Merge master.kernel.org:/home/rmk/linux-2.6-serial
[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 #include <linux/config.h>
70 #include <linux/version.h>
71
72 #include <linux/module.h>
73 #include <linux/kernel.h>
74 #include <linux/types.h>
75 #include <linux/compiler.h>
76 #include <linux/slab.h>
77 #include <linux/delay.h>
78 #include <linux/init.h>
79 #include <linux/ioport.h>
80 #include <linux/pci.h>
81 #include <linux/mm.h>
82 #include <linux/highmem.h>
83 #include <linux/list.h>
84 #include <linux/dma-mapping.h>
85
86 #include <linux/netdevice.h>
87 #include <linux/etherdevice.h>
88 #include <linux/skbuff.h>
89 #include <linux/ethtool.h>
90 #include <linux/crc32.h>
91 #include <linux/random.h>
92 #include <linux/mii.h>
93 #include <linux/ip.h>
94 #include <linux/tcp.h>
95
96 #include <net/checksum.h>
97
98 #include <asm/atomic.h>
99 #include <asm/system.h>
100 #include <asm/io.h>
101 #include <asm/byteorder.h>
102 #include <asm/uaccess.h>
103
104 #define cas_page_map(x)      kmap_atomic((x), KM_SKB_DATA_SOFTIRQ)
105 #define cas_page_unmap(x)    kunmap_atomic((x), KM_SKB_DATA_SOFTIRQ)
106 #define CAS_NCPUS            num_online_cpus()
107
108 #if defined(CONFIG_CASSINI_NAPI) && defined(HAVE_NETDEV_POLL)
109 #define USE_NAPI
110 #define cas_skb_release(x)  netif_receive_skb(x)
111 #else
112 #define cas_skb_release(x)  netif_rx(x)
113 #endif
114
115 /* select which firmware to use */
116 #define USE_HP_WORKAROUND     
117 #define HP_WORKAROUND_DEFAULT /* select which firmware to use as default */
118 #define CAS_HP_ALT_FIRMWARE   cas_prog_null /* alternate firmware */
119
120 #include "cassini.h"
121
122 #define USE_TX_COMPWB      /* use completion writeback registers */
123 #define USE_CSMA_CD_PROTO  /* standard CSMA/CD */
124 #define USE_RX_BLANK       /* hw interrupt mitigation */
125 #undef USE_ENTROPY_DEV     /* don't test for entropy device */
126
127 /* NOTE: these aren't useable unless PCI interrupts can be assigned.
128  * also, we need to make cp->lock finer-grained.
129  */
130 #undef  USE_PCI_INTB
131 #undef  USE_PCI_INTC
132 #undef  USE_PCI_INTD
133 #undef  USE_QOS
134
135 #undef  USE_VPD_DEBUG       /* debug vpd information if defined */
136
137 /* rx processing options */
138 #define USE_PAGE_ORDER      /* specify to allocate large rx pages */
139 #define RX_DONT_BATCH  0    /* if 1, don't batch flows */
140 #define RX_COPY_ALWAYS 0    /* if 0, use frags */
141 #define RX_COPY_MIN    64   /* copy a little to make upper layers happy */
142 #undef  RX_COUNT_BUFFERS    /* define to calculate RX buffer stats */
143
144 #define DRV_MODULE_NAME         "cassini"
145 #define PFX DRV_MODULE_NAME     ": "
146 #define DRV_MODULE_VERSION      "1.4"
147 #define DRV_MODULE_RELDATE      "1 July 2004"
148
149 #define CAS_DEF_MSG_ENABLE        \
150         (NETIF_MSG_DRV          | \
151          NETIF_MSG_PROBE        | \
152          NETIF_MSG_LINK         | \
153          NETIF_MSG_TIMER        | \
154          NETIF_MSG_IFDOWN       | \
155          NETIF_MSG_IFUP         | \
156          NETIF_MSG_RX_ERR       | \
157          NETIF_MSG_TX_ERR)
158
159 /* length of time before we decide the hardware is borked,
160  * and dev->tx_timeout() should be called to fix the problem
161  */
162 #define CAS_TX_TIMEOUT                  (HZ)
163 #define CAS_LINK_TIMEOUT                (22*HZ/10)
164 #define CAS_LINK_FAST_TIMEOUT           (1)
165
166 /* timeout values for state changing. these specify the number
167  * of 10us delays to be used before giving up.
168  */
169 #define STOP_TRIES_PHY 1000
170 #define STOP_TRIES     5000
171
172 /* specify a minimum frame size to deal with some fifo issues 
173  * max mtu == 2 * page size - ethernet header - 64 - swivel =
174  *            2 * page_size - 0x50
175  */
176 #define CAS_MIN_FRAME                   97
177 #define CAS_1000MB_MIN_FRAME            255
178 #define CAS_MIN_MTU                     60
179 #define CAS_MAX_MTU                     min(((cp->page_size << 1) - 0x50), 9000)
180
181 #if 1
182 /*
183  * Eliminate these and use separate atomic counters for each, to
184  * avoid a race condition.
185  */
186 #else
187 #define CAS_RESET_MTU                   1
188 #define CAS_RESET_ALL                   2
189 #define CAS_RESET_SPARE                 3
190 #endif
191
192 static char version[] __devinitdata =
193         DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
194
195 MODULE_AUTHOR("Adrian Sun (asun@darksunrising.com)");
196 MODULE_DESCRIPTION("Sun Cassini(+) ethernet driver");
197 MODULE_LICENSE("GPL");
198 MODULE_PARM(cassini_debug, "i");
199 MODULE_PARM_DESC(cassini_debug, "Cassini bitmapped debugging message enable value");
200 MODULE_PARM(link_mode, "i");
201 MODULE_PARM_DESC(link_mode, "default link mode");
202
203 /*
204  * Work around for a PCS bug in which the link goes down due to the chip
205  * being confused and never showing a link status of "up."
206  */
207 #define DEFAULT_LINKDOWN_TIMEOUT 5
208 /* 
209  * Value in seconds, for user input.
210  */
211 static int linkdown_timeout = DEFAULT_LINKDOWN_TIMEOUT;
212 MODULE_PARM(linkdown_timeout, "i");
213 MODULE_PARM_DESC(linkdown_timeout,
214 "min reset interval in sec. for PCS linkdown issue; disabled if not positive");
215
216 /*
217  * value in 'ticks' (units used by jiffies). Set when we init the
218  * module because 'HZ' in actually a function call on some flavors of
219  * Linux.  This will default to DEFAULT_LINKDOWN_TIMEOUT * HZ.
220  */
221 static int link_transition_timeout;
222
223
224 static int cassini_debug = -1;  /* -1 == use CAS_DEF_MSG_ENABLE as value */
225 static int link_mode;
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 int 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 int 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, %lx\n",
1930                         cp->dev->name, status, 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         char *p; 
1957
1958         hlen = CAS_VAL(RX_COMP2_HDR_SIZE, words[1]);
1959         dlen = CAS_VAL(RX_COMP1_DATA_SIZE, words[0]);
1960         len  = hlen + dlen;
1961
1962         if (RX_COPY_ALWAYS || (words[2] & RX_COMP3_SMALL_PKT)) 
1963                 alloclen = len;
1964         else 
1965                 alloclen = max(hlen, RX_COPY_MIN);
1966
1967         skb = dev_alloc_skb(alloclen + swivel + cp->crc_size);
1968         if (skb == NULL) 
1969                 return -1;
1970
1971         *skbref = skb;
1972         skb->dev = cp->dev;
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->len      += hlen - swivel;
2041
2042                 get_page(page->buffer);
2043                 frag->page = page->buffer;
2044                 frag->page_offset = off;
2045                 frag->size = hlen - swivel;
2046                 
2047                 /* any more data? */
2048                 if ((words[0] & RX_COMP1_SPLIT_PKT) && ((dlen -= hlen) > 0)) {
2049                         hlen = dlen;
2050                         off = 0;
2051
2052                         i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
2053                         page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
2054                         pci_dma_sync_single_for_cpu(cp->pdev, page->dma_addr, 
2055                                             hlen + cp->crc_size, 
2056                                             PCI_DMA_FROMDEVICE);
2057                         pci_dma_sync_single_for_device(cp->pdev, page->dma_addr,
2058                                             hlen + cp->crc_size,
2059                                             PCI_DMA_FROMDEVICE);
2060
2061                         skb_shinfo(skb)->nr_frags++;
2062                         skb->data_len += hlen;
2063                         skb->len      += hlen; 
2064                         frag++;
2065
2066                         get_page(page->buffer);
2067                         frag->page = page->buffer;
2068                         frag->page_offset = 0;
2069                         frag->size = hlen;
2070                         RX_USED_ADD(page, hlen + cp->crc_size);
2071                 }
2072
2073                 if (cp->crc_size) {
2074                         addr = cas_page_map(page->buffer);
2075                         crcaddr  = addr + off + hlen;
2076                 }
2077
2078         } else {
2079                 /* copying packet */
2080                 if (!dlen)
2081                         goto end_copy_pkt;
2082
2083                 i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
2084                 page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
2085                 off = CAS_VAL(RX_COMP1_DATA_OFF, words[0]) + swivel;
2086                 hlen = min(cp->page_size - off, dlen);
2087                 if (hlen < 0) {
2088                         if (netif_msg_rx_err(cp)) {
2089                                 printk(KERN_DEBUG "%s: rx page overflow: "
2090                                        "%d\n", cp->dev->name, hlen);
2091                         }
2092                         dev_kfree_skb_irq(skb);
2093                         return -1;
2094                 }
2095                 i = hlen;
2096                 if (i == dlen) /* attach FCS */
2097                         i += cp->crc_size;
2098                 pci_dma_sync_single_for_cpu(cp->pdev, page->dma_addr + off, i,
2099                                     PCI_DMA_FROMDEVICE);
2100                 addr = cas_page_map(page->buffer);
2101                 memcpy(p, addr + off, i);
2102                 pci_dma_sync_single_for_device(cp->pdev, page->dma_addr + off, i,
2103                                     PCI_DMA_FROMDEVICE);
2104                 cas_page_unmap(addr);
2105                 if (p == (char *) skb->data) /* not split */
2106                         RX_USED_ADD(page, cp->mtu_stride);
2107                 else
2108                         RX_USED_ADD(page, i);
2109         
2110                 /* any more data? */
2111                 if ((words[0] & RX_COMP1_SPLIT_PKT) && ((dlen -= hlen) > 0)) {
2112                         p += hlen;
2113                         i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
2114                         page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
2115                         pci_dma_sync_single_for_cpu(cp->pdev, page->dma_addr, 
2116                                             dlen + cp->crc_size, 
2117                                             PCI_DMA_FROMDEVICE);
2118                         addr = cas_page_map(page->buffer);
2119                         memcpy(p, addr, dlen + cp->crc_size);
2120                         pci_dma_sync_single_for_device(cp->pdev, page->dma_addr,
2121                                             dlen + cp->crc_size,
2122                                             PCI_DMA_FROMDEVICE);
2123                         cas_page_unmap(addr);
2124                         RX_USED_ADD(page, dlen + cp->crc_size); 
2125                 }
2126 end_copy_pkt:
2127                 if (cp->crc_size) {
2128                         addr    = NULL;
2129                         crcaddr = skb->data + alloclen;
2130                 }
2131                 skb_put(skb, alloclen);
2132         }
2133
2134         i = CAS_VAL(RX_COMP4_TCP_CSUM, words[3]);
2135         if (cp->crc_size) {
2136                 /* checksum includes FCS. strip it out. */
2137                 i = csum_fold(csum_partial(crcaddr, cp->crc_size, i));
2138                 if (addr)
2139                         cas_page_unmap(addr);
2140         }
2141         skb->csum = ntohs(i ^ 0xffff);
2142         skb->ip_summed = CHECKSUM_HW;
2143         skb->protocol = eth_type_trans(skb, cp->dev);
2144         return len;
2145 }
2146
2147
2148 /* we can handle up to 64 rx flows at a time. we do the same thing
2149  * as nonreassm except that we batch up the buffers. 
2150  * NOTE: we currently just treat each flow as a bunch of packets that
2151  *       we pass up. a better way would be to coalesce the packets
2152  *       into a jumbo packet. to do that, we need to do the following:
2153  *       1) the first packet will have a clean split between header and
2154  *          data. save both.
2155  *       2) each time the next flow packet comes in, extend the
2156  *          data length and merge the checksums.
2157  *       3) on flow release, fix up the header.
2158  *       4) make sure the higher layer doesn't care.
2159  * because packets get coalesced, we shouldn't run into fragment count 
2160  * issues.
2161  */
2162 static inline void cas_rx_flow_pkt(struct cas *cp, const u64 *words,
2163                                    struct sk_buff *skb)
2164 {
2165         int flowid = CAS_VAL(RX_COMP3_FLOWID, words[2]) & (N_RX_FLOWS - 1);
2166         struct sk_buff_head *flow = &cp->rx_flows[flowid];
2167         
2168         /* this is protected at a higher layer, so no need to 
2169          * do any additional locking here. stick the buffer
2170          * at the end.
2171          */
2172         __skb_insert(skb, flow->prev, (struct sk_buff *) flow, flow);
2173         if (words[0] & RX_COMP1_RELEASE_FLOW) {
2174                 while ((skb = __skb_dequeue(flow))) {
2175                         cas_skb_release(skb);
2176                 }
2177         }
2178 }
2179
2180 /* put rx descriptor back on ring. if a buffer is in use by a higher
2181  * layer, this will need to put in a replacement.
2182  */
2183 static void cas_post_page(struct cas *cp, const int ring, const int index)
2184 {
2185         cas_page_t *new;
2186         int entry;
2187
2188         entry = cp->rx_old[ring];
2189
2190         new = cas_page_swap(cp, ring, index);
2191         cp->init_rxds[ring][entry].buffer = cpu_to_le64(new->dma_addr);
2192         cp->init_rxds[ring][entry].index  =
2193                 cpu_to_le64(CAS_BASE(RX_INDEX_NUM, index) | 
2194                             CAS_BASE(RX_INDEX_RING, ring));
2195
2196         entry = RX_DESC_ENTRY(ring, entry + 1);
2197         cp->rx_old[ring] = entry;
2198         
2199         if (entry % 4)
2200                 return;
2201
2202         if (ring == 0)
2203                 writel(entry, cp->regs + REG_RX_KICK);
2204         else if ((N_RX_DESC_RINGS > 1) &&
2205                  (cp->cas_flags & CAS_FLAG_REG_PLUS)) 
2206                 writel(entry, cp->regs + REG_PLUS_RX_KICK1);
2207 }
2208
2209
2210 /* only when things are bad */
2211 static int cas_post_rxds_ringN(struct cas *cp, int ring, int num)
2212 {
2213         unsigned int entry, last, count, released;
2214         int cluster;
2215         cas_page_t **page = cp->rx_pages[ring];
2216
2217         entry = cp->rx_old[ring];
2218
2219         if (netif_msg_intr(cp))
2220                 printk(KERN_DEBUG "%s: rxd[%d] interrupt, done: %d\n",
2221                        cp->dev->name, ring, entry);
2222
2223         cluster = -1;
2224         count = entry & 0x3; 
2225         last = RX_DESC_ENTRY(ring, num ? entry + num - 4: entry - 4);
2226         released = 0;
2227         while (entry != last) {
2228                 /* make a new buffer if it's still in use */
2229                 if (page_count(page[entry]->buffer) > 1) {
2230                         cas_page_t *new = cas_page_dequeue(cp);
2231                         if (!new) {
2232                                 /* let the timer know that we need to 
2233                                  * do this again
2234                                  */
2235                                 cp->cas_flags |= CAS_FLAG_RXD_POST(ring);
2236                                 if (!timer_pending(&cp->link_timer))
2237                                         mod_timer(&cp->link_timer, jiffies + 
2238                                                   CAS_LINK_FAST_TIMEOUT);
2239                                 cp->rx_old[ring]  = entry;
2240                                 cp->rx_last[ring] = num ? num - released : 0;
2241                                 return -ENOMEM;
2242                         }
2243                         spin_lock(&cp->rx_inuse_lock);
2244                         list_add(&page[entry]->list, &cp->rx_inuse_list);
2245                         spin_unlock(&cp->rx_inuse_lock);
2246                         cp->init_rxds[ring][entry].buffer = 
2247                                 cpu_to_le64(new->dma_addr);
2248                         page[entry] = new;
2249                         
2250                 }
2251
2252                 if (++count == 4) {
2253                         cluster = entry;
2254                         count = 0;
2255                 }
2256                 released++;
2257                 entry = RX_DESC_ENTRY(ring, entry + 1);
2258         }
2259         cp->rx_old[ring] = entry;
2260
2261         if (cluster < 0) 
2262                 return 0;
2263
2264         if (ring == 0)
2265                 writel(cluster, cp->regs + REG_RX_KICK);
2266         else if ((N_RX_DESC_RINGS > 1) &&
2267                  (cp->cas_flags & CAS_FLAG_REG_PLUS)) 
2268                 writel(cluster, cp->regs + REG_PLUS_RX_KICK1);
2269         return 0;
2270 }
2271
2272
2273 /* process a completion ring. packets are set up in three basic ways:
2274  * small packets: should be copied header + data in single buffer.
2275  * large packets: header and data in a single buffer.
2276  * split packets: header in a separate buffer from data. 
2277  *                data may be in multiple pages. data may be > 256
2278  *                bytes but in a single page. 
2279  *
2280  * NOTE: RX page posting is done in this routine as well. while there's
2281  *       the capability of using multiple RX completion rings, it isn't
2282  *       really worthwhile due to the fact that the page posting will
2283  *       force serialization on the single descriptor ring. 
