p54: fix EEPROM structure endianness
[linux-2.6] / drivers / net / cassini.c
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 targetted 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
70 #include <linux/module.h>
71 #include <linux/kernel.h>
72 #include <linux/types.h>
73 #include <linux/compiler.h>
74 #include <linux/slab.h>
75 #include <linux/delay.h>
76 #include <linux/init.h>
77 #include <linux/ioport.h>
78 #include <linux/pci.h>
79 #include <linux/mm.h>
80 #include <linux/highmem.h>
81 #include <linux/list.h>
82 #include <linux/dma-mapping.h>
83
84 #include <linux/netdevice.h>
85 #include <linux/etherdevice.h>
86 #include <linux/skbuff.h>
87 #include <linux/ethtool.h>
88 #include <linux/crc32.h>
89 #include <linux/random.h>
90 #include <linux/mii.h>
91 #include <linux/ip.h>
92 #include <linux/tcp.h>
93 #include <linux/mutex.h>
94
95 #include <net/checksum.h>
96
97 #include <asm/atomic.h>
98 #include <asm/system.h>
99 #include <asm/io.h>
100 #include <asm/byteorder.h>
101 #include <asm/uaccess.h>
102
103 #define cas_page_map(x)      kmap_atomic((x), KM_SKB_DATA_SOFTIRQ)
104 #define cas_page_unmap(x)    kunmap_atomic((x), KM_SKB_DATA_SOFTIRQ)
105 #define CAS_NCPUS            num_online_cpus()
106
107 #if defined(CONFIG_CASSINI_NAPI) && defined(HAVE_NETDEV_POLL)
108 #define USE_NAPI
109 #define cas_skb_release(x)  netif_receive_skb(x)
110 #else
111 #define cas_skb_release(x)  netif_rx(x)
112 #endif
113
114 /* select which firmware to use */
115 #define USE_HP_WORKAROUND
116 #define HP_WORKAROUND_DEFAULT /* select which firmware to use as default */
117 #define CAS_HP_ALT_FIRMWARE   cas_prog_null /* alternate firmware */
118
119 #include "cassini.h"
120
121 #define USE_TX_COMPWB      /* use completion writeback registers */
122 #define USE_CSMA_CD_PROTO  /* standard CSMA/CD */
123 #define USE_RX_BLANK       /* hw interrupt mitigation */
124 #undef USE_ENTROPY_DEV     /* don't test for entropy device */
125
126 /* NOTE: these aren't useable unless PCI interrupts can be assigned.
127  * also, we need to make cp->lock finer-grained.
128  */
129 #undef  USE_PCI_INTB
130 #undef  USE_PCI_INTC
131 #undef  USE_PCI_INTD
132 #undef  USE_QOS
133
134 #undef  USE_VPD_DEBUG       /* debug vpd information if defined */
135
136 /* rx processing options */
137 #define USE_PAGE_ORDER      /* specify to allocate large rx pages */
138 #define RX_DONT_BATCH  0    /* if 1, don't batch flows */
139 #define RX_COPY_ALWAYS 0    /* if 0, use frags */
140 #define RX_COPY_MIN    64   /* copy a little to make upper layers happy */
141 #undef  RX_COUNT_BUFFERS    /* define to calculate RX buffer stats */
142
143 #define DRV_MODULE_NAME         "cassini"
144 #define PFX DRV_MODULE_NAME     ": "
145 #define DRV_MODULE_VERSION      "1.5"
146 #define DRV_MODULE_RELDATE      "4 Jan 2008"
147
148 #define CAS_DEF_MSG_ENABLE        \
149         (NETIF_MSG_DRV          | \
150          NETIF_MSG_PROBE        | \
151          NETIF_MSG_LINK         | \
152          NETIF_MSG_TIMER        | \
153          NETIF_MSG_IFDOWN       | \
154          NETIF_MSG_IFUP         | \
155          NETIF_MSG_RX_ERR       | \
156          NETIF_MSG_TX_ERR)
157
158 /* length of time before we decide the hardware is borked,
159  * and dev->tx_timeout() should be called to fix the problem
160  */
161 #define CAS_TX_TIMEOUT                  (HZ)
162 #define CAS_LINK_TIMEOUT                (22*HZ/10)
163 #define CAS_LINK_FAST_TIMEOUT           (1)
164
165 /* timeout values for state changing. these specify the number
166  * of 10us delays to be used before giving up.
167  */
168 #define STOP_TRIES_PHY 1000
169 #define STOP_TRIES     5000
170
171 /* specify a minimum frame size to deal with some fifo issues
172  * max mtu == 2 * page size - ethernet header - 64 - swivel =
173  *            2 * page_size - 0x50
174  */
175 #define CAS_MIN_FRAME                   97
176 #define CAS_1000MB_MIN_FRAME            255
177 #define CAS_MIN_MTU                     60
178 #define CAS_MAX_MTU                     min(((cp->page_size << 1) - 0x50), 9000)
179
180 #if 1
181 /*
182  * Eliminate these and use separate atomic counters for each, to
183  * avoid a race condition.
184  */
185 #else
186 #define CAS_RESET_MTU                   1
187 #define CAS_RESET_ALL                   2
188 #define CAS_RESET_SPARE                 3
189 #endif
190
191 static char version[] __devinitdata =
192         DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
193
194 static int cassini_debug = -1;  /* -1 == use CAS_DEF_MSG_ENABLE as value */
195 static int link_mode;
196
197 MODULE_AUTHOR("Adrian Sun (asun@darksunrising.com)");
198 MODULE_DESCRIPTION("Sun Cassini(+) ethernet driver");
199 MODULE_LICENSE("GPL");
200 module_param(cassini_debug, int, 0);
201 MODULE_PARM_DESC(cassini_debug, "Cassini bitmapped debugging message enable value");
202 module_param(link_mode, int, 0);
203 MODULE_PARM_DESC(link_mode, "default link mode");
204
205 /*
206  * Work around for a PCS bug in which the link goes down due to the chip
207  * being confused and never showing a link status of "up."
208  */
209 #define DEFAULT_LINKDOWN_TIMEOUT 5
210 /*
211  * Value in seconds, for user input.
212  */
213 static int linkdown_timeout = DEFAULT_LINKDOWN_TIMEOUT;
214 module_param(linkdown_timeout, int, 0);
215 MODULE_PARM_DESC(linkdown_timeout,
216 "min reset interval in sec. for PCS linkdown issue; disabled if not positive");
217
218 /*
219  * value in 'ticks' (units used by jiffies). Set when we init the
220  * module because 'HZ' in actually a function call on some flavors of
221  * Linux.  This will default to DEFAULT_LINKDOWN_TIMEOUT * HZ.
222  */
223 static int link_transition_timeout;
224
225
226
227 static u16 link_modes[] __devinitdata = {
228         BMCR_ANENABLE,                   /* 0 : autoneg */
229         0,                               /* 1 : 10bt half duplex */
230         BMCR_SPEED100,                   /* 2 : 100bt half duplex */
231         BMCR_FULLDPLX,                   /* 3 : 10bt full duplex */
232         BMCR_SPEED100|BMCR_FULLDPLX,     /* 4 : 100bt full duplex */
233         CAS_BMCR_SPEED1000|BMCR_FULLDPLX /* 5 : 1000bt full duplex */
234 };
235
236 static struct pci_device_id cas_pci_tbl[] __devinitdata = {
237         { PCI_VENDOR_ID_SUN, PCI_DEVICE_ID_SUN_CASSINI,
238           PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0UL },
239         { PCI_VENDOR_ID_NS, PCI_DEVICE_ID_NS_SATURN,
240           PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0UL },
241         { 0, }
242 };
243
244 MODULE_DEVICE_TABLE(pci, cas_pci_tbl);
245
246 static void cas_set_link_modes(struct cas *cp);
247
248 static inline void cas_lock_tx(struct cas *cp)
249 {
250         int i;
251
252         for (i = 0; i < N_TX_RINGS; i++)
253                 spin_lock(&cp->tx_lock[i]);
254 }
255
256 static inline void cas_lock_all(struct cas *cp)
257 {
258         spin_lock_irq(&cp->lock);
259         cas_lock_tx(cp);
260 }
261
262 /* WTZ: QA was finding deadlock problems with the previous
263  * versions after long test runs with multiple cards per machine.
264  * See if replacing cas_lock_all with safer versions helps. The
265  * symptoms QA is reporting match those we'd expect if interrupts
266  * aren't being properly restored, and we fixed a previous deadlock
267  * with similar symptoms by using save/restore versions in other
268  * places.
269  */
270 #define cas_lock_all_save(cp, flags) \
271 do { \
272         struct cas *xxxcp = (cp); \
273         spin_lock_irqsave(&xxxcp->lock, flags); \
274         cas_lock_tx(xxxcp); \
275 } while (0)
276
277 static inline void cas_unlock_tx(struct cas *cp)
278 {
279         int i;
280
281         for (i = N_TX_RINGS; i > 0; i--)
282                 spin_unlock(&cp->tx_lock[i - 1]);
283 }
284
285 static inline void cas_unlock_all(struct cas *cp)
286 {
287         cas_unlock_tx(cp);
288         spin_unlock_irq(&cp->lock);
289 }
290
291 #define cas_unlock_all_restore(cp, flags) \
292 do { \
293         struct cas *xxxcp = (cp); \
294         cas_unlock_tx(xxxcp); \
295         spin_unlock_irqrestore(&xxxcp->lock, flags); \
296 } while (0)
297
298 static void cas_disable_irq(struct cas *cp, const int ring)
299 {
300         /* Make sure we won't get any more interrupts */
301         if (ring == 0) {
302                 writel(0xFFFFFFFF, cp->regs + REG_INTR_MASK);
303                 return;
304         }
305
306         /* disable completion interrupts and selectively mask */
307         if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
308                 switch (ring) {
309 #if defined (USE_PCI_INTB) || defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
310 #ifdef USE_PCI_INTB
311                 case 1:
312 #endif
313 #ifdef USE_PCI_INTC
314                 case 2:
315 #endif
316 #ifdef USE_PCI_INTD
317                 case 3:
318 #endif
319                         writel(INTRN_MASK_CLEAR_ALL | INTRN_MASK_RX_EN,
320                                cp->regs + REG_PLUS_INTRN_MASK(ring));
321                         break;
322 #endif
323                 default:
324                         writel(INTRN_MASK_CLEAR_ALL, cp->regs +
325                                REG_PLUS_INTRN_MASK(ring));
326                         break;
327                 }
328         }
329 }
330
331 static inline void cas_mask_intr(struct cas *cp)
332 {
333         int i;
334
335         for (i = 0; i < N_RX_COMP_RINGS; i++)
336                 cas_disable_irq(cp, i);
337 }
338
339 static void cas_enable_irq(struct cas *cp, const int ring)
340 {
341         if (ring == 0) { /* all but TX_DONE */
342                 writel(INTR_TX_DONE, cp->regs + REG_INTR_MASK);
343                 return;
344         }
345
346         if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
347                 switch (ring) {
348 #if defined (USE_PCI_INTB) || defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
349 #ifdef USE_PCI_INTB
350                 case 1:
351 #endif
352 #ifdef USE_PCI_INTC
353                 case 2:
354 #endif
355 #ifdef USE_PCI_INTD
356                 case 3:
357 #endif
358                         writel(INTRN_MASK_RX_EN, cp->regs +
359                                REG_PLUS_INTRN_MASK(ring));
360                         break;
361 #endif
362                 default:
363                         break;
364                 }
365         }
366 }
367
368 static inline void cas_unmask_intr(struct cas *cp)
369 {
370         int i;
371
372         for (i = 0; i < N_RX_COMP_RINGS; i++)
373                 cas_enable_irq(cp, i);
374 }
375
376 static inline void cas_entropy_gather(struct cas *cp)
377 {
378 #ifdef USE_ENTROPY_DEV
379         if ((cp->cas_flags & CAS_FLAG_ENTROPY_DEV) == 0)
380                 return;
381
382         batch_entropy_store(readl(cp->regs + REG_ENTROPY_IV),
383                             readl(cp->regs + REG_ENTROPY_IV),
384                             sizeof(uint64_t)*8);
385 #endif
386 }
387
388 static inline void cas_entropy_reset(struct cas *cp)
389 {
390 #ifdef USE_ENTROPY_DEV
391         if ((cp->cas_flags & CAS_FLAG_ENTROPY_DEV) == 0)
392                 return;
393
394         writel(BIM_LOCAL_DEV_PAD | BIM_LOCAL_DEV_PROM | BIM_LOCAL_DEV_EXT,
395                cp->regs + REG_BIM_LOCAL_DEV_EN);
396         writeb(ENTROPY_RESET_STC_MODE, cp->regs + REG_ENTROPY_RESET);
397         writeb(0x55, cp->regs + REG_ENTROPY_RAND_REG);
398
399         /* if we read back 0x0, we don't have an entropy device */
400         if (readb(cp->regs + REG_ENTROPY_RAND_REG) == 0)
401                 cp->cas_flags &= ~CAS_FLAG_ENTROPY_DEV;
402 #endif
403 }
404
405 /* access to the phy. the following assumes that we've initialized the MIF to
406  * be in frame rather than bit-bang mode
407  */
408 static u16 cas_phy_read(struct cas *cp, int reg)
409 {
410         u32 cmd;
411         int limit = STOP_TRIES_PHY;
412
413         cmd = MIF_FRAME_ST | MIF_FRAME_OP_READ;
414         cmd |= CAS_BASE(MIF_FRAME_PHY_ADDR, cp->phy_addr);
415         cmd |= CAS_BASE(MIF_FRAME_REG_ADDR, reg);
416         cmd |= MIF_FRAME_TURN_AROUND_MSB;
417         writel(cmd, cp->regs + REG_MIF_FRAME);
418
419         /* poll for completion */
420         while (limit-- > 0) {
421                 udelay(10);
422                 cmd = readl(cp->regs + REG_MIF_FRAME);
423                 if (cmd & MIF_FRAME_TURN_AROUND_LSB)
424                         return (cmd & MIF_FRAME_DATA_MASK);
425         }
426         return 0xFFFF; /* -1 */
427 }
428
429 static int cas_phy_write(struct cas *cp, int reg, u16 val)
430 {
431         int limit = STOP_TRIES_PHY;
432         u32 cmd;
433
434         cmd = MIF_FRAME_ST | MIF_FRAME_OP_WRITE;
435         cmd |= CAS_BASE(MIF_FRAME_PHY_ADDR, cp->phy_addr);
436         cmd |= CAS_BASE(MIF_FRAME_REG_ADDR, reg);
437         cmd |= MIF_FRAME_TURN_AROUND_MSB;
438         cmd |= val & MIF_FRAME_DATA_MASK;
439         writel(cmd, cp->regs + REG_MIF_FRAME);
440
441         /* poll for completion */
442         while (limit-- > 0) {
443                 udelay(10);
444                 cmd = readl(cp->regs + REG_MIF_FRAME);
445                 if (cmd & MIF_FRAME_TURN_AROUND_LSB)
446                         return 0;
447         }
448         return -1;
449 }
450
451 static void cas_phy_powerup(struct cas *cp)
452 {
453         u16 ctl = cas_phy_read(cp, MII_BMCR);
454
455         if ((ctl & BMCR_PDOWN) == 0)
456                 return;
457         ctl &= ~BMCR_PDOWN;
458         cas_phy_write(cp, MII_BMCR, ctl);
459 }
460
461 static void cas_phy_powerdown(struct cas *cp)
462 {
463         u16 ctl = cas_phy_read(cp, MII_BMCR);
464
465         if (ctl & BMCR_PDOWN)
466                 return;
467         ctl |= BMCR_PDOWN;
468         cas_phy_write(cp, MII_BMCR, ctl);
469 }
470
471 /* cp->lock held. note: the last put_page will free the buffer */
472 static int cas_page_free(struct cas *cp, cas_page_t *page)
473 {
474         pci_unmap_page(cp->pdev, page->dma_addr, cp->page_size,
475                        PCI_DMA_FROMDEVICE);
476         __free_pages(page->buffer, cp->page_order);
477         kfree(page);
478         return 0;
479 }
480
481 #ifdef RX_COUNT_BUFFERS
482 #define RX_USED_ADD(x, y)       ((x)->used += (y))
483 #define RX_USED_SET(x, y)       ((x)->used  = (y))
484 #else
485 #define RX_USED_ADD(x, y)
486 #define RX_USED_SET(x, y)
487 #endif
488
489 /* local page allocation routines for the receive buffers. jumbo pages
490  * require at least 8K contiguous and 8K aligned buffers.
491  */
492 static cas_page_t *cas_page_alloc(struct cas *cp, const gfp_t flags)
493 {
494         cas_page_t *page;
495
496         page = kmalloc(sizeof(cas_page_t), flags);
497         if (!page)
498                 return NULL;
499
500         INIT_LIST_HEAD(&page->list);
501         RX_USED_SET(page, 0);
502         page->buffer = alloc_pages(flags, cp->page_order);
503         if (!page->buffer)
504                 goto page_err;
505         page->dma_addr = pci_map_page(cp->pdev, page->buffer, 0,
506                                       cp->page_size, PCI_DMA_FROMDEVICE);
507         return page;
508
509 page_err:
510         kfree(page);
511         return NULL;
512 }
513
514 /* initialize spare pool of rx buffers, but allocate during the open */
515 static void cas_spare_init(struct cas *cp)
516 {
517         spin_lock(&cp->rx_inuse_lock);
518         INIT_LIST_HEAD(&cp->rx_inuse_list);
519         spin_unlock(&cp->rx_inuse_lock);
520
521         spin_lock(&cp->rx_spare_lock);
522         INIT_LIST_HEAD(&cp->rx_spare_list);
523         cp->rx_spares_needed = RX_SPARE_COUNT;
524         spin_unlock(&cp->rx_spare_lock);
525 }
526
527 /* used on close. free all the spare buffers. */
528 static void cas_spare_free(struct cas *cp)
529 {
530         struct list_head list, *elem, *tmp;
531
532         /* free spare buffers */
533         INIT_LIST_HEAD(&list);
534         spin_lock(&cp->rx_spare_lock);
535         list_splice(&cp->rx_spare_list, &list);
536         INIT_LIST_HEAD(&cp->rx_spare_list);
537         spin_unlock(&cp->rx_spare_lock);
538         list_for_each_safe(elem, tmp, &list) {
539                 cas_page_free(cp, list_entry(elem, cas_page_t, list));
540         }
541
542         INIT_LIST_HEAD(&list);
543 #if 1
544         /*
545          * Looks like Adrian had protected this with a different
546          * lock than used everywhere else to manipulate this list.
547          */
548         spin_lock(&cp->rx_inuse_lock);
549         list_splice(&cp->rx_inuse_list, &list);
550         INIT_LIST_HEAD(&cp->rx_inuse_list);
551         spin_unlock(&cp->rx_inuse_lock);
552 #else
553         spin_lock(&cp->rx_spare_lock);
554         list_splice(&cp->rx_inuse_list, &list);
555         INIT_LIST_HEAD(&cp->rx_inuse_list);
556         spin_unlock(&cp->rx_spare_lock);
557 #endif
558         list_for_each_safe(elem, tmp, &list) {
559                 cas_page_free(cp, list_entry(elem, cas_page_t, list));
560         }
561 }
562
563 /* replenish spares if needed */
564 static void cas_spare_recover(struct cas *cp, const gfp_t flags)
565 {
566         struct list_head list, *elem, *tmp;
567         int needed, i;
568
569         /* check inuse list. if we don't need any more free buffers,
570          * just free it
571          */
572
573         /* make a local copy of the list */
574         INIT_LIST_HEAD(&list);
575         spin_lock(&cp->rx_inuse_lock);
576         list_splice(&cp->rx_inuse_list, &list);
577         INIT_LIST_HEAD(&cp->rx_inuse_list);
578         spin_unlock(&cp->rx_inuse_lock);
579
580         list_for_each_safe(elem, tmp, &list) {
581                 cas_page_t *page = list_entry(elem, cas_page_t, list);
582
583                 if (page_count(page->buffer) > 1)
584                         continue;
585
586                 list_del(elem);
587                 spin_lock(&cp->rx_spare_lock);
588                 if (cp->rx_spares_needed > 0) {
589                         list_add(elem, &cp->rx_spare_list);
590                         cp->rx_spares_needed--;
591                         spin_unlock(&cp->rx_spare_lock);
592                 } else {
593                         spin_unlock(&cp->rx_spare_lock);
594                         cas_page_free(cp, page);
595                 }
596         }
597
598         /* put any inuse buffers back on the list */
599         if (!list_empty(&list)) {
600                 spin_lock(&cp->rx_inuse_lock);
601                 list_splice(&list, &cp->rx_inuse_list);
602                 spin_unlock(&cp->rx_inuse_lock);
603         }
604
605         spin_lock(&cp->rx_spare_lock);
606         needed = cp->rx_spares_needed;
607         spin_unlock(&cp->rx_spare_lock);
608         if (!needed)
609                 return;
610
611         /* we still need spares, so try to allocate some */
612         INIT_LIST_HEAD(&list);
613         i = 0;
614         while (i < needed) {
615                 cas_page_t *spare = cas_page_alloc(cp, flags);
616                 if (!spare)
617                         break;
618                 list_add(&spare->list, &list);
619                 i++;
620         }
621
622         spin_lock(&cp->rx_spare_lock);
623         list_splice(&list, &cp->rx_spare_list);
624         cp->rx_spares_needed -= i;
625         spin_unlock(&cp->rx_spare_lock);
626 }
627
628 /* pull a page from the list. */
629 static cas_page_t *cas_page_dequeue(struct cas *cp)
630 {
631         struct list_head *entry;
632         int recover;
633
634         spin_lock(&cp->rx_spare_lock);
635         if (list_empty(&cp->rx_spare_list)) {
636                 /* try to do a quick recovery */
637                 spin_unlock(&cp->rx_spare_lock);
638                 cas_spare_recover(cp, GFP_ATOMIC);
639                 spin_lock(&cp->rx_spare_lock);
640                 if (list_empty(&cp->rx_spare_list)) {
641                         if (netif_msg_rx_err(cp))
642                                 printk(KERN_ERR "%s: no spare buffers "
643                                        "available.\n", cp->dev->name);
644                         spin_unlock(&cp->rx_spare_lock);
645                         return NULL;
646                 }
647         }
648
649         entry = cp->rx_spare_list.next;
650         list_del(entry);
651         recover = ++cp->rx_spares_needed;
652         spin_unlock(&cp->rx_spare_lock);
653
654         /* trigger the timer to do the recovery */
655         if ((recover & (RX_SPARE_RECOVER_VAL - 1)) == 0) {
656 #if 1
657                 atomic_inc(&cp->reset_task_pending);
658                 atomic_inc(&cp->reset_task_pending_spare);
659                 schedule_work(&cp->reset_task);
660 #else
661                 atomic_set(&cp->reset_task_pending, CAS_RESET_SPARE);
662                 schedule_work(&cp->reset_task);
663 #endif
664         }
665         return list_entry(entry, cas_page_t, list);
666 }
667
668
669 static void cas_mif_poll(struct cas *cp, const int enable)
670 {
671         u32 cfg;
672
673         cfg  = readl(cp->regs + REG_MIF_CFG);
674         cfg &= (MIF_CFG_MDIO_0 | MIF_CFG_MDIO_1);
675
676         if (cp->phy_type & CAS_PHY_MII_MDIO1)
677                 cfg |= MIF_CFG_PHY_SELECT;
678
679         /* poll and interrupt on link status change. */
680         if (enable) {
681                 cfg |= MIF_CFG_POLL_EN;
682                 cfg |= CAS_BASE(MIF_CFG_POLL_REG, MII_BMSR);
683                 cfg |= CAS_BASE(MIF_CFG_POLL_PHY, cp->phy_addr);
684         }
685         writel((enable) ? ~(BMSR_LSTATUS | BMSR_ANEGCOMPLETE) : 0xFFFF,
686                cp->regs + REG_MIF_MASK);
687         writel(cfg, cp->regs + REG_MIF_CFG);
688 }
689
690 /* Must be invoked under cp->lock */
691 static void cas_begin_auto_negotiation(struct cas *cp, struct ethtool_cmd *ep)
692 {
693         u16 ctl;
694 #if 1
695         int lcntl;
696         int changed = 0;
697         int oldstate = cp->lstate;
698         int link_was_not_down = !(oldstate == link_down);
699 #endif
700         /* Setup link parameters */
701         if (!ep)
702                 goto start_aneg;
703         lcntl = cp->link_cntl;
704         if (ep->autoneg == AUTONEG_ENABLE)
705                 cp->link_cntl = BMCR_ANENABLE;
706         else {
707                 cp->link_cntl = 0;
708                 if (ep->speed == SPEED_100)
709                         cp->link_cntl |= BMCR_SPEED100;
710                 else if (ep->speed == SPEED_1000)
711                         cp->link_cntl |= CAS_BMCR_SPEED1000;
712                 if (ep->duplex == DUPLEX_FULL)
713                         cp->link_cntl |= BMCR_FULLDPLX;
714         }
715 #if 1
716         changed = (lcntl != cp->link_cntl);
717 #endif
718 start_aneg:
719         if (cp->lstate == link_up) {
720                 printk(KERN_INFO "%s: PCS link down.\n",
721                        cp->dev->name);
722         } else {
723                 if (changed) {
724                         printk(KERN_INFO "%s: link configuration changed\n",
725                                cp->dev->name);
726                 }
727         }
728         cp->lstate = link_down;
729         cp->link_transition = LINK_TRANSITION_LINK_DOWN;
730         if (!cp->hw_running)
731                 return;
732 #if 1
733         /*
734          * WTZ: If the old state was link_up, we turn off the carrier
735          * to replicate everything we do elsewhere on a link-down
736          * event when we were already in a link-up state..
737          */
738         if (oldstate == link_up)
739                 netif_carrier_off(cp->dev);
740         if (changed  && link_was_not_down) {
741                 /*
742                  * WTZ: This branch will simply schedule a full reset after
743                  * we explicitly changed link modes in an ioctl. See if this
744                  * fixes the link-problems we were having for forced mode.
