xref: /netbsd-src/sys/dev/pci/if_bge.c (revision 93bf6008f8b7982c1d1a9486e4a4a0e687fe36eb)
1 /*	$NetBSD: if_bge.c,v 1.162 2009/04/19 11:10:36 msaitoh Exp $	*/
2 
3 /*
4  * Copyright (c) 2001 Wind River Systems
5  * Copyright (c) 1997, 1998, 1999, 2001
6  *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by Bill Paul.
19  * 4. Neither the name of the author nor the names of any co-contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
27  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
30  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
31  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
32  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
33  * THE POSSIBILITY OF SUCH DAMAGE.
34  *
35  * $FreeBSD: if_bge.c,v 1.13 2002/04/04 06:01:31 wpaul Exp $
36  */
37 
38 /*
39  * Broadcom BCM570x family gigabit ethernet driver for NetBSD.
40  *
41  * NetBSD version by:
42  *
43  *	Frank van der Linden <fvdl@wasabisystems.com>
44  *	Jason Thorpe <thorpej@wasabisystems.com>
45  *	Jonathan Stone <jonathan@dsg.stanford.edu>
46  *
47  * Originally written for FreeBSD by Bill Paul <wpaul@windriver.com>
48  * Senior Engineer, Wind River Systems
49  */
50 
51 /*
52  * The Broadcom BCM5700 is based on technology originally developed by
53  * Alteon Networks as part of the Tigon I and Tigon II gigabit ethernet
54  * MAC chips. The BCM5700, sometimes refered to as the Tigon III, has
55  * two on-board MIPS R4000 CPUs and can have as much as 16MB of external
56  * SSRAM. The BCM5700 supports TCP, UDP and IP checksum offload, jumbo
57  * frames, highly configurable RX filtering, and 16 RX and TX queues
58  * (which, along with RX filter rules, can be used for QOS applications).
59  * Other features, such as TCP segmentation, may be available as part
60  * of value-added firmware updates. Unlike the Tigon I and Tigon II,
61  * firmware images can be stored in hardware and need not be compiled
62  * into the driver.
63  *
64  * The BCM5700 supports the PCI v2.2 and PCI-X v1.0 standards, and will
65  * function in a 32-bit/64-bit 33/66MHz bus, or a 64-bit/133MHz bus.
66  *
67  * The BCM5701 is a single-chip solution incorporating both the BCM5700
68  * MAC and a BCM5401 10/100/1000 PHY. Unlike the BCM5700, the BCM5701
69  * does not support external SSRAM.
70  *
71  * Broadcom also produces a variation of the BCM5700 under the "Altima"
72  * brand name, which is functionally similar but lacks PCI-X support.
73  *
74  * Without external SSRAM, you can only have at most 4 TX rings,
75  * and the use of the mini RX ring is disabled. This seems to imply
76  * that these features are simply not available on the BCM5701. As a
77  * result, this driver does not implement any support for the mini RX
78  * ring.
79  */
80 
81 #include <sys/cdefs.h>
82 __KERNEL_RCSID(0, "$NetBSD: if_bge.c,v 1.162 2009/04/19 11:10:36 msaitoh Exp $");
83 
84 #include "bpfilter.h"
85 #include "vlan.h"
86 #include "rnd.h"
87 
88 #include <sys/param.h>
89 #include <sys/systm.h>
90 #include <sys/callout.h>
91 #include <sys/sockio.h>
92 #include <sys/mbuf.h>
93 #include <sys/malloc.h>
94 #include <sys/kernel.h>
95 #include <sys/device.h>
96 #include <sys/socket.h>
97 #include <sys/sysctl.h>
98 
99 #include <net/if.h>
100 #include <net/if_dl.h>
101 #include <net/if_media.h>
102 #include <net/if_ether.h>
103 
104 #if NRND > 0
105 #include <sys/rnd.h>
106 #endif
107 
108 #ifdef INET
109 #include <netinet/in.h>
110 #include <netinet/in_systm.h>
111 #include <netinet/in_var.h>
112 #include <netinet/ip.h>
113 #endif
114 
115 /* Headers for TCP  Segmentation Offload (TSO) */
116 #include <netinet/in_systm.h>		/* n_time for <netinet/ip.h>... */
117 #include <netinet/in.h>			/* ip_{src,dst}, for <netinet/ip.h> */
118 #include <netinet/ip.h>			/* for struct ip */
119 #include <netinet/tcp.h>		/* for struct tcphdr */
120 
121 
122 #if NBPFILTER > 0
123 #include <net/bpf.h>
124 #endif
125 
126 #include <dev/pci/pcireg.h>
127 #include <dev/pci/pcivar.h>
128 #include <dev/pci/pcidevs.h>
129 
130 #include <dev/mii/mii.h>
131 #include <dev/mii/miivar.h>
132 #include <dev/mii/miidevs.h>
133 #include <dev/mii/brgphyreg.h>
134 
135 #include <dev/pci/if_bgereg.h>
136 
137 #include <uvm/uvm_extern.h>
138 
139 #define ETHER_MIN_NOPAD (ETHER_MIN_LEN - ETHER_CRC_LEN) /* i.e., 60 */
140 
141 
142 /*
143  * Tunable thresholds for rx-side bge interrupt mitigation.
144  */
145 
146 /*
147  * The pairs of values below were obtained from empirical measurement
148  * on bcm5700 rev B2; they ar designed to give roughly 1 receive
149  * interrupt for every N packets received, where N is, approximately,
150  * the second value (rx_max_bds) in each pair.  The values are chosen
151  * such that moving from one pair to the succeeding pair was observed
152  * to roughly halve interrupt rate under sustained input packet load.
153  * The values were empirically chosen to avoid overflowing internal
154  * limits on the  bcm5700: inreasing rx_ticks much beyond 600
155  * results in internal wrapping and higher interrupt rates.
156  * The limit of 46 frames was chosen to match NFS workloads.
157  *
158  * These values also work well on bcm5701, bcm5704C, and (less
159  * tested) bcm5703.  On other chipsets, (including the Altima chip
160  * family), the larger values may overflow internal chip limits,
161  * leading to increasing interrupt rates rather than lower interrupt
162  * rates.
163  *
164  * Applications using heavy interrupt mitigation (interrupting every
165  * 32 or 46 frames) in both directions may need to increase the TCP
166  * windowsize to above 131072 bytes (e.g., to 199608 bytes) to sustain
167  * full link bandwidth, due to ACKs and window updates lingering
168  * in the RX queue during the 30-to-40-frame interrupt-mitigation window.