2284  */
2285 static int cas_rx_ringN(struct cas *cp, int ring, int budget)
2286 {
2287         struct cas_rx_comp *rxcs = cp->init_rxcs[ring];
2288         int entry, drops;
2289         int npackets = 0;
2290
2291         if (netif_msg_intr(cp))
2292                 printk(KERN_DEBUG "%s: rx[%d] interrupt, done: %d/%d\n",
2293                        cp->dev->name, ring,
2294                        readl(cp->regs + REG_RX_COMP_HEAD), 
2295                        cp->rx_new[ring]);
2296
2297         entry = cp->rx_new[ring];
2298         drops = 0;
2299         while (1) {
2300                 struct cas_rx_comp *rxc = rxcs + entry;
2301                 struct sk_buff *skb;
2302                 int type, len;
2303                 u64 words[4];
2304                 int i, dring;
2305
2306                 words[0] = le64_to_cpu(rxc->word1);
2307                 words[1] = le64_to_cpu(rxc->word2);
2308                 words[2] = le64_to_cpu(rxc->word3);
2309                 words[3] = le64_to_cpu(rxc->word4);
2310
2311                 /* don't touch if still owned by hw */
2312                 type = CAS_VAL(RX_COMP1_TYPE, words[0]);
2313                 if (type == 0)
2314                         break;
2315
2316                 /* hw hasn't cleared the zero bit yet */
2317                 if (words[3] & RX_COMP4_ZERO) {
2318                         break;
2319                 }
2320
2321                 /* get info on the packet */
2322                 if (words[3] & (RX_COMP4_LEN_MISMATCH | RX_COMP4_BAD)) {
2323                         spin_lock(&cp->stat_lock[ring]);
2324                         cp->net_stats[ring].rx_errors++;
2325                         if (words[3] & RX_COMP4_LEN_MISMATCH)
2326                                 cp->net_stats[ring].rx_length_errors++;
2327                         if (words[3] & RX_COMP4_BAD)
2328                                 cp->net_stats[ring].rx_crc_errors++;
2329                         spin_unlock(&cp->stat_lock[ring]);
2330
2331                         /* We'll just return it to Cassini. */
2332                 drop_it:
2333                         spin_lock(&cp->stat_lock[ring]);
2334                         ++cp->net_stats[ring].rx_dropped;
2335                         spin_unlock(&cp->stat_lock[ring]);
2336                         goto next;
2337                 }
2338
2339                 len = cas_rx_process_pkt(cp, rxc, entry, words, &skb);
2340                 if (len < 0) {
2341                         ++drops;
2342                         goto drop_it;
2343                 }
2344
2345                 /* see if it's a flow re-assembly or not. the driver
2346                  * itself handles release back up.
2347                  */
2348                 if (RX_DONT_BATCH || (type == 0x2)) {
2349                         /* non-reassm: these always get released */
2350                         cas_skb_release(skb); 
2351                 } else {
2352                         cas_rx_flow_pkt(cp, words, skb);
2353                 }
2354
2355                 spin_lock(&cp->stat_lock[ring]);
2356                 cp->net_stats[ring].rx_packets++;
2357                 cp->net_stats[ring].rx_bytes += len;
2358                 spin_unlock(&cp->stat_lock[ring]);
2359                 cp->dev->last_rx = jiffies;
2360
2361         next:
2362                 npackets++;
2363
2364                 /* should it be released? */
2365                 if (words[0] & RX_COMP1_RELEASE_HDR) {
2366                         i = CAS_VAL(RX_COMP2_HDR_INDEX, words[1]);
2367                         dring = CAS_VAL(RX_INDEX_RING, i);
2368                         i = CAS_VAL(RX_INDEX_NUM, i);
2369                         cas_post_page(cp, dring, i);
2370                 }
2371                 
2372                 if (words[0] & RX_COMP1_RELEASE_DATA) {
2373                         i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
2374                         dring = CAS_VAL(RX_INDEX_RING, i);
2375                         i = CAS_VAL(RX_INDEX_NUM, i);
2376                         cas_post_page(cp, dring, i);
2377                 }
2378
2379                 if (words[0] & RX_COMP1_RELEASE_NEXT) {
2380                         i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
2381                         dring = CAS_VAL(RX_INDEX_RING, i);
2382                         i = CAS_VAL(RX_INDEX_NUM, i);
2383                         cas_post_page(cp, dring, i);
2384                 }
2385
2386                 /* skip to the next entry */
2387                 entry = RX_COMP_ENTRY(ring, entry + 1 + 
2388                                       CAS_VAL(RX_COMP1_SKIP, words[0]));
2389 #ifdef USE_NAPI
2390                 if (budget && (npackets >= budget))
2391                         break;
2392 #endif
2393         }
2394         cp->rx_new[ring] = entry;
2395
2396         if (drops)
2397                 printk(KERN_INFO "%s: Memory squeeze, deferring packet.\n",
2398                        cp->dev->name);
2399         return npackets;
2400 }
2401
2402
2403 /* put completion entries back on the ring */
2404 static void cas_post_rxcs_ringN(struct net_device *dev,
2405                                 struct cas *cp, int ring)
2406 {
2407         struct cas_rx_comp *rxc = cp->init_rxcs[ring];
2408         int last, entry;
2409
2410         last = cp->rx_cur[ring];
2411         entry = cp->rx_new[ring]; 
2412         if (netif_msg_intr(cp))
2413                 printk(KERN_DEBUG "%s: rxc[%d] interrupt, done: %d/%d\n",
2414                        dev->name, ring, readl(cp->regs + REG_RX_COMP_HEAD),
2415                        entry);
2416         
2417         /* zero and re-mark descriptors */
2418         while (last != entry) {
2419                 cas_rxc_init(rxc + last);
2420                 last = RX_COMP_ENTRY(ring, last + 1);
2421         }
2422         cp->rx_cur[ring] = last;
2423
2424         if (ring == 0)
2425                 writel(last, cp->regs + REG_RX_COMP_TAIL);
2426         else if (cp->cas_flags & CAS_FLAG_REG_PLUS) 
2427                 writel(last, cp->regs + REG_PLUS_RX_COMPN_TAIL(ring));
2428 }
2429
2430
2431
2432 /* cassini can use all four PCI interrupts for the completion ring. 
2433  * rings 3 and 4 are identical
2434  */
2435 #if defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
2436 static inline void cas_handle_irqN(struct net_device *dev, 
2437                                    struct cas *cp, const u32 status,
2438                                    const int ring)
2439 {
2440         if (status & (INTR_RX_COMP_FULL_ALT | INTR_RX_COMP_AF_ALT)) 
2441                 cas_post_rxcs_ringN(dev, cp, ring);
2442 }
2443
2444 static irqreturn_t cas_interruptN(int irq, void *dev_id, struct pt_regs *regs)
2445 {
2446         struct net_device *dev = dev_id;
2447         struct cas *cp = netdev_priv(dev);
2448         unsigned long flags;
2449         int ring;
2450         u32 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(ring));
2451
2452         /* check for shared irq */
2453         if (status == 0)
2454                 return IRQ_NONE;
2455
2456         ring = (irq == cp->pci_irq_INTC) ? 2 : 3;
2457         spin_lock_irqsave(&cp->lock, flags);
2458         if (status & INTR_RX_DONE_ALT) { /* handle rx separately */
2459 #ifdef USE_NAPI
2460                 cas_mask_intr(cp);
2461                 netif_rx_schedule(dev);
2462 #else
2463                 cas_rx_ringN(cp, ring, 0);
2464 #endif
2465                 status &= ~INTR_RX_DONE_ALT;
2466         }
2467
2468         if (status)
2469                 cas_handle_irqN(dev, cp, status, ring);
2470         spin_unlock_irqrestore(&cp->lock, flags);
2471         return IRQ_HANDLED;
2472 }
2473 #endif
2474
2475 #ifdef USE_PCI_INTB
2476 /* everything but rx packets */
2477 static inline void cas_handle_irq1(struct cas *cp, const u32 status)
2478 {
2479         if (status & INTR_RX_BUF_UNAVAIL_1) {
2480                 /* Frame arrived, no free RX buffers available. 
2481                  * NOTE: we can get this on a link transition. */
2482                 cas_post_rxds_ringN(cp, 1, 0);
2483                 spin_lock(&cp->stat_lock[1]);
2484                 cp->net_stats[1].rx_dropped++;
2485                 spin_unlock(&cp->stat_lock[1]);
2486         }
2487
2488         if (status & INTR_RX_BUF_AE_1) 
2489                 cas_post_rxds_ringN(cp, 1, RX_DESC_RINGN_SIZE(1) - 
2490                                     RX_AE_FREEN_VAL(1));
2491
2492         if (status & (INTR_RX_COMP_AF | INTR_RX_COMP_FULL))
2493                 cas_post_rxcs_ringN(cp, 1);
2494 }
2495
2496 /* ring 2 handles a few more events than 3 and 4 */
2497 static irqreturn_t cas_interrupt1(int irq, void *dev_id, struct pt_regs *regs)
2498 {
2499         struct net_device *dev = dev_id;
2500         struct cas *cp = netdev_priv(dev);
2501         unsigned long flags;
2502         u32 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(1));
2503
2504         /* check for shared interrupt */
2505         if (status == 0)
2506                 return IRQ_NONE;
2507
2508         spin_lock_irqsave(&cp->lock, flags);
2509         if (status & INTR_RX_DONE_ALT) { /* handle rx separately */
2510 #ifdef USE_NAPI
2511                 cas_mask_intr(cp);
2512                 netif_rx_schedule(dev);
2513 #else
2514                 cas_rx_ringN(cp, 1, 0);
2515 #endif
2516                 status &= ~INTR_RX_DONE_ALT;
2517         }
2518         if (status)
2519                 cas_handle_irq1(cp, status);
2520         spin_unlock_irqrestore(&cp->lock, flags);
2521         return IRQ_HANDLED;
2522 }
2523 #endif
2524
2525 static inline void cas_handle_irq(struct net_device *dev,
2526                                   struct cas *cp, const u32 status)
2527 {
2528         /* housekeeping interrupts */
2529         if (status & INTR_ERROR_MASK)
2530                 cas_abnormal_irq(dev, cp, status);
2531
2532         if (status & INTR_RX_BUF_UNAVAIL) {
2533                 /* Frame arrived, no free RX buffers available. 
2534                  * NOTE: we can get this on a link transition.
2535                  */
2536                 cas_post_rxds_ringN(cp, 0, 0);
2537                 spin_lock(&cp->stat_lock[0]);
2538                 cp->net_stats[0].rx_dropped++;
2539                 spin_unlock(&cp->stat_lock[0]);
2540         } else if (status & INTR_RX_BUF_AE) {
2541                 cas_post_rxds_ringN(cp, 0, RX_DESC_RINGN_SIZE(0) -
2542                                     RX_AE_FREEN_VAL(0));
2543         }
2544
2545         if (status & (INTR_RX_COMP_AF | INTR_RX_COMP_FULL))
2546                 cas_post_rxcs_ringN(dev, cp, 0);
2547 }
2548
2549 static irqreturn_t cas_interrupt(int irq, void *dev_id, struct pt_regs *regs)
2550 {
2551         struct net_device *dev = dev_id;
2552         struct cas *cp = netdev_priv(dev);
2553         unsigned long flags;
2554         u32 status = readl(cp->regs + REG_INTR_STATUS);
2555
2556         if (status == 0)
2557                 return IRQ_NONE;
2558
2559         spin_lock_irqsave(&cp->lock, flags);
2560         if (status & (INTR_TX_ALL | INTR_TX_INTME)) {
2561                 cas_tx(dev, cp, status);
2562                 status &= ~(INTR_TX_ALL | INTR_TX_INTME);
2563         }
2564
2565         if (status & INTR_RX_DONE) {
2566 #ifdef USE_NAPI
2567                 cas_mask_intr(cp);
2568                 netif_rx_schedule(dev);
2569 #else
2570                 cas_rx_ringN(cp, 0, 0);
2571 #endif
2572                 status &= ~INTR_RX_DONE;
2573         }
2574
2575         if (status)
2576                 cas_handle_irq(dev, cp, status);
2577         spin_unlock_irqrestore(&cp->lock, flags);
2578         return IRQ_HANDLED;
2579 }
2580
2581
2582 #ifdef USE_NAPI
2583 static int cas_poll(struct net_device *dev, int *budget)
2584 {
2585         struct cas *cp = netdev_priv(dev);
2586         int i, enable_intr, todo, credits;
2587         u32 status = readl(cp->regs + REG_INTR_STATUS);
2588         unsigned long flags;
2589
2590         spin_lock_irqsave(&cp->lock, flags);
2591         cas_tx(dev, cp, status);
2592         spin_unlock_irqrestore(&cp->lock, flags);
2593
2594         /* NAPI rx packets. we spread the credits across all of the
2595          * rxc rings
2596          */
2597         todo = min(*budget, dev->quota);
2598
2599         /* to make sure we're fair with the work we loop through each
2600          * ring N_RX_COMP_RING times with a request of 
2601          * todo / N_RX_COMP_RINGS
2602          */
2603         enable_intr = 1;
2604         credits = 0;
2605         for (i = 0; i < N_RX_COMP_RINGS; i++) {
2606                 int j;
2607                 for (j = 0; j < N_RX_COMP_RINGS; j++) {
2608                         credits += cas_rx_ringN(cp, j, todo / N_RX_COMP_RINGS);
2609                         if (credits >= todo) {
2610                                 enable_intr = 0;
2611                                 goto rx_comp;
2612                         }
2613                 }
2614         }
2615
2616 rx_comp:
2617         *budget    -= credits;
2618         dev->quota -= credits;
2619
2620         /* final rx completion */
2621         spin_lock_irqsave(&cp->lock, flags);
2622         if (status)
2623                 cas_handle_irq(dev, cp, status);
2624
2625 #ifdef USE_PCI_INTB
2626         if (N_RX_COMP_RINGS > 1) {
2627                 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(1));
2628                 if (status)
2629                         cas_handle_irq1(dev, cp, status);
2630         }
2631 #endif
2632
2633 #ifdef USE_PCI_INTC
2634         if (N_RX_COMP_RINGS > 2) {
2635                 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(2));
2636                 if (status)
2637                         cas_handle_irqN(dev, cp, status, 2);
2638         }
2639 #endif
2640
2641 #ifdef USE_PCI_INTD
2642         if (N_RX_COMP_RINGS > 3) {
2643                 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(3));
2644                 if (status)
2645                         cas_handle_irqN(dev, cp, status, 3);
2646         }
2647 #endif
2648         spin_unlock_irqrestore(&cp->lock, flags);
2649         if (enable_intr) {
2650                 netif_rx_complete(dev);
2651                 cas_unmask_intr(cp);
2652                 return 0;
2653         }
2654         return 1;
2655 }
2656 #endif
2657
2658 #ifdef CONFIG_NET_POLL_CONTROLLER
2659 static void cas_netpoll(struct net_device *dev)
2660 {
2661         struct cas *cp = netdev_priv(dev);
2662
2663         cas_disable_irq(cp, 0);
2664         cas_interrupt(cp->pdev->irq, dev, NULL);
2665         cas_enable_irq(cp, 0);
2666
2667 #ifdef USE_PCI_INTB
2668         if (N_RX_COMP_RINGS > 1) {
2669                 /* cas_interrupt1(); */
2670         }
2671 #endif
2672 #ifdef USE_PCI_INTC
2673         if (N_RX_COMP_RINGS > 2) {
2674                 /* cas_interruptN(); */
2675         }
2676 #endif
2677 #ifdef USE_PCI_INTD
2678         if (N_RX_COMP_RINGS > 3) {
2679                 /* cas_interruptN(); */
2680         }
2681 #endif
2682 }
2683 #endif
2684
2685 static void cas_tx_timeout(struct net_device *dev)
2686 {
2687         struct cas *cp = netdev_priv(dev);
2688
2689         printk(KERN_ERR "%s: transmit timed out, resetting\n", dev->name);
2690         if (!cp->hw_running) {
2691                 printk("%s: hrm.. hw not running!\n", dev->name);
2692                 return;
2693         }
2694
2695         printk(KERN_ERR "%s: MIF_STATE[%08x]\n",
2696                dev->name, readl(cp->regs + REG_MIF_STATE_MACHINE));
2697
2698         printk(KERN_ERR "%s: MAC_STATE[%08x]\n",
2699                dev->name, readl(cp->regs + REG_MAC_STATE_MACHINE));
2700
2701         printk(KERN_ERR "%s: TX_STATE[%08x:%08x:%08x] "
2702                "FIFO[%08x:%08x:%08x] SM1[%08x] SM2[%08x]\n",
2703                dev->name,
2704                readl(cp->regs + REG_TX_CFG),
2705                readl(cp->regs + REG_MAC_TX_STATUS),
2706                readl(cp->regs + REG_MAC_TX_CFG),
2707                readl(cp->regs + REG_TX_FIFO_PKT_CNT),
2708                readl(cp->regs + REG_TX_FIFO_WRITE_PTR),
2709                readl(cp->regs + REG_TX_FIFO_READ_PTR),
2710                readl(cp->regs + REG_TX_SM_1),
2711                readl(cp->regs + REG_TX_SM_2));
2712
2713         printk(KERN_ERR "%s: RX_STATE[%08x:%08x:%08x]\n",
2714                dev->name,
2715                readl(cp->regs + REG_RX_CFG),
2716                readl(cp->regs + REG_MAC_RX_STATUS),
2717                readl(cp->regs + REG_MAC_RX_CFG));
2718
2719         printk(KERN_ERR "%s: HP_STATE[%08x:%08x:%08x:%08x]\n",
2720                dev->name,
2721                readl(cp->regs + REG_HP_STATE_MACHINE),
2722                readl(cp->regs + REG_HP_STATUS0),
2723                readl(cp->regs + REG_HP_STATUS1),
2724                readl(cp->regs + REG_HP_STATUS2));
2725
2726 #if 1
2727         atomic_inc(&cp->reset_task_pending);
2728         atomic_inc(&cp->reset_task_pending_all);
2729         schedule_work(&cp->reset_task);
2730 #else
2731         atomic_set(&cp->reset_task_pending, CAS_RESET_ALL);
2732         schedule_work(&cp->reset_task);
2733 #endif
2734 }
2735
2736 static inline int cas_intme(int ring, int entry)
2737 {
