1/* cassini.c: Sun Microsystems Cassini(+) ethernet driver.
2 *
3 * Copyright (C) 2004 Sun Microsystems Inc.
4 * Copyright (C) 2003 Adrian Sun (asun@darksunrising.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation; either version 2 of the
9 * License, or (at your option) any later version.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
19 * 02111-1307, USA.
20 *
21 * This driver uses the sungem driver (c) David Miller
22 * (davem@redhat.com) as its basis.
23 *
24 * The cassini chip has a number of features that distinguish it from
25 * the gem chip:
26 *  4 transmit descriptor rings that are used for either QoS (VLAN) or
27 *      load balancing (non-VLAN mode)
28 *  batching of multiple packets
29 *  multiple CPU dispatching
30 *  page-based RX descriptor engine with separate completion rings
31 *  Gigabit support (GMII and PCS interface)
32 *  MIF link up/down detection works
33 *
34 * RX is handled by page sized buffers that are attached as fragments to
35 * the skb. here's what's done:
36 *  -- driver allocates pages at a time and keeps reference counts
37 *     on them.
38 *  -- the upper protocol layers assume that the header is in the skb
39 *     itself. as a result, cassini will copy a small amount (64 bytes)
40 *     to make them happy.
41 *  -- driver appends the rest of the data pages as frags to skbuffs
42 *     and increments the reference count
43 *  -- on page reclamation, the driver swaps the page with a spare page.
44 *     if that page is still in use, it frees its reference to that page,
45 *     and allocates a new page for use. otherwise, it just recycles the
46 *     the page.
47 *
48 * NOTE: cassini can parse the header. however, it's not worth it
49 *       as long as the network stack requires a header copy.
50 *
51 * TX has 4 queues. currently these queues are used in a round-robin
52 * fashion for load balancing. They can also be used for QoS. for that
53 * to work, however, QoS information needs to be exposed down to the driver
54 * level so that subqueues get targeted to particular transmit rings.
55 * alternatively, the queues can be configured via use of the all-purpose
56 * ioctl.
57 *
58 * RX DATA: the rx completion ring has all the info, but the rx desc
59 * ring has all of the data. RX can conceivably come in under multiple
60 * interrupts, but the INT# assignment needs to be set up properly by
61 * the BIOS and conveyed to the driver. PCI BIOSes don't know how to do
62 * that. also, the two descriptor rings are designed to distinguish between
63 * encrypted and non-encrypted packets, but we use them for buffering
64 * instead.
65 *
66 * by default, the selective clear mask is set up to process rx packets.
67 */
68
69#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
70
71#include <linux/module.h>
72#include <linux/kernel.h>
73#include <linux/types.h>
74#include <linux/compiler.h>
75#include <linux/slab.h>
76#include <linux/delay.h>
77#include <linux/init.h>
78#include <linux/interrupt.h>
79#include <linux/vmalloc.h>
80#include <linux/ioport.h>
81#include <linux/pci.h>
82#include <linux/mm.h>
83#include <linux/highmem.h>
84#include <linux/list.h>
85#include <linux/dma-mapping.h>
86
87#include <linux/netdevice.h>
88#include <linux/etherdevice.h>
89#include <linux/skbuff.h>
90#include <linux/ethtool.h>
91#include <linux/crc32.h>
92#include <linux/random.h>
93#include <linux/mii.h>
94#include <linux/ip.h>
95#include <linux/tcp.h>
96#include <linux/mutex.h>
97#include <linux/firmware.h>
98
99#include <net/checksum.h>
100
101#include <linux/atomic.h>
102#include <asm/io.h>
103#include <asm/byteorder.h>
104#include <asm/uaccess.h>
105
106#define cas_page_map(x)      kmap_atomic((x))
107#define cas_page_unmap(x)    kunmap_atomic((x))
108#define CAS_NCPUS            num_online_cpus()
109
110#define cas_skb_release(x)  netif_rx(x)
111
112/* select which firmware to use */
113#define USE_HP_WORKAROUND
114#define HP_WORKAROUND_DEFAULT /* select which firmware to use as default */
115#define CAS_HP_ALT_FIRMWARE   cas_prog_null /* alternate firmware */
116
117#include "cassini.h"
118
119#define USE_TX_COMPWB      /* use completion writeback registers */
120#define USE_CSMA_CD_PROTO  /* standard CSMA/CD */
121#define USE_RX_BLANK       /* hw interrupt mitigation */
122#undef USE_ENTROPY_DEV     /* don't test for entropy device */
123
124/* NOTE: these aren't useable unless PCI interrupts can be assigned.
125 * also, we need to make cp->lock finer-grained.
126 */
127#undef  USE_PCI_INTB
128#undef  USE_PCI_INTC
129#undef  USE_PCI_INTD
130#undef  USE_QOS
131
132#undef  USE_VPD_DEBUG       /* debug vpd information if defined */
133
134/* rx processing options */
135#define USE_PAGE_ORDER      /* specify to allocate large rx pages */
136#define RX_DONT_BATCH  0    /* if 1, don't batch flows */
137#define RX_COPY_ALWAYS 0    /* if 0, use frags */
138#define RX_COPY_MIN    64   /* copy a little to make upper layers happy */
139#undef  RX_COUNT_BUFFERS    /* define to calculate RX buffer stats */
140
141#define DRV_MODULE_NAME		"cassini"
142#define DRV_MODULE_VERSION	"1.6"
143#define DRV_MODULE_RELDATE	"21 May 2008"
144
145#define CAS_DEF_MSG_ENABLE	  \
146	(NETIF_MSG_DRV		| \
147	 NETIF_MSG_PROBE	| \
148	 NETIF_MSG_LINK		| \
149	 NETIF_MSG_TIMER	| \
150	 NETIF_MSG_IFDOWN	| \
151	 NETIF_MSG_IFUP		| \
152	 NETIF_MSG_RX_ERR	| \
153	 NETIF_MSG_TX_ERR)
154
155/* length of time before we decide the hardware is borked,
156 * and dev->tx_timeout() should be called to fix the problem
157 */
158#define CAS_TX_TIMEOUT			(HZ)
159#define CAS_LINK_TIMEOUT                (22*HZ/10)
160#define CAS_LINK_FAST_TIMEOUT           (1)
161
162/* timeout values for state changing. these specify the number
163 * of 10us delays to be used before giving up.
164 */
165#define STOP_TRIES_PHY 1000
166#define STOP_TRIES     5000
167
168/* specify a minimum frame size to deal with some fifo issues
169 * max mtu == 2 * page size - ethernet header - 64 - swivel =
170 *            2 * page_size - 0x50
171 */
172#define CAS_MIN_FRAME			97
173#define CAS_1000MB_MIN_FRAME            255
174#define CAS_MIN_MTU                     60
175#define CAS_MAX_MTU                     min(((cp->page_size << 1) - 0x50), 9000)
176
177#if 1
178/*
179 * Eliminate these and use separate atomic counters for each, to
180 * avoid a race condition.
181 */
182#else
183#define CAS_RESET_MTU                   1
184#define CAS_RESET_ALL                   2
185#define CAS_RESET_SPARE                 3
186#endif
187
188static char version[] __devinitdata =
189	DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
190
191static int cassini_debug = -1;	/* -1 == use CAS_DEF_MSG_ENABLE as value */
192static int link_mode;
193
194MODULE_AUTHOR("Adrian Sun (asun@darksunrising.com)");
195MODULE_DESCRIPTION("Sun Cassini(+) ethernet driver");
196MODULE_LICENSE("GPL");
197MODULE_FIRMWARE("sun/cassini.bin");
198module_param(cassini_debug, int, 0);
199MODULE_PARM_DESC(cassini_debug, "Cassini bitmapped debugging message enable value");
200module_param(link_mode, int, 0);
201MODULE_PARM_DESC(link_mode, "default link mode");
202
203/*
204 * Work around for a PCS bug in which the link goes down due to the chip
205 * being confused and never showing a link status of "up."
206 */
207#define DEFAULT_LINKDOWN_TIMEOUT 5
208/*
209 * Value in seconds, for user input.
210 */
211static int linkdown_timeout = DEFAULT_LINKDOWN_TIMEOUT;
212module_param(linkdown_timeout, int, 0);
213MODULE_PARM_DESC(linkdown_timeout,
214"min reset interval in sec. for PCS linkdown issue; disabled if not positive");
215
216/*
217 * value in 'ticks' (units used by jiffies). Set when we init the
218 * module because 'HZ' in actually a function call on some flavors of
219 * Linux.  This will default to DEFAULT_LINKDOWN_TIMEOUT * HZ.
220 */
221static int link_transition_timeout;
222
223
224
225static u16 link_modes[] __devinitdata = {
226	BMCR_ANENABLE,			 /* 0 : autoneg */
227	0,				 /* 1 : 10bt half duplex */
228	BMCR_SPEED100,			 /* 2 : 100bt half duplex */
229	BMCR_FULLDPLX,			 /* 3 : 10bt full duplex */
230	BMCR_SPEED100|BMCR_FULLDPLX,	 /* 4 : 100bt full duplex */
231	CAS_BMCR_SPEED1000|BMCR_FULLDPLX /* 5 : 1000bt full duplex */
232};
233
234static DEFINE_PCI_DEVICE_TABLE(cas_pci_tbl) = {
235	{ PCI_VENDOR_ID_SUN, PCI_DEVICE_ID_SUN_CASSINI,
236	  PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0UL },
237	{ PCI_VENDOR_ID_NS, PCI_DEVICE_ID_NS_SATURN,
238	  PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0UL },
239	{ 0, }
240};
241
242MODULE_DEVICE_TABLE(pci, cas_pci_tbl);
243
244static void cas_set_link_modes(struct cas *cp);
245
246static inline void cas_lock_tx(struct cas *cp)
247{
248	int i;
249
250	for (i = 0; i < N_TX_RINGS; i++)
251		spin_lock(&cp->tx_lock[i]);
252}
253
254static inline void cas_lock_all(struct cas *cp)
255{
256	spin_lock_irq(&cp->lock);
257	cas_lock_tx(cp);
258}
259
260/* WTZ: QA was finding deadlock problems with the previous
261 * versions after long test runs with multiple cards per machine.
262 * See if replacing cas_lock_all with safer versions helps. The
263 * symptoms QA is reporting match those we'd expect if interrupts
264 * aren't being properly restored, and we fixed a previous deadlock
265 * with similar symptoms by using save/restore versions in other
266 * places.
267 */
268#define cas_lock_all_save(cp, flags) \
269do { \
270	struct cas *xxxcp = (cp); \
271	spin_lock_irqsave(&xxxcp->lock, flags); \
272	cas_lock_tx(xxxcp); \
273} while (0)
274
275static inline void cas_unlock_tx(struct cas *cp)
276{
277	int i;
278
279	for (i = N_TX_RINGS; i > 0; i--)
280		spin_unlock(&cp->tx_lock[i - 1]);
281}
282
283static inline void cas_unlock_all(struct cas *cp)
284{
285	cas_unlock_tx(cp);
286	spin_unlock_irq(&cp->lock);
287}
288
289#define cas_unlock_all_restore(cp, flags) \
290do { \
291	struct cas *xxxcp = (cp); \
292	cas_unlock_tx(xxxcp); \
293	spin_unlock_irqrestore(&xxxcp->lock, flags); \
294} while (0)
295
296static void cas_disable_irq(struct cas *cp, const int ring)
297{
298	/* Make sure we won't get any more interrupts */
299	if (ring == 0) {
300		writel(0xFFFFFFFF, cp->regs + REG_INTR_MASK);
301		return;
302	}
303
304	/* disable completion interrupts and selectively mask */
305	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
306		switch (ring) {
307#if defined (USE_PCI_INTB) || defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
308#ifdef USE_PCI_INTB
309		case 1:
310#endif
311#ifdef USE_PCI_INTC
312		case 2:
313#endif
314#ifdef USE_PCI_INTD
315		case 3:
316#endif
317			writel(INTRN_MASK_CLEAR_ALL | INTRN_MASK_RX_EN,
318			       cp->regs + REG_PLUS_INTRN_MASK(ring));
319			break;
320#endif
321		default:
322			writel(INTRN_MASK_CLEAR_ALL, cp->regs +
323			       REG_PLUS_INTRN_MASK(ring));
324			break;
325		}
326	}
327}
328
329static inline void cas_mask_intr(struct cas *cp)
330{
331	int i;
332
333	for (i = 0; i < N_RX_COMP_RINGS; i++)
334		cas_disable_irq(cp, i);
335}
336
337static void cas_enable_irq(struct cas *cp, const int ring)
338{
339	if (ring == 0) { /* all but TX_DONE */
340		writel(INTR_TX_DONE, cp->regs + REG_INTR_MASK);
341		return;
342	}
343
344	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
345		switch (ring) {
346#if defined (USE_PCI_INTB) || defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
347#ifdef USE_PCI_INTB
348		case 1:
349#endif
350#ifdef USE_PCI_INTC
351		case 2:
352#endif
353#ifdef USE_PCI_INTD
354		case 3:
355#endif
356			writel(INTRN_MASK_RX_EN, cp->regs +
357			       REG_PLUS_INTRN_MASK(ring));
358			break;
359#endif
360		default:
361			break;
362		}
363	}
364}
365
366static inline void cas_unmask_intr(struct cas *cp)
367{
368	int i;
369
370	for (i = 0; i < N_RX_COMP_RINGS; i++)
371		cas_enable_irq(cp, i);
372}
373
374static inline void cas_entropy_gather(struct cas *cp)
375{
376#ifdef USE_ENTROPY_DEV
377	if ((cp->cas_flags & CAS_FLAG_ENTROPY_DEV) == 0)
378		return;
379
380	batch_entropy_store(readl(cp->regs + REG_ENTROPY_IV),
381			    readl(cp->regs + REG_ENTROPY_IV),
382			    sizeof(uint64_t)*8);
383#endif
384}
385
386static inline void cas_entropy_reset(struct cas *cp)
387{
388#ifdef USE_ENTROPY_DEV
389	if ((cp->cas_flags & CAS_FLAG_ENTROPY_DEV) == 0)
390		return;
391
392	writel(BIM_LOCAL_DEV_PAD | BIM_LOCAL_DEV_PROM | BIM_LOCAL_DEV_EXT,
393	       cp->regs + REG_BIM_LOCAL_DEV_EN);
394	writeb(ENTROPY_RESET_STC_MODE, cp->regs + REG_ENTROPY_RESET);
395	writeb(0x55, cp->regs + REG_ENTROPY_RAND_REG);
396
397	/* if we read back 0x0, we don't have an entropy device */
398	if (readb(cp->regs + REG_ENTROPY_RAND_REG) == 0)
399		cp->cas_flags &= ~CAS_FLAG_ENTROPY_DEV;
400#endif
401}
402
403/* access to the phy. the following assumes that we've initialized the MIF to
404 * be in frame rather than bit-bang mode
405 */
406static u16 cas_phy_read(struct cas *cp, int reg)
407{
408	u32 cmd;
409	int limit = STOP_TRIES_PHY;
410
411	cmd = MIF_FRAME_ST | MIF_FRAME_OP_READ;
412	cmd |= CAS_BASE(MIF_FRAME_PHY_ADDR, cp->phy_addr);
413	cmd |= CAS_BASE(MIF_FRAME_REG_ADDR, reg);
414	cmd |= MIF_FRAME_TURN_AROUND_MSB;
415	writel(cmd, cp->regs + REG_MIF_FRAME);
416
417	/* poll for completion */
418	while (limit-- > 0) {
419		udelay(10);
420		cmd = readl(cp->regs + REG_MIF_FRAME);
421		if (cmd & MIF_FRAME_TURN_AROUND_LSB)
422			return cmd & MIF_FRAME_DATA_MASK;
423	}
424	return 0xFFFF; /* -1 */
425}
426
427static int cas_phy_write(struct cas *cp, int reg, u16 val)
428{
429	int limit = STOP_TRIES_PHY;
430	u32 cmd;
431
432	cmd = MIF_FRAME_ST | MIF_FRAME_OP_WRITE;
433	cmd |= CAS_BASE(MIF_FRAME_PHY_ADDR, cp->phy_addr);
434	cmd |= CAS_BASE(MIF_FRAME_REG_ADDR, reg);
435	cmd |= MIF_FRAME_TURN_AROUND_MSB;
436	cmd |= val & MIF_FRAME_DATA_MASK;
437	writel(cmd, cp->regs + REG_MIF_FRAME);
438
439	/* poll for completion */
440	while (limit-- > 0) {
441		udelay(10);
442		cmd = readl(cp->regs + REG_MIF_FRAME);
443		if (cmd & MIF_FRAME_TURN_AROUND_LSB)
444			return 0;
445	}
446	return -1;
447}
448
449static void cas_phy_powerup(struct cas *cp)
450{
451	u16 ctl = cas_phy_read(cp, MII_BMCR);
452
453	if ((ctl & BMCR_PDOWN) == 0)
454		return;
455	ctl &= ~BMCR_PDOWN;
456	cas_phy_write(cp, MII_BMCR, ctl);
457}
458
459static void cas_phy_powerdown(struct cas *cp)
460{
461	u16 ctl = cas_phy_read(cp, MII_BMCR);
462
463	if (ctl & BMCR_PDOWN)
464		return;
465	ctl |= BMCR_PDOWN;
466	cas_phy_write(cp, MII_BMCR, ctl);
467}
468
469/* cp->lock held. note: the last put_page will free the buffer */
470static int cas_page_free(struct cas *cp, cas_page_t *page)
471{
472	pci_unmap_page(cp->pdev, page->dma_addr, cp->page_size,
473		       PCI_DMA_FROMDEVICE);
474	__free_pages(page->buffer, cp->page_order);
475	kfree(page);
476	return 0;
477}
478
479#ifdef RX_COUNT_BUFFERS
480#define RX_USED_ADD(x, y)       ((x)->used += (y))
481#define RX_USED_SET(x, y)       ((x)->used  = (y))
482#else
483#define RX_USED_ADD(x, y)
484#define RX_USED_SET(x, y)
485#endif
486
487/* local page allocation routines for the receive buffers. jumbo pages
488 * require at least 8K contiguous and 8K aligned buffers.
489 */
490static cas_page_t *cas_page_alloc(struct cas *cp, const gfp_t flags)
491{
492	cas_page_t *page;
493
494	page = kmalloc(sizeof(cas_page_t), flags);
495	if (!page)
496		return NULL;
497
498	INIT_LIST_HEAD(&page->list);
499	RX_USED_SET(page, 0);
500	page->buffer = alloc_pages(flags, cp->page_order);
501	if (!page->buffer)
502		goto page_err;
503	page->dma_addr = pci_map_page(cp->pdev, page->buffer, 0,
504				      cp->page_size, PCI_DMA_FROMDEVICE);
505	return page;
506
507page_err:
508	kfree(page);
509	return NULL;
510}
511
512/* initialize spare pool of rx buffers, but allocate during the open */
513static void cas_spare_init(struct cas *cp)
514{
515  	spin_lock(&cp->rx_inuse_lock);
516	INIT_LIST_HEAD(&cp->rx_inuse_list);
517	spin_unlock(&cp->rx_inuse_lock);
518
519	spin_lock(&cp->rx_spare_lock);
520	INIT_LIST_HEAD(&cp->rx_spare_list);
521	cp->rx_spares_needed = RX_SPARE_COUNT;
522	spin_unlock(&cp->rx_spare_lock);
523}
524
525/* used on close. free all the spare buffers. */
526static void cas_spare_free(struct cas *cp)
527{
528	struct list_head list, *elem, *tmp;
529
530	/* free spare buffers */
531	INIT_LIST_HEAD(&list);
532	spin_lock(&cp->rx_spare_lock);
533	list_splice_init(&cp->rx_spare_list, &list);
534	spin_unlock(&cp->rx_spare_lock);
535	list_for_each_safe(elem, tmp, &list) {
536		cas_page_free(cp, list_entry(elem, cas_page_t, list));
537	}
538
539	INIT_LIST_HEAD(&list);
540#if 1
541	/*
542	 * Looks like Adrian had protected this with a different
543	 * lock than used everywhere else to manipulate this list.
544	 */
545	spin_lock(&cp->rx_inuse_lock);
546	list_splice_init(&cp->rx_inuse_list, &list);
547	spin_unlock(&cp->rx_inuse_lock);
548#else
549	spin_lock(&cp->rx_spare_lock);
550	list_splice_init(&cp->rx_inuse_list, &list);
551	spin_unlock(&cp->rx_spare_lock);
552#endif
553	list_for_each_safe(elem, tmp, &list) {
554		cas_page_free(cp, list_entry(elem, cas_page_t, list));
555	}
556}
557
558/* replenish spares if needed */
559static void cas_spare_recover(struct cas *cp, const gfp_t flags)
560{
561	struct list_head list, *elem, *tmp;
562	int needed, i;
563
564	/* check inuse list. if we don't need any more free buffers,
565	 * just free it
566	 */
567
568	/* make a local copy of the list */
569	INIT_LIST_HEAD(&list);
570	spin_lock(&cp->rx_inuse_lock);
571	list_splice_init(&cp->rx_inuse_list, &list);
572	spin_unlock(&cp->rx_inuse_lock);
573
574	list_for_each_safe(elem, tmp, &list) {
575		cas_page_t *page = list_entry(elem, cas_page_t, list);
576
577		/*
578		 * With the lockless pagecache, cassini buffering scheme gets
579		 * slightly less accurate: we might find that a page has an
580		 * elevated reference count here, due to a speculative ref,
581		 * and skip it as in-use. Ideally we would be able to reclaim
582		 * it. However this would be such a rare case, it doesn't
583		 * matter too much as we should pick it up the next time round.
584		 *
585		 * Importantly, if we find that the page has a refcount of 1
586		 * here (our refcount), then we know it is definitely not inuse
587		 * so we can reuse it.
588		 */
589		if (page_count(page->buffer) > 1)
590			continue;
591
592		list_del(elem);
593		spin_lock(&cp->rx_spare_lock);
594		if (cp->rx_spares_needed > 0) {
595			list_add(elem, &cp->rx_spare_list);
596			cp->rx_spares_needed--;
597			spin_unlock(&cp->rx_spare_lock);
598		} else {
599			spin_unlock(&cp->rx_spare_lock);
600			cas_page_free(cp, page);
601		}
602	}
603
604	/* put any inuse buffers back on the list */
605	if (!list_empty(&list)) {
606		spin_lock(&cp->rx_inuse_lock);
607		list_splice(&list, &cp->rx_inuse_list);
608		spin_unlock(&cp->rx_inuse_lock);
609	}
610
611	spin_lock(&cp->rx_spare_lock);
612	needed = cp->rx_spares_needed;
613	spin_unlock(&cp->rx_spare_lock);
614	if (!needed)
615		return;
616
617	/* we still need spares, so try to allocate some */
618	INIT_LIST_HEAD(&list);
619	i = 0;
620	while (i < needed) {
621		cas_page_t *spare = cas_page_alloc(cp, flags);
622		if (!spare)
623			break;
624		list_add(&spare->list, &list);
625		i++;
626	}
627
628	spin_lock(&cp->rx_spare_lock);
629	list_splice(&list, &cp->rx_spare_list);
630	cp->rx_spares_needed -= i;
631	spin_unlock(&cp->rx_spare_lock);
632}
633
634/* pull a page from the list. */
635static cas_page_t *cas_page_dequeue(struct cas *cp)
636{
637	struct list_head *entry;
638	int recover;
639
640	spin_lock(&cp->rx_spare_lock);
641	if (list_empty(&cp->rx_spare_list)) {
642		/* try to do a quick recovery */
643		spin_unlock(&cp->rx_spare_lock);
644		cas_spare_recover(cp, GFP_ATOMIC);
645		spin_lock(&cp->rx_spare_lock);
646		if (list_empty(&cp->rx_spare_list)) {
647			netif_err(cp, rx_err, cp->dev,
648				  "no spare buffers available\n");
649			spin_unlock(&cp->rx_spare_lock);
650			return NULL;
651		}
652	}
653
654	entry = cp->rx_spare_list.next;
655	list_del(entry);
656	recover = ++cp->rx_spares_needed;
657	spin_unlock(&cp->rx_spare_lock);
658
659	/* trigger the timer to do the recovery */
660	if ((recover & (RX_SPARE_RECOVER_VAL - 1)) == 0) {
661#if 1
662		atomic_inc(&cp->reset_task_pending);
663		atomic_inc(&cp->reset_task_pending_spare);
664		schedule_work(&cp->reset_task);
665#else
666		atomic_set(&cp->reset_task_pending, CAS_RESET_SPARE);
667		schedule_work(&cp->reset_task);
668#endif
669	}
670	return list_entry(entry, cas_page_t, list);
671}
672
673
674static void cas_mif_poll(struct cas *cp, const int enable)
675{
676	u32 cfg;
677
678	cfg  = readl(cp->regs + REG_MIF_CFG);
679	cfg &= (MIF_CFG_MDIO_0 | MIF_CFG_MDIO_1);
680
681	if (cp->phy_type & CAS_PHY_MII_MDIO1)
682		cfg |= MIF_CFG_PHY_SELECT;
683
684	/* poll and interrupt on link status change. */
685	if (enable) {
686		cfg |= MIF_CFG_POLL_EN;
687		cfg |= CAS_BASE(MIF_CFG_POLL_REG, MII_BMSR);
688		cfg |= CAS_BASE(MIF_CFG_POLL_PHY, cp->phy_addr);
689	}
690	writel((enable) ? ~(BMSR_LSTATUS | BMSR_ANEGCOMPLETE) : 0xFFFF,
691	       cp->regs + REG_MIF_MASK);
692	writel(cfg, cp->regs + REG_MIF_CFG);
693}
694
695/* Must be invoked under cp->lock */
696static void cas_begin_auto_negotiation(struct cas *cp, struct ethtool_cmd *ep)
697{
698	u16 ctl;
699#if 1
700	int lcntl;
701	int changed = 0;
702	int oldstate = cp->lstate;
703	int link_was_not_down = !(oldstate == link_down);
704#endif
705	/* Setup link parameters */
706	if (!ep)
707		goto start_aneg;
708	lcntl = cp->link_cntl;
709	if (ep->autoneg == AUTONEG_ENABLE)
710		cp->link_cntl = BMCR_ANENABLE;
711	else {
712		u32 speed = ethtool_cmd_speed(ep);
713		cp->link_cntl = 0;
714		if (speed == SPEED_100)
715			cp->link_cntl |= BMCR_SPEED100;
716		else if (speed == SPEED_1000)
717			cp->link_cntl |= CAS_BMCR_SPEED1000;
718		if (ep->duplex == DUPLEX_FULL)
719			cp->link_cntl |= BMCR_FULLDPLX;
720	}
721#if 1
722	changed = (lcntl != cp->link_cntl);
723#endif
724start_aneg:
725	if (cp->lstate == link_up) {
726		netdev_info(cp->dev, "PCS link down\n");
727	} else {
728		if (changed) {
729			netdev_info(cp->dev, "link configuration changed\n");
730		}
731	}
732	cp->lstate = link_down;
733	cp->link_transition = LINK_TRANSITION_LINK_DOWN;
734	if (!cp->hw_running)
735		return;
736#if 1
737	/*
738	 * WTZ: If the old state was link_up, we turn off the carrier
739	 * to replicate everything we do elsewhere on a link-down
740	 * event when we were already in a link-up state..
