode(cp, &full_duplex, &speed, &pause);
3581 		if ((val & PCS_MII_AUTONEG_EN) == 0) {
3582 			if (val & PCS_MII_CTRL_DUPLEX)
3583 				full_duplex = 1;
3584 		}
3585 	}
3586 
3587 	netif_info(cp, link, cp->dev, "Link up at %d Mbps, %s-duplex\n",
3588 		   speed, full_duplex ? "full" : "half");
3589 
3590 	val = MAC_XIF_TX_MII_OUTPUT_EN | MAC_XIF_LINK_LED;
3591 	if (CAS_PHY_MII(cp->phy_type)) {
3592 		val |= MAC_XIF_MII_BUFFER_OUTPUT_EN;
3593 		if (!full_duplex)
3594 			val |= MAC_XIF_DISABLE_ECHO;
3595 	}
3596 	if (full_duplex)
3597 		val |= MAC_XIF_FDPLX_LED;
3598 	if (speed == 1000)
3599 		val |= MAC_XIF_GMII_MODE;
3600 	writel(val, cp->regs + REG_MAC_XIF_CFG);
3601 
3602 	/* deal with carrier and collision detect. */
3603 	val = MAC_TX_CFG_IPG_EN;
3604 	if (full_duplex) {
3605 		val |= MAC_TX_CFG_IGNORE_CARRIER;
3606 		val |= MAC_TX_CFG_IGNORE_COLL;
3607 	} else {
3608 #ifndef USE_CSMA_CD_PROTO
3609 		val |= MAC_TX_CFG_NEVER_GIVE_UP_EN;
3610 		val |= MAC_TX_CFG_NEVER_GIVE_UP_LIM;
3611 #endif
3612 	}
3613 	/* val now set up for REG_MAC_TX_CFG */
3614 
3615 	/* If gigabit and half-duplex, enable carrier extension
3616 	 * mode.  increase slot time to 512 bytes as well.
3617 	 * else, disable it and make sure slot time is 64 bytes.
3618 	 * also activate checksum bug workaround
3619 	 */
3620 	if ((speed == 1000) && !full_duplex) {
3621 		writel(val | MAC_TX_CFG_CARRIER_EXTEND,
3622 		       cp->regs + REG_MAC_TX_CFG);
3623 
3624 		val = readl(cp->regs + REG_MAC_RX_CFG);
3625 		val &= ~MAC_RX_CFG_STRIP_FCS; /* checksum workaround */
3626 		writel(val | MAC_RX_CFG_CARRIER_EXTEND,
3627 		       cp->regs + REG_MAC_RX_CFG);
3628 
3629 		writel(0x200, cp->regs + REG_MAC_SLOT_TIME);
3630 
3631 		cp->crc_size = 4;
3632 		/* minimum size gigabit frame at half duplex */
3633 		cp->min_frame_size = CAS_1000MB_MIN_FRAME;
3634 
3635 	} else {
3636 		writel(val, cp->regs + REG_MAC_TX_CFG);
3637 
3638 		/* checksum bug workaround. don't strip FCS when in
3639 		 * half-duplex mode
3640 		 */
3641 		val = readl(cp->regs + REG_MAC_RX_CFG);
3642 		if (full_duplex) {
3643 			val |= MAC_RX_CFG_STRIP_FCS;
3644 			cp->crc_size = 0;
3645 			cp->min_frame_size = CAS_MIN_MTU;
3646 		} else {
3647 			val &= ~MAC_RX_CFG_STRIP_FCS;
3648 			cp->crc_size = 4;
3649 			cp->min_frame_size = CAS_MIN_FRAME;
3650 		}
3651 		writel(val & ~MAC_RX_CFG_CARRIER_EXTEND,
3652 		       cp->regs + REG_MAC_RX_CFG);
3653 		writel(0x40, cp->regs + REG_MAC_SLOT_TIME);
3654 	}
3655 
3656 	if (netif_msg_link(cp)) {
3657 		if (pause & 0x01) {
3658 			netdev_info(cp->dev, "Pause is enabled (rxfifo: %d off: %d on: %d)\n",
3659 				    cp->rx_fifo_size,
3660 				    cp->rx_pause_off,
3661 				    cp->rx_pause_on);
3662 		} else if (pause & 0x10) {
3663 			netdev_info(cp->dev, "TX pause enabled\n");
3664 		} else {
3665 			netdev_info(cp->dev, "Pause is disabled\n");
3666 		}
3667 	}
3668 
3669 	val = readl(cp->regs + REG_MAC_CTRL_CFG);
3670 	val &= ~(MAC_CTRL_CFG_SEND_PAUSE_EN | MAC_CTRL_CFG_RECV_PAUSE_EN);
3671 	if (pause) { /* symmetric or asymmetric pause */
3672 		val |= MAC_CTRL_CFG_SEND_PAUSE_EN;
3673 		if (pause & 0x01) { /* symmetric pause */
3674 			val |= MAC_CTRL_CFG_RECV_PAUSE_EN;
3675 		}
3676 	}
3677 	writel(val, cp->regs + REG_MAC_CTRL_CFG);
3678 	cas_start_dma(cp);
3679 }
3680 
3681 /* Must be invoked under cp->lock. */
3682 static void cas_init_hw(struct cas *cp, int restart_link)
3683 {
3684 	if (restart_link)
3685 		cas_phy_init(cp);
3686 
3687 	cas_init_pause_thresholds(cp);
3688 	cas_init_mac(cp);
3689 	cas_init_dma(cp);
3690 
3691 	if (restart_link) {
3692 		/* Default aneg parameters */
3693 		cp->timer_ticks = 0;
3694 		cas_begin_auto_negotiation(cp, NULL);
3695 	} else if (cp->lstate == link_up) {
3696 		cas_set_link_modes(cp);
3697 		netif_carrier_on(cp->dev);
3698 	}
3699 }
3700 
3701 /* Must be invoked under cp->lock. on earlier cassini boards,
3702  * SOFT_0 is tied to PCI reset. we use this to force a pci reset,
3703  * let it settle out, and then restore pci state.
3704  */
3705 static void cas_hard_reset(struct cas *cp)
3706 {
3707 	writel(BIM_LOCAL_DEV_SOFT_0, cp->regs + REG_BIM_LOCAL_DEV_EN);
3708 	udelay(20);
3709 	pci_restore_state(cp->pdev);
3710 }
3711 
3712 
3713 static void cas_global_reset(struct cas *cp, int blkflag)
3714 {
3715 	int limit;
3716 
3717 	/* issue a global reset. don't use RSTOUT. */
3718 	if (blkflag && !CAS_PHY_MII(cp->phy_type)) {
3719 		/* For PCS, when the blkflag is set, we should set the
3720 		 * SW_REST_BLOCK_PCS_SLINK bit to prevent the results of
3721 		 * the last autonegotiation from being cleared.  We'll
3722 		 * need some special handling if the chip is set into a
3723 		 * loopback mode.
3724 		 */
3725 		writel((SW_RESET_TX | SW_RESET_RX | SW_RESET_BLOCK_PCS_SLINK),
3726 		       cp->regs + REG_SW_RESET);
3727 	} else {
3728 		writel(SW_RESET_TX | SW_RESET_RX, cp->regs + REG_SW_RESET);
3729 	}
3730 
3731 	/* need to wait at least 3ms before polling register */
3732 	mdelay(3);
3733 
3734 	limit = STOP_TRIES;
3735 	while (limit-- > 0) {
3736 		u32 val = readl(cp->regs + REG_SW_RESET);
3737 		if ((val & (SW_RESET_TX | SW_RESET_RX)) == 0)
3738 			goto done;
3739 		udelay(10);
3740 	}
3741 	netdev_err(cp->dev, "sw reset failed\n");
3742 
3743 done:
3744 	/* enable various BIM interrupts */
3745 	writel(BIM_CFG_DPAR_INTR_ENABLE | BIM_CFG_RMA_INTR_ENABLE |
3746 	       BIM_CFG_RTA_INTR_ENABLE, cp->regs + REG_BIM_CFG);
3747 
3748 	/* clear out pci error status mask for handled errors.
3749 	 * we don't deal with DMA counter overflows as they happen
3750 	 * all the time.