2738         /* Algorithm: IRQ every 1/2 of descriptors. */
2739         if (!(entry & ((TX_DESC_RINGN_SIZE(ring) >> 1) - 1)))
2740                 return 1;
2741         return 0;
2742 }
2743
2744
2745 static void cas_write_txd(struct cas *cp, int ring, int entry,
2746                           dma_addr_t mapping, int len, u64 ctrl, int last)
2747 {
2748         struct cas_tx_desc *txd = cp->init_txds[ring] + entry;
2749
2750         ctrl |= CAS_BASE(TX_DESC_BUFLEN, len);
2751         if (cas_intme(ring, entry))
2752                 ctrl |= TX_DESC_INTME;
2753         if (last)
2754                 ctrl |= TX_DESC_EOF;
2755         txd->control = cpu_to_le64(ctrl);
2756         txd->buffer = cpu_to_le64(mapping);
2757 }
2758
2759 static inline void *tx_tiny_buf(struct cas *cp, const int ring, 
2760                                 const int entry)
2761 {
2762         return cp->tx_tiny_bufs[ring] + TX_TINY_BUF_LEN*entry;
2763 }
2764
2765 static inline dma_addr_t tx_tiny_map(struct cas *cp, const int ring, 
2766                                      const int entry, const int tentry)
2767 {
2768         cp->tx_tiny_use[ring][tentry].nbufs++;
2769         cp->tx_tiny_use[ring][entry].used = 1;
2770         return cp->tx_tiny_dvma[ring] + TX_TINY_BUF_LEN*entry;
2771 }
2772
2773 static inline int cas_xmit_tx_ringN(struct cas *cp, int ring, 
2774                                     struct sk_buff *skb)
2775 {
2776         struct net_device *dev = cp->dev;
2777         int entry, nr_frags, frag, tabort, tentry;
2778         dma_addr_t mapping;
2779         unsigned long flags;
2780         u64 ctrl;
2781         u32 len;
2782
2783         spin_lock_irqsave(&cp->tx_lock[ring], flags);
2784
2785         /* This is a hard error, log it. */
2786         if (TX_BUFFS_AVAIL(cp, ring) <= 
2787             CAS_TABORT(cp)*(skb_shinfo(skb)->nr_frags + 1)) {
2788                 netif_stop_queue(dev);
2789                 spin_unlock_irqrestore(&cp->tx_lock[ring], flags);
2790                 printk(KERN_ERR PFX "%s: BUG! Tx Ring full when "
2791                        "queue awake!\n", dev->name);
2792                 return 1;
2793         }
2794
2795         ctrl = 0;
2796         if (skb->ip_summed == CHECKSUM_HW) {
2797                 u64 csum_start_off, csum_stuff_off;
2798
2799                 csum_start_off = (u64) (skb->h.raw - skb->data);
2800                 csum_stuff_off = (u64) ((skb->h.raw + skb->csum) - skb->data);
2801
2802                 ctrl =  TX_DESC_CSUM_EN | 
2803                         CAS_BASE(TX_DESC_CSUM_START, csum_start_off) |
2804                         CAS_BASE(TX_DESC_CSUM_STUFF, csum_stuff_off);
2805         }
2806
2807         entry = cp->tx_new[ring];
2808         cp->tx_skbs[ring][entry] = skb;
2809
2810         nr_frags = skb_shinfo(skb)->nr_frags;
2811         len = skb_headlen(skb);
2812         mapping = pci_map_page(cp->pdev, virt_to_page(skb->data),
2813                                offset_in_page(skb->data), len,
2814                                PCI_DMA_TODEVICE);
2815
2816         tentry = entry;
2817         tabort = cas_calc_tabort(cp, (unsigned long) skb->data, len);
2818         if (unlikely(tabort)) {
2819                 /* NOTE: len is always >  tabort */
2820                 cas_write_txd(cp, ring, entry, mapping, len - tabort, 
2821                               ctrl | TX_DESC_SOF, 0);
2822                 entry = TX_DESC_NEXT(ring, entry);
2823
2824                 memcpy(tx_tiny_buf(cp, ring, entry), skb->data + 
2825                        len - tabort, tabort);
2826                 mapping = tx_tiny_map(cp, ring, entry, tentry);
2827                 cas_write_txd(cp, ring, entry, mapping, tabort, ctrl,
2828                               (nr_frags == 0));
2829         } else {
2830                 cas_write_txd(cp, ring, entry, mapping, len, ctrl | 
2831                               TX_DESC_SOF, (nr_frags == 0));
2832         }
2833         entry = TX_DESC_NEXT(ring, entry);
2834
2835         for (frag = 0; frag < nr_frags; frag++) {
2836                 skb_frag_t *fragp = &skb_shinfo(skb)->frags[frag];
2837
2838                 len = fragp->size;
2839                 mapping = pci_map_page(cp->pdev, fragp->page,
2840                                        fragp->page_offset, len,
2841                                        PCI_DMA_TODEVICE);
2842
2843                 tabort = cas_calc_tabort(cp, fragp->page_offset, len);
2844                 if (unlikely(tabort)) {
2845                         void *addr;
2846
2847                         /* NOTE: len is always > tabort */
2848                         cas_write_txd(cp, ring, entry, mapping, len - tabort,
2849                                       ctrl, 0);
2850                         entry = TX_DESC_NEXT(ring, entry);
2851                         
2852                         addr = cas_page_map(fragp->page);
2853                         memcpy(tx_tiny_buf(cp, ring, entry),
2854                                addr + fragp->page_offset + len - tabort, 
2855                                tabort);
2856                         cas_page_unmap(addr);
2857                         mapping = tx_tiny_map(cp, ring, entry, tentry);
2858                         len     = tabort;
2859                 }
2860
2861                 cas_write_txd(cp, ring, entry, mapping, len, ctrl,
2862                               (frag + 1 == nr_frags));
2863                 entry = TX_DESC_NEXT(ring, entry);
2864         }
2865
2866         cp->tx_new[ring] = entry;
2867         if (TX_BUFFS_AVAIL(cp, ring) <= CAS_TABORT(cp)*(MAX_SKB_FRAGS + 1))
2868                 netif_stop_queue(dev);
2869
2870         if (netif_msg_tx_queued(cp))
2871                 printk(KERN_DEBUG "%s: tx[%d] queued, slot %d, skblen %d, "
2872                        "avail %d\n",
2873                        dev->name, ring, entry, skb->len, 
2874                        TX_BUFFS_AVAIL(cp, ring));
2875         writel(entry, cp->regs + REG_TX_KICKN(ring));
2876         spin_unlock_irqrestore(&cp->tx_lock[ring], flags);
2877         return 0;
2878
2879
2880 static int cas_start_xmit(struct sk_buff *skb, struct net_device *dev)
2881 {
2882         struct cas *cp = netdev_priv(dev);
2883
2884         /* this is only used as a load-balancing hint, so it doesn't
2885          * need to be SMP safe
2886          */
2887         static int ring; 
2888
2889         skb = skb_padto(skb, cp->min_frame_size);
2890         if (!skb)
2891                 return 0;
2892
2893         /* XXX: we need some higher-level QoS hooks to steer packets to
2894          *      individual queues.
2895          */
2896         if (cas_xmit_tx_ringN(cp, ring++ & N_TX_RINGS_MASK, skb))
2897                 return 1;
2898         dev->trans_start = jiffies;
2899         return 0;
2900 }
2901
2902 static void cas_init_tx_dma(struct cas *cp)
2903 {
2904         u64 desc_dma = cp->block_dvma;
2905         unsigned long off;
2906         u32 val;
2907         int i;
2908
2909         /* set up tx completion writeback registers. must be 8-byte aligned */
2910 #ifdef USE_TX_COMPWB
2911         off = offsetof(struct cas_init_block, tx_compwb);
2912         writel((desc_dma + off) >> 32, cp->regs + REG_TX_COMPWB_DB_HI);
2913         writel((desc_dma + off) & 0xffffffff, cp->regs + REG_TX_COMPWB_DB_LOW);
2914 #endif
2915
2916         /* enable completion writebacks, enable paced mode,
2917          * disable read pipe, and disable pre-interrupt compwbs
2918          */
2919         val =   TX_CFG_COMPWB_Q1 | TX_CFG_COMPWB_Q2 | 
2920                 TX_CFG_COMPWB_Q3 | TX_CFG_COMPWB_Q4 |
2921                 TX_CFG_DMA_RDPIPE_DIS | TX_CFG_PACED_MODE | 
2922                 TX_CFG_INTR_COMPWB_DIS;
2923
2924         /* write out tx ring info and tx desc bases */
2925         for (i = 0; i < MAX_TX_RINGS; i++) {
2926                 off = (unsigned long) cp->init_txds[i] - 
2927                         (unsigned long) cp->init_block;
2928
2929                 val |= CAS_TX_RINGN_BASE(i);
2930                 writel((desc_dma + off) >> 32, cp->regs + REG_TX_DBN_HI(i));
2931                 writel((desc_dma + off) & 0xffffffff, cp->regs +
2932                        REG_TX_DBN_LOW(i));
2933                 /* don't zero out the kick register here as the system
2934                  * will wedge
2935                  */
2936         }
2937         writel(val, cp->regs + REG_TX_CFG);
2938
2939         /* program max burst sizes. these numbers should be different
2940          * if doing QoS.
2941          */
2942 #ifdef USE_QOS
2943         writel(0x800, cp->regs + REG_TX_MAXBURST_0);
2944         writel(0x1600, cp->regs + REG_TX_MAXBURST_1);
2945         writel(0x2400, cp->regs + REG_TX_MAXBURST_2);
2946         writel(0x4800, cp->regs + REG_TX_MAXBURST_3);
2947 #else
2948         writel(0x800, cp->regs + REG_TX_MAXBURST_0);
2949         writel(0x800, cp->regs + REG_TX_MAXBURST_1);
2950         writel(0x800, cp->regs + REG_TX_MAXBURST_2);
2951         writel(0x800, cp->regs + REG_TX_MAXBURST_3);
2952 #endif
2953 }
2954
2955 /* Must be invoked under cp->lock. */
2956 static inline void cas_init_dma(struct cas *cp)
2957 {
2958         cas_init_tx_dma(cp);
2959         cas_init_rx_dma(cp);
2960 }
2961
2962 /* Must be invoked under cp->lock. */
2963 static u32 cas_setup_multicast(struct cas *cp)
2964 {
2965         u32 rxcfg = 0;
2966         int i;
2967         
2968         if (cp->dev->flags & IFF_PROMISC) {
2969                 rxcfg |= MAC_RX_CFG_PROMISC_EN;
2970
2971         } else if (cp->dev->flags & IFF_ALLMULTI) {
2972                 for (i=0; i < 16; i++)
2973                         writel(0xFFFF, cp->regs + REG_MAC_HASH_TABLEN(i));
2974                 rxcfg |= MAC_RX_CFG_HASH_FILTER_EN;
2975
2976         } else {
2977                 u16 hash_table[16];
2978                 u32 crc;
2979                 struct dev_mc_list *dmi = cp->dev->mc_list;
2980                 int i;
2981
2982                 /* use the alternate mac address registers for the
2983                  * first 15 multicast addresses
2984                  */
2985                 for (i = 1; i <= CAS_MC_EXACT_MATCH_SIZE; i++) {
2986                         if (!dmi) {
2987                                 writel(0x0, cp->regs + REG_MAC_ADDRN(i*3 + 0));
2988                                 writel(0x0, cp->regs + REG_MAC_ADDRN(i*3 + 1));
2989                                 writel(0x0, cp->regs + REG_MAC_ADDRN(i*3 + 2));
2990                                 continue;
2991                         }
2992                         writel((dmi->dmi_addr[4] << 8) | dmi->dmi_addr[5], 
2993                                cp->regs + REG_MAC_ADDRN(i*3 + 0));
2994                         writel((dmi->dmi_addr[2] << 8) | dmi->dmi_addr[3], 
2995                                cp->regs + REG_MAC_ADDRN(i*3 + 1));
2996                         writel((dmi->dmi_addr[0] << 8) | dmi->dmi_addr[1], 
2997                                cp->regs + REG_MAC_ADDRN(i*3 + 2));
2998                         dmi = dmi->next;
2999                 }
3000
3001                 /* use hw hash table for the next series of 
3002                  * multicast addresses
3003                  */
3004                 memset(hash_table, 0, sizeof(hash_table));
3005                 while (dmi) {
3006                         crc = ether_crc_le(ETH_ALEN, dmi->dmi_addr);
3007                         crc >>= 24;
3008                         hash_table[crc >> 4] |= 1 << (15 - (crc & 0xf));
3009                         dmi = dmi->next;
3010                 }
3011                 for (i=0; i < 16; i++)
3012                         writel(hash_table[i], cp->regs + 
3013                                REG_MAC_HASH_TABLEN(i));
3014                 rxcfg |= MAC_RX_CFG_HASH_FILTER_EN;
3015         }
3016
3017         return rxcfg;
3018 }
3019
3020 /* must be invoked under cp->stat_lock[N_TX_RINGS] */
3021 static void cas_clear_mac_err(struct cas *cp)
3022 {
3023         writel(0, cp->regs + REG_MAC_COLL_NORMAL);
3024         writel(0, cp->regs + REG_MAC_COLL_FIRST);
3025         writel(0, cp->regs + REG_MAC_COLL_EXCESS);
3026         writel(0, cp->regs + REG_MAC_COLL_LATE);
3027         writel(0, cp->regs + REG_MAC_TIMER_DEFER);
3028         writel(0, cp->regs + REG_MAC_ATTEMPTS_PEAK);
3029         writel(0, cp->regs + REG_MAC_RECV_FRAME);
3030         writel(0, cp->regs + REG_MAC_LEN_ERR);
3031         writel(0, cp->regs + REG_MAC_ALIGN_ERR);
3032         writel(0, cp->regs + REG_MAC_FCS_ERR);
3033         writel(0, cp->regs + REG_MAC_RX_CODE_ERR);
3034 }
3035
3036
3037 static void cas_mac_reset(struct cas *cp)
3038 {
3039         int i;
3040
3041         /* do both TX and RX reset */
3042         writel(0x1, cp->regs + REG_MAC_TX_RESET);
3043         writel(0x1, cp->regs + REG_MAC_RX_RESET);
3044
3045         /* wait for TX */
3046         i = STOP_TRIES;
3047         while (i-- > 0) {
3048                 if (readl(cp->regs + REG_MAC_TX_RESET) == 0)
3049                         break;
3050                 udelay(10);
3051         }
3052
3053         /* wait for RX */
3054         i = STOP_TRIES;
3055         while (i-- > 0) {
3056                 if (readl(cp->regs + REG_MAC_RX_RESET) == 0)
3057                         break;
3058                 udelay(10);
3059         }
3060
3061         if (readl(cp->regs + REG_MAC_TX_RESET) |
3062             readl(cp->regs + REG_MAC_RX_RESET))
3063                 printk(KERN_ERR "%s: mac tx[%d]/rx[%d] reset failed [%08x]\n",
3064                        cp->dev->name, readl(cp->regs + REG_MAC_TX_RESET),
3065                        readl(cp->regs + REG_MAC_RX_RESET),
3066                        readl(cp->regs + REG_MAC_STATE_MACHINE));
3067 }
3068
3069
3070 /* Must be invoked under cp->lock. */
3071 static void cas_init_mac(struct cas *cp)
3072 {
3073         unsigned char *e = &cp->dev->dev_addr[0];
3074         int i;
3075 #ifdef CONFIG_CASSINI_MULTICAST_REG_WRITE
3076         u32 rxcfg;
3077 #endif
3078         cas_mac_reset(cp);
3079
3080         /* setup core arbitration weight register */
3081         writel(CAWR_RR_DIS, cp->regs + REG_CAWR);
3082
3083         /* XXX Use pci_dma_burst_advice() */
3084 #if !defined(CONFIG_SPARC64) && !defined(CONFIG_ALPHA)
3085         /* set the infinite burst register for chips that don't have
3086          * pci issues.
3087          */
3088         if ((cp->cas_flags & CAS_FLAG_TARGET_ABORT) == 0)
3089                 writel(INF_BURST_EN, cp->regs + REG_INF_BURST);
3090 #endif
3091
3092         writel(0x1BF0, cp->regs + REG_MAC_SEND_PAUSE);
3093
3094         writel(0x00, cp->regs + REG_MAC_IPG0);
3095         writel(0x08, cp->regs + REG_MAC_IPG1);
3096         writel(0x04, cp->regs + REG_MAC_IPG2);
3097         
3098         /* change later for 802.3z */
3099         writel(0x40, cp->regs + REG_MAC_SLOT_TIME); 
3100
3101         /* min frame + FCS */
3102         writel(ETH_ZLEN + 4, cp->regs + REG_MAC_FRAMESIZE_MIN);
3103
3104         /* Ethernet payload + header + FCS + optional VLAN tag. NOTE: we
3105          * specify the maximum frame size to prevent RX tag errors on 
3106          * oversized frames.
3107          */
3108         writel(CAS_BASE(MAC_FRAMESIZE_MAX_BURST, 0x2000) |
3109                CAS_BASE(MAC_FRAMESIZE_MAX_FRAME, 
3110                         (CAS_MAX_MTU + ETH_HLEN + 4 + 4)), 
3111                cp->regs + REG_MAC_FRAMESIZE_MAX);
3112
3113         /* NOTE: crc_size is used as a surrogate for half-duplex. 
3114          * workaround saturn half-duplex issue by increasing preamble
3115          * size to 65 bytes.
3116          */
3117         if ((cp->cas_flags & CAS_FLAG_SATURN) && cp->crc_size)
3118                 writel(0x41, cp->regs + REG_MAC_PA_SIZE);
3119         else
3120                 writel(0x07, cp->regs + REG_MAC_PA_SIZE);
3121         writel(0x04, cp->regs + REG_MAC_JAM_SIZE);
3122         writel(0x10, cp->regs + REG_MAC_ATTEMPT_LIMIT);
3123         writel(0x8808, cp->regs + REG_MAC_CTRL_TYPE);
3124
3125         writel((e[5] | (e[4] << 8)) & 0x3ff, cp->regs + REG_MAC_RANDOM_SEED);
3126
3127         writel(0, cp->regs + REG_MAC_ADDR_FILTER0);
3128         writel(0, cp->regs + REG_MAC_ADDR_FILTER1);
3129         writel(0, cp->regs + REG_MAC_ADDR_FILTER2);
3130         writel(0, cp->regs + REG_MAC_ADDR_FILTER2_1_MASK);
3131         writel(0, cp->regs + REG_MAC_ADDR_FILTER0_MASK);
3132
3133         /* setup mac address in perfect filter array */
3134         for (i = 0; i < 45; i++)
3135                 writel(0x0, cp->regs + REG_MAC_ADDRN(i));
3136
3137         writel((e[4] << 8) | e[5], cp->regs + REG_MAC_ADDRN(0));
3138         writel((e[2] << 8) | e[3], cp->regs + REG_MAC_ADDRN(1));
3139         writel((e[0] << 8) | e[1], cp->regs + REG_MAC_ADDRN(2));
3140
3141         writel(0x0001, cp->regs + REG_MAC_ADDRN(42));
3142         writel(0xc200, cp->regs + REG_MAC_ADDRN(43));
3143         writel(0x0180, cp->regs + REG_MAC_ADDRN(44));
3144
3145 #ifndef CONFIG_CASSINI_MULTICAST_REG_WRITE
3146         cp->mac_rx_cfg = cas_setup_multicast(cp);
3147 #else
3148         /* WTZ: Do what Adrian did in cas_set_multicast. Doing
3149          * a writel does not seem to be necessary because Cassini
3150          * seems to preserve the configuration when we do the reset.