741	 */
742	if (oldstate == link_up)
743		netif_carrier_off(cp->dev);
744	if (changed  && link_was_not_down) {
745		/*
746		 * WTZ: This branch will simply schedule a full reset after
747		 * we explicitly changed link modes in an ioctl. See if this
748		 * fixes the link-problems we were having for forced mode.
749		 */
750		atomic_inc(&cp->reset_task_pending);
751		atomic_inc(&cp->reset_task_pending_all);
752		schedule_work(&cp->reset_task);
753		cp->timer_ticks = 0;
754		mod_timer(&cp->link_timer, jiffies + CAS_LINK_TIMEOUT);
755		return;
756	}
757#endif
758	if (cp->phy_type & CAS_PHY_SERDES) {
759		u32 val = readl(cp->regs + REG_PCS_MII_CTRL);
760
761		if (cp->link_cntl & BMCR_ANENABLE) {
762			val |= (PCS_MII_RESTART_AUTONEG | PCS_MII_AUTONEG_EN);
763			cp->lstate = link_aneg;
764		} else {
765			if (cp->link_cntl & BMCR_FULLDPLX)
766				val |= PCS_MII_CTRL_DUPLEX;
767			val &= ~PCS_MII_AUTONEG_EN;
768			cp->lstate = link_force_ok;
769		}
770		cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
771		writel(val, cp->regs + REG_PCS_MII_CTRL);
772
773	} else {
774		cas_mif_poll(cp, 0);
775		ctl = cas_phy_read(cp, MII_BMCR);
776		ctl &= ~(BMCR_FULLDPLX | BMCR_SPEED100 |
777			 CAS_BMCR_SPEED1000 | BMCR_ANENABLE);
778		ctl |= cp->link_cntl;
779		if (ctl & BMCR_ANENABLE) {
780			ctl |= BMCR_ANRESTART;
781			cp->lstate = link_aneg;
782		} else {
783			cp->lstate = link_force_ok;
784		}
785		cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
786		cas_phy_write(cp, MII_BMCR, ctl);
787		cas_mif_poll(cp, 1);
788	}
789
790	cp->timer_ticks = 0;
791	mod_timer(&cp->link_timer, jiffies + CAS_LINK_TIMEOUT);
792}
793
794/* Must be invoked under cp->lock. */
795static int cas_reset_mii_phy(struct cas *cp)
796{
797	int limit = STOP_TRIES_PHY;
798	u16 val;
799
800	cas_phy_write(cp, MII_BMCR, BMCR_RESET);
801	udelay(100);
802	while (--limit) {
803		val = cas_phy_read(cp, MII_BMCR);
804		if ((val & BMCR_RESET) == 0)
805			break;
806		udelay(10);
807	}
808	return limit <= 0;
809}
810
811static int cas_saturn_firmware_init(struct cas *cp)
812{
813	const struct firmware *fw;
814	const char fw_name[] = "sun/cassini.bin";
815	int err;
816
817	if (PHY_NS_DP83065 != cp->phy_id)
818		return 0;
819
820	err = request_firmware(&fw, fw_name, &cp->pdev->dev);
821	if (err) {
822		pr_err("Failed to load firmware \"%s\"\n",
823		       fw_name);
824		return err;
825	}
826	if (fw->size < 2) {
827		pr_err("bogus length %zu in \"%s\"\n",
828		       fw->size, fw_name);
829		err = -EINVAL;
830		goto out;
831	}
832	cp->fw_load_addr= fw->data[1] << 8 | fw->data[0];
833	cp->fw_size = fw->size - 2;
834	cp->fw_data = vmalloc(cp->fw_size);
835	if (!cp->fw_data) {
836		err = -ENOMEM;
837		goto out;
838	}
839	memcpy(cp->fw_data, &fw->data[2], cp->fw_size);
840out:
841	release_firmware(fw);
842	return err;
843}
844
845static void cas_saturn_firmware_load(struct cas *cp)
846{
847	int i;
848
849	cas_phy_powerdown(cp);
850
851	/* expanded memory access mode */
852	cas_phy_write(cp, DP83065_MII_MEM, 0x0);
853
854	/* pointer configuration for new firmware */
855	cas_phy_write(cp, DP83065_MII_REGE, 0x8ff9);
856	cas_phy_write(cp, DP83065_MII_REGD, 0xbd);
857	cas_phy_write(cp, DP83065_MII_REGE, 0x8ffa);
858	cas_phy_write(cp, DP83065_MII_REGD, 0x82);
859	cas_phy_write(cp, DP83065_MII_REGE, 0x8ffb);
860	cas_phy_write(cp, DP83065_MII_REGD, 0x0);
861	cas_phy_write(cp, DP83065_MII_REGE, 0x8ffc);
862	cas_phy_write(cp, DP83065_MII_REGD, 0x39);
863
864	/* download new firmware */
865	cas_phy_write(cp, DP83065_MII_MEM, 0x1);
866	cas_phy_write(cp, DP83065_MII_REGE, cp->fw_load_addr);
867	for (i = 0; i < cp->fw_size; i++)
868		cas_phy_write(cp, DP83065_MII_REGD, cp->fw_data[i]);
869
870	/* enable firmware */
871	cas_phy_write(cp, DP83065_MII_REGE, 0x8ff8);
872	cas_phy_write(cp, DP83065_MII_REGD, 0x1);
873}
874
875
876/* phy initialization */
877static void cas_phy_init(struct cas *cp)
878{
879	u16 val;
880
881	/* if we're in MII/GMII mode, set up phy */
882	if (CAS_PHY_MII(cp->phy_type)) {
883		writel(PCS_DATAPATH_MODE_MII,
884		       cp->regs + REG_PCS_DATAPATH_MODE);
885
886		cas_mif_poll(cp, 0);
887		cas_reset_mii_phy(cp); /* take out of isolate mode */
888
889		if (PHY_LUCENT_B0 == cp->phy_id) {
890			/* workaround link up/down issue with lucent */
891			cas_phy_write(cp, LUCENT_MII_REG, 0x8000);
892			cas_phy_write(cp, MII_BMCR, 0x00f1);
893			cas_phy_write(cp, LUCENT_MII_REG, 0x0);
894
895		} else if (PHY_BROADCOM_B0 == (cp->phy_id & 0xFFFFFFFC)) {
896			/* workarounds for broadcom phy */
897			cas_phy_write(cp, BROADCOM_MII_REG8, 0x0C20);
898			cas_phy_write(cp, BROADCOM_MII_REG7, 0x0012);
899			cas_phy_write(cp, BROADCOM_MII_REG5, 0x1804);
900			cas_phy_write(cp, BROADCOM_MII_REG7, 0x0013);
901			cas_phy_write(cp, BROADCOM_MII_REG5, 0x1204);
902			cas_phy_write(cp, BROADCOM_MII_REG7, 0x8006);
903			cas_phy_write(cp, BROADCOM_MII_REG5, 0x0132);
904			cas_phy_write(cp, BROADCOM_MII_REG7, 0x8006);
905			cas_phy_write(cp, BROADCOM_MII_REG5, 0x0232);
906			cas_phy_write(cp, BROADCOM_MII_REG7, 0x201F);
907			cas_phy_write(cp, BROADCOM_MII_REG5, 0x0A20);
908
909		} else if (PHY_BROADCOM_5411 == cp->phy_id) {
910			val = cas_phy_read(cp, BROADCOM_MII_REG4);
911			val = cas_phy_read(cp, BROADCOM_MII_REG4);
912			if (val & 0x0080) {
913				/* link workaround */
914				cas_phy_write(cp, BROADCOM_MII_REG4,
915					      val & ~0x0080);
916			}
917
918		} else if (cp->cas_flags & CAS_FLAG_SATURN) {
919			writel((cp->phy_type & CAS_PHY_MII_MDIO0) ?
920			       SATURN_PCFG_FSI : 0x0,
921			       cp->regs + REG_SATURN_PCFG);
922
923			/* load firmware to address 10Mbps auto-negotiation
924			 * issue. NOTE: this will need to be changed if the
925			 * default firmware gets fixed.
926			 */
927			if (PHY_NS_DP83065 == cp->phy_id) {
928				cas_saturn_firmware_load(cp);
929			}
930			cas_phy_powerup(cp);
931		}
932
933		/* advertise capabilities */
934		val = cas_phy_read(cp, MII_BMCR);
935		val &= ~BMCR_ANENABLE;
936		cas_phy_write(cp, MII_BMCR, val);
937		udelay(10);
938
939		cas_phy_write(cp, MII_ADVERTISE,
940			      cas_phy_read(cp, MII_ADVERTISE) |
941			      (ADVERTISE_10HALF | ADVERTISE_10FULL |
942			       ADVERTISE_100HALF | ADVERTISE_100FULL |
943			       CAS_ADVERTISE_PAUSE |
944			       CAS_ADVERTISE_ASYM_PAUSE));
945
946		if (cp->cas_flags & CAS_FLAG_1000MB_CAP) {
947			/* make sure that we don't advertise half
948			 * duplex to avoid a chip issue
949			 */
950			val  = cas_phy_read(cp, CAS_MII_1000_CTRL);
951			val &= ~CAS_ADVERTISE_1000HALF;
952			val |= CAS_ADVERTISE_1000FULL;
953			cas_phy_write(cp, CAS_MII_1000_CTRL, val);
954		}
955
956	} else {
957		/* reset pcs for serdes */
958		u32 val;
959		int limit;
960
961		writel(PCS_DATAPATH_MODE_SERDES,
962		       cp->regs + REG_PCS_DATAPATH_MODE);
963
964		/* enable serdes pins on saturn */
965		if (cp->cas_flags & CAS_FLAG_SATURN)
966			writel(0, cp->regs + REG_SATURN_PCFG);
967
968		/* Reset PCS unit. */
969		val = readl(cp->regs + REG_PCS_MII_CTRL);
970		val |= PCS_MII_RESET;
971		writel(val, cp->regs + REG_PCS_MII_CTRL);
972
973		limit = STOP_TRIES;
974		while (--limit > 0) {
975			udelay(10);
976			if ((readl(cp->regs + REG_PCS_MII_CTRL) &
977			     PCS_MII_RESET) == 0)
978				break;
979		}
980		if (limit <= 0)
981			netdev_warn(cp->dev, "PCS reset bit would not clear [%08x]\n",
982				    readl(cp->regs + REG_PCS_STATE_MACHINE));
983
984		/* Make sure PCS is disabled while changing advertisement
985		 * configuration.
986		 */
987		writel(0x0, cp->regs + REG_PCS_CFG);
988
989		/* Advertise all capabilities except half-duplex. */
990		val  = readl(cp->regs + REG_PCS_MII_ADVERT);
991		val &= ~PCS_MII_ADVERT_HD;
992		val |= (PCS_MII_ADVERT_FD | PCS_MII_ADVERT_SYM_PAUSE |
993			PCS_MII_ADVERT_ASYM_PAUSE);
994		writel(val, cp->regs + REG_PCS_MII_ADVERT);
995
996		/* enable PCS */
997		writel(PCS_CFG_EN, cp->regs + REG_PCS_CFG);
998
999		/* pcs workaround: enable sync detect */
1000		writel(PCS_SERDES_CTRL_SYNCD_EN,
1001		       cp->regs + REG_PCS_SERDES_CTRL);
1002	}
1003}
1004
1005
1006static int cas_pcs_link_check(struct cas *cp)
1007{
1008	u32 stat, state_machine;
1009	int retval = 0;
1010
1011	/* The link status bit latches on zero, so you must
1012	 * read it twice in such a case to see a transition
1013	 * to the link being up.
1014	 */
1015	stat = readl(cp->regs + REG_PCS_MII_STATUS);
1016	if ((stat & PCS_MII_STATUS_LINK_STATUS) == 0)
1017		stat = readl(cp->regs + REG_PCS_MII_STATUS);
1018
1019	/* The remote-fault indication is only valid
1020	 * when autoneg has completed.
1021	 */
1022	if ((stat & (PCS_MII_STATUS_AUTONEG_COMP |
1023		     PCS_MII_STATUS_REMOTE_FAULT)) ==
1024	    (PCS_MII_STATUS_AUTONEG_COMP | PCS_MII_STATUS_REMOTE_FAULT))
1025		netif_info(cp, link, cp->dev, "PCS RemoteFault\n");
1026
1027	/* work around link detection issue by querying the PCS state
1028	 * machine directly.
1029	 */
1030	state_machine = readl(cp->regs + REG_PCS_STATE_MACHINE);
1031	if ((state_machine & PCS_SM_LINK_STATE_MASK) != SM_LINK_STATE_UP) {
1032		stat &= ~PCS_MII_STATUS_LINK_STATUS;
1033	} else if (state_machine & PCS_SM_WORD_SYNC_STATE_MASK) {
1034		stat |= PCS_MII_STATUS_LINK_STATUS;
1035	}
1036
1037	if (stat & PCS_MII_STATUS_LINK_STATUS) {
1038		if (cp->lstate != link_up) {
1039			if (cp->opened) {
1040				cp->lstate = link_up;
1041				cp->link_transition = LINK_TRANSITION_LINK_UP;
1042
1043				cas_set_link_modes(cp);
1044				netif_carrier_on(cp->dev);
1045			}
1046		}
1047	} else if (cp->lstate == link_up) {
1048		cp->lstate = link_down;
1049		if (link_transition_timeout != 0 &&
1050		    cp->link_transition != LINK_TRANSITION_REQUESTED_RESET &&
1051		    !cp->link_transition_jiffies_valid) {
1052			/*
1053			 * force a reset, as a workaround for the
1054			 * link-failure problem. May want to move this to a
1055			 * point a bit earlier in the sequence. If we had
1056			 * generated a reset a short time ago, we'll wait for
1057			 * the link timer to check the status until a
1058			 * timer expires (link_transistion_jiffies_valid is
1059			 * true when the timer is running.)  Instead of using
1060			 * a system timer, we just do a check whenever the
1061			 * link timer is running - this clears the flag after
1062			 * a suitable delay.
1063			 */
1064			retval = 1;
1065			cp->link_transition = LINK_TRANSITION_REQUESTED_RESET;
1066			cp->link_transition_jiffies = jiffies;
1067			cp->link_transition_jiffies_valid = 1;
1068		} else {
1069			cp->link_transition = LINK_TRANSITION_ON_FAILURE;
1070		}
1071		netif_carrier_off(cp->dev);
1072		if (cp->opened)
1073			netif_info(cp, link, cp->dev, "PCS link down\n");
1074
1075		/* Cassini only: if you force a mode, there can be
1076		 * sync problems on link down. to fix that, the following
1077		 * things need to be checked:
1078		 * 1) read serialink state register
1079		 * 2) read pcs status register to verify link down.
1080		 * 3) if link down and serial link == 0x03, then you need
1081		 *    to global reset the chip.
1082		 */
1083		if ((cp->cas_flags & CAS_FLAG_REG_PLUS) == 0) {
1084			/* should check to see if we're in a forced mode */
1085			stat = readl(cp->regs + REG_PCS_SERDES_STATE);
1086			if (stat == 0x03)
1087				return 1;
1088		}
1089	} else if (cp->lstate == link_down) {
1090		if (link_transition_timeout != 0 &&
1091		    cp->link_transition != LINK_TRANSITION_REQUESTED_RESET &&
1092		    !cp->link_transition_jiffies_valid) {
1093			/* force a reset, as a workaround for the
1094			 * link-failure problem.  May want to move
1095			 * this to a point a bit earlier in the
1096			 * sequence.
1097			 */
1098			retval = 1;
1099			cp->link_transition = LINK_TRANSITION_REQUESTED_RESET;
1100			cp->link_transition_jiffies = jiffies;
1101			cp->link_transition_jiffies_valid = 1;
1102		} else {
1103			cp->link_transition = LINK_TRANSITION_STILL_FAILED;
1104		}
1105	}
1106
1107	return retval;
1108}
1109
1110static int cas_pcs_interrupt(struct net_device *dev,
1111			     struct cas *cp, u32 status)
1112{
1113	u32 stat = readl(cp->regs + REG_PCS_INTR_STATUS);
1114
1115	if ((stat & PCS_INTR_STATUS_LINK_CHANGE) == 0)
1116		return 0;
1117	return cas_pcs_link_check(cp);
1118}
1119
1120static int cas_txmac_interrupt(struct net_device *dev,
1121			       struct cas *cp, u32 status)
1122{
1123	u32 txmac_stat = readl(cp->regs + REG_MAC_TX_STATUS);
1124
1125	if (!txmac_stat)
1126		return 0;
1127
1128	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
1129		     "txmac interrupt, txmac_stat: 0x%x\n", txmac_stat);
1130
1131	/* Defer timer expiration is quite normal,
1132	 * don't even log the event.
1133	 */
1134	if ((txmac_stat & MAC_TX_DEFER_TIMER) &&
1135	    !(txmac_stat & ~MAC_TX_DEFER_TIMER))
1136		return 0;
1137
1138	spin_lock(&cp->stat_lock[0]);
1139	if (txmac_stat & MAC_TX_UNDERRUN) {
1140		netdev_err(dev, "TX MAC xmit underrun\n");
1141		cp->net_stats[0].tx_fifo_errors++;
1142	}
1143
1144	if (txmac_stat & MAC_TX_MAX_PACKET_ERR) {
1145		netdev_err(dev, "TX MAC max packet size error\n");
1146		cp->net_stats[0].tx_errors++;
1147	}
1148
1149	/* The rest are all cases of one of the 16-bit TX
1150	 * counters expiring.
1151	 */
1152	if (txmac_stat & MAC_TX_COLL_NORMAL)
1153		cp->net_stats[0].collisions += 0x10000;
1154
1155	if (txmac_stat & MAC_TX_COLL_EXCESS) {
1156		cp->net_stats[0].tx_aborted_errors += 0x10000;
1157		cp->net_stats[0].collisions += 0x10000;
1158	}
1159
1160	if (txmac_stat & MAC_TX_COLL_LATE) {
1161		cp->net_stats[0].tx_aborted_errors += 0x10000;
1162		cp->net_stats[0].collisions += 0x10000;
1163	}
1164	spin_unlock(&cp->stat_lock[0]);
1165
1166	/* We do not keep track of MAC_TX_COLL_FIRST and
1167	 * MAC_TX_PEAK_ATTEMPTS events.
1168	 */
1169	return 0;
1170}
1171
1172static void cas_load_firmware(struct cas *cp, cas_hp_inst_t *firmware)
1173{
1174	cas_hp_inst_t *inst;
1175	u32 val;
1176	int i;
1177
1178	i = 0;
1179	while ((inst = firmware) && inst->note) {
1180		writel(i, cp->regs + REG_HP_INSTR_RAM_ADDR);
1181
1182		val = CAS_BASE(HP_INSTR_RAM_HI_VAL, inst->val);
1183		val |= CAS_BASE(HP_INSTR_RAM_HI_MASK, inst->mask);
1184		writel(val, cp->regs + REG_HP_INSTR_RAM_DATA_HI);
1185
1186		val = CAS_BASE(HP_INSTR_RAM_MID_OUTARG, inst->outarg >> 10);
1187		val |= CAS_BASE(HP_INSTR_RAM_MID_OUTOP, inst->outop);
1188		val |= CAS_BASE(HP_INSTR_RAM_MID_FNEXT, inst->fnext);
1189		val |= CAS_BASE(HP_INSTR_RAM_MID_FOFF, inst->foff);
1190		val |= CAS_BASE(HP_INSTR_RAM_MID_SNEXT, inst->snext);
1191		val |= CAS_BASE(HP_INSTR_RAM_MID_SOFF, inst->soff);
1192		val |= CAS_BASE(HP_INSTR_RAM_MID_OP, inst->op);
1193		writel(val, cp->regs + REG_HP_INSTR_RAM_DATA_MID);
1194
1195		val = CAS_BASE(HP_INSTR_RAM_LOW_OUTMASK, inst->outmask);
1196		val |= CAS_BASE(HP_INSTR_RAM_LOW_OUTSHIFT, inst->outshift);
1197		val |= CAS_BASE(HP_INSTR_RAM_LOW_OUTEN, inst->outenab);
1198		val |= CAS_BASE(HP_INSTR_RAM_LOW_OUTARG, inst->outarg);
1199		writel(val, cp->regs + REG_HP_INSTR_RAM_DATA_LOW);
1200		++firmware;
1201		++i;
1202	}
1203}
1204
1205static void cas_init_rx_dma(struct cas *cp)
1206{
1207	u64 desc_dma = cp->block_dvma;
1208	u32 val;
1209	int i, size;
1210
1211	/* rx free descriptors */
1212	val = CAS_BASE(RX_CFG_SWIVEL, RX_SWIVEL_OFF_VAL);
1213	val |= CAS_BASE(RX_CFG_DESC_RING, RX_DESC_RINGN_INDEX(0));
1214	val |= CAS_BASE(RX_CFG_COMP_RING, RX_COMP_RINGN_INDEX(0));
1215	if ((N_RX_DESC_RINGS > 1) &&
1216	    (cp->cas_flags & CAS_FLAG_REG_PLUS))  /* do desc 2 */
1217		val |= CAS_BASE(RX_CFG_DESC_RING1, RX_DESC_RINGN_INDEX(1));
1218	writel(val, cp->regs + REG_RX_CFG);
1219
1220	val = (unsigned long) cp->init_rxds[0] -
1221		(unsigned long) cp->init_block;
1222	writel((desc_dma + val) >> 32, cp->regs + REG_RX_DB_HI);
1223	writel((desc_dma + val) & 0xffffffff, cp->regs + REG_RX_DB_LOW);
1224	writel(RX_DESC_RINGN_SIZE(0) - 4, cp->regs + REG_RX_KICK);
1225
1226	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
1227		/* rx desc 2 is for IPSEC packets. however,
1228		 * we don't it that for that purpose.
1229		 */
1230		val = (unsigned long) cp->init_rxds[1] -
1231			(unsigned long) cp->init_block;
1232		writel((desc_dma + val) >> 32, cp->regs + REG_PLUS_RX_DB1_HI);
1233		writel((desc_dma + val) & 0xffffffff, cp->regs +
1234		       REG_PLUS_RX_DB1_LOW);
1235		writel(RX_DESC_RINGN_SIZE(1) - 4, cp->regs +
1236		       REG_PLUS_RX_KICK1);
1237	}
1238
1239	/* rx completion registers */
1240	val = (unsigned long) cp->init_rxcs[0] -
1241		(unsigned long) cp->init_block;
1242	writel((desc_dma + val) >> 32, cp->regs + REG_RX_CB_HI);
1243	writel((desc_dma + val) & 0xffffffff, cp->regs + REG_RX_CB_LOW);
1244
1245	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
1246		/* rx comp 2-4 */
1247		for (i = 1; i < MAX_RX_COMP_RINGS; i++) {
1248			val = (unsigned long) cp->init_rxcs[i] -
1249				(unsigned long) cp->init_block;
1250			writel((desc_dma + val) >> 32, cp->regs +
1251			       REG_PLUS_RX_CBN_HI(i));
1252			writel((desc_dma + val) & 0xffffffff, cp->regs +
1253			       REG_PLUS_RX_CBN_LOW(i));
1254		}
1255	}
1256
1257	/* read selective clear regs to prevent spurious interrupts
1258	 * on reset because complete == kick.
1259	 * selective clear set up to prevent interrupts on resets
1260	 */
1261	readl(cp->regs + REG_INTR_STATUS_ALIAS);
1262	writel(INTR_RX_DONE | INTR_RX_BUF_UNAVAIL, cp->regs + REG_ALIAS_CLEAR);
1263	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
1264		for (i = 1; i < N_RX_COMP_RINGS; i++)
1265			readl(cp->regs + REG_PLUS_INTRN_STATUS_ALIAS(i));
1266
1267		/* 2 is different from 3 and 4 */
1268		if (N_RX_COMP_RINGS > 1)
1269			writel(INTR_RX_DONE_ALT | INTR_RX_BUF_UNAVAIL_1,
1270			       cp->regs + REG_PLUS_ALIASN_CLEAR(1));
1271
1272		for (i = 2; i < N_RX_COMP_RINGS; i++)
1273			writel(INTR_RX_DONE_ALT,
1274			       cp->regs + REG_PLUS_ALIASN_CLEAR(i));
1275	}
1276
1277	/* set up pause thresholds */
1278	val  = CAS_BASE(RX_PAUSE_THRESH_OFF,
1279			cp->rx_pause_off / RX_PAUSE_THRESH_QUANTUM);
1280	val |= CAS_BASE(RX_PAUSE_THRESH_ON,
1281			cp->rx_pause_on / RX_PAUSE_THRESH_QUANTUM);
1282	writel(val, cp->regs + REG_RX_PAUSE_THRESH);
1283
1284	/* zero out dma reassembly buffers */
1285	for (i = 0; i < 64; i++) {
1286		writel(i, cp->regs + REG_RX_TABLE_ADDR);
1287		writel(0x0, cp->regs + REG_RX_TABLE_DATA_LOW);
1288		writel(0x0, cp->regs + REG_RX_TABLE_DATA_MID);
1289		writel(0x0, cp->regs + REG_RX_TABLE_DATA_HI);
1290	}
1291
1292	/* make sure address register is 0 for normal operation */
1293	writel(0x0, cp->regs + REG_RX_CTRL_FIFO_ADDR);
1294	writel(0x0, cp->regs + REG_RX_IPP_FIFO_ADDR);
1295
1296	/* interrupt mitigation */
1297#ifdef USE_RX_BLANK
1298	val = CAS_BASE(RX_BLANK_INTR_TIME, RX_BLANK_INTR_TIME_VAL);
1299	val |= CAS_BASE(RX_BLANK_INTR_PKT, RX_BLANK_INTR_PKT_VAL);
1300	writel(val, cp->regs + REG_RX_BLANK);
1301#else
1302	writel(0x0, cp->regs + REG_RX_BLANK);
1303#endif
1304
1305	/* interrupt generation as a function of low water marks for
1306	 * free desc and completion entries. these are used to trigger
1307	 * housekeeping for rx descs. we don't use the free interrupt
1308	 * as it's not very useful
1309	 */
1310	/* val = CAS_BASE(RX_AE_THRESH_FREE, RX_AE_FREEN_VAL(0)); */
1311	val = CAS_BASE(RX_AE_THRESH_COMP, RX_AE_COMP_VAL);
1312	writel(val, cp->regs + REG_RX_AE_THRESH);
1313	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
1314		val = CAS_BASE(RX_AE1_THRESH_FREE, RX_AE_FREEN_VAL(1));
1315		writel(val, cp->regs + REG_PLUS_RX_AE1_THRESH);
1316	}
1317
1318	/* Random early detect registers. useful for congestion avoidance.
1319	 * this should be tunable.