745                  */
746                 atomic_inc(&cp->reset_task_pending);
747                 atomic_inc(&cp->reset_task_pending_all);
748                 schedule_work(&cp->reset_task);
749                 cp->timer_ticks = 0;
750                 mod_timer(&cp->link_timer, jiffies + CAS_LINK_TIMEOUT);
751                 return;
752         }
753 #endif
754         if (cp->phy_type & CAS_PHY_SERDES) {
755                 u32 val = readl(cp->regs + REG_PCS_MII_CTRL);
756
757                 if (cp->link_cntl & BMCR_ANENABLE) {
758                         val |= (PCS_MII_RESTART_AUTONEG | PCS_MII_AUTONEG_EN);
759                         cp->lstate = link_aneg;
760                 } else {
761                         if (cp->link_cntl & BMCR_FULLDPLX)
762                                 val |= PCS_MII_CTRL_DUPLEX;
763                         val &= ~PCS_MII_AUTONEG_EN;
764                         cp->lstate = link_force_ok;
765                 }
766                 cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
767                 writel(val, cp->regs + REG_PCS_MII_CTRL);
768
769         } else {
770                 cas_mif_poll(cp, 0);
771                 ctl = cas_phy_read(cp, MII_BMCR);
772                 ctl &= ~(BMCR_FULLDPLX | BMCR_SPEED100 |
773                          CAS_BMCR_SPEED1000 | BMCR_ANENABLE);
774                 ctl |= cp->link_cntl;
775                 if (ctl & BMCR_ANENABLE) {
776                         ctl |= BMCR_ANRESTART;
777                         cp->lstate = link_aneg;
778                 } else {
779                         cp->lstate = link_force_ok;
780                 }
781                 cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
782                 cas_phy_write(cp, MII_BMCR, ctl);
783                 cas_mif_poll(cp, 1);
784         }
785
786         cp->timer_ticks = 0;
787         mod_timer(&cp->link_timer, jiffies + CAS_LINK_TIMEOUT);
788 }
789
790 /* Must be invoked under cp->lock. */
791 static int cas_reset_mii_phy(struct cas *cp)
792 {
793         int limit = STOP_TRIES_PHY;
794         u16 val;
795
796         cas_phy_write(cp, MII_BMCR, BMCR_RESET);
797         udelay(100);
798         while (limit--) {
799                 val = cas_phy_read(cp, MII_BMCR);
800                 if ((val & BMCR_RESET) == 0)
801                         break;
802                 udelay(10);
803         }
804         return (limit <= 0);
805 }
806
807 static void cas_saturn_firmware_load(struct cas *cp)
808 {
809         cas_saturn_patch_t *patch = cas_saturn_patch;
810
811         cas_phy_powerdown(cp);
812
813         /* expanded memory access mode */
814         cas_phy_write(cp, DP83065_MII_MEM, 0x0);
815
816         /* pointer configuration for new firmware */
817         cas_phy_write(cp, DP83065_MII_REGE, 0x8ff9);
818         cas_phy_write(cp, DP83065_MII_REGD, 0xbd);
819         cas_phy_write(cp, DP83065_MII_REGE, 0x8ffa);
820         cas_phy_write(cp, DP83065_MII_REGD, 0x82);
821         cas_phy_write(cp, DP83065_MII_REGE, 0x8ffb);
822         cas_phy_write(cp, DP83065_MII_REGD, 0x0);
823         cas_phy_write(cp, DP83065_MII_REGE, 0x8ffc);
824         cas_phy_write(cp, DP83065_MII_REGD, 0x39);
825
826         /* download new firmware */
827         cas_phy_write(cp, DP83065_MII_MEM, 0x1);
828         cas_phy_write(cp, DP83065_MII_REGE, patch->addr);
829         while (patch->addr) {
830                 cas_phy_write(cp, DP83065_MII_REGD, patch->val);
831                 patch++;
832         }
833
834         /* enable firmware */
835         cas_phy_write(cp, DP83065_MII_REGE, 0x8ff8);
836         cas_phy_write(cp, DP83065_MII_REGD, 0x1);
837 }
838
839
840 /* phy initialization */
841 static void cas_phy_init(struct cas *cp)
842 {
843         u16 val;
844
845         /* if we're in MII/GMII mode, set up phy */
846         if (CAS_PHY_MII(cp->phy_type)) {
847                 writel(PCS_DATAPATH_MODE_MII,
848                        cp->regs + REG_PCS_DATAPATH_MODE);
849
850                 cas_mif_poll(cp, 0);
851                 cas_reset_mii_phy(cp); /* take out of isolate mode */
852
853                 if (PHY_LUCENT_B0 == cp->phy_id) {
854                         /* workaround link up/down issue with lucent */
855                         cas_phy_write(cp, LUCENT_MII_REG, 0x8000);
856                         cas_phy_write(cp, MII_BMCR, 0x00f1);
857                         cas_phy_write(cp, LUCENT_MII_REG, 0x0);
858
859                 } else if (PHY_BROADCOM_B0 == (cp->phy_id & 0xFFFFFFFC)) {
860                         /* workarounds for broadcom phy */
861                         cas_phy_write(cp, BROADCOM_MII_REG8, 0x0C20);
862                         cas_phy_write(cp, BROADCOM_MII_REG7, 0x0012);
863                         cas_phy_write(cp, BROADCOM_MII_REG5, 0x1804);
864                         cas_phy_write(cp, BROADCOM_MII_REG7, 0x0013);
865                         cas_phy_write(cp, BROADCOM_MII_REG5, 0x1204);
866                         cas_phy_write(cp, BROADCOM_MII_REG7, 0x8006);
867                         cas_phy_write(cp, BROADCOM_MII_REG5, 0x0132);
868                         cas_phy_write(cp, BROADCOM_MII_REG7, 0x8006);
869                         cas_phy_write(cp, BROADCOM_MII_REG5, 0x0232);
870                         cas_phy_write(cp, BROADCOM_MII_REG7, 0x201F);
871                         cas_phy_write(cp, BROADCOM_MII_REG5, 0x0A20);
872
873                 } else if (PHY_BROADCOM_5411 == cp->phy_id) {
874                         val = cas_phy_read(cp, BROADCOM_MII_REG4);
875                         val = cas_phy_read(cp, BROADCOM_MII_REG4);
876                         if (val & 0x0080) {
877                                 /* link workaround */
878                                 cas_phy_write(cp, BROADCOM_MII_REG4,
879                                               val & ~0x0080);
880                         }
881
882                 } else if (cp->cas_flags & CAS_FLAG_SATURN) {
883                         writel((cp->phy_type & CAS_PHY_MII_MDIO0) ?
884                                SATURN_PCFG_FSI : 0x0,
885                                cp->regs + REG_SATURN_PCFG);
886
887                         /* load firmware to address 10Mbps auto-negotiation
888                          * issue. NOTE: this will need to be changed if the
889                          * default firmware gets fixed.
890                          */
891                         if (PHY_NS_DP83065 == cp->phy_id) {
892                                 cas_saturn_firmware_load(cp);
893                         }
894                         cas_phy_powerup(cp);
895                 }
896
897                 /* advertise capabilities */
898                 val = cas_phy_read(cp, MII_BMCR);
899                 val &= ~BMCR_ANENABLE;
900                 cas_phy_write(cp, MII_BMCR, val);
901                 udelay(10);
902
903                 cas_phy_write(cp, MII_ADVERTISE,
904                               cas_phy_read(cp, MII_ADVERTISE) |
905                               (ADVERTISE_10HALF | ADVERTISE_10FULL |
906                                ADVERTISE_100HALF | ADVERTISE_100FULL |
907                                CAS_ADVERTISE_PAUSE |
908                                CAS_ADVERTISE_ASYM_PAUSE));
909
910                 if (cp->cas_flags & CAS_FLAG_1000MB_CAP) {
911                         /* make sure that we don't advertise half
912                          * duplex to avoid a chip issue
913                          */
914                         val  = cas_phy_read(cp, CAS_MII_1000_CTRL);
915                         val &= ~CAS_ADVERTISE_1000HALF;
916                         val |= CAS_ADVERTISE_1000FULL;
917                         cas_phy_write(cp, CAS_MII_1000_CTRL, val);
918                 }
919
920         } else {
921                 /* reset pcs for serdes */
922                 u32 val;
923                 int limit;
924
925                 writel(PCS_DATAPATH_MODE_SERDES,
926                        cp->regs + REG_PCS_DATAPATH_MODE);
927
928                 /* enable serdes pins on saturn */
929                 if (cp->cas_flags & CAS_FLAG_SATURN)
930                         writel(0, cp->regs + REG_SATURN_PCFG);
931
932                 /* Reset PCS unit. */
933                 val = readl(cp->regs + REG_PCS_MII_CTRL);
934                 val |= PCS_MII_RESET;
935                 writel(val, cp->regs + REG_PCS_MII_CTRL);
936
937                 limit = STOP_TRIES;
938                 while (limit-- > 0) {
939                         udelay(10);
940                         if ((readl(cp->regs + REG_PCS_MII_CTRL) &
941                              PCS_MII_RESET) == 0)
942                                 break;
943                 }
944                 if (limit <= 0)
945                         printk(KERN_WARNING "%s: PCS reset bit would not "
946                                "clear [%08x].\n", cp->dev->name,
947                                readl(cp->regs + REG_PCS_STATE_MACHINE));
948
949                 /* Make sure PCS is disabled while changing advertisement
950                  * configuration.
951                  */
952                 writel(0x0, cp->regs + REG_PCS_CFG);
953
954                 /* Advertise all capabilities except half-duplex. */
955                 val  = readl(cp->regs + REG_PCS_MII_ADVERT);
956                 val &= ~PCS_MII_ADVERT_HD;
957                 val |= (PCS_MII_ADVERT_FD | PCS_MII_ADVERT_SYM_PAUSE |
958                         PCS_MII_ADVERT_ASYM_PAUSE);
959                 writel(val, cp->regs + REG_PCS_MII_ADVERT);
960
961                 /* enable PCS */
962                 writel(PCS_CFG_EN, cp->regs + REG_PCS_CFG);
963
964                 /* pcs workaround: enable sync detect */
965                 writel(PCS_SERDES_CTRL_SYNCD_EN,
966                        cp->regs + REG_PCS_SERDES_CTRL);
967         }
968 }
969
970
971 static int cas_pcs_link_check(struct cas *cp)
972 {
973         u32 stat, state_machine;
974         int retval = 0;
975
976         /* The link status bit latches on zero, so you must
977          * read it twice in such a case to see a transition
978          * to the link being up.
979          */
980         stat = readl(cp->regs + REG_PCS_MII_STATUS);
981         if ((stat & PCS_MII_STATUS_LINK_STATUS) == 0)
982                 stat = readl(cp->regs + REG_PCS_MII_STATUS);
983
984         /* The remote-fault indication is only valid
985          * when autoneg has completed.
986          */
987         if ((stat & (PCS_MII_STATUS_AUTONEG_COMP |
988                      PCS_MII_STATUS_REMOTE_FAULT)) ==
989             (PCS_MII_STATUS_AUTONEG_COMP | PCS_MII_STATUS_REMOTE_FAULT)) {
990                 if (netif_msg_link(cp))
991                         printk(KERN_INFO "%s: PCS RemoteFault\n",
992                                cp->dev->name);
993         }
994
995         /* work around link detection issue by querying the PCS state
996          * machine directly.
997          */
998         state_machine = readl(cp->regs + REG_PCS_STATE_MACHINE);
999         if ((state_machine & PCS_SM_LINK_STATE_MASK) != SM_LINK_STATE_UP) {
1000                 stat &= ~PCS_MII_STATUS_LINK_STATUS;
1001         } else if (state_machine & PCS_SM_WORD_SYNC_STATE_MASK) {
1002                 stat |= PCS_MII_STATUS_LINK_STATUS;
1003         }
1004
1005         if (stat & PCS_MII_STATUS_LINK_STATUS) {
1006                 if (cp->lstate != link_up) {
1007                         if (cp->opened) {
1008                                 cp->lstate = link_up;
1009                                 cp->link_transition = LINK_TRANSITION_LINK_UP;
1010
1011                                 cas_set_link_modes(cp);
1012                                 netif_carrier_on(cp->dev);
1013                         }
1014                 }
1015         } else if (cp->lstate == link_up) {
1016                 cp->lstate = link_down;
1017                 if (link_transition_timeout != 0 &&
1018                     cp->link_transition != LINK_TRANSITION_REQUESTED_RESET &&
1019                     !cp->link_transition_jiffies_valid) {
1020                         /*
1021                          * force a reset, as a workaround for the
1022                          * link-failure problem. May want to move this to a
1023                          * point a bit earlier in the sequence. If we had
1024                          * generated a reset a short time ago, we'll wait for
1025                          * the link timer to check the status until a
1026                          * timer expires (link_transistion_jiffies_valid is
1027                          * true when the timer is running.)  Instead of using
1028                          * a system timer, we just do a check whenever the
1029                          * link timer is running - this clears the flag after
1030                          * a suitable delay.
1031                          */
1032                         retval = 1;
1033                         cp->link_transition = LINK_TRANSITION_REQUESTED_RESET;
1034                         cp->link_transition_jiffies = jiffies;
1035                         cp->link_transition_jiffies_valid = 1;
1036                 } else {
1037                         cp->link_transition = LINK_TRANSITION_ON_FAILURE;
1038                 }
1039                 netif_carrier_off(cp->dev);
1040                 if (cp->opened && netif_msg_link(cp)) {
1041                         printk(KERN_INFO "%s: PCS link down.\n",
1042                                cp->dev->name);
1043                 }
1044
1045                 /* Cassini only: if you force a mode, there can be
1046                  * sync problems on link down. to fix that, the following
1047                  * things need to be checked:
1048                  * 1) read serialink state register
1049                  * 2) read pcs status register to verify link down.
1050                  * 3) if link down and serial link == 0x03, then you need
1051                  *    to global reset the chip.
1052                  */
1053                 if ((cp->cas_flags & CAS_FLAG_REG_PLUS) == 0) {
1054                         /* should check to see if we're in a forced mode */
1055                         stat = readl(cp->regs + REG_PCS_SERDES_STATE);
1056                         if (stat == 0x03)
1057                                 return 1;
1058                 }
1059         } else if (cp->lstate == link_down) {
1060                 if (link_transition_timeout != 0 &&
1061                     cp->link_transition != LINK_TRANSITION_REQUESTED_RESET &&
1062                     !cp->link_transition_jiffies_valid) {
1063                         /* force a reset, as a workaround for the
1064                          * link-failure problem.  May want to move
1065                          * this to a point a bit earlier in the
1066                          * sequence.
1067                          */
1068                         retval = 1;
1069                         cp->link_transition = LINK_TRANSITION_REQUESTED_RESET;
1070                         cp->link_transition_jiffies = jiffies;
1071                         cp->link_transition_jiffies_valid = 1;
1072                 } else {
1073                         cp->link_transition = LINK_TRANSITION_STILL_FAILED;
1074                 }
1075         }
1076
1077         return retval;
1078 }
1079
1080 static int cas_pcs_interrupt(struct net_device *dev,
1081                              struct cas *cp, u32 status)
1082 {
1083         u32 stat = readl(cp->regs + REG_PCS_INTR_STATUS);
1084
1085         if ((stat & PCS_INTR_STATUS_LINK_CHANGE) == 0)
1086                 return 0;
1087         return cas_pcs_link_check(cp);
1088 }
1089
1090 static int cas_txmac_interrupt(struct net_device *dev,
1091                                struct cas *cp, u32 status)
1092 {
1093         u32 txmac_stat = readl(cp->regs + REG_MAC_TX_STATUS);
1094
1095         if (!txmac_stat)
1096                 return 0;
1097
1098         if (netif_msg_intr(cp))
1099                 printk(KERN_DEBUG "%s: txmac interrupt, txmac_stat: 0x%x\n",
1100                         cp->dev->name, txmac_stat);
1101
1102         /* Defer timer expiration is quite normal,
1103          * don't even log the event.
1104          */
1105         if ((txmac_stat & MAC_TX_DEFER_TIMER) &&
1106             !(txmac_stat & ~MAC_TX_DEFER_TIMER))
1107                 return 0;
1108
1109         spin_lock(&cp->stat_lock[0]);
1110         if (txmac_stat & MAC_TX_UNDERRUN) {
1111                 printk(KERN_ERR "%s: TX MAC xmit underrun.\n",
1112                        dev->name);
1113                 cp->net_stats[0].tx_fifo_errors++;
1114         }
1115
1116         if (txmac_stat & MAC_TX_MAX_PACKET_ERR) {
1117                 printk(KERN_ERR "%s: TX MAC max packet size error.\n",
1118                        dev->name);
1119                 cp->net_stats[0].tx_errors++;
1120         }
1121
1122         /* The rest are all cases of one of the 16-bit TX
1123          * counters expiring.
1124          */
1125         if (txmac_stat & MAC_TX_COLL_NORMAL)
1126                 cp->net_stats[0].collisions += 0x10000;
1127
1128         if (txmac_stat & MAC_TX_COLL_EXCESS) {
1129                 cp->net_stats[0].tx_aborted_errors += 0x10000;
1130                 cp->net_stats[0].collisions += 0x10000;
1131         }
1132
1133         if (txmac_stat & MAC_TX_COLL_LATE) {
1134                 cp->net_stats[0].tx_aborted_errors += 0x10000;
1135                 cp->net_stats[0].collisions += 0x10000;
1136         }
1137         spin_unlock(&cp->stat_lock[0]);
1138
1139         /* We do not keep track of MAC_TX_COLL_FIRST and
1140          * MAC_TX_PEAK_ATTEMPTS events.
1141          */
1142         return 0;
1143 }
1144
1145 static void cas_load_firmware(struct cas *cp, cas_hp_inst_t *firmware)
1146 {
1147         cas_hp_inst_t *inst;
1148         u32 val;
1149         int i;
1150
1151         i = 0;
1152         while ((inst = firmware) && inst->note) {
1153                 writel(i, cp->regs + REG_HP_INSTR_RAM_ADDR);
1154
1155                 val = CAS_BASE(HP_INSTR_RAM_HI_VAL, inst->val);
1156                 val |= CAS_BASE(HP_INSTR_RAM_HI_MASK, inst->mask);
1157                 writel(val, cp->regs + REG_HP_INSTR_RAM_DATA_HI);
1158
1159                 val = CAS_BASE(HP_INSTR_RAM_MID_OUTARG, inst->outarg >> 10);
1160                 val |= CAS_BASE(HP_INSTR_RAM_MID_OUTOP, inst->outop);
1161                 val |= CAS_BASE(HP_INSTR_RAM_MID_FNEXT, inst->fnext);
1162                 val |= CAS_BASE(HP_INSTR_RAM_MID_FOFF, inst->foff);
1163                 val |= CAS_BASE(HP_INSTR_RAM_MID_SNEXT, inst->snext);
1164                 val |= CAS_BASE(HP_INSTR_RAM_MID_SOFF, inst->soff);
1165                 val |= CAS_BASE(HP_INSTR_RAM_MID_OP, inst->op);
1166                 writel(val, cp->regs + REG_HP_INSTR_RAM_DATA_MID);
1167
1168                 val = CAS_BASE(HP_INSTR_RAM_LOW_OUTMASK, inst->outmask);
1169                 val |= CAS_BASE(HP_INSTR_RAM_LOW_OUTSHIFT, inst->outshift);
1170                 val |= CAS_BASE(HP_INSTR_RAM_LOW_OUTEN, inst->outenab);
1171                 val |= CAS_BASE(HP_INSTR_RAM_LOW_OUTARG, inst->outarg);
1172                 writel(val, cp->regs + REG_HP_INSTR_RAM_DATA_LOW);
1173                 ++firmware;
1174                 ++i;
1175         }
1176 }
1177
1178 static void cas_init_rx_dma(struct cas *cp)
1179 {
1180         u64 desc_dma = cp->block_dvma;
1181         u32 val;
1182         int i, size;
1183
1184         /* rx free descriptors */
1185         val = CAS_BASE(RX_CFG_SWIVEL, RX_SWIVEL_OFF_VAL);
1186         val |= CAS_BASE(RX_CFG_DESC_RING, RX_DESC_RINGN_INDEX(0));
1187         val |= CAS_BASE(RX_CFG_COMP_RING, RX_COMP_RINGN_INDEX(0));
1188         if ((N_RX_DESC_RINGS > 1) &&
1189             (cp->cas_flags & CAS_FLAG_REG_PLUS))  /* do desc 2 */
1190                 val |= CAS_BASE(RX_CFG_DESC_RING1, RX_DESC_RINGN_INDEX(1));
1191         writel(val, cp->regs + REG_RX_CFG);
1192
1193         val = (unsigned long) cp->init_rxds[0] -
1194                 (unsigned long) cp->init_block;
1195         writel((desc_dma + val) >> 32, cp->regs + REG_RX_DB_HI);
1196         writel((desc_dma + val) & 0xffffffff, cp->regs + REG_RX_DB_LOW);
1197         writel(RX_DESC_RINGN_SIZE(0) - 4, cp->regs + REG_RX_KICK);
1198
1199         if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
1200                 /* rx desc 2 is for IPSEC packets. however,
1201                  * we don't it that for that purpose.
1202                  */
1203                 val = (unsigned long) cp->init_rxds[1] -
1204                         (unsigned long) cp->init_block;
1205                 writel((desc_dma + val) >> 32, cp->regs + REG_PLUS_RX_DB1_HI);
1206                 writel((desc_dma + val) & 0xffffffff, cp->regs +
1207                        REG_PLUS_RX_DB1_LOW);
1208                 writel(RX_DESC_RINGN_SIZE(1) - 4, cp->regs +
1209                        REG_PLUS_RX_KICK1);
1210         }
1211
1212         /* rx completion registers */
1213         val = (unsigned long) cp->init_rxcs[0] -
1214                 (unsigned long) cp->init_block;
1215         writel((desc_dma + val) >> 32, cp->regs + REG_RX_CB_HI);
1216         writel((desc_dma + val) & 0xffffffff, cp->regs + REG_RX_CB_LOW);
1217
1218         if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
1219                 /* rx comp 2-4 */
1220                 for (i = 1; i < MAX_RX_COMP_RINGS; i++) {
1221                         val = (unsigned long) cp->init_rxcs[i] -
1222                                 (unsigned long) cp->init_block;
1223                         writel((desc_dma + val) >> 32, cp->regs +
1224                                REG_PLUS_RX_CBN_HI(i));
1225                         writel((desc_dma + val) & 0xffffffff, cp->regs +
1226                                REG_PLUS_RX_CBN_LOW(i));
1227                 }
1228         }
1229
1230         /* read selective clear regs to prevent spurious interrupts
1231          * on reset because complete == kick.
1232          * selective clear set up to prevent interrupts on resets
1233          */
1234         readl(cp->regs + REG_INTR_STATUS_ALIAS);
1235         writel(INTR_RX_DONE | INTR_RX_BUF_UNAVAIL, cp->regs + REG_ALIAS_CLEAR);
1236         if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
1237                 for (i = 1; i < N_RX_COMP_RINGS; i++)
1238                         readl(cp->regs + REG_PLUS_INTRN_STATUS_ALIAS(i));
1239
1240                 /* 2 is different from 3 and 4 */
1241                 if (N_RX_COMP_RINGS > 1)
1242                         writel(INTR_RX_DONE_ALT | INTR_RX_BUF_UNAVAIL_1,
1243                                cp->regs + REG_PLUS_ALIASN_CLEAR(1));
1244
1245                 for (i = 2; i < N_RX_COMP_RINGS; i++)
1246                         writel(INTR_RX_DONE_ALT,
1247                                cp->regs + REG_PLUS_ALIASN_CLEAR(i));
1248         }
1249
1250         /* set up pause thresholds */
1251         val  = CAS_BASE(RX_PAUSE_THRESH_OFF,
1252                         cp->rx_pause_off / RX_PAUSE_THRESH_QUANTUM);
1253         val |= CAS_BASE(RX_PAUSE_THRESH_ON,
1254                         cp->rx_pause_on / RX_PAUSE_THRESH_QUANTUM);
1255         writel(val, cp->regs + REG_RX_PAUSE_THRESH);
1256
1257         /* zero out dma reassembly buffers */
1258         for (i = 0; i < 64; i++) {
1259                 writel(i, cp->regs + REG_RX_TABLE_ADDR);
1260                 writel(0x0, cp->regs + REG_RX_TABLE_DATA_LOW);
1261                 writel(0x0, cp->regs + REG_RX_TABLE_DATA_MID);
1262                 writel(0x0, cp->regs + REG_RX_TABLE_DATA_HI);
1263         }
1264
1265         /* make sure address register is 0 for normal operation */
1266         writel(0x0, cp->regs + REG_RX_CTRL_FIFO_ADDR);
1267         writel(0x0, cp->regs + REG_RX_IPP_FIFO_ADDR);
1268
1269         /* interrupt mitigation */
1270 #ifdef USE_RX_BLANK
1271         val = CAS_BASE(RX_BLANK_INTR_TIME, RX_BLANK_INTR_TIME_VAL);
1272         val |= CAS_BASE(RX_BLANK_INTR_PKT, RX_BLANK_INTR_PKT_VAL);
1273         writel(val, cp->regs + REG_RX_BLANK);
1274 #else
1275         writel(0x0, cp->regs + REG_RX_BLANK);
1276 #endif
1277
1278         /* interrupt generation as a function of low water marks for
1279          * free desc and completion entries. these are used to trigger
1280          * housekeeping for rx descs. we don't use the free interrupt
1281          * as it's not very useful
1282          */
1283         /* val = CAS_BASE(RX_AE_THRESH_FREE, RX_AE_FREEN_VAL(0)); */
1284         val = CAS_BASE(RX_AE_THRESH_COMP, RX_AE_COMP_VAL);
1285         writel(val, cp->regs + REG_RX_AE_THRESH);
1286         if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
1287                 val = CAS_BASE(RX_AE1_THRESH_FREE, RX_AE_FREEN_VAL(1));
1288                 writel(val, cp->regs + REG_PLUS_RX_AE1_THRESH);
1289         }
1290
1291         /* Random early detect registers. useful for congestion avoidance.
1292          * this should be tunable.
1293          */
1294         writel(0x0, cp->regs + REG_RX_RED);
1295
1296         /* receive page sizes. default == 2K (0x800) */
1297         val = 0;
1298         if (cp->page_size == 0x1000)
1299                 val = 0x1;
1300         else if (cp->page_size == 0x2000)
1301                 val = 0x2;
1302         else if (cp->page_size == 0x4000)
1303                 val = 0x3;
1304
1305         /* round mtu + offset. constrain to page size. */
1306         size = cp->dev->mtu + 64;
1307         if (size > cp->page_size)
1308                 size = cp->page_size;
1309
1310         if (size <= 0x400)
1311                 i = 0x0;
1312         else if (size <= 0x800)
1313                 i = 0x1;
1314         else if (size <= 0x1000)
1315                 i = 0x2;
1316         else
1317                 i = 0x3;
1318
1319         cp->mtu_stride = 1 << (i + 10);
1320         val  = CAS_BASE(RX_PAGE_SIZE, val);
1321         val |= CAS_BASE(RX_PAGE_SIZE_MTU_STRIDE, i);
1322         val |= CAS_BASE(RX_PAGE_SIZE_MTU_COUNT, cp->page_size >> (i + 10));
1323         val |= CAS_BASE(RX_PAGE_SIZE_MTU_OFF, 0x1);
1324         writel(val, cp->regs + REG_RX_PAGE_SIZE);
1325
1326         /* enable the header parser if desired */
1327         if (CAS_HP_FIRMWARE == cas_prog_null)
1328                 return;
1329
1330         val = CAS_BASE(HP_CFG_NUM_CPU, CAS_NCPUS > 63 ? 0 : CAS_NCPUS);
1331         val |= HP_CFG_PARSE_EN | HP_CFG_SYN_INC_MASK;
1332         val |= CAS_BASE(HP_CFG_TCP_THRESH, HP_TCP_THRESH_VAL);
1333         writel(val, cp->regs + REG_HP_CFG);
1334 }
1335
1336 static inline void cas_rxc_init(struct cas_rx_comp *rxc)
1337 {
1338         memset(rxc, 0, sizeof(*rxc));
1339         rxc->word4 = cpu_to_le64(RX_COMP4_ZERO);
1340 }
1341
1342 /* NOTE: we use the ENC RX DESC ring for spares. the rx_page[0,1]
1343  * flipping is protected by the fact that the chip will not
1344  * hand back the same page index while it's being processed.