3151          * If the chip is in trouble, though, it is not clear if we
3152          * can really count on this behavior. cas_set_multicast uses
3153          * spin_lock_irqsave, but we are called only in cas_init_hw and
3154          * cas_init_hw is protected by cas_lock_all, which calls
3155          * spin_lock_irq (so it doesn't need to save the flags, and
3156          * we should be OK for the writel, as that is the only 
3157          * difference).
3158          */
3159         cp->mac_rx_cfg = rxcfg = cas_setup_multicast(cp);
3160         writel(rxcfg, cp->regs + REG_MAC_RX_CFG);
3161 #endif
3162         spin_lock(&cp->stat_lock[N_TX_RINGS]);
3163         cas_clear_mac_err(cp);
3164         spin_unlock(&cp->stat_lock[N_TX_RINGS]);
3165
3166         /* Setup MAC interrupts.  We want to get all of the interesting
3167          * counter expiration events, but we do not want to hear about
3168          * normal rx/tx as the DMA engine tells us that.
3169          */
3170         writel(MAC_TX_FRAME_XMIT, cp->regs + REG_MAC_TX_MASK);
3171         writel(MAC_RX_FRAME_RECV, cp->regs + REG_MAC_RX_MASK);
3172
3173         /* Don't enable even the PAUSE interrupts for now, we
3174          * make no use of those events other than to record them.
3175          */
3176         writel(0xffffffff, cp->regs + REG_MAC_CTRL_MASK);
3177 }
3178
3179 /* Must be invoked under cp->lock. */
3180 static void cas_init_pause_thresholds(struct cas *cp)
3181 {
3182         /* Calculate pause thresholds.  Setting the OFF threshold to the
3183          * full RX fifo size effectively disables PAUSE generation
3184          */
3185         if (cp->rx_fifo_size <= (2 * 1024)) {
3186                 cp->rx_pause_off = cp->rx_pause_on = cp->rx_fifo_size;
3187         } else {
3188                 int max_frame = (cp->dev->mtu + ETH_HLEN + 4 + 4 + 64) & ~63;
3189                 if (max_frame * 3 > cp->rx_fifo_size) {
3190                         cp->rx_pause_off = 7104;
3191                         cp->rx_pause_on  = 960;
3192                 } else {
3193                         int off = (cp->rx_fifo_size - (max_frame * 2));
3194                         int on = off - max_frame;
3195                         cp->rx_pause_off = off;
3196                         cp->rx_pause_on = on;
3197                 }
3198         }
3199 }
3200
3201 static int cas_vpd_match(const void __iomem *p, const char *str)
3202 {
3203         int len = strlen(str) + 1;
3204         int i;
3205         
3206         for (i = 0; i < len; i++) {
3207                 if (readb(p + i) != str[i])
3208                         return 0;
3209         }
3210         return 1;
3211 }
3212
3213
3214 /* get the mac address by reading the vpd information in the rom.
3215  * also get the phy type and determine if there's an entropy generator.
3216  * NOTE: this is a bit convoluted for the following reasons:
3217  *  1) vpd info has order-dependent mac addresses for multinic cards
3218  *  2) the only way to determine the nic order is to use the slot
3219  *     number.
3220  *  3) fiber cards don't have bridges, so their slot numbers don't
3221  *     mean anything.
3222  *  4) we don't actually know we have a fiber card until after 
3223  *     the mac addresses are parsed.
3224  */
3225 static int cas_get_vpd_info(struct cas *cp, unsigned char *dev_addr,
3226                             const int offset)
3227 {
3228         void __iomem *p = cp->regs + REG_EXPANSION_ROM_RUN_START;
3229         void __iomem *base, *kstart;
3230         int i, len;
3231         int found = 0;
3232 #define VPD_FOUND_MAC        0x01
3233 #define VPD_FOUND_PHY        0x02
3234
3235         int phy_type = CAS_PHY_MII_MDIO0; /* default phy type */
3236         int mac_off  = 0;
3237
3238         /* give us access to the PROM */
3239         writel(BIM_LOCAL_DEV_PROM | BIM_LOCAL_DEV_PAD,
3240                cp->regs + REG_BIM_LOCAL_DEV_EN);
3241
3242         /* check for an expansion rom */
3243         if (readb(p) != 0x55 || readb(p + 1) != 0xaa)
3244                 goto use_random_mac_addr;
3245
3246         /* search for beginning of vpd */
3247         base = 0;
3248         for (i = 2; i < EXPANSION_ROM_SIZE; i++) {
3249                 /* check for PCIR */
3250                 if ((readb(p + i + 0) == 0x50) &&
3251                     (readb(p + i + 1) == 0x43) &&
3252                     (readb(p + i + 2) == 0x49) &&
3253                     (readb(p + i + 3) == 0x52)) {
3254                         base = p + (readb(p + i + 8) | 
3255                                     (readb(p + i + 9) << 8));
3256                         break;
3257                 }               
3258         }
3259
3260         if (!base || (readb(base) != 0x82))
3261                 goto use_random_mac_addr;
3262         
3263         i = (readb(base + 1) | (readb(base + 2) << 8)) + 3;
3264         while (i < EXPANSION_ROM_SIZE) {
3265                 if (readb(base + i) != 0x90) /* no vpd found */
3266                         goto use_random_mac_addr;
3267
3268                 /* found a vpd field */
3269                 len = readb(base + i + 1) | (readb(base + i + 2) << 8);
3270
3271                 /* extract keywords */
3272                 kstart = base + i + 3;
3273                 p = kstart;
3274                 while ((p - kstart) < len) {
3275                         int klen = readb(p + 2);
3276                         int j;
3277                         char type;
3278
3279                         p += 3;
3280                         
3281                         /* look for the following things:
3282                          * -- correct length == 29
3283                          * 3 (type) + 2 (size) + 
3284                          * 18 (strlen("local-mac-address") + 1) + 
3285                          * 6 (mac addr) 
3286                          * -- VPD Instance 'I'
3287                          * -- VPD Type Bytes 'B'
3288                          * -- VPD data length == 6
3289                          * -- property string == local-mac-address
3290                          * 
3291                          * -- correct length == 24
3292                          * 3 (type) + 2 (size) + 
3293                          * 12 (strlen("entropy-dev") + 1) + 
3294                          * 7 (strlen("vms110") + 1)
3295                          * -- VPD Instance 'I'
3296                          * -- VPD Type String 'B'
3297                          * -- VPD data length == 7
3298                          * -- property string == entropy-dev
3299                          *
3300                          * -- correct length == 18
3301                          * 3 (type) + 2 (size) + 
3302                          * 9 (strlen("phy-type") + 1) + 
3303                          * 4 (strlen("pcs") + 1)
3304                          * -- VPD Instance 'I'
3305                          * -- VPD Type String 'S'
3306                          * -- VPD data length == 4
3307                          * -- property string == phy-type
3308                          * 
3309                          * -- correct length == 23
3310                          * 3 (type) + 2 (size) + 
3311                          * 14 (strlen("phy-interface") + 1) + 
3312                          * 4 (strlen("pcs") + 1)
3313                          * -- VPD Instance 'I'
3314                          * -- VPD Type String 'S'
3315                          * -- VPD data length == 4
3316                          * -- property string == phy-interface
3317                          */
3318                         if (readb(p) != 'I')
3319                                 goto next;
3320
3321                         /* finally, check string and length */
3322                         type = readb(p + 3);
3323                         if (type == 'B') {
3324                                 if ((klen == 29) && readb(p + 4) == 6 &&
3325                                     cas_vpd_match(p + 5, 
3326                                                   "local-mac-address")) {
3327                                         if (mac_off++ > offset) 
3328                                                 goto next;
3329
3330                                         /* set mac address */
3331                                         for (j = 0; j < 6; j++) 
3332                                                 dev_addr[j] = 
3333                                                         readb(p + 23 + j);
3334                                         goto found_mac;
3335                                 }
3336                         }
3337
3338                         if (type != 'S')
3339                                 goto next;
3340
3341 #ifdef USE_ENTROPY_DEV
3342                         if ((klen == 24) && 
3343                             cas_vpd_match(p + 5, "entropy-dev") &&
3344                             cas_vpd_match(p + 17, "vms110")) {
3345                                 cp->cas_flags |= CAS_FLAG_ENTROPY_DEV;
3346                                 goto next;
3347                         }
3348 #endif
3349
3350                         if (found & VPD_FOUND_PHY)
3351                                 goto next;
3352
3353                         if ((klen == 18) && readb(p + 4) == 4 &&
3354                             cas_vpd_match(p + 5, "phy-type")) {
3355                                 if (cas_vpd_match(p + 14, "pcs")) {
3356                                         phy_type = CAS_PHY_SERDES;
3357                                         goto found_phy;
3358                                 }
3359                         }
3360                         
3361                         if ((klen == 23) && readb(p + 4) == 4 &&
3362                             cas_vpd_match(p + 5, "phy-interface")) {
3363                                 if (cas_vpd_match(p + 19, "pcs")) {
3364                                         phy_type = CAS_PHY_SERDES;
3365                                         goto found_phy;
3366                                 }
3367                         }
3368 found_mac:
3369                         found |= VPD_FOUND_MAC;
3370                         goto next;
3371
3372 found_phy:
3373                         found |= VPD_FOUND_PHY;
3374
3375 next:
3376                         p += klen;
3377                 }
3378                 i += len + 3;
3379         }
3380
3381 use_random_mac_addr:
3382         if (found & VPD_FOUND_MAC)
3383                 goto done;
3384
3385         /* Sun MAC prefix then 3 random bytes. */
3386         printk(PFX "MAC address not found in ROM VPD\n");
3387         dev_addr[0] = 0x08;
3388         dev_addr[1] = 0x00;
3389         dev_addr[2] = 0x20;
3390         get_random_bytes(dev_addr + 3, 3);
3391
3392 done:
3393         writel(0, cp->regs + REG_BIM_LOCAL_DEV_EN);
3394         return phy_type;
3395 }
3396
3397 /* check pci invariants */
3398 static void cas_check_pci_invariants(struct cas *cp)
3399 {
3400         struct pci_dev *pdev = cp->pdev;
3401         u8 rev;
3402
3403         cp->cas_flags = 0;
3404         pci_read_config_byte(pdev, PCI_REVISION_ID, &rev);
3405         if ((pdev->vendor == PCI_VENDOR_ID_SUN) &&
3406             (pdev->device == PCI_DEVICE_ID_SUN_CASSINI)) {
3407                 if (rev >= CAS_ID_REVPLUS)
3408                         cp->cas_flags |= CAS_FLAG_REG_PLUS;
3409                 if (rev < CAS_ID_REVPLUS02u)
3410                         cp->cas_flags |= CAS_FLAG_TARGET_ABORT;
3411
3412                 /* Original Cassini supports HW CSUM, but it's not
3413                  * enabled by default as it can trigger TX hangs.
3414                  */
3415                 if (rev < CAS_ID_REV2)
3416                         cp->cas_flags |= CAS_FLAG_NO_HW_CSUM;
3417         } else {
3418                 /* Only sun has original cassini chips.  */
3419                 cp->cas_flags |= CAS_FLAG_REG_PLUS;
3420
3421                 /* We use a flag because the same phy might be externally
3422                  * connected.
3423                  */
3424                 if ((pdev->vendor == PCI_VENDOR_ID_NS) &&
3425                     (pdev->device == PCI_DEVICE_ID_NS_SATURN))
3426                         cp->cas_flags |= CAS_FLAG_SATURN;
3427         }
3428 }
3429
3430
3431 static int cas_check_invariants(struct cas *cp)
3432 {
3433         struct pci_dev *pdev = cp->pdev;
3434         u32 cfg;
3435         int i;
3436
3437         /* get page size for rx buffers. */
3438         cp->page_order = 0; 
3439 #ifdef USE_PAGE_ORDER
3440         if (PAGE_SHIFT < CAS_JUMBO_PAGE_SHIFT) {
3441                 /* see if we can allocate larger pages */
3442                 struct page *page = alloc_pages(GFP_ATOMIC, 
3443                                                 CAS_JUMBO_PAGE_SHIFT - 
3444                                                 PAGE_SHIFT);
3445                 if (page) {
3446                         __free_pages(page, CAS_JUMBO_PAGE_SHIFT - PAGE_SHIFT);
3447                         cp->page_order = CAS_JUMBO_PAGE_SHIFT - PAGE_SHIFT;
3448                 } else {
3449                         printk(PFX "MTU limited to %d bytes\n", CAS_MAX_MTU);
3450                 }
3451         }
3452 #endif
3453         cp->page_size = (PAGE_SIZE << cp->page_order);
3454
3455         /* Fetch the FIFO configurations. */
3456         cp->tx_fifo_size = readl(cp->regs + REG_TX_FIFO_SIZE) * 64;
3457         cp->rx_fifo_size = RX_FIFO_SIZE;
3458
3459         /* finish phy determination. MDIO1 takes precedence over MDIO0 if 
3460          * they're both connected.
3461          */
3462         cp->phy_type = cas_get_vpd_info(cp, cp->dev->dev_addr, 
3463                                         PCI_SLOT(pdev->devfn));
3464         if (cp->phy_type & CAS_PHY_SERDES) {
3465                 cp->cas_flags |= CAS_FLAG_1000MB_CAP;
3466                 return 0; /* no more checking needed */
3467         } 
3468
3469         /* MII */
3470         cfg = readl(cp->regs + REG_MIF_CFG);
3471         if (cfg & MIF_CFG_MDIO_1) {
3472                 cp->phy_type = CAS_PHY_MII_MDIO1;
3473         } else if (cfg & MIF_CFG_MDIO_0) {
3474                 cp->phy_type = CAS_PHY_MII_MDIO0;
3475         }
3476
3477         cas_mif_poll(cp, 0);
3478         writel(PCS_DATAPATH_MODE_MII, cp->regs + REG_PCS_DATAPATH_MODE);
3479
3480         for (i = 0; i < 32; i++) {
3481                 u32 phy_id;
3482                 int j;
3483
3484                 for (j = 0; j < 3; j++) {
3485                         cp->phy_addr = i;
3486                         phy_id = cas_phy_read(cp, MII_PHYSID1) << 16;
3487                         phy_id |= cas_phy_read(cp, MII_PHYSID2);
3488                         if (phy_id && (phy_id != 0xFFFFFFFF)) {
3489                                 cp->phy_id = phy_id;
3490                                 goto done;
3491                         }
3492                 }
3493         }
3494         printk(KERN_ERR PFX "MII phy did not respond [%08x]\n",
3495                readl(cp->regs + REG_MIF_STATE_MACHINE));
3496         return -1;
3497
3498 done:
3499         /* see if we can do gigabit */
3500         cfg = cas_phy_read(cp, MII_BMSR);
3501         if ((cfg & CAS_BMSR_1000_EXTEND) && 
3502             cas_phy_read(cp, CAS_MII_1000_EXTEND))
3503                 cp->cas_flags |= CAS_FLAG_1000MB_CAP;
3504         return 0;
3505 }
3506
3507 /* Must be invoked under cp->lock. */
3508 static inline void cas_start_dma(struct cas *cp)
3509 {
3510         int i;
3511         u32 val;
3512         int txfailed = 0;
3513         
3514         /* enable dma */
3515         val = readl(cp->regs + REG_TX_CFG) | TX_CFG_DMA_EN;
3516         writel(val, cp->regs + REG_TX_CFG);
3517         val = readl(cp->regs + REG_RX_CFG) | RX_CFG_DMA_EN;
3518         writel(val, cp->regs + REG_RX_CFG);
3519
3520         /* enable the mac */
3521         val = readl(cp->regs + REG_MAC_TX_CFG) | MAC_TX_CFG_EN;
3522         writel(val, cp->regs + REG_MAC_TX_CFG);
3523         val = readl(cp->regs + REG_MAC_RX_CFG) | MAC_RX_CFG_EN;
3524         writel(val, cp->regs + REG_MAC_RX_CFG);
3525
3526         i = STOP_TRIES;
3527         while (i-- > 0) {
3528                 val = readl(cp->regs + REG_MAC_TX_CFG);
3529                 if ((val & MAC_TX_CFG_EN))
3530                         break;
3531                 udelay(10);
3532         }
3533         if (i < 0) txfailed = 1;
3534         i = STOP_TRIES;
3535         while (i-- > 0) {
3536                 val = readl(cp->regs + REG_MAC_RX_CFG);
3537                 if ((val & MAC_RX_CFG_EN)) {
3538                         if (txfailed) {
3539                           printk(KERN_ERR 
3540                                  "%s: enabling mac failed [tx:%08x:%08x].\n", 
3541                                  cp->dev->name,
3542                                  readl(cp->regs + REG_MIF_STATE_MACHINE),
3543                                  readl(cp->regs + REG_MAC_STATE_MACHINE));
3544                         }
3545                         goto enable_rx_done;
3546                 }
3547                 udelay(10);
3548         }
3549         printk(KERN_ERR "%s: enabling mac failed [%s:%08x:%08x].\n", 
3550                cp->dev->name,
3551                (txfailed? "tx,rx":"rx"),
3552                readl(cp->regs + REG_MIF_STATE_MACHINE),
3553                readl(cp->regs + REG_MAC_STATE_MACHINE));
3554
3555 enable_rx_done:
3556         cas_unmask_intr(cp); /* enable interrupts */
3557         writel(RX_DESC_RINGN_SIZE(0) - 4, cp->regs + REG_RX_KICK);
3558         writel(0, cp->regs + REG_RX_COMP_TAIL);
3559
3560         if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
3561                 if (N_RX_DESC_RINGS > 1) 
3562                         writel(RX_DESC_RINGN_SIZE(1) - 4, 
3563                                cp->regs + REG_PLUS_RX_KICK1);
3564
3565                 for (i = 1; i < N_RX_COMP_RINGS; i++) 
3566                         writel(0, cp->regs + REG_PLUS_RX_COMPN_TAIL(i));
3567         }
3568 }
3569
3570 /* Must be invoked under cp->lock. */
3571 static void cas_read_pcs_link_mode(struct cas *cp, int *fd, int *spd,
3572                                    int *pause)
3573 {
3574         u32 val = readl(cp->regs + REG_PCS_MII_LPA);
3575         *fd     = (val & PCS_MII_LPA_FD) ? 1 : 0;
3576         *pause  = (val & PCS_MII_LPA_SYM_PAUSE) ? 0x01 : 0x00;
3577         if (val & PCS_MII_LPA_ASYM_PAUSE)
3578                 *pause |= 0x10;
3579         *spd = 1000;
3580 }
3581
3582 /* Must be invoked under cp->lock. */
3583 static void cas_read_mii_link_mode(struct cas *cp, int *fd, int *spd,
3584                                    int *pause)
3585 {
3586         u32 val;
3587
3588         *fd = 0;
3589         *spd = 10;
3590         *pause = 0;
3591         
3592         /* use GMII registers */
3593         val = cas_phy_read(cp, MII_LPA);
3594         if (val & CAS_LPA_PAUSE)
3595                 *pause = 0x01;
3596
3597         if (val & CAS_LPA_ASYM_PAUSE)
3598                 *pause |= 0x10;
3599
3600         if (val & LPA_DUPLEX)
3601                 *fd = 1;
3602         if (val & LPA_100)
3603                 *spd = 100;
3604
3605         if (cp->cas_flags & CAS_FLAG_1000MB_CAP) {
3606                 val = cas_phy_read(cp, CAS_MII_1000_STATUS);
3607                 if (val & (CAS_LPA_1000FULL | CAS_LPA_1000HALF))
3608                         *spd = 1000;
3609                 if (val & CAS_LPA_1000FULL)
3610                         *fd = 1;
3611         }
3612 }
3613
3614 /* A link-up condition has occurred, initialize and enable the
3615  * rest of the chip.