1320	 */
1321	writel(0x0, cp->regs + REG_RX_RED);
1322
1323	/* receive page sizes. default == 2K (0x800) */
1324	val = 0;
1325	if (cp->page_size == 0x1000)
1326		val = 0x1;
1327	else if (cp->page_size == 0x2000)
1328		val = 0x2;
1329	else if (cp->page_size == 0x4000)
1330		val = 0x3;
1331
1332	/* round mtu + offset. constrain to page size. */
1333	size = cp->dev->mtu + 64;
1334	if (size > cp->page_size)
1335		size = cp->page_size;
1336
1337	if (size <= 0x400)
1338		i = 0x0;
1339	else if (size <= 0x800)
1340		i = 0x1;
1341	else if (size <= 0x1000)
1342		i = 0x2;
1343	else
1344		i = 0x3;
1345
1346	cp->mtu_stride = 1 << (i + 10);
1347	val  = CAS_BASE(RX_PAGE_SIZE, val);
1348	val |= CAS_BASE(RX_PAGE_SIZE_MTU_STRIDE, i);
1349	val |= CAS_BASE(RX_PAGE_SIZE_MTU_COUNT, cp->page_size >> (i + 10));
1350	val |= CAS_BASE(RX_PAGE_SIZE_MTU_OFF, 0x1);
1351	writel(val, cp->regs + REG_RX_PAGE_SIZE);
1352
1353	/* enable the header parser if desired */
1354	if (CAS_HP_FIRMWARE == cas_prog_null)
1355		return;
1356
1357	val = CAS_BASE(HP_CFG_NUM_CPU, CAS_NCPUS > 63 ? 0 : CAS_NCPUS);
1358	val |= HP_CFG_PARSE_EN | HP_CFG_SYN_INC_MASK;
1359	val |= CAS_BASE(HP_CFG_TCP_THRESH, HP_TCP_THRESH_VAL);
1360	writel(val, cp->regs + REG_HP_CFG);
1361}
1362
1363static inline void cas_rxc_init(struct cas_rx_comp *rxc)
1364{
1365	memset(rxc, 0, sizeof(*rxc));
1366	rxc->word4 = cpu_to_le64(RX_COMP4_ZERO);
1367}
1368
1369/* NOTE: we use the ENC RX DESC ring for spares. the rx_page[0,1]
1370 * flipping is protected by the fact that the chip will not
1371 * hand back the same page index while it's being processed.
1372 */
1373static inline cas_page_t *cas_page_spare(struct cas *cp, const int index)
1374{
1375	cas_page_t *page = cp->rx_pages[1][index];
1376	cas_page_t *new;
1377
1378	if (page_count(page->buffer) == 1)
1379		return page;
1380
1381	new = cas_page_dequeue(cp);
1382	if (new) {
1383		spin_lock(&cp->rx_inuse_lock);
1384		list_add(&page->list, &cp->rx_inuse_list);
1385		spin_unlock(&cp->rx_inuse_lock);
1386	}
1387	return new;
1388}
1389
1390/* this needs to be changed if we actually use the ENC RX DESC ring */
1391static cas_page_t *cas_page_swap(struct cas *cp, const int ring,
1392				 const int index)
1393{
1394	cas_page_t **page0 = cp->rx_pages[0];
1395	cas_page_t **page1 = cp->rx_pages[1];
1396
1397	/* swap if buffer is in use */
1398	if (page_count(page0[index]->buffer) > 1) {
1399		cas_page_t *new = cas_page_spare(cp, index);
1400		if (new) {
1401			page1[index] = page0[index];
1402			page0[index] = new;
1403		}
1404	}
1405	RX_USED_SET(page0[index], 0);
1406	return page0[index];
1407}
1408
1409static void cas_clean_rxds(struct cas *cp)
1410{
1411	/* only clean ring 0 as ring 1 is used for spare buffers */
1412        struct cas_rx_desc *rxd = cp->init_rxds[0];
1413	int i, size;
1414
1415	/* release all rx flows */
1416	for (i = 0; i < N_RX_FLOWS; i++) {
1417		struct sk_buff *skb;
1418		while ((skb = __skb_dequeue(&cp->rx_flows[i]))) {
1419			cas_skb_release(skb);
1420		}
1421	}
1422
1423	/* initialize descriptors */
1424	size = RX_DESC_RINGN_SIZE(0);
1425	for (i = 0; i < size; i++) {
1426		cas_page_t *page = cas_page_swap(cp, 0, i);
1427		rxd[i].buffer = cpu_to_le64(page->dma_addr);
1428		rxd[i].index  = cpu_to_le64(CAS_BASE(RX_INDEX_NUM, i) |
1429					    CAS_BASE(RX_INDEX_RING, 0));
1430	}
1431
1432	cp->rx_old[0]  = RX_DESC_RINGN_SIZE(0) - 4;
1433	cp->rx_last[0] = 0;
1434	cp->cas_flags &= ~CAS_FLAG_RXD_POST(0);
1435}
1436
1437static void cas_clean_rxcs(struct cas *cp)
1438{
1439	int i, j;
1440
1441	/* take ownership of rx comp descriptors */
1442	memset(cp->rx_cur, 0, sizeof(*cp->rx_cur)*N_RX_COMP_RINGS);
1443	memset(cp->rx_new, 0, sizeof(*cp->rx_new)*N_RX_COMP_RINGS);
1444	for (i = 0; i < N_RX_COMP_RINGS; i++) {
1445		struct cas_rx_comp *rxc = cp->init_rxcs[i];
1446		for (j = 0; j < RX_COMP_RINGN_SIZE(i); j++) {
1447			cas_rxc_init(rxc + j);
1448		}
1449	}
1450}
1451
1452#if 0
1453/* When we get a RX fifo overflow, the RX unit is probably hung
1454 * so we do the following.
1455 *
1456 * If any part of the reset goes wrong, we return 1 and that causes the
1457 * whole chip to be reset.
1458 */
1459static int cas_rxmac_reset(struct cas *cp)
1460{
1461	struct net_device *dev = cp->dev;
1462	int limit;
1463	u32 val;
1464
1465	/* First, reset MAC RX. */
1466	writel(cp->mac_rx_cfg & ~MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
1467	for (limit = 0; limit < STOP_TRIES; limit++) {
1468		if (!(readl(cp->regs + REG_MAC_RX_CFG) & MAC_RX_CFG_EN))
1469			break;
1470		udelay(10);
1471	}
1472	if (limit == STOP_TRIES) {
1473		netdev_err(dev, "RX MAC will not disable, resetting whole chip\n");
1474		return 1;
1475	}
1476
1477	/* Second, disable RX DMA. */
1478	writel(0, cp->regs + REG_RX_CFG);
1479	for (limit = 0; limit < STOP_TRIES; limit++) {
1480		if (!(readl(cp->regs + REG_RX_CFG) & RX_CFG_DMA_EN))
1481			break;
1482		udelay(10);
1483	}
1484	if (limit == STOP_TRIES) {
1485		netdev_err(dev, "RX DMA will not disable, resetting whole chip\n");
1486		return 1;
1487	}
1488
1489	mdelay(5);
1490
1491	/* Execute RX reset command. */
1492	writel(SW_RESET_RX, cp->regs + REG_SW_RESET);
1493	for (limit = 0; limit < STOP_TRIES; limit++) {
1494		if (!(readl(cp->regs + REG_SW_RESET) & SW_RESET_RX))
1495			break;
1496		udelay(10);
1497	}
1498	if (limit == STOP_TRIES) {
1499		netdev_err(dev, "RX reset command will not execute, resetting whole chip\n");
1500		return 1;
1501	}
1502
1503	/* reset driver rx state */
1504	cas_clean_rxds(cp);
1505	cas_clean_rxcs(cp);
1506
1507	/* Now, reprogram the rest of RX unit. */
1508	cas_init_rx_dma(cp);
1509
1510	/* re-enable */
1511	val = readl(cp->regs + REG_RX_CFG);
1512	writel(val | RX_CFG_DMA_EN, cp->regs + REG_RX_CFG);
1513	writel(MAC_RX_FRAME_RECV, cp->regs + REG_MAC_RX_MASK);
1514	val = readl(cp->regs + REG_MAC_RX_CFG);
1515	writel(val | MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
1516	return 0;
1517}
1518#endif
1519
1520static int cas_rxmac_interrupt(struct net_device *dev, struct cas *cp,
1521			       u32 status)
1522{
1523	u32 stat = readl(cp->regs + REG_MAC_RX_STATUS);
1524
1525	if (!stat)
1526		return 0;
1527
1528	netif_dbg(cp, intr, cp->dev, "rxmac interrupt, stat: 0x%x\n", stat);
1529
1530	/* these are all rollovers */
1531	spin_lock(&cp->stat_lock[0]);
1532	if (stat & MAC_RX_ALIGN_ERR)
1533		cp->net_stats[0].rx_frame_errors += 0x10000;
1534
1535	if (stat & MAC_RX_CRC_ERR)
1536		cp->net_stats[0].rx_crc_errors += 0x10000;
1537
1538	if (stat & MAC_RX_LEN_ERR)
1539		cp->net_stats[0].rx_length_errors += 0x10000;
1540
1541	if (stat & MAC_RX_OVERFLOW) {
1542		cp->net_stats[0].rx_over_errors++;
1543		cp->net_stats[0].rx_fifo_errors++;
1544	}
1545
1546	/* We do not track MAC_RX_FRAME_COUNT and MAC_RX_VIOL_ERR
1547	 * events.
1548	 */
1549	spin_unlock(&cp->stat_lock[0]);
1550	return 0;
1551}
1552
1553static int cas_mac_interrupt(struct net_device *dev, struct cas *cp,
1554			     u32 status)
1555{
1556	u32 stat = readl(cp->regs + REG_MAC_CTRL_STATUS);
1557
1558	if (!stat)
1559		return 0;
1560
1561	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
1562		     "mac interrupt, stat: 0x%x\n", stat);
1563
1564	/* This interrupt is just for pause frame and pause
1565	 * tracking.  It is useful for diagnostics and debug
1566	 * but probably by default we will mask these events.
1567	 */
1568	if (stat & MAC_CTRL_PAUSE_STATE)
1569		cp->pause_entered++;
1570
1571	if (stat & MAC_CTRL_PAUSE_RECEIVED)
1572		cp->pause_last_time_recvd = (stat >> 16);
1573
1574	return 0;
1575}
1576
1577
1578/* Must be invoked under cp->lock. */
1579static inline int cas_mdio_link_not_up(struct cas *cp)
1580{
1581	u16 val;
1582
1583	switch (cp->lstate) {
1584	case link_force_ret:
1585		netif_info(cp, link, cp->dev, "Autoneg failed again, keeping forced mode\n");
1586		cas_phy_write(cp, MII_BMCR, cp->link_fcntl);
1587		cp->timer_ticks = 5;
1588		cp->lstate = link_force_ok;
1589		cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
1590		break;
1591
1592	case link_aneg:
1593		val = cas_phy_read(cp, MII_BMCR);
1594
1595		/* Try forced modes. we try things in the following order:
1596		 * 1000 full -> 100 full/half -> 10 half
1597		 */
1598		val &= ~(BMCR_ANRESTART | BMCR_ANENABLE);
1599		val |= BMCR_FULLDPLX;
1600		val |= (cp->cas_flags & CAS_FLAG_1000MB_CAP) ?
1601			CAS_BMCR_SPEED1000 : BMCR_SPEED100;
1602		cas_phy_write(cp, MII_BMCR, val);
1603		cp->timer_ticks = 5;
1604		cp->lstate = link_force_try;
1605		cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
1606		break;
1607
1608	case link_force_try:
1609		/* Downgrade from 1000 to 100 to 10 Mbps if necessary. */
1610		val = cas_phy_read(cp, MII_BMCR);
1611		cp->timer_ticks = 5;
1612		if (val & CAS_BMCR_SPEED1000) { /* gigabit */
1613			val &= ~CAS_BMCR_SPEED1000;
1614			val |= (BMCR_SPEED100 | BMCR_FULLDPLX);
1615			cas_phy_write(cp, MII_BMCR, val);
1616			break;
1617		}
1618
1619		if (val & BMCR_SPEED100) {
1620			if (val & BMCR_FULLDPLX) /* fd failed */
1621				val &= ~BMCR_FULLDPLX;
1622			else { /* 100Mbps failed */
1623				val &= ~BMCR_SPEED100;
1624			}
1625			cas_phy_write(cp, MII_BMCR, val);
1626			break;
1627		}
1628	default:
1629		break;
1630	}
1631	return 0;
1632}
1633
1634
1635/* must be invoked with cp->lock held */
1636static int cas_mii_link_check(struct cas *cp, const u16 bmsr)
1637{
1638	int restart;
1639
1640	if (bmsr & BMSR_LSTATUS) {
1641		/* Ok, here we got a link. If we had it due to a forced
1642		 * fallback, and we were configured for autoneg, we
1643		 * retry a short autoneg pass. If you know your hub is
1644		 * broken, use ethtool ;)
1645		 */
1646		if ((cp->lstate == link_force_try) &&
1647		    (cp->link_cntl & BMCR_ANENABLE)) {
1648			cp->lstate = link_force_ret;
1649			cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
1650			cas_mif_poll(cp, 0);
1651			cp->link_fcntl = cas_phy_read(cp, MII_BMCR);
1652			cp->timer_ticks = 5;
1653			if (cp->opened)
1654				netif_info(cp, link, cp->dev,
1655					   "Got link after fallback, retrying autoneg once...\n");
1656			cas_phy_write(cp, MII_BMCR,
1657				      cp->link_fcntl | BMCR_ANENABLE |
1658				      BMCR_ANRESTART);
1659			cas_mif_poll(cp, 1);
1660
1661		} else if (cp->lstate != link_up) {
1662			cp->lstate = link_up;
1663			cp->link_transition = LINK_TRANSITION_LINK_UP;
1664
1665			if (cp->opened) {
1666				cas_set_link_modes(cp);
1667				netif_carrier_on(cp->dev);
1668			}
1669		}
1670		return 0;
1671	}
1672
1673	/* link not up. if the link was previously up, we restart the
1674	 * whole process
1675	 */
1676	restart = 0;
1677	if (cp->lstate == link_up) {
1678		cp->lstate = link_down;
1679		cp->link_transition = LINK_TRANSITION_LINK_DOWN;
1680
1681		netif_carrier_off(cp->dev);
1682		if (cp->opened)
1683			netif_info(cp, link, cp->dev, "Link down\n");
1684		restart = 1;
1685
1686	} else if (++cp->timer_ticks > 10)
1687		cas_mdio_link_not_up(cp);
1688
1689	return restart;
1690}
1691
1692static int cas_mif_interrupt(struct net_device *dev, struct cas *cp,
1693			     u32 status)
1694{
1695	u32 stat = readl(cp->regs + REG_MIF_STATUS);
1696	u16 bmsr;
1697
1698	/* check for a link change */
1699	if (CAS_VAL(MIF_STATUS_POLL_STATUS, stat) == 0)
1700		return 0;
1701
1702	bmsr = CAS_VAL(MIF_STATUS_POLL_DATA, stat);
1703	return cas_mii_link_check(cp, bmsr);
1704}
1705
1706static int cas_pci_interrupt(struct net_device *dev, struct cas *cp,
1707			     u32 status)
1708{
1709	u32 stat = readl(cp->regs + REG_PCI_ERR_STATUS);
1710
1711	if (!stat)
1712		return 0;
1713
1714	netdev_err(dev, "PCI error [%04x:%04x]",
1715		   stat, readl(cp->regs + REG_BIM_DIAG));
1716
1717	/* cassini+ has this reserved */
1718	if ((stat & PCI_ERR_BADACK) &&
1719	    ((cp->cas_flags & CAS_FLAG_REG_PLUS) == 0))
1720		pr_cont(" <No ACK64# during ABS64 cycle>");
1721
1722	if (stat & PCI_ERR_DTRTO)
1723		pr_cont(" <Delayed transaction timeout>");
1724	if (stat & PCI_ERR_OTHER)
1725		pr_cont(" <other>");
1726	if (stat & PCI_ERR_BIM_DMA_WRITE)
1727		pr_cont(" <BIM DMA 0 write req>");
1728	if (stat & PCI_ERR_BIM_DMA_READ)
1729		pr_cont(" <BIM DMA 0 read req>");
1730	pr_cont("\n");
1731
1732	if (stat & PCI_ERR_OTHER) {
1733		u16 cfg;
1734
1735		/* Interrogate PCI config space for the
1736		 * true cause.
1737		 */
1738		pci_read_config_word(cp->pdev, PCI_STATUS, &cfg);
1739		netdev_err(dev, "Read PCI cfg space status [%04x]\n", cfg);
1740		if (cfg & PCI_STATUS_PARITY)
1741			netdev_err(dev, "PCI parity error detected\n");
1742		if (cfg & PCI_STATUS_SIG_TARGET_ABORT)
1743			netdev_err(dev, "PCI target abort\n");
1744		if (cfg & PCI_STATUS_REC_TARGET_ABORT)
1745			netdev_err(dev, "PCI master acks target abort\n");
1746		if (cfg & PCI_STATUS_REC_MASTER_ABORT)
1747			netdev_err(dev, "PCI master abort\n");
1748		if (cfg & PCI_STATUS_SIG_SYSTEM_ERROR)
1749			netdev_err(dev, "PCI system error SERR#\n");
1750		if (cfg & PCI_STATUS_DETECTED_PARITY)
1751			netdev_err(dev, "PCI parity error\n");
1752
1753		/* Write the error bits back to clear them. */
1754		cfg &= (PCI_STATUS_PARITY |
1755			PCI_STATUS_SIG_TARGET_ABORT |
1756			PCI_STATUS_REC_TARGET_ABORT |
1757			PCI_STATUS_REC_MASTER_ABORT |
1758			PCI_STATUS_SIG_SYSTEM_ERROR |
1759			PCI_STATUS_DETECTED_PARITY);
1760		pci_write_config_word(cp->pdev, PCI_STATUS, cfg);
1761	}
1762
1763	/* For all PCI errors, we should reset the chip. */
1764	return 1;
1765}
1766
1767/* All non-normal interrupt conditions get serviced here.
1768 * Returns non-zero if we should just exit the interrupt
1769 * handler right now (ie. if we reset the card which invalidates
1770 * all of the other original irq status bits).
1771 */
1772static int cas_abnormal_irq(struct net_device *dev, struct cas *cp,
1773			    u32 status)
1774{
1775	if (status & INTR_RX_TAG_ERROR) {
1776		/* corrupt RX tag framing */
1777		netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
1778			     "corrupt rx tag framing\n");
1779		spin_lock(&cp->stat_lock[0]);
1780		cp->net_stats[0].rx_errors++;
1781		spin_unlock(&cp->stat_lock[0]);
1782		goto do_reset;
1783	}
1784
1785	if (status & INTR_RX_LEN_MISMATCH) {
1786		/* length mismatch. */
1787		netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
1788			     "length mismatch for rx frame\n");
1789		spin_lock(&cp->stat_lock[0]);
1790		cp->net_stats[0].rx_errors++;
1791		spin_unlock(&cp->stat_lock[0]);
1792		goto do_reset;
1793	}
1794
1795	if (status & INTR_PCS_STATUS) {
1796		if (cas_pcs_interrupt(dev, cp, status))
1797			goto do_reset;
1798	}
1799
1800	if (status & INTR_TX_MAC_STATUS) {
1801		if (cas_txmac_interrupt(dev, cp, status))
1802			goto do_reset;
1803	}
1804
1805	if (status & INTR_RX_MAC_STATUS) {
1806		if (cas_rxmac_interrupt(dev, cp, status))
1807			goto do_reset;
1808	}
1809
1810	if (status & INTR_MAC_CTRL_STATUS) {
1811		if (cas_mac_interrupt(dev, cp, status))
1812			goto do_reset;
1813	}
1814
1815	if (status & INTR_MIF_STATUS) {
1816		if (cas_mif_interrupt(dev, cp, status))
1817			goto do_reset;
1818	}
1819
1820	if (status & INTR_PCI_ERROR_STATUS) {
1821		if (cas_pci_interrupt(dev, cp, status))
1822			goto do_reset;
1823	}
1824	return 0;
1825
1826do_reset:
1827#if 1
1828	atomic_inc(&cp->reset_task_pending);
1829	atomic_inc(&cp->reset_task_pending_all);
1830	netdev_err(dev, "reset called in cas_abnormal_irq [0x%x]\n", status);
1831	schedule_work(&cp->reset_task);
1832#else
1833	atomic_set(&cp->reset_task_pending, CAS_RESET_ALL);
1834	netdev_err(dev, "reset called in cas_abnormal_irq\n");
1835	schedule_work(&cp->reset_task);
1836#endif
1837	return 1;
1838}
1839
1840/* NOTE: CAS_TABORT returns 1 or 2 so that it can be used when
1841 *       determining whether to do a netif_stop/wakeup
1842 */
1843#define CAS_TABORT(x)      (((x)->cas_flags & CAS_FLAG_TARGET_ABORT) ? 2 : 1)
1844#define CAS_ROUND_PAGE(x)  (((x) + PAGE_SIZE - 1) & PAGE_MASK)
1845static inline int cas_calc_tabort(struct cas *cp, const unsigned long addr,
1846				  const int len)
1847{
1848	unsigned long off = addr + len;
1849
1850	if (CAS_TABORT(cp) == 1)
1851		return 0;
1852	if ((CAS_ROUND_PAGE(off) - off) > TX_TARGET_ABORT_LEN)
1853		return 0;
1854	return TX_TARGET_ABORT_LEN;
1855}
1856
1857static inline void cas_tx_ringN(struct cas *cp, int ring, int limit)
1858{
1859	struct cas_tx_desc *txds;
1860	struct sk_buff **skbs;
1861	struct net_device *dev = cp->dev;
1862	int entry, count;
1863
1864	spin_lock(&cp->tx_lock[ring]);
1865	txds = cp->init_txds[ring];
1866	skbs = cp->tx_skbs[ring];
1867	entry = cp->tx_old[ring];
1868
1869	count = TX_BUFF_COUNT(ring, entry, limit);
1870	while (entry != limit) {
1871		struct sk_buff *skb = skbs[entry];
1872		dma_addr_t daddr;
1873		u32 dlen;
1874		int frag;
1875
1876		if (!skb) {
1877			/* this should never occur */
1878			entry = TX_DESC_NEXT(ring, entry);
1879			continue;
1880		}
1881
1882		/* however, we might get only a partial skb release. */
1883		count -= skb_shinfo(skb)->nr_frags +
1884			+ cp->tx_tiny_use[ring][entry].nbufs + 1;
1885		if (count < 0)
1886			break;
1887
1888		netif_printk(cp, tx_done, KERN_DEBUG, cp->dev,
1889			     "tx[%d] done, slot %d\n", ring, entry);
1890
1891		skbs[entry] = NULL;
1892		cp->tx_tiny_use[ring][entry].nbufs = 0;
1893
1894		for (frag = 0; frag <= skb_shinfo(skb)->nr_frags; frag++) {
1895			struct cas_tx_desc *txd = txds + entry;
1896
1897			daddr = le64_to_cpu(txd->buffer);
1898			dlen = CAS_VAL(TX_DESC_BUFLEN,
1899				       le64_to_cpu(txd->control));
1900			pci_unmap_page(cp->pdev, daddr, dlen,
1901				       PCI_DMA_TODEVICE);
1902			entry = TX_DESC_NEXT(ring, entry);
1903
1904			/* tiny buffer may follow */
1905			if (cp->tx_tiny_use[ring][entry].used) {
1906				cp->tx_tiny_use[ring][entry].used = 0;
1907				entry = TX_DESC_NEXT(ring, entry);
1908			}
1909		}
1910
1911		spin_lock(&cp->stat_lock[ring]);
1912		cp->net_stats[ring].tx_packets++;
1913		cp->net_stats[ring].tx_bytes += skb->len;
1914		spin_unlock(&cp->stat_lock[ring]);
1915		dev_kfree_skb_irq(skb);
1916	}
1917	cp->tx_old[ring] = entry;
1918
1919	/* this is wrong for multiple tx rings. the net device needs
1920	 * multiple queues for this to do the right thing.  we wait
1921	 * for 2*packets to be available when using tiny buffers
1922	 */
1923	if (netif_queue_stopped(dev) &&
1924	    (TX_BUFFS_AVAIL(cp, ring) > CAS_TABORT(cp)*(MAX_SKB_FRAGS + 1)))
1925		netif_wake_queue(dev);
1926	spin_unlock(&cp->tx_lock[ring]);
1927}
1928
1929static void cas_tx(struct net_device *dev, struct cas *cp,
1930		   u32 status)
1931{
1932        int limit, ring;
1933#ifdef USE_TX_COMPWB
1934	u64 compwb = le64_to_cpu(cp->init_block->tx_compwb);
1935#endif
1936	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
1937		     "tx interrupt, status: 0x%x, %llx\n",
1938		     status, (unsigned long long)compwb);
1939	/* process all the rings */
1940	for (ring = 0; ring < N_TX_RINGS; ring++) {
1941#ifdef USE_TX_COMPWB
1942		/* use the completion writeback registers */
1943		limit = (CAS_VAL(TX_COMPWB_MSB, compwb) << 8) |
1944			CAS_VAL(TX_COMPWB_LSB, compwb);
1945		compwb = TX_COMPWB_NEXT(compwb);
1946#else
1947		limit = readl(cp->regs + REG_TX_COMPN(ring));
1948#endif
1949		if (cp->tx_old[ring] != limit)
1950			cas_tx_ringN(cp, ring, limit);
1951	}
1952}
1953
1954
1955static int cas_rx_process_pkt(struct cas *cp, struct cas_rx_comp *rxc,
1956			      int entry, const u64 *words,
1957			      struct sk_buff **skbref)
1958{
1959	int dlen, hlen, len, i, alloclen;
1960	int off, swivel = RX_SWIVEL_OFF_VAL;
1961	struct cas_page *page;
1962	struct sk_buff *skb;
1963	void *addr, *crcaddr;
1964	__sum16 csum;
1965	char *p;
1966
1967	hlen = CAS_VAL(RX_COMP2_HDR_SIZE, words[1]);
1968	dlen = CAS_VAL(RX_COMP1_DATA_SIZE, words[0]);
1969	len  = hlen + dlen;
1970
1971	if (RX_COPY_ALWAYS || (words[2] & RX_COMP3_SMALL_PKT))
1972		alloclen = len;
1973	else
1974		alloclen = max(hlen, RX_COPY_MIN);
1975
1976	skb = netdev_alloc_skb(cp->dev, alloclen + swivel + cp->crc_size);
1977	if (skb == NULL)
1978		return -1;
1979
1980	*skbref = skb;
1981	skb_reserve(skb, swivel);
1982
1983	p = skb->data;
1984	addr = crcaddr = NULL;
1985	if (hlen) { /* always copy header pages */
1986		i = CAS_VAL(RX_COMP2_HDR_INDEX, words[1]);
1987		page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
1988		off = CAS_VAL(RX_COMP2_HDR_OFF, words[1]) * 0x100 +
1989			swivel;
1990
1991		i = hlen;
1992		if (!dlen) /* attach FCS */
1993			i += cp->crc_size;
1994		pci_dma_sync_single_for_cpu(cp->pdev, page->dma_addr + off, i,
1995				    PCI_DMA_FROMDEVICE);
1996		addr = cas_page_map(page->buffer);
1997		memcpy(p, addr + off, i);
1998		pci_dma_sync_single_for_device(cp->pdev, page->dma_addr + off, i,
1999				    PCI_DMA_FROMDEVICE);
2000		cas_page_unmap(addr);
2001		RX_USED_ADD(page, 0x100);
2002		p += hlen;
2003		swivel = 0;
2004	}
2005
2006
2007	if (alloclen < (hlen + dlen)) {
2008		skb_frag_t *frag = skb_shinfo(skb)->frags;
2009
2010		/* normal or jumbo packets. we use frags */
2011		i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
2012		page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
2013		off = CAS_VAL(RX_COMP1_DATA_OFF, words[0]) + swivel;
2014
2015		hlen = min(cp->page_size - off, dlen);
2016		if (hlen < 0) {
2017			netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
2018				     "rx page overflow: %d\n", hlen);
2019			dev_kfree_skb_irq(skb);
2020			return -1;
2021		}
2022		i = hlen;
2023		if (i == dlen)  /* attach FCS */
2024			i += cp->crc_size;
2025		pci_dma_sync_single_for_cpu(cp->pdev, page->dma_addr + off, i,
2026				    PCI_DMA_FROMDEVICE);
2027
2028		/* make sure we always copy a header */
2029		swivel = 0;
2030		if (p == (char *) skb->data) { /* not split */
2031			addr = cas_page_map(page->buffer);
2032			memcpy(p, addr + off, RX_COPY_MIN);
2033			pci_dma_sync_single_for_device(cp->pdev, page->dma_addr + off, i,
2034					PCI_DMA_FROMDEVICE);
2035			cas_page_unmap(addr);
2036			off += RX_COPY_MIN;
2037			swivel = RX_COPY_MIN;
2038			RX_USED_ADD(page, cp->mtu_stride);
2039		} else {
2040			RX_USED_ADD(page, hlen);
2041		}
2042		skb_put(skb, alloclen);
2043
2044		skb_shinfo(skb)->nr_frags++;
2045		skb->data_len += hlen - swivel;
2046		skb->truesize += hlen - swivel;
2047		skb->len      += hlen - swivel;
2048
2049		__skb_frag_set_page(frag, page->buffer);
2050		__skb_frag_ref(frag);
2051		frag->page_offset = off;
2052		skb_frag_size_set(frag, hlen - swivel);
2053
2054		/* any more data? */
2055		if ((words[0] & RX_COMP1_SPLIT_PKT) && ((dlen -= hlen) > 0)) {
2056			hlen = dlen;
2057			off = 0;
2058
2059			i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
2060			page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
2061			pci_dma_sync_single_for_cpu(cp->pdev, page->dma_addr,
2062					    hlen + cp->crc_size,
2063					    PCI_DMA_FROMDEVICE);
2064			pci_dma_sync_single_for_device(cp->pdev, page->dma_addr,
2065					    hlen + cp->crc_size,
2066					    PCI_DMA_FROMDEVICE);
2067
2068			skb_shinfo(skb)->nr_frags++;
2069			skb->data_len += hlen;
2070			skb->len      += hlen;
2071			frag++;
2072
2073			__skb_frag_set_page(frag, page->buffer);
2074			__skb_frag_ref(frag);
2075			frag->page_offset = 0;
2076			skb_frag_size_set(frag, hlen);
2077			RX_USED_ADD(page, hlen + cp->crc_size);
2078		}
2079
2080		if (cp->crc_size) {
2081			addr = cas_page_map(page->buffer);
2082			crcaddr  = addr + off + hlen;
2083		}
2084
2085	} else {
2086		/* copying packet */
2087		if (!dlen)
2088			goto end_copy_pkt;
2089
2090		i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
2091		page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
2092		off = CAS_VAL(RX_COMP1_DATA_OFF, words[0]) + swivel;
2093		hlen = min(cp->page_size - off, dlen);
2094		if (hlen < 0) {
2095			netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
2096				     "rx page overflow: %d\n", hlen);
2097			dev_kfree_skb_irq(skb);
2098			return -1;
2099		}
2100		i = hlen;
2101		if (i == dlen) /* attach FCS */
2102			i += cp->crc_size;
2103		pci_dma_sync_single_for_cpu(cp->pdev, page->dma_addr + off, i,
2104				    PCI_DMA_FROMDEVICE);
2105		addr = cas_page_map(page->buffer);
2106		memcpy(p, addr + off, i);
2107		pci_dma_sync_single_for_device(cp->pdev, page->dma_addr + off, i,
2108				    PCI_DMA_FROMDEVICE);
2109		cas_page_unmap(addr);
2110		if (p == (char *) skb->data) /* not split */
2111			RX_USED_ADD(page, cp->mtu_stride);
2112		else
2113			RX_USED_ADD(page, i);
2114
2115		/* any more data? */
2116		if ((words[0] & RX_COMP1_SPLIT_PKT) && ((dlen -= hlen) > 0)) {
2117			p += hlen;
2118			i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
2119			page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
2120			pci_dma_sync_single_for_cpu(cp->pdev, page->dma_addr,
2121					    dlen + cp->crc_size,
2122					    PCI_DMA_FROMDEVICE);
2123			addr = cas_page_map(page->buffer);
2124			memcpy(p, addr, dlen + cp->crc_size);
2125			pci_dma_sync_single_for_device(cp->pdev, page->dma_addr,
2126					    dlen + cp->crc_size,
2127					    PCI_DMA_FROMDEVICE);
2128			cas_page_unmap(addr);
2129			RX_USED_ADD(page, dlen + cp->crc_size);
2130		}
2131end_copy_pkt:
2132		if (cp->crc_size) {
2133			addr    = NULL;
2134			crcaddr = skb->data + alloclen;
2135		}
2136		skb_put(skb, alloclen);
2137	}
2138
2139	csum = (__force __sum16)htons(CAS_VAL(RX_COMP4_TCP_CSUM, words[3]));
2140	if (cp->crc_size) {
2141		/* checksum includes FCS. strip it out. */
2142		csum = csum_fold(csum_partial(crcaddr, cp->crc_size,
2143					      csum_unfold(csum)));
2144		if (addr)
2145			cas_page_unmap(addr);
2146	}
2147	skb->protocol = eth_type_trans(skb, cp->dev);
2148	if (skb->protocol == htons(ETH_P_IP)) {
2149		skb->csum = csum_unfold(~csum);
2150		skb->ip_summed = CHECKSUM_COMPLETE;
2151	} else
2152		skb_checksum_none_assert(skb);
2153	return len;
2154}
2155
2156
2157/* we can handle up to 64 rx flows at a time. we do the same thing
2158 * as nonreassm except that we batch up the buffers.