169  */
170 static const struct bge_load_rx_thresh {
171 	int rx_ticks;
172 	int rx_max_bds; }
173 bge_rx_threshes[] = {
174 	{ 32,   2 },
175 	{ 50,   4 },
176 	{ 100,  8 },
177 	{ 192, 16 },
178 	{ 416, 32 },
179 	{ 598, 46 }
180 };
181 #define NBGE_RX_THRESH (sizeof(bge_rx_threshes) / sizeof(bge_rx_threshes[0]))
182 
183 /* XXX patchable; should be sysctl'able */
184 static int	bge_auto_thresh = 1;
185 static int	bge_rx_thresh_lvl;
186 
187 static int	bge_rxthresh_nodenum;
188 
189 typedef int (*bge_eaddr_fcn_t)(struct bge_softc *, u_int8_t[]);
190 
191 static int	bge_probe(device_t, cfdata_t, void *);
192 static void	bge_attach(device_t, device_t, void *);
193 static void	bge_release_resources(struct bge_softc *);
194 static void	bge_txeof(struct bge_softc *);
195 static void	bge_rxeof(struct bge_softc *);
196 
197 static int 	bge_get_eaddr_mem(struct bge_softc *, u_int8_t[]);
198 static int 	bge_get_eaddr_nvram(struct bge_softc *, u_int8_t[]);
199 static int 	bge_get_eaddr_eeprom(struct bge_softc *, u_int8_t[]);
200 static int 	bge_get_eaddr(struct bge_softc *, u_int8_t[]);
201 
202 static void	bge_tick(void *);
203 static void	bge_stats_update(struct bge_softc *);
204 static int	bge_encap(struct bge_softc *, struct mbuf *, u_int32_t *);
205 
206 static int	bge_intr(void *);
207 static void	bge_start(struct ifnet *);
208 static int	bge_ioctl(struct ifnet *, u_long, void *);
209 static int	bge_init(struct ifnet *);
210 static void	bge_stop(struct ifnet *, int);
211 static void	bge_watchdog(struct ifnet *);
212 static int	bge_ifmedia_upd(struct ifnet *);
213 static void	bge_ifmedia_sts(struct ifnet *, struct ifmediareq *);
214 
215 static void	bge_setmulti(struct bge_softc *);
216 
217 static void	bge_handle_events(struct bge_softc *);
218 static int	bge_alloc_jumbo_mem(struct bge_softc *);
219 #if 0 /* XXX */
220 static void	bge_free_jumbo_mem(struct bge_softc *);
221 #endif
222 static void	*bge_jalloc(struct bge_softc *);
223 static void	bge_jfree(struct mbuf *, void *, size_t, void *);
224 static int	bge_newbuf_std(struct bge_softc *, int, struct mbuf *,
225 			       bus_dmamap_t);
226 static int	bge_newbuf_jumbo(struct bge_softc *, int, struct mbuf *);
227 static int	bge_init_rx_ring_std(struct bge_softc *);
228 static void	bge_free_rx_ring_std(struct bge_softc *);
229 static int	bge_init_rx_ring_jumbo(struct bge_softc *);
230 static void	bge_free_rx_ring_jumbo(struct bge_softc *);
231 static void	bge_free_tx_ring(struct bge_softc *);
232 static int	bge_init_tx_ring(struct bge_softc *);
233 
234 static int	bge_chipinit(struct bge_softc *);
235 static int	bge_blockinit(struct bge_softc *);
236 static int	bge_setpowerstate(struct bge_softc *, int);
237 
238 static void	bge_reset(struct bge_softc *);
239 static void	bge_link_upd(struct bge_softc *);
240 
241 #define BGE_DEBUG
242 #ifdef BGE_DEBUG
243 #define DPRINTF(x)	if (bgedebug) printf x
244 #define DPRINTFN(n,x)	if (bgedebug >= (n)) printf x
245 #define BGE_TSO_PRINTF(x)  do { if (bge_tso_debug) printf x ;} while (0)
246 int	bgedebug = 0;
247 int	bge_tso_debug = 0;
248 #else
249 #define DPRINTF(x)
250 #define DPRINTFN(n,x)
251 #define BGE_TSO_PRINTF(x)
252 #endif
253 
254 #ifdef BGE_EVENT_COUNTERS
255 #define	BGE_EVCNT_INCR(ev)	(ev).ev_count++
256 #define	BGE_EVCNT_ADD(ev, val)	(ev).ev_count += (val)
257 #define	BGE_EVCNT_UPD(ev, val)	(ev).ev_count = (val)
258 #else
259 #define	BGE_EVCNT_INCR(ev)	/* nothing */
260 #define	BGE_EVCNT_ADD(ev, val)	/* nothing */
261 #define	BGE_EVCNT_UPD(ev, val)	/* nothing */
262 #endif
263 
264 static const struct bge_product {
265 	pci_vendor_id_t		bp_vendor;
266 	pci_product_id_t	bp_product;
267 	const char		*bp_name;
268 } bge_products[] = {
269 	/*
270 	 * The BCM5700 documentation seems to indicate that the hardware
271 	 * still has the Alteon vendor ID burned into it, though it
272 	 * should always be overridden by the value in the EEPROM.  We'll
273 	 * check for it anyway.
274 	 */
275 	{ PCI_VENDOR_ALTEON,
276 	  PCI_PRODUCT_ALTEON_BCM5700,
277 	  "Broadcom BCM5700 Gigabit Ethernet",
278 	  },
279 	{ PCI_VENDOR_ALTEON,
280 	  PCI_PRODUCT_ALTEON_BCM5701,
281 	  "Broadcom BCM5701 Gigabit Ethernet",
282 	  },
283 	{ PCI_VENDOR_ALTIMA,
284 	  PCI_PRODUCT_ALTIMA_AC1000,
285 	  "Altima AC1000 Gigabit Ethernet",
286 	  },
287 	{ PCI_VENDOR_ALTIMA,
288 	  PCI_PRODUCT_ALTIMA_AC1001,
289 	  "Altima AC1001 Gigabit Ethernet",
290 	   },
291 	{ PCI_VENDOR_ALTIMA,
292 	  PCI_PRODUCT_ALTIMA_AC9100,
293 	  "Altima AC9100 Gigabit Ethernet",
294 	  },
295 	{ PCI_VENDOR_BROADCOM,
296 	  PCI_PRODUCT_BROADCOM_BCM5700,
297 	  "Broadcom BCM5700 Gigabit Ethernet",
298 	  },
299 	{ PCI_VENDOR_BROADCOM,
300 	  PCI_PRODUCT_BROADCOM_BCM5701,
301 	  "Broadcom BCM5701 Gigabit Ethernet",
302 	  },
303 	{ PCI_VENDOR_BROADCOM,
304 	  PCI_PRODUCT_BROADCOM_BCM5702,
305 	  "Broadcom BCM5702 Gigabit Ethernet",
306 	  },
307 	{ PCI_VENDOR_BROADCOM,
308 	  PCI_PRODUCT_BROADCOM_BCM5702X,
309 	  "Broadcom BCM5702X Gigabit Ethernet" },
310 	{ PCI_VENDOR_BROADCOM,
311 	  PCI_PRODUCT_BROADCOM_BCM5703,
312 	  "Broadcom BCM5703 Gigabit Ethernet",
313 	  },