3751 	 */
3752 	writel(0xFFFFFFFFU & ~(PCI_ERR_BADACK | PCI_ERR_DTRTO |
3753 			       PCI_ERR_OTHER | PCI_ERR_BIM_DMA_WRITE |
3754 			       PCI_ERR_BIM_DMA_READ), cp->regs +
3755 	       REG_PCI_ERR_STATUS_MASK);
3756 
3757 	/* set up for MII by default to address mac rx reset timeout
3758 	 * issue
3759 	 */
3760 	writel(PCS_DATAPATH_MODE_MII, cp->regs + REG_PCS_DATAPATH_MODE);
3761 }
3762 
3763 static void cas_reset(struct cas *cp, int blkflag)
3764 {
3765 	u32 val;
3766 
3767 	cas_mask_intr(cp);
3768 	cas_global_reset(cp, blkflag);
3769 	cas_mac_reset(cp);
3770 	cas_entropy_reset(cp);
3771 
3772 	/* disable dma engines. */
3773 	val = readl(cp->regs + REG_TX_CFG);
3774 	val &= ~TX_CFG_DMA_EN;
3775 	writel(val, cp->regs + REG_TX_CFG);
3776 
3777 	val = readl(cp->regs + REG_RX_CFG);
3778 	val &= ~RX_CFG_DMA_EN;
3779 	writel(val, cp->regs + REG_RX_CFG);
3780 
3781 	/* program header parser */
3782 	if ((cp->cas_flags & CAS_FLAG_TARGET_ABORT) ||
3783 	    (&CAS_HP_ALT_FIRMWARE[0] == &cas_prog_null[0])) {
3784 		cas_load_firmware(cp, CAS_HP_FIRMWARE);
3785 	} else {
3786 		cas_load_firmware(cp, CAS_HP_ALT_FIRMWARE);
3787 	}
3788 
3789 	/* clear out error registers */
3790 	spin_lock(&cp->stat_lock[N_TX_RINGS]);
3791 	cas_clear_mac_err(cp);
3792 	spin_unlock(&cp->stat_lock[N_TX_RINGS]);
3793 }
3794 
3795 /* Shut down the chip, must be called with pm_mutex held.  */
3796 static void cas_shutdown(struct cas *cp)
3797 {
3798 	unsigned long flags;
3799 
3800 	/* Make us not-running to avoid timers respawning */
3801 	cp->hw_running = 0;
3802 
3803 	del_timer_sync(&cp->link_timer);
3804 
3805 	/* Stop the reset task */
3806 #if 0
3807 	while (atomic_read(&cp->reset_task_pending_mtu) ||
3808 	       atomic_read(&cp->reset_task_pending_spare) ||
3809 	       atomic_read(&cp->reset_task_pending_all))
3810 		schedule();
3811 
3812 #else
3813 	while (atomic_read(&cp->reset_task_pending))
3814 		schedule();
3815 #endif
3816 	/* Actually stop the chip */
3817 	cas_lock_all_save(cp, flags);
3818 	cas_reset(cp, 0);
3819 	if (cp->cas_flags & CAS_FLAG_SATURN)
3820 		cas_phy_powerdown(cp);
3821 	cas_unlock_all_restore(cp, flags);
3822 }
3823 
3824 static int cas_change_mtu(struct net_device *dev, int new_mtu)
3825 {
3826 	struct cas *cp = netdev_priv(dev);
3827 
3828 	dev->mtu = new_mtu;
3829 	if (!netif_running(dev) || !netif_device_present(dev))
3830 		return 0;
3831 
3832 	/* let the reset task handle it */
3833 #if 1
3834 	atomic_inc(&cp->reset_task_pending);
3835 	if ((cp->phy_type & CAS_PHY_SERDES)) {
3836 		atomic_inc(&cp->reset_task_pending_all);
3837 	} else {
3838 		atomic_inc(&cp->reset_task_pending_mtu);
3839 	}
3840 	schedule_work(&cp->reset_task);
3841 #else
3842 	atomic_set(&cp->reset_task_pending, (cp->phy_type & CAS_PHY_SERDES) ?
3843 		   CAS_RESET_ALL : CAS_RESET_MTU);
3844 	pr_err("reset called in cas_change_mtu\n");
3845 	schedule_work(&cp->reset_task);
3846 #endif
3847 
3848 	flush_work(&cp->reset_task);
3849 	return 0;
3850 }
3851 
3852 static void cas_clean_txd(struct cas *cp, int ring)
3853 {
3854 	struct cas_tx_desc *txd = cp->init_txds[ring];
3855 	struct sk_buff *skb, **skbs = cp->tx_skbs[ring];
3856 	u64 daddr, dlen;
3857 	int i, size;
3858 
3859 	size = TX_DESC_RINGN_SIZE(ring);
3860 	for (i = 0; i < size; i++) {
3861 		int frag;
3862 
3863 		if (skbs[i] == NULL)
3864 			continue;
3865 
3866 		skb = skbs[i];
3867 		skbs[i] = NULL;
3868 
3869 		for (frag = 0; frag <= skb_shinfo(skb)->nr_frags;  frag++) {
3870 			int ent = i & (size - 1);
3871 
3872 			/* first buffer is never a tiny buffer and so
3873 			 * needs to be unmapped.
3874 			 */
3875 			daddr = le64_to_cpu(txd[ent].buffer);
3876 			dlen  =  CAS_VAL(TX_DESC_BUFLEN,
3877 					 le64_to_cpu(txd[ent].control));
3878 			dma_unmap_page(&cp->pdev->dev, daddr, dlen,
3879 				       DMA_TO_DEVICE);
3880 
3881 			if (frag != skb_shinfo(skb)->nr_frags) {
3882 				i++;
3883 
3884 				/* next buffer might by a tiny buffer.
3885 				 * skip past it.
3886 				 */
3887 				ent = i & (size - 1);
3888 				if (cp->tx_tiny_use[ring][ent].used)
3889 					i++;
3890 			}
3891 		}
3892 		dev_kfree_skb_any(skb);
3893 	}
3894 
3895 	/* zero out tiny buf usage */
3896 	memset(cp->tx_tiny_use[ring], 0, size*sizeof(*cp->tx_tiny_use[ring]));
3897 }
3898 
3899 /* freed on close */
3900 static inline void cas_free_rx_desc(struct cas *cp, int ring)
3901 {
3902 	cas_page_t **page = cp->rx_pages[ring];
3903 	int i, size;
3904 
3905 	size = RX_DESC_RINGN_SIZE(ring);
3906 	for (i = 0; i < size; i++) {
3907 		if (page[i]) {
3908 			cas_page_free(cp, page[i]);
3909 			page[i] = NULL;
3910 		}
3911 	}
3912 }
3913 
3914 static void cas_free_rxds(struct cas *cp)
3915 {
3916 	int i;
3917 
3918 	for (i = 0; i < N_RX_DESC_RINGS; i++)
3919 		cas_free_rx_desc(cp, i);
3920 }
3921 
3922 /* Must be invoked under cp->lock. */
3923 static void cas_clean_rings(struct cas *cp)
3924 {
3925 	int i;
3926 
3927 	/* need to clean all tx rings */
3928 	memset(cp->tx_old, 0, sizeof(*cp->tx_old)*N_TX_RINGS);
3929 	memset(cp->tx_new, 0, sizeof(*cp->tx_new)*N_TX_RINGS);
3930 	for (i = 0; i < N_TX_RINGS; i++)
3931 		cas_clean_txd(cp, i);
3932 
3933 	/* zero out init block */
3934 	memset(cp->init_block, 0, sizeof(struct cas_init_block));
3935 	cas_clean_rxds(cp);
3936 	cas_clean_rxcs(cp);
3937 }
3938 
3939 /* allocated on open */
3940 static inline int cas_alloc_rx_desc(struct cas *cp, int ring)
3941 {
3942 	cas_page_t **page = cp->rx_pages[ring];
3943 	int size, i = 0;
3944 
3945 	size = RX_DESC_RINGN_SIZE(ring);
3946 	for (i = 0; i < size; i++) {
3947 		if ((page[i] = cas_page_alloc(cp, GFP_KERNEL)) == NULL)
3948 			return -1;
3949 	}
3950 	return 0;
3951 }
3952 
3953 static int cas_alloc_rxds(struct cas *cp)
3954 {
3955 	int i;
3956 
3957 	for (i = 0; i < N_RX_DESC_RINGS; i++) {
3958 		if (cas_alloc_rx_desc(cp, i) < 0) {
3959 			cas_free_rxds(cp);
3960 			return -1;
3961 		}
3962 	}
3963 	return 0;
3964 }
3965 
3966 static void cas_reset_task(struct work_struct *work)
3967 {
3968 	struct cas *cp = container_of(work, struct cas, reset_task);
3969 #if 0
3970 	int pending = atomic_read(&cp->reset_task_pending);
3971 #else
3972 	int pending_all = atomic_read(&cp->reset_task_pending_all);
3973 	int pending_spare = atomic_read(&cp->reset_task_pending_spare);
3974 	int pending_mtu = atomic_read(&cp->reset_task_pending_mtu);
3975 
3976 	if (pending_all == 0 && pending_spare == 0 && pending_mtu == 0) {
3977 		/* We can have more tasks scheduled than actually
3978 		 * needed.
3979 		 */
3980 		atomic_dec(&cp->reset_task_pending);
3981 		return;
3982 	}
3983 #endif
3984 	/* The link went down, we reset the ring, but keep
3985 	 * DMA stopped. Use this function for reset
3986 	 * on error as well.
3987 	 */
3988 	if (cp->hw_running) {
3989 		unsigned long flags;
3990 
3991 		/* Make sure we don't get interrupts or tx packets */
3992 		netif_device_detach(cp->dev);
3993 		cas_lock_all_save(cp, flags);
3994 
3995 		if (cp->opened) {
3996 			/* We call cas_spare_recover when we call cas_open.