2159 * NOTE: we currently just treat each flow as a bunch of packets that
2160 *       we pass up. a better way would be to coalesce the packets
2161 *       into a jumbo packet. to do that, we need to do the following:
2162 *       1) the first packet will have a clean split between header and
2163 *          data. save both.
2164 *       2) each time the next flow packet comes in, extend the
2165 *          data length and merge the checksums.
2166 *       3) on flow release, fix up the header.
2167 *       4) make sure the higher layer doesn't care.
2168 * because packets get coalesced, we shouldn't run into fragment count
2169 * issues.
2170 */
2171static inline void cas_rx_flow_pkt(struct cas *cp, const u64 *words,
2172				   struct sk_buff *skb)
2173{
2174	int flowid = CAS_VAL(RX_COMP3_FLOWID, words[2]) & (N_RX_FLOWS - 1);
2175	struct sk_buff_head *flow = &cp->rx_flows[flowid];
2176
2177	/* this is protected at a higher layer, so no need to
2178	 * do any additional locking here. stick the buffer
2179	 * at the end.
2180	 */
2181	__skb_queue_tail(flow, skb);
2182	if (words[0] & RX_COMP1_RELEASE_FLOW) {
2183		while ((skb = __skb_dequeue(flow))) {
2184			cas_skb_release(skb);
2185		}
2186	}
2187}
2188
2189/* put rx descriptor back on ring. if a buffer is in use by a higher
2190 * layer, this will need to put in a replacement.
2191 */
2192static void cas_post_page(struct cas *cp, const int ring, const int index)
2193{
2194	cas_page_t *new;
2195	int entry;
2196
2197	entry = cp->rx_old[ring];
2198
2199	new = cas_page_swap(cp, ring, index);
2200	cp->init_rxds[ring][entry].buffer = cpu_to_le64(new->dma_addr);
2201	cp->init_rxds[ring][entry].index  =
2202		cpu_to_le64(CAS_BASE(RX_INDEX_NUM, index) |
2203			    CAS_BASE(RX_INDEX_RING, ring));
2204
2205	entry = RX_DESC_ENTRY(ring, entry + 1);
2206	cp->rx_old[ring] = entry;
2207
2208	if (entry % 4)
2209		return;
2210
2211	if (ring == 0)
2212		writel(entry, cp->regs + REG_RX_KICK);
2213	else if ((N_RX_DESC_RINGS > 1) &&
2214		 (cp->cas_flags & CAS_FLAG_REG_PLUS))
2215		writel(entry, cp->regs + REG_PLUS_RX_KICK1);
2216}
2217
2218
2219/* only when things are bad */
2220static int cas_post_rxds_ringN(struct cas *cp, int ring, int num)
2221{
2222	unsigned int entry, last, count, released;
2223	int cluster;
2224	cas_page_t **page = cp->rx_pages[ring];
2225
2226	entry = cp->rx_old[ring];
2227
2228	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
2229		     "rxd[%d] interrupt, done: %d\n", ring, entry);
2230
2231	cluster = -1;
2232	count = entry & 0x3;
2233	last = RX_DESC_ENTRY(ring, num ? entry + num - 4: entry - 4);
2234	released = 0;
2235	while (entry != last) {
2236		/* make a new buffer if it's still in use */
2237		if (page_count(page[entry]->buffer) > 1) {
2238			cas_page_t *new = cas_page_dequeue(cp);
2239			if (!new) {
2240				/* let the timer know that we need to
2241				 * do this again
2242				 */
2243				cp->cas_flags |= CAS_FLAG_RXD_POST(ring);
2244				if (!timer_pending(&cp->link_timer))
2245					mod_timer(&cp->link_timer, jiffies +
2246						  CAS_LINK_FAST_TIMEOUT);
2247				cp->rx_old[ring]  = entry;
2248				cp->rx_last[ring] = num ? num - released : 0;
2249				return -ENOMEM;
2250			}
2251			spin_lock(&cp->rx_inuse_lock);
2252			list_add(&page[entry]->list, &cp->rx_inuse_list);
2253			spin_unlock(&cp->rx_inuse_lock);
2254			cp->init_rxds[ring][entry].buffer =
2255				cpu_to_le64(new->dma_addr);
2256			page[entry] = new;
2257
2258		}
2259
2260		if (++count == 4) {
2261			cluster = entry;
2262			count = 0;
2263		}
2264		released++;
2265		entry = RX_DESC_ENTRY(ring, entry + 1);
2266	}
2267	cp->rx_old[ring] = entry;
2268
2269	if (cluster < 0)
2270		return 0;
2271
2272	if (ring == 0)
2273		writel(cluster, cp->regs + REG_RX_KICK);
2274	else if ((N_RX_DESC_RINGS > 1) &&
2275		 (cp->cas_flags & CAS_FLAG_REG_PLUS))
2276		writel(cluster, cp->regs + REG_PLUS_RX_KICK1);
2277	return 0;
2278}
2279
2280
2281/* process a completion ring. packets are set up in three basic ways:
2282 * small packets: should be copied header + data in single buffer.
2283 * large packets: header and data in a single buffer.
2284 * split packets: header in a separate buffer from data.
2285 *                data may be in multiple pages. data may be > 256
2286 *                bytes but in a single page.
2287 *
2288 * NOTE: RX page posting is done in this routine as well. while there's
2289 *       the capability of using multiple RX completion rings, it isn't
2290 *       really worthwhile due to the fact that the page posting will
2291 *       force serialization on the single descriptor ring.
2292 */
2293static int cas_rx_ringN(struct cas *cp, int ring, int budget)
2294{
2295	struct cas_rx_comp *rxcs = cp->init_rxcs[ring];
2296	int entry, drops;
2297	int npackets = 0;
2298
2299	netif_printk(cp, intr, KERN_DEBUG, cp->dev,
2300		     "rx[%d] interrupt, done: %d/%d\n",
2301		     ring,
2302		     readl(cp->regs + REG_RX_COMP_HEAD), cp->rx_new[ring]);
2303
2304	entry = cp->rx_new[ring];
2305	drops = 0;
2306	while (1) {
2307		struct cas_rx_comp *rxc = rxcs + entry;
2308		struct sk_buff *uninitialized_var(skb);
2309		int type, len;
2310		u64 words[4];
2311		int i, dring;
2312
2313		words[0] = le64_to_cpu(rxc->word1);
2314		words[1] = le64_to_cpu(rxc->word2);
2315		words[2] = le64_to_cpu(rxc->word3);
2316		words[3] = le64_to_cpu(rxc->word4);
2317
2318		/* don't touch if still owned by hw */
2319		type = CAS_VAL(RX_COMP1_TYPE, words[0]);
2320		if (type == 0)
2321			break;
2322
2323		/* hw hasn't cleared the zero bit yet */
2324		if (words[3] & RX_COMP4_ZERO) {
2325			break;
2326		}
2327
2328		/* get info on the packet */
2329		if (words[3] & (RX_COMP4_LEN_MISMATCH | RX_COMP4_BAD)) {
2330			spin_lock(&cp->stat_lock[ring]);
2331			cp->net_stats[ring].rx_errors++;
2332			if (words[3] & RX_COMP4_LEN_MISMATCH)
2333				cp->net_stats[ring].rx_length_errors++;
2334			if (words[3] & RX_COMP4_BAD)
2335				cp->net_stats[ring].rx_crc_errors++;
2336			spin_unlock(&cp->stat_lock[ring]);
2337
2338			/* We'll just return it to Cassini. */
2339		drop_it:
2340			spin_lock(&cp->stat_lock[ring]);
2341			++cp->net_stats[ring].rx_dropped;
2342			spin_unlock(&cp->stat_lock[ring]);
2343			goto next;
2344		}
2345
2346		len = cas_rx_process_pkt(cp, rxc, entry, words, &skb);
2347		if (len < 0) {
2348			++drops;
2349			goto drop_it;
2350		}
2351
2352		/* see if it's a flow re-assembly or not. the driver
2353		 * itself handles release back up.
2354		 */
2355		if (RX_DONT_BATCH || (type == 0x2)) {
2356			/* non-reassm: these always get released */
2357			cas_skb_release(skb);
2358		} else {
2359			cas_rx_flow_pkt(cp, words, skb);
2360		}
2361
2362		spin_lock(&cp->stat_lock[ring]);
2363		cp->net_stats[ring].rx_packets++;
2364		cp->net_stats[ring].rx_bytes += len;
2365		spin_unlock(&cp->stat_lock[ring]);
2366
2367	next:
2368		npackets++;
2369
2370		/* should it be released? */
2371		if (words[0] & RX_COMP1_RELEASE_HDR) {
2372			i = CAS_VAL(RX_COMP2_HDR_INDEX, words[1]);
2373			dring = CAS_VAL(RX_INDEX_RING, i);
2374			i = CAS_VAL(RX_INDEX_NUM, i);
2375			cas_post_page(cp, dring, i);
2376		}
2377
2378		if (words[0] & RX_COMP1_RELEASE_DATA) {
2379			i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
2380			dring = CAS_VAL(RX_INDEX_RING, i);
2381			i = CAS_VAL(RX_INDEX_NUM, i);
2382			cas_post_page(cp, dring, i);
2383		}
2384
2385		if (words[0] & RX_COMP1_RELEASE_NEXT) {
2386			i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
2387			dring = CAS_VAL(RX_INDEX_RING, i);
2388			i = CAS_VAL(RX_INDEX_NUM, i);
2389			cas_post_page(cp, dring, i);
2390		}
2391
2392		/* skip to the next entry */
2393		entry = RX_COMP_ENTRY(ring, entry + 1 +
2394				      CAS_VAL(RX_COMP1_SKIP, words[0]));
2395#ifdef USE_NAPI
2396		if (budget && (npackets >= budget))
2397			break;
2398#endif
2399	}
2400	cp->rx_new[ring] = entry;
2401
2402	if (drops)
2403		netdev_info(cp->dev, "Memory squeeze, deferring packet\n");
2404	return npackets;
2405}
2406
2407
2408/* put completion entries back on the ring */
2409static void cas_post_rxcs_ringN(struct net_device *dev,
2410				struct cas *cp, int ring)
2411{
2412	struct cas_rx_comp *rxc = cp->init_rxcs[ring];
2413	int last, entry;
2414
2415	last = cp->rx_cur[ring];
2416	entry = cp->rx_new[ring];
2417	netif_printk(cp, intr, KERN_DEBUG, dev,
2418		     "rxc[%d] interrupt, done: %d/%d\n",
2419		     ring, readl(cp->regs + REG_RX_COMP_HEAD), entry);
2420
2421	/* zero and re-mark descriptors */
2422	while (last != entry) {
2423		cas_rxc_init(rxc + last);
2424		last = RX_COMP_ENTRY(ring, last + 1);
2425	}
2426	cp->rx_cur[ring] = last;
2427
2428	if (ring == 0)
2429		writel(last, cp->regs + REG_RX_COMP_TAIL);
2430	else if (cp->cas_flags & CAS_FLAG_REG_PLUS)
2431		writel(last, cp->regs + REG_PLUS_RX_COMPN_TAIL(ring));
2432}
2433
2434
2435
2436/* cassini can use all four PCI interrupts for the completion ring.
2437 * rings 3 and 4 are identical
2438 */
2439#if defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
2440static inline void cas_handle_irqN(struct net_device *dev,
2441				   struct cas *cp, const u32 status,
2442				   const int ring)
2443{
2444	if (status & (INTR_RX_COMP_FULL_ALT | INTR_RX_COMP_AF_ALT))
2445		cas_post_rxcs_ringN(dev, cp, ring);
2446}
2447
2448static irqreturn_t cas_interruptN(int irq, void *dev_id)
2449{
2450	struct net_device *dev = dev_id;
2451	struct cas *cp = netdev_priv(dev);
2452	unsigned long flags;
2453	int ring = (irq == cp->pci_irq_INTC) ? 2 : 3;
2454	u32 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(ring));
2455
2456	/* check for shared irq */
2457	if (status == 0)
2458		return IRQ_NONE;
2459
2460	spin_lock_irqsave(&cp->lock, flags);
2461	if (status & INTR_RX_DONE_ALT) { /* handle rx separately */
2462#ifdef USE_NAPI
2463		cas_mask_intr(cp);
2464		napi_schedule(&cp->napi);
2465#else
2466		cas_rx_ringN(cp, ring, 0);
2467#endif
2468		status &= ~INTR_RX_DONE_ALT;
2469	}
2470
2471	if (status)
2472		cas_handle_irqN(dev, cp, status, ring);
2473	spin_unlock_irqrestore(&cp->lock, flags);
2474	return IRQ_HANDLED;
2475}
2476#endif
2477
2478#ifdef USE_PCI_INTB
2479/* everything but rx packets */
2480static inline void cas_handle_irq1(struct cas *cp, const u32 status)
2481{
2482	if (status & INTR_RX_BUF_UNAVAIL_1) {
2483		/* Frame arrived, no free RX buffers available.
2484		 * NOTE: we can get this on a link transition. */
2485		cas_post_rxds_ringN(cp, 1, 0);
2486		spin_lock(&cp->stat_lock[1]);
2487		cp->net_stats[1].rx_dropped++;
2488		spin_unlock(&cp->stat_lock[1]);
2489	}
2490
2491	if (status & INTR_RX_BUF_AE_1)
2492		cas_post_rxds_ringN(cp, 1, RX_DESC_RINGN_SIZE(1) -
2493				    RX_AE_FREEN_VAL(1));
2494
2495	if (status & (INTR_RX_COMP_AF | INTR_RX_COMP_FULL))
2496		cas_post_rxcs_ringN(cp, 1);
2497}
2498
2499/* ring 2 handles a few more events than 3 and 4 */
2500static irqreturn_t cas_interrupt1(int irq, void *dev_id)
2501{
2502	struct net_device *dev = dev_id;
2503	struct cas *cp = netdev_priv(dev);
2504	unsigned long flags;
2505	u32 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(1));
2506
2507	/* check for shared interrupt */
2508	if (status == 0)
2509		return IRQ_NONE;
2510
2511	spin_lock_irqsave(&cp->lock, flags);
2512	if (status & INTR_RX_DONE_ALT) { /* handle rx separately */
2513#ifdef USE_NAPI
2514		cas_mask_intr(cp);
2515		napi_schedule(&cp->napi);
2516#else
2517		cas_rx_ringN(cp, 1, 0);
2518#endif
2519		status &= ~INTR_RX_DONE_ALT;
2520	}
2521	if (status)
2522		cas_handle_irq1(cp, status);
2523	spin_unlock_irqrestore(&cp->lock, flags);
2524	return IRQ_HANDLED;
2525}
2526#endif
2527
2528static inline void cas_handle_irq(struct net_device *dev,
2529				  struct cas *cp, const u32 status)
2530{
2531	/* housekeeping interrupts */
2532	if (status & INTR_ERROR_MASK)
2533		cas_abnormal_irq(dev, cp, status);
2534
2535	if (status & INTR_RX_BUF_UNAVAIL) {
2536		/* Frame arrived, no free RX buffers available.
2537		 * NOTE: we can get this on a link transition.