314 	{ PCI_VENDOR_BROADCOM,
315 	  PCI_PRODUCT_BROADCOM_BCM5703X,
316 	  "Broadcom BCM5703X Gigabit Ethernet",
317 	  },
318 	{ PCI_VENDOR_BROADCOM,
319 	  PCI_PRODUCT_BROADCOM_BCM5703_ALT,
320 	  "Broadcom BCM5703 Gigabit Ethernet",
321 	  },
322    	{ PCI_VENDOR_BROADCOM,
323 	  PCI_PRODUCT_BROADCOM_BCM5704C,
324 	  "Broadcom BCM5704C Dual Gigabit Ethernet",
325 	  },
326    	{ PCI_VENDOR_BROADCOM,
327 	  PCI_PRODUCT_BROADCOM_BCM5704S,
328 	  "Broadcom BCM5704S Dual Gigabit Ethernet",
329 	  },
330    	{ PCI_VENDOR_BROADCOM,
331 	  PCI_PRODUCT_BROADCOM_BCM5705,
332 	  "Broadcom BCM5705 Gigabit Ethernet",
333 	  },
334    	{ PCI_VENDOR_BROADCOM,
335 	  PCI_PRODUCT_BROADCOM_BCM5705K,
336 	  "Broadcom BCM5705K Gigabit Ethernet",
337 	  },
338    	{ PCI_VENDOR_BROADCOM,
339 	  PCI_PRODUCT_BROADCOM_BCM5705M,
340 	  "Broadcom BCM5705M Gigabit Ethernet",
341 	  },
342    	{ PCI_VENDOR_BROADCOM,
343 	  PCI_PRODUCT_BROADCOM_BCM5705M_ALT,
344 	  "Broadcom BCM5705M Gigabit Ethernet",
345 	  },
346 	{ PCI_VENDOR_BROADCOM,
347 	  PCI_PRODUCT_BROADCOM_BCM5714,
348 	  "Broadcom BCM5714/5715 Gigabit Ethernet",
349 	  },
350 	{ PCI_VENDOR_BROADCOM,
351 	  PCI_PRODUCT_BROADCOM_BCM5715,
352 	  "Broadcom BCM5714/5715 Gigabit Ethernet",
353 	  },
354 	{ PCI_VENDOR_BROADCOM,
355 	  PCI_PRODUCT_BROADCOM_BCM5789,
356 	  "Broadcom BCM5789 Gigabit Ethernet",
357 	  },
358 	{ PCI_VENDOR_BROADCOM,
359 	  PCI_PRODUCT_BROADCOM_BCM5721,
360 	  "Broadcom BCM5721 Gigabit Ethernet",
361 	  },
362 	{ PCI_VENDOR_BROADCOM,
363 	  PCI_PRODUCT_BROADCOM_BCM5722,
364 	  "Broadcom BCM5722 Gigabit Ethernet",
365 	  },
366 	{ PCI_VENDOR_BROADCOM,
367 	  PCI_PRODUCT_BROADCOM_BCM5750,
368 	  "Broadcom BCM5750 Gigabit Ethernet",
369 	  },
370 	{ PCI_VENDOR_BROADCOM,
371 	  PCI_PRODUCT_BROADCOM_BCM5750M,
372 	  "Broadcom BCM5750M Gigabit Ethernet",
373 	  },
374 	{ PCI_VENDOR_BROADCOM,
375 	  PCI_PRODUCT_BROADCOM_BCM5751,
376 	  "Broadcom BCM5751 Gigabit Ethernet",
377 	  },
378 	{ PCI_VENDOR_BROADCOM,
379 	  PCI_PRODUCT_BROADCOM_BCM5751M,
380 	  "Broadcom BCM5751M Gigabit Ethernet",
381 	  },
382 	{ PCI_VENDOR_BROADCOM,
383 	  PCI_PRODUCT_BROADCOM_BCM5752,
384 	  "Broadcom BCM5752 Gigabit Ethernet",
385 	  },
386 	{ PCI_VENDOR_BROADCOM,
387 	  PCI_PRODUCT_BROADCOM_BCM5752M,
388 	  "Broadcom BCM5752M Gigabit Ethernet",
389 	  },
390 	{ PCI_VENDOR_BROADCOM,
391 	  PCI_PRODUCT_BROADCOM_BCM5753,
392 	  "Broadcom BCM5753 Gigabit Ethernet",
393 	  },
394 	{ PCI_VENDOR_BROADCOM,
395 	  PCI_PRODUCT_BROADCOM_BCM5753M,
396 	  "Broadcom BCM5753M Gigabit Ethernet",
397 	  },
398 	{ PCI_VENDOR_BROADCOM,
399 	  PCI_PRODUCT_BROADCOM_BCM5754,
400 	  "Broadcom BCM5754 Gigabit Ethernet",
401 	},
402 	{ PCI_VENDOR_BROADCOM,
403 	  PCI_PRODUCT_BROADCOM_BCM5754M,
404 	  "Broadcom BCM5754M Gigabit Ethernet",
405 	},
406 	{ PCI_VENDOR_BROADCOM,
407 	  PCI_PRODUCT_BROADCOM_BCM5755,
408 	  "Broadcom BCM5755 Gigabit Ethernet",
409 	},
410 	{ PCI_VENDOR_BROADCOM,
411 	  PCI_PRODUCT_BROADCOM_BCM5755M,
412 	  "Broadcom BCM5755M Gigabit Ethernet",
413 	},
414    	{ PCI_VENDOR_BROADCOM,
415 	  PCI_PRODUCT_BROADCOM_BCM5780,
416 	  "Broadcom BCM5780 Gigabit Ethernet",
417 	  },
418    	{ PCI_VENDOR_BROADCOM,
419 	  PCI_PRODUCT_BROADCOM_BCM5780S,
420 	  "Broadcom BCM5780S Gigabit Ethernet",
421 	  },
422    	{ PCI_VENDOR_BROADCOM,
423 	  PCI_PRODUCT_BROADCOM_BCM5782,
424 	  "Broadcom BCM5782 Gigabit Ethernet",
425 	},
426 	{ PCI_VENDOR_BROADCOM,
427 	  PCI_PRODUCT_BROADCOM_BCM5786,
428 	  "Broadcom BCM5786 Gigabit Ethernet",
429 	},
430 	{ PCI_VENDOR_BROADCOM,
431 	  PCI_PRODUCT_BROADCOM_BCM5787,
432 	  "Broadcom BCM5787 Gigabit Ethernet",
433 	},
434 	{ PCI_VENDOR_BROADCOM,
435 	  PCI_PRODUCT_BROADCOM_BCM5787M,
436 	  "Broadcom BCM5787M Gigabit Ethernet",
437 	},
438    	{ PCI_VENDOR_BROADCOM,
439 	  PCI_PRODUCT_BROADCOM_BCM5788,
440 	  "Broadcom BCM5788 Gigabit Ethernet",
441 	  },
442    	{ PCI_VENDOR_BROADCOM,
443 	  PCI_PRODUCT_BROADCOM_BCM5789,
444 	  "Broadcom BCM5789 Gigabit Ethernet",
445 	  },
446    	{ PCI_VENDOR_BROADCOM,
447 	  PCI_PRODUCT_BROADCOM_BCM5901,
448 	  "Broadcom BCM5901 Fast Ethernet",
449 	  },
450    	{ PCI_VENDOR_BROADCOM,
451 	  PCI_PRODUCT_BROADCOM_BCM5901A2,
452 	  "Broadcom BCM5901A2 Fast Ethernet",
453 	  },
454 	{ PCI_VENDOR_SCHNEIDERKOCH,
455 	  PCI_PRODUCT_SCHNEIDERKOCH_SK_9DX1,
456 	  "SysKonnect SK-9Dx1 Gigabit Ethernet",
457 	  },
458 	{ PCI_VENDOR_3COM,
459 	  PCI_PRODUCT_3COM_3C996,
460 	  "3Com 3c996 Gigabit Ethernet",
461 	  },
462 	{ PCI_VENDOR_BROADCOM,
463 	  PCI_PRODUCT_BROADCOM_BCM5906,
464 	  "Broadcom BCM5906 Fast Ethernet",
465 	  },
466 	{ PCI_VENDOR_BROADCOM,
467 	  PCI_PRODUCT_BROADCOM_BCM5906M,
468 	  "Broadcom BCM5906M Fast Ethernet",
469 	  },
470 	{ 0,
471 	  0,
472 	  NULL },
473 };
474 
475 /*
476  * XXX: how to handle variants based on 5750 and derivatives:
477  * 5750 5751, 5721, possibly 5714, 5752, and 5708?, which
478  * in general behave like a 5705, except with additional quirks.
479  * This driver's current handling of the 5721 is wrong;
480  * how we map ASIC revision to "quirks" needs more thought.
481  * (defined here until the thought is done).