3997 			 * but we do not initialize the lists cas_spare_recover
3998 			 * uses until cas_open is called.
3999 			 */
4000 			cas_spare_recover(cp, GFP_ATOMIC);
4001 		}
4002 #if 1
4003 		/* test => only pending_spare set */
4004 		if (!pending_all && !pending_mtu)
4005 			goto done;
4006 #else
4007 		if (pending == CAS_RESET_SPARE)
4008 			goto done;
4009 #endif
4010 		/* when pending == CAS_RESET_ALL, the following
4011 		 * call to cas_init_hw will restart auto negotiation.
4012 		 * Setting the second argument of cas_reset to
4013 		 * !(pending == CAS_RESET_ALL) will set this argument
4014 		 * to 1 (avoiding reinitializing the PHY for the normal
4015 		 * PCS case) when auto negotiation is not restarted.
4016 		 */
4017 #if 1
4018 		cas_reset(cp, !(pending_all > 0));
4019 		if (cp->opened)
4020 			cas_clean_rings(cp);
4021 		cas_init_hw(cp, (pending_all > 0));
4022 #else
4023 		cas_reset(cp, !(pending == CAS_RESET_ALL));
4024 		if (cp->opened)
4025 			cas_clean_rings(cp);
4026 		cas_init_hw(cp, pending == CAS_RESET_ALL);
4027 #endif
4028 
4029 done:
4030 		cas_unlock_all_restore(cp, flags);
4031 		netif_device_attach(cp->dev);
4032 	}
4033 #if 1
4034 	atomic_sub(pending_all, &cp->reset_task_pending_all);
4035 	atomic_sub(pending_spare, &cp->reset_task_pending_spare);
4036 	atomic_sub(pending_mtu, &cp->reset_task_pending_mtu);
4037 	atomic_dec(&cp->reset_task_pending);
4038 #else
4039 	atomic_set(&cp->reset_task_pending, 0);
4040 #endif
4041 }
4042 
4043 static void cas_link_timer(struct timer_list *t)
4044 {
4045 	struct cas *cp = from_timer(cp, t, link_timer);
4046 	int mask, pending = 0, reset = 0;
4047 	unsigned long flags;
4048 
4049 	if (link_transition_timeout != 0 &&
4050 	    cp->link_transition_jiffies_valid &&
4051 	    time_is_before_jiffies(cp->link_transition_jiffies +
4052 	      link_transition_timeout)) {
4053 		/* One-second counter so link-down workaround doesn't
4054 		 * cause resets to occur so fast as to fool the switch
4055 		 * into thinking the link is down.
4056 		 */
4057 		cp->link_transition_jiffies_valid = 0;
4058 	}
4059 
4060 	if (!cp->hw_running)
4061 		return;
4062 
4063 	spin_lock_irqsave(&cp->lock, flags);
4064 	cas_lock_tx(cp);
4065 	cas_entropy_gather(cp);
4066 
4067 	/* If the link task is still pending, we just
4068 	 * reschedule the link timer
4069 	 */
4070 #if 1
4071 	if (atomic_read(&cp->reset_task_pending_all) ||
4072 	    atomic_read(&cp->reset_task_pending_spare) ||
4073 	    atomic_read(&cp->reset_task_pending_mtu))
4074 		goto done;
4075 #else
4076 	if (atomic_read(&cp->reset_task_pending))
4077 		goto done;
4078 #endif
4079 
4080 	/* check for rx cleaning */
4081 	if ((mask = (cp->cas_flags & CAS_FLAG_RXD_POST_MASK))) {
4082 		int i, rmask;
4083 
4084 		for (i = 0; i < MAX_RX_DESC_RINGS; i++) {
4085 			rmask = CAS_FLAG_RXD_POST(i);
4086 			if ((mask & rmask) == 0)
4087 				continue;
4088 
4089 			/* post_rxds will do a mod_timer */
4090 			if (cas_post_rxds_ringN(cp, i, cp->rx_last[i]) < 0) {
4091 				pending = 1;
4092 				continue;
4093 			}
4094 			cp->cas_flags &= ~rmask;
4095 		}
4096 	}
4097 
4098 	if (CAS_PHY_MII(cp->phy_type)) {
4099 		u16 bmsr;
4100 		cas_mif_poll(cp, 0);
4101 		bmsr = cas_phy_read(cp, MII_BMSR);
4102 		/* WTZ: Solaris driver reads this twice, but that
4103 		 * may be due to the PCS case and the use of a
4104 		 * common implementation. Read it twice here to be
4105 		 * safe.
4106 		 */
4107 		bmsr = cas_phy_read(cp, MII_BMSR);
4108 		cas_mif_poll(cp, 1);
4109 		readl(cp->regs + REG_MIF_STATUS); /* avoid dups */
4110 		reset = cas_mii_link_check(cp, bmsr);
4111 	} else {
4112 		reset = cas_pcs_link_check(cp);
4113 	}
4114 
4115 	if (reset)
4116 		goto done;
4117 
4118 	/* check for tx state machine confusion */
4119 	if ((readl(cp->regs + REG_MAC_TX_STATUS) & MAC_TX_FRAME_XMIT) == 0) {
4120 		u32 val = readl(cp->regs + REG_MAC_STATE_MACHINE);
4121 		u32 wptr, rptr;
4122 		int tlm  = CAS_VAL(MAC_SM_TLM, val);
4123 
4124 		if (((tlm == 0x5) || (tlm == 0x3)) &&
4125 		    (CAS_VAL(MAC_SM_ENCAP_SM, val) == 0)) {
4126 			netif_printk(cp, tx_err, KERN_DEBUG, cp->dev,
4127 				     "tx err: MAC_STATE[%08x]\n", val);
4128 			reset = 1;
4129 			goto done;
4130 		}
4131 
4132 		val  = readl(cp->regs + REG_TX_FIFO_PKT_CNT);
4133 		wptr = readl(cp->regs + REG_TX_FIFO_WRITE_PTR);
4134 		rptr = readl(cp->regs + REG_TX_FIFO_READ_PTR);
4135 		if ((val == 0) && (wptr != rptr)) {
4136 			netif_printk(cp, tx_err, KERN_DEBUG, cp->dev,
4137 				     "tx err: TX_FIFO[%08x:%08x:%08x]\n",
4138 				     val, wptr, rptr);
4139 			reset = 1;
4140 		}
4141 
4142 		if (reset)
4143 			cas_hard_reset(cp);
4144 	}
4145 
4146 done:
4147 	if (reset) {
4148 #if 1
4149 		atomic_inc(&cp->reset_task_pending);
4150 		atomic_inc(&cp->reset_task_pending_all);
4151 		schedule_work(&cp->reset_task);
4152 #else
4153 		atomic_set(&cp->reset_task_pending, CAS_RESET_ALL);
4154 		pr_err("reset called in cas_link_timer\n");
4155 		schedule_work(&cp->reset_task);
4156 #endif
4157 	}
4158 
4159 	if (!pending)
4160 		mod_timer(&cp->link_timer, jiffies + CAS_LINK_TIMEOUT);
4161 	cas_unlock_tx(cp);
4162 	spin_unlock_irqrestore(&cp->lock, flags);
4163 }
4164 
4165 /* tiny buffers are used to avoid target abort issues with
4166  * older cassini's
4167  */
4168 static void cas_tx_tiny_free(struct cas *cp)
4169 {
4170 	struct pci_dev *pdev = cp->pdev;
4171 	int i;
4172 
4173 	for (i = 0; i < N_TX_RINGS; i++) {
4174 		if (!cp->tx_tiny_bufs[i])
4175 			continue;
4176 
4177 		dma_free_coherent(&pdev->dev, TX_TINY_BUF_BLOCK,
4178 				  cp->tx_tiny_bufs[i], cp->tx_tiny_dvma[i]);
4179 		cp->tx_tiny_bufs[i] = NULL;
4180 	}
4181 }
4182 
4183 static int cas_tx_tiny_alloc(struct cas *cp)
4184 {
4185 	struct pci_dev *pdev = cp->pdev;
4186 	int i;
4187 
4188 	for (i = 0; i < N_TX_RINGS; i++) {
4189 		cp->tx_tiny_bufs[i] =
4190 			dma_alloc_coherent(&pdev->dev, TX_TINY_BUF_BLOCK,
4191 					   &cp->tx_tiny_dvma[i], GFP_KERNEL);
4192 		if (!cp->tx_tiny_bufs[i]) {
4193 			cas_tx_tiny_free(cp);
4194 			return -1;
4195 		}
4196 	}
4197 	return 0;
4198 }
4199 
4200 
4201 static int cas_open(struct net_device *dev)
4202 {
4203 	struct cas *cp = netdev_priv(dev);
4204 	int hw_was_up, err;
4205 	unsigned long flags;
4206 
4207 	mutex_lock(&cp->pm_mutex);
4208 
4209 	hw_was_up = cp->hw_running;
4210 
4211 	/* The power-management mutex protects the hw_running
4212 	 * etc. state so it is safe to do this bit without cp->lock
4213 	 */
4214 	if (!cp->hw_running) {
4215 		/* Reset the chip */
4216 		cas_lock_all_save(cp, flags);
4217 		/* We set the second arg to cas_reset to zero
4218 		 * because cas_init_hw below will have its second
4219 		 * argument set to non-zero, which will force
4220 		 * autonegotiation to start.