1345  */
1346 static inline cas_page_t *cas_page_spare(struct cas *cp, const int index)
1347 {
1348         cas_page_t *page = cp->rx_pages[1][index];
1349         cas_page_t *new;
1350
1351         if (page_count(page->buffer) == 1)
1352                 return page;
1353
1354         new = cas_page_dequeue(cp);
1355         if (new) {
1356                 spin_lock(&cp->rx_inuse_lock);
1357                 list_add(&page->list, &cp->rx_inuse_list);
1358                 spin_unlock(&cp->rx_inuse_lock);
1359         }
1360         return new;
1361 }
1362
1363 /* this needs to be changed if we actually use the ENC RX DESC ring */
1364 static cas_page_t *cas_page_swap(struct cas *cp, const int ring,
1365                                  const int index)
1366 {
1367         cas_page_t **page0 = cp->rx_pages[0];
1368         cas_page_t **page1 = cp->rx_pages[1];
1369
1370         /* swap if buffer is in use */
1371         if (page_count(page0[index]->buffer) > 1) {
1372                 cas_page_t *new = cas_page_spare(cp, index);
1373                 if (new) {
1374                         page1[index] = page0[index];
1375                         page0[index] = new;
1376                 }
1377         }
1378         RX_USED_SET(page0[index], 0);
1379         return page0[index];
1380 }
1381
1382 static void cas_clean_rxds(struct cas *cp)
1383 {
1384         /* only clean ring 0 as ring 1 is used for spare buffers */
1385         struct cas_rx_desc *rxd = cp->init_rxds[0];
1386         int i, size;
1387
1388         /* release all rx flows */
1389         for (i = 0; i < N_RX_FLOWS; i++) {
1390                 struct sk_buff *skb;
1391                 while ((skb = __skb_dequeue(&cp->rx_flows[i]))) {
1392                         cas_skb_release(skb);
1393                 }
1394         }
1395
1396         /* initialize descriptors */
1397         size = RX_DESC_RINGN_SIZE(0);
1398         for (i = 0; i < size; i++) {
1399                 cas_page_t *page = cas_page_swap(cp, 0, i);
1400                 rxd[i].buffer = cpu_to_le64(page->dma_addr);
1401                 rxd[i].index  = cpu_to_le64(CAS_BASE(RX_INDEX_NUM, i) |
1402                                             CAS_BASE(RX_INDEX_RING, 0));
1403         }
1404
1405         cp->rx_old[0]  = RX_DESC_RINGN_SIZE(0) - 4;
1406         cp->rx_last[0] = 0;
1407         cp->cas_flags &= ~CAS_FLAG_RXD_POST(0);
1408 }
1409
1410 static void cas_clean_rxcs(struct cas *cp)
1411 {
1412         int i, j;
1413
1414         /* take ownership of rx comp descriptors */
1415         memset(cp->rx_cur, 0, sizeof(*cp->rx_cur)*N_RX_COMP_RINGS);
1416         memset(cp->rx_new, 0, sizeof(*cp->rx_new)*N_RX_COMP_RINGS);
1417         for (i = 0; i < N_RX_COMP_RINGS; i++) {
1418                 struct cas_rx_comp *rxc = cp->init_rxcs[i];
1419                 for (j = 0; j < RX_COMP_RINGN_SIZE(i); j++) {
1420                         cas_rxc_init(rxc + j);
1421                 }
1422         }
1423 }
1424
1425 #if 0
1426 /* When we get a RX fifo overflow, the RX unit is probably hung
1427  * so we do the following.
1428  *
1429  * If any part of the reset goes wrong, we return 1 and that causes the
1430  * whole chip to be reset.
1431  */
1432 static int cas_rxmac_reset(struct cas *cp)
1433 {
1434         struct net_device *dev = cp->dev;
1435         int limit;
1436         u32 val;
1437
1438         /* First, reset MAC RX. */
1439         writel(cp->mac_rx_cfg & ~MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
1440         for (limit = 0; limit < STOP_TRIES; limit++) {
1441                 if (!(readl(cp->regs + REG_MAC_RX_CFG) & MAC_RX_CFG_EN))
1442                         break;
1443                 udelay(10);
1444         }
1445         if (limit == STOP_TRIES) {
1446                 printk(KERN_ERR "%s: RX MAC will not disable, resetting whole "
1447                        "chip.\n", dev->name);
1448                 return 1;
1449         }
1450
1451         /* Second, disable RX DMA. */
1452         writel(0, cp->regs + REG_RX_CFG);
1453         for (limit = 0; limit < STOP_TRIES; limit++) {
1454                 if (!(readl(cp->regs + REG_RX_CFG) & RX_CFG_DMA_EN))
1455                         break;
1456                 udelay(10);
1457         }
1458         if (limit == STOP_TRIES) {
1459                 printk(KERN_ERR "%s: RX DMA will not disable, resetting whole "
1460                        "chip.\n", dev->name);
1461                 return 1;
1462         }
1463
1464         mdelay(5);
1465
1466         /* Execute RX reset command. */
1467         writel(SW_RESET_RX, cp->regs + REG_SW_RESET);
1468         for (limit = 0; limit < STOP_TRIES; limit++) {
1469                 if (!(readl(cp->regs + REG_SW_RESET) & SW_RESET_RX))
1470                         break;
1471                 udelay(10);
1472         }
1473         if (limit == STOP_TRIES) {
1474                 printk(KERN_ERR "%s: RX reset command will not execute, "
1475                        "resetting whole chip.\n", dev->name);
1476                 return 1;
1477         }
1478
1479         /* reset driver rx state */
1480         cas_clean_rxds(cp);
1481         cas_clean_rxcs(cp);
1482
1483         /* Now, reprogram the rest of RX unit. */
1484         cas_init_rx_dma(cp);
1485
1486         /* re-enable */
1487         val = readl(cp->regs + REG_RX_CFG);
1488         writel(val | RX_CFG_DMA_EN, cp->regs + REG_RX_CFG);
1489         writel(MAC_RX_FRAME_RECV, cp->regs + REG_MAC_RX_MASK);
1490         val = readl(cp->regs + REG_MAC_RX_CFG);
1491         writel(val | MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
1492         return 0;
1493 }
1494 #endif
1495
1496 static int cas_rxmac_interrupt(struct net_device *dev, struct cas *cp,
1497                                u32 status)
1498 {
1499         u32 stat = readl(cp->regs + REG_MAC_RX_STATUS);
1500
1501         if (!stat)
1502                 return 0;
1503
1504         if (netif_msg_intr(cp))
1505                 printk(KERN_DEBUG "%s: rxmac interrupt, stat: 0x%x\n",
1506                         cp->dev->name, stat);
1507
1508         /* these are all rollovers */
1509         spin_lock(&cp->stat_lock[0]);
1510         if (stat & MAC_RX_ALIGN_ERR)
1511                 cp->net_stats[0].rx_frame_errors += 0x10000;
1512
1513         if (stat & MAC_RX_CRC_ERR)
1514                 cp->net_stats[0].rx_crc_errors += 0x10000;
1515
1516         if (stat & MAC_RX_LEN_ERR)
1517                 cp->net_stats[0].rx_length_errors += 0x10000;
1518
1519         if (stat & MAC_RX_OVERFLOW) {
1520                 cp->net_stats[0].rx_over_errors++;
1521                 cp->net_stats[0].rx_fifo_errors++;
1522         }
1523
1524         /* We do not track MAC_RX_FRAME_COUNT and MAC_RX_VIOL_ERR
1525          * events.
1526          */
1527         spin_unlock(&cp->stat_lock[0]);
1528         return 0;
1529 }
1530
1531 static int cas_mac_interrupt(struct net_device *dev, struct cas *cp,
1532                              u32 status)
1533 {
1534         u32 stat = readl(cp->regs + REG_MAC_CTRL_STATUS);
1535
1536         if (!stat)
1537                 return 0;
1538
1539         if (netif_msg_intr(cp))
1540                 printk(KERN_DEBUG "%s: mac interrupt, stat: 0x%x\n",
1541                         cp->dev->name, stat);
1542
1543         /* This interrupt is just for pause frame and pause
1544          * tracking.  It is useful for diagnostics and debug
1545          * but probably by default we will mask these events.
1546          */
1547         if (stat & MAC_CTRL_PAUSE_STATE)
1548                 cp->pause_entered++;
1549
1550         if (stat & MAC_CTRL_PAUSE_RECEIVED)
1551                 cp->pause_last_time_recvd = (stat >> 16);
1552
1553         return 0;
1554 }
1555
1556
1557 /* Must be invoked under cp->lock. */
1558 static inline int cas_mdio_link_not_up(struct cas *cp)
1559 {
1560         u16 val;
1561
1562         switch (cp->lstate) {
1563         case link_force_ret:
1564                 if (netif_msg_link(cp))
1565                         printk(KERN_INFO "%s: Autoneg failed again, keeping"
1566                                 " forced mode\n", cp->dev->name);
1567                 cas_phy_write(cp, MII_BMCR, cp->link_fcntl);
1568                 cp->timer_ticks = 5;
1569                 cp->lstate = link_force_ok;
1570                 cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
1571                 break;
1572
1573         case link_aneg:
1574                 val = cas_phy_read(cp, MII_BMCR);
1575
1576                 /* Try forced modes. we try things in the following order:
1577                  * 1000 full -> 100 full/half -> 10 half
1578                  */
1579                 val &= ~(BMCR_ANRESTART | BMCR_ANENABLE);
1580                 val |= BMCR_FULLDPLX;
1581                 val |= (cp->cas_flags & CAS_FLAG_1000MB_CAP) ?
1582                         CAS_BMCR_SPEED1000 : BMCR_SPEED100;
1583                 cas_phy_write(cp, MII_BMCR, val);
1584                 cp->timer_ticks = 5;
1585                 cp->lstate = link_force_try;
1586                 cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
1587                 break;
1588
1589         case link_force_try:
1590                 /* Downgrade from 1000 to 100 to 10 Mbps if necessary. */
1591                 val = cas_phy_read(cp, MII_BMCR);
1592                 cp->timer_ticks = 5;
1593                 if (val & CAS_BMCR_SPEED1000) { /* gigabit */
1594                         val &= ~CAS_BMCR_SPEED1000;
1595                         val |= (BMCR_SPEED100 | BMCR_FULLDPLX);
1596                         cas_phy_write(cp, MII_BMCR, val);
1597                         break;
1598                 }
1599
1600                 if (val & BMCR_SPEED100) {
1601                         if (val & BMCR_FULLDPLX) /* fd failed */
1602                                 val &= ~BMCR_FULLDPLX;
1603                         else { /* 100Mbps failed */
1604                                 val &= ~BMCR_SPEED100;
1605                         }
1606                         cas_phy_write(cp, MII_BMCR, val);
1607                         break;
1608                 }
1609         default:
1610                 break;
1611         }
1612         return 0;
1613 }
1614
1615
1616 /* must be invoked with cp->lock held */
1617 static int cas_mii_link_check(struct cas *cp, const u16 bmsr)
1618 {
1619         int restart;
1620
1621         if (bmsr & BMSR_LSTATUS) {
1622                 /* Ok, here we got a link. If we had it due to a forced
1623                  * fallback, and we were configured for autoneg, we
1624                  * retry a short autoneg pass. If you know your hub is
1625                  * broken, use ethtool ;)
1626                  */
1627                 if ((cp->lstate == link_force_try) &&
1628                     (cp->link_cntl & BMCR_ANENABLE)) {
1629                         cp->lstate = link_force_ret;
1630                         cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
1631                         cas_mif_poll(cp, 0);
1632                         cp->link_fcntl = cas_phy_read(cp, MII_BMCR);
1633                         cp->timer_ticks = 5;
1634                         if (cp->opened && netif_msg_link(cp))
1635                                 printk(KERN_INFO "%s: Got link after fallback, retrying"
1636                                        " autoneg once...\n", cp->dev->name);
1637                         cas_phy_write(cp, MII_BMCR,
1638                                       cp->link_fcntl | BMCR_ANENABLE |
1639                                       BMCR_ANRESTART);
1640                         cas_mif_poll(cp, 1);
1641
1642                 } else if (cp->lstate != link_up) {
1643                         cp->lstate = link_up;
1644                         cp->link_transition = LINK_TRANSITION_LINK_UP;
1645
1646                         if (cp->opened) {
1647                                 cas_set_link_modes(cp);
1648                                 netif_carrier_on(cp->dev);
1649                         }
1650                 }
1651                 return 0;
1652         }
1653
1654         /* link not up. if the link was previously up, we restart the
1655          * whole process
1656          */
1657         restart = 0;
1658         if (cp->lstate == link_up) {
1659                 cp->lstate = link_down;
1660                 cp->link_transition = LINK_TRANSITION_LINK_DOWN;
1661
1662                 netif_carrier_off(cp->dev);
1663                 if (cp->opened && netif_msg_link(cp))
1664                         printk(KERN_INFO "%s: Link down\n",
1665                                cp->dev->name);
1666                 restart = 1;
1667
1668         } else if (++cp->timer_ticks > 10)
1669                 cas_mdio_link_not_up(cp);
1670
1671         return restart;
1672 }
1673
1674 static int cas_mif_interrupt(struct net_device *dev, struct cas *cp,
1675                              u32 status)
1676 {
1677         u32 stat = readl(cp->regs + REG_MIF_STATUS);
1678         u16 bmsr;
1679
1680         /* check for a link change */
1681         if (CAS_VAL(MIF_STATUS_POLL_STATUS, stat) == 0)
1682                 return 0;
1683
1684         bmsr = CAS_VAL(MIF_STATUS_POLL_DATA, stat);
1685         return cas_mii_link_check(cp, bmsr);
1686 }
1687
1688 static int cas_pci_interrupt(struct net_device *dev, struct cas *cp,
1689                              u32 status)
1690 {
1691         u32 stat = readl(cp->regs + REG_PCI_ERR_STATUS);
1692
1693         if (!stat)
1694                 return 0;
1695
1696         printk(KERN_ERR "%s: PCI error [%04x:%04x] ", dev->name, stat,
1697                readl(cp->regs + REG_BIM_DIAG));
1698
1699         /* cassini+ has this reserved */
1700         if ((stat & PCI_ERR_BADACK) &&
1701             ((cp->cas_flags & CAS_FLAG_REG_PLUS) == 0))
1702                 printk("<No ACK64# during ABS64 cycle> ");
1703
1704         if (stat & PCI_ERR_DTRTO)
1705                 printk("<Delayed transaction timeout> ");
1706         if (stat & PCI_ERR_OTHER)
1707                 printk("<other> ");
1708         if (stat & PCI_ERR_BIM_DMA_WRITE)
1709                 printk("<BIM DMA 0 write req> ");
1710         if (stat & PCI_ERR_BIM_DMA_READ)
1711                 printk("<BIM DMA 0 read req> ");
1712         printk("\n");
1713
1714         if (stat & PCI_ERR_OTHER) {
1715                 u16 cfg;
1716
1717                 /* Interrogate PCI config space for the
1718                  * true cause.
1719                  */
1720                 pci_read_config_word(cp->pdev, PCI_STATUS, &cfg);
1721                 printk(KERN_ERR "%s: Read PCI cfg space status [%04x]\n",
1722                        dev->name, cfg);
1723                 if (cfg & PCI_STATUS_PARITY)
1724                         printk(KERN_ERR "%s: PCI parity error detected.\n",
1725                                dev->name);
1726                 if (cfg & PCI_STATUS_SIG_TARGET_ABORT)
1727                         printk(KERN_ERR "%s: PCI target abort.\n",
1728                                dev->name);
1729                 if (cfg & PCI_STATUS_REC_TARGET_ABORT)
1730                         printk(KERN_ERR "%s: PCI master acks target abort.\n",
1731                                dev->name);
1732                 if (cfg & PCI_STATUS_REC_MASTER_ABORT)
1733                         printk(KERN_ERR "%s: PCI master abort.\n", dev->name);
1734                 if (cfg & PCI_STATUS_SIG_SYSTEM_ERROR)
1735                         printk(KERN_ERR "%s: PCI system error SERR#.\n",
1736                                dev->name);
1737                 if (cfg & PCI_STATUS_DETECTED_PARITY)
1738                         printk(KERN_ERR "%s: PCI parity error.\n",
1739                                dev->name);
1740
1741                 /* Write the error bits back to clear them. */
1742                 cfg &= (PCI_STATUS_PARITY |
1743                         PCI_STATUS_SIG_TARGET_ABORT |
1744                         PCI_STATUS_REC_TARGET_ABORT |
1745                         PCI_STATUS_REC_MASTER_ABORT |
1746                         PCI_STATUS_SIG_SYSTEM_ERROR |
1747                         PCI_STATUS_DETECTED_PARITY);
1748                 pci_write_config_word(cp->pdev, PCI_STATUS, cfg);
1749         }
1750
1751         /* For all PCI errors, we should reset the chip. */
1752         return 1;
1753 }
1754
1755 /* All non-normal interrupt conditions get serviced here.
1756  * Returns non-zero if we should just exit the interrupt
1757  * handler right now (ie. if we reset the card which invalidates
1758  * all of the other original irq status bits).
1759  */
1760 static int cas_abnormal_irq(struct net_device *dev, struct cas *cp,
1761                             u32 status)
1762 {
1763         if (status & INTR_RX_TAG_ERROR) {
1764                 /* corrupt RX tag framing */
1765                 if (netif_msg_rx_err(cp))
1766                         printk(KERN_DEBUG "%s: corrupt rx tag framing\n",
1767                                 cp->dev->name);
1768                 spin_lock(&cp->stat_lock[0]);
1769                 cp->net_stats[0].rx_errors++;
1770                 spin_unlock(&cp->stat_lock[0]);
1771                 goto do_reset;
1772         }
1773
1774         if (status & INTR_RX_LEN_MISMATCH) {
1775                 /* length mismatch. */
1776                 if (netif_msg_rx_err(cp))
1777                         printk(KERN_DEBUG "%s: length mismatch for rx frame\n",
1778                                 cp->dev->name);
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_PCS_STATUS) {
1786                 if (cas_pcs_interrupt(dev, cp, status))
1787                         goto do_reset;
1788         }
1789
1790         if (status & INTR_TX_MAC_STATUS) {
1791                 if (cas_txmac_interrupt(dev, cp, status))
1792                         goto do_reset;
1793         }
1794
1795         if (status & INTR_RX_MAC_STATUS) {
1796                 if (cas_rxmac_interrupt(dev, cp, status))
1797                         goto do_reset;
1798         }
1799
1800         if (status & INTR_MAC_CTRL_STATUS) {
1801                 if (cas_mac_interrupt(dev, cp, status))
1802                         goto do_reset;
1803         }
1804
1805         if (status & INTR_MIF_STATUS) {
1806                 if (cas_mif_interrupt(dev, cp, status))
1807                         goto do_reset;
1808         }
1809
1810         if (status & INTR_PCI_ERROR_STATUS) {
1811                 if (cas_pci_interrupt(dev, cp, status))
1812                         goto do_reset;
1813         }
1814         return 0;
1815
1816 do_reset:
1817 #if 1
1818         atomic_inc(&cp->reset_task_pending);
1819         atomic_inc(&cp->reset_task_pending_all);
1820         printk(KERN_ERR "%s:reset called in cas_abnormal_irq [0x%x]\n",
1821                dev->name, status);
1822         schedule_work(&cp->reset_task);
1823 #else
1824         atomic_set(&cp->reset_task_pending, CAS_RESET_ALL);
1825         printk(KERN_ERR "reset called in cas_abnormal_irq\n");
1826         schedule_work(&cp->reset_task);
1827 #endif
1828         return 1;
1829 }
1830
1831 /* NOTE: CAS_TABORT returns 1 or 2 so that it can be used when
1832  *       determining whether to do a netif_stop/wakeup
1833  */
1834 #define CAS_TABORT(x)      (((x)->cas_flags & CAS_FLAG_TARGET_ABORT) ? 2 : 1)
1835 #define CAS_ROUND_PAGE(x)  (((x) + PAGE_SIZE - 1) & PAGE_MASK)
1836 static inline int cas_calc_tabort(struct cas *cp, const unsigned long addr,
1837                                   const int len)
1838 {
1839         unsigned long off = addr + len;
1840
1841         if (CAS_TABORT(cp) == 1)
1842                 return 0;
1843         if ((CAS_ROUND_PAGE(off) - off) > TX_TARGET_ABORT_LEN)
1844                 return 0;
1845         return TX_TARGET_ABORT_LEN;
1846 }
1847
1848 static inline void cas_tx_ringN(struct cas *cp, int ring, int limit)
1849 {
1850         struct cas_tx_desc *txds;
1851         struct sk_buff **skbs;
1852         struct net_device *dev = cp->dev;
1853         int entry, count;
1854
1855         spin_lock(&cp->tx_lock[ring]);
1856         txds = cp->init_txds[ring];
1857         skbs = cp->tx_skbs[ring];
1858         entry = cp->tx_old[ring];
1859
1860         count = TX_BUFF_COUNT(ring, entry, limit);
1861         while (entry != limit) {
1862                 struct sk_buff *skb = skbs[entry];
1863                 dma_addr_t daddr;
1864                 u32 dlen;
1865                 int frag;
1866
1867                 if (!skb) {
1868                         /* this should never occur */
1869                         entry = TX_DESC_NEXT(ring, entry);
1870                         continue;
1871                 }
1872
1873                 /* however, we might get only a partial skb release. */
1874                 count -= skb_shinfo(skb)->nr_frags +
1875                         + cp->tx_tiny_use[ring][entry].nbufs + 1;
1876                 if (count < 0)
1877                         break;
1878
1879                 if (netif_msg_tx_done(cp))
1880                         printk(KERN_DEBUG "%s: tx[%d] done, slot %d\n",
1881                                cp->dev->name, ring, entry);
1882
1883                 skbs[entry] = NULL;
1884                 cp->tx_tiny_use[ring][entry].nbufs = 0;
1885
1886                 for (frag = 0; frag <= skb_shinfo(skb)->nr_frags; frag++) {
1887                         struct cas_tx_desc *txd = txds + entry;
1888
1889                         daddr = le64_to_cpu(txd->buffer);
1890                         dlen = CAS_VAL(TX_DESC_BUFLEN,
1891                                        le64_to_cpu(txd->control));
1892                         pci_unmap_page(cp->pdev, daddr, dlen,
1893                                        PCI_DMA_TODEVICE);
1894                         entry = TX_DESC_NEXT(ring, entry);
1895
1896                         /* tiny buffer may follow */
1897                         if (cp->tx_tiny_use[ring][entry].used) {
1898                                 cp->tx_tiny_use[ring][entry].used = 0;
1899                                 entry = TX_DESC_NEXT(ring, entry);
1900                         }
1901                 }
1902
1903                 spin_lock(&cp->stat_lock[ring]);
1904                 cp->net_stats[ring].tx_packets++;
1905                 cp->net_stats[ring].tx_bytes += skb->len;
1906                 spin_unlock(&cp->stat_lock[ring]);
1907                 dev_kfree_skb_irq(skb);
1908         }
1909         cp->tx_old[ring] = entry;
1910
1911         /* this is wrong for multiple tx rings. the net device needs
1912          * multiple queues for this to do the right thing.  we wait
1913          * for 2*packets to be available when using tiny buffers
1914          */
1915         if (netif_queue_stopped(dev) &&
1916             (TX_BUFFS_AVAIL(cp, ring) > CAS_TABORT(cp)*(MAX_SKB_FRAGS + 1)))
1917                 netif_wake_queue(dev);
1918         spin_unlock(&cp->tx_lock[ring]);
1919 }
1920
1921 static void cas_tx(struct net_device *dev, struct cas *cp,
1922                    u32 status)
1923 {
1924         int limit, ring;
1925 #ifdef USE_TX_COMPWB
1926         u64 compwb = le64_to_cpu(cp->init_block->tx_compwb);
1927 #endif
1928         if (netif_msg_intr(cp))
1929                 printk(KERN_DEBUG "%s: tx interrupt, status: 0x%x, %llx\n",
1930                         cp->dev->name, status, (unsigned long long)compwb);
1931         /* process all the rings */
1932         for (ring = 0; ring < N_TX_RINGS; ring++) {
1933 #ifdef USE_TX_COMPWB
1934                 /* use the completion writeback registers */
1935                 limit = (CAS_VAL(TX_COMPWB_MSB, compwb) << 8) |
1936                         CAS_VAL(TX_COMPWB_LSB, compwb);
1937                 compwb = TX_COMPWB_NEXT(compwb);
1938 #else
1939                 limit = readl(cp->regs + REG_TX_COMPN(ring));
1940 #endif
1941                 if (cp->tx_old[ring] != limit)
1942                         cas_tx_ringN(cp, ring, limit);
1943         }
1944 }
1945
1946
1947 static int cas_rx_process_pkt(struct cas *cp, struct cas_rx_comp *rxc,
1948                               int entry, const u64 *words,
1949                               struct sk_buff **skbref)
1950 {
1951         int dlen, hlen, len, i, alloclen;
1952         int off, swivel = RX_SWIVEL_OFF_VAL;
1953         struct cas_page *page;
1954         struct sk_buff *skb;
1955         void *addr, *crcaddr;
1956         __sum16 csum;
1957         char *p;
1958
1959         hlen = CAS_VAL(RX_COMP2_HDR_SIZE, words[1]);
1960         dlen = CAS_VAL(RX_COMP1_DATA_SIZE, words[0]);
1961         len  = hlen + dlen;
1962
1963         if (RX_COPY_ALWAYS || (words[2] & RX_COMP3_SMALL_PKT))
1964                 alloclen = len;
1965         else
1966                 alloclen = max(hlen, RX_COPY_MIN);
1967
1968         skb = dev_alloc_skb(alloclen + swivel + cp->crc_size);
1969         if (skb == NULL)
1970                 return -1;
1971
1972         *skbref = skb;
1973         skb_reserve(skb, swivel);
1974
1975         p = skb->data;
1976         addr = crcaddr = NULL;
1977         if (hlen) { /* always copy header pages */
1978                 i = CAS_VAL(RX_COMP2_HDR_INDEX, words[1]);
1979                 page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
1980                 off = CAS_VAL(RX_COMP2_HDR_OFF, words[1]) * 0x100 +
1981                         swivel;
1982
1983                 i = hlen;
1984                 if (!dlen) /* attach FCS */
1985                         i += cp->crc_size;
1986                 pci_dma_sync_single_for_cpu(cp->pdev, page->dma_addr + off, i,
1987                                     PCI_DMA_FROMDEVICE);
1988                 addr = cas_page_map(page->buffer);
1989                 memcpy(p, addr + off, i);
1990                 pci_dma_sync_single_for_device(cp->pdev, page->dma_addr + off, i,
1991                                     PCI_DMA_FROMDEVICE);
1992                 cas_page_unmap(addr);
1993                 RX_USED_ADD(page, 0x100);
1994                 p += hlen;
1995                 swivel = 0;
1996         }
1997
1998
1999         if (alloclen < (hlen + dlen)) {
2000                 skb_frag_t *frag = skb_shinfo(skb)->frags;
2001
2002                 /* normal or jumbo packets. we use frags */
2003                 i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
2004                 page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
2005                 off = CAS_VAL(RX_COMP1_DATA_OFF, words[0]) + swivel;
2006
2007                 hlen = min(cp->page_size - off, dlen);
2008                 if (hlen < 0) {
2009                         if (netif_msg_rx_err(cp)) {
2010                                 printk(KERN_DEBUG "%s: rx page overflow: "
2011                                        "%d\n", cp->dev->name, hlen);
2012                         }
2013                         dev_kfree_skb_irq(skb);
2014                         return -1;
2015                 }
2016                 i = hlen;
2017                 if (i == dlen)  /* attach FCS */
2018                         i += cp->crc_size;
2019                 pci_dma_sync_single_for_cpu(cp->pdev, page->dma_addr + off, i,
2020                                     PCI_DMA_FROMDEVICE);
2021
2022                 /* make sure we always copy a header */
2023                 swivel = 0;
2024                 if (p == (char *) skb->data) { /* not split */
2025                         addr = cas_page_map(page->buffer);
2026                         memcpy(p, addr + off, RX_COPY_MIN);
2027                         pci_dma_sync_single_for_device(cp->pdev, page->dma_addr + off, i,
2028                                         PCI_DMA_FROMDEVICE);
2029                         cas_page_unmap(addr);
2030                         off += RX_COPY_MIN;
2031                         swivel = RX_COPY_MIN;
2032                         RX_USED_ADD(page, cp->mtu_stride);
2033                 } else {
2034                         RX_USED_ADD(page, hlen);
2035                 }
2036                 skb_put(skb, alloclen);
2037
2038                 skb_shinfo(skb)->nr_frags++;
2039                 skb->data_len += hlen - swivel;
2040                 skb->truesize += hlen - swivel;
2041                 skb->len      += hlen - swivel;
2042
2043                 get_page(page->buffer);
2044                 frag->page = page->buffer;
2045                 frag->page_offset = off;
2046                 frag->size = hlen - swivel;
2047
2048                 /* any more data? */
2049                 if ((words[0] & RX_COMP1_SPLIT_PKT) && ((dlen -= hlen) > 0)) {
2050                         hlen = dlen;
2051                         off = 0;
2052
2053                         i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
2054                         page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
2055                         pci_dma_sync_single_for_cpu(cp->pdev, page->dma_addr,
2056                                             hlen + cp->crc_size,
2057                                             PCI_DMA_FROMDEVICE);
2058                         pci_dma_sync_single_for_device(cp->pdev, page->dma_addr,
2059                                             hlen + cp->crc_size,
2060                                             PCI_DMA_FROMDEVICE);
2061
2062                         skb_shinfo(skb)->nr_frags++;
2063                         skb->data_len += hlen;
2064                         skb->len      += hlen;
2065                         frag++;
2066
2067                         get_page(page->buffer);
2068                         frag->page = page->buffer;
2069                         frag->page_offset = 0;
2070                         frag->size = hlen;
2071                         RX_USED_ADD(page, hlen + cp->crc_size);
2072                 }
2073
2074                 if (cp->crc_size) {
2075                         addr = cas_page_map(page->buffer);
2076                         crcaddr  = addr + off + hlen;
2077                 }
2078
2079         } else {
2080                 /* copying packet */
2081                 if (!dlen)
2082                         goto end_copy_pkt;
2083
2084                 i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
2085                 page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
2086                 off = CAS_VAL(RX_COMP1_DATA_OFF, words[0]) + swivel;
2087                 hlen = min(cp->page_size - off, dlen);
2088                 if (hlen < 0) {
2089                         if (netif_msg_rx_err(cp)) {
2090                                 printk(KERN_DEBUG "%s: rx page overflow: "
2091                                        "%d\n", cp->dev->name, hlen);
2092                         }
2093                         dev_kfree_skb_irq(skb);
2094                         return -1;
2095                 }
2096                 i = hlen;
2097                 if (i == dlen) /* attach FCS */
2098                         i += cp->crc_size;
2099                 pci_dma_sync_single_for_cpu(cp->pdev, page->dma_addr + off, i,
2100                                     PCI_DMA_FROMDEVICE);
2101                 addr = cas_page_map(page->buffer);
2102                 memcpy(p, addr + off, i);
2103                 pci_dma_sync_single_for_device(cp->pdev, page->dma_addr + off, i,
2104                                     PCI_DMA_FROMDEVICE);
2105                 cas_page_unmap(addr);
2106                 if (p == (char *) skb->data) /* not split */
2107                         RX_USED_ADD(page, cp->mtu_stride);
2108                 else
2109                         RX_USED_ADD(page, i);
2110
2111                 /* any more data? */
2112                 if ((words[0] & RX_COMP1_SPLIT_PKT) && ((dlen -= hlen) > 0)) {
2113                         p += hlen;
2114                         i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
2115                         page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
2116                         pci_dma_sync_single_for_cpu(cp->pdev, page->dma_addr,
2117                                             dlen + cp->crc_size,
2118                                             PCI_DMA_FROMDEVICE);
2119                         addr = cas_page_map(page->buffer);
2120                         memcpy(p, addr, dlen + cp->crc_size);
2121                         pci_dma_sync_single_for_device(cp->pdev, page->dma_addr,
2122                                             dlen + cp->crc_size,
2123                                             PCI_DMA_FROMDEVICE);
2124                         cas_page_unmap(addr);
2125                         RX_USED_ADD(page, dlen + cp->crc_size);
2126                 }
2127 end_copy_pkt:
2128                 if (cp->crc_size) {
2129                         addr    = NULL;
2130                         crcaddr = skb->data + alloclen;
2131                 }
2132                 skb_put(skb, alloclen);
2133         }
2134
2135         csum = (__force __sum16)htons(CAS_VAL(RX_COMP4_TCP_CSUM, words[3]));
2136         if (cp->crc_size) {
2137                 /* checksum includes FCS. strip it out. */
2138                 csum = csum_fold(csum_partial(crcaddr, cp->crc_size,
2139                                               csum_unfold(csum)));
2140                 if (addr)
2141                         cas_page_unmap(addr);
2142         }
2143         skb->csum = csum_unfold(~csum);
2144         skb->ip_summed = CHECKSUM_COMPLETE;
2145         skb->protocol = eth_type_trans(skb, cp->dev);
2146         return len;
2147 }
2148
2149
2150 /* we can handle up to 64 rx flows at a time. we do the same thing
2151  * as nonreassm except that we batch up the buffers.