482  */
483 #define BGE_IS_5714_FAMILY(sc) \
484 	(BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5714_A0 || \
485 	 BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5780 ||	\
486 	 BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5714 )
487 
488 #define BGE_IS_5750_OR_BEYOND(sc)  \
489 	(BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5750 || \
490 	 BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5752 || \
491 	 BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 || \
492 	 BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5787 || \
493 	 BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906 || \
494 	 BGE_IS_5714_FAMILY(sc) )
495 
496 #define BGE_IS_5705_OR_BEYOND(sc)  \
497 	(BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5705 || \
498 	    (BGE_IS_5750_OR_BEYOND(sc)))
499 
500 static const struct bge_revision {
501 	uint32_t		br_chipid;
502 	const char		*br_name;
503 } bge_revisions[] = {
504 	{ BGE_CHIPID_BCM5700_A0, "BCM5700 A0" },
505 	{ BGE_CHIPID_BCM5700_A1, "BCM5700 A1" },
506 	{ BGE_CHIPID_BCM5700_B0, "BCM5700 B0" },
507 	{ BGE_CHIPID_BCM5700_B1, "BCM5700 B1" },
508 	{ BGE_CHIPID_BCM5700_B2, "BCM5700 B2" },
509 	{ BGE_CHIPID_BCM5700_B3, "BCM5700 B3" },
510 	/* This is treated like a BCM5700 Bx */
511 	{ BGE_CHIPID_BCM5700_ALTIMA, "BCM5700 Altima" },
512 	{ BGE_CHIPID_BCM5700_C0, "BCM5700 C0" },
513 	{ BGE_CHIPID_BCM5701_A0, "BCM5701 A0" },
514 	{ BGE_CHIPID_BCM5701_B0, "BCM5701 B0" },
515 	{ BGE_CHIPID_BCM5701_B2, "BCM5701 B2" },
516 	{ BGE_CHIPID_BCM5701_B5, "BCM5701 B5" },
517 	{ BGE_CHIPID_BCM5703_A0, "BCM5703 A0" },
518 	{ BGE_CHIPID_BCM5703_A1, "BCM5703 A1" },
519 	{ BGE_CHIPID_BCM5703_A2, "BCM5703 A2" },
520 	{ BGE_CHIPID_BCM5703_A3, "BCM5703 A3" },
521 	{ BGE_CHIPID_BCM5703_B0, "BCM5703 B0" },
522 	{ BGE_CHIPID_BCM5704_A0, "BCM5704 A0" },
523 	{ BGE_CHIPID_BCM5704_A1, "BCM5704 A1" },
524 	{ BGE_CHIPID_BCM5704_A2, "BCM5704 A2" },
525 	{ BGE_CHIPID_BCM5704_A3, "BCM5704 A3" },
526 	{ BGE_CHIPID_BCM5704_B0, "BCM5704 B0" },
527 	{ BGE_CHIPID_BCM5705_A0, "BCM5705 A0" },
528 	{ BGE_CHIPID_BCM5705_A1, "BCM5705 A1" },
529 	{ BGE_CHIPID_BCM5705_A2, "BCM5705 A2" },
530 	{ BGE_CHIPID_BCM5705_A3, "BCM5705 A3" },
531 	{ BGE_CHIPID_BCM5750_A0, "BCM5750 A0" },
532 	{ BGE_CHIPID_BCM5750_A1, "BCM5750 A1" },
533 	{ BGE_CHIPID_BCM5750_A3, "BCM5750 A3" },
534 	{ BGE_CHIPID_BCM5750_B0, "BCM5750 B0" },
535 	{ BGE_CHIPID_BCM5750_B1, "BCM5750 B1" },
536 	{ BGE_CHIPID_BCM5750_C0, "BCM5750 C0" },
537 	{ BGE_CHIPID_BCM5750_C1, "BCM5750 C1" },
538 	{ BGE_CHIPID_BCM5750_C2, "BCM5750 C2" },
539 	{ BGE_CHIPID_BCM5752_A0, "BCM5752 A0" },
540 	{ BGE_CHIPID_BCM5752_A1, "BCM5752 A1" },
541 	{ BGE_CHIPID_BCM5752_A2, "BCM5752 A2" },
542 	{ BGE_CHIPID_BCM5714_A0, "BCM5714 A0" },
543 	{ BGE_CHIPID_BCM5714_B0, "BCM5714 B0" },
544 	{ BGE_CHIPID_BCM5714_B3, "BCM5714 B3" },
545 	{ BGE_CHIPID_BCM5715_A0, "BCM5715 A0" },
546 	{ BGE_CHIPID_BCM5715_A1, "BCM5715 A1" },
547 	{ BGE_CHIPID_BCM5715_A3, "BCM5715 A3" },
548 	{ BGE_CHIPID_BCM5755_A0, "BCM5755 A0" },
549 	{ BGE_CHIPID_BCM5755_A1, "BCM5755 A1" },
550 	{ BGE_CHIPID_BCM5755_A2, "BCM5755 A2" },
551 	{ BGE_CHIPID_BCM5755_C0, "BCM5755 C0" },
552 	{ BGE_CHIPID_BCM5787_A0, "BCM5754/5787 A0" },
553 	{ BGE_CHIPID_BCM5787_A1, "BCM5754/5787 A1" },
554 	{ BGE_CHIPID_BCM5787_A2, "BCM5754/5787 A2" },
555 	{ BGE_CHIPID_BCM5906_A1, "BCM5906 A1" },
556 	{ BGE_CHIPID_BCM5906_A2, "BCM5906 A2" },
557 	{ 0, NULL }
558 };
559 
560 /*
561  * Some defaults for major revisions, so that newer steppings
562  * that we don't know about have a shot at working.
563  */
564 static const struct bge_revision bge_majorrevs[] = {
565 	{ BGE_ASICREV_BCM5700, "unknown BCM5700" },
566 	{ BGE_ASICREV_BCM5701, "unknown BCM5701" },
567 	{ BGE_ASICREV_BCM5703, "unknown BCM5703" },
568 	{ BGE_ASICREV_BCM5704, "unknown BCM5704" },
569 	{ BGE_ASICREV_BCM5705, "unknown BCM5705" },
570 	{ BGE_ASICREV_BCM5750, "unknown BCM5750" },
571 	{ BGE_ASICREV_BCM5714_A0, "unknown BCM5714" },
572 	{ BGE_ASICREV_BCM5714, "unknown BCM5714" },
573 	{ BGE_ASICREV_BCM5752, "unknown BCM5752 family" },
574 	{ BGE_ASICREV_BCM5755, "unknown BCM5755" },
575 	{ BGE_ASICREV_BCM5780, "unknown BCM5780" },
576 	/* 5754 and 5787 share the same ASIC ID */
577 	{ BGE_ASICREV_BCM5787, "unknown BCM5787/5787" },
578 	{ BGE_ASICREV_BCM5906, "unknown BCM5906" },
579 	{ 0, NULL }
580 };
581 
582 CFATTACH_DECL_NEW(bge, sizeof(struct bge_softc),
583     bge_probe, bge_attach, NULL, NULL);
584 
585 static u_int32_t
586 bge_readmem_ind(struct bge_softc *sc, int off)
587 {
588 	pcireg_t val;
589 
590 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, off);
591 	val = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_DATA);
592 	return val;
593 }
594 
595 static void
596 bge_writemem_ind(struct bge_softc *sc, int off, int val)
597 {
598 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, off);
599 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_DATA, val);
600 }
601 
602 #ifdef notdef
603 static u_int32_t
604 bge_readreg_ind(struct bge_softc *sc, int off)
605 {
606 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_BASEADDR, off);
607 	return (pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_DATA));
608 }
609 #endif
610 
611 static void
612 bge_writereg_ind(struct bge_softc *sc, int off, int val)
613 {
614 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_BASEADDR, off);
615 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_DATA, val);
616 }
617 
618 static void
619 bge_writemem_direct(struct bge_softc *sc, int off, int val)
620 {
621 	CSR_WRITE_4(sc, off, val);
622 }
623 
624 static void
625 bge_writembx(struct bge_softc *sc, int off, int val)
626 {
627 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906)
628 		off += BGE_LPMBX_IRQ0_HI - BGE_MBX_IRQ0_HI;
629 
630 	CSR_WRITE_4(sc, off, val);
631 }
632 
633 static u_int8_t
634 bge_nvram_getbyte(struct bge_softc *sc, int addr, u_int8_t *dest)
635 {
636 	u_int32_t access, byte = 0;
637 	int i;
638 
639 	/* Lock. */
640 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_SET1);
641 	for (i = 0; i < 8000; i++) {
642 		if (CSR_READ_4(sc, BGE_NVRAM_SWARB) & BGE_NVRAMSWARB_GNT1)
643 			break;
644 		DELAY(20);
645 	}
646 	if (i == 8000)
647 		return (1);
648 
649 	/* Enable access. */
650 	access = CSR_READ_4(sc, BGE_NVRAM_ACCESS);
651 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access | BGE_NVRAMACC_ENABLE);
652 
653 	CSR_WRITE_4(sc, BGE_NVRAM_ADDR, addr & 0xfffffffc);
654 	CSR_WRITE_4(sc, BGE_NVRAM_CMD, BGE_NVRAM_READCMD);
655 	for (i = 0; i < BGE_TIMEOUT * 10; i++) {
656 		DELAY(10);
657 		if (CSR_READ_4(sc, BGE_NVRAM_CMD) & BGE_NVRAMCMD_DONE) {
658 			DELAY(10);
659 			break;
660 		}
661 	}
662 
663 	if (i == BGE_TIMEOUT * 10) {
664 		aprint_error_dev(sc->bge_dev, "nvram read timed out\n");
665 		return (1);
666 	}
667 
668 	/* Get result. */
669 	byte = CSR_READ_4(sc, BGE_NVRAM_RDDATA);
670 
671 	*dest = (bswap32(byte) >> ((addr % 4) * 8)) & 0xFF;
672 
673 	/* Disable access. */
674 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access);
675 
676 	/* Unlock. */
677 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_CLR1);
678 	CSR_READ_4(sc, BGE_NVRAM_SWARB);
679 
680 	return (0);
681 }
682 
683 /*
684  * Read a sequence of bytes from NVRAM.