4221 		 */
4222 		cas_reset(cp, 0);
4223 		cp->hw_running = 1;
4224 		cas_unlock_all_restore(cp, flags);
4225 	}
4226 
4227 	err = -ENOMEM;
4228 	if (cas_tx_tiny_alloc(cp) < 0)
4229 		goto err_unlock;
4230 
4231 	/* alloc rx descriptors */
4232 	if (cas_alloc_rxds(cp) < 0)
4233 		goto err_tx_tiny;
4234 
4235 	/* allocate spares */
4236 	cas_spare_init(cp);
4237 	cas_spare_recover(cp, GFP_KERNEL);
4238 
4239 	/* We can now request the interrupt as we know it's masked
4240 	 * on the controller. cassini+ has up to 4 interrupts
4241 	 * that can be used, but you need to do explicit pci interrupt
4242 	 * mapping to expose them
4243 	 */
4244 	if (request_irq(cp->pdev->irq, cas_interrupt,
4245 			IRQF_SHARED, dev->name, (void *) dev)) {
4246 		netdev_err(cp->dev, "failed to request irq !\n");
4247 		err = -EAGAIN;
4248 		goto err_spare;
4249 	}
4250 
4251 #ifdef USE_NAPI
4252 	napi_enable(&cp->napi);
4253 #endif
4254 	/* init hw */
4255 	cas_lock_all_save(cp, flags);
4256 	cas_clean_rings(cp);
4257 	cas_init_hw(cp, !hw_was_up);
4258 	cp->opened = 1;
4259 	cas_unlock_all_restore(cp, flags);
4260 
4261 	netif_start_queue(dev);
4262 	mutex_unlock(&cp->pm_mutex);
4263 	return 0;
4264 
4265 err_spare:
4266 	cas_spare_free(cp);
4267 	cas_free_rxds(cp);
4268 err_tx_tiny:
4269 	cas_tx_tiny_free(cp);
4270 err_unlock:
4271 	mutex_unlock(&cp->pm_mutex);
4272 	return err;
4273 }
4274 
4275 static int cas_close(struct net_device *dev)
4276 {
4277 	unsigned long flags;
4278 	struct cas *cp = netdev_priv(dev);
4279 
4280 #ifdef USE_NAPI
4281 	napi_disable(&cp->napi);
4282 #endif
4283 	/* Make sure we don't get distracted by suspend/resume */
4284 	mutex_lock(&cp->pm_mutex);
4285 
4286 	netif_stop_queue(dev);
4287 
4288 	/* Stop traffic, mark us closed */
4289 	cas_lock_all_save(cp, flags);
4290 	cp->opened = 0;
4291 	cas_reset(cp, 0);
4292 	cas_phy_init(cp);
4293 	cas_begin_auto_negotiation(cp, NULL);
4294 	cas_clean_rings(cp);
4295 	cas_unlock_all_restore(cp, flags);
4296 
4297 	free_irq(cp->pdev->irq, (void *) dev);
4298 	cas_spare_free(cp);
4299 	cas_free_rxds(cp);
4300 	cas_tx_tiny_free(cp);
4301 	mutex_unlock(&cp->pm_mutex);
4302 	return 0;
4303 }
4304 
4305 static struct {
4306 	const char name[ETH_GSTRING_LEN];
4307 } ethtool_cassini_statnames[] = {
4308 	{"collisions"},
4309 	{"rx_bytes"},
4310 	{"rx_crc_errors"},
4311 	{"rx_dropped"},
4312 	{"rx_errors"},
4313 	{"rx_fifo_errors"},
4314 	{"rx_frame_errors"},
4315 	{"rx_length_errors"},
4316 	{"rx_over_errors"},
4317 	{"rx_packets"},
4318 	{"tx_aborted_errors"},
4319 	{"tx_bytes"},
4320 	{"tx_dropped"},
4321 	{"tx_errors"},
4322 	{"tx_fifo_errors"},
4323 	{"tx_packets"}
4324 };
4325 #define CAS_NUM_STAT_KEYS ARRAY_SIZE(ethtool_cassini_statnames)
4326 
4327 static struct {
4328 	const int offsets;	/* neg. values for 2nd arg to cas_read_phy */
4329 } ethtool_register_table[] = {
4330 	{-MII_BMSR},
4331 	{-MII_BMCR},
4332 	{REG_CAWR},
4333 	{REG_INF_BURST},
4334 	{REG_BIM_CFG},
4335 	{REG_RX_CFG},
4336 	{REG_HP_CFG},
4337 	{REG_MAC_TX_CFG},
4338 	{REG_MAC_RX_CFG},
4339 	{REG_MAC_CTRL_CFG},
4340 	{REG_MAC_XIF_CFG},
4341 	{REG_MIF_CFG},
4342 	{REG_PCS_CFG},
4343 	{REG_SATURN_PCFG},
4344 	{REG_PCS_MII_STATUS},
4345 	{REG_PCS_STATE_MACHINE},
4346 	{REG_MAC_COLL_EXCESS},
4347 	{REG_MAC_COLL_LATE}
4348 };
4349 #define CAS_REG_LEN 	ARRAY_SIZE(ethtool_register_table)
4350 #define CAS_MAX_REGS 	(sizeof (u32)*CAS_REG_LEN)
4351 
4352 static void cas_read_regs(struct cas *cp, u8 *ptr, int len)
4353 {
4354 	u8 *p;
4355 	int i;
4356 	unsigned long flags;
4357 
4358 	spin_lock_irqsave(&cp->lock, flags);
4359 	for (i = 0, p = ptr; i < len ; i ++, p += sizeof(u32)) {
4360 		u16 hval;
4361 		u32 val;
4362 		if (ethtool_register_table[i].offsets < 0) {
4363 			hval = cas_phy_read(cp,
4364 				    -ethtool_register_table[i].offsets);
4365 			val = hval;
4366 		} else {
4367 			val= readl(cp->regs+ethtool_register_table[i].offsets);
4368 		}
4369 		memcpy(p, (u8 *)&val, sizeof(u32));
4370 	}
4371 	spin_unlock_irqrestore(&cp->lock, flags);
4372 }
4373 
4374 static struct net_device_stats *cas_get_stats(struct net_device *dev)
4375 {
4376 	struct cas *cp = netdev_priv(dev);
4377 	struct net_device_stats *stats = cp->net_stats;
4378 	unsigned long flags;
4379 	int i;
4380 	unsigned long tmp;
4381 
4382 	/* we collate all of the stats into net_stats[N_TX_RING] */
4383 	if (!cp->hw_running)
4384 		return stats + N_TX_RINGS;
4385 
4386 	/* collect outstanding stats */
4387 	/* WTZ: the Cassini spec gives these as 16 bit counters but
4388 	 * stored in 32-bit words.  Added a mask of 0xffff to be safe,
4389 	 * in case the chip somehow puts any garbage in the other bits.
4390 	 * Also, counter usage didn't seem to mach what Adrian did
4391 	 * in the parts of the code that set these quantities. Made
4392 	 * that consistent.