2152  * NOTE: we currently just treat each flow as a bunch of packets that
2153  *       we pass up. a better way would be to coalesce the packets
2154  *       into a jumbo packet. to do that, we need to do the following:
2155  *       1) the first packet will have a clean split between header and
2156  *          data. save both.
2157  *       2) each time the next flow packet comes in, extend the
2158  *          data length and merge the checksums.
2159  *       3) on flow release, fix up the header.
2160  *       4) make sure the higher layer doesn't care.
2161  * because packets get coalesced, we shouldn't run into fragment count
2162  * issues.
2163  */
2164 static inline void cas_rx_flow_pkt(struct cas *cp, const u64 *words,
2165                                    struct sk_buff *skb)
2166 {
2167         int flowid = CAS_VAL(RX_COMP3_FLOWID, words[2]) & (N_RX_FLOWS - 1);
2168         struct sk_buff_head *flow = &cp->rx_flows[flowid];
2169
2170         /* this is protected at a higher layer, so no need to
2171          * do any additional locking here. stick the buffer
2172          * at the end.
2173          */
2174         __skb_insert(skb, flow->prev, (struct sk_buff *) flow, flow);
2175         if (words[0] & RX_COMP1_RELEASE_FLOW) {
2176                 while ((skb = __skb_dequeue(flow))) {
2177                         cas_skb_release(skb);
2178                 }
2179         }
2180 }
2181
2182 /* put rx descriptor back on ring. if a buffer is in use by a higher
2183  * layer, this will need to put in a replacement.
2184  */
2185 static void cas_post_page(struct cas *cp, const int ring, const int index)
2186 {
2187         cas_page_t *new;
2188         int entry;
2189
2190         entry = cp->rx_old[ring];
2191
2192         new = cas_page_swap(cp, ring, index);
2193         cp->init_rxds[ring][entry].buffer = cpu_to_le64(new->dma_addr);
2194         cp->init_rxds[ring][entry].index  =
2195                 cpu_to_le64(CAS_BASE(RX_INDEX_NUM, index) |
2196                             CAS_BASE(RX_INDEX_RING, ring));
2197
2198         entry = RX_DESC_ENTRY(ring, entry + 1);
2199         cp->rx_old[ring] = entry;
2200
2201         if (entry % 4)
2202                 return;
2203
2204         if (ring == 0)
2205                 writel(entry, cp->regs + REG_RX_KICK);
2206         else if ((N_RX_DESC_RINGS > 1) &&
2207                  (cp->cas_flags & CAS_FLAG_REG_PLUS))
2208                 writel(entry, cp->regs + REG_PLUS_RX_KICK1);
2209 }
2210
2211
2212 /* only when things are bad */
2213 static int cas_post_rxds_ringN(struct cas *cp, int ring, int num)
2214 {
2215         unsigned int entry, last, count, released;
2216         int cluster;
2217         cas_page_t **page = cp->rx_pages[ring];
2218
2219         entry = cp->rx_old[ring];
2220
2221         if (netif_msg_intr(cp))
2222                 printk(KERN_DEBUG "%s: rxd[%d] interrupt, done: %d\n",
2223                        cp->dev->name, ring, entry);
2224
2225         cluster = -1;
2226         count = entry & 0x3;
2227         last = RX_DESC_ENTRY(ring, num ? entry + num - 4: entry - 4);
2228         released = 0;
2229         while (entry != last) {
2230                 /* make a new buffer if it's still in use */
2231                 if (page_count(page[entry]->buffer) > 1) {
2232                         cas_page_t *new = cas_page_dequeue(cp);
2233                         if (!new) {
2234                                 /* let the timer know that we need to
2235                                  * do this again
2236                                  */
2237                                 cp->cas_flags |= CAS_FLAG_RXD_POST(ring);
2238                                 if (!timer_pending(&cp->link_timer))
2239                                         mod_timer(&cp->link_timer, jiffies +
2240                                                   CAS_LINK_FAST_TIMEOUT);
2241                                 cp->rx_old[ring]  = entry;
2242                                 cp->rx_last[ring] = num ? num - released : 0;
2243                                 return -ENOMEM;
2244                         }
2245                         spin_lock(&cp->rx_inuse_lock);
2246                         list_add(&page[entry]->list, &cp->rx_inuse_list);
2247                         spin_unlock(&cp->rx_inuse_lock);
2248                         cp->init_rxds[ring][entry].buffer =
2249                                 cpu_to_le64(new->dma_addr);
2250                         page[entry] = new;
2251
2252                 }
2253
2254                 if (++count == 4) {
2255                         cluster = entry;
2256                         count = 0;
2257                 }
2258                 released++;
2259                 entry = RX_DESC_ENTRY(ring, entry + 1);
2260         }
2261         cp->rx_old[ring] = entry;
2262
2263         if (cluster < 0)
2264                 return 0;
2265
2266         if (ring == 0)
2267                 writel(cluster, cp->regs + REG_RX_KICK);
2268         else if ((N_RX_DESC_RINGS > 1) &&
2269                  (cp->cas_flags & CAS_FLAG_REG_PLUS))
2270                 writel(cluster, cp->regs + REG_PLUS_RX_KICK1);
2271         return 0;
2272 }
2273
2274
2275 /* process a completion ring. packets are set up in three basic ways:
2276  * small packets: should be copied header + data in single buffer.
2277  * large packets: header and data in a single buffer.
2278  * split packets: header in a separate buffer from data.
2279  *                data may be in multiple pages. data may be > 256
2280  *                bytes but in a single page.
2281  *
2282  * NOTE: RX page posting is done in this routine as well. while there's
2283  *       the capability of using multiple RX completion rings, it isn't
2284  *       really worthwhile due to the fact that the page posting will
2285  *       force serialization on the single descriptor ring.
2286  */
2287 static int cas_rx_ringN(struct cas *cp, int ring, int budget)
2288 {
2289         struct cas_rx_comp *rxcs = cp->init_rxcs[ring];
2290         int entry, drops;
2291         int npackets = 0;
2292
2293         if (netif_msg_intr(cp))
2294                 printk(KERN_DEBUG "%s: rx[%d] interrupt, done: %d/%d\n",
2295                        cp->dev->name, ring,
2296                        readl(cp->regs + REG_RX_COMP_HEAD),
2297                        cp->rx_new[ring]);
2298
2299         entry = cp->rx_new[ring];
2300         drops = 0;
2301         while (1) {
2302                 struct cas_rx_comp *rxc = rxcs + entry;
2303                 struct sk_buff *skb;
2304                 int type, len;
2305                 u64 words[4];
2306                 int i, dring;
2307
2308                 words[0] = le64_to_cpu(rxc->word1);
2309                 words[1] = le64_to_cpu(rxc->word2);
2310                 words[2] = le64_to_cpu(rxc->word3);
2311                 words[3] = le64_to_cpu(rxc->word4);
2312
2313                 /* don't touch if still owned by hw */
2314                 type = CAS_VAL(RX_COMP1_TYPE, words[0]);
2315                 if (type == 0)
2316                         break;
2317
2318                 /* hw hasn't cleared the zero bit yet */
2319                 if (words[3] & RX_COMP4_ZERO) {
2320                         break;
2321                 }
2322
2323                 /* get info on the packet */
2324                 if (words[3] & (RX_COMP4_LEN_MISMATCH | RX_COMP4_BAD)) {
2325                         spin_lock(&cp->stat_lock[ring]);
2326                         cp->net_stats[ring].rx_errors++;
2327                         if (words[3] & RX_COMP4_LEN_MISMATCH)
2328                                 cp->net_stats[ring].rx_length_errors++;
2329                         if (words[3] & RX_COMP4_BAD)
2330                                 cp->net_stats[ring].rx_crc_errors++;
2331                         spin_unlock(&cp->stat_lock[ring]);
2332
2333                         /* We'll just return it to Cassini. */
2334                 drop_it:
2335                         spin_lock(&cp->stat_lock[ring]);
2336                         ++cp->net_stats[ring].rx_dropped;
2337                         spin_unlock(&cp->stat_lock[ring]);
2338                         goto next;
2339                 }
2340
2341                 len = cas_rx_process_pkt(cp, rxc, entry, words, &skb);
2342                 if (len < 0) {
2343                         ++drops;
2344                         goto drop_it;
2345                 }
2346
2347                 /* see if it's a flow re-assembly or not. the driver
2348                  * itself handles release back up.
2349                  */
2350                 if (RX_DONT_BATCH || (type == 0x2)) {
2351                         /* non-reassm: these always get released */
2352                         cas_skb_release(skb);
2353                 } else {
2354                         cas_rx_flow_pkt(cp, words, skb);
2355                 }
2356
2357                 spin_lock(&cp->stat_lock[ring]);
2358                 cp->net_stats[ring].rx_packets++;
2359                 cp->net_stats[ring].rx_bytes += len;
2360                 spin_unlock(&cp->stat_lock[ring]);
2361                 cp->dev->last_rx = jiffies;
2362
2363         next:
2364                 npackets++;
2365
2366                 /* should it be released? */
2367                 if (words[0] & RX_COMP1_RELEASE_HDR) {
2368                         i = CAS_VAL(RX_COMP2_HDR_INDEX, words[1]);
2369                         dring = CAS_VAL(RX_INDEX_RING, i);
2370                         i = CAS_VAL(RX_INDEX_NUM, i);
2371                         cas_post_page(cp, dring, i);
2372                 }
2373
2374                 if (words[0] & RX_COMP1_RELEASE_DATA) {
2375                         i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
2376                         dring = CAS_VAL(RX_INDEX_RING, i);
2377                         i = CAS_VAL(RX_INDEX_NUM, i);
2378                         cas_post_page(cp, dring, i);
2379                 }
2380
2381                 if (words[0] & RX_COMP1_RELEASE_NEXT) {
2382                         i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
2383                         dring = CAS_VAL(RX_INDEX_RING, i);
2384                         i = CAS_VAL(RX_INDEX_NUM, i);
2385                         cas_post_page(cp, dring, i);
2386                 }
2387
2388                 /* skip to the next entry */
2389                 entry = RX_COMP_ENTRY(ring, entry + 1 +
2390                                       CAS_VAL(RX_COMP1_SKIP, words[0]));
2391 #ifdef USE_NAPI
2392                 if (budget && (npackets >= budget))
2393                         break;
2394 #endif
2395         }
2396         cp->rx_new[ring] = entry;
2397
2398         if (drops)
2399                 printk(KERN_INFO "%s: Memory squeeze, deferring packet.\n",
2400                        cp->dev->name);
2401         return npackets;
2402 }
2403
2404
2405 /* put completion entries back on the ring */
2406 static void cas_post_rxcs_ringN(struct net_device *dev,
2407                                 struct cas *cp, int ring)
2408 {
2409         struct cas_rx_comp *rxc = cp->init_rxcs[ring];
2410         int last, entry;
2411
2412         last = cp->rx_cur[ring];
2413         entry = cp->rx_new[ring];
2414         if (netif_msg_intr(cp))
2415                 printk(KERN_DEBUG "%s: rxc[%d] interrupt, done: %d/%d\n",
2416                        dev->name, ring, readl(cp->regs + REG_RX_COMP_HEAD),
2417                        entry);
2418
2419         /* zero and re-mark descriptors */
2420         while (last != entry) {
2421                 cas_rxc_init(rxc + last);
2422                 last = RX_COMP_ENTRY(ring, last + 1);
2423         }
2424         cp->rx_cur[ring] = last;
2425
2426         if (ring == 0)
2427                 writel(last, cp->regs + REG_RX_COMP_TAIL);
2428         else if (cp->cas_flags & CAS_FLAG_REG_PLUS)
2429                 writel(last, cp->regs + REG_PLUS_RX_COMPN_TAIL(ring));
2430 }
2431
2432
2433
2434 /* cassini can use all four PCI interrupts for the completion ring.
2435  * rings 3 and 4 are identical
2436  */
2437 #if defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
2438 static inline void cas_handle_irqN(struct net_device *dev,
2439                                    struct cas *cp, const u32 status,
2440                                    const int ring)
2441 {
2442         if (status & (INTR_RX_COMP_FULL_ALT | INTR_RX_COMP_AF_ALT))
2443                 cas_post_rxcs_ringN(dev, cp, ring);
2444 }
2445
2446 static irqreturn_t cas_interruptN(int irq, void *dev_id)
2447 {
2448         struct net_device *dev = dev_id;
2449         struct cas *cp = netdev_priv(dev);
2450         unsigned long flags;
2451         int ring;
2452         u32 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(ring));
2453
2454         /* check for shared irq */
2455         if (status == 0)
2456                 return IRQ_NONE;
2457
2458         ring = (irq == cp->pci_irq_INTC) ? 2 : 3;
2459         spin_lock_irqsave(&cp->lock, flags);
2460         if (status & INTR_RX_DONE_ALT) { /* handle rx separately */
2461 #ifdef USE_NAPI
2462                 cas_mask_intr(cp);
2463                 netif_rx_schedule(dev, &cp->napi);
2464 #else
2465                 cas_rx_ringN(cp, ring, 0);
2466 #endif
2467                 status &= ~INTR_RX_DONE_ALT;
2468         }
2469
2470         if (status)
2471                 cas_handle_irqN(dev, cp, status, ring);
2472         spin_unlock_irqrestore(&cp->lock, flags);
2473         return IRQ_HANDLED;
2474 }
2475 #endif
2476
2477 #ifdef USE_PCI_INTB
2478 /* everything but rx packets */
2479 static inline void cas_handle_irq1(struct cas *cp, const u32 status)
2480 {
2481         if (status & INTR_RX_BUF_UNAVAIL_1) {
2482                 /* Frame arrived, no free RX buffers available.
2483                  * NOTE: we can get this on a link transition. */
2484                 cas_post_rxds_ringN(cp, 1, 0);
2485                 spin_lock(&cp->stat_lock[1]);
2486                 cp->net_stats[1].rx_dropped++;
2487                 spin_unlock(&cp->stat_lock[1]);
2488         }
2489
2490         if (status & INTR_RX_BUF_AE_1)
2491                 cas_post_rxds_ringN(cp, 1, RX_DESC_RINGN_SIZE(1) -
2492                                     RX_AE_FREEN_VAL(1));
2493
2494         if (status & (INTR_RX_COMP_AF | INTR_RX_COMP_FULL))
2495                 cas_post_rxcs_ringN(cp, 1);
2496 }
2497
2498 /* ring 2 handles a few more events than 3 and 4 */
2499 static irqreturn_t cas_interrupt1(int irq, void *dev_id)
2500 {
2501         struct net_device *dev = dev_id;
2502         struct cas *cp = netdev_priv(dev);
2503         unsigned long flags;
2504         u32 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(1));
2505
2506         /* check for shared interrupt */
2507         if (status == 0)
2508                 return IRQ_NONE;
2509
2510         spin_lock_irqsave(&cp->lock, flags);
2511         if (status & INTR_RX_DONE_ALT) { /* handle rx separately */
2512 #ifdef USE_NAPI
2513                 cas_mask_intr(cp);
2514                 netif_rx_schedule(dev, &cp->napi);
2515 #else
2516                 cas_rx_ringN(cp, 1, 0);
2517 #endif
2518                 status &= ~INTR_RX_DONE_ALT;
2519         }
2520         if (status)
2521                 cas_handle_irq1(cp, status);
2522         spin_unlock_irqrestore(&cp->lock, flags);
2523         return IRQ_HANDLED;
2524 }
2525 #endif
2526
2527 static inline void cas_handle_irq(struct net_device *dev,
2528                                   struct cas *cp, const u32 status)
2529 {
2530         /* housekeeping interrupts */
2531         if (status & INTR_ERROR_MASK)
2532                 cas_abnormal_irq(dev, cp, status);
2533
2534         if (status & INTR_RX_BUF_UNAVAIL) {
2535                 /* Frame arrived, no free RX buffers available.
2536                  * NOTE: we can get this on a link transition.