3616  *
3617  * Must be invoked under cp->lock.
3618  */
3619 static void cas_set_link_modes(struct cas *cp)
3620 {
3621         u32 val;
3622         int full_duplex, speed, pause;
3623
3624         full_duplex = 0;
3625         speed = 10;
3626         pause = 0;
3627
3628         if (CAS_PHY_MII(cp->phy_type)) {
3629                 cas_mif_poll(cp, 0);
3630                 val = cas_phy_read(cp, MII_BMCR);
3631                 if (val & BMCR_ANENABLE) {
3632                         cas_read_mii_link_mode(cp, &full_duplex, &speed, 
3633                                                &pause);
3634                 } else {
3635                         if (val & BMCR_FULLDPLX)
3636                                 full_duplex = 1;
3637
3638                         if (val & BMCR_SPEED100)
3639                                 speed = 100;
3640                         else if (val & CAS_BMCR_SPEED1000)
3641                                 speed = (cp->cas_flags & CAS_FLAG_1000MB_CAP) ?
3642                                         1000 : 100;
3643                 }
3644                 cas_mif_poll(cp, 1);
3645
3646         } else {
3647                 val = readl(cp->regs + REG_PCS_MII_CTRL);
3648                 cas_read_pcs_link_mode(cp, &full_duplex, &speed, &pause);
3649                 if ((val & PCS_MII_AUTONEG_EN) == 0) {
3650                         if (val & PCS_MII_CTRL_DUPLEX)
3651                                 full_duplex = 1;
3652                 }
3653         }
3654
3655         if (netif_msg_link(cp))
3656                 printk(KERN_INFO "%s: Link up at %d Mbps, %s-duplex.\n",
3657                        cp->dev->name, speed, (full_duplex ? "full" : "half"));
3658
3659         val = MAC_XIF_TX_MII_OUTPUT_EN | MAC_XIF_LINK_LED;
3660         if (CAS_PHY_MII(cp->phy_type)) {
3661                 val |= MAC_XIF_MII_BUFFER_OUTPUT_EN;
3662                 if (!full_duplex)
3663                         val |= MAC_XIF_DISABLE_ECHO;
3664         }
3665         if (full_duplex) 
3666                 val |= MAC_XIF_FDPLX_LED;
3667         if (speed == 1000)
3668                 val |= MAC_XIF_GMII_MODE;
3669         writel(val, cp->regs + REG_MAC_XIF_CFG);
3670
3671         /* deal with carrier and collision detect. */
3672         val = MAC_TX_CFG_IPG_EN;
3673         if (full_duplex) {
3674                 val |= MAC_TX_CFG_IGNORE_CARRIER;
3675                 val |= MAC_TX_CFG_IGNORE_COLL;
3676         } else {
3677 #ifndef USE_CSMA_CD_PROTO
3678                 val |= MAC_TX_CFG_NEVER_GIVE_UP_EN;
3679                 val |= MAC_TX_CFG_NEVER_GIVE_UP_LIM;
3680 #endif
3681         }
3682         /* val now set up for REG_MAC_TX_CFG */
3683
3684         /* If gigabit and half-duplex, enable carrier extension
3685          * mode.  increase slot time to 512 bytes as well. 
3686          * else, disable it and make sure slot time is 64 bytes.
3687          * also activate checksum bug workaround
3688          */
3689         if ((speed == 1000) && !full_duplex) {
3690                 writel(val | MAC_TX_CFG_CARRIER_EXTEND, 
3691                        cp->regs + REG_MAC_TX_CFG);
3692
3693                 val = readl(cp->regs + REG_MAC_RX_CFG);
3694                 val &= ~MAC_RX_CFG_STRIP_FCS; /* checksum workaround */
3695                 writel(val | MAC_RX_CFG_CARRIER_EXTEND, 
3696                        cp->regs + REG_MAC_RX_CFG);
3697
3698                 writel(0x200, cp->regs + REG_MAC_SLOT_TIME);
3699
3700                 cp->crc_size = 4;
3701                 /* minimum size gigabit frame at half duplex */
3702                 cp->min_frame_size = CAS_1000MB_MIN_FRAME;
3703
3704         } else {
3705                 writel(val, cp->regs + REG_MAC_TX_CFG);
3706
3707                 /* checksum bug workaround. don't strip FCS when in 
3708                  * half-duplex mode
3709                  */
3710                 val = readl(cp->regs + REG_MAC_RX_CFG);
3711                 if (full_duplex) {
3712                         val |= MAC_RX_CFG_STRIP_FCS;
3713                         cp->crc_size = 0;
3714                         cp->min_frame_size = CAS_MIN_MTU;
3715                 } else {
3716                         val &= ~MAC_RX_CFG_STRIP_FCS;
3717                         cp->crc_size = 4;
3718                         cp->min_frame_size = CAS_MIN_FRAME;
3719                 }
3720                 writel(val & ~MAC_RX_CFG_CARRIER_EXTEND, 
3721                        cp->regs + REG_MAC_RX_CFG);
3722                 writel(0x40, cp->regs + REG_MAC_SLOT_TIME);
3723         }
3724
3725         if (netif_msg_link(cp)) {
3726                 if (pause & 0x01) {
3727                         printk(KERN_INFO "%s: Pause is enabled "
3728                                "(rxfifo: %d off: %d on: %d)\n",
3729                                cp->dev->name,
3730                                cp->rx_fifo_size,
3731                                cp->rx_pause_off,
3732                                cp->rx_pause_on);
3733                 } else if (pause & 0x10) {
3734                         printk(KERN_INFO "%s: TX pause enabled\n",
3735                                cp->dev->name);
3736                 } else {
3737                         printk(KERN_INFO "%s: Pause is disabled\n",
3738                                cp->dev->name);
3739                 }
3740         }
3741
3742         val = readl(cp->regs + REG_MAC_CTRL_CFG);
3743         val &= ~(MAC_CTRL_CFG_SEND_PAUSE_EN | MAC_CTRL_CFG_RECV_PAUSE_EN);
3744         if (pause) { /* symmetric or asymmetric pause */
3745                 val |= MAC_CTRL_CFG_SEND_PAUSE_EN;
3746                 if (pause & 0x01) { /* symmetric pause */
3747                         val |= MAC_CTRL_CFG_RECV_PAUSE_EN;
3748                 } 
3749         }
3750         writel(val, cp->regs + REG_MAC_CTRL_CFG);
3751         cas_start_dma(cp);
3752 }
3753
3754 /* Must be invoked under cp->lock. */
3755 static void cas_init_hw(struct cas *cp, int restart_link)
3756 {
3757         if (restart_link)
3758                 cas_phy_init(cp);
3759
3760         cas_init_pause_thresholds(cp);
3761         cas_init_mac(cp);
3762         cas_init_dma(cp);
3763
3764         if (restart_link) {
3765                 /* Default aneg parameters */
3766                 cp->timer_ticks = 0;
3767                 cas_begin_auto_negotiation(cp, NULL);
3768         } else if (cp->lstate == link_up) {
3769                 cas_set_link_modes(cp);
3770                 netif_carrier_on(cp->dev);
3771         }
3772 }
3773
3774 /* Must be invoked under cp->lock. on earlier cassini boards,
3775  * SOFT_0 is tied to PCI reset. we use this to force a pci reset,
3776  * let it settle out, and then restore pci state.
3777  */
3778 static void cas_hard_reset(struct cas *cp)
3779 {
3780         writel(BIM_LOCAL_DEV_SOFT_0, cp->regs + REG_BIM_LOCAL_DEV_EN); 
3781         udelay(20);
3782         pci_restore_state(cp->pdev);
3783 }
3784
3785
3786 static void cas_global_reset(struct cas *cp, int blkflag)
3787 {
3788         int limit;
3789
3790         /* issue a global reset. don't use RSTOUT. */
3791         if (blkflag && !CAS_PHY_MII(cp->phy_type)) {
3792                 /* For PCS, when the blkflag is set, we should set the
3793                  * SW_REST_BLOCK_PCS_SLINK bit to prevent the results of
3794                  * the last autonegotiation from being cleared.  We'll
3795                  * need some special handling if the chip is set into a
3796                  * loopback mode.
3797                  */
3798                 writel((SW_RESET_TX | SW_RESET_RX | SW_RESET_BLOCK_PCS_SLINK), 
3799                        cp->regs + REG_SW_RESET);
3800         } else {
3801                 writel(SW_RESET_TX | SW_RESET_RX, cp->regs + REG_SW_RESET);
3802         }
3803
3804         /* need to wait at least 3ms before polling register */
3805         mdelay(3);
3806
3807         limit = STOP_TRIES;
3808         while (limit-- > 0) {
3809                 u32 val = readl(cp->regs + REG_SW_RESET);
3810                 if ((val & (SW_RESET_TX | SW_RESET_RX)) == 0)
3811                         goto done;
3812                 udelay(10);
3813         }
3814         printk(KERN_ERR "%s: sw reset failed.\n", cp->dev->name);
3815
3816 done:
3817         /* enable various BIM interrupts */
3818         writel(BIM_CFG_DPAR_INTR_ENABLE | BIM_CFG_RMA_INTR_ENABLE | 
3819                BIM_CFG_RTA_INTR_ENABLE, cp->regs + REG_BIM_CFG);
3820
3821         /* clear out pci error status mask for handled errors.
3822          * we don't deal with DMA counter overflows as they happen
3823          * all the time.
3824          */
3825         writel(0xFFFFFFFFU & ~(PCI_ERR_BADACK | PCI_ERR_DTRTO | 
3826                                PCI_ERR_OTHER | PCI_ERR_BIM_DMA_WRITE | 
3827                                PCI_ERR_BIM_DMA_READ), cp->regs + 
3828                REG_PCI_ERR_STATUS_MASK);
3829
3830         /* set up for MII by default to address mac rx reset timeout
3831          * issue
3832          */
3833         writel(PCS_DATAPATH_MODE_MII, cp->regs + REG_PCS_DATAPATH_MODE);
3834 }
3835
3836 static void cas_reset(struct cas *cp, int blkflag)
3837 {
3838         u32 val;
3839
3840         cas_mask_intr(cp);
3841         cas_global_reset(cp, blkflag);
3842         cas_mac_reset(cp);
3843         cas_entropy_reset(cp);
3844
3845         /* disable dma engines. */
3846         val = readl(cp->regs + REG_TX_CFG);
3847         val &= ~TX_CFG_DMA_EN;
3848         writel(val, cp->regs + REG_TX_CFG);
3849
3850         val = readl(cp->regs + REG_RX_CFG);
3851         val &= ~RX_CFG_DMA_EN;
3852         writel(val, cp->regs + REG_RX_CFG);
3853
3854         /* program header parser */
3855         if ((cp->cas_flags & CAS_FLAG_TARGET_ABORT) ||
3856             (CAS_HP_ALT_FIRMWARE == cas_prog_null)) {
3857                 cas_load_firmware(cp, CAS_HP_FIRMWARE);
3858         } else {
3859                 cas_load_firmware(cp, CAS_HP_ALT_FIRMWARE);
3860         }
3861
3862         /* clear out error registers */
3863         spin_lock(&cp->stat_lock[N_TX_RINGS]);
3864         cas_clear_mac_err(cp);
3865         spin_unlock(&cp->stat_lock[N_TX_RINGS]);
3866 }
3867
3868 /* Shut down the chip, must be called with pm_sem held.  */
3869 static void cas_shutdown(struct cas *cp)
3870 {
3871         unsigned long flags;
3872
3873         /* Make us not-running to avoid timers respawning */
3874         cp->hw_running = 0;
3875
3876         del_timer_sync(&cp->link_timer);
3877
3878         /* Stop the reset task */
3879 #if 0
3880         while (atomic_read(&cp->reset_task_pending_mtu) ||
3881                atomic_read(&cp->reset_task_pending_spare) ||
3882                atomic_read(&cp->reset_task_pending_all))
3883                 schedule();
3884
3885 #else
3886         while (atomic_read(&cp->reset_task_pending))
3887                 schedule();
3888 #endif  
3889         /* Actually stop the chip */
3890         cas_lock_all_save(cp, flags);
3891         cas_reset(cp, 0);
3892         if (cp->cas_flags & CAS_FLAG_SATURN)
3893                 cas_phy_powerdown(cp);
3894         cas_unlock_all_restore(cp, flags);
3895 }
3896
3897 static int cas_change_mtu(struct net_device *dev, int new_mtu)
3898 {
3899         struct cas *cp = netdev_priv(dev);
3900
3901         if (new_mtu < CAS_MIN_MTU || new_mtu > CAS_MAX_MTU)
3902                 return -EINVAL;
3903
3904         dev->mtu = new_mtu;
3905         if (!netif_running(dev) || !netif_device_present(dev))
3906                 return 0;
3907
3908         /* let the reset task handle it */
3909 #if 1
3910         atomic_inc(&cp->reset_task_pending);
3911         if ((cp->phy_type & CAS_PHY_SERDES)) {
3912                 atomic_inc(&cp->reset_task_pending_all);
3913         } else {
3914                 atomic_inc(&cp->reset_task_pending_mtu);
3915         }
3916         schedule_work(&cp->reset_task);
3917 #else
3918         atomic_set(&cp->reset_task_pending, (cp->phy_type & CAS_PHY_SERDES) ? 
3919                    CAS_RESET_ALL : CAS_RESET_MTU);
3920         printk(KERN_ERR "reset called in cas_change_mtu\n");
3921         schedule_work(&cp->reset_task);
3922 #endif
3923
3924         flush_scheduled_work();
3925         return 0;
3926 }
3927
3928 static void cas_clean_txd(struct cas *cp, int ring)
3929 {
3930         struct cas_tx_desc *txd = cp->init_txds[ring];
3931         struct sk_buff *skb, **skbs = cp->tx_skbs[ring];
3932         u64 daddr, dlen;
3933         int i, size;
3934
3935         size = TX_DESC_RINGN_SIZE(ring);
3936         for (i = 0; i < size; i++) {
3937                 int frag;
3938
3939                 if (skbs[i] == NULL)
3940                         continue;
3941
3942                 skb = skbs[i];
3943                 skbs[i] = NULL;
3944
3945                 for (frag = 0; frag <= skb_shinfo(skb)->nr_frags;  frag++) {
3946                         int ent = i & (size - 1);
3947
3948                         /* first buffer is never a tiny buffer and so
3949                          * needs to be unmapped.
3950                          */
3951                         daddr = le64_to_cpu(txd[ent].buffer);
3952                         dlen  =  CAS_VAL(TX_DESC_BUFLEN, 
3953                                          le64_to_cpu(txd[ent].control));
3954                         pci_unmap_page(cp->pdev, daddr, dlen,
3955                                        PCI_DMA_TODEVICE);
3956
3957                         if (frag != skb_shinfo(skb)->nr_frags) {
3958                                 i++;
3959
3960                                 /* next buffer might by a tiny buffer.
3961                                  * skip past it.