685  */
686 static int
687 bge_read_nvram(struct bge_softc *sc, u_int8_t *dest, int off, int cnt)
688 {
689 	int err = 0, i;
690 	u_int8_t byte = 0;
691 
692 	if (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5906)
693 		return (1);
694 
695 	for (i = 0; i < cnt; i++) {
696 		err = bge_nvram_getbyte(sc, off + i, &byte);
697 		if (err)
698 			break;
699 		*(dest + i) = byte;
700 	}
701 
702 	return (err ? 1 : 0);
703 }
704 
705 /*
706  * Read a byte of data stored in the EEPROM at address 'addr.' The
707  * BCM570x supports both the traditional bitbang interface and an
708  * auto access interface for reading the EEPROM. We use the auto
709  * access method.
710  */
711 static u_int8_t
712 bge_eeprom_getbyte(struct bge_softc *sc, int addr, u_int8_t *dest)
713 {
714 	int i;
715 	u_int32_t byte = 0;
716 
717 	/*
718 	 * Enable use of auto EEPROM access so we can avoid
719 	 * having to use the bitbang method.
720 	 */
721 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_AUTO_EEPROM);
722 
723 	/* Reset the EEPROM, load the clock period. */
724 	CSR_WRITE_4(sc, BGE_EE_ADDR,
725 	    BGE_EEADDR_RESET | BGE_EEHALFCLK(BGE_HALFCLK_384SCL));
726 	DELAY(20);
727 
728 	/* Issue the read EEPROM command. */
729 	CSR_WRITE_4(sc, BGE_EE_ADDR, BGE_EE_READCMD | addr);
730 
731 	/* Wait for completion */
732 	for(i = 0; i < BGE_TIMEOUT * 10; i++) {
733 		DELAY(10);
734 		if (CSR_READ_4(sc, BGE_EE_ADDR) & BGE_EEADDR_DONE)
735 			break;
736 	}
737 
738 	if (i == BGE_TIMEOUT) {
739 		aprint_error_dev(sc->bge_dev, "eeprom read timed out\n");
740 		return (0);
741 	}
742 
743 	/* Get result. */
744 	byte = CSR_READ_4(sc, BGE_EE_DATA);
745 
746 	*dest = (byte >> ((addr % 4) * 8)) & 0xFF;
747 
748 	return (0);
749 }
750 
751 /*
752  * Read a sequence of bytes from the EEPROM.
753  */
754 static int
755 bge_read_eeprom(struct bge_softc *sc, void *destv, int off, int cnt)
756 {
757 	int err = 0, i;
758 	u_int8_t byte = 0;
759 	char *dest = destv;
760 
761 	for (i = 0; i < cnt; i++) {
762 		err = bge_eeprom_getbyte(sc, off + i, &byte);
763 		if (err)
764 			break;
765 		*(dest + i) = byte;
766 	}
767 
768 	return (err ? 1 : 0);
769 }
770 
771 static int
772 bge_miibus_readreg(device_t dev, int phy, int reg)
773 {
774 	struct bge_softc *sc = device_private(dev);
775 	u_int32_t val;
776 	u_int32_t saved_autopoll;
777 	int i;
778 
779 	/*
780 	 * Broadcom's own driver always assumes the internal
781 	 * PHY is at GMII address 1. On some chips, the PHY responds
782 	 * to accesses at all addresses, which could cause us to
783 	 * bogusly attach the PHY 32 times at probe type. Always
784 	 * restricting the lookup to address 1 is simpler than
785 	 * trying to figure out which chips revisions should be
786 	 * special-cased.