4393 	 */
4394 	spin_lock_irqsave(&cp->stat_lock[N_TX_RINGS], flags);
4395 	stats[N_TX_RINGS].rx_crc_errors +=
4396 	  readl(cp->regs + REG_MAC_FCS_ERR) & 0xffff;
4397 	stats[N_TX_RINGS].rx_frame_errors +=
4398 		readl(cp->regs + REG_MAC_ALIGN_ERR) &0xffff;
4399 	stats[N_TX_RINGS].rx_length_errors +=
4400 		readl(cp->regs + REG_MAC_LEN_ERR) & 0xffff;
4401 #if 1
4402 	tmp = (readl(cp->regs + REG_MAC_COLL_EXCESS) & 0xffff) +
4403 		(readl(cp->regs + REG_MAC_COLL_LATE) & 0xffff);
4404 	stats[N_TX_RINGS].tx_aborted_errors += tmp;
4405 	stats[N_TX_RINGS].collisions +=
4406 	  tmp + (readl(cp->regs + REG_MAC_COLL_NORMAL) & 0xffff);
4407 #else
4408 	stats[N_TX_RINGS].tx_aborted_errors +=
4409 		readl(cp->regs + REG_MAC_COLL_EXCESS);
4410 	stats[N_TX_RINGS].collisions += readl(cp->regs + REG_MAC_COLL_EXCESS) +
4411 		readl(cp->regs + REG_MAC_COLL_LATE);
4412 #endif
4413 	cas_clear_mac_err(cp);
4414 
4415 	/* saved bits that are unique to ring 0 */
4416 	spin_lock(&cp->stat_lock[0]);
4417 	stats[N_TX_RINGS].collisions        += stats[0].collisions;
4418 	stats[N_TX_RINGS].rx_over_errors    += stats[0].rx_over_errors;
4419 	stats[N_TX_RINGS].rx_frame_errors   += stats[0].rx_frame_errors;
4420 	stats[N_TX_RINGS].rx_fifo_errors    += stats[0].rx_fifo_errors;
4421 	stats[N_TX_RINGS].tx_aborted_errors += stats[0].tx_aborted_errors;
4422 	stats[N_TX_RINGS].tx_fifo_errors    += stats[0].tx_fifo_errors;
4423 	spin_unlock(&cp->stat_lock[0]);
4424 
4425 	for (i = 0; i < N_TX_RINGS; i++) {
4426 		spin_lock(&cp->stat_lock[i]);
4427 		stats[N_TX_RINGS].rx_length_errors +=
4428 			stats[i].rx_length_errors;
4429 		stats[N_TX_RINGS].rx_crc_errors += stats[i].rx_crc_errors;
4430 		stats[N_TX_RINGS].rx_packets    += stats[i].rx_packets;
4431 		stats[N_TX_RINGS].tx_packets    += stats[i].tx_packets;
4432 		stats[N_TX_RINGS].rx_bytes      += stats[i].rx_bytes;
4433 		stats[N_TX_RINGS].tx_bytes      += stats[i].tx_bytes;
4434 		stats[N_TX_RINGS].rx_errors     += stats[i].rx_errors;
4435 		stats[N_TX_RINGS].tx_errors     += stats[i].tx_errors;
4436 		stats[N_TX_RINGS].rx_dropped    += stats[i].rx_dropped;
4437 		stats[N_TX_RINGS].tx_dropped    += stats[i].tx_dropped;
4438 		memset(stats + i, 0, sizeof(struct net_device_stats));
4439 		spin_unlock(&cp->stat_lock[i]);
4440 	}
4441 	spin_unlock_irqrestore(&cp->stat_lock[N_TX_RINGS], flags);
4442 	return stats + N_TX_RINGS;
4443 }
4444 
4445 
4446 static void cas_set_multicast(struct net_device *dev)
4447 {
4448 	struct cas *cp = netdev_priv(dev);
4449 	u32 rxcfg, rxcfg_new;
4450 	unsigned long flags;
4451 	int limit = STOP_TRIES;
4452 
4453 	if (!cp->hw_running)
4454 		return;
4455 
4456 	spin_lock_irqsave(&cp->lock, flags);
4457 	rxcfg = readl(cp->regs + REG_MAC_RX_CFG);
4458 
4459 	/* disable RX MAC and wait for completion */
4460 	writel(rxcfg & ~MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
4461 	while (readl(cp->regs + REG_MAC_RX_CFG) & MAC_RX_CFG_EN) {
4462 		if (!limit--)
4463 			break;
4464 		udelay(10);
4465 	}
4466 
4467 	/* disable hash filter and wait for completion */
4468 	limit = STOP_TRIES;
4469 	rxcfg &= ~(MAC_RX_CFG_PROMISC_EN | MAC_RX_CFG_HASH_FILTER_EN);
4470 	writel(rxcfg & ~MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
4471 	while (readl(cp->regs + REG_MAC_RX_CFG) & MAC_RX_CFG_HASH_FILTER_EN) {
4472 		if (!limit--)
4473 			break;
4474 		udelay(10);
4475 	}
4476 
4477 	/* program hash filters */
4478 	cp->mac_rx_cfg = rxcfg_new = cas_setup_multicast(cp);
4479 	rxcfg |= rxcfg_new;
4480 	writel(rxcfg, cp->regs + REG_MAC_RX_CFG);
4481 	spin_unlock_irqrestore(&cp->lock, flags);
4482 }
4483 
4484 static void cas_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
4485 {
4486 	struct cas *cp = netdev_priv(dev);
4487 	strscpy(info->driver, DRV_MODULE_NAME, sizeof(info->driver));
4488 	strscpy(info->version, DRV_MODULE_VERSION, sizeof(info->version));
4489 	strscpy(info->bus_info, pci_name(cp->pdev), sizeof(info->bus_info));
4490 }
4491 
4492 static int cas_get_link_ksettings(struct net_device *dev,
4493 				  struct ethtool_link_ksettings *cmd)
4494 {
4495 	struct cas *cp = netdev_priv(dev);
4496 	u16 bmcr;
4497 	int full_duplex, speed, pause;
4498 	unsigned long flags;
4499 	enum link_state linkstate = link_up;
4500 	u32 supported, advertising;
4501 
4502 	advertising = 0;
4503 	supported = SUPPORTED_Autoneg;
4504 	if (cp->cas_flags & CAS_FLAG_1000MB_CAP) {
4505 		supported |= SUPPORTED_1000baseT_Full;
4506 		advertising |= ADVERTISED_1000baseT_Full;
4507 	}
4508 
4509 	/* Record PHY settings if HW is on. */
4510 	spin_lock_irqsave(&cp->lock, flags);
4511 	bmcr = 0;
4512 	linkstate = cp->lstate;
4513 	if (CAS_PHY_MII(cp->phy_type)) {
4514 		cmd->base.port = PORT_MII;
4515 		cmd->base.phy_address = cp->phy_addr;
4516 		advertising |= ADVERTISED_TP | ADVERTISED_MII |
4517 			ADVERTISED_10baseT_Half |
4518 			ADVERTISED_10baseT_Full |
4519 			ADVERTISED_100baseT_Half |
4520 			ADVERTISED_100baseT_Full;
4521 
4522 		supported |=
4523 			(SUPPORTED_10baseT_Half |
4524 			 SUPPORTED_10baseT_Full |
4525 			 SUPPORTED_100baseT_Half |
4526 			 SUPPORTED_100baseT_Full |
4527 			 SUPPORTED_TP | SUPPORTED_MII);
4528 
4529 		if (cp->hw_running) {
4530 			cas_mif_poll(cp, 0);
4531 			bmcr = cas_phy_read(cp, MII_BMCR);
4532 			cas_read_mii_link_mode(cp, &full_duplex,
4533 					       &speed, &pause);
4534 			cas_mif_poll(cp, 1);
4535 		}
4536 
4537 	} else {
4538 		cmd->base.port = PORT_FIBRE;
4539 		cmd->base.phy_address = 0;
4540 		supported   |= SUPPORTED_FIBRE;
4541 		advertising |= ADVERTISED_FIBRE;
4542 
4543 		if (cp->hw_running) {
4544 			/* pcs uses the same bits as mii */
4545 			bmcr = readl(cp->regs + REG_PCS_MII_CTRL);
4546 			cas_read_pcs_link_mode(cp, &full_duplex,
4547 					       &speed, &pause);
4548 		}
4549 	}
4550 	spin_unlock_irqrestore(&cp->lock, flags);
4551 
4552 	if (bmcr & BMCR_ANENABLE) {
4553 		advertising |= ADVERTISED_Autoneg;
4554 		cmd->base.autoneg = AUTONEG_ENABLE;
4555 		cmd->base.speed =  ((speed == 10) ?
4556 					    SPEED_10 :
4557 					    ((speed == 1000) ?
4558 					     SPEED_1000 : SPEED_100));
4559 		cmd->base.duplex = full_duplex ? DUPLEX_FULL : DUPLEX_HALF;
4560 	} else {
4561 		cmd->base.autoneg = AUTONEG_DISABLE;
4562 		cmd->base.speed = ((bmcr & CAS_BMCR_SPEED1000) ?
4563 					    SPEED_1000 :
4564 					    ((bmcr & BMCR_SPEED100) ?
4565 					     SPEED_100 : SPEED_10));
4566 		cmd->base.duplex = (bmcr & BMCR_FULLDPLX) ?
4567 			DUPLEX_FULL : DUPLEX_HALF;
4568 	}
4569 	if (linkstate != link_up) {
4570 		/* Force these to "unknown" if the link is not up and
4571 		 * autonogotiation in enabled. We can set the link
4572 		 * speed to 0, but not cmd->duplex,
4573 		 * because its legal values are 0 and 1.  Ethtool will
4574 		 * print the value reported in parentheses after the
4575 		 * word "Unknown" for unrecognized values.
4576 		 *
4577 		 * If in forced mode, we report the speed and duplex
4578 		 * settings that we configured.
4579 		 */
4580 		if (cp->link_cntl & BMCR_ANENABLE) {
4581 			cmd->base.speed = 0;
4582 			cmd->base.duplex = 0xff;
4583 		} else {
4584 			cmd->base.speed = SPEED_10;
4585 			if (cp->link_cntl & BMCR_SPEED100) {
4586 				cmd->base.speed = SPEED_100;
4587 			} else if (cp->link_cntl & CAS_BMCR_SPEED1000) {
4588 				cmd->base.speed = SPEED_1000;
4589 			}
4590 			cmd->base.duplex = (cp->link_cntl & BMCR_FULLDPLX) ?