3962                                  */
3963                                 ent = i & (size - 1);
3964                                 if (cp->tx_tiny_use[ring][ent].used)
3965                                         i++;
3966                         }
3967                 }
3968                 dev_kfree_skb_any(skb);
3969         }
3970
3971         /* zero out tiny buf usage */
3972         memset(cp->tx_tiny_use[ring], 0, size*sizeof(*cp->tx_tiny_use[ring]));
3973 }
3974
3975 /* freed on close */
3976 static inline void cas_free_rx_desc(struct cas *cp, int ring)
3977 {
3978         cas_page_t **page = cp->rx_pages[ring];
3979         int i, size;
3980
3981         size = RX_DESC_RINGN_SIZE(ring);
3982         for (i = 0; i < size; i++) {
3983                 if (page[i]) {
3984                         cas_page_free(cp, page[i]);
3985                         page[i] = NULL;
3986                 }
3987         }
3988 }
3989
3990 static void cas_free_rxds(struct cas *cp)
3991 {
3992         int i;
3993
3994         for (i = 0; i < N_RX_DESC_RINGS; i++)
3995                 cas_free_rx_desc(cp, i);
3996 }
3997
3998 /* Must be invoked under cp->lock. */
3999 static void cas_clean_rings(struct cas *cp)
4000 {
4001         int i;
4002
4003         /* need to clean all tx rings */
4004         memset(cp->tx_old, 0, sizeof(*cp->tx_old)*N_TX_RINGS);
4005         memset(cp->tx_new, 0, sizeof(*cp->tx_new)*N_TX_RINGS);
4006         for (i = 0; i < N_TX_RINGS; i++)
4007                 cas_clean_txd(cp, i);
4008
4009         /* zero out init block */
4010         memset(cp->init_block, 0, sizeof(struct cas_init_block));
4011         cas_clean_rxds(cp);
4012         cas_clean_rxcs(cp);
4013 }
4014
4015 /* allocated on open */
4016 static inline int cas_alloc_rx_desc(struct cas *cp, int ring)
4017 {
4018         cas_page_t **page = cp->rx_pages[ring];
4019         int size, i = 0;
4020
4021         size = RX_DESC_RINGN_SIZE(ring);
4022         for (i = 0; i < size; i++) {
4023                 if ((page[i] = cas_page_alloc(cp, GFP_KERNEL)) == NULL) 
4024                         return -1;
4025         }
4026         return 0;
4027 }
4028
4029 static int cas_alloc_rxds(struct cas *cp)
4030 {
4031         int i;
4032
4033         for (i = 0; i < N_RX_DESC_RINGS; i++) {
4034                 if (cas_alloc_rx_desc(cp, i) < 0) {
4035                         cas_free_rxds(cp);
4036                         return -1;
4037                 }
4038         }
4039         return 0;
4040 }
4041
4042 static void cas_reset_task(void *data)
4043 {
4044         struct cas *cp = (struct cas *) data;
4045 #if 0
4046         int pending = atomic_read(&cp->reset_task_pending);
4047 #else
4048         int pending_all = atomic_read(&cp->reset_task_pending_all);
4049         int pending_spare = atomic_read(&cp->reset_task_pending_spare);
4050         int pending_mtu = atomic_read(&cp->reset_task_pending_mtu);
4051
4052         if (pending_all == 0 && pending_spare == 0 && pending_mtu == 0) {
4053                 /* We can have more tasks scheduled than actually
4054                  * needed.
4055                  */
4056                 atomic_dec(&cp->reset_task_pending);
4057                 return;
4058         }
4059 #endif
4060         /* The link went down, we reset the ring, but keep
4061          * DMA stopped. Use this function for reset
4062          * on error as well.
4063          */
4064         if (cp->hw_running) {
4065                 unsigned long flags;
4066
4067                 /* Make sure we don't get interrupts or tx packets */
4068                 netif_device_detach(cp->dev);
4069                 cas_lock_all_save(cp, flags);
4070
4071                 if (cp->opened) {
4072                         /* We call cas_spare_recover when we call cas_open.
4073                          * but we do not initialize the lists cas_spare_recover
4074                          * uses until cas_open is called.
4075                          */
4076                         cas_spare_recover(cp, GFP_ATOMIC);
4077                 }
4078 #if 1
4079                 /* test => only pending_spare set */
4080                 if (!pending_all && !pending_mtu)
4081                         goto done;
4082 #else
4083                 if (pending == CAS_RESET_SPARE)
4084                         goto done;
4085 #endif
4086                 /* when pending == CAS_RESET_ALL, the following
4087                  * call to cas_init_hw will restart auto negotiation.
4088                  * Setting the second argument of cas_reset to
4089                  * !(pending == CAS_RESET_ALL) will set this argument
4090                  * to 1 (avoiding reinitializing the PHY for the normal 
4091                  * PCS case) when auto negotiation is not restarted.
4092                  */
4093 #if 1
4094                 cas_reset(cp, !(pending_all > 0));
4095                 if (cp->opened)
4096                         cas_clean_rings(cp);
4097                 cas_init_hw(cp, (pending_all > 0));
4098 #else
4099                 cas_reset(cp, !(pending == CAS_RESET_ALL));
4100                 if (cp->opened)
4101                         cas_clean_rings(cp);
4102                 cas_init_hw(cp, pending == CAS_RESET_ALL);
4103 #endif
4104
4105 done:
4106                 cas_unlock_all_restore(cp, flags);
4107                 netif_device_attach(cp->dev);
4108         }
4109 #if 1
4110         atomic_sub(pending_all, &cp->reset_task_pending_all);
4111         atomic_sub(pending_spare, &cp->reset_task_pending_spare);
4112         atomic_sub(pending_mtu, &cp->reset_task_pending_mtu);
4113         atomic_dec(&cp->reset_task_pending);
4114 #else
4115         atomic_set(&cp->reset_task_pending, 0);
4116 #endif
4117 }
4118
4119 static void cas_link_timer(unsigned long data)
4120 {
4121         struct cas *cp = (struct cas *) data;
4122         int mask, pending = 0, reset = 0;
4123         unsigned long flags;
4124
4125         if (link_transition_timeout != 0 &&
4126             cp->link_transition_jiffies_valid &&
4127             ((jiffies - cp->link_transition_jiffies) > 
4128               (link_transition_timeout))) {
4129                 /* One-second counter so link-down workaround doesn't 
4130                  * cause resets to occur so fast as to fool the switch
4131                  * into thinking the link is down.
4132                  */
4133                 cp->link_transition_jiffies_valid = 0;
4134         }
4135
4136         if (!cp->hw_running)
4137                 return;
4138
4139         spin_lock_irqsave(&cp->lock, flags);
4140         cas_lock_tx(cp);
4141         cas_entropy_gather(cp);
4142
4143         /* If the link task is still pending, we just
4144          * reschedule the link timer
4145          */
4146 #if 1
4147         if (atomic_read(&cp->reset_task_pending_all) ||
4148             atomic_read(&cp->reset_task_pending_spare) ||
4149             atomic_read(&cp->reset_task_pending_mtu)) 
4150                 goto done;
4151 #else
4152         if (atomic_read(&cp->reset_task_pending)) 
4153                 goto done;
4154 #endif
4155
4156         /* check for rx cleaning */
4157         if ((mask = (cp->cas_flags & CAS_FLAG_RXD_POST_MASK))) {
4158                 int i, rmask;
4159
4160                 for (i = 0; i < MAX_RX_DESC_RINGS; i++) {
4161                         rmask = CAS_FLAG_RXD_POST(i);
4162                         if ((mask & rmask) == 0)
4163                                 continue;
4164
4165                         /* post_rxds will do a mod_timer */
4166                         if (cas_post_rxds_ringN(cp, i, cp->rx_last[i]) < 0) {
4167                                 pending = 1;
4168                                 continue;
4169                         }
4170                         cp->cas_flags &= ~rmask;
4171                 }
4172         }
4173
4174         if (CAS_PHY_MII(cp->phy_type)) {
4175                 u16 bmsr;
4176                 cas_mif_poll(cp, 0);
4177                 bmsr = cas_phy_read(cp, MII_BMSR);
4178                 /* WTZ: Solaris driver reads this twice, but that
4179                  * may be due to the PCS case and the use of a
4180                  * common implementation. Read it twice here to be
4181                  * safe.
4182                  */
4183                 bmsr = cas_phy_read(cp, MII_BMSR);
4184                 cas_mif_poll(cp, 1);
4185                 readl(cp->regs + REG_MIF_STATUS); /* avoid dups */
4186                 reset = cas_mii_link_check(cp, bmsr);
4187         } else {
4188                 reset = cas_pcs_link_check(cp);
4189         }
4190
4191         if (reset)
4192                 goto done;
4193
4194         /* check for tx state machine confusion */
4195         if ((readl(cp->regs + REG_MAC_TX_STATUS) & MAC_TX_FRAME_XMIT) == 0) {
4196                 u32 val = readl(cp->regs + REG_MAC_STATE_MACHINE);
4197                 u32 wptr, rptr;
4198                 int tlm  = CAS_VAL(MAC_SM_TLM, val);
4199
4200                 if (((tlm == 0x5) || (tlm == 0x3)) &&
4201                     (CAS_VAL(MAC_SM_ENCAP_SM, val) == 0)) {
4202                         if (netif_msg_tx_err(cp))
4203                                 printk(KERN_DEBUG "%s: tx err: "
4204                                        "MAC_STATE[%08x]\n",
4205                                        cp->dev->name, val);
4206                         reset = 1;
4207                         goto done;
4208                 }
4209
4210                 val  = readl(cp->regs + REG_TX_FIFO_PKT_CNT);
4211                 wptr = readl(cp->regs + REG_TX_FIFO_WRITE_PTR);
4212                 rptr = readl(cp->regs + REG_TX_FIFO_READ_PTR);
4213                 if ((val == 0) && (wptr != rptr)) {
4214                         if (netif_msg_tx_err(cp))
4215                                 printk(KERN_DEBUG "%s: tx err: "
4216                                        "TX_FIFO[%08x:%08x:%08x]\n",
4217                                        cp->dev->name, val, wptr, rptr);
4218                         reset = 1;
4219                 }
4220
4221                 if (reset)
4222                         cas_hard_reset(cp);
4223         }
4224
4225 done:
4226         if (reset) {
4227 #if 1
4228                 atomic_inc(&cp->reset_task_pending);
4229                 atomic_inc(&cp->reset_task_pending_all);
4230                 schedule_work(&cp->reset_task);
4231 #else
4232                 atomic_set(&cp->reset_task_pending, CAS_RESET_ALL);
4233                 printk(KERN_ERR "reset called in cas_link_timer\n");
4234                 schedule_work(&cp->reset_task);
4235 #endif
4236         }
4237
4238         if (!pending)
4239                 mod_timer(&cp->link_timer, jiffies + CAS_LINK_TIMEOUT);
4240         cas_unlock_tx(cp);
4241         spin_unlock_irqrestore(&cp->lock, flags);
4242 }
4243
4244 /* tiny buffers are used to avoid target abort issues with 
4245  * older cassini's
4246  */
4247 static void cas_tx_tiny_free(struct cas *cp)
4248 {
4249         struct pci_dev *pdev = cp->pdev;
4250         int i;
4251
4252         for (i = 0; i < N_TX_RINGS; i++) {
4253                 if (!cp->tx_tiny_bufs[i])
4254                         continue;
4255
4256                 pci_free_consistent(pdev, TX_TINY_BUF_BLOCK, 
4257                                     cp->tx_tiny_bufs[i],
4258                                     cp->tx_tiny_dvma[i]);
4259                 cp->tx_tiny_bufs[i] = NULL;
4260         }
4261 }
4262
4263 static int cas_tx_tiny_alloc(struct cas *cp)
4264 {
4265         struct pci_dev *pdev = cp->pdev;
4266         int i;
4267
4268         for (i = 0; i < N_TX_RINGS; i++) {
4269                 cp->tx_tiny_bufs[i] = 
4270                         pci_alloc_consistent(pdev, TX_TINY_BUF_BLOCK,
4271                                              &cp->tx_tiny_dvma[i]);
4272                 if (!cp->tx_tiny_bufs[i]) {
4273                         cas_tx_tiny_free(cp);
4274                         return -1;
4275                 }
4276         }
4277         return 0;
4278 }
4279
4280
4281 static int cas_open(struct net_device *dev)
4282 {
4283         struct cas *cp = netdev_priv(dev);
4284         int hw_was_up, err;
4285         unsigned long flags;
4286
4287         down(&cp->pm_sem);
4288
4289         hw_was_up = cp->hw_running;
4290
4291         /* The power-management semaphore protects the hw_running
4292          * etc. state so it is safe to do this bit without cp->lock
4293          */
4294         if (!cp->hw_running) {
4295                 /* Reset the chip */
4296                 cas_lock_all_save(cp, flags);
4297                 /* We set the second arg to cas_reset to zero
4298                  * because cas_init_hw below will have its second 
4299                  * argument set to non-zero, which will force
4300                  * autonegotiation to start.
4301                  */
4302                 cas_reset(cp, 0);
4303                 cp->hw_running = 1;
4304                 cas_unlock_all_restore(cp, flags);
4305         }
4306
4307         if (cas_tx_tiny_alloc(cp) < 0)
4308                 return -ENOMEM;
4309
4310         /* alloc rx descriptors */
4311         err = -ENOMEM;
4312         if (cas_alloc_rxds(cp) < 0)
4313                 goto err_tx_tiny;
4314         
4315         /* allocate spares */
4316         cas_spare_init(cp);
4317         cas_spare_recover(cp, GFP_KERNEL);
4318
4319         /* We can now request the interrupt as we know it's masked
4320          * on the controller. cassini+ has up to 4 interrupts
4321          * that can be used, but you need to do explicit pci interrupt 
4322          * mapping to expose them
4323          */
4324         if (request_irq(cp->pdev->irq, cas_interrupt,
4325                         SA_SHIRQ, dev->name, (void *) dev)) {
4326                 printk(KERN_ERR "%s: failed to request irq !\n", 
4327                        cp->dev->name);
4328                 err = -EAGAIN;
4329                 goto err_spare;
4330         }
4331
4332         /* init hw */
4333         cas_lock_all_save(cp, flags);
4334         cas_clean_rings(cp);
4335         cas_init_hw(cp, !hw_was_up);
4336         cp->opened = 1;
4337         cas_unlock_all_restore(cp, flags);
4338
4339         netif_start_queue(dev);
4340         up(&cp->pm_sem);
4341         return 0;
4342
4343 err_spare:
4344         cas_spare_free(cp);
4345         cas_free_rxds(cp);
4346 err_tx_tiny:
4347         cas_tx_tiny_free(cp);
4348         up(&cp->pm_sem);
4349         return err;
4350 }
4351
4352 static int cas_close(struct net_device *dev)
4353 {
4354         unsigned long flags;
4355         struct cas *cp = netdev_priv(dev);
4356
4357         /* Make sure we don't get distracted by suspend/resume */
4358         down(&cp->pm_sem);
4359
4360         netif_stop_queue(dev);
4361
4362         /* Stop traffic, mark us closed */
4363         cas_lock_all_save(cp, flags);
4364         cp->opened = 0; 
4365         cas_reset(cp, 0);
4366         cas_phy_init(cp); 
4367         cas_begin_auto_negotiation(cp, NULL);
4368         cas_clean_rings(cp);
4369         cas_unlock_all_restore(cp, flags);
4370
4371         free_irq(cp->pdev->irq, (void *) dev);
4372         cas_spare_free(cp);
4373         cas_free_rxds(cp);
4374         cas_tx_tiny_free(cp);
4375         up(&cp->pm_sem);
4376         return 0;
4377 }
4378
4379 static struct {
4380         const char name[ETH_GSTRING_LEN];
4381 } ethtool_cassini_statnames[] = {
4382         {"collisions"},
4383         {"rx_bytes"},
4384         {"rx_crc_errors"},
4385         {"rx_dropped"},
4386         {"rx_errors"},
4387         {"rx_fifo_errors"},
4388         {"rx_frame_errors"},
4389         {"rx_length_errors"},
4390         {"rx_over_errors"},
4391         {"rx_packets"},
4392         {"tx_aborted_errors"},
4393         {"tx_bytes"},
4394         {"tx_dropped"},
4395         {"tx_errors"},
4396         {"tx_fifo_errors"},
4397         {"tx_packets"}
4398 };
4399 #define CAS_NUM_STAT_KEYS (sizeof(ethtool_cassini_statnames)/ETH_GSTRING_LEN)
4400
4401 static struct {
4402         const int offsets;      /* neg. values for 2nd arg to cas_read_phy */
4403 } ethtool_register_table[] = {
4404         {-MII_BMSR},
4405         {-MII_BMCR},
4406         {REG_CAWR},
4407         {REG_INF_BURST},
4408         {REG_BIM_CFG},
4409         {REG_RX_CFG},
4410         {REG_HP_CFG},
4411         {REG_MAC_TX_CFG},
4412         {REG_MAC_RX_CFG},
4413         {REG_MAC_CTRL_CFG},
4414         {REG_MAC_XIF_CFG},
4415         {REG_MIF_CFG},
4416         {REG_PCS_CFG},
4417         {REG_SATURN_PCFG},
4418         {REG_PCS_MII_STATUS},
4419         {REG_PCS_STATE_MACHINE},
4420         {REG_MAC_COLL_EXCESS},
4421         {REG_MAC_COLL_LATE}
4422 };
4423 #define CAS_REG_LEN     (sizeof(ethtool_register_table)/sizeof(int))
4424 #define CAS_MAX_REGS    (sizeof (u32)*CAS_REG_LEN)
4425
4426 static u8 *cas_get_regs(struct cas *cp)
4427 {
4428         u8 *ptr = kmalloc(CAS_MAX_REGS, GFP_KERNEL);
4429         u8 *p;
4430         int i;
4431         unsigned long flags;
4432
4433         if (!ptr)
4434                 return NULL;
4435
4436         spin_lock_irqsave(&cp->lock, flags);
4437         for (i = 0, p = ptr; i < CAS_REG_LEN ; i ++, p += sizeof(u32)) {
4438                 u16 hval;
4439                 u32 val;
4440                 if (ethtool_register_table[i].offsets < 0) {
4441                         hval = cas_phy_read(cp,
4442                                     -ethtool_register_table[i].offsets);
4443                         val = hval;
4444                 } else {
4445                         val= readl(cp->regs+ethtool_register_table[i].offsets);
4446                 }
4447                 memcpy(p, (u8 *)&val, sizeof(u32));
4448         }
4449         spin_unlock_irqrestore(&cp->lock, flags);
4450
4451         return ptr;
4452 }
4453
4454 static struct net_device_stats *cas_get_stats(struct net_device *dev)
4455 {
4456         struct cas *cp = netdev_priv(dev);
4457         struct net_device_stats *stats = cp->net_stats;
4458         unsigned long flags;
4459         int i;
4460         unsigned long tmp;
4461
4462         /* we collate all of the stats into net_stats[N_TX_RING] */
4463         if (!cp->hw_running)
4464                 return stats + N_TX_RINGS;
4465         
4466         /* collect outstanding stats */
4467         /* WTZ: the Cassini spec gives these as 16 bit counters but
4468          * stored in 32-bit words.  Added a mask of 0xffff to be safe,
4469          * in case the chip somehow puts any garbage in the other bits.