787 	 */
788 	if (phy != 1)
789 		return (0);
790 
791 	/* Reading with autopolling on may trigger PCI errors */
792 	saved_autopoll = CSR_READ_4(sc, BGE_MI_MODE);
793 	if (saved_autopoll & BGE_MIMODE_AUTOPOLL) {
794 		BGE_STS_CLRBIT(sc, BGE_STS_AUTOPOLL);
795 		CSR_WRITE_4(sc, BGE_MI_MODE,
796 		    saved_autopoll &~ BGE_MIMODE_AUTOPOLL);
797 		DELAY(40);
798 	}
799 
800 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_READ|BGE_MICOMM_BUSY|
801 	    BGE_MIPHY(phy)|BGE_MIREG(reg));
802 
803 	for (i = 0; i < BGE_TIMEOUT; i++) {
804 		val = CSR_READ_4(sc, BGE_MI_COMM);
805 		if (!(val & BGE_MICOMM_BUSY))
806 			break;
807 		delay(10);
808 	}
809 
810 	if (i == BGE_TIMEOUT) {
811 		aprint_error_dev(sc->bge_dev, "PHY read timed out\n");
812 		val = 0;
813 		goto done;
814 	}
815 
816 	val = CSR_READ_4(sc, BGE_MI_COMM);
817 
818 done:
819 	if (saved_autopoll & BGE_MIMODE_AUTOPOLL) {
820 		BGE_STS_SETBIT(sc, BGE_STS_AUTOPOLL);
821 		CSR_WRITE_4(sc, BGE_MI_MODE, saved_autopoll);
822 		DELAY(40);
823 	}
824 
825 	if (val & BGE_MICOMM_READFAIL)
826 		return (0);
827 
828 	return (val & 0xFFFF);
829 }
830 
831 static void
832 bge_miibus_writereg(device_t dev, int phy, int reg, int val)
833 {
834 	struct bge_softc *sc = device_private(dev);
835 	u_int32_t saved_autopoll;
836 	int i;
837 
838 	if (phy!=1) {
839 		return;
840 	}
841 
842 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906 &&
843 	    (reg == BRGPHY_MII_1000CTL || reg == BRGPHY_MII_AUXCTL)) {
844 		return;
845 	}
846 
847 	/* Reading with autopolling on may trigger PCI errors */
848 	saved_autopoll = CSR_READ_4(sc, BGE_MI_MODE);
849 	if (saved_autopoll & BGE_MIMODE_AUTOPOLL) {
850 		delay(40);
851 		BGE_STS_CLRBIT(sc, BGE_STS_AUTOPOLL);
852 		CSR_WRITE_4(sc, BGE_MI_MODE,
853 		    saved_autopoll & (~BGE_MIMODE_AUTOPOLL));
854 		delay(10); /* 40 usec is supposed to be adequate */
855 	}
856 
857 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_WRITE | BGE_MICOMM_BUSY |
858 	    BGE_MIPHY(phy) | BGE_MIREG(reg)|val);
859 
860 	for (i = 0; i < BGE_TIMEOUT; i++) {
861 		delay(10);
862 		if (!(CSR_READ_4(sc, BGE_MI_COMM) & BGE_MICOMM_BUSY)) {
863 			delay(5);
864 			CSR_READ_4(sc, BGE_MI_COMM);
865 			break;
866 		}
867 	}
868 
869 	if (saved_autopoll & BGE_MIMODE_AUTOPOLL) {
870 		BGE_STS_SETBIT(sc, BGE_STS_AUTOPOLL);
871 		CSR_WRITE_4(sc, BGE_MI_MODE, saved_autopoll);
872 		delay(40);
873 	}
874 
875 	if (i == BGE_TIMEOUT)
876 		aprint_error_dev(sc->bge_dev, "PHY read timed out\n");
877 }
878 
879 static void
880 bge_miibus_statchg(device_t dev)
881 {
882 	struct bge_softc *sc = device_private(dev);
883 	struct mii_data *mii = &sc->bge_mii;
884 
885 	/*
886 	 * Get flow control negotiation result.
887 	 */
888 	if (IFM_SUBTYPE(mii->mii_media.ifm_cur->ifm_media) == IFM_AUTO &&
889 	    (mii->mii_media_active & IFM_ETH_FMASK) != sc->bge_flowflags) {
890 		sc->bge_flowflags = mii->mii_media_active & IFM_ETH_FMASK;
891 		mii->mii_media_active &= ~IFM_ETH_FMASK;
892 	}
893 
894 	BGE_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_PORTMODE);
895 	if (IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_T ||
896 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_SX)
897 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_GMII);
898 	else
899 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_MII);
900 
901 	if ((mii->mii_media_active & IFM_GMASK) == IFM_FDX)
902 		BGE_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
903 	else
904 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
905 
906 	/*
907 	 * 802.3x flow control
908 	 */
909 	if (sc->bge_flowflags & IFM_ETH_RXPAUSE)
910 		BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_FLOWCTL_ENABLE);
911 	else
912 		BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_FLOWCTL_ENABLE);
913 
914 	if (sc->bge_flowflags & IFM_ETH_TXPAUSE)
915 		BGE_SETBIT(sc, BGE_TX_MODE, BGE_TXMODE_FLOWCTL_ENABLE);
916 	else
917 		BGE_CLRBIT(sc, BGE_TX_MODE, BGE_TXMODE_FLOWCTL_ENABLE);
918 }
919 
920 /*
921  * Update rx threshold levels to values in a particular slot
922  * of the interrupt-mitigation table bge_rx_threshes.
923  */
924 static void
925 bge_set_thresh(struct ifnet *ifp, int lvl)
926 {
927 	struct bge_softc *sc = ifp->if_softc;
928 	int s;
929 
930 	/* For now, just save the new Rx-intr thresholds and record
931 	 * that a threshold update is pending.  Updating the hardware
932 	 * registers here (even at splhigh()) is observed to
933 	 * occasionaly cause glitches where Rx-interrupts are not
934 	 * honoured for up to 10 seconds. jonathan@NetBSD.org, 2003-04-05
935 	 */
936 	s = splnet();
937 	sc->bge_rx_coal_ticks = bge_rx_threshes[lvl].rx_ticks;
938 	sc->bge_rx_max_coal_bds = bge_rx_threshes[lvl].rx_max_bds;
939 	sc->bge_pending_rxintr_change = 1;
940 	splx(s);
941 
942 	 return;
943 }
944 
945 
946 /*
947  * Update Rx thresholds of all bge devices
948  */
949 static void
950 bge_update_all_threshes(int lvl)
951 {
952 	struct ifnet *ifp;
953 	const char * const namebuf = "bge";
954 	int namelen;
955 
956 	if (lvl < 0)
957 		lvl = 0;
958 	else if( lvl >= NBGE_RX_THRESH)
959 		lvl = NBGE_RX_THRESH - 1;
960 
961 	namelen = strlen(namebuf);
962 	/*
963 	 * Now search all the interfaces for this name/number
964 	 */
965 	IFNET_FOREACH(ifp) {
966 		if (strncmp(ifp->if_xname, namebuf, namelen) != 0)
967 		      continue;
968 		/* We got a match: update if doing auto-threshold-tuning */
969 		if (bge_auto_thresh)
970 			bge_set_thresh(ifp, lvl);
971 	}
972 }
973 
974 /*
975  * Handle events that have triggered interrupts.
976  */
977 static void
978 bge_handle_events(struct bge_softc *sc)
979 {
980 
981 	return;
982 }
983 
984 /*
985  * Memory management for jumbo frames.
986  */
987 
988 static int
989 bge_alloc_jumbo_mem(struct bge_softc *sc)
990 {
991 	char *ptr, *kva;
992 	bus_dma_segment_t	seg;
993 	int		i, rseg, state, error;
994 	struct bge_jpool_entry   *entry;
995 
996 	state = error = 0;
997 
998 	/* Grab a big chunk o' storage. */
999 	if (bus_dmamem_alloc(sc->bge_dmatag, BGE_JMEM, PAGE_SIZE, 0,
1000 	     &seg, 1, &rseg, BUS_DMA_NOWAIT)) {
1001 		aprint_error_dev(sc->bge_dev, "can't alloc rx buffers\n");
1002 		return ENOBUFS;
1003 	}
1004 
1005 	state = 1;
1006 	if (bus_dmamem_map(sc->bge_dmatag, &seg, rseg, BGE_JMEM, (void **)&kva,
1007 	    BUS_DMA_NOWAIT)) {
1008 		aprint_error_dev(sc->bge_dev,
1009 		    "can't map DMA buffers (%d bytes)\n", (int)BGE_JMEM);
1010 		error = ENOBUFS;
1011 		goto out;
1012 	}
1013 
1014 	state = 2;
1015 	if (bus_dmamap_create(sc->bge_dmatag, BGE_JMEM, 1, BGE_JMEM, 0,
1016 	    BUS_DMA_NOWAIT, &sc->bge_cdata.bge_rx_jumbo_map)) {
1017 		aprint_error_dev(sc->bge_dev, "can't create DMA map\n");
1018 		error = ENOBUFS;
1019 		goto out;
1020 	}
1021 
1022 	state = 3;
1023 	if (bus_dmamap_load(sc->bge_dmatag, sc->bge_cdata.bge_rx_jumbo_map,
1024 	    kva, BGE_JMEM, NULL, BUS_DMA_NOWAIT)) {
1025 		aprint_error_dev(sc->bge_dev, "can't load DMA map\n");
1026 		error = ENOBUFS;
1027 		goto out;
1028 	}
1029 
1030 	state = 4;
1031 	sc->bge_cdata.bge_jumbo_buf = (void *)kva;
1032 	DPRINTFN(1,("bge_jumbo_buf = %p\n", sc->bge_cdata.bge_jumbo_buf));
1033 
1034 	SLIST_INIT(&sc->bge_jfree_listhead);
1035 	SLIST_INIT(&sc->bge_jinuse_listhead);
1036 
1037 	/*
1038 	 * Now divide it up into 9K pieces and save the addresses
1039 	 * in an array.