2537                  */
2538                 cas_post_rxds_ringN(cp, 0, 0);
2539                 spin_lock(&cp->stat_lock[0]);
2540                 cp->net_stats[0].rx_dropped++;
2541                 spin_unlock(&cp->stat_lock[0]);
2542         } else if (status & INTR_RX_BUF_AE) {
2543                 cas_post_rxds_ringN(cp, 0, RX_DESC_RINGN_SIZE(0) -
2544                                     RX_AE_FREEN_VAL(0));
2545         }
2546
2547         if (status & (INTR_RX_COMP_AF | INTR_RX_COMP_FULL))
2548                 cas_post_rxcs_ringN(dev, cp, 0);
2549 }
2550
2551 static irqreturn_t cas_interrupt(int irq, void *dev_id)
2552 {
2553         struct net_device *dev = dev_id;
2554         struct cas *cp = netdev_priv(dev);
2555         unsigned long flags;
2556         u32 status = readl(cp->regs + REG_INTR_STATUS);
2557
2558         if (status == 0)
2559                 return IRQ_NONE;
2560
2561         spin_lock_irqsave(&cp->lock, flags);
2562         if (status & (INTR_TX_ALL | INTR_TX_INTME)) {
2563                 cas_tx(dev, cp, status);
2564                 status &= ~(INTR_TX_ALL | INTR_TX_INTME);
2565         }
2566
2567         if (status & INTR_RX_DONE) {
2568 #ifdef USE_NAPI
2569                 cas_mask_intr(cp);
2570                 netif_rx_schedule(dev, &cp->napi);
2571 #else
2572                 cas_rx_ringN(cp, 0, 0);
2573 #endif
2574                 status &= ~INTR_RX_DONE;
2575         }
2576
2577         if (status)
2578                 cas_handle_irq(dev, cp, status);
2579         spin_unlock_irqrestore(&cp->lock, flags);
2580         return IRQ_HANDLED;
2581 }
2582
2583
2584 #ifdef USE_NAPI
2585 static int cas_poll(struct napi_struct *napi, int budget)
2586 {
2587         struct cas *cp = container_of(napi, struct cas, napi);
2588         struct net_device *dev = cp->dev;
2589         int i, enable_intr, credits;
2590         u32 status = readl(cp->regs + REG_INTR_STATUS);
2591         unsigned long flags;
2592
2593         spin_lock_irqsave(&cp->lock, flags);
2594         cas_tx(dev, cp, status);
2595         spin_unlock_irqrestore(&cp->lock, flags);
2596
2597         /* NAPI rx packets. we spread the credits across all of the
2598          * rxc rings
2599          *
2600          * to make sure we're fair with the work we loop through each
2601          * ring N_RX_COMP_RING times with a request of
2602          * budget / N_RX_COMP_RINGS
2603          */
2604         enable_intr = 1;
2605         credits = 0;
2606         for (i = 0; i < N_RX_COMP_RINGS; i++) {
2607                 int j;
2608                 for (j = 0; j < N_RX_COMP_RINGS; j++) {
2609                         credits += cas_rx_ringN(cp, j, budget / N_RX_COMP_RINGS);
2610                         if (credits >= budget) {
2611                                 enable_intr = 0;
2612                                 goto rx_comp;
2613                         }
2614                 }
2615         }
2616
2617 rx_comp:
2618         /* final rx completion */
2619         spin_lock_irqsave(&cp->lock, flags);
2620         if (status)
2621                 cas_handle_irq(dev, cp, status);
2622
2623 #ifdef USE_PCI_INTB
2624         if (N_RX_COMP_RINGS > 1) {
2625                 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(1));
2626                 if (status)
2627                         cas_handle_irq1(dev, cp, status);
2628         }
2629 #endif
2630
2631 #ifdef USE_PCI_INTC
2632         if (N_RX_COMP_RINGS > 2) {
2633                 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(2));
2634                 if (status)
2635                         cas_handle_irqN(dev, cp, status, 2);
2636         }
2637 #endif
2638
2639 #ifdef USE_PCI_INTD
2640         if (N_RX_COMP_RINGS > 3) {
2641                 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(3));
2642                 if (status)
2643                         cas_handle_irqN(dev, cp, status, 3);
2644         }
2645 #endif
2646         spin_unlock_irqrestore(&cp->lock, flags);
2647         if (enable_intr) {
2648                 netif_rx_complete(dev, napi);
2649                 cas_unmask_intr(cp);
2650         }
2651         return credits;
2652 }
2653 #endif
2654
2655 #ifdef CONFIG_NET_POLL_CONTROLLER
2656 static void cas_netpoll(struct net_device *dev)
2657 {
2658         struct cas *cp = netdev_priv(dev);
2659
2660         cas_disable_irq(cp, 0);
2661         cas_interrupt(cp->pdev->irq, dev);
2662         cas_enable_irq(cp, 0);
2663
2664 #ifdef USE_PCI_INTB
2665         if (N_RX_COMP_RINGS > 1) {
2666                 /* cas_interrupt1(); */
2667         }
2668 #endif
2669 #ifdef USE_PCI_INTC
2670         if (N_RX_COMP_RINGS > 2) {
2671                 /* cas_interruptN(); */
2672         }
2673 #endif
2674 #ifdef USE_PCI_INTD
2675         if (N_RX_COMP_RINGS > 3) {
2676                 /* cas_interruptN(); */
2677         }
2678 #endif
2679 }
2680 #endif
2681
2682 static void cas_tx_timeout(struct net_device *dev)
2683 {
2684         struct cas *cp = netdev_priv(dev);
2685
2686         printk(KERN_ERR "%s: transmit timed out, resetting\n", dev->name);
2687         if (!cp->hw_running) {
2688                 printk("%s: hrm.. hw not running!\n", dev->name);
2689                 return;
2690         }
2691
2692         printk(KERN_ERR "%s: MIF_STATE[%08x]\n",
2693                dev->name, readl(cp->regs + REG_MIF_STATE_MACHINE));
2694
2695         printk(KERN_ERR "%s: MAC_STATE[%08x]\n",
2696                dev->name, readl(cp->regs + REG_MAC_STATE_MACHINE));
2697
2698         printk(KERN_ERR "%s: TX_STATE[%08x:%08x:%08x] "
2699                "FIFO[%08x:%08x:%08x] SM1[%08x] SM2[%08x]\n",
2700                dev->name,
2701                readl(cp->regs + REG_TX_CFG),
2702                readl(cp->regs + REG_MAC_TX_STATUS),
2703                readl(cp->regs + REG_MAC_TX_CFG),
2704                readl(cp->regs + REG_TX_FIFO_PKT_CNT),
2705                readl(cp->regs + REG_TX_FIFO_WRITE_PTR),
2706                readl(cp->regs + REG_TX_FIFO_READ_PTR),
2707                readl(cp->regs + REG_TX_SM_1),
2708                readl(cp->regs + REG_TX_SM_2));
2709
2710         printk(KERN_ERR "%s: RX_STATE[%08x:%08x:%08x]\n",
2711                dev->name,
2712                readl(cp->regs + REG_RX_CFG),
2713                readl(cp->regs + REG_MAC_RX_STATUS),
2714                readl(cp->regs + REG_MAC_RX_CFG));
2715
2716         printk(KERN_ERR "%s: HP_STATE[%08x:%08x:%08x:%08x]\n",
2717                dev->name,
2718                readl(cp->regs + REG_HP_STATE_MACHINE),
2719                readl(cp->regs + REG_HP_STATUS0),
2720                readl(cp->regs + REG_HP_STATUS1),
2721                readl(cp->regs + REG_HP_STATUS2));
2722
2723 #if 1
2724         atomic_inc(&cp->reset_task_pending);
2725         atomic_inc(&cp->reset_task_pending_all);
2726         schedule_work(&cp->reset_task);
2727 #else
2728         atomic_set(&cp->reset_task_pending, CAS_RESET_ALL);
2729         schedule_work(&cp->reset_task);
2730 #endif
2731 }
2732
2733 static inline int cas_intme(int ring, int entry)
2734 {
2735         /* Algorithm: IRQ every 1/2 of descriptors. */
2736         if (!(entry & ((TX_DESC_RINGN_SIZE(ring) >> 1) - 1)))
2737                 return 1;
2738         return 0;
2739 }
2740
2741
2742 static void cas_write_txd(struct cas *cp, int ring, int entry,
2743                           dma_addr_t mapping, int len, u64 ctrl, int last)
2744 {
2745         struct cas_tx_desc *txd = cp->init_txds[ring] + entry;
2746
2747         ctrl |= CAS_BASE(TX_DESC_BUFLEN, len);
2748         if (cas_intme(ring, entry))
2749                 ctrl |= TX_DESC_INTME;
2750         if (last)
2751                 ctrl |= TX_DESC_EOF;
2752         txd->control = cpu_to_le64(ctrl);
2753         txd->buffer = cpu_to_le64(mapping);
2754 }
2755
2756 static inline void *tx_tiny_buf(struct cas *cp, const int ring,
2757                                 const int entry)
2758 {
2759         return cp->tx_tiny_bufs[ring] + TX_TINY_BUF_LEN*entry;
2760 }
2761
2762 static inline dma_addr_t tx_tiny_map(struct cas *cp, const int ring,
2763                                      const int entry, const int tentry)
2764 {
2765         cp->tx_tiny_use[ring][tentry].nbufs++;
2766         cp->tx_tiny_use[ring][entry].used = 1;
2767         return cp->tx_tiny_dvma[ring] + TX_TINY_BUF_LEN*entry;
2768 }
2769
2770 static inline int cas_xmit_tx_ringN(struct cas *cp, int ring,
2771                                     struct sk_buff *skb)
2772 {
2773         struct net_device *dev = cp->dev;
2774         int entry, nr_frags, frag, tabort, tentry;
2775         dma_addr_t mapping;
2776         unsigned long flags;
2777         u64 ctrl;
2778         u32 len;
2779
2780         spin_lock_irqsave(&cp->tx_lock[ring], flags);
2781
2782         /* This is a hard error, log it. */
2783         if (TX_BUFFS_AVAIL(cp, ring) <=
2784             CAS_TABORT(cp)*(skb_shinfo(skb)->nr_frags + 1)) {
2785                 netif_stop_queue(dev);
2786                 spin_unlock_irqrestore(&cp->tx_lock[ring], flags);
2787                 printk(KERN_ERR PFX "%s: BUG! Tx Ring full when "
2788                        "queue awake!\n", dev->name);
2789                 return 1;
2790         }
2791
2792         ctrl = 0;
2793         if (skb->ip_summed == CHECKSUM_PARTIAL) {
2794                 const u64 csum_start_off = skb_transport_offset(skb);
2795                 const u64 csum_stuff_off = csum_start_off + skb->csum_offset;
2796
2797                 ctrl =  TX_DESC_CSUM_EN |
2798                         CAS_BASE(TX_DESC_CSUM_START, csum_start_off) |
2799                         CAS_BASE(TX_DESC_CSUM_STUFF, csum_stuff_off);
2800         }
2801
2802         entry = cp->tx_new[ring];
2803         cp->tx_skbs[ring][entry] = skb;
2804
2805         nr_frags = skb_shinfo(skb)->nr_frags;
2806         len = skb_headlen(skb);
2807         mapping = pci_map_page(cp->pdev, virt_to_page(skb->data),
2808                                offset_in_page(skb->data), len,
2809                                PCI_DMA_TODEVICE);
2810
2811         tentry = entry;
2812         tabort = cas_calc_tabort(cp, (unsigned long) skb->data, len);
2813         if (unlikely(tabort)) {
2814                 /* NOTE: len is always >  tabort */
2815                 cas_write_txd(cp, ring, entry, mapping, len - tabort,
2816                               ctrl | TX_DESC_SOF, 0);
2817                 entry = TX_DESC_NEXT(ring, entry);
2818
2819                 skb_copy_from_linear_data_offset(skb, len - tabort,
2820                               tx_tiny_buf(cp, ring, entry), tabort);
2821                 mapping = tx_tiny_map(cp, ring, entry, tentry);
2822                 cas_write_txd(cp, ring, entry, mapping, tabort, ctrl,
2823                               (nr_frags == 0));
2824         } else {
2825                 cas_write_txd(cp, ring, entry, mapping, len, ctrl |
2826                               TX_DESC_SOF, (nr_frags == 0));
2827         }
2828         entry = TX_DESC_NEXT(ring, entry);
2829
2830         for (frag = 0; frag < nr_frags; frag++) {
2831                 skb_frag_t *fragp = &skb_shinfo(skb)->frags[frag];
2832
2833                 len = fragp->size;
2834                 mapping = pci_map_page(cp->pdev, fragp->page,
2835                                        fragp->page_offset, len,
2836                                        PCI_DMA_TODEVICE);
2837
2838                 tabort = cas_calc_tabort(cp, fragp->page_offset, len);
2839                 if (unlikely(tabort)) {
2840                         void *addr;
2841
2842                         /* NOTE: len is always > tabort */
2843                         cas_write_txd(cp, ring, entry, mapping, len - tabort,
2844                                       ctrl, 0);
2845                         entry = TX_DESC_NEXT(ring, entry);
2846
2847                         addr = cas_page_map(fragp->page);
2848                         memcpy(tx_tiny_buf(cp, ring, entry),
2849                                addr + fragp->page_offset + len - tabort,
2850                                tabort);
2851                         cas_page_unmap(addr);
2852                         mapping = tx_tiny_map(cp, ring, entry, tentry);
2853                         len     = tabort;
2854                 }
2855
2856                 cas_write_txd(cp, ring, entry, mapping, len, ctrl,
2857                               (frag + 1 == nr_frags));
2858                 entry = TX_DESC_NEXT(ring, entry);
2859         }
2860
2861         cp->tx_new[ring] = entry;
2862         if (TX_BUFFS_AVAIL(cp, ring) <= CAS_TABORT(cp)*(MAX_SKB_FRAGS + 1))
2863                 netif_stop_queue(dev);
2864
2865         if (netif_msg_tx_queued(cp))
2866                 printk(KERN_DEBUG "%s: tx[%d] queued, slot %d, skblen %d, "
2867                        "avail %d\n",
2868                        dev->name, ring, entry, skb->len,
2869                        TX_BUFFS_AVAIL(cp, ring));
2870         writel(entry, cp->regs + REG_TX_KICKN(ring));
2871         spin_unlock_irqrestore(&cp->tx_lock[ring], flags);
2872         return 0;
2873 }
2874
2875 static int cas_start_xmit(struct sk_buff *skb, struct net_device *dev)
2876 {
2877         struct cas *cp = netdev_priv(dev);
2878
2879         /* this is only used as a load-balancing hint, so it doesn't
2880          * need to be SMP safe
2881          */
2882         static int ring;
2883
2884         if (skb_padto(skb, cp->min_frame_size))
2885                 return 0;
2886
2887         /* XXX: we need some higher-level QoS hooks to steer packets to
2888          *      individual queues.
2889          */
2890         if (cas_xmit_tx_ringN(cp, ring++ & N_TX_RINGS_MASK, skb))
2891                 return 1;
2892         dev->trans_start = jiffies;
2893         return 0;
2894 }
2895
2896 static void cas_init_tx_dma(struct cas *cp)
2897 {
2898         u64 desc_dma = cp->block_dvma;
2899         unsigned long off;
2900         u32 val;
2901         int i;
2902
2903         /* set up tx completion writeback registers. must be 8-byte aligned */
2904 #ifdef USE_TX_COMPWB
2905         off = offsetof(struct cas_init_block, tx_compwb);
2906         writel((desc_dma + off) >> 32, cp->regs + REG_TX_COMPWB_DB_HI);
2907         writel((desc_dma + off) & 0xffffffff, cp->regs + REG_TX_COMPWB_DB_LOW);
2908 #endif
2909
2910         /* enable completion writebacks, enable paced mode,
2911          * disable read pipe, and disable pre-interrupt compwbs
2912          */
2913         val =   TX_CFG_COMPWB_Q1 | TX_CFG_COMPWB_Q2 |
2914                 TX_CFG_COMPWB_Q3 | TX_CFG_COMPWB_Q4 |
2915                 TX_CFG_DMA_RDPIPE_DIS | TX_CFG_PACED_MODE |
2916                 TX_CFG_INTR_COMPWB_DIS;
2917
2918         /* write out tx ring info and tx desc bases */
2919         for (i = 0; i < MAX_TX_RINGS; i++) {
2920                 off = (unsigned long) cp->init_txds[i] -
2921                         (unsigned long) cp->init_block;
2922
2923                 val |= CAS_TX_RINGN_BASE(i);
2924                 writel((desc_dma + off) >> 32, cp->regs + REG_TX_DBN_HI(i));
2925                 writel((desc_dma + off) & 0xffffffff, cp->regs +
2926                        REG_TX_DBN_LOW(i));
2927                 /* don't zero out the kick register here as the system
2928                  * will wedge
2929                  */
2930         }
2931         writel(val, cp->regs + REG_TX_CFG);
2932
2933         /* program max burst sizes. these numbers should be different
2934          * if doing QoS.
2935          */
2936 #ifdef USE_QOS
2937         writel(0x800, cp->regs + REG_TX_MAXBURST_0);
2938         writel(0x1600, cp->regs + REG_TX_MAXBURST_1);
2939         writel(0x2400, cp->regs + REG_TX_MAXBURST_2);
2940         writel(0x4800, cp->regs + REG_TX_MAXBURST_3);
2941 #else
2942         writel(0x800, cp->regs + REG_TX_MAXBURST_0);
2943         writel(0x800, cp->regs + REG_TX_MAXBURST_1);
2944         writel(0x800, cp->regs + REG_TX_MAXBURST_2);
2945         writel(0x800, cp->regs + REG_TX_MAXBURST_3);
2946 #endif
2947 }
2948
2949 /* Must be invoked under cp->lock. */
2950 static inline void cas_init_dma(struct cas *cp)
2951 {
2952         cas_init_tx_dma(cp);
2953         cas_init_rx_dma(cp);
2954 }
2955
2956 /* Must be invoked under cp->lock. */
2957 static u32 cas_setup_multicast(struct cas *cp)
2958 {
2959         u32 rxcfg = 0;
2960         int i;
2961
2962         if (cp->dev->flags & IFF_PROMISC) {
2963                 rxcfg |= MAC_RX_CFG_PROMISC_EN;
2964
2965         } else if (cp->dev->flags & IFF_ALLMULTI) {
2966                 for (i=0; i < 16; i++)
2967                         writel(0xFFFF, cp->regs + REG_MAC_HASH_TABLEN(i));
2968                 rxcfg |= MAC_RX_CFG_HASH_FILTER_EN;
2969
2970         } else {
2971                 u16 hash_table[16];
2972                 u32 crc;
2973                 struct dev_mc_list *dmi = cp->dev->mc_list;
2974                 int i;
2975
2976                 /* use the alternate mac address registers for the
2977                  * first 15 multicast addresses
2978                  */
2979                 for (i = 1; i <= CAS_MC_EXACT_MATCH_SIZE; i++) {
2980                         if (!dmi) {
2981                                 writel(0x0, cp->regs + REG_MAC_ADDRN(i*3 + 0));
2982                                 writel(0x0, cp->regs + REG_MAC_ADDRN(i*3 + 1));
2983                                 writel(0x0, cp->regs + REG_MAC_ADDRN(i*3 + 2));
2984                                 continue;
2985                         }
2986                         writel((dmi->dmi_addr[4] << 8) | dmi->dmi_addr[5],
2987                                cp->regs + REG_MAC_ADDRN(i*3 + 0));
2988                         writel((dmi->dmi_addr[2] << 8) | dmi->dmi_addr[3],
2989                                cp->regs + REG_MAC_ADDRN(i*3 + 1));
2990                         writel((dmi->dmi_addr[0] << 8) | dmi->dmi_addr[1],
2991                                cp->regs + REG_MAC_ADDRN(i*3 + 2));
2992                         dmi = dmi->next;
2993                 }
2994
2995                 /* use hw hash table for the next series of
2996                  * multicast addresses
2997                  */
2998                 memset(hash_table, 0, sizeof(hash_table));
2999                 while (dmi) {
3000                         crc = ether_crc_le(ETH_ALEN, dmi->dmi_addr);
3001                         crc >>= 24;
3002                         hash_table[crc >> 4] |= 1 << (15 - (crc & 0xf));
3003                         dmi = dmi->next;
3004                 }
3005                 for (i=0; i < 16; i++)
3006                         writel(hash_table[i], cp->regs +
3007                                REG_MAC_HASH_TABLEN(i));
3008                 rxcfg |= MAC_RX_CFG_HASH_FILTER_EN;
3009         }
3010
3011         return rxcfg;
3012 }
3013
3014 /* must be invoked under cp->stat_lock[N_TX_RINGS] */
3015 static void cas_clear_mac_err(struct cas *cp)
3016 {
3017         writel(0, cp->regs + REG_MAC_COLL_NORMAL);
3018         writel(0, cp->regs + REG_MAC_COLL_FIRST);
3019         writel(0, cp->regs + REG_MAC_COLL_EXCESS);
3020         writel(0, cp->regs + REG_MAC_COLL_LATE);
3021         writel(0, cp->regs + REG_MAC_TIMER_DEFER);
3022         writel(0, cp->regs + REG_MAC_ATTEMPTS_PEAK);
3023         writel(0, cp->regs + REG_MAC_RECV_FRAME);
3024         writel(0, cp->regs + REG_MAC_LEN_ERR);
3025         writel(0, cp->regs + REG_MAC_ALIGN_ERR);
3026         writel(0, cp->regs + REG_MAC_FCS_ERR);
3027         writel(0, cp->regs + REG_MAC_RX_CODE_ERR);
3028 }
3029
3030
3031 static void cas_mac_reset(struct cas *cp)
3032 {
3033         int i;
3034
3035         /* do both TX and RX reset */
3036         writel(0x1, cp->regs + REG_MAC_TX_RESET);
3037         writel(0x1, cp->regs + REG_MAC_RX_RESET);
3038
3039         /* wait for TX */
3040         i = STOP_TRIES;
3041         while (i-- > 0) {
3042                 if (readl(cp->regs + REG_MAC_TX_RESET) == 0)
3043                         break;
3044                 udelay(10);
3045         }
3046
3047         /* wait for RX */
3048         i = STOP_TRIES;
3049         while (i-- > 0) {
3050                 if (readl(cp->regs + REG_MAC_RX_RESET) == 0)
3051                         break;
3052                 udelay(10);
3053         }
3054
3055         if (readl(cp->regs + REG_MAC_TX_RESET) |
3056             readl(cp->regs + REG_MAC_RX_RESET))
3057                 printk(KERN_ERR "%s: mac tx[%d]/rx[%d] reset failed [%08x]\n",
3058                        cp->dev->name, readl(cp->regs + REG_MAC_TX_RESET),
3059                        readl(cp->regs + REG_MAC_RX_RESET),
3060                        readl(cp->regs + REG_MAC_STATE_MACHINE));
3061 }
3062
3063
3064 /* Must be invoked under cp->lock. */
3065 static void cas_init_mac(struct cas *cp)
3066 {
3067         unsigned char *e = &cp->dev->dev_addr[0];
3068         int i;
3069 #ifdef CONFIG_CASSINI_MULTICAST_REG_WRITE
3070         u32 rxcfg;
3071 #endif
3072         cas_mac_reset(cp);
3073
3074         /* setup core arbitration weight register */
3075         writel(CAWR_RR_DIS, cp->regs + REG_CAWR);
3076
3077         /* XXX Use pci_dma_burst_advice() */
3078 #if !defined(CONFIG_SPARC64) && !defined(CONFIG_ALPHA)
3079         /* set the infinite burst register for chips that don't have
3080          * pci issues.
3081          */
3082         if ((cp->cas_flags & CAS_FLAG_TARGET_ABORT) == 0)
3083                 writel(INF_BURST_EN, cp->regs + REG_INF_BURST);
3084 #endif
3085
3086         writel(0x1BF0, cp->regs + REG_MAC_SEND_PAUSE);
3087
3088         writel(0x00, cp->regs + REG_MAC_IPG0);
3089         writel(0x08, cp->regs + REG_MAC_IPG1);
3090         writel(0x04, cp->regs + REG_MAC_IPG2);
3091
3092         /* change later for 802.3z */
3093         writel(0x40, cp->regs + REG_MAC_SLOT_TIME);
3094
3095         /* min frame + FCS */
3096         writel(ETH_ZLEN + 4, cp->regs + REG_MAC_FRAMESIZE_MIN);
3097
3098         /* Ethernet payload + header + FCS + optional VLAN tag. NOTE: we
3099          * specify the maximum frame size to prevent RX tag errors on
3100          * oversized frames.
3101          */
3102         writel(CAS_BASE(MAC_FRAMESIZE_MAX_BURST, 0x2000) |
3103                CAS_BASE(MAC_FRAMESIZE_MAX_FRAME,
3104                         (CAS_MAX_MTU + ETH_HLEN + 4 + 4)),
3105                cp->regs + REG_MAC_FRAMESIZE_MAX);
3106
3107         /* NOTE: crc_size is used as a surrogate for half-duplex.
3108          * workaround saturn half-duplex issue by increasing preamble
3109          * size to 65 bytes.
3110          */
3111         if ((cp->cas_flags & CAS_FLAG_SATURN) && cp->crc_size)
3112                 writel(0x41, cp->regs + REG_MAC_PA_SIZE);
3113         else
3114                 writel(0x07, cp->regs + REG_MAC_PA_SIZE);
3115         writel(0x04, cp->regs + REG_MAC_JAM_SIZE);
3116         writel(0x10, cp->regs + REG_MAC_ATTEMPT_LIMIT);
3117         writel(0x8808, cp->regs + REG_MAC_CTRL_TYPE);
3118
3119         writel((e[5] | (e[4] << 8)) & 0x3ff, cp->regs + REG_MAC_RANDOM_SEED);
3120
3121         writel(0, cp->regs + REG_MAC_ADDR_FILTER0);
3122         writel(0, cp->regs + REG_MAC_ADDR_FILTER1);
3123         writel(0, cp->regs + REG_MAC_ADDR_FILTER2);
3124         writel(0, cp->regs + REG_MAC_ADDR_FILTER2_1_MASK);
3125         writel(0, cp->regs + REG_MAC_ADDR_FILTER0_MASK);
3126
3127         /* setup mac address in perfect filter array */
3128         for (i = 0; i < 45; i++)
3129                 writel(0x0, cp->regs + REG_MAC_ADDRN(i));
3130
3131         writel((e[4] << 8) | e[5], cp->regs + REG_MAC_ADDRN(0));
3132         writel((e[2] << 8) | e[3], cp->regs + REG_MAC_ADDRN(1));
3133         writel((e[0] << 8) | e[1], cp->regs + REG_MAC_ADDRN(2));
3134
3135         writel(0x0001, cp->regs + REG_MAC_ADDRN(42));
3136         writel(0xc200, cp->regs + REG_MAC_ADDRN(43));
3137         writel(0x0180, cp->regs + REG_MAC_ADDRN(44));
3138
3139 #ifndef CONFIG_CASSINI_MULTICAST_REG_WRITE
3140         cp->mac_rx_cfg = cas_setup_multicast(cp);
3141 #else
3142         /* WTZ: Do what Adrian did in cas_set_multicast. Doing
3143          * a writel does not seem to be necessary because Cassini
3144          * seems to preserve the configuration when we do the reset.
3145          * If the chip is in trouble, though, it is not clear if we
3146          * can really count on this behavior. cas_set_multicast uses
3147          * spin_lock_irqsave, but we are called only in cas_init_hw and
3148          * cas_init_hw is protected by cas_lock_all, which calls
3149          * spin_lock_irq (so it doesn't need to save the flags, and
3150          * we should be OK for the writel, as that is the only
3151          * difference).
3152          */
3153         cp->mac_rx_cfg = rxcfg = cas_setup_multicast(cp);
3154         writel(rxcfg, cp->regs + REG_MAC_RX_CFG);
3155 #endif
3156         spin_lock(&cp->stat_lock[N_TX_RINGS]);
3157         cas_clear_mac_err(cp);
3158         spin_unlock(&cp->stat_lock[N_TX_RINGS]);
3159
3160         /* Setup MAC interrupts.  We want to get all of the interesting
3161          * counter expiration events, but we do not want to hear about
3162          * normal rx/tx as the DMA engine tells us that.
3163          */
3164         writel(MAC_TX_FRAME_XMIT, cp->regs + REG_MAC_TX_MASK);
3165         writel(MAC_RX_FRAME_RECV, cp->regs + REG_MAC_RX_MASK);
3166
3167         /* Don't enable even the PAUSE interrupts for now, we
3168          * make no use of those events other than to record them.
3169          */
3170         writel(0xffffffff, cp->regs + REG_MAC_CTRL_MASK);
3171 }
3172
3173 /* Must be invoked under cp->lock. */
3174 static void cas_init_pause_thresholds(struct cas *cp)
3175 {
3176         /* Calculate pause thresholds.  Setting the OFF threshold to the
3177          * full RX fifo size effectively disables PAUSE generation
3178          */
3179         if (cp->rx_fifo_size <= (2 * 1024)) {
3180                 cp->rx_pause_off = cp->rx_pause_on = cp->rx_fifo_size;
3181         } else {
3182                 int max_frame = (cp->dev->mtu + ETH_HLEN + 4 + 4 + 64) & ~63;
3183                 if (max_frame * 3 > cp->rx_fifo_size) {
3184                         cp->rx_pause_off = 7104;
3185                         cp->rx_pause_on  = 960;
3186                 } else {
3187                         int off = (cp->rx_fifo_size - (max_frame * 2));
3188                         int on = off - max_frame;
3189                         cp->rx_pause_off = off;
3190                         cp->rx_pause_on = on;
3191                 }
3192         }
3193 }
3194
3195 static int cas_vpd_match(const void __iomem *p, const char *str)
3196 {
3197         int len = strlen(str) + 1;
3198         int i;
3199
3200         for (i = 0; i < len; i++) {
3201                 if (readb(p + i) != str[i])
3202                         return 0;
3203         }
3204         return 1;
3205 }
3206
3207
3208 /* get the mac address by reading the vpd information in the rom.
3209  * also get the phy type and determine if there's an entropy generator.
3210  * NOTE: this is a bit convoluted for the following reasons:
3211  *  1) vpd info has order-dependent mac addresses for multinic cards
3212  *  2) the only way to determine the nic order is to use the slot
3213  *     number.
3214  *  3) fiber cards don't have bridges, so their slot numbers don't
3215  *     mean anything.
3216  *  4) we don't actually know we have a fiber card until after
3217  *     the mac addresses are parsed.