4470          * Also, counter usage didn't seem to mach what Adrian did
4471          * in the parts of the code that set these quantities. Made
4472          * that consistent.
4473          */
4474         spin_lock_irqsave(&cp->stat_lock[N_TX_RINGS], flags);
4475         stats[N_TX_RINGS].rx_crc_errors += 
4476           readl(cp->regs + REG_MAC_FCS_ERR) & 0xffff;
4477         stats[N_TX_RINGS].rx_frame_errors += 
4478                 readl(cp->regs + REG_MAC_ALIGN_ERR) &0xffff;
4479         stats[N_TX_RINGS].rx_length_errors += 
4480                 readl(cp->regs + REG_MAC_LEN_ERR) & 0xffff;
4481 #if 1
4482         tmp = (readl(cp->regs + REG_MAC_COLL_EXCESS) & 0xffff) +
4483                 (readl(cp->regs + REG_MAC_COLL_LATE) & 0xffff);
4484         stats[N_TX_RINGS].tx_aborted_errors += tmp;
4485         stats[N_TX_RINGS].collisions +=
4486           tmp + (readl(cp->regs + REG_MAC_COLL_NORMAL) & 0xffff);
4487 #else
4488         stats[N_TX_RINGS].tx_aborted_errors += 
4489                 readl(cp->regs + REG_MAC_COLL_EXCESS);
4490         stats[N_TX_RINGS].collisions += readl(cp->regs + REG_MAC_COLL_EXCESS) +
4491                 readl(cp->regs + REG_MAC_COLL_LATE);
4492 #endif
4493         cas_clear_mac_err(cp);
4494
4495         /* saved bits that are unique to ring 0 */
4496         spin_lock(&cp->stat_lock[0]);
4497         stats[N_TX_RINGS].collisions        += stats[0].collisions;
4498         stats[N_TX_RINGS].rx_over_errors    += stats[0].rx_over_errors;
4499         stats[N_TX_RINGS].rx_frame_errors   += stats[0].rx_frame_errors;
4500         stats[N_TX_RINGS].rx_fifo_errors    += stats[0].rx_fifo_errors;
4501         stats[N_TX_RINGS].tx_aborted_errors += stats[0].tx_aborted_errors;
4502         stats[N_TX_RINGS].tx_fifo_errors    += stats[0].tx_fifo_errors;
4503         spin_unlock(&cp->stat_lock[0]);
4504
4505         for (i = 0; i < N_TX_RINGS; i++) {
4506                 spin_lock(&cp->stat_lock[i]);
4507                 stats[N_TX_RINGS].rx_length_errors += 
4508                         stats[i].rx_length_errors;
4509                 stats[N_TX_RINGS].rx_crc_errors += stats[i].rx_crc_errors;
4510                 stats[N_TX_RINGS].rx_packets    += stats[i].rx_packets;
4511                 stats[N_TX_RINGS].tx_packets    += stats[i].tx_packets;
4512                 stats[N_TX_RINGS].rx_bytes      += stats[i].rx_bytes;
4513                 stats[N_TX_RINGS].tx_bytes      += stats[i].tx_bytes;
4514                 stats[N_TX_RINGS].rx_errors     += stats[i].rx_errors;
4515                 stats[N_TX_RINGS].tx_errors     += stats[i].tx_errors;
4516                 stats[N_TX_RINGS].rx_dropped    += stats[i].rx_dropped;
4517                 stats[N_TX_RINGS].tx_dropped    += stats[i].tx_dropped;
4518                 memset(stats + i, 0, sizeof(struct net_device_stats));
4519                 spin_unlock(&cp->stat_lock[i]);
4520         }
4521         spin_unlock_irqrestore(&cp->stat_lock[N_TX_RINGS], flags);
4522         return stats + N_TX_RINGS;
4523 }
4524
4525
4526 static void cas_set_multicast(struct net_device *dev)
4527 {
4528         struct cas *cp = netdev_priv(dev);
4529         u32 rxcfg, rxcfg_new;
4530         unsigned long flags;
4531         int limit = STOP_TRIES;
4532         
4533         if (!cp->hw_running)
4534                 return;
4535                 
4536         spin_lock_irqsave(&cp->lock, flags);
4537         rxcfg = readl(cp->regs + REG_MAC_RX_CFG);
4538
4539         /* disable RX MAC and wait for completion */
4540         writel(rxcfg & ~MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
4541         while (readl(cp->regs + REG_MAC_RX_CFG) & MAC_RX_CFG_EN) {
4542                 if (!limit--)
4543                         break;
4544                 udelay(10);
4545         }
4546
4547         /* disable hash filter and wait for completion */
4548         limit = STOP_TRIES;
4549         rxcfg &= ~(MAC_RX_CFG_PROMISC_EN | MAC_RX_CFG_HASH_FILTER_EN);
4550         writel(rxcfg & ~MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
4551         while (readl(cp->regs + REG_MAC_RX_CFG) & MAC_RX_CFG_HASH_FILTER_EN) {
4552                 if (!limit--)
4553                         break;
4554                 udelay(10);
4555         }
4556
4557         /* program hash filters */
4558         cp->mac_rx_cfg = rxcfg_new = cas_setup_multicast(cp);
4559         rxcfg |= rxcfg_new;
4560         writel(rxcfg, cp->regs + REG_MAC_RX_CFG);
4561         spin_unlock_irqrestore(&cp->lock, flags);
4562 }
4563
4564 /* Eventually add support for changing the advertisement
4565  * on autoneg.
4566  */
4567 static int cas_ethtool_ioctl(struct net_device *dev, void *ep_user)
4568 {
4569         struct cas *cp = netdev_priv(dev);
4570         u16 bmcr;
4571         int full_duplex, speed, pause;
4572         struct ethtool_cmd ecmd;
4573         unsigned long flags;
4574         enum link_state linkstate = link_up;
4575
4576         if (copy_from_user(&ecmd, ep_user, sizeof(ecmd)))
4577                 return -EFAULT;
4578                 
4579         switch(ecmd.cmd) {
4580         case ETHTOOL_GDRVINFO: {
4581                 struct ethtool_drvinfo info = { cmd: ETHTOOL_GDRVINFO };
4582
4583                 strncpy(info.driver, DRV_MODULE_NAME,
4584                         ETHTOOL_BUSINFO_LEN);
4585                 strncpy(info.version, DRV_MODULE_VERSION,
4586                         ETHTOOL_BUSINFO_LEN);
4587                 info.fw_version[0] = '\0';
4588                 strncpy(info.bus_info, pci_name(cp->pdev),
4589                         ETHTOOL_BUSINFO_LEN);
4590                 info.regdump_len = cp->casreg_len < CAS_MAX_REGS ?
4591                         cp->casreg_len : CAS_MAX_REGS;
4592                 info.n_stats = CAS_NUM_STAT_KEYS;
4593                 if (copy_to_user(ep_user, &info, sizeof(info)))
4594                         return -EFAULT;
4595
4596                 return 0;
4597         }
4598
4599         case ETHTOOL_GSET:
4600                 ecmd.advertising = 0;
4601                 ecmd.supported = SUPPORTED_Autoneg;
4602                 if (cp->cas_flags & CAS_FLAG_1000MB_CAP) {
4603                         ecmd.supported |= SUPPORTED_1000baseT_Full;
4604                         ecmd.advertising |= ADVERTISED_1000baseT_Full;
4605                 }
4606
4607                 /* Record PHY settings if HW is on. */
4608                 spin_lock_irqsave(&cp->lock, flags);
4609                 bmcr = 0;
4610                 linkstate = cp->lstate;
4611                 if (CAS_PHY_MII(cp->phy_type)) {
4612                         ecmd.port = PORT_MII;
4613                         ecmd.transceiver = (cp->cas_flags & CAS_FLAG_SATURN) ?
4614                                 XCVR_INTERNAL : XCVR_EXTERNAL;
4615                         ecmd.phy_address = cp->phy_addr;
4616                         ecmd.advertising |= ADVERTISED_TP | ADVERTISED_MII |
4617                                 ADVERTISED_10baseT_Half | 
4618                                 ADVERTISED_10baseT_Full | 
4619                                 ADVERTISED_100baseT_Half | 
4620                                 ADVERTISED_100baseT_Full;
4621
4622                         ecmd.supported |=
4623                                 (SUPPORTED_10baseT_Half | 
4624                                  SUPPORTED_10baseT_Full |
4625                                  SUPPORTED_100baseT_Half | 
4626                                  SUPPORTED_100baseT_Full |
4627                                  SUPPORTED_TP | SUPPORTED_MII);
4628
4629                         if (cp->hw_running) {
4630                                 cas_mif_poll(cp, 0);
4631                                 bmcr = cas_phy_read(cp, MII_BMCR);
4632                                 cas_read_mii_link_mode(cp, &full_duplex, 
4633                                                        &speed, &pause);
4634                                 cas_mif_poll(cp, 1);
4635                         }
4636
4637                 } else {
4638                         ecmd.port = PORT_FIBRE;
4639                         ecmd.transceiver = XCVR_INTERNAL;
4640                         ecmd.phy_address = 0;
4641                         ecmd.supported   |= SUPPORTED_FIBRE;
4642                         ecmd.advertising |= ADVERTISED_FIBRE;
4643
4644                         if (cp->hw_running) {
4645                                 /* pcs uses the same bits as mii */ 
4646                                 bmcr = readl(cp->regs + REG_PCS_MII_CTRL);
4647                                 cas_read_pcs_link_mode(cp, &full_duplex, 
4648                                                        &speed, &pause);
4649                         }
4650                 }
4651                 spin_unlock_irqrestore(&cp->lock, flags);
4652
4653                 if (bmcr & BMCR_ANENABLE) {
4654                         ecmd.advertising |= ADVERTISED_Autoneg;
4655                         ecmd.autoneg = AUTONEG_ENABLE;
4656                         ecmd.speed = ((speed == 10) ?
4657                                       SPEED_10 :
4658                                       ((speed == 1000) ?
4659                                        SPEED_1000 : SPEED_100));
4660                         ecmd.duplex = full_duplex ? DUPLEX_FULL : DUPLEX_HALF;
4661                 } else {
4662                         ecmd.autoneg = AUTONEG_DISABLE;
4663                         ecmd.speed =
4664                                 (bmcr & CAS_BMCR_SPEED1000) ?
4665                                 SPEED_1000 : 
4666                                 ((bmcr & BMCR_SPEED100) ? SPEED_100: 
4667                                  SPEED_10);
4668                         ecmd.duplex =
4669                                 (bmcr & BMCR_FULLDPLX) ?
4670                                 DUPLEX_FULL : DUPLEX_HALF;
4671                 }
4672                 if (linkstate != link_up) {
4673                         /* Force these to "unknown" if the link is not up and
4674                          * autonogotiation in enabled. We can set the link 
4675                          * speed to 0, but not ecmd.duplex,
4676                          * because its legal values are 0 and 1.  Ethtool will
4677                          * print the value reported in parentheses after the
4678                          * word "Unknown" for unrecognized values.
4679                          *
4680                          * If in forced mode, we report the speed and duplex
4681                          * settings that we configured.
4682                          */
4683                         if (cp->link_cntl & BMCR_ANENABLE) {
4684                                 ecmd.speed = 0;
4685                                 ecmd.duplex = 0xff;
4686                         } else {
4687                                 ecmd.speed = SPEED_10;
4688                                 if (cp->link_cntl & BMCR_SPEED100) {
4689                                         ecmd.speed = SPEED_100;
4690                                 } else if (cp->link_cntl & CAS_BMCR_SPEED1000) {
4691                                         ecmd.speed = SPEED_1000;
4692                                 }
4693                                 ecmd.duplex = (cp->link_cntl & BMCR_FULLDPLX)?
4694                                         DUPLEX_FULL : DUPLEX_HALF;
4695                         }
4696                 }
4697                 if (copy_to_user(ep_user, &ecmd, sizeof(ecmd)))
4698                         return -EFAULT;
4699                 return 0;
4700
4701         case ETHTOOL_SSET:
4702                 if (!capable(CAP_NET_ADMIN))
4703                         return -EPERM;
4704
4705                 /* Verify the settings we care about. */
4706                 if (ecmd.autoneg != AUTONEG_ENABLE &&
4707                     ecmd.autoneg != AUTONEG_DISABLE)
4708                         return -EINVAL;
4709
4710                 if (ecmd.autoneg == AUTONEG_DISABLE &&
4711                     ((ecmd.speed != SPEED_1000 &&
4712                       ecmd.speed != SPEED_100 &&
4713                       ecmd.speed != SPEED_10) ||
4714                      (ecmd.duplex != DUPLEX_HALF &&
4715                       ecmd.duplex != DUPLEX_FULL)))
4716                         return -EINVAL;
4717
4718                 /* Apply settings and restart link process. */
4719                 spin_lock_irqsave(&cp->lock, flags);
4720                 cas_begin_auto_negotiation(cp, &ecmd);
4721                 spin_unlock_irqrestore(&cp->lock, flags);
4722                 return 0;
4723
4724         case ETHTOOL_NWAY_RST:
4725                 if ((cp->link_cntl & BMCR_ANENABLE) == 0)
4726                         return -EINVAL;
4727
4728                 /* Restart link process. */
4729                 spin_lock_irqsave(&cp->lock, flags);
4730                 cas_begin_auto_negotiation(cp, NULL);
4731                 spin_unlock_irqrestore(&cp->lock, flags);
4732
4733                 return 0;
4734
4735         case ETHTOOL_GWOL:
4736         case ETHTOOL_SWOL:
4737                 break; /* doesn't exist */
4738
4739         /* get link status */
4740         case ETHTOOL_GLINK: {
4741                 struct ethtool_value edata = { cmd: ETHTOOL_GLINK };
4742
4743                 edata.data = (cp->lstate == link_up);
4744                 if (copy_to_user(ep_user, &edata, sizeof(edata)))
4745                         return -EFAULT;
4746                 return 0;
4747         }
4748
4749         /* get message-level */
4750         case ETHTOOL_GMSGLVL: {
4751                 struct ethtool_value edata = { cmd: ETHTOOL_GMSGLVL };
4752
4753                 edata.data = cp->msg_enable;
4754                 if (copy_to_user(ep_user, &edata, sizeof(edata)))
4755                         return -EFAULT;
4756                 return 0;
4757         }
4758
4759         /* set message-level */
4760         case ETHTOOL_SMSGLVL: {
4761                 struct ethtool_value edata;
4762
4763                 if (!capable(CAP_NET_ADMIN)) {
4764                         return (-EPERM);
4765                 }
4766                 if (copy_from_user(&edata, ep_user, sizeof(edata)))
4767                         return -EFAULT;
4768                 cp->msg_enable = edata.data;
4769                 return 0;
4770         }
4771
4772         case ETHTOOL_GREGS: {
4773                 struct ethtool_regs edata;
4774                 u8 *ptr;
4775                 int len = cp->casreg_len < CAS_MAX_REGS ?
4776                         cp->casreg_len: CAS_MAX_REGS;
4777
4778                 if (copy_from_user(&edata, ep_user, sizeof (edata)))
4779                         return -EFAULT;
4780
4781                 if (edata.len > len)
4782                         edata.len = len;
4783                 edata.version = 0;
4784                 if (copy_to_user (ep_user, &edata, sizeof(edata)))
4785                         return -EFAULT;
4786
4787                 /* cas_get_regs handles locks (cp->lock).  */
4788                 ptr = cas_get_regs(cp);
4789                 if (ptr == NULL)
4790                         return -ENOMEM;
4791                 if (copy_to_user(ep_user + sizeof (edata), ptr, edata.len))
4792                         return -EFAULT;
4793
4794                 kfree(ptr);
4795                 return (0);
4796         }
4797         case ETHTOOL_GSTRINGS: {
4798                 struct ethtool_gstrings edata;
4799                 int len;
4800
4801                 if (copy_from_user(&edata, ep_user, sizeof(edata)))
4802                         return -EFAULT;
4803
4804                 len = edata.len;
4805                 switch(edata.string_set) {
4806                 case ETH_SS_STATS:
4807                         edata.len = (len < CAS_NUM_STAT_KEYS) ?
4808                                 len : CAS_NUM_STAT_KEYS;
4809                         if (copy_to_user(ep_user, &edata, sizeof(edata)))
4810                                 return -EFAULT;
4811
4812                         if (copy_to_user(ep_user + sizeof(edata),
4813                                          &ethtool_cassini_statnames, 
4814                                          (edata.len * ETH_GSTRING_LEN)))
4815                                 return -EFAULT;
4816                         return 0;
4817                 default:
4818                         return -EINVAL;
4819                 }
4820         }
4821         case ETHTOOL_GSTATS: {
4822                 int i = 0;
4823                 u64 *tmp;
4824                 struct ethtool_stats edata;
4825                 struct net_device_stats *stats;
4826                 int len;
4827
4828                 if (copy_from_user(&edata, ep_user, sizeof(edata)))
4829                         return -EFAULT;
4830
4831                 len = edata.n_stats;
4832                 stats = cas_get_stats(cp->dev);
4833                 edata.cmd = ETHTOOL_GSTATS;
4834                 edata.n_stats = (len < CAS_NUM_STAT_KEYS) ?