2538		 */
2539		cas_post_rxds_ringN(cp, 0, 0);
2540		spin_lock(&cp->stat_lock[0]);
2541		cp->net_stats[0].rx_dropped++;
2542		spin_unlock(&cp->stat_lock[0]);
2543	} else if (status & INTR_RX_BUF_AE) {
2544		cas_post_rxds_ringN(cp, 0, RX_DESC_RINGN_SIZE(0) -
2545				    RX_AE_FREEN_VAL(0));
2546	}
2547
2548	if (status & (INTR_RX_COMP_AF | INTR_RX_COMP_FULL))
2549		cas_post_rxcs_ringN(dev, cp, 0);
2550}
2551
2552static irqreturn_t cas_interrupt(int irq, void *dev_id)
2553{
2554	struct net_device *dev = dev_id;
2555	struct cas *cp = netdev_priv(dev);
2556	unsigned long flags;
2557	u32 status = readl(cp->regs + REG_INTR_STATUS);
2558
2559	if (status == 0)
2560		return IRQ_NONE;
2561
2562	spin_lock_irqsave(&cp->lock, flags);
2563	if (status & (INTR_TX_ALL | INTR_TX_INTME)) {
2564		cas_tx(dev, cp, status);
2565		status &= ~(INTR_TX_ALL | INTR_TX_INTME);
2566	}
2567
2568	if (status & INTR_RX_DONE) {
2569#ifdef USE_NAPI
2570		cas_mask_intr(cp);
2571		napi_schedule(&cp->napi);
2572#else
2573		cas_rx_ringN(cp, 0, 0);
2574#endif
2575		status &= ~INTR_RX_DONE;
2576	}
2577
2578	if (status)
2579		cas_handle_irq(dev, cp, status);
2580	spin_unlock_irqrestore(&cp->lock, flags);
2581	return IRQ_HANDLED;
2582}
2583
2584
2585#ifdef USE_NAPI
2586static int cas_poll(struct napi_struct *napi, int budget)
2587{
2588	struct cas *cp = container_of(napi, struct cas, napi);
2589	struct net_device *dev = cp->dev;
2590	int i, enable_intr, credits;
2591	u32 status = readl(cp->regs + REG_INTR_STATUS);
2592	unsigned long flags;
2593
2594	spin_lock_irqsave(&cp->lock, flags);
2595	cas_tx(dev, cp, status);
2596	spin_unlock_irqrestore(&cp->lock, flags);
2597
2598	/* NAPI rx packets. we spread the credits across all of the
2599	 * rxc rings
2600	 *
2601	 * to make sure we're fair with the work we loop through each
2602	 * ring N_RX_COMP_RING times with a request of
2603	 * budget / N_RX_COMP_RINGS
2604	 */
2605	enable_intr = 1;
2606	credits = 0;
2607	for (i = 0; i < N_RX_COMP_RINGS; i++) {
2608		int j;
2609		for (j = 0; j < N_RX_COMP_RINGS; j++) {
2610			credits += cas_rx_ringN(cp, j, budget / N_RX_COMP_RINGS);
2611			if (credits >= budget) {
2612				enable_intr = 0;
2613				goto rx_comp;
2614			}
2615		}
2616	}
2617
2618rx_comp:
2619	/* final rx completion */
2620	spin_lock_irqsave(&cp->lock, flags);
2621	if (status)
2622		cas_handle_irq(dev, cp, status);
2623
2624#ifdef USE_PCI_INTB
2625	if (N_RX_COMP_RINGS > 1) {
2626		status = readl(cp->regs + REG_PLUS_INTRN_STATUS(1));
2627		if (status)
2628			cas_handle_irq1(dev, cp, status);
2629	}
2630#endif
2631
2632#ifdef USE_PCI_INTC
2633	if (N_RX_COMP_RINGS > 2) {
2634		status = readl(cp->regs + REG_PLUS_INTRN_STATUS(2));
2635		if (status)
2636			cas_handle_irqN(dev, cp, status, 2);
2637	}
2638#endif
2639
2640#ifdef USE_PCI_INTD
2641	if (N_RX_COMP_RINGS > 3) {
2642		status = readl(cp->regs + REG_PLUS_INTRN_STATUS(3));
2643		if (status)
2644			cas_handle_irqN(dev, cp, status, 3);
2645	}
2646#endif
2647	spin_unlock_irqrestore(&cp->lock, flags);
2648	if (enable_intr) {
2649		napi_complete(napi);
2650		cas_unmask_intr(cp);
2651	}
2652	return credits;
2653}
2654#endif
2655
2656#ifdef CONFIG_NET_POLL_CONTROLLER
2657static void cas_netpoll(struct net_device *dev)
2658{
2659	struct cas *cp = netdev_priv(dev);
2660
2661	cas_disable_irq(cp, 0);
2662	cas_interrupt(cp->pdev->irq, dev);
2663	cas_enable_irq(cp, 0);
2664
2665#ifdef USE_PCI_INTB
2666	if (N_RX_COMP_RINGS > 1) {
2667		/* cas_interrupt1(); */
2668	}
2669#endif
2670#ifdef USE_PCI_INTC
2671	if (N_RX_COMP_RINGS > 2) {
2672		/* cas_interruptN(); */
2673	}
2674#endif
2675#ifdef USE_PCI_INTD
2676	if (N_RX_COMP_RINGS > 3) {
2677		/* cas_interruptN(); */
2678	}
2679#endif
2680}
2681#endif
2682
2683static void cas_tx_timeout(struct net_device *dev)
2684{
2685	struct cas *cp = netdev_priv(dev);
2686
2687	netdev_err(dev, "transmit timed out, resetting\n");
2688	if (!cp->hw_running) {
2689		netdev_err(dev, "hrm.. hw not running!\n");
2690		return;
2691	}
2692
2693	netdev_err(dev, "MIF_STATE[%08x]\n",
2694		   readl(cp->regs + REG_MIF_STATE_MACHINE));
2695
2696	netdev_err(dev, "MAC_STATE[%08x]\n",
2697		   readl(cp->regs + REG_MAC_STATE_MACHINE));
2698
2699	netdev_err(dev, "TX_STATE[%08x:%08x:%08x] FIFO[%08x:%08x:%08x] SM1[%08x] SM2[%08x]\n",
2700		   readl(cp->regs + REG_TX_CFG),
2701		   readl(cp->regs + REG_MAC_TX_STATUS),
2702		   readl(cp->regs + REG_MAC_TX_CFG),
2703		   readl(cp->regs + REG_TX_FIFO_PKT_CNT),
2704		   readl(cp->regs + REG_TX_FIFO_WRITE_PTR),
2705		   readl(cp->regs + REG_TX_FIFO_READ_PTR),
2706		   readl(cp->regs + REG_TX_SM_1),
2707		   readl(cp->regs + REG_TX_SM_2));
2708
2709	netdev_err(dev, "RX_STATE[%08x:%08x:%08x]\n",
2710		   readl(cp->regs + REG_RX_CFG),
2711		   readl(cp->regs + REG_MAC_RX_STATUS),
2712		   readl(cp->regs + REG_MAC_RX_CFG));
2713
2714	netdev_err(dev, "HP_STATE[%08x:%08x:%08x:%08x]\n",
2715		   readl(cp->regs + REG_HP_STATE_MACHINE),
2716		   readl(cp->regs + REG_HP_STATUS0),
2717		   readl(cp->regs + REG_HP_STATUS1),
2718		   readl(cp->regs + REG_HP_STATUS2));
2719
2720#if 1
2721	atomic_inc(&cp->reset_task_pending);
2722	atomic_inc(&cp->reset_task_pending_all);
2723	schedule_work(&cp->reset_task);
2724#else
2725	atomic_set(&cp->reset_task_pending, CAS_RESET_ALL);
2726	schedule_work(&cp->reset_task);
2727#endif
2728}
2729
2730static inline int cas_intme(int ring, int entry)
2731{
2732	/* Algorithm: IRQ every 1/2 of descriptors. */
2733	if (!(entry & ((TX_DESC_RINGN_SIZE(ring) >> 1) - 1)))
2734		return 1;
2735	return 0;
2736}
2737
2738
2739static void cas_write_txd(struct cas *cp, int ring, int entry,
2740			  dma_addr_t mapping, int len, u64 ctrl, int last)
2741{
2742	struct cas_tx_desc *txd = cp->init_txds[ring] + entry;
2743
2744	ctrl |= CAS_BASE(TX_DESC_BUFLEN, len);
2745	if (cas_intme(ring, entry))
2746		ctrl |= TX_DESC_INTME;
2747	if (last)
2748		ctrl |= TX_DESC_EOF;
2749	txd->control = cpu_to_le64(ctrl);
2750	txd->buffer = cpu_to_le64(mapping);
2751}
2752
2753static inline void *tx_tiny_buf(struct cas *cp, const int ring,
2754				const int entry)
2755{
2756	return cp->tx_tiny_bufs[ring] + TX_TINY_BUF_LEN*entry;
2757}
2758
2759static inline dma_addr_t tx_tiny_map(struct cas *cp, const int ring,
2760				     const int entry, const int tentry)
2761{
2762	cp->tx_tiny_use[ring][tentry].nbufs++;
2763	cp->tx_tiny_use[ring][entry].used = 1;
2764	return cp->tx_tiny_dvma[ring] + TX_TINY_BUF_LEN*entry;
2765}
2766
2767static inline int cas_xmit_tx_ringN(struct cas *cp, int ring,
2768				    struct sk_buff *skb)
2769{
2770	struct net_device *dev = cp->dev;
2771	int entry, nr_frags, frag, tabort, tentry;
2772	dma_addr_t mapping;
2773	unsigned long flags;
2774	u64 ctrl;
2775	u32 len;
2776
2777	spin_lock_irqsave(&cp->tx_lock[ring], flags);
2778
2779	/* This is a hard error, log it. */
2780	if (TX_BUFFS_AVAIL(cp, ring) <=
2781	    CAS_TABORT(cp)*(skb_shinfo(skb)->nr_frags + 1)) {
2782		netif_stop_queue(dev);
2783		spin_unlock_irqrestore(&cp->tx_lock[ring], flags);
2784		netdev_err(dev, "BUG! Tx Ring full when queue awake!\n");
2785		return 1;
2786	}
2787
2788	ctrl = 0;
2789	if (skb->ip_summed == CHECKSUM_PARTIAL) {
2790		const u64 csum_start_off = skb_checksum_start_offset(skb);
2791		const u64 csum_stuff_off = csum_start_off + skb->csum_offset;
2792
2793		ctrl =  TX_DESC_CSUM_EN |
2794			CAS_BASE(TX_DESC_CSUM_START, csum_start_off) |
2795			CAS_BASE(TX_DESC_CSUM_STUFF, csum_stuff_off);
2796	}
2797
2798	entry = cp->tx_new[ring];
2799	cp->tx_skbs[ring][entry] = skb;
2800
2801	nr_frags = skb_shinfo(skb)->nr_frags;
2802	len = skb_headlen(skb);
2803	mapping = pci_map_page(cp->pdev, virt_to_page(skb->data),
2804			       offset_in_page(skb->data), len,
2805			       PCI_DMA_TODEVICE);
2806
2807	tentry = entry;
2808	tabort = cas_calc_tabort(cp, (unsigned long) skb->data, len);
2809	if (unlikely(tabort)) {
2810		/* NOTE: len is always >  tabort */
2811		cas_write_txd(cp, ring, entry, mapping, len - tabort,
2812			      ctrl | TX_DESC_SOF, 0);
2813		entry = TX_DESC_NEXT(ring, entry);
2814
2815		skb_copy_from_linear_data_offset(skb, len - tabort,
2816			      tx_tiny_buf(cp, ring, entry), tabort);
2817		mapping = tx_tiny_map(cp, ring, entry, tentry);
2818		cas_write_txd(cp, ring, entry, mapping, tabort, ctrl,
2819			      (nr_frags == 0));
2820	} else {
2821		cas_write_txd(cp, ring, entry, mapping, len, ctrl |
2822			      TX_DESC_SOF, (nr_frags == 0));
2823	}
2824	entry = TX_DESC_NEXT(ring, entry);
2825
2826	for (frag = 0; frag < nr_frags; frag++) {
2827		const skb_frag_t *fragp = &skb_shinfo(skb)->frags[frag];
2828
2829		len = skb_frag_size(fragp);
2830		mapping = skb_frag_dma_map(&cp->pdev->dev, fragp, 0, len,
2831					   DMA_TO_DEVICE);
2832
2833		tabort = cas_calc_tabort(cp, fragp->page_offset, len);
2834		if (unlikely(tabort)) {
2835			void *addr;
2836
2837			/* NOTE: len is always > tabort */
2838			cas_write_txd(cp, ring, entry, mapping, len - tabort,
2839				      ctrl, 0);
2840			entry = TX_DESC_NEXT(ring, entry);
2841
2842			addr = cas_page_map(skb_frag_page(fragp));
2843			memcpy(tx_tiny_buf(cp, ring, entry),
2844			       addr + fragp->page_offset + len - tabort,
2845			       tabort);
2846			cas_page_unmap(addr);
2847			mapping = tx_tiny_map(cp, ring, entry, tentry);
2848			len     = tabort;
2849		}
2850
2851		cas_write_txd(cp, ring, entry, mapping, len, ctrl,
2852			      (frag + 1 == nr_frags));
2853		entry = TX_DESC_NEXT(ring, entry);
2854	}
2855
2856	cp->tx_new[ring] = entry;
2857	if (TX_BUFFS_AVAIL(cp, ring) <= CAS_TABORT(cp)*(MAX_SKB_FRAGS + 1))
2858		netif_stop_queue(dev);
2859
2860	netif_printk(cp, tx_queued, KERN_DEBUG, dev,
2861		     "tx[%d] queued, slot %d, skblen %d, avail %d\n",
2862		     ring, entry, skb->len, TX_BUFFS_AVAIL(cp, ring));
2863	writel(entry, cp->regs + REG_TX_KICKN(ring));
2864	spin_unlock_irqrestore(&cp->tx_lock[ring], flags);
2865	return 0;
2866}
2867
2868static netdev_tx_t cas_start_xmit(struct sk_buff *skb, struct net_device *dev)
2869{
2870	struct cas *cp = netdev_priv(dev);
2871
2872	/* this is only used as a load-balancing hint, so it doesn't
2873	 * need to be SMP safe
2874	 */
2875	static int ring;
2876
2877	if (skb_padto(skb, cp->min_frame_size))
2878		return NETDEV_TX_OK;
2879
2880	/* XXX: we need some higher-level QoS hooks to steer packets to
2881	 *      individual queues.
2882	 */
2883	if (cas_xmit_tx_ringN(cp, ring++ & N_TX_RINGS_MASK, skb))
2884		return NETDEV_TX_BUSY;
2885	return NETDEV_TX_OK;
2886}
2887
2888static void cas_init_tx_dma(struct cas *cp)
2889{
2890	u64 desc_dma = cp->block_dvma;
2891	unsigned long off;
2892	u32 val;
2893	int i;
2894
2895	/* set up tx completion writeback registers. must be 8-byte aligned */
2896#ifdef USE_TX_COMPWB
2897	off = offsetof(struct cas_init_block, tx_compwb);
2898	writel((desc_dma + off) >> 32, cp->regs + REG_TX_COMPWB_DB_HI);
2899	writel((desc_dma + off) & 0xffffffff, cp->regs + REG_TX_COMPWB_DB_LOW);
2900#endif
2901
2902	/* enable completion writebacks, enable paced mode,
2903	 * disable read pipe, and disable pre-interrupt compwbs
2904	 */
2905	val =   TX_CFG_COMPWB_Q1 | TX_CFG_COMPWB_Q2 |
2906		TX_CFG_COMPWB_Q3 | TX_CFG_COMPWB_Q4 |
2907		TX_CFG_DMA_RDPIPE_DIS | TX_CFG_PACED_MODE |
2908		TX_CFG_INTR_COMPWB_DIS;
2909
2910	/* write out tx ring info and tx desc bases */
2911	for (i = 0; i < MAX_TX_RINGS; i++) {
2912		off = (unsigned long) cp->init_txds[i] -
2913			(unsigned long) cp->init_block;
2914
2915		val |= CAS_TX_RINGN_BASE(i);
2916		writel((desc_dma + off) >> 32, cp->regs + REG_TX_DBN_HI(i));
2917		writel((desc_dma + off) & 0xffffffff, cp->regs +
2918		       REG_TX_DBN_LOW(i));
2919		/* don't zero out the kick register here as the system
2920		 * will wedge
2921		 */
2922	}
2923	writel(val, cp->regs + REG_TX_CFG);
2924
2925	/* program max burst sizes. these numbers should be different
2926	 * if doing QoS.
2927	 */
2928#ifdef USE_QOS
2929	writel(0x800, cp->regs + REG_TX_MAXBURST_0);
2930	writel(0x1600, cp->regs + REG_TX_MAXBURST_1);
2931	writel(0x2400, cp->regs + REG_TX_MAXBURST_2);
2932	writel(0x4800, cp->regs + REG_TX_MAXBURST_3);
2933#else
2934	writel(0x800, cp->regs + REG_TX_MAXBURST_0);
2935	writel(0x800, cp->regs + REG_TX_MAXBURST_1);
2936	writel(0x800, cp->regs + REG_TX_MAXBURST_2);
2937	writel(0x800, cp->regs + REG_TX_MAXBURST_3);
2938#endif
2939}
2940
2941/* Must be invoked under cp->lock. */
2942static inline void cas_init_dma(struct cas *cp)
2943{
2944	cas_init_tx_dma(cp);
2945	cas_init_rx_dma(cp);
2946}
2947
2948static void cas_process_mc_list(struct cas *cp)
2949{
2950	u16 hash_table[16];
2951	u32 crc;
2952	struct netdev_hw_addr *ha;
2953	int i = 1;
2954
2955	memset(hash_table, 0, sizeof(hash_table));
2956	netdev_for_each_mc_addr(ha, cp->dev) {
2957		if (i <= CAS_MC_EXACT_MATCH_SIZE) {
2958			/* use the alternate mac address registers for the
2959			 * first 15 multicast addresses
2960			 */
2961			writel((ha->addr[4] << 8) | ha->addr[5],
2962			       cp->regs + REG_MAC_ADDRN(i*3 + 0));
2963			writel((ha->addr[2] << 8) | ha->addr[3],
2964			       cp->regs + REG_MAC_ADDRN(i*3 + 1));
2965			writel((ha->addr[0] << 8) | ha->addr[1],
2966			       cp->regs + REG_MAC_ADDRN(i*3 + 2));
2967			i++;
2968		}
2969		else {
2970			/* use hw hash table for the next series of
2971			 * multicast addresses
2972			 */
2973			crc = ether_crc_le(ETH_ALEN, ha->addr);
2974			crc >>= 24;
2975			hash_table[crc >> 4] |= 1 << (15 - (crc & 0xf));
2976		}
2977	}
2978	for (i = 0; i < 16; i++)
2979		writel(hash_table[i], cp->regs + REG_MAC_HASH_TABLEN(i));
2980}
2981
2982/* Must be invoked under cp->lock. */
2983static u32 cas_setup_multicast(struct cas *cp)
2984{
2985	u32 rxcfg = 0;
2986	int i;
2987
2988	if (cp->dev->flags & IFF_PROMISC) {
2989		rxcfg |= MAC_RX_CFG_PROMISC_EN;
2990
2991	} else if (cp->dev->flags & IFF_ALLMULTI) {
2992	    	for (i=0; i < 16; i++)
2993			writel(0xFFFF, cp->regs + REG_MAC_HASH_TABLEN(i));
2994		rxcfg |= MAC_RX_CFG_HASH_FILTER_EN;
2995
2996	} else {
2997		cas_process_mc_list(cp);
2998		rxcfg |= MAC_RX_CFG_HASH_FILTER_EN;
2999	}
3000
3001	return rxcfg;
3002}
3003
3004/* must be invoked under cp->stat_lock[N_TX_RINGS] */
3005static void cas_clear_mac_err(struct cas *cp)
3006{
3007	writel(0, cp->regs + REG_MAC_COLL_NORMAL);
3008	writel(0, cp->regs + REG_MAC_COLL_FIRST);
3009	writel(0, cp->regs + REG_MAC_COLL_EXCESS);
3010	writel(0, cp->regs + REG_MAC_COLL_LATE);
3011	writel(0, cp->regs + REG_MAC_TIMER_DEFER);
3012	writel(0, cp->regs + REG_MAC_ATTEMPTS_PEAK);
3013	writel(0, cp->regs + REG_MAC_RECV_FRAME);
3014	writel(0, cp->regs + REG_MAC_LEN_ERR);
3015	writel(0, cp->regs + REG_MAC_ALIGN_ERR);
3016	writel(0, cp->regs + REG_MAC_FCS_ERR);
3017	writel(0, cp->regs + REG_MAC_RX_CODE_ERR);
3018}
3019
3020
3021static void cas_mac_reset(struct cas *cp)
3022{
3023	int i;
3024
3025	/* do both TX and RX reset */
3026	writel(0x1, cp->regs + REG_MAC_TX_RESET);
3027	writel(0x1, cp->regs + REG_MAC_RX_RESET);
3028
3029	/* wait for TX */
3030	i = STOP_TRIES;
3031	while (i-- > 0) {
3032		if (readl(cp->regs + REG_MAC_TX_RESET) == 0)
3033			break;
3034		udelay(10);
3035	}
3036
3037	/* wait for RX */
3038	i = STOP_TRIES;
3039	while (i-- > 0) {
3040		if (readl(cp->regs + REG_MAC_RX_RESET) == 0)
3041			break;
3042		udelay(10);
3043	}
3044
3045	if (readl(cp->regs + REG_MAC_TX_RESET) |
3046	    readl(cp->regs + REG_MAC_RX_RESET))
3047		netdev_err(cp->dev, "mac tx[%d]/rx[%d] reset failed [%08x]\n",
3048			   readl(cp->regs + REG_MAC_TX_RESET),
3049			   readl(cp->regs + REG_MAC_RX_RESET),
3050			   readl(cp->regs + REG_MAC_STATE_MACHINE));
3051}
3052
3053
3054/* Must be invoked under cp->lock. */
3055static void cas_init_mac(struct cas *cp)
3056{
3057	unsigned char *e = &cp->dev->dev_addr[0];
3058	int i;
3059	cas_mac_reset(cp);
3060
3061	/* setup core arbitration weight register */
3062	writel(CAWR_RR_DIS, cp->regs + REG_CAWR);
3063
3064	/* XXX Use pci_dma_burst_advice() */
3065#if !defined(CONFIG_SPARC64) && !defined(CONFIG_ALPHA)
3066	/* set the infinite burst register for chips that don't have
3067	 * pci issues.
3068	 */
3069	if ((cp->cas_flags & CAS_FLAG_TARGET_ABORT) == 0)
3070		writel(INF_BURST_EN, cp->regs + REG_INF_BURST);
3071#endif
3072
3073	writel(0x1BF0, cp->regs + REG_MAC_SEND_PAUSE);
3074
3075	writel(0x00, cp->regs + REG_MAC_IPG0);
3076	writel(0x08, cp->regs + REG_MAC_IPG1);
3077	writel(0x04, cp->regs + REG_MAC_IPG2);
3078
3079	/* change later for 802.3z */
3080	writel(0x40, cp->regs + REG_MAC_SLOT_TIME);
3081
3082	/* min frame + FCS */
3083	writel(ETH_ZLEN + 4, cp->regs + REG_MAC_FRAMESIZE_MIN);
3084
3085	/* Ethernet payload + header + FCS + optional VLAN tag. NOTE: we
3086	 * specify the maximum frame size to prevent RX tag errors on
3087	 * oversized frames.
3088	 */
3089	writel(CAS_BASE(MAC_FRAMESIZE_MAX_BURST, 0x2000) |
3090	       CAS_BASE(MAC_FRAMESIZE_MAX_FRAME,
3091			(CAS_MAX_MTU + ETH_HLEN + 4 + 4)),
3092	       cp->regs + REG_MAC_FRAMESIZE_MAX);
3093
3094	/* NOTE: crc_size is used as a surrogate for half-duplex.
3095	 * workaround saturn half-duplex issue by increasing preamble
3096	 * size to 65 bytes.
3097	 */
3098	if ((cp->cas_flags & CAS_FLAG_SATURN) && cp->crc_size)
3099		writel(0x41, cp->regs + REG_MAC_PA_SIZE);
3100	else
3101		writel(0x07, cp->regs + REG_MAC_PA_SIZE);
3102	writel(0x04, cp->regs + REG_MAC_JAM_SIZE);
3103	writel(0x10, cp->regs + REG_MAC_ATTEMPT_LIMIT);
3104	writel(0x8808, cp->regs + REG_MAC_CTRL_TYPE);
3105
3106	writel((e[5] | (e[4] << 8)) & 0x3ff, cp->regs + REG_MAC_RANDOM_SEED);
3107
3108	writel(0, cp->regs + REG_MAC_ADDR_FILTER0);
3109	writel(0, cp->regs + REG_MAC_ADDR_FILTER1);
3110	writel(0, cp->regs + REG_MAC_ADDR_FILTER2);
3111	writel(0, cp->regs + REG_MAC_ADDR_FILTER2_1_MASK);
3112	writel(0, cp->regs + REG_MAC_ADDR_FILTER0_MASK);
3113
3114	/* setup mac address in perfect filter array */
3115	for (i = 0; i < 45; i++)
3116		writel(0x0, cp->regs + REG_MAC_ADDRN(i));
3117
3118	writel((e[4] << 8) | e[5], cp->regs + REG_MAC_ADDRN(0));
3119	writel((e[2] << 8) | e[3], cp->regs + REG_MAC_ADDRN(1));
3120	writel((e[0] << 8) | e[1], cp->regs + REG_MAC_ADDRN(2));
3121
3122	writel(0x0001, cp->regs + REG_MAC_ADDRN(42));
3123	writel(0xc200, cp->regs + REG_MAC_ADDRN(43));
3124	writel(0x0180, cp->regs + REG_MAC_ADDRN(44));
3125
3126	cp->mac_rx_cfg = cas_setup_multicast(cp);
3127
3128	spin_lock(&cp->stat_lock[N_TX_RINGS]);
3129	cas_clear_mac_err(cp);
3130	spin_unlock(&cp->stat_lock[N_TX_RINGS]);
3131
3132	/* Setup MAC interrupts.  We want to get all of the interesting
3133	 * counter expiration events, but we do not want to hear about
3134	 * normal rx/tx as the DMA engine tells us that.
3135	 */
3136	writel(MAC_TX_FRAME_XMIT, cp->regs + REG_MAC_TX_MASK);
3137	writel(MAC_RX_FRAME_RECV, cp->regs + REG_MAC_RX_MASK);
3138
3139	/* Don't enable even the PAUSE interrupts for now, we
3140	 * make no use of those events other than to record them.
3141	 */
3142	writel(0xffffffff, cp->regs + REG_MAC_CTRL_MASK);
3143}
3144
3145/* Must be invoked under cp->lock. */
3146static void cas_init_pause_thresholds(struct cas *cp)
3147{
3148	/* Calculate pause thresholds.  Setting the OFF threshold to the
3149	 * full RX fifo size effectively disables PAUSE generation
3150	 */
3151	if (cp->rx_fifo_size <= (2 * 1024)) {
3152		cp->rx_pause_off = cp->rx_pause_on = cp->rx_fifo_size;
3153	} else {
3154		int max_frame = (cp->dev->mtu + ETH_HLEN + 4 + 4 + 64) & ~63;
3155		if (max_frame * 3 > cp->rx_fifo_size) {
3156			cp->rx_pause_off = 7104;
3157			cp->rx_pause_on  = 960;
3158		} else {
3159			int off = (cp->rx_fifo_size - (max_frame * 2));
3160			int on = off - max_frame;
3161			cp->rx_pause_off = off;
3162			cp->rx_pause_on = on;
3163		}
3164	}
3165}
3166
3167static int cas_vpd_match(const void __iomem *p, const char *str)
3168{
3169	int len = strlen(str) + 1;
3170	int i;
3171
3172	for (i = 0; i < len; i++) {
3173		if (readb(p + i) != str[i])
3174			return 0;
3175	}
3176	return 1;
3177}
3178
3179
3180/* get the mac address by reading the vpd information in the rom.
3181 * also get the phy type and determine if there's an entropy generator.
3182 * NOTE: this is a bit convoluted for the following reasons:
3183 *  1) vpd info has order-dependent mac addresses for multinic cards
3184 *  2) the only way to determine the nic order is to use the slot
3185 *     number.
3186 *  3) fiber cards don't have bridges, so their slot numbers don't
3187 *     mean anything.
3188 *  4) we don't actually know we have a fiber card until after
3189 *     the mac addresses are parsed.
3190 */
3191static int cas_get_vpd_info(struct cas *cp, unsigned char *dev_addr,
3192			    const int offset)
3193{
3194	void __iomem *p = cp->regs + REG_EXPANSION_ROM_RUN_START;
3195	void __iomem *base, *kstart;
3196	int i, len;
3197	int found = 0;
3198#define VPD_FOUND_MAC        0x01
3199#define VPD_FOUND_PHY        0x02
3200
3201	int phy_type = CAS_PHY_MII_MDIO0; /* default phy type */
3202	int mac_off  = 0;
3203
3204#if defined(CONFIG_SPARC)
3205	const unsigned char *addr;
3206#endif
3207
3208	/* give us access to the PROM */
3209	writel(BIM_LOCAL_DEV_PROM | BIM_LOCAL_DEV_PAD,
3210	       cp->regs + REG_BIM_LOCAL_DEV_EN);
3211
3212	/* check for an expansion rom */
3213	if (readb(p) != 0x55 || readb(p + 1) != 0xaa)
3214		goto use_random_mac_addr;
3215
3216	/* search for beginning of vpd */
3217	base = NULL;
3218	for (i = 2; i < EXPANSION_ROM_SIZE; i++) {
3219		/* check for PCIR */
3220		if ((readb(p + i + 0) == 0x50) &&
3221		    (readb(p + i + 1) == 0x43) &&
3222		    (readb(p + i + 2) == 0x49) &&
3223		    (readb(p + i + 3) == 0x52)) {
3224			base = p + (readb(p + i + 8) |
3225				    (readb(p + i + 9) << 8));
3226			break;
3227		}
3228	}
3229
3230	if (!base || (readb(base) != 0x82))
3231		goto use_random_mac_addr;
3232
3233	i = (readb(base + 1) | (readb(base + 2) << 8)) + 3;
3234	while (i < EXPANSION_ROM_SIZE) {
3235		if (readb(base + i) != 0x90) /* no vpd found */
3236			goto use_random_mac_addr;
3237
3238		/* found a vpd field */
3239		len = readb(base + i + 1) | (readb(base + i + 2) << 8);
3240
3241		/* extract keywords */
3242		kstart = base + i + 3;
3243		p = kstart;
3244		while ((p - kstart) < len) {
3245			int klen = readb(p + 2);
3246			int j;
3247			char type;
3248
3249			p += 3;
3250
3251			/* look for the following things:
3252			 * -- correct length == 29
3253			 * 3 (type) + 2 (size) +
3254			 * 18 (strlen("local-mac-address") + 1) +
3255			 * 6 (mac addr)
3256			 * -- VPD Instance 'I'
3257			 * -- VPD Type Bytes 'B'
3258			 * -- VPD data length == 6
3259			 * -- property string == local-mac-address
3260			 *
3261			 * -- correct length == 24
3262			 * 3 (type) + 2 (size) +
3263			 * 12 (strlen("entropy-dev") + 1) +
3264			 * 7 (strlen("vms110") + 1)
3265			 * -- VPD Instance 'I'
3266			 * -- VPD Type String 'B'
3267			 * -- VPD data length == 7
3268			 * -- property string == entropy-dev
3269			 *
3270			 * -- correct length == 18
3271			 * 3 (type) + 2 (size) +
3272			 * 9 (strlen("phy-type") + 1) +
3273			 * 4 (strlen("pcs") + 1)
3274			 * -- VPD Instance 'I'
3275			 * -- VPD Type String 'S'
3276			 * -- VPD data length == 4
3277			 * -- property string == phy-type
3278			 *
3279			 * -- correct length == 23
3280			 * 3 (type) + 2 (size) +
3281			 * 14 (strlen("phy-interface") + 1) +
3282			 * 4 (strlen("pcs") + 1)
3283			 * -- VPD Instance 'I'
3284			 * -- VPD Type String 'S'
3285			 * -- VPD data length == 4
3286			 * -- property string == phy-interface
3287			 */
3288			if (readb(p) != 'I')
3289				goto next;
3290
3291			/* finally, check string and length */
3292			type = readb(p + 3);
3293			if (type == 'B') {
3294				if ((klen == 29) && readb(p + 4) == 6 &&
3295				    cas_vpd_match(p + 5,
3296						  "local-mac-address")) {
3297					if (mac_off++ > offset)
3298						goto next;
3299
3300					/* set mac address */
3301					for (j = 0; j < 6; j++)
3302						dev_addr[j] =
3303							readb(p + 23 + j);
3304					goto found_mac;
3305				}
3306			}
3307
3308			if (type != 'S')
3309				goto next;
3310
3311#ifdef USE_ENTROPY_DEV
3312			if ((klen == 24) &&
3313			    cas_vpd_match(p + 5, "entropy-dev") &&
3314			    cas_vpd_match(p + 17, "vms110")) {
3315				cp->cas_flags |= CAS_FLAG_ENTROPY_DEV;
3316				goto next;
3317			}
3318#endif
3319
3320			if (found & VPD_FOUND_PHY)
3321				goto next;
3322
3323			if ((klen == 18) && readb(p + 4) == 4 &&
3324			    cas_vpd_match(p + 5, "phy-type")) {
3325				if (cas_vpd_match(p + 14, "pcs")) {
3326					phy_type = CAS_PHY_SERDES;
3327					goto found_phy;
3328				}
3329			}
3330
3331			if ((klen == 23) && readb(p + 4) == 4 &&
3332			    cas_vpd_match(p + 5, "phy-interface")) {
3333				if (cas_vpd_match(p + 19, "pcs")) {
3334					phy_type = CAS_PHY_SERDES;
3335					goto found_phy;
3336				}
3337			}
3338found_mac:
3339			found |= VPD_FOUND_MAC;
3340			goto next;
3341
3342found_phy:
3343			found |= VPD_FOUND_PHY;
3344
3345next:
3346			p += klen;
3347		}
3348		i += len + 3;
3349	}
3350
3351use_random_mac_addr:
3352	if (found & VPD_FOUND_MAC)
3353		goto done;
3354
3355#if defined(CONFIG_SPARC)
3356	addr = of_get_property(cp->of_node, "local-mac-address", NULL);
3357	if (addr != NULL) {
3358		memcpy(dev_addr, addr, 6);
3359		goto done;
3360	}
3361#endif
3362
3363	/* Sun MAC prefix then 3 random bytes. */
3364	pr_info("MAC address not found in ROM VPD\n");
3365	dev_addr[0] = 0x08;
3366	dev_addr[1] = 0x00;
3367	dev_addr[2] = 0x20;
3368	get_random_bytes(dev_addr + 3, 3);
3369
3370done:
3371	writel(0, cp->regs + REG_BIM_LOCAL_DEV_EN);
3372	return phy_type;
3373}
3374
3375/* check pci invariants */
3376static void cas_check_pci_invariants(struct cas *cp)
3377{
3378	struct pci_dev *pdev = cp->pdev;
3379
3380	cp->cas_flags = 0;
3381	if ((pdev->vendor == PCI_VENDOR_ID_SUN) &&
3382	    (pdev->device == PCI_DEVICE_ID_SUN_CASSINI)) {
3383		if (pdev->revision >= CAS_ID_REVPLUS)
3384			cp->cas_flags |= CAS_FLAG_REG_PLUS;
3385		if (pdev->revision < CAS_ID_REVPLUS02u)
3386			cp->cas_flags |= CAS_FLAG_TARGET_ABORT;
3387
3388		/* Original Cassini supports HW CSUM, but it's not
3389		 * enabled by default as it can trigger TX hangs.