4591 				DUPLEX_FULL : DUPLEX_HALF;
4592 		}
4593 	}
4594 
4595 	ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.supported,
4596 						supported);
4597 	ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.advertising,
4598 						advertising);
4599 
4600 	return 0;
4601 }
4602 
4603 static int cas_set_link_ksettings(struct net_device *dev,
4604 				  const struct ethtool_link_ksettings *cmd)
4605 {
4606 	struct cas *cp = netdev_priv(dev);
4607 	unsigned long flags;
4608 	u32 speed = cmd->base.speed;
4609 
4610 	/* Verify the settings we care about. */
4611 	if (cmd->base.autoneg != AUTONEG_ENABLE &&
4612 	    cmd->base.autoneg != AUTONEG_DISABLE)
4613 		return -EINVAL;
4614 
4615 	if (cmd->base.autoneg == AUTONEG_DISABLE &&
4616 	    ((speed != SPEED_1000 &&
4617 	      speed != SPEED_100 &&
4618 	      speed != SPEED_10) ||
4619 	     (cmd->base.duplex != DUPLEX_HALF &&
4620 	      cmd->base.duplex != DUPLEX_FULL)))
4621 		return -EINVAL;
4622 
4623 	/* Apply settings and restart link process. */
4624 	spin_lock_irqsave(&cp->lock, flags);
4625 	cas_begin_auto_negotiation(cp, cmd);
4626 	spin_unlock_irqrestore(&cp->lock, flags);
4627 	return 0;
4628 }
4629 
4630 static int cas_nway_reset(struct net_device *dev)
4631 {
4632 	struct cas *cp = netdev_priv(dev);
4633 	unsigned long flags;
4634 
4635 	if ((cp->link_cntl & BMCR_ANENABLE) == 0)
4636 		return -EINVAL;
4637 
4638 	/* Restart link process. */
4639 	spin_lock_irqsave(&cp->lock, flags);
4640 	cas_begin_auto_negotiation(cp, NULL);
4641 	spin_unlock_irqrestore(&cp->lock, flags);
4642 
4643 	return 0;
4644 }
4645 
4646 static u32 cas_get_link(struct net_device *dev)
4647 {
4648 	struct cas *cp = netdev_priv(dev);
4649 	return cp->lstate == link_up;
4650 }
4651 
4652 static u32 cas_get_msglevel(struct net_device *dev)
4653 {
4654 	struct cas *cp = netdev_priv(dev);
4655 	return cp->msg_enable;
4656 }
4657 
4658 static void cas_set_msglevel(struct net_device *dev, u32 value)
4659 {
4660 	struct cas *cp = netdev_priv(dev);
4661 	cp->msg_enable = value;
4662 }
4663 
4664 static int cas_get_regs_len(struct net_device *dev)
4665 {
4666 	struct cas *cp = netdev_priv(dev);
4667 	return min_t(int, cp->casreg_len, CAS_MAX_REGS);
4668 }
4669 
4670 static void cas_get_regs(struct net_device *dev, struct ethtool_regs *regs,
4671 			     void *p)
4672 {
4673 	struct cas *cp = netdev_priv(dev);
4674 	regs->version = 0;
4675 	/* cas_read_regs handles locks (cp->lock).  */
4676 	cas_read_regs(cp, p, regs->len / sizeof(u32));
4677 }
4678 
4679 static int cas_get_sset_count(struct net_device *dev, int sset)
4680 {
4681 	switch (sset) {
4682 	case ETH_SS_STATS:
4683 		return CAS_NUM_STAT_KEYS;
4684 	default:
4685 		return -EOPNOTSUPP;
4686 	}
4687 }
4688 
4689 static void cas_get_strings(struct net_device *dev, u32 stringset, u8 *data)
4690 {
4691 	 memcpy(data, &ethtool_cassini_statnames,
4692 					 CAS_NUM_STAT_KEYS * ETH_GSTRING_LEN);
4693 }
4694 
4695 static void cas_get_ethtool_stats(struct net_device *dev,
4696 				      struct ethtool_stats *estats, u64 *data)
4697 {
4698 	struct cas *cp = netdev_priv(dev);
4699 	struct net_device_stats *stats = cas_get_stats(cp->dev);
4700 	int i = 0;
4701 	data[i++] = stats->collisions;
4702 	data[i++] = stats->rx_bytes;
4703 	data[i++] = stats->rx_crc_errors;
4704 	data[i++] = stats->rx_dropped;
4705 	data[i++] = stats->rx_errors;
4706 	data[i++] = stats->rx_fifo_errors;
4707 	data[i++] = stats->rx_frame_errors;
4708 	data[i++] = stats->rx_length_errors;
4709 	data[i++] = stats->rx_over_errors;
4710 	data[i++] = stats->rx_packets;
4711 	data[i++] = stats->tx_aborted_errors;
4712 	data[i++] = stats->tx_bytes;
4713 	data[i++] = stats->tx_dropped;
4714 	data[i++] = stats->tx_errors;
4715 	data[i++] = stats->tx_fifo_errors;
4716 	data[i++] = stats->tx_packets;
4717 	BUG_ON(i != CAS_NUM_STAT_KEYS);
4718 }
4719 
4720 static const struct ethtool_ops cas_ethtool_ops = {
4721 	.get_drvinfo		= cas_get_drvinfo,
4722 	.nway_reset		= cas_nway_reset,
4723 	.get_link		= cas_get_link,
4724 	.get_msglevel		= cas_get_msglevel,
4725 	.set_msglevel		= cas_set_msglevel,
4726 	.get_regs_len		= cas_get_regs_len,
4727 	.get_regs		= cas_get_regs,
4728 	.get_sset_count		= cas_get_sset_count,
4729 	.get_strings		= cas_get_strings,
4730 	.get_ethtool_stats	= cas_get_ethtool_stats,
4731 	.get_link_ksettings	= cas_get_link_ksettings,
4732 	.set_link_ksettings	= cas_set_link_ksettings,
4733 };
4734 
4735 static int cas_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
4736 {
4737 	struct cas *cp = netdev_priv(dev);
4738 	struct mii_ioctl_data *data = if_mii(ifr);
4739 	unsigned long flags;
4740 	int rc = -EOPNOTSUPP;
4741 
4742 	/* Hold the PM mutex while doing ioctl's or we may collide
4743 	 * with open/close and power management and oops.
4744 	 */
4745 	mutex_lock(&cp->pm_mutex);
4746 	switch (cmd) {
4747 	case SIOCGMIIPHY:		/* Get address of MII PHY in use. */
4748 		data->phy_id = cp->phy_addr;
4749 		fallthrough;
4750 
4751 	case SIOCGMIIREG:		/* Read MII PHY register. */
4752 		spin_lock_irqsave(&cp->lock, flags);
4753 		cas_mif_poll(cp, 0);
4754 		data->val_out = cas_phy_read(cp, data->reg_num & 0x1f);
4755 		cas_mif_poll(cp, 1);
4756 		spin_unlock_irqrestore(&cp->lock, flags);
4757 		rc = 0;
4758 		break;
4759 
4760 	case SIOCSMIIREG:		/* Write MII PHY register. */
4761 		spin_lock_irqsave(&cp->lock, flags);
4762 		cas_mif_poll(cp, 0);
4763 		rc = cas_phy_write(cp, data->reg_num & 0x1f, data->val_in);
4764 		cas_mif_poll(cp, 1);
4765 		spin_unlock_irqrestore(&cp->lock, flags);
4766 		break;
4767 	default:
4768 		break;
4769 	}
4770 
4771 	mutex_unlock(&cp->pm_mutex);
4772 	return rc;
4773 }
4774 
4775 /* When this chip sits underneath an Intel 31154 bridge, it is the
4776  * only subordinate device and we can tweak the bridge settings to
4777  * reflect that fact.
4778  */
4779 static void cas_program_bridge(struct pci_dev *cas_pdev)
4780 {
4781 	struct pci_dev *pdev = cas_pdev->bus->self;
4782 	u32 val;
4783 
4784 	if (!pdev)
4785 		return;
4786 
4787 	if (pdev->vendor != 0x8086 || pdev->device != 0x537c)
4788 		return;
4789 
4790 	/* Clear bit 10 (Bus Parking Control) in the Secondary
4791 	 * Arbiter Control/Status Register which lives at offset
4792 	 * 0x41.  Using a 32-bit word read/modify/write at 0x40
4793 	 * is much simpler so that's how we do this.
4794 	 */
4795 	pci_read_config_dword(pdev, 0x40, &val);
4796 	val &= ~0x00040000;
4797 	pci_write_config_dword(pdev, 0x40, val);
4798 
4799 	/* Max out the Multi-Transaction Timer settings since
4800 	 * Cassini is the only device present.
4801 	 *
4802 	 * The register is 16-bit and lives at 0x50.  When the
4803 	 * settings are enabled, it extends the GRANT# signal
4804 	 * for a requestor after a transaction is complete.  This
4805 	 * allows the next request to run without first needing
4806 	 * to negotiate the GRANT# signal back.