1040 	 */
1041 	ptr = sc->bge_cdata.bge_jumbo_buf;
1042 	for (i = 0; i < BGE_JSLOTS; i++) {
1043 		sc->bge_cdata.bge_jslots[i] = ptr;
1044 		ptr += BGE_JLEN;
1045 		entry = malloc(sizeof(struct bge_jpool_entry),
1046 		    M_DEVBUF, M_NOWAIT);
1047 		if (entry == NULL) {
1048 			aprint_error_dev(sc->bge_dev,
1049 			    "no memory for jumbo buffer queue!\n");
1050 			error = ENOBUFS;
1051 			goto out;
1052 		}
1053 		entry->slot = i;
1054 		SLIST_INSERT_HEAD(&sc->bge_jfree_listhead,
1055 				 entry, jpool_entries);
1056 	}
1057 out:
1058 	if (error != 0) {
1059 		switch (state) {
1060 		case 4:
1061 			bus_dmamap_unload(sc->bge_dmatag,
1062 			    sc->bge_cdata.bge_rx_jumbo_map);
1063 		case 3:
1064 			bus_dmamap_destroy(sc->bge_dmatag,
1065 			    sc->bge_cdata.bge_rx_jumbo_map);
1066 		case 2:
1067 			bus_dmamem_unmap(sc->bge_dmatag, kva, BGE_JMEM);
1068 		case 1:
1069 			bus_dmamem_free(sc->bge_dmatag, &seg, rseg);
1070 			break;
1071 		default:
1072 			break;
1073 		}
1074 	}
1075 
1076 	return error;
1077 }
1078 
1079 /*
1080  * Allocate a jumbo buffer.
1081  */
1082 static void *
1083 bge_jalloc(struct bge_softc *sc)
1084 {
1085 	struct bge_jpool_entry   *entry;
1086 
1087 	entry = SLIST_FIRST(&sc->bge_jfree_listhead);
1088 
1089 	if (entry == NULL) {
1090 		aprint_error_dev(sc->bge_dev, "no free jumbo buffers\n");
1091 		return (NULL);
1092 	}
1093 
1094 	SLIST_REMOVE_HEAD(&sc->bge_jfree_listhead, jpool_entries);
1095 	SLIST_INSERT_HEAD(&sc->bge_jinuse_listhead, entry, jpool_entries);
1096 	return (sc->bge_cdata.bge_jslots[entry->slot]);
1097 }
1098 
1099 /*
1100  * Release a jumbo buffer.
1101  */
1102 static void
1103 bge_jfree(struct mbuf *m, void *buf, size_t size, void *arg)
1104 {
1105 	struct bge_jpool_entry *entry;
1106 	struct bge_softc *sc;
1107 	int i, s;
1108 
1109 	/* Extract the softc struct pointer. */
1110 	sc = (struct bge_softc *)arg;
1111 
1112 	if (sc == NULL)
1113 		panic("bge_jfree: can't find softc pointer!");
1114 
1115 	/* calculate the slot this buffer belongs to */
1116 
1117 	i = ((char *)buf
1118 	     - (char *)sc->bge_cdata.bge_jumbo_buf) / BGE_JLEN;
1119 
1120 	if ((i < 0) || (i >= BGE_JSLOTS))
1121 		panic("bge_jfree: asked to free buffer that we don't manage!");
1122 
1123 	s = splvm();
1124 	entry = SLIST_FIRST(&sc->bge_jinuse_listhead);
1125 	if (entry == NULL)
1126 		panic("bge_jfree: buffer not in use!");
1127 	entry->slot = i;
1128 	SLIST_REMOVE_HEAD(&sc->bge_jinuse_listhead, jpool_entries);
1129 	SLIST_INSERT_HEAD(&sc->bge_jfree_listhead, entry, jpool_entries);
1130 
1131 	if (__predict_true(m != NULL))
1132   		pool_cache_put(mb_cache, m);
1133 	splx(s);
1134 }
1135 
1136 
1137 /*
1138  * Intialize a standard receive ring descriptor.
1139  */
1140 static int
1141 bge_newbuf_std(struct bge_softc *sc, int i, struct mbuf *m, bus_dmamap_t dmamap)
1142 {
1143 	struct mbuf		*m_new = NULL;
1144 	struct bge_rx_bd	*r;
1145 	int			error;
1146 
1147 	if (dmamap == NULL) {
1148 		error = bus_dmamap_create(sc->bge_dmatag, MCLBYTES, 1,
1149 		    MCLBYTES, 0, BUS_DMA_NOWAIT, &dmamap);
1150 		if (error != 0)
1151 			return error;
1152 	}
1153 
1154 	sc->bge_cdata.bge_rx_std_map[i] = dmamap;
1155 
1156 	if (m == NULL) {
1157 		MGETHDR(m_new, M_DONTWAIT, MT_DATA);
1158 		if (m_new == NULL)
1159 			return (ENOBUFS);
1160 
1161 		MCLGET(m_new, M_DONTWAIT);
1162 		if (!(m_new->m_flags & M_EXT)) {
1163 			m_freem(m_new);
1164 			return (ENOBUFS);
1165 		}
1166 		m_new->m_len = m_new->m_pkthdr.len = MCLBYTES;
1167 
1168 	} else {
1169 		m_new = m;
1170 		m_new->m_len = m_new->m_pkthdr.len = MCLBYTES;
1171 		m_new->m_data = m_new->m_ext.ext_buf;
1172 	}
1173 	if (!(sc->bge_flags & BGE_RX_ALIGNBUG))
1174 	    m_adj(m_new, ETHER_ALIGN);
1175 	if (bus_dmamap_load_mbuf(sc->bge_dmatag, dmamap, m_new,
1176 	    BUS_DMA_READ|BUS_DMA_NOWAIT))
1177 		return (ENOBUFS);
1178 	bus_dmamap_sync(sc->bge_dmatag, dmamap, 0, dmamap->dm_mapsize,
1179 	    BUS_DMASYNC_PREREAD);
1180 
1181 	sc->bge_cdata.bge_rx_std_chain[i] = m_new;
1182 	r = &sc->bge_rdata->bge_rx_std_ring[i];
1183 	bge_set_hostaddr(&r->bge_addr,
1184 	    dmamap->dm_segs[0].ds_addr);
1185 	r->bge_flags = BGE_RXBDFLAG_END;
1186 	r->bge_len = m_new->m_len;
1187 	r->bge_idx = i;
1188 
1189 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
1190 	    offsetof(struct bge_ring_data, bge_rx_std_ring) +
1191 		i * sizeof (struct bge_rx_bd),
1192 	    sizeof (struct bge_rx_bd),
1193 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1194 
1195 	return (0);
1196 }
1197 
1198 /*
1199  * Initialize a jumbo receive ring descriptor. This allocates
1200  * a jumbo buffer from the pool managed internally by the driver.