3390		 */
3391		if (pdev->revision < CAS_ID_REV2)
3392			cp->cas_flags |= CAS_FLAG_NO_HW_CSUM;
3393	} else {
3394		/* Only sun has original cassini chips.  */
3395		cp->cas_flags |= CAS_FLAG_REG_PLUS;
3396
3397		/* We use a flag because the same phy might be externally
3398		 * connected.
3399		 */
3400		if ((pdev->vendor == PCI_VENDOR_ID_NS) &&
3401		    (pdev->device == PCI_DEVICE_ID_NS_SATURN))
3402			cp->cas_flags |= CAS_FLAG_SATURN;
3403	}
3404}
3405
3406
3407static int cas_check_invariants(struct cas *cp)
3408{
3409	struct pci_dev *pdev = cp->pdev;
3410	u32 cfg;
3411	int i;
3412
3413	/* get page size for rx buffers. */
3414	cp->page_order = 0;
3415#ifdef USE_PAGE_ORDER
3416	if (PAGE_SHIFT < CAS_JUMBO_PAGE_SHIFT) {
3417		/* see if we can allocate larger pages */
3418		struct page *page = alloc_pages(GFP_ATOMIC,
3419						CAS_JUMBO_PAGE_SHIFT -
3420						PAGE_SHIFT);
3421		if (page) {
3422			__free_pages(page, CAS_JUMBO_PAGE_SHIFT - PAGE_SHIFT);
3423			cp->page_order = CAS_JUMBO_PAGE_SHIFT - PAGE_SHIFT;
3424		} else {
3425			printk("MTU limited to %d bytes\n", CAS_MAX_MTU);
3426		}
3427	}
3428#endif
3429	cp->page_size = (PAGE_SIZE << cp->page_order);
3430
3431	/* Fetch the FIFO configurations. */
3432	cp->tx_fifo_size = readl(cp->regs + REG_TX_FIFO_SIZE) * 64;
3433	cp->rx_fifo_size = RX_FIFO_SIZE;
3434
3435	/* finish phy determination. MDIO1 takes precedence over MDIO0 if
3436	 * they're both connected.
3437	 */
3438	cp->phy_type = cas_get_vpd_info(cp, cp->dev->dev_addr,
3439					PCI_SLOT(pdev->devfn));
3440	if (cp->phy_type & CAS_PHY_SERDES) {
3441		cp->cas_flags |= CAS_FLAG_1000MB_CAP;
3442		return 0; /* no more checking needed */
3443	}
3444
3445	/* MII */
3446	cfg = readl(cp->regs + REG_MIF_CFG);
3447	if (cfg & MIF_CFG_MDIO_1) {
3448		cp->phy_type = CAS_PHY_MII_MDIO1;
3449	} else if (cfg & MIF_CFG_MDIO_0) {
3450		cp->phy_type = CAS_PHY_MII_MDIO0;
3451	}
3452
3453	cas_mif_poll(cp, 0);
3454	writel(PCS_DATAPATH_MODE_MII, cp->regs + REG_PCS_DATAPATH_MODE);
3455
3456	for (i = 0; i < 32; i++) {
3457		u32 phy_id;
3458		int j;
3459
3460		for (j = 0; j < 3; j++) {
3461			cp->phy_addr = i;
3462			phy_id = cas_phy_read(cp, MII_PHYSID1) << 16;
3463			phy_id |= cas_phy_read(cp, MII_PHYSID2);
3464			if (phy_id && (phy_id != 0xFFFFFFFF)) {
3465				cp->phy_id = phy_id;
3466				goto done;
3467			}
3468		}
3469	}
3470	pr_err("MII phy did not respond [%08x]\n",
3471	       readl(cp->regs + REG_MIF_STATE_MACHINE));
3472	return -1;
3473
3474done:
3475	/* see if we can do gigabit */
3476	cfg = cas_phy_read(cp, MII_BMSR);
3477	if ((cfg & CAS_BMSR_1000_EXTEND) &&
3478	    cas_phy_read(cp, CAS_MII_1000_EXTEND))
3479		cp->cas_flags |= CAS_FLAG_1000MB_CAP;
3480	return 0;
3481}
3482
3483/* Must be invoked under cp->lock. */
3484static inline void cas_start_dma(struct cas *cp)
3485{
3486	int i;
3487	u32 val;
3488	int txfailed = 0;
3489
3490	/* enable dma */
3491	val = readl(cp->regs + REG_TX_CFG) | TX_CFG_DMA_EN;
3492	writel(val, cp->regs + REG_TX_CFG);
3493	val = readl(cp->regs + REG_RX_CFG) | RX_CFG_DMA_EN;
3494	writel(val, cp->regs + REG_RX_CFG);
3495
3496	/* enable the mac */
3497	val = readl(cp->regs + REG_MAC_TX_CFG) | MAC_TX_CFG_EN;
3498	writel(val, cp->regs + REG_MAC_TX_CFG);
3499	val = readl(cp->regs + REG_MAC_RX_CFG) | MAC_RX_CFG_EN;
3500	writel(val, cp->regs + REG_MAC_RX_CFG);
3501
3502	i = STOP_TRIES;
3503	while (i-- > 0) {
3504		val = readl(cp->regs + REG_MAC_TX_CFG);
3505		if ((val & MAC_TX_CFG_EN))
3506			break;
3507		udelay(10);
3508	}
3509	if (i < 0) txfailed = 1;
3510	i = STOP_TRIES;
3511	while (i-- > 0) {
3512		val = readl(cp->regs + REG_MAC_RX_CFG);
3513		if ((val & MAC_RX_CFG_EN)) {
3514			if (txfailed) {
3515				netdev_err(cp->dev,
3516					   "enabling mac failed [tx:%08x:%08x]\n",
3517					   readl(cp->regs + REG_MIF_STATE_MACHINE),
3518					   readl(cp->regs + REG_MAC_STATE_MACHINE));
3519			}
3520			goto enable_rx_done;
3521		}
3522		udelay(10);
3523	}
3524	netdev_err(cp->dev, "enabling mac failed [%s:%08x:%08x]\n",
3525		   (txfailed ? "tx,rx" : "rx"),
3526		   readl(cp->regs + REG_MIF_STATE_MACHINE),
3527		   readl(cp->regs + REG_MAC_STATE_MACHINE));
3528
3529enable_rx_done:
3530	cas_unmask_intr(cp); /* enable interrupts */
3531	writel(RX_DESC_RINGN_SIZE(0) - 4, cp->regs + REG_RX_KICK);
3532	writel(0, cp->regs + REG_RX_COMP_TAIL);
3533
3534	if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
3535		if (N_RX_DESC_RINGS > 1)
3536			writel(RX_DESC_RINGN_SIZE(1) - 4,
3537			       cp->regs + REG_PLUS_RX_KICK1);
3538
3539		for (i = 1; i < N_RX_COMP_RINGS; i++)
3540			writel(0, cp->regs + REG_PLUS_RX_COMPN_TAIL(i));
3541	}
3542}
3543
3544/* Must be invoked under cp->lock. */
3545static void cas_read_pcs_link_mode(struct cas *cp, int *fd, int *spd,
3546				   int *pause)
3547{
3548	u32 val = readl(cp->regs + REG_PCS_MII_LPA);
3549	*fd     = (val & PCS_MII_LPA_FD) ? 1 : 0;
3550	*pause  = (val & PCS_MII_LPA_SYM_PAUSE) ? 0x01 : 0x00;
3551	if (val & PCS_MII_LPA_ASYM_PAUSE)
3552		*pause |= 0x10;
3553	*spd = 1000;
3554}
3555
3556/* Must be invoked under cp->lock. */
3557static void cas_read_mii_link_mode(struct cas *cp, int *fd, int *spd,
3558				   int *pause)
3559{
3560	u32 val;
3561
3562	*fd = 0;
3563	*spd = 10;
3564	*pause = 0;
3565
3566	/* use GMII registers */
3567	val = cas_phy_read(cp, MII_LPA);
3568	if (val & CAS_LPA_PAUSE)
3569		*pause = 0x01;
3570
3571	if (val & CAS_LPA_ASYM_PAUSE)
3572		*pause |= 0x10;
3573
3574	if (val & LPA_DUPLEX)
3575		*fd = 1;
3576	if (val & LPA_100)
3577		*spd = 100;
3578
3579	if (cp->cas_flags & CAS_FLAG_1000MB_CAP) {
3580		val = cas_phy_read(cp, CAS_MII_1000_STATUS);
3581		if (val & (CAS_LPA_1000FULL | CAS_LPA_1000HALF))
3582			*spd = 1000;
3583		if (val & CAS_LPA_1000FULL)
3584			*fd = 1;
3585	}
3586}
3587
3588/* A link-up condition has occurred, initialize and enable the
3589 * rest of the chip.
3590 *
3591 * Must be invoked under cp->lock.
3592 */
3593static void cas_set_link_modes(struct cas *cp)
3594{
3595	u32 val;
3596	int full_duplex, speed, pause;
3597
3598	full_duplex = 0;
3599	speed = 10;
3600	pause = 0;
3601
3602	if (CAS_PHY_MII(cp->phy_type)) {
3603		cas_mif_poll(cp, 0);
3604		val = cas_phy_read(cp, MII_BMCR);
3605		if (val & BMCR_ANENABLE) {
3606			cas_read_mii_link_mode(cp, &full_duplex, &speed,
3607					       &pause);
3608		} else {
3609			if (val & BMCR_FULLDPLX)
3610				full_duplex = 1;
3611
3612			if (val & BMCR_SPEED100)
3613				speed = 100;
3614			else if (val & CAS_BMCR_SPEED1000)
3615				speed = (cp->cas_flags & CAS_FLAG_1000MB_CAP) ?
3616					1000 : 100;
3617		}
3618		cas_mif_poll(cp, 1);
3619
3620	} else {
3621		val = readl(cp->regs + REG_PCS_MII_CTRL);
3622		cas_read_pcs_link_mode(cp, &full_duplex, &speed, &pause);
3623		if ((val & PCS_MII_AUTONEG_EN) == 0) {
3624			if (val & PCS_MII_CTRL_DUPLEX)
3625				full_duplex = 1;
3626		}
3627	}
3628
3629	netif_info(cp, link, cp->dev, "Link up at %d Mbps, %s-duplex\n",
3630		   speed, full_duplex ? "full" : "half");
3631
3632	val = MAC_XIF_TX_MII_OUTPUT_EN | MAC_XIF_LINK_LED;
3633	if (CAS_PHY_MII(cp->phy_type)) {
3634		val |= MAC_XIF_MII_BUFFER_OUTPUT_EN;
3635		if (!full_duplex)
3636			val |= MAC_XIF_DISABLE_ECHO;
3637	}
3638	if (full_duplex)
3639		val |= MAC_XIF_FDPLX_LED;
3640	if (speed == 1000)
3641		val |= MAC_XIF_GMII_MODE;
3642	writel(val, cp->regs + REG_MAC_XIF_CFG);
3643
3644	/* deal with carrier and collision detect. */
3645	val = MAC_TX_CFG_IPG_EN;
3646	if (full_duplex) {
3647		val |= MAC_TX_CFG_IGNORE_CARRIER;
3648		val |= MAC_TX_CFG_IGNORE_COLL;
3649	} else {
3650#ifndef USE_CSMA_CD_PROTO
3651		val |= MAC_TX_CFG_NEVER_GIVE_UP_EN;
3652		val |= MAC_TX_CFG_NEVER_GIVE_UP_LIM;
3653#endif
3654	}
3655	/* val now set up for REG_MAC_TX_CFG */
3656
3657	/* If gigabit and half-duplex, enable carrier extension
3658	 * mode.  increase slot time to 512 bytes as well.
3659	 * else, disable it and make sure slot time is 64 bytes.
3660	 * also activate checksum bug workaround
3661	 */
3662	if ((speed == 1000) && !full_duplex) {
3663		writel(val | MAC_TX_CFG_CARRIER_EXTEND,
3664		       cp->regs + REG_MAC_TX_CFG);
3665
3666		val = readl(cp->regs + REG_MAC_RX_CFG);
3667		val &= ~MAC_RX_CFG_STRIP_FCS; /* checksum workaround */
3668		writel(val | MAC_RX_CFG_CARRIER_EXTEND,
3669		       cp->regs + REG_MAC_RX_CFG);
3670
3671		writel(0x200, cp->regs + REG_MAC_SLOT_TIME);
3672
3673		cp->crc_size = 4;
3674		/* minimum size gigabit frame at half duplex */
3675		cp->min_frame_size = CAS_1000MB_MIN_FRAME;
3676
3677	} else {
3678		writel(val, cp->regs + REG_MAC_TX_CFG);
3679
3680		/* checksum bug workaround. don't strip FCS when in
3681		 * half-duplex mode
3682		 */
3683		val = readl(cp->regs + REG_MAC_RX_CFG);
3684		if (full_duplex) {
3685			val |= MAC_RX_CFG_STRIP_FCS;
3686			cp->crc_size = 0;
3687			cp->min_frame_size = CAS_MIN_MTU;
3688		} else {
3689			val &= ~MAC_RX_CFG_STRIP_FCS;
3690			cp->crc_size = 4;
3691			cp->min_frame_size = CAS_MIN_FRAME;
3692		}
3693		writel(val & ~MAC_RX_CFG_CARRIER_EXTEND,
3694		       cp->regs + REG_MAC_RX_CFG);
3695		writel(0x40, cp->regs + REG_MAC_SLOT_TIME);
3696	}
3697
3698	if (netif_msg_link(cp)) {
3699		if (pause & 0x01) {
3700			netdev_info(cp->dev, "Pause is enabled (rxfifo: %d off: %d on: %d)\n",
3701				    cp->rx_fifo_size,
3702				    cp->rx_pause_off,
3703				    cp->rx_pause_on);
3704		} else if (pause & 0x10) {
3705			netdev_info(cp->dev, "TX pause enabled\n");
3706		} else {
3707			netdev_info(cp->dev, "Pause is disabled\n");
3708		}
3709	}
3710
3711	val = readl(cp->regs + REG_MAC_CTRL_CFG);
3712	val &= ~(MAC_CTRL_CFG_SEND_PAUSE_EN | MAC_CTRL_CFG_RECV_PAUSE_EN);
3713	if (pause) { /* symmetric or asymmetric pause */
3714		val |= MAC_CTRL_CFG_SEND_PAUSE_EN;
3715		if (pause & 0x01) { /* symmetric pause */
3716			val |= MAC_CTRL_CFG_RECV_PAUSE_EN;
3717		}
3718	}
3719	writel(val, cp->regs + REG_MAC_CTRL_CFG);
3720	cas_start_dma(cp);
3721}
3722
3723/* Must be invoked under cp->lock. */
3724static void cas_init_hw(struct cas *cp, int restart_link)
3725{
3726	if (restart_link)
3727		cas_phy_init(cp);
3728
3729	cas_init_pause_thresholds(cp);
3730	cas_init_mac(cp);
3731	cas_init_dma(cp);
3732
3733	if (restart_link) {
3734		/* Default aneg parameters */
3735		cp->timer_ticks = 0;
3736		cas_begin_auto_negotiation(cp, NULL);
3737	} else if (cp->lstate == link_up) {
3738		cas_set_link_modes(cp);
3739		netif_carrier_on(cp->dev);
3740	}
3741}
3742
3743/* Must be invoked under cp->lock. on earlier cassini boards,
3744 * SOFT_0 is tied to PCI reset. we use this to force a pci reset,
3745 * let it settle out, and then restore pci state.
3746 */
3747static void cas_hard_reset(struct cas *cp)
3748{
3749	writel(BIM_LOCAL_DEV_SOFT_0, cp->regs + REG_BIM_LOCAL_DEV_EN);
3750	udelay(20);
3751	pci_restore_state(cp->pdev);
3752}
3753
3754
3755static void cas_global_reset(struct cas *cp, int blkflag)
3756{
3757	int limit;
3758
3759	/* issue a global reset. don't use RSTOUT. */
3760	if (blkflag && !CAS_PHY_MII(cp->phy_type)) {
3761		/* For PCS, when the blkflag is set, we should set the
3762		 * SW_REST_BLOCK_PCS_SLINK bit to prevent the results of
3763		 * the last autonegotiation from being cleared.  We'll
3764		 * need some special handling if the chip is set into a
3765		 * loopback mode.
3766		 */
3767		writel((SW_RESET_TX | SW_RESET_RX | SW_RESET_BLOCK_PCS_SLINK),
3768		       cp->regs + REG_SW_RESET);
3769	} else {
3770		writel(SW_RESET_TX | SW_RESET_RX, cp->regs + REG_SW_RESET);
3771	}
3772
3773	/* need to wait at least 3ms before polling register */
3774	mdelay(3);
3775
3776	limit = STOP_TRIES;
3777	while (limit-- > 0) {
3778		u32 val = readl(cp->regs + REG_SW_RESET);
3779		if ((val & (SW_RESET_TX | SW_RESET_RX)) == 0)
3780			goto done;
3781		udelay(10);
3782	}
3783	netdev_err(cp->dev, "sw reset failed\n");
3784
3785done:
3786	/* enable various BIM interrupts */
3787	writel(BIM_CFG_DPAR_INTR_ENABLE | BIM_CFG_RMA_INTR_ENABLE |
3788	       BIM_CFG_RTA_INTR_ENABLE, cp->regs + REG_BIM_CFG);
3789
3790	/* clear out pci error status mask for handled errors.
3791	 * we don't deal with DMA counter overflows as they happen
3792	 * all the time.
3793	 */
3794	writel(0xFFFFFFFFU & ~(PCI_ERR_BADACK | PCI_ERR_DTRTO |
3795			       PCI_ERR_OTHER | PCI_ERR_BIM_DMA_WRITE |
3796			       PCI_ERR_BIM_DMA_READ), cp->regs +
3797	       REG_PCI_ERR_STATUS_MASK);
3798
3799	/* set up for MII by default to address mac rx reset timeout
3800	 * issue
3801	 */
3802	writel(PCS_DATAPATH_MODE_MII, cp->regs + REG_PCS_DATAPATH_MODE);
3803}
3804
3805static void cas_reset(struct cas *cp, int blkflag)
3806{
3807	u32 val;
3808
3809	cas_mask_intr(cp);
3810	cas_global_reset(cp, blkflag);
3811	cas_mac_reset(cp);
3812	cas_entropy_reset(cp);
3813
3814	/* disable dma engines. */
3815	val = readl(cp->regs + REG_TX_CFG);
3816	val &= ~TX_CFG_DMA_EN;
3817	writel(val, cp->regs + REG_TX_CFG);
3818
3819	val = readl(cp->regs + REG_RX_CFG);
3820	val &= ~RX_CFG_DMA_EN;
3821	writel(val, cp->regs + REG_RX_CFG);
3822
3823	/* program header parser */
3824	if ((cp->cas_flags & CAS_FLAG_TARGET_ABORT) ||
3825	    (CAS_HP_ALT_FIRMWARE == cas_prog_null)) {
3826		cas_load_firmware(cp, CAS_HP_FIRMWARE);
3827	} else {
3828		cas_load_firmware(cp, CAS_HP_ALT_FIRMWARE);
3829	}
3830
3831	/* clear out error registers */
3832	spin_lock(&cp->stat_lock[N_TX_RINGS]);
3833	cas_clear_mac_err(cp);
3834	spin_unlock(&cp->stat_lock[N_TX_RINGS]);
3835}
3836
3837/* Shut down the chip, must be called with pm_mutex held.  */
3838static void cas_shutdown(struct cas *cp)
3839{
3840	unsigned long flags;
3841
3842	/* Make us not-running to avoid timers respawning */
3843	cp->hw_running = 0;
3844
3845	del_timer_sync(&cp->link_timer);
3846
3847	/* Stop the reset task */
3848#if 0
3849	while (atomic_read(&cp->reset_task_pending_mtu) ||
3850	       atomic_read(&cp->reset_task_pending_spare) ||
3851	       atomic_read(&cp->reset_task_pending_all))
3852		schedule();
3853
3854#else
3855	while (atomic_read(&cp->reset_task_pending))
3856		schedule();
3857#endif
3858	/* Actually stop the chip */
3859	cas_lock_all_save(cp, flags);
3860	cas_reset(cp, 0);
3861	if (cp->cas_flags & CAS_FLAG_SATURN)
3862		cas_phy_powerdown(cp);
3863	cas_unlock_all_restore(cp, flags);
3864}
3865
3866static int cas_change_mtu(struct net_device *dev, int new_mtu)
3867{
3868	struct cas *cp = netdev_priv(dev);
3869
3870	if (new_mtu < CAS_MIN_MTU || new_mtu > CAS_MAX_MTU)
3871		return -EINVAL;
3872
3873	dev->mtu = new_mtu;
3874	if (!netif_running(dev) || !netif_device_present(dev))
3875		return 0;
3876
3877	/* let the reset task handle it */
3878#if 1
3879	atomic_inc(&cp->reset_task_pending);
3880	if ((cp->phy_type & CAS_PHY_SERDES)) {
3881		atomic_inc(&cp->reset_task_pending_all);
3882	} else {
3883		atomic_inc(&cp->reset_task_pending_mtu);
3884	}
3885	schedule_work(&cp->reset_task);
3886#else
3887	atomic_set(&cp->reset_task_pending, (cp->phy_type & CAS_PHY_SERDES) ?
3888		   CAS_RESET_ALL : CAS_RESET_MTU);
3889	pr_err("reset called in cas_change_mtu\n");
3890	schedule_work(&cp->reset_task);
3891#endif
3892
3893	flush_work_sync(&cp->reset_task);
3894	return 0;
3895}
3896
3897static void cas_clean_txd(struct cas *cp, int ring)
3898{
3899	struct cas_tx_desc *txd = cp->init_txds[ring];
3900	struct sk_buff *skb, **skbs = cp->tx_skbs[ring];
3901	u64 daddr, dlen;
3902	int i, size;
3903
3904	size = TX_DESC_RINGN_SIZE(ring);
3905	for (i = 0; i < size; i++) {
3906		int frag;
3907
3908		if (skbs[i] == NULL)
3909			continue;
3910
3911		skb = skbs[i];
3912		skbs[i] = NULL;
3913
3914		for (frag = 0; frag <= skb_shinfo(skb)->nr_frags;  frag++) {
3915			int ent = i & (size - 1);
3916
3917			/* first buffer is never a tiny buffer and so
3918			 * needs to be unmapped.
3919			 */
3920			daddr = le64_to_cpu(txd[ent].buffer);
3921			dlen  =  CAS_VAL(TX_DESC_BUFLEN,
3922					 le64_to_cpu(txd[ent].control));
3923			pci_unmap_page(cp->pdev, daddr, dlen,
3924				       PCI_DMA_TODEVICE);
3925
3926			if (frag != skb_shinfo(skb)->nr_frags) {
3927				i++;
3928
3929				/* next buffer might by a tiny buffer.
3930				 * skip past it.