4807 	 *
4808 	 * Bits 12:10 define the grant duration:
4809 	 *
4810 	 *	1	--	16 clocks
4811 	 *	2	--	32 clocks
4812 	 *	3	--	64 clocks
4813 	 *	4	--	128 clocks
4814 	 *	5	--	256 clocks
4815 	 *
4816 	 * All other values are illegal.
4817 	 *
4818 	 * Bits 09:00 define which REQ/GNT signal pairs get the
4819 	 * GRANT# signal treatment.  We set them all.
4820 	 */
4821 	pci_write_config_word(pdev, 0x50, (5 << 10) | 0x3ff);
4822 
4823 	/* The Read Prefecth Policy register is 16-bit and sits at
4824 	 * offset 0x52.  It enables a "smart" pre-fetch policy.  We
4825 	 * enable it and max out all of the settings since only one
4826 	 * device is sitting underneath and thus bandwidth sharing is
4827 	 * not an issue.
4828 	 *
4829 	 * The register has several 3 bit fields, which indicates a
4830 	 * multiplier applied to the base amount of prefetching the
4831 	 * chip would do.  These fields are at:
4832 	 *
4833 	 *	15:13	---	ReRead Primary Bus
4834 	 *	12:10	---	FirstRead Primary Bus
4835 	 *	09:07	---	ReRead Secondary Bus
4836 	 *	06:04	---	FirstRead Secondary Bus
4837 	 *
4838 	 * Bits 03:00 control which REQ/GNT pairs the prefetch settings
4839 	 * get enabled on.  Bit 3 is a grouped enabler which controls
4840 	 * all of the REQ/GNT pairs from [8:3].  Bits 2 to 0 control
4841 	 * the individual REQ/GNT pairs [2:0].
4842 	 */
4843 	pci_write_config_word(pdev, 0x52,
4844 			      (0x7 << 13) |
4845 			      (0x7 << 10) |
4846 			      (0x7 <<  7) |
4847 			      (0x7 <<  4) |
4848 			      (0xf <<  0));
4849 
4850 	/* Force cacheline size to 0x8 */
4851 	pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE, 0x08);
4852 
4853 	/* Force latency timer to maximum setting so Cassini can
4854 	 * sit on the bus as long as it likes.
4855 	 */
4856 	pci_write_config_byte(pdev, PCI_LATENCY_TIMER, 0xff);
4857 }
4858 
4859 static const struct net_device_ops cas_netdev_ops = {
4860 	.ndo_open		= cas_open,
4861 	.ndo_stop		= cas_close,
4862 	.ndo_start_xmit		= cas_start_xmit,
4863 	.ndo_get_stats 		= cas_get_stats,
4864 	.ndo_set_rx_mode	= cas_set_multicast,
4865 	.ndo_eth_ioctl		= cas_ioctl,
4866 	.ndo_tx_timeout		= cas_tx_timeout,
4867 	.ndo_change_mtu		= cas_change_mtu,
4868 	.ndo_set_mac_address	= eth_mac_addr,
4869 	.ndo_validate_addr	= eth_validate_addr,
4870 #ifdef CONFIG_NET_POLL_CONTROLLER
4871 	.ndo_poll_controller	= cas_netpoll,
4872 #endif
4873 };
4874 
4875 static int cas_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
4876 {
4877 	static int cas_version_printed = 0;
4878 	unsigned long casreg_len;
4879 	struct net_device *dev;
4880 	struct cas *cp;
4881 	u16 pci_cmd;
4882 	int i, err;
4883 	u8 orig_cacheline_size = 0, cas_cacheline_size = 0;
4884 
4885 	if (cas_version_printed++ == 0)
4886 		pr_info("%s", version);
4887 
4888 	err = pci_enable_device(pdev);
4889 	if (err) {
4890 		dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
4891 		return err;
4892 	}
4893 
4894 	if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
4895 		dev_err(&pdev->dev, "Cannot find proper PCI device "
4896 		       "base address, aborting\n");
4897 		err = -ENODEV;
4898 		goto err_out_disable_pdev;
4899 	}
4900 
4901 	dev = alloc_etherdev(sizeof(*cp));
4902 	if (!dev) {
4903 		err = -ENOMEM;
4904 		goto err_out_disable_pdev;
4905 	}
4906 	SET_NETDEV_DEV(dev, &pdev->dev);
4907 
4908 	err = pci_request_regions(pdev, dev->name);
4909 	if (err) {
4910 		dev_err(&pdev->dev, "Cannot obtain PCI resources, aborting\n");
4911 		goto err_out_free_netdev;
4912 	}
4913 	pci_set_master(pdev);
4914 
4915 	/* we must always turn on parity response or else parity
4916 	 * doesn't get generated properly. disable SERR/PERR as well.
4917 	 * in addition, we want to turn MWI on.
4918 	 */
4919 	pci_read_config_word(pdev, PCI_COMMAND, &pci_cmd);
4920 	pci_cmd &= ~PCI_COMMAND_SERR;
4921 	pci_cmd |= PCI_COMMAND_PARITY;
4922 	pci_write_config_word(pdev, PCI_COMMAND, pci_cmd);
4923 	if (pci_try_set_mwi(pdev))
4924 		pr_warn("Could not enable MWI for %s\n", pci_name(pdev));
4925 
4926 	cas_program_bridge(pdev);
4927 
4928 	/*
4929 	 * On some architectures, the default cache line size set
4930 	 * by pci_try_set_mwi reduces perforamnce.  We have to increase
4931 	 * it for this case.  To start, we'll print some configuration
4932 	 * data.
4933 	 */
4934 #if 1
4935 	pci_read_config_byte(pdev, PCI_CACHE_LINE_SIZE,
4936 			     &orig_cacheline_size);
4937 	if (orig_cacheline_size < CAS_PREF_CACHELINE_SIZE) {
4938 		cas_cacheline_size =
4939 			(CAS_PREF_CACHELINE_SIZE < SMP_CACHE_BYTES) ?
4940 			CAS_PREF_CACHELINE_SIZE : SMP_CACHE_BYTES;
4941 		if (pci_write_config_byte(pdev,
4942 					  PCI_CACHE_LINE_SIZE,
4943 					  cas_cacheline_size)) {
4944 			dev_err(&pdev->dev, "Could not set PCI cache "
4945 			       "line size\n");
4946 			goto err_out_free_res;
4947 		}
4948 	}
4949 #endif
4950 
4951 
4952 	/* Configure DMA attributes. */
4953 	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
4954 	if (err) {
4955 		dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
4956 		goto err_out_free_res;
4957 	}
4958 
4959 	casreg_len = pci_resource_len(pdev, 0);
4960 
4961 	cp = netdev_priv(dev);
4962 	cp->pdev = pdev;
4963 #if 1
4964 	/* A value of 0 indicates we never explicitly set it */
4965 	cp->orig_cacheline_size = cas_cacheline_size ? orig_cacheline_size: 0;
4966 #endif
4967 	cp->dev = dev;
4968 	cp->msg_enable = (cassini_debug < 0) ? CAS_DEF_MSG_ENABLE :
4969 	  cassini_debug;
4970 
4971 #if defined(CONFIG_SPARC)
4972 	cp->of_node = pci_device_to_OF_node(pdev);
4973 #endif
4974 
4975 	cp->link_transition = LINK_TRANSITION_UNKNOWN;
4976 	cp->link_transition_jiffies_valid = 0;
4977 
4978 	spin_lock_init(&cp->lock);
4979 	spin_lock_init(&cp->rx_inuse_lock);
4980 	spin_lock_init(&cp->rx_spare_lock);
4981 	for (i = 0; i < N_TX_RINGS; i++) {
4982 		spin_lock_init(&cp->stat_lock[i]);
4983 		spin_lock_init(&cp->tx_lock[i]);
4984 	}
4985 	spin_lock_init(&cp->stat_lock[N_TX_RINGS]);
4986 	mutex_init(&cp->pm_mutex);
4987 
4988 	timer_setup(&cp->link_timer, cas_link_timer, 0);
4989 
4990 #if 1
4991 	/* Just in case the implementation of atomic operations
4992 	 * change so that an explicit initialization is necessary.