3218  */
3219 static int cas_get_vpd_info(struct cas *cp, unsigned char *dev_addr,
3220                             const int offset)
3221 {
3222         void __iomem *p = cp->regs + REG_EXPANSION_ROM_RUN_START;
3223         void __iomem *base, *kstart;
3224         int i, len;
3225         int found = 0;
3226 #define VPD_FOUND_MAC        0x01
3227 #define VPD_FOUND_PHY        0x02
3228
3229         int phy_type = CAS_PHY_MII_MDIO0; /* default phy type */
3230         int mac_off  = 0;
3231
3232         /* give us access to the PROM */
3233         writel(BIM_LOCAL_DEV_PROM | BIM_LOCAL_DEV_PAD,
3234                cp->regs + REG_BIM_LOCAL_DEV_EN);
3235
3236         /* check for an expansion rom */
3237         if (readb(p) != 0x55 || readb(p + 1) != 0xaa)
3238                 goto use_random_mac_addr;
3239
3240         /* search for beginning of vpd */
3241         base = NULL;
3242         for (i = 2; i < EXPANSION_ROM_SIZE; i++) {
3243                 /* check for PCIR */
3244                 if ((readb(p + i + 0) == 0x50) &&
3245                     (readb(p + i + 1) == 0x43) &&
3246                     (readb(p + i + 2) == 0x49) &&
3247                     (readb(p + i + 3) == 0x52)) {
3248                         base = p + (readb(p + i + 8) |
3249                                     (readb(p + i + 9) << 8));
3250                         break;
3251                 }
3252         }
3253
3254         if (!base || (readb(base) != 0x82))
3255                 goto use_random_mac_addr;
3256
3257         i = (readb(base + 1) | (readb(base + 2) << 8)) + 3;
3258         while (i < EXPANSION_ROM_SIZE) {
3259                 if (readb(base + i) != 0x90) /* no vpd found */
3260                         goto use_random_mac_addr;
3261
3262                 /* found a vpd field */
3263                 len = readb(base + i + 1) | (readb(base + i + 2) << 8);
3264
3265                 /* extract keywords */
3266                 kstart = base + i + 3;
3267                 p = kstart;
3268                 while ((p - kstart) < len) {
3269                         int klen = readb(p + 2);
3270                         int j;
3271                         char type;
3272
3273                         p += 3;
3274
3275                         /* look for the following things:
3276                          * -- correct length == 29
3277                          * 3 (type) + 2 (size) +
3278                          * 18 (strlen("local-mac-address") + 1) +
3279                          * 6 (mac addr)
3280                          * -- VPD Instance 'I'
3281                          * -- VPD Type Bytes 'B'
3282                          * -- VPD data length == 6
3283                          * -- property string == local-mac-address
3284                          *
3285                          * -- correct length == 24
3286                          * 3 (type) + 2 (size) +
3287                          * 12 (strlen("entropy-dev") + 1) +
3288                          * 7 (strlen("vms110") + 1)
3289                          * -- VPD Instance 'I'
3290                          * -- VPD Type String 'B'
3291                          * -- VPD data length == 7
3292                          * -- property string == entropy-dev
3293                          *
3294                          * -- correct length == 18
3295                          * 3 (type) + 2 (size) +
3296                          * 9 (strlen("phy-type") + 1) +
3297                          * 4 (strlen("pcs") + 1)
3298                          * -- VPD Instance 'I'
3299                          * -- VPD Type String 'S'
3300                          * -- VPD data length == 4
3301                          * -- property string == phy-type
3302                          *
3303                          * -- correct length == 23
3304                          * 3 (type) + 2 (size) +
3305                          * 14 (strlen("phy-interface") + 1) +
3306                          * 4 (strlen("pcs") + 1)
3307                          * -- VPD Instance 'I'
3308                          * -- VPD Type String 'S'
3309                          * -- VPD data length == 4
3310                          * -- property string == phy-interface
3311                          */
3312                         if (readb(p) != 'I')
3313                                 goto next;
3314
3315                         /* finally, check string and length */
3316                         type = readb(p + 3);
3317                         if (type == 'B') {
3318                                 if ((klen == 29) && readb(p + 4) == 6 &&
3319                                     cas_vpd_match(p + 5,
3320                                                   "local-mac-address")) {
3321                                         if (mac_off++ > offset)
3322                                                 goto next;
3323
3324                                         /* set mac address */
3325                                         for (j = 0; j < 6; j++)
3326                                                 dev_addr[j] =
3327                                                         readb(p + 23 + j);
3328                                         goto found_mac;
3329                                 }
3330                         }
3331
3332                         if (type != 'S')
3333                                 goto next;
3334
3335 #ifdef USE_ENTROPY_DEV
3336                         if ((klen == 24) &&
3337                             cas_vpd_match(p + 5, "entropy-dev") &&
3338                             cas_vpd_match(p + 17, "vms110")) {
3339                                 cp->cas_flags |= CAS_FLAG_ENTROPY_DEV;
3340                                 goto next;
3341                         }
3342 #endif
3343
3344                         if (found & VPD_FOUND_PHY)
3345                                 goto next;
3346
3347                         if ((klen == 18) && readb(p + 4) == 4 &&
3348                             cas_vpd_match(p + 5, "phy-type")) {
3349                                 if (cas_vpd_match(p + 14, "pcs")) {
3350                                         phy_type = CAS_PHY_SERDES;
3351                                         goto found_phy;
3352                                 }
3353                         }
3354
3355                         if ((klen == 23) && readb(p + 4) == 4 &&
3356                             cas_vpd_match(p + 5, "phy-interface")) {
3357                                 if (cas_vpd_match(p + 19, "pcs")) {
3358                                         phy_type = CAS_PHY_SERDES;
3359                                         goto found_phy;
3360                                 }
3361                         }
3362 found_mac:
3363                         found |= VPD_FOUND_MAC;
3364                         goto next;
3365
3366 found_phy:
3367                         found |= VPD_FOUND_PHY;
3368
3369 next:
3370                         p += klen;
3371                 }
3372                 i += len + 3;
3373         }
3374
3375 use_random_mac_addr:
3376         if (found & VPD_FOUND_MAC)
3377                 goto done;
3378
3379         /* Sun MAC prefix then 3 random bytes. */
3380         printk(PFX "MAC address not found in ROM VPD\n");
3381         dev_addr[0] = 0x08;
3382         dev_addr[1] = 0x00;
3383         dev_addr[2] = 0x20;
3384         get_random_bytes(dev_addr + 3, 3);
3385
3386 done:
3387         writel(0, cp->regs + REG_BIM_LOCAL_DEV_EN);
3388         return phy_type;
3389 }
3390
3391 /* check pci invariants */
3392 static void cas_check_pci_invariants(struct cas *cp)
3393 {
3394         struct pci_dev *pdev = cp->pdev;
3395
3396         cp->cas_flags = 0;
3397         if ((pdev->vendor == PCI_VENDOR_ID_SUN) &&
3398             (pdev->device == PCI_DEVICE_ID_SUN_CASSINI)) {
3399                 if (pdev->revision >= CAS_ID_REVPLUS)
3400                         cp->cas_flags |= CAS_FLAG_REG_PLUS;
3401                 if (pdev->revision < CAS_ID_REVPLUS02u)
3402                         cp->cas_flags |= CAS_FLAG_TARGET_ABORT;
3403
3404                 /* Original Cassini supports HW CSUM, but it's not
3405                  * enabled by default as it can trigger TX hangs.
3406                  */
3407                 if (pdev->revision < CAS_ID_REV2)
3408                         cp->cas_flags |= CAS_FLAG_NO_HW_CSUM;
3409         } else {
3410                 /* Only sun has original cassini chips.  */
3411                 cp->cas_flags |= CAS_FLAG_REG_PLUS;
3412
3413                 /* We use a flag because the same phy might be externally
3414                  * connected.
3415                  */
3416                 if ((pdev->vendor == PCI_VENDOR_ID_NS) &&
3417                     (pdev->device == PCI_DEVICE_ID_NS_SATURN))
3418                         cp->cas_flags |= CAS_FLAG_SATURN;
3419         }
3420 }
3421
3422
3423 static int cas_check_invariants(struct cas *cp)
3424 {
3425         struct pci_dev *pdev = cp->pdev;
3426         u32 cfg;
3427         int i;
3428
3429         /* get page size for rx buffers. */
3430         cp->page_order = 0;
3431 #ifdef USE_PAGE_ORDER
3432         if (PAGE_SHIFT < CAS_JUMBO_PAGE_SHIFT) {
3433                 /* see if we can allocate larger pages */
3434                 struct page *page = alloc_pages(GFP_ATOMIC,
3435                                                 CAS_JUMBO_PAGE_SHIFT -
3436                                                 PAGE_SHIFT);
3437                 if (page) {
3438                         __free_pages(page, CAS_JUMBO_PAGE_SHIFT - PAGE_SHIFT);
3439                         cp->page_order = CAS_JUMBO_PAGE_SHIFT - PAGE_SHIFT;
3440                 } else {
3441                         printk(PFX "MTU limited to %d bytes\n", CAS_MAX_MTU);
3442                 }
3443         }
3444 #endif
3445         cp->page_size = (PAGE_SIZE << cp->page_order);
3446
3447         /* Fetch the FIFO configurations. */
3448         cp->tx_fifo_size = readl(cp->regs + REG_TX_FIFO_SIZE) * 64;
3449         cp->rx_fifo_size = RX_FIFO_SIZE;
3450
3451         /* finish phy determination. MDIO1 takes precedence over MDIO0 if
3452          * they're both connected.
3453          */
3454         cp->phy_type = cas_get_vpd_info(cp, cp->dev->dev_addr,
3455                                         PCI_SLOT(pdev->devfn));
3456         if (cp->phy_type & CAS_PHY_SERDES) {
3457                 cp->cas_flags |= CAS_FLAG_1000MB_CAP;
3458                 return 0; /* no more checking needed */
3459         }
3460
3461         /* MII */
3462         cfg = readl(cp->regs + REG_MIF_CFG);
3463         if (cfg & MIF_CFG_MDIO_1) {
3464                 cp->phy_type = CAS_PHY_MII_MDIO1;
3465         } else if (cfg & MIF_CFG_MDIO_0) {
3466                 cp->phy_type = CAS_PHY_MII_MDIO0;
3467         }
3468
3469         cas_mif_poll(cp, 0);
3470         writel(PCS_DATAPATH_MODE_MII, cp->regs + REG_PCS_DATAPATH_MODE);
3471
3472         for (i = 0; i < 32; i++) {
3473                 u32 phy_id;
3474                 int j;
3475
3476                 for (j = 0; j < 3; j++) {
3477                         cp->phy_addr = i;
3478                         phy_id = cas_phy_read(cp, MII_PHYSID1) << 16;
3479                         phy_id |= cas_phy_read(cp, MII_PHYSID2);
3480                         if (phy_id && (phy_id != 0xFFFFFFFF)) {
3481                                 cp->phy_id = phy_id;
3482                                 goto done;
3483                         }
3484                 }
3485         }
3486         printk(KERN_ERR PFX "MII phy did not respond [%08x]\n",
3487                readl(cp->regs + REG_MIF_STATE_MACHINE));
3488         return -1;
3489
3490 done:
3491         /* see if we can do gigabit */
3492         cfg = cas_phy_read(cp, MII_BMSR);
3493         if ((cfg & CAS_BMSR_1000_EXTEND) &&
3494             cas_phy_read(cp, CAS_MII_1000_EXTEND))
3495                 cp->cas_flags |= CAS_FLAG_1000MB_CAP;
3496         return 0;
3497 }
3498
3499 /* Must be invoked under cp->lock. */
3500 static inline void cas_start_dma(struct cas *cp)
3501 {
3502         int i;
3503         u32 val;
3504         int txfailed = 0;
3505
3506         /* enable dma */
3507         val = readl(cp->regs + REG_TX_CFG) | TX_CFG_DMA_EN;
3508         writel(val, cp->regs + REG_TX_CFG);
3509         val = readl(cp->regs + REG_RX_CFG) | RX_CFG_DMA_EN;
3510         writel(val, cp->regs + REG_RX_CFG);
3511
3512         /* enable the mac */
3513         val = readl(cp->regs + REG_MAC_TX_CFG) | MAC_TX_CFG_EN;
3514         writel(val, cp->regs + REG_MAC_TX_CFG);
3515         val = readl(cp->regs + REG_MAC_RX_CFG) | MAC_RX_CFG_EN;
3516         writel(val, cp->regs + REG_MAC_RX_CFG);
3517
3518         i = STOP_TRIES;
3519         while (i-- > 0) {
3520                 val = readl(cp->regs + REG_MAC_TX_CFG);
3521                 if ((val & MAC_TX_CFG_EN))
3522                         break;
3523                 udelay(10);
3524         }
3525         if (i < 0) txfailed = 1;
3526         i = STOP_TRIES;
3527         while (i-- > 0) {
3528                 val = readl(cp->regs + REG_MAC_RX_CFG);
3529                 if ((val & MAC_RX_CFG_EN)) {
3530                         if (txfailed) {
3531                           printk(KERN_ERR
3532                                  "%s: enabling mac failed [tx:%08x:%08x].\n",
3533                                  cp->dev->name,
3534                                  readl(cp->regs + REG_MIF_STATE_MACHINE),
3535                                  readl(cp->regs + REG_MAC_STATE_MACHINE));
3536                         }
3537                         goto enable_rx_done;
3538                 }
3539                 udelay(10);
3540         }
3541         printk(KERN_ERR "%s: enabling mac failed [%s:%08x:%08x].\n",
3542                cp->dev->name,
3543                (txfailed? "tx,rx":"rx"),
3544                readl(cp->regs + REG_MIF_STATE_MACHINE),
3545                readl(cp->regs + REG_MAC_STATE_MACHINE));
3546
3547 enable_rx_done:
3548         cas_unmask_intr(cp); /* enable interrupts */
3549         writel(RX_DESC_RINGN_SIZE(0) - 4, cp->regs + REG_RX_KICK);
3550         writel(0, cp->regs + REG_RX_COMP_TAIL);
3551
3552         if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
3553                 if (N_RX_DESC_RINGS > 1)
3554                         writel(RX_DESC_RINGN_SIZE(1) - 4,
3555                                cp->regs + REG_PLUS_RX_KICK1);
3556
3557                 for (i = 1; i < N_RX_COMP_RINGS; i++)
3558                         writel(0, cp->regs + REG_PLUS_RX_COMPN_TAIL(i));
3559         }
3560 }
3561
3562 /* Must be invoked under cp->lock. */
3563 static void cas_read_pcs_link_mode(struct cas *cp, int *fd, int *spd,
3564                                    int *pause)
3565 {
3566         u32 val = readl(cp->regs + REG_PCS_MII_LPA);
3567         *fd     = (val & PCS_MII_LPA_FD) ? 1 : 0;
3568         *pause  = (val & PCS_MII_LPA_SYM_PAUSE) ? 0x01 : 0x00;
3569         if (val & PCS_MII_LPA_ASYM_PAUSE)
3570                 *pause |= 0x10;
3571         *spd = 1000;
3572 }
3573
3574 /* Must be invoked under cp->lock. */
3575 static void cas_read_mii_link_mode(struct cas *cp, int *fd, int *spd,
3576                                    int *pause)
3577 {
3578         u32 val;
3579
3580         *fd = 0;
3581         *spd = 10;
3582         *pause = 0;
3583
3584         /* use GMII registers */
3585         val = cas_phy_read(cp, MII_LPA);
3586         if (val & CAS_LPA_PAUSE)
3587                 *pause = 0x01;
3588
3589         if (val & CAS_LPA_ASYM_PAUSE)
3590                 *pause |= 0x10;
3591
3592         if (val & LPA_DUPLEX)
3593                 *fd = 1;
3594         if (val & LPA_100)
3595                 *spd = 100;
3596
3597         if (cp->cas_flags & CAS_FLAG_1000MB_CAP) {
3598                 val = cas_phy_read(cp, CAS_MII_1000_STATUS);
3599                 if (val & (CAS_LPA_1000FULL | CAS_LPA_1000HALF))
3600                         *spd = 1000;
3601                 if (val & CAS_LPA_1000FULL)
3602                         *fd = 1;
3603         }
3604 }
3605
3606 /* A link-up condition has occurred, initialize and enable the
3607  * rest of the chip.
3608  *
3609  * Must be invoked under cp->lock.
3610  */
3611 static void cas_set_link_modes(struct cas *cp)
3612 {
3613         u32 val;
3614         int full_duplex, speed, pause;
3615
3616         full_duplex = 0;
3617         speed = 10;
3618         pause = 0;
3619
3620         if (CAS_PHY_MII(cp->phy_type)) {
3621                 cas_mif_poll(cp, 0);
3622                 val = cas_phy_read(cp, MII_BMCR);
3623                 if (val & BMCR_ANENABLE) {
3624                         cas_read_mii_link_mode(cp, &full_duplex, &speed,
3625                                                &pause);
3626                 } else {
3627                         if (val & BMCR_FULLDPLX)
3628                                 full_duplex = 1;
3629
3630                         if (val & BMCR_SPEED100)
3631                                 speed = 100;
3632                         else if (val & CAS_BMCR_SPEED1000)
3633                                 speed = (cp->cas_flags & CAS_FLAG_1000MB_CAP) ?
3634                                         1000 : 100;
3635                 }
3636                 cas_mif_poll(cp, 1);
3637
3638         } else {
3639                 val = readl(cp->regs + REG_PCS_MII_CTRL);
3640                 cas_read_pcs_link_mode(cp, &full_duplex, &speed, &pause);
3641                 if ((val & PCS_MII_AUTONEG_EN) == 0) {
3642                         if (val & PCS_MII_CTRL_DUPLEX)
3643                                 full_duplex = 1;
3644                 }
3645         }
3646
3647         if (netif_msg_link(cp))
3648                 printk(KERN_INFO "%s: Link up at %d Mbps, %s-duplex.\n",
3649                        cp->dev->name, speed, (full_duplex ? "full" : "half"));
3650
3651         val = MAC_XIF_TX_MII_OUTPUT_EN | MAC_XIF_LINK_LED;
3652         if (CAS_PHY_MII(cp->phy_type)) {
3653                 val |= MAC_XIF_MII_BUFFER_OUTPUT_EN;
3654                 if (!full_duplex)
3655                         val |= MAC_XIF_DISABLE_ECHO;
3656         }
3657         if (full_duplex)
3658                 val |= MAC_XIF_FDPLX_LED;
3659         if (speed == 1000)
3660                 val |= MAC_XIF_GMII_MODE;
3661         writel(val, cp->regs + REG_MAC_XIF_CFG);
3662
3663         /* deal with carrier and collision detect. */
3664         val = MAC_TX_CFG_IPG_EN;
3665         if (full_duplex) {
3666                 val |= MAC_TX_CFG_IGNORE_CARRIER;
3667                 val |= MAC_TX_CFG_IGNORE_COLL;
3668         } else {
3669 #ifndef USE_CSMA_CD_PROTO
3670                 val |= MAC_TX_CFG_NEVER_GIVE_UP_EN;
3671                 val |= MAC_TX_CFG_NEVER_GIVE_UP_LIM;
3672 #endif
3673         }
3674         /* val now set up for REG_MAC_TX_CFG */
3675
3676         /* If gigabit and half-duplex, enable carrier extension
3677          * mode.  increase slot time to 512 bytes as well.
3678          * else, disable it and make sure slot time is 64 bytes.
3679          * also activate checksum bug workaround
3680          */
3681         if ((speed == 1000) && !full_duplex) {
3682                 writel(val | MAC_TX_CFG_CARRIER_EXTEND,
3683                        cp->regs + REG_MAC_TX_CFG);
3684
3685                 val = readl(cp->regs + REG_MAC_RX_CFG);
3686                 val &= ~MAC_RX_CFG_STRIP_FCS; /* checksum workaround */
3687                 writel(val | MAC_RX_CFG_CARRIER_EXTEND,
3688                        cp->regs + REG_MAC_RX_CFG);
3689
3690                 writel(0x200, cp->regs + REG_MAC_SLOT_TIME);
3691
3692                 cp->crc_size = 4;
3693                 /* minimum size gigabit frame at half duplex */
3694                 cp->min_frame_size = CAS_1000MB_MIN_FRAME;
3695
3696         } else {
3697                 writel(val, cp->regs + REG_MAC_TX_CFG);
3698
3699                 /* checksum bug workaround. don't strip FCS when in
3700                  * half-duplex mode
3701                  */
3702                 val = readl(cp->regs + REG_MAC_RX_CFG);
3703                 if (full_duplex) {
3704                         val |= MAC_RX_CFG_STRIP_FCS;
3705                         cp->crc_size = 0;
3706                         cp->min_frame_size = CAS_MIN_MTU;
3707                 } else {
3708                         val &= ~MAC_RX_CFG_STRIP_FCS;
3709                         cp->crc_size = 4;
3710                         cp->min_frame_size = CAS_MIN_FRAME;
3711                 }
3712                 writel(val & ~MAC_RX_CFG_CARRIER_EXTEND,
3713                        cp->regs + REG_MAC_RX_CFG);
3714                 writel(0x40, cp->regs + REG_MAC_SLOT_TIME);
3715         }
3716
3717         if (netif_msg_link(cp)) {
3718                 if (pause & 0x01) {
3719                         printk(KERN_INFO "%s: Pause is enabled "
3720                                "(rxfifo: %d off: %d on: %d)\n",
3721                                cp->dev->name,
3722                                cp->rx_fifo_size,
3723                                cp->rx_pause_off,
3724                                cp->rx_pause_on);
3725                 } else if (pause & 0x10) {
3726                         printk(KERN_INFO "%s: TX pause enabled\n",
3727                                cp->dev->name);
3728                 } else {
3729                         printk(KERN_INFO "%s: Pause is disabled\n",
3730                                cp->dev->name);
3731                 }
3732         }
3733
3734         val = readl(cp->regs + REG_MAC_CTRL_CFG);
3735         val &= ~(MAC_CTRL_CFG_SEND_PAUSE_EN | MAC_CTRL_CFG_RECV_PAUSE_EN);
3736         if (pause) { /* symmetric or asymmetric pause */
3737                 val |= MAC_CTRL_CFG_SEND_PAUSE_EN;
3738                 if (pause & 0x01) { /* symmetric pause */
3739                         val |= MAC_CTRL_CFG_RECV_PAUSE_EN;
3740                 }
3741         }
3742         writel(val, cp->regs + REG_MAC_CTRL_CFG);
3743         cas_start_dma(cp);
3744 }
3745
3746 /* Must be invoked under cp->lock. */
3747 static void cas_init_hw(struct cas *cp, int restart_link)
3748 {
3749         if (restart_link)
3750                 cas_phy_init(cp);
3751
3752         cas_init_pause_thresholds(cp);
3753         cas_init_mac(cp);
3754         cas_init_dma(cp);
3755
3756         if (restart_link) {
3757                 /* Default aneg parameters */
3758                 cp->timer_ticks = 0;
3759                 cas_begin_auto_negotiation(cp, NULL);
3760         } else if (cp->lstate == link_up) {
3761                 cas_set_link_modes(cp);
3762                 netif_carrier_on(cp->dev);
3763         }
3764 }
3765
3766 /* Must be invoked under cp->lock. on earlier cassini boards,
3767  * SOFT_0 is tied to PCI reset. we use this to force a pci reset,
3768  * let it settle out, and then restore pci state.
3769  */
3770 static void cas_hard_reset(struct cas *cp)
3771 {
3772         writel(BIM_LOCAL_DEV_SOFT_0, cp->regs + REG_BIM_LOCAL_DEV_EN);
3773         udelay(20);
3774         pci_restore_state(cp->pdev);
3775 }
3776
3777
3778 static void cas_global_reset(struct cas *cp, int blkflag)
3779 {
3780         int limit;
3781
3782         /* issue a global reset. don't use RSTOUT. */
3783         if (blkflag && !CAS_PHY_MII(cp->phy_type)) {
3784                 /* For PCS, when the blkflag is set, we should set the
3785                  * SW_REST_BLOCK_PCS_SLINK bit to prevent the results of
3786                  * the last autonegotiation from being cleared.  We'll
3787                  * need some special handling if the chip is set into a
3788                  * loopback mode.
3789                  */
3790                 writel((SW_RESET_TX | SW_RESET_RX | SW_RESET_BLOCK_PCS_SLINK),
3791                        cp->regs + REG_SW_RESET);
3792         } else {
3793                 writel(SW_RESET_TX | SW_RESET_RX, cp->regs + REG_SW_RESET);
3794         }
3795
3796         /* need to wait at least 3ms before polling register */
3797         mdelay(3);
3798
3799         limit = STOP_TRIES;
3800         while (limit-- > 0) {
3801                 u32 val = readl(cp->regs + REG_SW_RESET);
3802                 if ((val & (SW_RESET_TX | SW_RESET_RX)) == 0)
3803                         goto done;
3804                 udelay(10);
3805         }
3806         printk(KERN_ERR "%s: sw reset failed.\n", cp->dev->name);
3807
3808 done:
3809         /* enable various BIM interrupts */
3810         writel(BIM_CFG_DPAR_INTR_ENABLE | BIM_CFG_RMA_INTR_ENABLE |
3811                BIM_CFG_RTA_INTR_ENABLE, cp->regs + REG_BIM_CFG);
3812
3813         /* clear out pci error status mask for handled errors.
3814          * we don't deal with DMA counter overflows as they happen
3815          * all the time.
3816          */
3817         writel(0xFFFFFFFFU & ~(PCI_ERR_BADACK | PCI_ERR_DTRTO |
3818                                PCI_ERR_OTHER | PCI_ERR_BIM_DMA_WRITE |
3819                                PCI_ERR_BIM_DMA_READ), cp->regs +
3820                REG_PCI_ERR_STATUS_MASK);
3821
3822         /* set up for MII by default to address mac rx reset timeout
3823          * issue
3824          */
3825         writel(PCS_DATAPATH_MODE_MII, cp->regs + REG_PCS_DATAPATH_MODE);
3826 }
3827
3828 static void cas_reset(struct cas *cp, int blkflag)
3829 {
3830         u32 val;
3831
3832         cas_mask_intr(cp);
3833         cas_global_reset(cp, blkflag);
3834         cas_mac_reset(cp);
3835         cas_entropy_reset(cp);
3836
3837         /* disable dma engines. */
3838         val = readl(cp->regs + REG_TX_CFG);
3839         val &= ~TX_CFG_DMA_EN;
3840         writel(val, cp->regs + REG_TX_CFG);
3841
3842         val = readl(cp->regs + REG_RX_CFG);
3843         val &= ~RX_CFG_DMA_EN;
3844         writel(val, cp->regs + REG_RX_CFG);
3845
3846         /* program header parser */
3847         if ((cp->cas_flags & CAS_FLAG_TARGET_ABORT) ||
3848             (CAS_HP_ALT_FIRMWARE == cas_prog_null)) {
3849                 cas_load_firmware(cp, CAS_HP_FIRMWARE);
3850         } else {
3851                 cas_load_firmware(cp, CAS_HP_ALT_FIRMWARE);
3852         }
3853
3854         /* clear out error registers */
3855         spin_lock(&cp->stat_lock[N_TX_RINGS]);
3856         cas_clear_mac_err(cp);
3857         spin_unlock(&cp->stat_lock[N_TX_RINGS]);
3858 }
3859
3860 /* Shut down the chip, must be called with pm_mutex held.  */
3861 static void cas_shutdown(struct cas *cp)
3862 {
3863         unsigned long flags;
3864
3865         /* Make us not-running to avoid timers respawning */
3866         cp->hw_running = 0;
3867
3868         del_timer_sync(&cp->link_timer);
3869
3870         /* Stop the reset task */
3871 #if 0
3872         while (atomic_read(&cp->reset_task_pending_mtu) ||
3873                atomic_read(&cp->reset_task_pending_spare) ||
3874                atomic_read(&cp->reset_task_pending_all))
3875                 schedule();
3876
3877 #else
3878         while (atomic_read(&cp->reset_task_pending))
3879                 schedule();
3880 #endif
3881         /* Actually stop the chip */
3882         cas_lock_all_save(cp, flags);
3883         cas_reset(cp, 0);
3884         if (cp->cas_flags & CAS_FLAG_SATURN)
3885                 cas_phy_powerdown(cp);
3886         cas_unlock_all_restore(cp, flags);
3887 }
3888
3889 static int cas_change_mtu(struct net_device *dev, int new_mtu)
3890 {
3891         struct cas *cp = netdev_priv(dev);
3892
3893         if (new_mtu < CAS_MIN_MTU || new_mtu > CAS_MAX_MTU)
3894                 return -EINVAL;
3895
3896         dev->mtu = new_mtu;
3897         if (!netif_running(dev) || !netif_device_present(dev))
3898                 return 0;
3899
3900         /* let the reset task handle it */
3901 #if 1
3902         atomic_inc(&cp->reset_task_pending);
3903         if ((cp->phy_type & CAS_PHY_SERDES)) {
3904                 atomic_inc(&cp->reset_task_pending_all);
3905         } else {
3906                 atomic_inc(&cp->reset_task_pending_mtu);
3907         }
3908         schedule_work(&cp->reset_task);
3909 #else
3910         atomic_set(&cp->reset_task_pending, (cp->phy_type & CAS_PHY_SERDES) ?
3911                    CAS_RESET_ALL : CAS_RESET_MTU);
3912         printk(KERN_ERR "reset called in cas_change_mtu\n");
3913         schedule_work(&cp->reset_task);
3914 #endif
3915
3916         flush_scheduled_work();
3917         return 0;
3918 }
3919
3920 static void cas_clean_txd(struct cas *cp, int ring)
3921 {
3922         struct cas_tx_desc *txd = cp->init_txds[ring];
3923         struct sk_buff *skb, **skbs = cp->tx_skbs[ring];
3924         u64 daddr, dlen;
3925         int i, size;
3926
3927         size = TX_DESC_RINGN_SIZE(ring);
3928         for (i = 0; i < size; i++) {
3929                 int frag;
3930
3931                 if (skbs[i] == NULL)
3932                         continue;
3933
3934                 skb = skbs[i];
3935                 skbs[i] = NULL;
3936
3937                 for (frag = 0; frag <= skb_shinfo(skb)->nr_frags;  frag++) {
3938                         int ent = i & (size - 1);
3939
3940                         /* first buffer is never a tiny buffer and so
3941                          * needs to be unmapped.
3942                          */
3943                         daddr = le64_to_cpu(txd[ent].buffer);
3944                         dlen  =  CAS_VAL(TX_DESC_BUFLEN,
3945                                          le64_to_cpu(txd[ent].control));
3946                         pci_unmap_page(cp->pdev, daddr, dlen,
3947                                        PCI_DMA_TODEVICE);
3948
3949                         if (frag != skb_shinfo(skb)->nr_frags) {
3950                                 i++;
3951
3952                                 /* next buffer might by a tiny buffer.
3953                                  * skip past it.