3931				 */
3932				ent = i & (size - 1);
3933				if (cp->tx_tiny_use[ring][ent].used)
3934					i++;
3935			}
3936		}
3937		dev_kfree_skb_any(skb);
3938	}
3939
3940	/* zero out tiny buf usage */
3941	memset(cp->tx_tiny_use[ring], 0, size*sizeof(*cp->tx_tiny_use[ring]));
3942}
3943
3944/* freed on close */
3945static inline void cas_free_rx_desc(struct cas *cp, int ring)
3946{
3947	cas_page_t **page = cp->rx_pages[ring];
3948	int i, size;
3949
3950	size = RX_DESC_RINGN_SIZE(ring);
3951	for (i = 0; i < size; i++) {
3952		if (page[i]) {
3953			cas_page_free(cp, page[i]);
3954			page[i] = NULL;
3955		}
3956	}
3957}
3958
3959static void cas_free_rxds(struct cas *cp)
3960{
3961	int i;
3962
3963	for (i = 0; i < N_RX_DESC_RINGS; i++)
3964		cas_free_rx_desc(cp, i);
3965}
3966
3967/* Must be invoked under cp->lock. */
3968static void cas_clean_rings(struct cas *cp)
3969{
3970	int i;
3971
3972	/* need to clean all tx rings */
3973	memset(cp->tx_old, 0, sizeof(*cp->tx_old)*N_TX_RINGS);
3974	memset(cp->tx_new, 0, sizeof(*cp->tx_new)*N_TX_RINGS);
3975	for (i = 0; i < N_TX_RINGS; i++)
3976		cas_clean_txd(cp, i);
3977
3978	/* zero out init block */
3979	memset(cp->init_block, 0, sizeof(struct cas_init_block));
3980	cas_clean_rxds(cp);
3981	cas_clean_rxcs(cp);
3982}
3983
3984/* allocated on open */
3985static inline int cas_alloc_rx_desc(struct cas *cp, int ring)
3986{
3987	cas_page_t **page = cp->rx_pages[ring];
3988	int size, i = 0;
3989
3990	size = RX_DESC_RINGN_SIZE(ring);
3991	for (i = 0; i < size; i++) {
3992		if ((page[i] = cas_page_alloc(cp, GFP_KERNEL)) == NULL)
3993			return -1;
3994	}
3995	return 0;
3996}
3997
3998static int cas_alloc_rxds(struct cas *cp)
3999{
4000	int i;
4001
4002	for (i = 0; i < N_RX_DESC_RINGS; i++) {
4003		if (cas_alloc_rx_desc(cp, i) < 0) {
4004			cas_free_rxds(cp);
4005			return -1;
4006		}
4007	}
4008	return 0;
4009}
4010
4011static void cas_reset_task(struct work_struct *work)
4012{
4013	struct cas *cp = container_of(work, struct cas, reset_task);
4014#if 0
4015	int pending = atomic_read(&cp->reset_task_pending);
4016#else
4017	int pending_all = atomic_read(&cp->reset_task_pending_all);
4018	int pending_spare = atomic_read(&cp->reset_task_pending_spare);
4019	int pending_mtu = atomic_read(&cp->reset_task_pending_mtu);
4020
4021	if (pending_all == 0 && pending_spare == 0 && pending_mtu == 0) {
4022		/* We can have more tasks scheduled than actually
4023		 * needed.
4024		 */
4025		atomic_dec(&cp->reset_task_pending);
4026		return;
4027	}
4028#endif
4029	/* The link went down, we reset the ring, but keep
4030	 * DMA stopped. Use this function for reset
4031	 * on error as well.
4032	 */
4033	if (cp->hw_running) {
4034		unsigned long flags;
4035
4036		/* Make sure we don't get interrupts or tx packets */
4037		netif_device_detach(cp->dev);
4038		cas_lock_all_save(cp, flags);
4039
4040		if (cp->opened) {
4041			/* We call cas_spare_recover when we call cas_open.
4042			 * but we do not initialize the lists cas_spare_recover
4043			 * uses until cas_open is called.
4044			 */
4045			cas_spare_recover(cp, GFP_ATOMIC);
4046		}
4047#if 1
4048		/* test => only pending_spare set */
4049		if (!pending_all && !pending_mtu)
4050			goto done;
4051#else
4052		if (pending == CAS_RESET_SPARE)
4053			goto done;
4054#endif
4055		/* when pending == CAS_RESET_ALL, the following
4056		 * call to cas_init_hw will restart auto negotiation.
4057		 * Setting the second argument of cas_reset to
4058		 * !(pending == CAS_RESET_ALL) will set this argument
4059		 * to 1 (avoiding reinitializing the PHY for the normal
4060		 * PCS case) when auto negotiation is not restarted.
4061		 */
4062#if 1
4063		cas_reset(cp, !(pending_all > 0));
4064		if (cp->opened)
4065			cas_clean_rings(cp);
4066		cas_init_hw(cp, (pending_all > 0));
4067#else
4068		cas_reset(cp, !(pending == CAS_RESET_ALL));
4069		if (cp->opened)
4070			cas_clean_rings(cp);
4071		cas_init_hw(cp, pending == CAS_RESET_ALL);
4072#endif
4073
4074done:
4075		cas_unlock_all_restore(cp, flags);
4076		netif_device_attach(cp->dev);
4077	}
4078#if 1
4079	atomic_sub(pending_all, &cp->reset_task_pending_all);
4080	atomic_sub(pending_spare, &cp->reset_task_pending_spare);
4081	atomic_sub(pending_mtu, &cp->reset_task_pending_mtu);
4082	atomic_dec(&cp->reset_task_pending);
4083#else
4084	atomic_set(&cp->reset_task_pending, 0);
4085#endif
4086}
4087
4088static void cas_link_timer(unsigned long data)
4089{
4090	struct cas *cp = (struct cas *) data;
4091	int mask, pending = 0, reset = 0;
4092	unsigned long flags;
4093
4094	if (link_transition_timeout != 0 &&
4095	    cp->link_transition_jiffies_valid &&
4096	    ((jiffies - cp->link_transition_jiffies) >
4097	      (link_transition_timeout))) {
4098		/* One-second counter so link-down workaround doesn't
4099		 * cause resets to occur so fast as to fool the switch
4100		 * into thinking the link is down.
4101		 */
4102		cp->link_transition_jiffies_valid = 0;
4103	}
4104
4105	if (!cp->hw_running)
4106		return;
4107
4108	spin_lock_irqsave(&cp->lock, flags);
4109	cas_lock_tx(cp);
4110	cas_entropy_gather(cp);
4111
4112	/* If the link task is still pending, we just
4113	 * reschedule the link timer
4114	 */
4115#if 1
4116	if (atomic_read(&cp->reset_task_pending_all) ||
4117	    atomic_read(&cp->reset_task_pending_spare) ||
4118	    atomic_read(&cp->reset_task_pending_mtu))
4119		goto done;
4120#else
4121	if (atomic_read(&cp->reset_task_pending))
4122		goto done;
4123#endif
4124
4125	/* check for rx cleaning */
4126	if ((mask = (cp->cas_flags & CAS_FLAG_RXD_POST_MASK))) {
4127		int i, rmask;
4128
4129		for (i = 0; i < MAX_RX_DESC_RINGS; i++) {
4130			rmask = CAS_FLAG_RXD_POST(i);
4131			if ((mask & rmask) == 0)
4132				continue;
4133
4134			/* post_rxds will do a mod_timer */
4135			if (cas_post_rxds_ringN(cp, i, cp->rx_last[i]) < 0) {
4136				pending = 1;
4137				continue;
4138			}
4139			cp->cas_flags &= ~rmask;
4140		}
4141	}
4142
4143	if (CAS_PHY_MII(cp->phy_type)) {
4144		u16 bmsr;
4145		cas_mif_poll(cp, 0);
4146		bmsr = cas_phy_read(cp, MII_BMSR);
4147		/* WTZ: Solaris driver reads this twice, but that
4148		 * may be due to the PCS case and the use of a
4149		 * common implementation. Read it twice here to be
4150		 * safe.
4151		 */
4152		bmsr = cas_phy_read(cp, MII_BMSR);
4153		cas_mif_poll(cp, 1);
4154		readl(cp->regs + REG_MIF_STATUS); /* avoid dups */
4155		reset = cas_mii_link_check(cp, bmsr);
4156	} else {
4157		reset = cas_pcs_link_check(cp);
4158	}
4159
4160	if (reset)
4161		goto done;
4162
4163	/* check for tx state machine confusion */
4164	if ((readl(cp->regs + REG_MAC_TX_STATUS) & MAC_TX_FRAME_XMIT) == 0) {
4165		u32 val = readl(cp->regs + REG_MAC_STATE_MACHINE);
4166		u32 wptr, rptr;
4167		int tlm  = CAS_VAL(MAC_SM_TLM, val);
4168
4169		if (((tlm == 0x5) || (tlm == 0x3)) &&
4170		    (CAS_VAL(MAC_SM_ENCAP_SM, val) == 0)) {
4171			netif_printk(cp, tx_err, KERN_DEBUG, cp->dev,
4172				     "tx err: MAC_STATE[%08x]\n", val);
4173			reset = 1;
4174			goto done;
4175		}
4176
4177		val  = readl(cp->regs + REG_TX_FIFO_PKT_CNT);
4178		wptr = readl(cp->regs + REG_TX_FIFO_WRITE_PTR);
4179		rptr = readl(cp->regs + REG_TX_FIFO_READ_PTR);
4180		if ((val == 0) && (wptr != rptr)) {
4181			netif_printk(cp, tx_err, KERN_DEBUG, cp->dev,
4182				     "tx err: TX_FIFO[%08x:%08x:%08x]\n",
4183				     val, wptr, rptr);
4184			reset = 1;
4185		}
4186
4187		if (reset)
4188			cas_hard_reset(cp);
4189	}
4190
4191done:
4192	if (reset) {
4193#if 1
4194		atomic_inc(&cp->reset_task_pending);
4195		atomic_inc(&cp->reset_task_pending_all);
4196		schedule_work(&cp->reset_task);
4197#else
4198		atomic_set(&cp->reset_task_pending, CAS_RESET_ALL);
4199		pr_err("reset called in cas_link_timer\n");
4200		schedule_work(&cp->reset_task);
4201#endif
4202	}
4203
4204	if (!pending)
4205		mod_timer(&cp->link_timer, jiffies + CAS_LINK_TIMEOUT);
4206	cas_unlock_tx(cp);
4207	spin_unlock_irqrestore(&cp->lock, flags);
4208}
4209
4210/* tiny buffers are used to avoid target abort issues with
4211 * older cassini's
4212 */
4213static void cas_tx_tiny_free(struct cas *cp)
4214{
4215	struct pci_dev *pdev = cp->pdev;
4216	int i;
4217
4218	for (i = 0; i < N_TX_RINGS; i++) {
4219		if (!cp->tx_tiny_bufs[i])
4220			continue;
4221
4222		pci_free_consistent(pdev, TX_TINY_BUF_BLOCK,
4223				    cp->tx_tiny_bufs[i],
4224				    cp->tx_tiny_dvma[i]);
4225		cp->tx_tiny_bufs[i] = NULL;
4226	}
4227}
4228
4229static int cas_tx_tiny_alloc(struct cas *cp)
4230{
4231	struct pci_dev *pdev = cp->pdev;
4232	int i;
4233
4234	for (i = 0; i < N_TX_RINGS; i++) {
4235		cp->tx_tiny_bufs[i] =
4236			pci_alloc_consistent(pdev, TX_TINY_BUF_BLOCK,
4237					     &cp->tx_tiny_dvma[i]);
4238		if (!cp->tx_tiny_bufs[i]) {
4239			cas_tx_tiny_free(cp);
4240			return -1;
4241		}
4242	}
4243	return 0;
4244}
4245
4246
4247static int cas_open(struct net_device *dev)
4248{
4249	struct cas *cp = netdev_priv(dev);
4250	int hw_was_up, err;
4251	unsigned long flags;
4252
4253	mutex_lock(&cp->pm_mutex);
4254
4255	hw_was_up = cp->hw_running;
4256
4257	/* The power-management mutex protects the hw_running
4258	 * etc. state so it is safe to do this bit without cp->lock
4259	 */
4260	if (!cp->hw_running) {
4261		/* Reset the chip */
4262		cas_lock_all_save(cp, flags);
4263		/* We set the second arg to cas_reset to zero
4264		 * because cas_init_hw below will have its second
4265		 * argument set to non-zero, which will force
4266		 * autonegotiation to start.
4267		 */
4268		cas_reset(cp, 0);
4269		cp->hw_running = 1;
4270		cas_unlock_all_restore(cp, flags);
4271	}
4272
4273	err = -ENOMEM;
4274	if (cas_tx_tiny_alloc(cp) < 0)
4275		goto err_unlock;
4276
4277	/* alloc rx descriptors */
4278	if (cas_alloc_rxds(cp) < 0)
4279		goto err_tx_tiny;
4280
4281	/* allocate spares */
4282	cas_spare_init(cp);
4283	cas_spare_recover(cp, GFP_KERNEL);
4284
4285	/* We can now request the interrupt as we know it's masked
4286	 * on the controller. cassini+ has up to 4 interrupts
4287	 * that can be used, but you need to do explicit pci interrupt
4288	 * mapping to expose them
4289	 */
4290	if (request_irq(cp->pdev->irq, cas_interrupt,
4291			IRQF_SHARED, dev->name, (void *) dev)) {
4292		netdev_err(cp->dev, "failed to request irq !\n");
4293		err = -EAGAIN;
4294		goto err_spare;
4295	}
4296
4297#ifdef USE_NAPI
4298	napi_enable(&cp->napi);
4299#endif
4300	/* init hw */
4301	cas_lock_all_save(cp, flags);
4302	cas_clean_rings(cp);
4303	cas_init_hw(cp, !hw_was_up);
4304	cp->opened = 1;
4305	cas_unlock_all_restore(cp, flags);
4306
4307	netif_start_queue(dev);
4308	mutex_unlock(&cp->pm_mutex);
4309	return 0;
4310
4311err_spare:
4312	cas_spare_free(cp);
4313	cas_free_rxds(cp);
4314err_tx_tiny:
4315	cas_tx_tiny_free(cp);
4316err_unlock:
4317	mutex_unlock(&cp->pm_mutex);
4318	return err;
4319}
4320
4321static int cas_close(struct net_device *dev)
4322{
4323	unsigned long flags;
4324	struct cas *cp = netdev_priv(dev);
4325
4326#ifdef USE_NAPI
4327	napi_disable(&cp->napi);
4328#endif
4329	/* Make sure we don't get distracted by suspend/resume */
4330	mutex_lock(&cp->pm_mutex);
4331
4332	netif_stop_queue(dev);
4333
4334	/* Stop traffic, mark us closed */
4335	cas_lock_all_save(cp, flags);
4336	cp->opened = 0;
4337	cas_reset(cp, 0);
4338	cas_phy_init(cp);
4339	cas_begin_auto_negotiation(cp, NULL);
4340	cas_clean_rings(cp);
4341	cas_unlock_all_restore(cp, flags);
4342
4343	free_irq(cp->pdev->irq, (void *) dev);
4344	cas_spare_free(cp);
4345	cas_free_rxds(cp);
4346	cas_tx_tiny_free(cp);
4347	mutex_unlock(&cp->pm_mutex);
4348	return 0;
4349}
4350
4351static struct {
4352	const char name[ETH_GSTRING_LEN];
4353} ethtool_cassini_statnames[] = {
4354	{"collisions"},
4355	{"rx_bytes"},
4356	{"rx_crc_errors"},
4357	{"rx_dropped"},
4358	{"rx_errors"},
4359	{"rx_fifo_errors"},
4360	{"rx_frame_errors"},
4361	{"rx_length_errors"},
4362	{"rx_over_errors"},
4363	{"rx_packets"},
4364	{"tx_aborted_errors"},
4365	{"tx_bytes"},
4366	{"tx_dropped"},
4367	{"tx_errors"},
4368	{"tx_fifo_errors"},
4369	{"tx_packets"}
4370};
4371#define CAS_NUM_STAT_KEYS ARRAY_SIZE(ethtool_cassini_statnames)
4372
4373static struct {
4374	const int offsets;	/* neg. values for 2nd arg to cas_read_phy */
4375} ethtool_register_table[] = {
4376	{-MII_BMSR},
4377	{-MII_BMCR},
4378	{REG_CAWR},
4379	{REG_INF_BURST},
4380	{REG_BIM_CFG},
4381	{REG_RX_CFG},
4382	{REG_HP_CFG},
4383	{REG_MAC_TX_CFG},
4384	{REG_MAC_RX_CFG},
4385	{REG_MAC_CTRL_CFG},
4386	{REG_MAC_XIF_CFG},
4387	{REG_MIF_CFG},
4388	{REG_PCS_CFG},
4389	{REG_SATURN_PCFG},
4390	{REG_PCS_MII_STATUS},
4391	{REG_PCS_STATE_MACHINE},
4392	{REG_MAC_COLL_EXCESS},
4393	{REG_MAC_COLL_LATE}
4394};
4395#define CAS_REG_LEN 	ARRAY_SIZE(ethtool_register_table)
4396#define CAS_MAX_REGS 	(sizeof (u32)*CAS_REG_LEN)
4397
4398static void cas_read_regs(struct cas *cp, u8 *ptr, int len)
4399{
4400	u8 *p;
4401	int i;
4402	unsigned long flags;
4403
4404	spin_lock_irqsave(&cp->lock, flags);
4405	for (i = 0, p = ptr; i < len ; i ++, p += sizeof(u32)) {
4406		u16 hval;
4407		u32 val;
4408		if (ethtool_register_table[i].offsets < 0) {
4409			hval = cas_phy_read(cp,
4410				    -ethtool_register_table[i].offsets);
4411			val = hval;
4412		} else {
4413			val= readl(cp->regs+ethtool_register_table[i].offsets);
4414		}
4415		memcpy(p, (u8 *)&val, sizeof(u32));
4416	}
4417	spin_unlock_irqrestore(&cp->lock, flags);
4418}
4419
4420static struct net_device_stats *cas_get_stats(struct net_device *dev)
4421{
4422	struct cas *cp = netdev_priv(dev);
4423	struct net_device_stats *stats = cp->net_stats;
4424	unsigned long flags;
4425	int i;
4426	unsigned long tmp;
4427
4428	/* we collate all of the stats into net_stats[N_TX_RING] */
4429	if (!cp->hw_running)
4430		return stats + N_TX_RINGS;
4431
4432	/* collect outstanding stats */
4433	/* WTZ: the Cassini spec gives these as 16 bit counters but
4434	 * stored in 32-bit words.  Added a mask of 0xffff to be safe,
4435	 * in case the chip somehow puts any garbage in the other bits.
4436	 * Also, counter usage didn't seem to mach what Adrian did
4437	 * in the parts of the code that set these quantities. Made
4438	 * that consistent.
4439	 */
4440	spin_lock_irqsave(&cp->stat_lock[N_TX_RINGS], flags);
4441	stats[N_TX_RINGS].rx_crc_errors +=
4442	  readl(cp->regs + REG_MAC_FCS_ERR) & 0xffff;
4443	stats[N_TX_RINGS].rx_frame_errors +=
4444		readl(cp->regs + REG_MAC_ALIGN_ERR) &0xffff;
4445	stats[N_TX_RINGS].rx_length_errors +=
4446		readl(cp->regs + REG_MAC_LEN_ERR) & 0xffff;
4447#if 1
4448	tmp = (readl(cp->regs + REG_MAC_COLL_EXCESS) & 0xffff) +
4449		(readl(cp->regs + REG_MAC_COLL_LATE) & 0xffff);
4450	stats[N_TX_RINGS].tx_aborted_errors += tmp;
4451	stats[N_TX_RINGS].collisions +=
4452	  tmp + (readl(cp->regs + REG_MAC_COLL_NORMAL) & 0xffff);
4453#else
4454	stats[N_TX_RINGS].tx_aborted_errors +=
4455		readl(cp->regs + REG_MAC_COLL_EXCESS);
4456	stats[N_TX_RINGS].collisions += readl(cp->regs + REG_MAC_COLL_EXCESS) +
4457		readl(cp->regs + REG_MAC_COLL_LATE);
4458#endif
4459	cas_clear_mac_err(cp);
4460
4461	/* saved bits that are unique to ring 0 */
4462	spin_lock(&cp->stat_lock[0]);
4463	stats[N_TX_RINGS].collisions        += stats[0].collisions;
4464	stats[N_TX_RINGS].rx_over_errors    += stats[0].rx_over_errors;
4465	stats[N_TX_RINGS].rx_frame_errors   += stats[0].rx_frame_errors;
4466	stats[N_TX_RINGS].rx_fifo_errors    += stats[0].rx_fifo_errors;
4467	stats[N_TX_RINGS].tx_aborted_errors += stats[0].tx_aborted_errors;
4468	stats[N_TX_RINGS].tx_fifo_errors    += stats[0].tx_fifo_errors;
4469	spin_unlock(&cp->stat_lock[0]);
4470
4471	for (i = 0; i < N_TX_RINGS; i++) {
4472		spin_lock(&cp->stat_lock[i]);
4473		stats[N_TX_RINGS].rx_length_errors +=
4474			stats[i].rx_length_errors;
4475		stats[N_TX_RINGS].rx_crc_errors += stats[i].rx_crc_errors;
4476		stats[N_TX_RINGS].rx_packets    += stats[i].rx_packets;
4477		stats[N_TX_RINGS].tx_packets    += stats[i].tx_packets;
4478		stats[N_TX_RINGS].rx_bytes      += stats[i].rx_bytes;
4479		stats[N_TX_RINGS].tx_bytes      += stats[i].tx_bytes;
4480		stats[N_TX_RINGS].rx_errors     += stats[i].rx_errors;
4481		stats[N_TX_RINGS].tx_errors     += stats[i].tx_errors;
4482		stats[N_TX_RINGS].rx_dropped    += stats[i].rx_dropped;
4483		stats[N_TX_RINGS].tx_dropped    += stats[i].tx_dropped;
4484		memset(stats + i, 0, sizeof(struct net_device_stats));
4485		spin_unlock(&cp->stat_lock[i]);
4486	}
4487	spin_unlock_irqrestore(&cp->stat_lock[N_TX_RINGS], flags);
4488	return stats + N_TX_RINGS;
4489}
4490
4491
4492static void cas_set_multicast(struct net_device *dev)
4493{
4494	struct cas *cp = netdev_priv(dev);
4495	u32 rxcfg, rxcfg_new;
4496	unsigned long flags;
4497	int limit = STOP_TRIES;
4498
4499	if (!cp->hw_running)
4500		return;
4501
4502	spin_lock_irqsave(&cp->lock, flags);
4503	rxcfg = readl(cp->regs + REG_MAC_RX_CFG);
4504
4505	/* disable RX MAC and wait for completion */
4506	writel(rxcfg & ~MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
4507	while (readl(cp->regs + REG_MAC_RX_CFG) & MAC_RX_CFG_EN) {
4508		if (!limit--)
4509			break;
4510		udelay(10);
4511	}
4512
4513	/* disable hash filter and wait for completion */
4514	limit = STOP_TRIES;
4515	rxcfg &= ~(MAC_RX_CFG_PROMISC_EN | MAC_RX_CFG_HASH_FILTER_EN);
4516	writel(rxcfg & ~MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
4517	while (readl(cp->regs + REG_MAC_RX_CFG) & MAC_RX_CFG_HASH_FILTER_EN) {
4518		if (!limit--)
4519			break;
4520		udelay(10);
4521	}
4522
4523	/* program hash filters */
4524	cp->mac_rx_cfg = rxcfg_new = cas_setup_multicast(cp);
4525	rxcfg |= rxcfg_new;
4526	writel(rxcfg, cp->regs + REG_MAC_RX_CFG);
4527	spin_unlock_irqrestore(&cp->lock, flags);
4528}
4529
4530static void cas_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
4531{
4532	struct cas *cp = netdev_priv(dev);
4533	strlcpy(info->driver, DRV_MODULE_NAME, sizeof(info->driver));
4534	strlcpy(info->version, DRV_MODULE_VERSION, sizeof(info->version));
4535	strlcpy(info->bus_info, pci_name(cp->pdev), sizeof(info->bus_info));
4536	info->regdump_len = cp->casreg_len < CAS_MAX_REGS ?
4537		cp->casreg_len : CAS_MAX_REGS;
4538	info->n_stats = CAS_NUM_STAT_KEYS;
4539}
4540
4541static int cas_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
4542{
4543	struct cas *cp = netdev_priv(dev);
4544	u16 bmcr;
4545	int full_duplex, speed, pause;
4546	unsigned long flags;
4547	enum link_state linkstate = link_up;
4548
4549	cmd->advertising = 0;
4550	cmd->supported = SUPPORTED_Autoneg;
4551	if (cp->cas_flags & CAS_FLAG_1000MB_CAP) {
4552		cmd->supported |= SUPPORTED_1000baseT_Full;
4553		cmd->advertising |= ADVERTISED_1000baseT_Full;
4554	}
4555
4556	/* Record PHY settings if HW is on. */
4557	spin_lock_irqsave(&cp->lock, flags);
4558	bmcr = 0;
4559	linkstate = cp->lstate;
4560	if (CAS_PHY_MII(cp->phy_type)) {
4561		cmd->port = PORT_MII;
4562		cmd->transceiver = (cp->cas_flags & CAS_FLAG_SATURN) ?
4563			XCVR_INTERNAL : XCVR_EXTERNAL;
4564		cmd->phy_address = cp->phy_addr;
4565		cmd->advertising |= ADVERTISED_TP | ADVERTISED_MII |
4566			ADVERTISED_10baseT_Half |
4567			ADVERTISED_10baseT_Full |
4568			ADVERTISED_100baseT_Half |
4569			ADVERTISED_100baseT_Full;
4570
4571		cmd->supported |=
4572			(SUPPORTED_10baseT_Half |
4573			 SUPPORTED_10baseT_Full |
4574			 SUPPORTED_100baseT_Half |
4575			 SUPPORTED_100baseT_Full |
4576			 SUPPORTED_TP | SUPPORTED_MII);
4577
4578		if (cp->hw_running) {
4579			cas_mif_poll(cp, 0);
4580			bmcr = cas_phy_read(cp, MII_BMCR);
4581			cas_read_mii_link_mode(cp, &full_duplex,
4582					       &speed, &pause);
4583			cas_mif_poll(cp, 1);
4584		}
4585
4586	} else {
4587		cmd->port = PORT_FIBRE;
4588		cmd->transceiver = XCVR_INTERNAL;
4589		cmd->phy_address = 0;
4590		cmd->supported   |= SUPPORTED_FIBRE;
4591		cmd->advertising |= ADVERTISED_FIBRE;
4592
4593		if (cp->hw_running) {
4594			/* pcs uses the same bits as mii */
4595			bmcr = readl(cp->regs + REG_PCS_MII_CTRL);
4596			cas_read_pcs_link_mode(cp, &full_duplex,
4597					       &speed, &pause);
4598		}
4599	}
4600	spin_unlock_irqrestore(&cp->lock, flags);
4601
4602	if (bmcr & BMCR_ANENABLE) {
4603		cmd->advertising |= ADVERTISED_Autoneg;
4604		cmd->autoneg = AUTONEG_ENABLE;
4605		ethtool_cmd_speed_set(cmd, ((speed == 10) ?
4606					    SPEED_10 :
4607					    ((speed == 1000) ?