4835                         len : CAS_NUM_STAT_KEYS;
4836                 if (copy_to_user(ep_user, &edata, sizeof (edata)))
4837                         return -EFAULT;
4838
4839                 tmp = kmalloc(sizeof(u64)*CAS_NUM_STAT_KEYS, GFP_KERNEL);
4840                 if (tmp) {
4841                         tmp[i++] = stats->collisions;
4842                         tmp[i++] = stats->rx_bytes;
4843                         tmp[i++] = stats->rx_crc_errors;
4844                         tmp[i++] = stats->rx_dropped;
4845                         tmp[i++] = stats->rx_errors;
4846                         tmp[i++] = stats->rx_fifo_errors;
4847                         tmp[i++] = stats->rx_frame_errors;
4848                         tmp[i++] = stats->rx_length_errors;
4849                         tmp[i++] = stats->rx_over_errors;
4850                         tmp[i++] = stats->rx_packets;
4851                         tmp[i++] = stats->tx_aborted_errors;
4852                         tmp[i++] = stats->tx_bytes;
4853                         tmp[i++] = stats->tx_dropped;
4854                         tmp[i++] = stats->tx_errors;
4855                         tmp[i++] = stats->tx_fifo_errors;
4856                         tmp[i++] = stats->tx_packets;
4857                         BUG_ON(i != CAS_NUM_STAT_KEYS);
4858
4859                         i = copy_to_user(ep_user + sizeof(edata),
4860                                          tmp, sizeof(u64)*edata.n_stats);
4861                         kfree(tmp);
4862                 } else {
4863                         return -ENOMEM;
4864                 }
4865                 if (i)
4866                         return -EFAULT;
4867                 return 0;
4868         }
4869         }
4870
4871         return -EOPNOTSUPP;
4872 }
4873
4874 static int cas_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
4875 {
4876         struct cas *cp = netdev_priv(dev);
4877         struct mii_ioctl_data *data = (struct mii_ioctl_data *)&ifr->ifr_data;
4878         unsigned long flags;
4879         int rc = -EOPNOTSUPP;
4880         
4881         /* Hold the PM semaphore while doing ioctl's or we may collide
4882          * with open/close and power management and oops.
4883          */
4884         down(&cp->pm_sem);
4885         switch (cmd) {
4886         case SIOCETHTOOL:
4887                 rc = cas_ethtool_ioctl(dev, ifr->ifr_data);
4888                 break;
4889
4890         case SIOCGMIIPHY:               /* Get address of MII PHY in use. */
4891                 data->phy_id = cp->phy_addr;
4892                 /* Fallthrough... */
4893
4894         case SIOCGMIIREG:               /* Read MII PHY register. */
4895                 spin_lock_irqsave(&cp->lock, flags);
4896                 cas_mif_poll(cp, 0);
4897                 data->val_out = cas_phy_read(cp, data->reg_num & 0x1f);
4898                 cas_mif_poll(cp, 1);
4899                 spin_unlock_irqrestore(&cp->lock, flags);
4900                 rc = 0;
4901                 break;
4902
4903         case SIOCSMIIREG:               /* Write MII PHY register. */
4904                 if (!capable(CAP_NET_ADMIN)) {
4905                         rc = -EPERM;
4906                         break;
4907                 }
4908                 spin_lock_irqsave(&cp->lock, flags);
4909                 cas_mif_poll(cp, 0);
4910                 rc = cas_phy_write(cp, data->reg_num & 0x1f, data->val_in);
4911                 cas_mif_poll(cp, 1);
4912                 spin_unlock_irqrestore(&cp->lock, flags);
4913                 break;
4914         default:
4915                 break;
4916         };
4917
4918         up(&cp->pm_sem);
4919         return rc;
4920 }
4921
4922 static int __devinit cas_init_one(struct pci_dev *pdev,
4923                                   const struct pci_device_id *ent)
4924 {
4925         static int cas_version_printed = 0;
4926         unsigned long casreg_base, casreg_len;
4927         struct net_device *dev;
4928         struct cas *cp;
4929         int i, err, pci_using_dac;
4930         u16 pci_cmd;
4931         u8 orig_cacheline_size = 0, cas_cacheline_size = 0;
4932
4933         if (cas_version_printed++ == 0)
4934                 printk(KERN_INFO "%s", version);
4935
4936         err = pci_enable_device(pdev);
4937         if (err) {
4938                 printk(KERN_ERR PFX "Cannot enable PCI device, "
4939                        "aborting.\n");
4940                 return err;
4941         }
4942
4943         if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
4944                 printk(KERN_ERR PFX "Cannot find proper PCI device "
4945                        "base address, aborting.\n");
4946                 err = -ENODEV;
4947                 goto err_out_disable_pdev;
4948         }
4949
4950         dev = alloc_etherdev(sizeof(*cp));
4951         if (!dev) {
4952                 printk(KERN_ERR PFX "Etherdev alloc failed, aborting.\n");
4953                 err = -ENOMEM;
4954                 goto err_out_disable_pdev;
4955         }
4956         SET_MODULE_OWNER(dev);
4957         SET_NETDEV_DEV(dev, &pdev->dev);
4958
4959         err = pci_request_regions(pdev, dev->name);
4960         if (err) {
4961                 printk(KERN_ERR PFX "Cannot obtain PCI resources, "
4962                        "aborting.\n");
4963                 goto err_out_free_netdev;
4964         }
4965         pci_set_master(pdev);
4966
4967         /* we must always turn on parity response or else parity
4968          * doesn't get generated properly. disable SERR/PERR as well.
4969          * in addition, we want to turn MWI on.
4970          */
4971         pci_read_config_word(pdev, PCI_COMMAND, &pci_cmd);
4972         pci_cmd &= ~PCI_COMMAND_SERR;
4973         pci_cmd |= PCI_COMMAND_PARITY;
4974         pci_write_config_word(pdev, PCI_COMMAND, pci_cmd);
4975         pci_set_mwi(pdev);
4976         /*
4977          * On some architectures, the default cache line size set
4978          * by pci_set_mwi reduces perforamnce.  We have to increase
4979          * it for this case.  To start, we'll print some configuration
4980          * data.
4981          */
4982 #if 1
4983         pci_read_config_byte(pdev, PCI_CACHE_LINE_SIZE,
4984                              &orig_cacheline_size);
4985         if (orig_cacheline_size < CAS_PREF_CACHELINE_SIZE) {
4986                 cas_cacheline_size = 
4987                         (CAS_PREF_CACHELINE_SIZE < SMP_CACHE_BYTES) ? 
4988                         CAS_PREF_CACHELINE_SIZE : SMP_CACHE_BYTES;
4989                 if (pci_write_config_byte(pdev, 
4990                                           PCI_CACHE_LINE_SIZE, 
4991                                           cas_cacheline_size)) {
4992                         printk(KERN_ERR PFX "Could not set PCI cache "
4993                                "line size\n");
4994                         goto err_write_cacheline;
4995                 }
4996         }
4997 #endif
4998
4999
5000         /* Configure DMA attributes. */
5001         if (!pci_set_dma_mask(pdev, DMA_64BIT_MASK)) {
5002                 pci_using_dac = 1;
5003                 err = pci_set_consistent_dma_mask(pdev,
5004                                                   DMA_64BIT_MASK);
5005                 if (err < 0) {
5006                         printk(KERN_ERR PFX "Unable to obtain 64-bit DMA "
5007                                "for consistent allocations\n");
5008                         goto err_out_free_res;
5009                 }
5010
5011         } else {
5012                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
5013                 if (err) {
5014                         printk(KERN_ERR PFX "No usable DMA configuration, "
5015                                "aborting.\n");
5016                         goto err_out_free_res;
5017                 }
5018                 pci_using_dac = 0;
5019         }
5020
5021         casreg_base = pci_resource_start(pdev, 0);
5022         casreg_len = pci_resource_len(pdev, 0);
5023
5024         cp = netdev_priv(dev);
5025         cp->pdev = pdev;
5026 #if 1
5027         /* A value of 0 indicates we never explicitly set it */
5028         cp->orig_cacheline_size = cas_cacheline_size ? orig_cacheline_size: 0;
5029 #endif
5030         cp->dev = dev;
5031         cp->msg_enable = (cassini_debug < 0) ? CAS_DEF_MSG_ENABLE : 
5032           cassini_debug;
5033
5034         cp->link_transition = LINK_TRANSITION_UNKNOWN;
5035         cp->link_transition_jiffies_valid = 0;
5036
5037         spin_lock_init(&cp->lock);
5038         spin_lock_init(&cp->rx_inuse_lock);
5039         spin_lock_init(&cp->rx_spare_lock);
5040         for (i = 0; i < N_TX_RINGS; i++) {
5041                 spin_lock_init(&cp->stat_lock[i]);
5042                 spin_lock_init(&cp->tx_lock[i]);
5043         }
5044         spin_lock_init(&cp->stat_lock[N_TX_RINGS]);
5045         init_MUTEX(&cp->pm_sem);
5046
5047         init_timer(&cp->link_timer);
5048         cp->link_timer.function = cas_link_timer;
5049         cp->link_timer.data = (unsigned long) cp;
5050
5051 #if 1
5052         /* Just in case the implementation of atomic operations
5053          * change so that an explicit initialization is necessary.
5054          */
5055         atomic_set(&cp->reset_task_pending, 0);
5056         atomic_set(&cp->reset_task_pending_all, 0);
5057         atomic_set(&cp->reset_task_pending_spare, 0);
5058         atomic_set(&cp->reset_task_pending_mtu, 0);
5059 #endif
5060         INIT_WORK(&cp->reset_task, cas_reset_task, cp);
5061
5062         /* Default link parameters */
5063         if (link_mode >= 0 && link_mode <= 6)
5064                 cp->link_cntl = link_modes[link_mode];
5065         else
5066                 cp->link_cntl = BMCR_ANENABLE;
5067         cp->lstate = link_down;
5068         cp->link_transition = LINK_TRANSITION_LINK_DOWN;
5069         netif_carrier_off(cp->dev);
5070         cp->timer_ticks = 0;
5071
5072         /* give us access to cassini registers */
5073         cp->regs = ioremap(casreg_base, casreg_len);
5074         if (cp->regs == 0UL) {
5075                 printk(KERN_ERR PFX "Cannot map device registers, "
5076                        "aborting.\n");
5077                 goto err_out_free_res;
5078         }
5079         cp->casreg_len = casreg_len;
5080
5081         pci_save_state(pdev);
5082         cas_check_pci_invariants(cp);
5083         cas_hard_reset(cp);
5084         cas_reset(cp, 0);
5085         if (cas_check_invariants(cp))
5086                 goto err_out_iounmap;
5087
5088         cp->init_block = (struct cas_init_block *)
5089                 pci_alloc_consistent(pdev, sizeof(struct cas_init_block),
5090                                      &cp->block_dvma);
5091         if (!cp->init_block) {
5092                 printk(KERN_ERR PFX "Cannot allocate init block, "
5093                        "aborting.\n");
5094                 goto err_out_iounmap;
5095         }
5096
5097         for (i = 0; i < N_TX_RINGS; i++) 
5098                 cp->init_txds[i] = cp->init_block->txds[i];
5099
5100         for (i = 0; i < N_RX_DESC_RINGS; i++) 
5101                 cp->init_rxds[i] = cp->init_block->rxds[i];
5102
5103         for (i = 0; i < N_RX_COMP_RINGS; i++) 
5104                 cp->init_rxcs[i] = cp->init_block->rxcs[i];
5105
5106         for (i = 0; i < N_RX_FLOWS; i++)
5107                 skb_queue_head_init(&cp->rx_flows[i]);
5108
5109         dev->open = cas_open;
5110         dev->stop = cas_close;
5111         dev->hard_start_xmit = cas_start_xmit;
5112         dev->get_stats = cas_get_stats;
5113         dev->set_multicast_list = cas_set_multicast;
5114         dev->do_ioctl = cas_ioctl;
5115         dev->tx_timeout = cas_tx_timeout;
5116         dev->watchdog_timeo = CAS_TX_TIMEOUT;
5117         dev->change_mtu = cas_change_mtu;
5118 #ifdef USE_NAPI
5119         dev->poll = cas_poll;
5120         dev->weight = 64;
5121 #endif
5122 #ifdef CONFIG_NET_POLL_CONTROLLER
5123         dev->poll_controller = cas_netpoll;
5124 #endif
5125         dev->irq = pdev->irq;
5126         dev->dma = 0;
5127
5128         /* Cassini features. */
5129         if ((cp->cas_flags & CAS_FLAG_NO_HW_CSUM) == 0)
5130                 dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG;
5131
5132         if (pci_using_dac)
5133                 dev->features |= NETIF_F_HIGHDMA;
5134
5135         if (register_netdev(dev)) {
5136                 printk(KERN_ERR PFX "Cannot register net device, "
5137                        "aborting.\n");
5138                 goto err_out_free_consistent;
5139         }
5140
5141         i = readl(cp->regs + REG_BIM_CFG);
5142         printk(KERN_INFO "%s: Sun Cassini%s (%sbit/%sMHz PCI/%s) "
5143                "Ethernet[%d] ",  dev->name, 
5144                (cp->cas_flags & CAS_FLAG_REG_PLUS) ? "+" : "", 
5145                (i & BIM_CFG_32BIT) ? "32" : "64",
5146                (i & BIM_CFG_66MHZ) ? "66" : "33",
5147                (cp->phy_type == CAS_PHY_SERDES) ? "Fi" : "Cu", pdev->irq); 
5148
5149         for (i = 0; i < 6; i++)
5150                 printk("%2.2x%c", dev->dev_addr[i],
5151                        i == 5 ? ' ' : ':');
5152         printk("\n");
5153
5154         pci_set_drvdata(pdev, dev);
5155         cp->hw_running = 1;
5156         cas_entropy_reset(cp);
5157         cas_phy_init(cp);
5158         cas_begin_auto_negotiation(cp, NULL);
5159         return 0;
5160
5161 err_out_free_consistent:
5162         pci_free_consistent(pdev, sizeof(struct cas_init_block),
5163                             cp->init_block, cp->block_dvma);
5164
5165 err_out_iounmap:
5166         down(&cp->pm_sem);
5167         if (cp->hw_running)
5168                 cas_shutdown(cp);
5169         up(&cp->pm_sem);
5170
5171         iounmap((void *) cp->regs);
5172
5173
5174 err_out_free_res:
5175         pci_release_regions(pdev);
5176
5177 err_write_cacheline:
5178         /* Try to restore it in case the error occured after we
5179          * set it. 
5180          */
5181         pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE, orig_cacheline_size);
5182
5183 err_out_free_netdev:
5184         free_netdev(dev);
5185
5186 err_out_disable_pdev:
5187         pci_disable_device(pdev);
5188         pci_set_drvdata(pdev, NULL);
5189         return -ENODEV;
5190 }
5191
5192 static void __devexit cas_remove_one(struct pci_dev *pdev)
5193 {
5194         struct net_device *dev = pci_get_drvdata(pdev);
5195         struct cas *cp;
5196         if (!dev)
5197                 return;
5198
5199         cp = netdev_priv(dev);
5200         unregister_netdev(dev);
5201
5202         down(&cp->pm_sem);
5203         flush_scheduled_work();
5204         if (cp->hw_running)
5205                 cas_shutdown(cp);
5206         up(&cp->pm_sem);
5207
5208 #if 1
5209         if (cp->orig_cacheline_size) {
5210                 /* Restore the cache line size if we had modified
5211                  * it.
5212                  */
5213                 pci_write_config_byte(pdev, PCI_CACHE_LINE_SIZE, 
5214                                       cp->orig_cacheline_size);
5215         }
5216 #endif
5217         pci_free_consistent(pdev, sizeof(struct cas_init_block),
5218                             cp->init_block, cp->block_dvma);
5219         iounmap((void *) cp->regs);
5220         free_netdev(dev);
5221         pci_release_regions(pdev);
5222         pci_disable_device(pdev);
5223         pci_set_drvdata(pdev, NULL);
5224 }
5225
5226 #ifdef CONFIG_PM
5227 static int cas_suspend(struct pci_dev *pdev, u32 state)
5228 {
5229         struct net_device *dev = pci_get_drvdata(pdev);
5230         struct cas *cp = netdev_priv(dev);
5231         unsigned long flags;
5232
5233         /* We hold the PM semaphore during entire driver
5234          * sleep time
5235          */
5236         down(&cp->pm_sem);
5237         
5238         /* If the driver is opened, we stop the DMA */
5239         if (cp->opened) {
5240                 netif_device_detach(dev);
5241
5242                 cas_lock_all_save(cp, flags);
5243
5244                 /* We can set the second arg of cas_reset to 0
5245                  * because on resume, we'll call cas_init_hw with
5246                  * its second arg set so that autonegotiation is
5247                  * restarted.
5248                  */
5249                 cas_reset(cp, 0);
5250                 cas_clean_rings(cp);
5251                 cas_unlock_all_restore(cp, flags);
5252         }
5253
5254         if (cp->hw_running)
5255                 cas_shutdown(cp);
5256
5257         return 0;
5258 }
5259
5260 static int cas_resume(struct pci_dev *pdev)
5261 {
5262         struct net_device *dev = pci_get_drvdata(pdev);
5263         struct cas *cp = netdev_priv(dev);
5264
5265         printk(KERN_INFO "%s: resuming\n", dev->name);
5266
5267         cas_hard_reset(cp);
5268         if (cp->opened) {
5269                 unsigned long flags;
5270                 cas_lock_all_save(cp, flags);
5271                 cas_reset(cp, 0);
5272                 cp->hw_running = 1;
5273                 cas_clean_rings(cp);
5274                 cas_init_hw(cp, 1);
5275                 cas_unlock_all_restore(cp, flags);
5276
5277                 netif_device_attach(dev);
5278         }
5279         up(&cp->pm_sem);
5280         return 0;
5281 }
5282 #endif /* CONFIG_PM */
5283
5284 static struct pci_driver cas_driver = {
5285         .name           = DRV_MODULE_NAME,
5286         .id_table       = cas_pci_tbl,
5287         .probe          = cas_init_one,
5288         .remove         = __devexit_p(cas_remove_one),
5289 #ifdef CONFIG_PM
5290         .suspend        = cas_suspend,
5291         .resume         = cas_resume
5292 #endif
5293 };
5294
5295 static int __init cas_init(void)
5296 {
5297         if (linkdown_timeout > 0)
5298                 link_transition_timeout = linkdown_timeout * HZ;
5299         else
5300                 link_transition_timeout = 0;
5301
5302         return pci_module_init(&cas_driver);
5303 }
5304
5305 static void __exit cas_cleanup(void)
5306 {
5307         pci_unregister_driver(&cas_driver);
5308 }
5309
5310 module_init(cas_init);
5311 module_exit(cas_cleanup);