3954                                  */
3955                                 ent = i & (size - 1);
3956                                 if (cp->tx_tiny_use[ring][ent].used)
3957                                         i++;
3958                         }
3959                 }
3960                 dev_kfree_skb_any(skb);
3961         }
3962
3963         /* zero out tiny buf usage */
3964         memset(cp->tx_tiny_use[ring], 0, size*sizeof(*cp->tx_tiny_use[ring]));
3965 }
3966
3967 /* freed on close */
3968 static inline void cas_free_rx_desc(struct cas *cp, int ring)
3969 {
3970         cas_page_t **page = cp->rx_pages[ring];
3971         int i, size;
3972
3973         size = RX_DESC_RINGN_SIZE(ring);
3974         for (i = 0; i < size; i++) {
3975                 if (page[i]) {
3976                         cas_page_free(cp, page[i]);
3977                         page[i] = NULL;
3978                 }
3979         }
3980 }
3981
3982 static void cas_free_rxds(struct cas *cp)
3983 {
3984         int i;
3985
3986         for (i = 0; i < N_RX_DESC_RINGS; i++)
3987                 cas_free_rx_desc(cp, i);
3988 }
3989
3990 /* Must be invoked under cp->lock. */
3991 static void cas_clean_rings(struct cas *cp)
3992 {
3993         int i;
3994
3995         /* need to clean all tx rings */
3996         memset(cp->tx_old, 0, sizeof(*cp->tx_old)*N_TX_RINGS);
3997         memset(cp->tx_new, 0, sizeof(*cp->tx_new)*N_TX_RINGS);
3998         for (i = 0; i < N_TX_RINGS; i++)
3999                 cas_clean_txd(cp, i);
4000
4001         /* zero out init block */
4002         memset(cp->init_block, 0, sizeof(struct cas_init_block));
4003         cas_clean_rxds(cp);
4004         cas_clean_rxcs(cp);
4005 }
4006
4007 /* allocated on open */
4008 static inline int cas_alloc_rx_desc(struct cas *cp, int ring)
4009 {
4010         cas_page_t **page = cp->rx_pages[ring];
4011         int size, i = 0;
4012
4013         size = RX_DESC_RINGN_SIZE(ring);
4014         for (i = 0; i < size; i++) {
4015                 if ((page[i] = cas_page_alloc(cp, GFP_KERNEL)) == NULL)
4016                         return -1;
4017         }
4018         return 0;
4019 }
4020
4021 static int cas_alloc_rxds(struct cas *cp)
4022 {
4023         int i;
4024
4025         for (i = 0; i < N_RX_DESC_RINGS; i++) {
4026                 if (cas_alloc_rx_desc(cp, i) < 0) {
4027                         cas_free_rxds(cp);
4028                         return -1;
4029                 }
4030         }
4031         return 0;
4032 }
4033
4034 static void cas_reset_task(struct work_struct *work)
4035 {
4036         struct cas *cp = container_of(work, struct cas, reset_task);
4037 #if 0
4038         int pending = atomic_read(&cp->reset_task_pending);
4039 #else
4040         int pending_all = atomic_read(&cp->reset_task_pending_all);
4041         int pending_spare = atomic_read(&cp->reset_task_pending_spare);
4042         int pending_mtu = atomic_read(&cp->reset_task_pending_mtu);
4043
4044         if (pending_all == 0 && pending_spare == 0 && pending_mtu == 0) {
4045                 /* We can have more tasks scheduled than actually
4046                  * needed.
4047                  */
4048                 atomic_dec(&cp->reset_task_pending);
4049                 return;
4050         }
4051 #endif
4052         /* The link went down, we reset the ring, but keep
4053          * DMA stopped. Use this function for reset
4054          * on error as well.
4055          */
4056         if (cp->hw_running) {
4057                 unsigned long flags;
4058
4059                 /* Make sure we don't get interrupts or tx packets */
4060                 netif_device_detach(cp->dev);
4061                 cas_lock_all_save(cp, flags);
4062
4063                 if (cp->opened) {
4064                         /* We call cas_spare_recover when we call cas_open.
4065                          * but we do not initialize the lists cas_spare_recover
4066                          * uses until cas_open is called.
4067                          */
4068                         cas_spare_recover(cp, GFP_ATOMIC);
4069                 }
4070 #if 1
4071                 /* test => only pending_spare set */
4072                 if (!pending_all && !pending_mtu)
4073                         goto done;
4074 #else
4075                 if (pending == CAS_RESET_SPARE)
4076                         goto done;
4077 #endif
4078                 /* when pending == CAS_RESET_ALL, the following
4079                  * call to cas_init_hw will restart auto negotiation.
4080                  * Setting the second argument of cas_reset to
4081                  * !(pending == CAS_RESET_ALL) will set this argument
4082                  * to 1 (avoiding reinitializing the PHY for the normal
4083                  * PCS case) when auto negotiation is not restarted.
4084                  */
4085 #if 1
4086                 cas_reset(cp, !(pending_all > 0));
4087                 if (cp->opened)
4088                         cas_clean_rings(cp);
4089                 cas_init_hw(cp, (pending_all > 0));
4090 #else
4091                 cas_reset(cp, !(pending == CAS_RESET_ALL));
4092                 if (cp->opened)
4093                         cas_clean_rings(cp);
4094                 cas_init_hw(cp, pending == CAS_RESET_ALL);
4095 #endif
4096
4097 done:
4098                 cas_unlock_all_restore(cp, flags);
4099                 netif_device_attach(cp->dev);
4100         }
4101 #if 1
4102         atomic_sub(pending_all, &cp->reset_task_pending_all);
4103         atomic_sub(pending_spare, &cp->reset_task_pending_spare);
4104         atomic_sub(pending_mtu, &cp->reset_task_pending_mtu);
4105         atomic_dec(&cp->reset_task_pending);
4106 #else
4107         atomic_set(&cp->reset_task_pending, 0);
4108 #endif
4109 }
4110
4111 static void cas_link_timer(unsigned long data)
4112 {
4113         struct cas *cp = (struct cas *) data;
4114         int mask, pending = 0, reset = 0;
4115         unsigned long flags;
4116
4117         if (link_transition_timeout != 0 &&
4118             cp->link_transition_jiffies_valid &&
4119             ((jiffies - cp->link_transition_jiffies) >
4120               (link_transition_timeout))) {
4121                 /* One-second counter so link-down workaround doesn't
4122                  * cause resets to occur so fast as to fool the switch
4123                  * into thinking the link is down.
4124                  */
4125                 cp->link_transition_jiffies_valid = 0;
4126         }
4127
4128         if (!cp->hw_running)
4129                 return;
4130
4131         spin_lock_irqsave(&cp->lock, flags);
4132         cas_lock_tx(cp);
4133         cas_entropy_gather(cp);
4134
4135         /* If the link task is still pending, we just
4136          * reschedule the link timer
4137          */
4138 #if 1
4139         if (atomic_read(&cp->reset_task_pending_all) ||
4140             atomic_read(&cp->reset_task_pending_spare) ||
4141             atomic_read(&cp->reset_task_pending_mtu))
4142                 goto done;
4143 #else
4144         if (atomic_read(&cp->reset_task_pending))
4145                 goto done;
4146 #endif
4147
4148         /* check for rx cleaning */
4149         if ((mask = (cp->cas_flags & CAS_FLAG_RXD_POST_MASK))) {
4150                 int i, rmask;
4151
4152                 for (i = 0; i < MAX_RX_DESC_RINGS; i++) {
4153                         rmask = CAS_FLAG_RXD_POST(i);
4154                         if ((mask & rmask) == 0)
4155                                 continue;
4156
4157                         /* post_rxds will do a mod_timer */
4158                         if (cas_post_rxds_ringN(cp, i, cp->rx_last[i]) < 0) {
4159                                 pending = 1;
4160                                 continue;
4161                         }
4162                         cp->cas_flags &= ~rmask;
4163                 }
4164         }
4165
4166         if (CAS_PHY_MII(cp->phy_type)) {
4167                 u16 bmsr;
4168                 cas_mif_poll(cp, 0);
4169                 bmsr = cas_phy_read(cp, MII_BMSR);
4170                 /* WTZ: Solaris driver reads this twice, but that
4171                  * may be due to the PCS case and the use of a
4172                  * common implementation. Read it twice here to be
4173                  * safe.
4174                  */
4175                 bmsr = cas_phy_read(cp, MII_BMSR);
4176                 cas_mif_poll(cp, 1);
4177                 readl(cp->regs + REG_MIF_STATUS); /* avoid dups */
4178                 reset = cas_mii_link_check(cp, bmsr);
4179         } else {
4180                 reset = cas_pcs_link_check(cp);
4181         }
4182
4183         if (reset)
4184                 goto done;
4185
4186         /* check for tx state machine confusion */
4187         if ((readl(cp->regs + REG_MAC_TX_STATUS) & MAC_TX_FRAME_XMIT) == 0) {
4188                 u32 val = readl(cp->regs + REG_MAC_STATE_MACHINE);
4189                 u32 wptr, rptr;
4190                 int tlm  = CAS_VAL(MAC_SM_TLM, val);
4191
4192                 if (((tlm == 0x5) || (tlm == 0x3)) &&
4193                     (CAS_VAL(MAC_SM_ENCAP_SM, val) == 0)) {
4194                         if (netif_msg_tx_err(cp))
4195                                 printk(KERN_DEBUG "%s: tx err: "
4196                                        "MAC_STATE[%08x]\n",
4197                                        cp->dev->name, val);
4198                         reset = 1;
4199                         goto done;
4200                 }
4201
4202                 val  = readl(cp->regs + REG_TX_FIFO_PKT_CNT);
4203                 wptr = readl(cp->regs + REG_TX_FIFO_WRITE_PTR);
4204                 rptr = readl(cp->regs + REG_TX_FIFO_READ_PTR);
4205                 if ((val == 0) && (wptr != rptr)) {
4206                         if (netif_msg_tx_err(cp))
4207                                 printk(KERN_DEBUG "%s: tx err: "
4208                                        "TX_FIFO[%08x:%08x:%08x]\n",
4209                                        cp->dev->name, val, wptr, rptr);
4210                         reset = 1;
4211                 }
4212
4213                 if (reset)
4214                         cas_hard_reset(cp);
4215         }
4216
4217 done:
4218         if (reset) {
4219 #if 1
4220                 atomic_inc(&cp->reset_task_pending);
4221                 atomic_inc(&cp->reset_task_pending_all);
4222                 schedule_work(&cp->reset_task);
4223 #else
4224                 atomic_set(&cp->reset_task_pending, CAS_RESET_ALL);
4225                 printk(KERN_ERR "reset called in cas_link_timer\n");
4226                 schedule_work(&cp->reset_task);
4227 #endif
4228         }
4229
4230         if (!pending)
4231                 mod_timer(&cp->link_timer, jiffies + CAS_LINK_TIMEOUT);
4232         cas_unlock_tx(cp);
4233         spin_unlock_irqrestore(&cp->lock, flags);
4234 }
4235
4236 /* tiny buffers are used to avoid target abort issues with
4237  * older cassini's
4238  */
4239 static void cas_tx_tiny_free(struct cas *cp)
4240 {
4241         struct pci_dev *pdev = cp->pdev;
4242         int i;
4243
4244         for (i = 0; i < N_TX_RINGS; i++) {
4245                 if (!cp->tx_tiny_bufs[i])
4246                         continue;
4247
4248                 pci_free_consistent(pdev, TX_TINY_BUF_BLOCK,
4249                                     cp->tx_tiny_bufs[i],
4250                                     cp->tx_tiny_dvma[i]);
4251                 cp->tx_tiny_bufs[i] = NULL;
4252         }
4253 }
4254
4255 static int cas_tx_tiny_alloc(struct cas *cp)
4256 {
4257         struct pci_dev *pdev = cp->pdev;
4258         int i;
4259
4260         for (i = 0; i < N_TX_RINGS; i++) {
4261                 cp->tx_tiny_bufs[i] =
4262                         pci_alloc_consistent(pdev, TX_TINY_BUF_BLOCK,
4263                                              &cp->tx_tiny_dvma[i]);
4264                 if (!cp->tx_tiny_bufs[i]) {
4265                         cas_tx_tiny_free(cp);
4266                         return -1;
4267                 }
4268         }
4269         return 0;
4270 }
4271
4272
4273 static int cas_open(struct net_device *dev)
4274 {
4275         struct cas *cp = netdev_priv(dev);
4276         int hw_was_up, err;
4277         unsigned long flags;
4278
4279         mutex_lock(&cp->pm_mutex);
4280
4281         hw_was_up = cp->hw_running;
4282
4283         /* The power-management mutex protects the hw_running
4284          * etc. state so it is safe to do this bit without cp->lock
4285          */
4286         if (!cp->hw_running) {
4287                 /* Reset the chip */
4288                 cas_lock_all_save(cp, flags);
4289                 /* We set the second arg to cas_reset to zero
4290                  * because cas_init_hw below will have its second
4291                  * argument set to non-zero, which will force
4292                  * autonegotiation to start.
4293                  */
4294                 cas_reset(cp, 0);
4295                 cp->hw_running = 1;
4296                 cas_unlock_all_restore(cp, flags);
4297         }
4298
4299         if (cas_tx_tiny_alloc(cp) < 0)
4300                 return -ENOMEM;
4301
4302         /* alloc rx descriptors */
4303         err = -ENOMEM;
4304         if (cas_alloc_rxds(cp) < 0)
4305                 goto err_tx_tiny;
4306
4307         /* allocate spares */
4308         cas_spare_init(cp);
4309         cas_spare_recover(cp, GFP_KERNEL);
4310
4311         /* We can now request the interrupt as we know it's masked
4312          * on the controller. cassini+ has up to 4 interrupts
4313          * that can be used, but you need to do explicit pci interrupt
4314          * mapping to expose them
4315          */
4316         if (request_irq(cp->pdev->irq, cas_interrupt,
4317                         IRQF_SHARED, dev->name, (void *) dev)) {
4318                 printk(KERN_ERR "%s: failed to request irq !\n",
4319                        cp->dev->name);
4320                 err = -EAGAIN;
4321                 goto err_spare;
4322         }
4323
4324 #ifdef USE_NAPI
4325         napi_enable(&cp->napi);
4326 #endif
4327         /* init hw */
4328         cas_lock_all_save(cp, flags);
4329         cas_clean_rings(cp);
4330         cas_init_hw(cp, !hw_was_up);
4331         cp->opened = 1;
4332         cas_unlock_all_restore(cp, flags);
4333
4334         netif_start_queue(dev);
4335         mutex_unlock(&cp->pm_mutex);
4336         return 0;
4337
4338 err_spare:
4339         cas_spare_free(cp);
4340         cas_free_rxds(cp);
4341 err_tx_tiny:
4342         cas_tx_tiny_free(cp);
4343         mutex_unlock(&cp->pm_mutex);
4344         return err;
4345 }
4346
4347 static int cas_close(struct net_device *dev)
4348 {
4349         unsigned long flags;
4350         struct cas *cp = netdev_priv(dev);
4351
4352 #ifdef USE_NAPI
4353         napi_disable(&cp->napi);
4354 #endif
4355         /* Make sure we don't get distracted by suspend/resume */
4356         mutex_lock(&cp->pm_mutex);
4357
4358         netif_stop_queue(dev);
4359
4360         /* Stop traffic, mark us closed */
4361         cas_lock_all_save(cp, flags);
4362         cp->opened = 0;
4363         cas_reset(cp, 0);
4364         cas_phy_init(cp);
4365         cas_begin_auto_negotiation(cp, NULL);
4366         cas_clean_rings(cp);
4367         cas_unlock_all_restore(cp, flags);
4368
4369         free_irq(cp->pdev->irq, (void *) dev);
4370         cas_spare_free(cp);
4371         cas_free_rxds(cp);
4372         cas_tx_tiny_free(cp);
4373         mutex_unlock(&cp->pm_mutex);
4374         return 0;
4375 }
4376
4377 static struct {
4378         const char name[ETH_GSTRING_LEN];
4379 } ethtool_cassini_statnames[] = {
4380         {"collisions"},
4381         {"rx_bytes"},
4382         {"rx_crc_errors"},
4383         {"rx_dropped"},
4384         {"rx_errors"},
4385         {"rx_fifo_errors"},
4386         {"rx_frame_errors"},
4387         {"rx_length_errors"},
4388         {"rx_over_errors"},
4389         {"rx_packets"},
4390         {"tx_aborted_errors"},
4391         {"tx_bytes"},
4392         {"tx_dropped"},
4393         {"tx_errors"},
4394         {"tx_fifo_errors"},
4395         {"tx_packets"}
4396 };
4397 #define CAS_NUM_STAT_KEYS ARRAY_SIZE(ethtool_cassini_statnames)
4398
4399 static struct {
4400         const int offsets;      /* neg. values for 2nd arg to cas_read_phy */
4401 } ethtool_register_table[] = {
4402         {-MII_BMSR},
4403         {-MII_BMCR},
4404         {REG_CAWR},
4405         {REG_INF_BURST},
4406         {REG_BIM_CFG},
4407         {REG_RX_CFG},
4408         {REG_HP_CFG},
4409         {REG_MAC_TX_CFG},
4410         {REG_MAC_RX_CFG},
4411         {REG_MAC_CTRL_CFG},
4412         {REG_MAC_XIF_CFG},
4413         {REG_MIF_CFG},
4414         {REG_PCS_CFG},
4415         {REG_SATURN_PCFG},
4416         {REG_PCS_MII_STATUS},
4417         {REG_PCS_STATE_MACHINE},
4418         {REG_MAC_COLL_EXCESS},
4419         {REG_MAC_COLL_LATE}
4420 };
4421 #define CAS_REG_LEN     ARRAY_SIZE(ethtool_register_table)
4422 #define CAS_MAX_REGS    (sizeof (u32)*CAS_REG_LEN)
4423
4424 static void cas_read_regs(struct cas *cp, u8 *ptr, int len)
4425 {
4426         u8 *p;
4427         int i;
4428         unsigned long flags;
4429
4430         spin_lock_irqsave(&cp->lock, flags);
4431         for (i = 0, p = ptr; i < len ; i ++, p += sizeof(u32)) {
4432                 u16 hval;
4433                 u32 val;
4434                 if (ethtool_register_table[i].offsets < 0) {
4435                         hval = cas_phy_read(cp,
4436                                     -ethtool_register_table[i].offsets);
4437                         val = hval;
4438                 } else {
4439                         val= readl(cp->regs+ethtool_register_table[i].offsets);
4440                 }
4441                 memcpy(p, (u8 *)&val, sizeof(u32));
4442         }
4443         spin_unlock_irqrestore(&cp->lock, flags);
4444 }
4445
4446 static struct net_device_stats *cas_get_stats(struct net_device *dev)
4447 {
4448         struct cas *cp = netdev_priv(dev);
4449         struct net_device_stats *stats = cp->net_stats;
4450         unsigned long flags;
4451         int i;
4452         unsigned long tmp;
4453
4454         /* we collate all of the stats into net_stats[N_TX_RING] */
4455         if (!cp->hw_running)
4456                 return stats + N_TX_RINGS;
4457
4458         /* collect outstanding stats */
4459         /* WTZ: the Cassini spec gives these as 16 bit counters but
4460          * stored in 32-bit words.  Added a mask of 0xffff to be safe,
4461          * in case the chip somehow puts any garbage in the other bits.
4462          * Also, counter usage didn't seem to mach what Adrian did
4463          * in the parts of the code that set these quantities. Made
4464          * that consistent.
4465          */
4466         spin_lock_irqsave(&cp->stat_lock[N_TX_RINGS], flags);
4467         stats[N_TX_RINGS].rx_crc_errors +=
4468           readl(cp->regs + REG_MAC_FCS_ERR) & 0xffff;
4469         stats[N_TX_RINGS].rx_frame_errors +=
4470                 readl(cp->regs + REG_MAC_ALIGN_ERR) &0xffff;
4471         stats[N_TX_RINGS].rx_length_errors +=
4472                 readl(cp->regs + REG_MAC_LEN_ERR) & 0xffff;
4473 #if 1
4474         tmp = (readl(cp->regs + REG_MAC_COLL_EXCESS) & 0xffff) +
4475                 (readl(cp->regs + REG_MAC_COLL_LATE) & 0xffff);
4476         stats[N_TX_RINGS].tx_aborted_errors += tmp;
4477         stats[N_TX_RINGS].collisions +=
4478           tmp + (readl(cp->regs + REG_MAC_COLL_NORMAL) & 0xffff);
4479 #else
4480         stats[N_TX_RINGS].tx_aborted_errors +=
4481                 readl(cp->regs + REG_MAC_COLL_EXCESS);
4482         stats[N_TX_RINGS].collisions += readl(cp->regs + REG_MAC_COLL_EXCESS) +
4483                 readl(cp->regs + REG_MAC_COLL_LATE);
4484 #endif
4485         cas_clear_mac_err(cp);
4486
4487         /* saved bits that are unique to ring 0 */
4488         spin_lock(&cp->stat_lock[0]);
4489         stats[N_TX_RINGS].collisions        += stats[0].collisions;
4490         stats[N_TX_RINGS].rx_over_errors    += stats[0].rx_over_errors;
4491         stats[N_TX_RINGS].rx_frame_errors   += stats[0].rx_frame_errors;
4492         stats[N_TX_RINGS].rx_fifo_errors    += stats[0].rx_fifo_errors;
4493         stats[N_TX_RINGS].tx_aborted_errors += stats[0].tx_aborted_errors;
4494         stats[N_TX_RINGS].tx_fifo_errors    += stats[0].tx_fifo_errors;
4495         spin_unlock(&cp->stat_lock[0]);
4496
4497         for (i = 0; i < N_TX_RINGS; i++) {
4498                 spin_lock(&cp->stat_lock[i]);
4499                 stats[N_TX_RINGS].rx_length_errors +=
4500                         stats[i].rx_length_errors;
4501                 stats[N_TX_RINGS].rx_crc_errors += stats[i].rx_crc_errors;
4502                 stats[N_TX_RINGS].rx_packets    += stats[i].rx_packets;
4503                 stats[N_TX_RINGS].tx_packets    += stats[i].tx_packets;
4504                 stats[N_TX_RINGS].rx_bytes      += stats[i].rx_bytes;
4505                 stats[N_TX_RINGS].tx_bytes      += stats[i].tx_bytes;
4506                 stats[N_TX_RINGS].rx_errors     += stats[i].rx_errors;
4507                 stats[N_TX_RINGS].tx_errors     += stats[i].tx_errors;
4508                 stats[N_TX_RINGS].rx_dropped    += stats[i].rx_dropped;
4509                 stats[N_TX_RINGS].tx_dropped    += stats[i].tx_dropped;
4510                 memset(stats + i, 0, sizeof(struct net_device_stats));
4511                 spin_unlock(&cp->stat_lock[i]);
4512         }
4513         spin_unlock_irqrestore(&cp->stat_lock[N_TX_RINGS], flags);
4514         return stats + N_TX_RINGS;
4515 }
4516
4517
4518 static void cas_set_multicast(struct net_device *dev)
4519 {
4520         struct cas *cp = netdev_priv(dev);
4521         u32 rxcfg, rxcfg_new;
4522         unsigned long flags;
4523         int limit = STOP_TRIES;
4524
4525         if (!cp->hw_running)
4526                 return;
4527
4528         spin_lock_irqsave(&cp->lock, flags);
4529         rxcfg = readl(cp->regs + REG_MAC_RX_CFG);
4530
4531         /* disable RX MAC and wait for completion */
4532         writel(rxcfg & ~MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
4533         while (readl(cp->regs + REG_MAC_RX_CFG) & MAC_RX_CFG_EN) {
4534                 if (!limit--)
4535                         break;
4536                 udelay(10);
4537         }
4538
4539         /* disable hash filter and wait for completion */
4540         limit = STOP_TRIES;
4541         rxcfg &= ~(MAC_RX_CFG_PROMISC_EN | MAC_RX_CFG_HASH_FILTER_EN);
4542         writel(rxcfg & ~MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
4543         while (readl(cp->regs + REG_MAC_RX_CFG) & MAC_RX_CFG_HASH_FILTER_EN) {
4544                 if (!limit--)
4545                         break;
4546                 udelay(10);
4547         }
4548
4549         /* program hash filters */
4550         cp->mac_rx_cfg = rxcfg_new = cas_setup_multicast(cp);
4551         rxcfg |= rxcfg_new;
4552         writel(rxcfg, cp->regs + REG_MAC_RX_CFG);
4553         spin_unlock_irqrestore(&cp->lock, flags);
4554 }
4555
4556 static void cas_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
4557 {
4558         struct cas *cp = netdev_priv(dev);
4559         strncpy(info->driver, DRV_MODULE_NAME, ETHTOOL_BUSINFO_LEN);
4560         strncpy(info->version, DRV_MODULE_VERSION, ETHTOOL_BUSINFO_LEN);
4561         info->fw_version[0] = '\0';
4562         strncpy(info->bus_info, pci_name(cp->pdev), ETHTOOL_BUSINFO_LEN);
4563         info->regdump_len = cp->casreg_len < CAS_MAX_REGS ?
4564                 cp->casreg_len : CAS_MAX_REGS;
4565         info->n_stats = CAS_NUM_STAT_KEYS;
4566 }
4567
4568 static int cas_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
4569 {
4570         struct cas *cp = netdev_priv(dev);
4571         u16 bmcr;
4572         int full_duplex, speed, pause;
4573         unsigned long flags;
4574         enum link_state linkstate = link_up;
4575
4576         cmd->advertising = 0;
4577         cmd->supported = SUPPORTED_Autoneg;
4578         if (cp->cas_flags & CAS_FLAG_1000MB_CAP) {
4579                 cmd->supported |= SUPPORTED_1000baseT_Full;
4580                 cmd->advertising |= ADVERTISED_1000baseT_Full;
4581         }
4582
4583         /* Record PHY settings if HW is on. */
4584         spin_lock_irqsave(&cp->lock, flags);
4585         bmcr = 0;
4586         linkstate = cp->lstate;
4587         if (CAS_PHY_MII(cp->phy_type)) {
4588                 cmd->port = PORT_MII;
4589                 cmd->transceiver = (cp->cas_flags & CAS_FLAG_SATURN) ?
4590                         XCVR_INTERNAL : XCVR_EXTERNAL;
4591                 cmd->phy_address = cp->phy_addr;
4592                 cmd->advertising |= ADVERTISED_TP | ADVERTISED_MII |
4593                         ADVERTISED_10baseT_Half |
4594                         ADVERTISED_10baseT_Full |
4595                         ADVERTISED_100baseT_Half |
4596                         ADVERTISED_100baseT_Full;
4597
4598                 cmd->supported |=
4599                         (SUPPORTED_10baseT_Half |
4600                          SUPPORTED_10baseT_Full |
4601                          SUPPORTED_100baseT_Half |
4602                          SUPPORTED_100baseT_Full |
4603                          SUPPORTED_TP | SUPPORTED_MII);
4604
4605                 if (cp->hw_running) {
4606                         cas_mif_poll(cp, 0);
4607                         bmcr = cas_phy_read(cp, MII_BMCR);
4608                         cas_read_mii_link_mode(cp, &full_duplex,
4609                                                &speed, &pause);
4610                         cas_mif_poll(cp, 1);
4611                 }
4612
4613         } else {
4614                 cmd->port = PORT_FIBRE;
4615                 cmd->transceiver = XCVR_INTERNAL;
4616                 cmd->phy_address = 0;
4617                 cmd->supported   |= SUPPORTED_FIBRE;
4618                 cmd->advertising |= ADVERTISED_FIBRE;
4619
4620                 if (cp->hw_running) {
4621                         /* pcs uses the same bits as mii */
4622                         bmcr = readl(cp->regs + REG_PCS_MII_CTRL);
4623                         cas_read_pcs_link_mode(cp, &full_duplex,
4624                                                &speed, &pause);
4625                 }
4626         }
4627         spin_unlock_irqrestore(&cp->lock, flags);
4628
4629         if (bmcr & BMCR_ANENABLE) {
4630                 cmd->advertising |= ADVERTISED_Autoneg;
4631                 cmd->autoneg = AUTONEG_ENABLE;
4632                 cmd->speed = ((speed == 10) ?