4993 	 */
4994 	atomic_set(&cp->reset_task_pending, 0);
4995 	atomic_set(&cp->reset_task_pending_all, 0);
4996 	atomic_set(&cp->reset_task_pending_spare, 0);
4997 	atomic_set(&cp->reset_task_pending_mtu, 0);
4998 #endif
4999 	INIT_WORK(&cp->reset_task, cas_reset_task);
5000 
5001 	/* Default link parameters */
5002 	if (link_mode >= 0 && link_mode < 6)
5003 		cp->link_cntl = link_modes[link_mode];
5004 	else
5005 		cp->link_cntl = BMCR_ANENABLE;
5006 	cp->lstate = link_down;
5007 	cp->link_transition = LINK_TRANSITION_LINK_DOWN;
5008 	netif_carrier_off(cp->dev);
5009 	cp->timer_ticks = 0;
5010 
5011 	/* give us access to cassini registers */
5012 	cp->regs = pci_iomap(pdev, 0, casreg_len);
5013 	if (!cp->regs) {
5014 		dev_err(&pdev->dev, "Cannot map device registers, aborting\n");
5015 		goto err_out_free_res;
5016 	}
5017 	cp->casreg_len = casreg_len;
5018 
5019 	pci_save_state(pdev);
5020 	cas_check_pci_invariants(cp);
5021 	cas_hard_reset(cp);
5022 	cas_reset(cp, 0);
5023 	if (cas_check_invariants(cp))
5024 		goto err_out_iounmap;
5025 	if (cp->cas_flags & CAS_FLAG_SATURN)
5026 		cas_saturn_firmware_init(cp);
5027 
5028 	cp->init_block =
5029 		dma_alloc_coherent(&pdev->dev, sizeof(struct cas_init_block),
5030 				   &cp->block_dvma, GFP_KERNEL);
5031 	if (!cp->init_block) {
5032 		dev_err(&pdev->dev, "Cannot allocate init block, aborting\n");
5033 		goto err_out_iounmap;
5034 	}
5035 
5036 	for (i = 0; i < N_TX_RINGS; i++)
5037 		cp->init_txds[i] = cp->init_block->txds[i];
5038 
5039 	for (i = 0; i < N_RX_DESC_RINGS; i++)
5040 		cp->init_rxds[i] = cp->init_block->rxds[i];
5041 
5042 	for (i = 0; i < N_RX_COMP_RINGS; i++)
5043 		cp->init_rxcs[i] = cp->init_block->rxcs[i];
5044 
5045 	for (i = 0; i < N_RX_FLOWS; i++)
5046 		skb_queue_head_init(&cp->rx_flows[i]);
5047 
5048 	dev->netdev_ops = &cas_netdev_ops;
5049 	dev->ethtool_ops = &cas_ethtool_ops;
5050 	dev->watchdog_timeo = CAS_TX_TIMEOUT;
5051 
5052 #ifdef USE_NAPI
5053 	netif_napi_add(dev, &cp->napi, cas_poll);
5054 #endif
5055 	dev->irq = pdev->irq;
5056 	dev->dma = 0;
5057 
5058 	/* Cassini features. */
5059 	if ((cp->cas_flags & CAS_FLAG_NO_HW_CSUM) == 0)
5060 		dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG;
5061 
5062 	dev->features |= NETIF_F_HIGHDMA;
5063 
5064 	/* MTU range: 60 - varies or 9000 */
5065 	dev->min_mtu = CAS_MIN_MTU;
5066 	dev->max_mtu = CAS_MAX_MTU;
5067 
5068 	if (register_netdev(dev)) {
5069 		dev_err(&pdev->dev, "Cannot register net device, aborting\n");
5070 		goto err_out_free_consistent;
5071 	}
5072 
5073 	i = readl(cp->regs + REG_BIM_CFG);
5074 	netdev_info(dev, "Sun Cassini%s (%sbit/%sMHz PCI/%s) Ethernet[%d] %pM\n",
5075 		    (cp->cas_flags & CAS_FLAG_REG_PLUS) ? "+" : "",
5076 		    (i & BIM_CFG_32BIT) ? "32" : "64",
5077 		    (i & BIM_CFG_66MHZ) ? "66" : "33",
5078 		    (cp->phy_type == CAS_PHY_SERDES) ? "Fi" : "Cu", pdev->irq,
5079 		    dev->dev_addr);
5080 
5081 	pci_set_drvdata(pdev, dev);
5082 	cp->hw_running = 1;
5083 	cas_entropy_reset(cp);
5084 	cas_phy_init(cp);
5085 	cas_begin_auto_negotiation(cp, NULL);
5086 	return 0;
5087 
5088 err_out_free_consistent:
5089 	dma_free_coherent(&pdev->dev, sizeof(struct cas_init_block),
5090 			  cp->init_block, cp->block_dvma);
5091 
5092 err_out_iounmap:
5093 	mutex_lock(&cp->pm_mutex);
5094 	if (cp->hw_running)
5095 		cas_shutdown(cp);
5096 	mutex_unlock(&cp->pm_mutex);
5097 
5098 	pci_iounmap(pdev, cp->regs);
5099 
5100 
5101 err_out_free_res:
5102 	pci_release_regions(pdev);
5103 
5104 	/* Try to restore it in case the error occurred after we
5105 	 * set it.
5106 	 */
5107 	pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE, orig_cacheline_size);
5108 
5109 err_out_free_netdev:
5110 	free_netdev(dev);
5111 
5112 err_out_disable_pdev:
5113 	pci_disable_device(pdev);
5114 	return -ENODEV;
5115 }
5116 
5117 static void cas_remove_one(struct pci_dev *pdev)
5118 {
5119 	struct net_device *dev = pci_get_drvdata(pdev);
5120 	struct cas *cp;
5121 	if (!dev)
5122 		return;
5123 
5124 	cp = netdev_priv(dev);
5125 	unregister_netdev(dev);
5126 
5127 	vfree(cp->fw_data);
5128 
5129 	mutex_lock(&cp->pm_mutex);
5130 	cancel_work_sync(&cp->reset_task);
5131 	if (cp->hw_running)
5132 		cas_shutdown(cp);
5133 	mutex_unlock(&cp->pm_mutex);
5134 
5135 #if 1
5136 	if (cp->orig_cacheline_size) {
5137 		/* Restore the cache line size if we had modified
5138 		 * it.
5139 		 */
5140 		pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE,
5141 				      cp->orig_cacheline_size);
5142 	}
5143 #endif
5144 	dma_free_coherent(&pdev->dev, sizeof(struct cas_init_block),
5145 			  cp->init_block, cp->block_dvma);
5146 	pci_iounmap(pdev, cp->regs);
5147 	free_netdev(dev);
5148 	pci_release_regions(pdev);
5149 	pci_disable_device(pdev);
5150 }
5151 
5152 static int __maybe_unused cas_suspend(struct device *dev_d)
5153 {
5154 	struct net_device *dev = dev_get_drvdata(dev_d);
5155 	struct cas *cp = netdev_priv(dev);
5156 	unsigned long flags;
5157 
5158 	mutex_lock(&cp->pm_mutex);
5159 
5160 	/* If the driver is opened, we stop the DMA */
5161 	if (cp->opened) {
5162 		netif_device_detach(dev);
5163 
5164 		cas_lock_all_save(cp, flags);
5165 
5166 		/* We can set the second arg of cas_reset to 0
5167 		 * because on resume, we'll call cas_init_hw with
5168 		 * its second arg set so that autonegotiation is
5169 		 * restarted.
5170 		 */
5171 		cas_reset(cp, 0);
5172 		cas_clean_rings(cp);
5173 		cas_unlock_all_restore(cp, flags);
5174 	}
5175 
5176 	if (cp->hw_running)
5177 		cas_shutdown(cp);
5178 	mutex_unlock(&cp->pm_mutex);
5179 
5180 	return 0;
5181 }
5182 
5183 static int __maybe_unused cas_resume(struct device *dev_d)
5184 {
5185 	struct net_device *dev = dev_get_drvdata(dev_d);
5186 	struct cas *cp = netdev_priv(dev);
5187 
5188 	netdev_info(dev, "resuming\n");
5189 
5190 	mutex_lock(&cp->pm_mutex);
5191 	cas_hard_reset(cp);
5192 	if (cp->opened) {
5193 		unsigned long flags;
5194 		cas_lock_all_save(cp, flags);
5195 		cas_reset(cp, 0);
5196 		cp->hw_running = 1;
5197 		cas_clean_rings(cp);
5198 		cas_init_hw(cp, 1);
5199 		cas_unlock_all_restore(cp, flags);
5200 
5201 		netif_device_attach(dev);
5202 	}
5203 	mutex_unlock(&cp->pm_mutex);
5204 	return 0;
5205 }
5206 
5207 static SIMPLE_DEV_PM_OPS(cas_pm_ops, cas_suspend, cas_resume);
5208 
5209 static struct pci_driver cas_driver = {
5210 	.name		= DRV_MODULE_NAME,
5211 	.id_table	= cas_pci_tbl,
5212 	.probe		= cas_init_one,
5213 	.remove		= cas_remove_one,
5214 	.driver.pm	= &cas_pm_ops,
5215 };
5216 
5217 static int __init cas_init(void)
5218 {
5219 	if (linkdown_timeout > 0)
5220 		link_transition_timeout = linkdown_timeout * HZ;
5221 	else
5222 		link_transition_timeout = 0;
5223 
5224 	return pci_register_driver(&cas_driver);
5225 }
5226 
5227 static void __exit cas_cleanup(void)
5228 {
5229 	pci_unregister_driver(&cas_driver);
5230 }
5231 
5232 module_init(cas_init);
5233 module_exit(cas_cleanup);
5234