4608					     SPEED_1000 : SPEED_100)));
4609		cmd->duplex = full_duplex ? DUPLEX_FULL : DUPLEX_HALF;
4610	} else {
4611		cmd->autoneg = AUTONEG_DISABLE;
4612		ethtool_cmd_speed_set(cmd, ((bmcr & CAS_BMCR_SPEED1000) ?
4613					    SPEED_1000 :
4614					    ((bmcr & BMCR_SPEED100) ?
4615					     SPEED_100 : SPEED_10)));
4616		cmd->duplex =
4617			(bmcr & BMCR_FULLDPLX) ?
4618			DUPLEX_FULL : DUPLEX_HALF;
4619	}
4620	if (linkstate != link_up) {
4621		/* Force these to "unknown" if the link is not up and
4622		 * autonogotiation in enabled. We can set the link
4623		 * speed to 0, but not cmd->duplex,
4624		 * because its legal values are 0 and 1.  Ethtool will
4625		 * print the value reported in parentheses after the
4626		 * word "Unknown" for unrecognized values.
4627		 *
4628		 * If in forced mode, we report the speed and duplex
4629		 * settings that we configured.
4630		 */
4631		if (cp->link_cntl & BMCR_ANENABLE) {
4632			ethtool_cmd_speed_set(cmd, 0);
4633			cmd->duplex = 0xff;
4634		} else {
4635			ethtool_cmd_speed_set(cmd, SPEED_10);
4636			if (cp->link_cntl & BMCR_SPEED100) {
4637				ethtool_cmd_speed_set(cmd, SPEED_100);
4638			} else if (cp->link_cntl & CAS_BMCR_SPEED1000) {
4639				ethtool_cmd_speed_set(cmd, SPEED_1000);
4640			}
4641			cmd->duplex = (cp->link_cntl & BMCR_FULLDPLX)?
4642				DUPLEX_FULL : DUPLEX_HALF;
4643		}
4644	}
4645	return 0;
4646}
4647
4648static int cas_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
4649{
4650	struct cas *cp = netdev_priv(dev);
4651	unsigned long flags;
4652	u32 speed = ethtool_cmd_speed(cmd);
4653
4654	/* Verify the settings we care about. */
4655	if (cmd->autoneg != AUTONEG_ENABLE &&
4656	    cmd->autoneg != AUTONEG_DISABLE)
4657		return -EINVAL;
4658
4659	if (cmd->autoneg == AUTONEG_DISABLE &&
4660	    ((speed != SPEED_1000 &&
4661	      speed != SPEED_100 &&
4662	      speed != SPEED_10) ||
4663	     (cmd->duplex != DUPLEX_HALF &&
4664	      cmd->duplex != DUPLEX_FULL)))
4665		return -EINVAL;
4666
4667	/* Apply settings and restart link process. */
4668	spin_lock_irqsave(&cp->lock, flags);
4669	cas_begin_auto_negotiation(cp, cmd);
4670	spin_unlock_irqrestore(&cp->lock, flags);
4671	return 0;
4672}
4673
4674static int cas_nway_reset(struct net_device *dev)
4675{
4676	struct cas *cp = netdev_priv(dev);
4677	unsigned long flags;
4678
4679	if ((cp->link_cntl & BMCR_ANENABLE) == 0)
4680		return -EINVAL;
4681
4682	/* Restart link process. */
4683	spin_lock_irqsave(&cp->lock, flags);
4684	cas_begin_auto_negotiation(cp, NULL);
4685	spin_unlock_irqrestore(&cp->lock, flags);
4686
4687	return 0;
4688}
4689
4690static u32 cas_get_link(struct net_device *dev)
4691{
4692	struct cas *cp = netdev_priv(dev);
4693	return cp->lstate == link_up;
4694}
4695
4696static u32 cas_get_msglevel(struct net_device *dev)
4697{
4698	struct cas *cp = netdev_priv(dev);
4699	return cp->msg_enable;
4700}
4701
4702static void cas_set_msglevel(struct net_device *dev, u32 value)
4703{
4704	struct cas *cp = netdev_priv(dev);
4705	cp->msg_enable = value;
4706}
4707
4708static int cas_get_regs_len(struct net_device *dev)
4709{
4710	struct cas *cp = netdev_priv(dev);
4711	return cp->casreg_len < CAS_MAX_REGS ? cp->casreg_len: CAS_MAX_REGS;
4712}
4713
4714static void cas_get_regs(struct net_device *dev, struct ethtool_regs *regs,
4715			     void *p)
4716{
4717	struct cas *cp = netdev_priv(dev);
4718	regs->version = 0;
4719	/* cas_read_regs handles locks (cp->lock).  */
4720	cas_read_regs(cp, p, regs->len / sizeof(u32));
4721}
4722
4723static int cas_get_sset_count(struct net_device *dev, int sset)
4724{
4725	switch (sset) {
4726	case ETH_SS_STATS:
4727		return CAS_NUM_STAT_KEYS;
4728	default:
4729		return -EOPNOTSUPP;
4730	}
4731}
4732
4733static void cas_get_strings(struct net_device *dev, u32 stringset, u8 *data)
4734{
4735	 memcpy(data, &ethtool_cassini_statnames,
4736					 CAS_NUM_STAT_KEYS * ETH_GSTRING_LEN);
4737}
4738
4739static void cas_get_ethtool_stats(struct net_device *dev,
4740				      struct ethtool_stats *estats, u64 *data)
4741{
4742	struct cas *cp = netdev_priv(dev);
4743	struct net_device_stats *stats = cas_get_stats(cp->dev);
4744	int i = 0;
4745	data[i++] = stats->collisions;
4746	data[i++] = stats->rx_bytes;
4747	data[i++] = stats->rx_crc_errors;
4748	data[i++] = stats->rx_dropped;
4749	data[i++] = stats->rx_errors;
4750	data[i++] = stats->rx_fifo_errors;
4751	data[i++] = stats->rx_frame_errors;
4752	data[i++] = stats->rx_length_errors;
4753	data[i++] = stats->rx_over_errors;
4754	data[i++] = stats->rx_packets;
4755	data[i++] = stats->tx_aborted_errors;
4756	data[i++] = stats->tx_bytes;
4757	data[i++] = stats->tx_dropped;
4758	data[i++] = stats->tx_errors;
4759	data[i++] = stats->tx_fifo_errors;
4760	data[i++] = stats->tx_packets;
4761	BUG_ON(i != CAS_NUM_STAT_KEYS);
4762}
4763
4764static const struct ethtool_ops cas_ethtool_ops = {
4765	.get_drvinfo		= cas_get_drvinfo,
4766	.get_settings		= cas_get_settings,
4767	.set_settings		= cas_set_settings,
4768	.nway_reset		= cas_nway_reset,
4769	.get_link		= cas_get_link,
4770	.get_msglevel		= cas_get_msglevel,
4771	.set_msglevel		= cas_set_msglevel,
4772	.get_regs_len		= cas_get_regs_len,
4773	.get_regs		= cas_get_regs,
4774	.get_sset_count		= cas_get_sset_count,
4775	.get_strings		= cas_get_strings,
4776	.get_ethtool_stats	= cas_get_ethtool_stats,
4777};
4778
4779static int cas_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
4780{
4781	struct cas *cp = netdev_priv(dev);
4782	struct mii_ioctl_data *data = if_mii(ifr);
4783	unsigned long flags;
4784	int rc = -EOPNOTSUPP;
4785
4786	/* Hold the PM mutex while doing ioctl's or we may collide
4787	 * with open/close and power management and oops.
4788	 */
4789	mutex_lock(&cp->pm_mutex);
4790	switch (cmd) {
4791	case SIOCGMIIPHY:		/* Get address of MII PHY in use. */
4792		data->phy_id = cp->phy_addr;
4793		/* Fallthrough... */
4794
4795	case SIOCGMIIREG:		/* Read MII PHY register. */
4796		spin_lock_irqsave(&cp->lock, flags);
4797		cas_mif_poll(cp, 0);
4798		data->val_out = cas_phy_read(cp, data->reg_num & 0x1f);
4799		cas_mif_poll(cp, 1);
4800		spin_unlock_irqrestore(&cp->lock, flags);
4801		rc = 0;
4802		break;
4803
4804	case SIOCSMIIREG:		/* Write MII PHY register. */
4805		spin_lock_irqsave(&cp->lock, flags);
4806		cas_mif_poll(cp, 0);
4807		rc = cas_phy_write(cp, data->reg_num & 0x1f, data->val_in);
4808		cas_mif_poll(cp, 1);
4809		spin_unlock_irqrestore(&cp->lock, flags);
4810		break;
4811	default:
4812		break;
4813	}
4814
4815	mutex_unlock(&cp->pm_mutex);
4816	return rc;
4817}
4818
4819/* When this chip sits underneath an Intel 31154 bridge, it is the
4820 * only subordinate device and we can tweak the bridge settings to
4821 * reflect that fact.
4822 */
4823static void __devinit cas_program_bridge(struct pci_dev *cas_pdev)
4824{
4825	struct pci_dev *pdev = cas_pdev->bus->self;
4826	u32 val;
4827
4828	if (!pdev)
4829		return;
4830
4831	if (pdev->vendor != 0x8086 || pdev->device != 0x537c)
4832		return;
4833
4834	/* Clear bit 10 (Bus Parking Control) in the Secondary
4835	 * Arbiter Control/Status Register which lives at offset
4836	 * 0x41.  Using a 32-bit word read/modify/write at 0x40
4837	 * is much simpler so that's how we do this.
4838	 */
4839	pci_read_config_dword(pdev, 0x40, &val);
4840	val &= ~0x00040000;
4841	pci_write_config_dword(pdev, 0x40, val);
4842
4843	/* Max out the Multi-Transaction Timer settings since
4844	 * Cassini is the only device present.
4845	 *
4846	 * The register is 16-bit and lives at 0x50.  When the
4847	 * settings are enabled, it extends the GRANT# signal
4848	 * for a requestor after a transaction is complete.  This
4849	 * allows the next request to run without first needing
4850	 * to negotiate the GRANT# signal back.
4851	 *
4852	 * Bits 12:10 define the grant duration:
4853	 *
4854	 *	1	--	16 clocks
4855	 *	2	--	32 clocks
4856	 *	3	--	64 clocks
4857	 *	4	--	128 clocks
4858	 *	5	--	256 clocks
4859	 *
4860	 * All other values are illegal.
4861	 *
4862	 * Bits 09:00 define which REQ/GNT signal pairs get the
4863	 * GRANT# signal treatment.  We set them all.
4864	 */
4865	pci_write_config_word(pdev, 0x50, (5 << 10) | 0x3ff);
4866
4867	/* The Read Prefecth Policy register is 16-bit and sits at
4868	 * offset 0x52.  It enables a "smart" pre-fetch policy.  We
4869	 * enable it and max out all of the settings since only one
4870	 * device is sitting underneath and thus bandwidth sharing is
4871	 * not an issue.
4872	 *
4873	 * The register has several 3 bit fields, which indicates a
4874	 * multiplier applied to the base amount of prefetching the
4875	 * chip would do.  These fields are at:
4876	 *
4877	 *	15:13	---	ReRead Primary Bus
4878	 *	12:10	---	FirstRead Primary Bus
4879	 *	09:07	---	ReRead Secondary Bus
4880	 *	06:04	---	FirstRead Secondary Bus
4881	 *
4882	 * Bits 03:00 control which REQ/GNT pairs the prefetch settings
4883	 * get enabled on.  Bit 3 is a grouped enabler which controls
4884	 * all of the REQ/GNT pairs from [8:3].  Bits 2 to 0 control
4885	 * the individual REQ/GNT pairs [2:0].
4886	 */
4887	pci_write_config_word(pdev, 0x52,
4888			      (0x7 << 13) |
4889			      (0x7 << 10) |
4890			      (0x7 <<  7) |
4891			      (0x7 <<  4) |
4892			      (0xf <<  0));
4893
4894	/* Force cacheline size to 0x8 */
4895	pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE, 0x08);
4896
4897	/* Force latency timer to maximum setting so Cassini can
4898	 * sit on the bus as long as it likes.
4899	 */
4900	pci_write_config_byte(pdev, PCI_LATENCY_TIMER, 0xff);
4901}
4902
4903static const struct net_device_ops cas_netdev_ops = {
4904	.ndo_open		= cas_open,
4905	.ndo_stop		= cas_close,
4906	.ndo_start_xmit		= cas_start_xmit,
4907	.ndo_get_stats 		= cas_get_stats,
4908	.ndo_set_rx_mode	= cas_set_multicast,
4909	.ndo_do_ioctl		= cas_ioctl,
4910	.ndo_tx_timeout		= cas_tx_timeout,
4911	.ndo_change_mtu		= cas_change_mtu,
4912	.ndo_set_mac_address	= eth_mac_addr,
4913	.ndo_validate_addr	= eth_validate_addr,
4914#ifdef CONFIG_NET_POLL_CONTROLLER
4915	.ndo_poll_controller	= cas_netpoll,
4916#endif
4917};
4918
4919static int __devinit cas_init_one(struct pci_dev *pdev,
4920				  const struct pci_device_id *ent)
4921{
4922	static int cas_version_printed = 0;
4923	unsigned long casreg_len;
4924	struct net_device *dev;
4925	struct cas *cp;
4926	int i, err, pci_using_dac;
4927	u16 pci_cmd;
4928	u8 orig_cacheline_size = 0, cas_cacheline_size = 0;
4929
4930	if (cas_version_printed++ == 0)
4931		pr_info("%s", version);
4932
4933	err = pci_enable_device(pdev);
4934	if (err) {
4935		dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
4936		return err;
4937	}
4938
4939	if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
4940		dev_err(&pdev->dev, "Cannot find proper PCI device "
4941		       "base address, aborting\n");
4942		err = -ENODEV;
4943		goto err_out_disable_pdev;
4944	}
4945
4946	dev = alloc_etherdev(sizeof(*cp));
4947	if (!dev) {
4948		err = -ENOMEM;
4949		goto err_out_disable_pdev;
4950	}
4951	SET_NETDEV_DEV(dev, &pdev->dev);
4952
4953	err = pci_request_regions(pdev, dev->name);
4954	if (err) {
4955		dev_err(&pdev->dev, "Cannot obtain PCI resources, aborting\n");
4956		goto err_out_free_netdev;
4957	}
4958	pci_set_master(pdev);
4959
4960	/* we must always turn on parity response or else parity
4961	 * doesn't get generated properly. disable SERR/PERR as well.
4962	 * in addition, we want to turn MWI on.
4963	 */
4964	pci_read_config_word(pdev, PCI_COMMAND, &pci_cmd);
4965	pci_cmd &= ~PCI_COMMAND_SERR;
4966	pci_cmd |= PCI_COMMAND_PARITY;
4967	pci_write_config_word(pdev, PCI_COMMAND, pci_cmd);
4968	if (pci_try_set_mwi(pdev))
4969		pr_warning("Could not enable MWI for %s\n", pci_name(pdev));
4970
4971	cas_program_bridge(pdev);
4972
4973	/*
4974	 * On some architectures, the default cache line size set
4975	 * by pci_try_set_mwi reduces perforamnce.  We have to increase
4976	 * it for this case.  To start, we'll print some configuration
4977	 * data.
4978	 */
4979#if 1
4980	pci_read_config_byte(pdev, PCI_CACHE_LINE_SIZE,
4981			     &orig_cacheline_size);
4982	if (orig_cacheline_size < CAS_PREF_CACHELINE_SIZE) {
4983		cas_cacheline_size =
4984			(CAS_PREF_CACHELINE_SIZE < SMP_CACHE_BYTES) ?
4985			CAS_PREF_CACHELINE_SIZE : SMP_CACHE_BYTES;
4986		if (pci_write_config_byte(pdev,
4987					  PCI_CACHE_LINE_SIZE,
4988					  cas_cacheline_size)) {
4989			dev_err(&pdev->dev, "Could not set PCI cache "
4990			       "line size\n");
4991			goto err_write_cacheline;
4992		}
4993	}
4994#endif
4995
4996
4997	/* Configure DMA attributes. */
4998	if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
4999		pci_using_dac = 1;
5000		err = pci_set_consistent_dma_mask(pdev,
5001						  DMA_BIT_MASK(64));
5002		if (err < 0) {
5003			dev_err(&pdev->dev, "Unable to obtain 64-bit DMA "
5004			       "for consistent allocations\n");
5005			goto err_out_free_res;
5006		}
5007
5008	} else {
5009		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
5010		if (err) {
5011			dev_err(&pdev->dev, "No usable DMA configuration, "
5012			       "aborting\n");
5013			goto err_out_free_res;
5014		}
5015		pci_using_dac = 0;
5016	}
5017
5018	casreg_len = pci_resource_len(pdev, 0);
5019
5020	cp = netdev_priv(dev);
5021	cp->pdev = pdev;
5022#if 1
5023	/* A value of 0 indicates we never explicitly set it */
5024	cp->orig_cacheline_size = cas_cacheline_size ? orig_cacheline_size: 0;
5025#endif
5026	cp->dev = dev;
5027	cp->msg_enable = (cassini_debug < 0) ? CAS_DEF_MSG_ENABLE :
5028	  cassini_debug;
5029
5030#if defined(CONFIG_SPARC)
5031	cp->of_node = pci_device_to_OF_node(pdev);
5032#endif
5033
5034	cp->link_transition = LINK_TRANSITION_UNKNOWN;
5035	cp->link_transition_jiffies_valid = 0;
5036
5037	spin_lock_init(&cp->lock);
5038	spin_lock_init(&cp->rx_inuse_lock);
5039	spin_lock_init(&cp->rx_spare_lock);
5040	for (i = 0; i < N_TX_RINGS; i++) {
5041		spin_lock_init(&cp->stat_lock[i]);
5042		spin_lock_init(&cp->tx_lock[i]);
5043	}
5044	spin_lock_init(&cp->stat_lock[N_TX_RINGS]);
5045	mutex_init(&cp->pm_mutex);
5046
5047	init_timer(&cp->link_timer);
5048	cp->link_timer.function = cas_link_timer;
5049	cp->link_timer.data = (unsigned long) cp;
5050
5051#if 1
5052	/* Just in case the implementation of atomic operations
5053	 * change so that an explicit initialization is necessary.
5054	 */
5055	atomic_set(&cp->reset_task_pending, 0);
5056	atomic_set(&cp->reset_task_pending_all, 0);
5057	atomic_set(&cp->reset_task_pending_spare, 0);
5058	atomic_set(&cp->reset_task_pending_mtu, 0);
5059#endif
5060	INIT_WORK(&cp->reset_task, cas_reset_task);
5061
5062	/* Default link parameters */
5063	if (link_mode >= 0 && link_mode < 6)
5064		cp->link_cntl = link_modes[link_mode];
5065	else
5066		cp->link_cntl = BMCR_ANENABLE;
5067	cp->lstate = link_down;
5068	cp->link_transition = LINK_TRANSITION_LINK_DOWN;
5069	netif_carrier_off(cp->dev);
5070	cp->timer_ticks = 0;
5071
5072	/* give us access to cassini registers */
5073	cp->regs = pci_iomap(pdev, 0, casreg_len);
5074	if (!cp->regs) {
5075		dev_err(&pdev->dev, "Cannot map device registers, aborting\n");
5076		goto err_out_free_res;
5077	}
5078	cp->casreg_len = casreg_len;
5079
5080	pci_save_state(pdev);
5081	cas_check_pci_invariants(cp);
5082	cas_hard_reset(cp);
5083	cas_reset(cp, 0);
5084	if (cas_check_invariants(cp))
5085		goto err_out_iounmap;
5086	if (cp->cas_flags & CAS_FLAG_SATURN)
5087		if (cas_saturn_firmware_init(cp))
5088			goto err_out_iounmap;
5089
5090	cp->init_block = (struct cas_init_block *)
5091		pci_alloc_consistent(pdev, sizeof(struct cas_init_block),
5092				     &cp->block_dvma);
5093	if (!cp->init_block) {
5094		dev_err(&pdev->dev, "Cannot allocate init block, aborting\n");
5095		goto err_out_iounmap;
5096	}
5097
5098	for (i = 0; i < N_TX_RINGS; i++)
5099		cp->init_txds[i] = cp->init_block->txds[i];
5100
5101	for (i = 0; i < N_RX_DESC_RINGS; i++)
5102		cp->init_rxds[i] = cp->init_block->rxds[i];
5103
5104	for (i = 0; i < N_RX_COMP_RINGS; i++)
5105		cp->init_rxcs[i] = cp->init_block->rxcs[i];
5106
5107	for (i = 0; i < N_RX_FLOWS; i++)
5108		skb_queue_head_init(&cp->rx_flows[i]);
5109
5110	dev->netdev_ops = &cas_netdev_ops;
5111	dev->ethtool_ops = &cas_ethtool_ops;
5112	dev->watchdog_timeo = CAS_TX_TIMEOUT;
5113
5114#ifdef USE_NAPI
5115	netif_napi_add(dev, &cp->napi, cas_poll, 64);
5116#endif
5117	dev->irq = pdev->irq;
5118	dev->dma = 0;
5119
5120	/* Cassini features. */
5121	if ((cp->cas_flags & CAS_FLAG_NO_HW_CSUM) == 0)
5122		dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG;
5123
5124	if (pci_using_dac)
5125		dev->features |= NETIF_F_HIGHDMA;
5126
5127	if (register_netdev(dev)) {
5128		dev_err(&pdev->dev, "Cannot register net device, aborting\n");
5129		goto err_out_free_consistent;
5130	}
5131
5132	i = readl(cp->regs + REG_BIM_CFG);
5133	netdev_info(dev, "Sun Cassini%s (%sbit/%sMHz PCI/%s) Ethernet[%d] %pM\n",
5134		    (cp->cas_flags & CAS_FLAG_REG_PLUS) ? "+" : "",
5135		    (i & BIM_CFG_32BIT) ? "32" : "64",
5136		    (i & BIM_CFG_66MHZ) ? "66" : "33",
5137		    (cp->phy_type == CAS_PHY_SERDES) ? "Fi" : "Cu", pdev->irq,
5138		    dev->dev_addr);
5139
5140	pci_set_drvdata(pdev, dev);
5141	cp->hw_running = 1;
5142	cas_entropy_reset(cp);
5143	cas_phy_init(cp);
5144	cas_begin_auto_negotiation(cp, NULL);
5145	return 0;
5146
5147err_out_free_consistent:
5148	pci_free_consistent(pdev, sizeof(struct cas_init_block),
5149			    cp->init_block, cp->block_dvma);
5150
5151err_out_iounmap:
5152	mutex_lock(&cp->pm_mutex);
5153	if (cp->hw_running)
5154		cas_shutdown(cp);
5155	mutex_unlock(&cp->pm_mutex);
5156
5157	pci_iounmap(pdev, cp->regs);
5158
5159
5160err_out_free_res:
5161	pci_release_regions(pdev);
5162
5163err_write_cacheline:
5164	/* Try to restore it in case the error occurred after we
5165	 * set it.
5166	 */
5167	pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE, orig_cacheline_size);
5168
5169err_out_free_netdev:
5170	free_netdev(dev);
5171
5172err_out_disable_pdev:
5173	pci_disable_device(pdev);
5174	pci_set_drvdata(pdev, NULL);
5175	return -ENODEV;
5176}
5177
5178static void __devexit cas_remove_one(struct pci_dev *pdev)
5179{
5180	struct net_device *dev = pci_get_drvdata(pdev);
5181	struct cas *cp;
5182	if (!dev)
5183		return;
5184
5185	cp = netdev_priv(dev);
5186	unregister_netdev(dev);
5187
5188	if (cp->fw_data)
5189		vfree(cp->fw_data);
5190
5191	mutex_lock(&cp->pm_mutex);
5192	cancel_work_sync(&cp->reset_task);
5193	if (cp->hw_running)
5194		cas_shutdown(cp);
5195	mutex_unlock(&cp->pm_mutex);
5196
5197#if 1
5198	if (cp->orig_cacheline_size) {
5199		/* Restore the cache line size if we had modified
5200		 * it.
5201		 */
5202		pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE,
5203				      cp->orig_cacheline_size);
5204	}
5205#endif
5206	pci_free_consistent(pdev, sizeof(struct cas_init_block),
5207			    cp->init_block, cp->block_dvma);
5208	pci_iounmap(pdev, cp->regs);
5209	free_netdev(dev);
5210	pci_release_regions(pdev);
5211	pci_disable_device(pdev);
5212	pci_set_drvdata(pdev, NULL);
5213}
5214
5215#ifdef CONFIG_PM
5216static int cas_suspend(struct pci_dev *pdev, pm_message_t state)
5217{
5218	struct net_device *dev = pci_get_drvdata(pdev);
5219	struct cas *cp = netdev_priv(dev);
5220	unsigned long flags;
5221
5222	mutex_lock(&cp->pm_mutex);
5223
5224	/* If the driver is opened, we stop the DMA */
5225	if (cp->opened) {
5226		netif_device_detach(dev);
5227
5228		cas_lock_all_save(cp, flags);
5229
5230		/* We can set the second arg of cas_reset to 0
5231		 * because on resume, we'll call cas_init_hw with
5232		 * its second arg set so that autonegotiation is
5233		 * restarted.
5234		 */
5235		cas_reset(cp, 0);
5236		cas_clean_rings(cp);
5237		cas_unlock_all_restore(cp, flags);
5238	}
5239
5240	if (cp->hw_running)
5241		cas_shutdown(cp);
5242	mutex_unlock(&cp->pm_mutex);
5243
5244	return 0;
5245}
5246
5247static int cas_resume(struct pci_dev *pdev)
5248{
5249	struct net_device *dev = pci_get_drvdata(pdev);
5250	struct cas *cp = netdev_priv(dev);
5251
5252	netdev_info(dev, "resuming\n");
5253
5254	mutex_lock(&cp->pm_mutex);
5255	cas_hard_reset(cp);
5256	if (cp->opened) {
5257		unsigned long flags;
5258		cas_lock_all_save(cp, flags);
5259		cas_reset(cp, 0);
5260		cp->hw_running = 1;
5261		cas_clean_rings(cp);
5262		cas_init_hw(cp, 1);
5263		cas_unlock_all_restore(cp, flags);
5264
5265		netif_device_attach(dev);
5266	}
5267	mutex_unlock(&cp->pm_mutex);
5268	return 0;
5269}
5270#endif /* CONFIG_PM */
5271
5272static struct pci_driver cas_driver = {
5273	.name		= DRV_MODULE_NAME,
5274	.id_table	= cas_pci_tbl,
5275	.probe		= cas_init_one,
5276	.remove		= __devexit_p(cas_remove_one),
5277#ifdef CONFIG_PM
5278	.suspend	= cas_suspend,
5279	.resume		= cas_resume
5280#endif
5281};
5282
5283static int __init cas_init(void)
5284{
5285	if (linkdown_timeout > 0)
5286		link_transition_timeout = linkdown_timeout * HZ;
5287	else
5288		link_transition_timeout = 0;
5289
5290	return pci_register_driver(&cas_driver);
5291}
5292
5293static void __exit cas_cleanup(void)
5294{
5295	pci_unregister_driver(&cas_driver);
5296}
5297
5298module_init(cas_init);
5299module_exit(cas_cleanup);
5300