4633                               SPEED_10 :
4634                               ((speed == 1000) ?
4635                                SPEED_1000 : SPEED_100));
4636                 cmd->duplex = full_duplex ? DUPLEX_FULL : DUPLEX_HALF;
4637         } else {
4638                 cmd->autoneg = AUTONEG_DISABLE;
4639                 cmd->speed =
4640                         (bmcr & CAS_BMCR_SPEED1000) ?
4641                         SPEED_1000 :
4642                         ((bmcr & BMCR_SPEED100) ? SPEED_100:
4643                          SPEED_10);
4644                 cmd->duplex =
4645                         (bmcr & BMCR_FULLDPLX) ?
4646                         DUPLEX_FULL : DUPLEX_HALF;
4647         }
4648         if (linkstate != link_up) {
4649                 /* Force these to "unknown" if the link is not up and
4650                  * autonogotiation in enabled. We can set the link
4651                  * speed to 0, but not cmd->duplex,
4652                  * because its legal values are 0 and 1.  Ethtool will
4653                  * print the value reported in parentheses after the
4654                  * word "Unknown" for unrecognized values.
4655                  *
4656                  * If in forced mode, we report the speed and duplex
4657                  * settings that we configured.
4658                  */
4659                 if (cp->link_cntl & BMCR_ANENABLE) {
4660                         cmd->speed = 0;
4661                         cmd->duplex = 0xff;
4662                 } else {
4663                         cmd->speed = SPEED_10;
4664                         if (cp->link_cntl & BMCR_SPEED100) {
4665                                 cmd->speed = SPEED_100;
4666                         } else if (cp->link_cntl & CAS_BMCR_SPEED1000) {
4667                                 cmd->speed = SPEED_1000;
4668                         }
4669                         cmd->duplex = (cp->link_cntl & BMCR_FULLDPLX)?
4670                                 DUPLEX_FULL : DUPLEX_HALF;
4671                 }
4672         }
4673         return 0;
4674 }
4675
4676 static int cas_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
4677 {
4678         struct cas *cp = netdev_priv(dev);
4679         unsigned long flags;
4680
4681         /* Verify the settings we care about. */
4682         if (cmd->autoneg != AUTONEG_ENABLE &&
4683             cmd->autoneg != AUTONEG_DISABLE)
4684                 return -EINVAL;
4685
4686         if (cmd->autoneg == AUTONEG_DISABLE &&
4687             ((cmd->speed != SPEED_1000 &&
4688               cmd->speed != SPEED_100 &&
4689               cmd->speed != SPEED_10) ||
4690              (cmd->duplex != DUPLEX_HALF &&
4691               cmd->duplex != DUPLEX_FULL)))
4692                 return -EINVAL;
4693
4694         /* Apply settings and restart link process. */
4695         spin_lock_irqsave(&cp->lock, flags);
4696         cas_begin_auto_negotiation(cp, cmd);
4697         spin_unlock_irqrestore(&cp->lock, flags);
4698         return 0;
4699 }
4700
4701 static int cas_nway_reset(struct net_device *dev)
4702 {
4703         struct cas *cp = netdev_priv(dev);
4704         unsigned long flags;
4705
4706         if ((cp->link_cntl & BMCR_ANENABLE) == 0)
4707                 return -EINVAL;
4708
4709         /* Restart link process. */
4710         spin_lock_irqsave(&cp->lock, flags);
4711         cas_begin_auto_negotiation(cp, NULL);
4712         spin_unlock_irqrestore(&cp->lock, flags);
4713
4714         return 0;
4715 }
4716
4717 static u32 cas_get_link(struct net_device *dev)
4718 {
4719         struct cas *cp = netdev_priv(dev);
4720         return cp->lstate == link_up;
4721 }
4722
4723 static u32 cas_get_msglevel(struct net_device *dev)
4724 {
4725         struct cas *cp = netdev_priv(dev);
4726         return cp->msg_enable;
4727 }
4728
4729 static void cas_set_msglevel(struct net_device *dev, u32 value)
4730 {
4731         struct cas *cp = netdev_priv(dev);
4732         cp->msg_enable = value;
4733 }
4734
4735 static int cas_get_regs_len(struct net_device *dev)
4736 {
4737         struct cas *cp = netdev_priv(dev);
4738         return cp->casreg_len < CAS_MAX_REGS ? cp->casreg_len: CAS_MAX_REGS;
4739 }
4740
4741 static void cas_get_regs(struct net_device *dev, struct ethtool_regs *regs,
4742                              void *p)
4743 {
4744         struct cas *cp = netdev_priv(dev);
4745         regs->version = 0;
4746         /* cas_read_regs handles locks (cp->lock).  */
4747         cas_read_regs(cp, p, regs->len / sizeof(u32));
4748 }
4749
4750 static int cas_get_sset_count(struct net_device *dev, int sset)
4751 {
4752         switch (sset) {
4753         case ETH_SS_STATS:
4754                 return CAS_NUM_STAT_KEYS;
4755         default:
4756                 return -EOPNOTSUPP;
4757         }
4758 }
4759
4760 static void cas_get_strings(struct net_device *dev, u32 stringset, u8 *data)
4761 {
4762          memcpy(data, &ethtool_cassini_statnames,
4763                                          CAS_NUM_STAT_KEYS * ETH_GSTRING_LEN);
4764 }
4765
4766 static void cas_get_ethtool_stats(struct net_device *dev,
4767                                       struct ethtool_stats *estats, u64 *data)
4768 {
4769         struct cas *cp = netdev_priv(dev);
4770         struct net_device_stats *stats = cas_get_stats(cp->dev);
4771         int i = 0;
4772         data[i++] = stats->collisions;
4773         data[i++] = stats->rx_bytes;
4774         data[i++] = stats->rx_crc_errors;
4775         data[i++] = stats->rx_dropped;
4776         data[i++] = stats->rx_errors;
4777         data[i++] = stats->rx_fifo_errors;
4778         data[i++] = stats->rx_frame_errors;
4779         data[i++] = stats->rx_length_errors;
4780         data[i++] = stats->rx_over_errors;
4781         data[i++] = stats->rx_packets;
4782         data[i++] = stats->tx_aborted_errors;
4783         data[i++] = stats->tx_bytes;
4784         data[i++] = stats->tx_dropped;
4785         data[i++] = stats->tx_errors;
4786         data[i++] = stats->tx_fifo_errors;
4787         data[i++] = stats->tx_packets;
4788         BUG_ON(i != CAS_NUM_STAT_KEYS);
4789 }
4790
4791 static const struct ethtool_ops cas_ethtool_ops = {
4792         .get_drvinfo            = cas_get_drvinfo,
4793         .get_settings           = cas_get_settings,
4794         .set_settings           = cas_set_settings,
4795         .nway_reset             = cas_nway_reset,
4796         .get_link               = cas_get_link,
4797         .get_msglevel           = cas_get_msglevel,
4798         .set_msglevel           = cas_set_msglevel,
4799         .get_regs_len           = cas_get_regs_len,
4800         .get_regs               = cas_get_regs,
4801         .get_sset_count         = cas_get_sset_count,
4802         .get_strings            = cas_get_strings,
4803         .get_ethtool_stats      = cas_get_ethtool_stats,
4804 };
4805
4806 static int cas_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
4807 {
4808         struct cas *cp = netdev_priv(dev);
4809         struct mii_ioctl_data *data = if_mii(ifr);
4810         unsigned long flags;
4811         int rc = -EOPNOTSUPP;
4812
4813         /* Hold the PM mutex while doing ioctl's or we may collide
4814          * with open/close and power management and oops.
4815          */
4816         mutex_lock(&cp->pm_mutex);
4817         switch (cmd) {
4818         case SIOCGMIIPHY:               /* Get address of MII PHY in use. */
4819                 data->phy_id = cp->phy_addr;
4820                 /* Fallthrough... */
4821
4822         case SIOCGMIIREG:               /* Read MII PHY register. */
4823                 spin_lock_irqsave(&cp->lock, flags);
4824                 cas_mif_poll(cp, 0);
4825                 data->val_out = cas_phy_read(cp, data->reg_num & 0x1f);
4826                 cas_mif_poll(cp, 1);
4827                 spin_unlock_irqrestore(&cp->lock, flags);
4828                 rc = 0;
4829                 break;
4830
4831         case SIOCSMIIREG:               /* Write MII PHY register. */
4832                 if (!capable(CAP_NET_ADMIN)) {
4833                         rc = -EPERM;
4834                         break;
4835                 }
4836                 spin_lock_irqsave(&cp->lock, flags);
4837                 cas_mif_poll(cp, 0);
4838                 rc = cas_phy_write(cp, data->reg_num & 0x1f, data->val_in);
4839                 cas_mif_poll(cp, 1);
4840                 spin_unlock_irqrestore(&cp->lock, flags);
4841                 break;
4842         default:
4843                 break;
4844         };
4845
4846         mutex_unlock(&cp->pm_mutex);
4847         return rc;
4848 }
4849
4850 /* When this chip sits underneath an Intel 31154 bridge, it is the
4851  * only subordinate device and we can tweak the bridge settings to
4852  * reflect that fact.
4853  */
4854 static void __devinit cas_program_bridge(struct pci_dev *cas_pdev)
4855 {
4856         struct pci_dev *pdev = cas_pdev->bus->self;
4857         u32 val;
4858
4859         if (!pdev)
4860                 return;
4861
4862         if (pdev->vendor != 0x8086 || pdev->device != 0x537c)
4863                 return;
4864
4865         /* Clear bit 10 (Bus Parking Control) in the Secondary
4866          * Arbiter Control/Status Register which lives at offset
4867          * 0x41.  Using a 32-bit word read/modify/write at 0x40
4868          * is much simpler so that's how we do this.
4869          */
4870         pci_read_config_dword(pdev, 0x40, &val);
4871         val &= ~0x00040000;
4872         pci_write_config_dword(pdev, 0x40, val);
4873
4874         /* Max out the Multi-Transaction Timer settings since
4875          * Cassini is the only device present.
4876          *
4877          * The register is 16-bit and lives at 0x50.  When the
4878          * settings are enabled, it extends the GRANT# signal
4879          * for a requestor after a transaction is complete.  This
4880          * allows the next request to run without first needing
4881          * to negotiate the GRANT# signal back.
4882          *
4883          * Bits 12:10 define the grant duration:
4884          *
4885          *      1       --      16 clocks
4886          *      2       --      32 clocks
4887          *      3       --      64 clocks
4888          *      4       --      128 clocks
4889          *      5       --      256 clocks
4890          *
4891          * All other values are illegal.
4892          *
4893          * Bits 09:00 define which REQ/GNT signal pairs get the
4894          * GRANT# signal treatment.  We set them all.
4895          */
4896         pci_write_config_word(pdev, 0x50, (5 << 10) | 0x3ff);
4897
4898         /* The Read Prefecth Policy register is 16-bit and sits at
4899          * offset 0x52.  It enables a "smart" pre-fetch policy.  We
4900          * enable it and max out all of the settings since only one
4901          * device is sitting underneath and thus bandwidth sharing is
4902          * not an issue.
4903          *
4904          * The register has several 3 bit fields, which indicates a
4905          * multiplier applied to the base amount of prefetching the
4906          * chip would do.  These fields are at:
4907          *
4908          *      15:13   ---     ReRead Primary Bus
4909          *      12:10   ---     FirstRead Primary Bus
4910          *      09:07   ---     ReRead Secondary Bus
4911          *      06:04   ---     FirstRead Secondary Bus
4912          *
4913          * Bits 03:00 control which REQ/GNT pairs the prefetch settings
4914          * get enabled on.  Bit 3 is a grouped enabler which controls
4915          * all of the REQ/GNT pairs from [8:3].  Bits 2 to 0 control
4916          * the individual REQ/GNT pairs [2:0].
4917          */
4918         pci_write_config_word(pdev, 0x52,
4919                               (0x7 << 13) |
4920                               (0x7 << 10) |
4921                               (0x7 <<  7) |
4922                               (0x7 <<  4) |
4923                               (0xf <<  0));
4924
4925         /* Force cacheline size to 0x8 */
4926         pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE, 0x08);
4927
4928         /* Force latency timer to maximum setting so Cassini can
4929          * sit on the bus as long as it likes.
4930          */
4931         pci_write_config_byte(pdev, PCI_LATENCY_TIMER, 0xff);
4932 }
4933
4934 static int __devinit cas_init_one(struct pci_dev *pdev,
4935                                   const struct pci_device_id *ent)
4936 {
4937         static int cas_version_printed = 0;
4938         unsigned long casreg_len;
4939         struct net_device *dev;
4940         struct cas *cp;
4941         int i, err, pci_using_dac;
4942         u16 pci_cmd;
4943         u8 orig_cacheline_size = 0, cas_cacheline_size = 0;
4944         DECLARE_MAC_BUF(mac);
4945
4946         if (cas_version_printed++ == 0)
4947                 printk(KERN_INFO "%s", version);
4948
4949         err = pci_enable_device(pdev);
4950         if (err) {
4951                 dev_err(&pdev->dev, "Cannot enable PCI device, aborting.\n");
4952                 return err;
4953         }
4954
4955         if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
4956                 dev_err(&pdev->dev, "Cannot find proper PCI device "
4957                        "base address, aborting.\n");
4958                 err = -ENODEV;
4959                 goto err_out_disable_pdev;
4960         }
4961
4962         dev = alloc_etherdev(sizeof(*cp));
4963         if (!dev) {
4964                 dev_err(&pdev->dev, "Etherdev alloc failed, aborting.\n");
4965                 err = -ENOMEM;
4966                 goto err_out_disable_pdev;
4967         }
4968         SET_NETDEV_DEV(dev, &pdev->dev);
4969
4970         err = pci_request_regions(pdev, dev->name);
4971         if (err) {
4972                 dev_err(&pdev->dev, "Cannot obtain PCI resources, aborting.\n");
4973                 goto err_out_free_netdev;
4974         }
4975         pci_set_master(pdev);
4976
4977         /* we must always turn on parity response or else parity
4978          * doesn't get generated properly. disable SERR/PERR as well.
4979          * in addition, we want to turn MWI on.
4980          */
4981         pci_read_config_word(pdev, PCI_COMMAND, &pci_cmd);
4982         pci_cmd &= ~PCI_COMMAND_SERR;
4983         pci_cmd |= PCI_COMMAND_PARITY;
4984         pci_write_config_word(pdev, PCI_COMMAND, pci_cmd);
4985         if (pci_try_set_mwi(pdev))
4986                 printk(KERN_WARNING PFX "Could not enable MWI for %s\n",
4987                        pci_name(pdev));
4988
4989         cas_program_bridge(pdev);
4990
4991         /*
4992          * On some architectures, the default cache line size set
4993          * by pci_try_set_mwi reduces perforamnce.  We have to increase
4994          * it for this case.  To start, we'll print some configuration
4995          * data.
4996          */
4997 #if 1
4998         pci_read_config_byte(pdev, PCI_CACHE_LINE_SIZE,
4999                              &orig_cacheline_size);
5000         if (orig_cacheline_size < CAS_PREF_CACHELINE_SIZE) {
5001                 cas_cacheline_size =
5002                         (CAS_PREF_CACHELINE_SIZE < SMP_CACHE_BYTES) ?
5003                         CAS_PREF_CACHELINE_SIZE : SMP_CACHE_BYTES;
5004                 if (pci_write_config_byte(pdev,
5005                                           PCI_CACHE_LINE_SIZE,
5006                                           cas_cacheline_size)) {
5007                         dev_err(&pdev->dev, "Could not set PCI cache "
5008                                "line size\n");
5009                         goto err_write_cacheline;
5010                 }
5011         }
5012 #endif
5013
5014
5015         /* Configure DMA attributes. */
5016         if (!pci_set_dma_mask(pdev, DMA_64BIT_MASK)) {
5017                 pci_using_dac = 1;
5018                 err = pci_set_consistent_dma_mask(pdev,
5019                                                   DMA_64BIT_MASK);
5020                 if (err < 0) {
5021                         dev_err(&pdev->dev, "Unable to obtain 64-bit DMA "
5022                                "for consistent allocations\n");
5023                         goto err_out_free_res;
5024                 }
5025
5026         } else {
5027                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
5028                 if (err) {
5029                         dev_err(&pdev->dev, "No usable DMA configuration, "
5030                                "aborting.\n");
5031                         goto err_out_free_res;
5032                 }
5033                 pci_using_dac = 0;
5034         }
5035
5036         casreg_len = pci_resource_len(pdev, 0);
5037
5038         cp = netdev_priv(dev);
5039         cp->pdev = pdev;
5040 #if 1
5041         /* A value of 0 indicates we never explicitly set it */
5042         cp->orig_cacheline_size = cas_cacheline_size ? orig_cacheline_size: 0;
5043 #endif
5044         cp->dev = dev;
5045         cp->msg_enable = (cassini_debug < 0) ? CAS_DEF_MSG_ENABLE :
5046           cassini_debug;
5047
5048         cp->link_transition = LINK_TRANSITION_UNKNOWN;
5049         cp->link_transition_jiffies_valid = 0;
5050
5051         spin_lock_init(&cp->lock);
5052         spin_lock_init(&cp->rx_inuse_lock);
5053         spin_lock_init(&cp->rx_spare_lock);
5054         for (i = 0; i < N_TX_RINGS; i++) {
5055                 spin_lock_init(&cp->stat_lock[i]);
5056                 spin_lock_init(&cp->tx_lock[i]);
5057         }
5058         spin_lock_init(&cp->stat_lock[N_TX_RINGS]);
5059         mutex_init(&cp->pm_mutex);
5060
5061         init_timer(&cp->link_timer);
5062         cp->link_timer.function = cas_link_timer;
5063         cp->link_timer.data = (unsigned long) cp;
5064
5065 #if 1
5066         /* Just in case the implementation of atomic operations
5067          * change so that an explicit initialization is necessary.
5068          */
5069         atomic_set(&cp->reset_task_pending, 0);
5070         atomic_set(&cp->reset_task_pending_all, 0);
5071         atomic_set(&cp->reset_task_pending_spare, 0);
5072         atomic_set(&cp->reset_task_pending_mtu, 0);
5073 #endif
5074         INIT_WORK(&cp->reset_task, cas_reset_task);
5075
5076         /* Default link parameters */
5077         if (link_mode >= 0 && link_mode <= 6)
5078                 cp->link_cntl = link_modes[link_mode];
5079         else
5080                 cp->link_cntl = BMCR_ANENABLE;
5081         cp->lstate = link_down;
5082         cp->link_transition = LINK_TRANSITION_LINK_DOWN;
5083         netif_carrier_off(cp->dev);
5084         cp->timer_ticks = 0;
5085
5086         /* give us access to cassini registers */
5087         cp->regs = pci_iomap(pdev, 0, casreg_len);
5088         if (!cp->regs) {
5089                 dev_err(&pdev->dev, "Cannot map device registers, aborting.\n");
5090                 goto err_out_free_res;
5091         }
5092         cp->casreg_len = casreg_len;
5093
5094         pci_save_state(pdev);
5095         cas_check_pci_invariants(cp);
5096         cas_hard_reset(cp);
5097         cas_reset(cp, 0);
5098         if (cas_check_invariants(cp))
5099                 goto err_out_iounmap;
5100
5101         cp->init_block = (struct cas_init_block *)
5102                 pci_alloc_consistent(pdev, sizeof(struct cas_init_block),
5103                                      &cp->block_dvma);
5104         if (!cp->init_block) {
5105                 dev_err(&pdev->dev, "Cannot allocate init block, aborting.\n");
5106                 goto err_out_iounmap;
5107         }
5108
5109         for (i = 0; i < N_TX_RINGS; i++)
5110                 cp->init_txds[i] = cp->init_block->txds[i];
5111
5112         for (i = 0; i < N_RX_DESC_RINGS; i++)
5113                 cp->init_rxds[i] = cp->init_block->rxds[i];
5114
5115         for (i = 0; i < N_RX_COMP_RINGS; i++)
5116                 cp->init_rxcs[i] = cp->init_block->rxcs[i];
5117
5118         for (i = 0; i < N_RX_FLOWS; i++)
5119                 skb_queue_head_init(&cp->rx_flows[i]);
5120
5121         dev->open = cas_open;
5122         dev->stop = cas_close;
5123         dev->hard_start_xmit = cas_start_xmit;
5124         dev->get_stats = cas_get_stats;
5125         dev->set_multicast_list = cas_set_multicast;
5126         dev->do_ioctl = cas_ioctl;
5127         dev->ethtool_ops = &cas_ethtool_ops;
5128         dev->tx_timeout = cas_tx_timeout;
5129         dev->watchdog_timeo = CAS_TX_TIMEOUT;
5130         dev->change_mtu = cas_change_mtu;
5131 #ifdef USE_NAPI
5132         netif_napi_add(dev, &cp->napi, cas_poll, 64);
5133 #endif
5134 #ifdef CONFIG_NET_POLL_CONTROLLER
5135         dev->poll_controller = cas_netpoll;
5136 #endif
5137         dev->irq = pdev->irq;
5138         dev->dma = 0;
5139
5140         /* Cassini features. */
5141         if ((cp->cas_flags & CAS_FLAG_NO_HW_CSUM) == 0)
5142                 dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG;
5143
5144         if (pci_using_dac)
5145                 dev->features |= NETIF_F_HIGHDMA;
5146
5147         if (register_netdev(dev)) {
5148                 dev_err(&pdev->dev, "Cannot register net device, aborting.\n");
5149                 goto err_out_free_consistent;
5150         }
5151
5152         i = readl(cp->regs + REG_BIM_CFG);
5153         printk(KERN_INFO "%s: Sun Cassini%s (%sbit/%sMHz PCI/%s) "
5154                "Ethernet[%d] %s\n",  dev->name,
5155                (cp->cas_flags & CAS_FLAG_REG_PLUS) ? "+" : "",
5156                (i & BIM_CFG_32BIT) ? "32" : "64",
5157                (i & BIM_CFG_66MHZ) ? "66" : "33",
5158                (cp->phy_type == CAS_PHY_SERDES) ? "Fi" : "Cu", pdev->irq,
5159                print_mac(mac, dev->dev_addr));
5160
5161         pci_set_drvdata(pdev, dev);
5162         cp->hw_running = 1;
5163         cas_entropy_reset(cp);
5164         cas_phy_init(cp);
5165         cas_begin_auto_negotiation(cp, NULL);
5166         return 0;
5167
5168 err_out_free_consistent:
5169         pci_free_consistent(pdev, sizeof(struct cas_init_block),
5170                             cp->init_block, cp->block_dvma);
5171
5172 err_out_iounmap:
5173         mutex_lock(&cp->pm_mutex);
5174         if (cp->hw_running)
5175                 cas_shutdown(cp);
5176         mutex_unlock(&cp->pm_mutex);
5177
5178         pci_iounmap(pdev, cp->regs);
5179
5180
5181 err_out_free_res:
5182         pci_release_regions(pdev);
5183
5184 err_write_cacheline:
5185         /* Try to restore it in case the error occured after we
5186          * set it.
5187          */
5188         pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE, orig_cacheline_size);
5189
5190 err_out_free_netdev:
5191         free_netdev(dev);
5192
5193 err_out_disable_pdev:
5194         pci_disable_device(pdev);
5195         pci_set_drvdata(pdev, NULL);
5196         return -ENODEV;
5197 }
5198
5199 static void __devexit cas_remove_one(struct pci_dev *pdev)
5200 {
5201         struct net_device *dev = pci_get_drvdata(pdev);
5202         struct cas *cp;
5203         if (!dev)
5204                 return;
5205
5206         cp = netdev_priv(dev);
5207         unregister_netdev(dev);
5208
5209         mutex_lock(&cp->pm_mutex);
5210         flush_scheduled_work();
5211         if (cp->hw_running)
5212                 cas_shutdown(cp);
5213         mutex_unlock(&cp->pm_mutex);
5214
5215 #if 1
5216         if (cp->orig_cacheline_size) {
5217                 /* Restore the cache line size if we had modified
5218                  * it.
5219                  */
5220                 pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE,
5221                                       cp->orig_cacheline_size);
5222         }
5223 #endif
5224         pci_free_consistent(pdev, sizeof(struct cas_init_block),
5225                             cp->init_block, cp->block_dvma);
5226         pci_iounmap(pdev, cp->regs);
5227         free_netdev(dev);
5228         pci_release_regions(pdev);
5229         pci_disable_device(pdev);
5230         pci_set_drvdata(pdev, NULL);
5231 }
5232
5233 #ifdef CONFIG_PM
5234 static int cas_suspend(struct pci_dev *pdev, pm_message_t state)
5235 {
5236         struct net_device *dev = pci_get_drvdata(pdev);
5237         struct cas *cp = netdev_priv(dev);
5238         unsigned long flags;
5239
5240         mutex_lock(&cp->pm_mutex);
5241
5242         /* If the driver is opened, we stop the DMA */
5243         if (cp->opened) {
5244                 netif_device_detach(dev);
5245
5246                 cas_lock_all_save(cp, flags);
5247
5248                 /* We can set the second arg of cas_reset to 0
5249                  * because on resume, we'll call cas_init_hw with
5250                  * its second arg set so that autonegotiation is
5251                  * restarted.
5252                  */
5253                 cas_reset(cp, 0);
5254                 cas_clean_rings(cp);
5255                 cas_unlock_all_restore(cp, flags);
5256         }
5257
5258         if (cp->hw_running)
5259                 cas_shutdown(cp);
5260         mutex_unlock(&cp->pm_mutex);
5261
5262         return 0;
5263 }
5264
5265 static int cas_resume(struct pci_dev *pdev)
5266 {
5267         struct net_device *dev = pci_get_drvdata(pdev);
5268         struct cas *cp = netdev_priv(dev);
5269
5270         printk(KERN_INFO "%s: resuming\n", dev->name);
5271
5272         mutex_lock(&cp->pm_mutex);
5273         cas_hard_reset(cp);
5274         if (cp->opened) {
5275                 unsigned long flags;
5276                 cas_lock_all_save(cp, flags);
5277                 cas_reset(cp, 0);
5278                 cp->hw_running = 1;
5279                 cas_clean_rings(cp);
5280                 cas_init_hw(cp, 1);
5281                 cas_unlock_all_restore(cp, flags);
5282
5283                 netif_device_attach(dev);
5284         }
5285         mutex_unlock(&cp->pm_mutex);
5286         return 0;
5287 }
5288 #endif /* CONFIG_PM */
5289
5290 static struct pci_driver cas_driver = {
5291         .name           = DRV_MODULE_NAME,
5292         .id_table       = cas_pci_tbl,
5293         .probe          = cas_init_one,
5294         .remove         = __devexit_p(cas_remove_one),
5295 #ifdef CONFIG_PM
5296         .suspend        = cas_suspend,
5297         .resume         = cas_resume
5298 #endif
5299 };
5300
5301 static int __init cas_init(void)
5302 {
5303         if (linkdown_timeout > 0)
5304                 link_transition_timeout = linkdown_timeout * HZ;
5305         else
5306                 link_transition_timeout = 0;
5307
5308         return pci_register_driver(&cas_driver);
5309 }
5310
5311 static void __exit cas_cleanup(void)
5312 {
5313         pci_unregister_driver(&cas_driver);
5314 }
5315
5316 module_init(cas_init);
5317 module_exit(cas_cleanup);