1201  */
1202 static int
1203 bge_newbuf_jumbo(struct bge_softc *sc, int i, struct mbuf *m)
1204 {
1205 	struct mbuf *m_new = NULL;
1206 	struct bge_rx_bd *r;
1207 	void *buf = NULL;
1208 
1209 	if (m == NULL) {
1210 
1211 		/* Allocate the mbuf. */
1212 		MGETHDR(m_new, M_DONTWAIT, MT_DATA);
1213 		if (m_new == NULL)
1214 			return (ENOBUFS);
1215 
1216 		/* Allocate the jumbo buffer */
1217 		buf = bge_jalloc(sc);
1218 		if (buf == NULL) {
1219 			m_freem(m_new);
1220 			aprint_error_dev(sc->bge_dev,
1221 			    "jumbo allocation failed -- packet dropped!\n");
1222 			return (ENOBUFS);
1223 		}
1224 
1225 		/* Attach the buffer to the mbuf. */
1226 		m_new->m_len = m_new->m_pkthdr.len = BGE_JUMBO_FRAMELEN;
1227 		MEXTADD(m_new, buf, BGE_JUMBO_FRAMELEN, M_DEVBUF,
1228 		    bge_jfree, sc);
1229 		m_new->m_flags |= M_EXT_RW;
1230 	} else {
1231 		m_new = m;
1232 		buf = m_new->m_data = m_new->m_ext.ext_buf;
1233 		m_new->m_ext.ext_size = BGE_JUMBO_FRAMELEN;
1234 	}
1235 	if (!(sc->bge_flags & BGE_RX_ALIGNBUG))
1236 	    m_adj(m_new, ETHER_ALIGN);
1237 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_cdata.bge_rx_jumbo_map,
1238 	    mtod(m_new, char *) - (char *)sc->bge_cdata.bge_jumbo_buf, BGE_JLEN,
1239 	    BUS_DMASYNC_PREREAD);
1240 	/* Set up the descriptor. */
1241 	r = &sc->bge_rdata->bge_rx_jumbo_ring[i];
1242 	sc->bge_cdata.bge_rx_jumbo_chain[i] = m_new;
1243 	bge_set_hostaddr(&r->bge_addr, BGE_JUMBO_DMA_ADDR(sc, m_new));
1244 	r->bge_flags = BGE_RXBDFLAG_END|BGE_RXBDFLAG_JUMBO_RING;
1245 	r->bge_len = m_new->m_len;
1246 	r->bge_idx = i;
1247 
1248 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
1249 	    offsetof(struct bge_ring_data, bge_rx_jumbo_ring) +
1250 		i * sizeof (struct bge_rx_bd),
1251 	    sizeof (struct bge_rx_bd),
1252 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1253 
1254 	return (0);
1255 }
1256 
1257 /*
1258  * The standard receive ring has 512 entries in it. At 2K per mbuf cluster,
1259  * that's 1MB or memory, which is a lot. For now, we fill only the first
1260  * 256 ring entries and hope that our CPU is fast enough to keep up with
1261  * the NIC.
1262  */
1263 static int
1264 bge_init_rx_ring_std(struct bge_softc *sc)
1265 {
1266 	int i;
1267 
1268 	if (sc->bge_flags & BGE_RXRING_VALID)
1269 		return 0;
1270 
1271 	for (i = 0; i < BGE_SSLOTS; i++) {
1272 		if (bge_newbuf_std(sc, i, NULL, 0) == ENOBUFS)
1273 			return (ENOBUFS);
1274 	}
1275 
1276 	sc->bge_std = i - 1;
1277 	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bge_std);
1278 
1279 	sc->bge_flags |= BGE_RXRING_VALID;
1280 
1281 	return (0);
1282 }
1283 
1284 static void
1285 bge_free_rx_ring_std(struct bge_softc *sc)
1286 {
1287 	int i;
1288 
1289 	if (!(sc->bge_flags & BGE_RXRING_VALID))
1290 		return;
1291 
1292 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1293 		if (sc->bge_cdata.bge_rx_std_chain[i] != NULL) {
1294 			m_freem(sc->bge_cdata.bge_rx_std_chain[i]);
1295 			sc->bge_cdata.bge_rx_std_chain[i] = NULL;
1296 			bus_dmamap_destroy(sc->bge_dmatag,
1297 			    sc->bge_cdata.bge_rx_std_map[i]);
1298 		}
1299 		memset((char *)&sc->bge_rdata->bge_rx_std_ring[i], 0,
1300 		    sizeof(struct bge_rx_bd));
1301 	}
1302 
1303 	sc->bge_flags &= ~BGE_RXRING_VALID;
1304 }
1305 
1306 static int
1307 bge_init_rx_ring_jumbo(struct bge_softc *sc)
1308 {
1309 	int i;
1310 	volatile struct bge_rcb *rcb;
1311 
1312 	if (sc->bge_flags & BGE_JUMBO_RXRING_VALID)
1313 		return 0;
1314 
1315 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1316 		if (bge_newbuf_jumbo(sc, i, NULL) == ENOBUFS)
1317 			return (ENOBUFS);
1318 	};
1319 
1320 	sc->bge_jumbo = i - 1;
1321 	sc->bge_flags |= BGE_JUMBO_RXRING_VALID;
1322 
1323 	rcb = &sc->bge_rdata->bge_info.bge_jumbo_rx_rcb;
1324 	rcb->bge_maxlen_flags = 0;
1325 	CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1326 
1327 	bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
1328 
1329 	return (0);
1330 }
1331 
1332 static void
1333 bge_free_rx_ring_jumbo(struct bge_softc *sc)
1334 {
1335 	int i;
1336 
1337 	if (!(sc->bge_flags & BGE_JUMBO_RXRING_VALID))
1338 		return;
1339 
1340 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1341 		if (sc->bge_cdata.bge_rx_jumbo_chain[i] != NULL) {
1342 			m_freem(sc->bge_cdata.bge_rx_jumbo_chain[i]);
1343 			sc->bge_cdata.bge_rx_jumbo_chain[i] = NULL;
1344 		}
1345 		memset((char *)&sc->bge_rdata->bge_rx_jumbo_ring[i], 0,
1346 		    sizeof(struct bge_rx_bd));
1347 	}
1348 
1349 	sc->bge_flags &= ~BGE_JUMBO_RXRING_VALID;
1350 }
1351 
1352 static void
1353 bge_free_tx_ring(struct bge_softc *sc)
1354 {
1355 	int i, freed;
1356 	struct txdmamap_pool_entry *dma;
1357 
1358 	if (!(sc->bge_flags & BGE_TXRING_VALID))
1359 		return;
1360 
1361 	freed = 0;
1362 
1363 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
1364 		if (sc->bge_cdata.bge_tx_chain[i] != NULL) {
1365 			freed++;
1366 			m_freem(sc->bge_cdata.bge_tx_chain[i]);
1367 			sc->bge_cdata.bge_tx_chain[i] = NULL;
1368 			SLIST_INSERT_HEAD(&sc->txdma_list, sc->txdma[i],
1369 					    link);
1370 			sc->txdma[i] = 0;
1371 		}
1372 		memset((char *)&sc->bge_rdata->bge_tx_ring[i], 0,
1373 		    sizeof(struct bge_tx_bd));
1374 	}
1375 
1376 	while ((dma = SLIST_FIRST(&sc->txdma_list))) {
1377 		SLIST_REMOVE_HEAD(&sc->txdma_list, link);
1378 		bus_dmamap_destroy(sc->bge_dmatag, dma->dmamap);
1379 		free(dma, M_DEVBUF);
1380 	}
1381 
1382 	sc->bge_flags &= ~BGE_TXRING_VALID;
1383 }
1384 
1385 static int
1386 bge_init_tx_ring(struct bge_softc *sc)
1387 {
1388 	int i;
1389 	bus_dmamap_t dmamap;
1390 	struct txdmamap_pool_entry *dma;
1391 
1392 	if (sc->bge_flags & BGE_TXRING_VALID)
1393 		return 0;
1394 
1395 	sc->bge_txcnt = 0;
1396 	sc->bge_tx_saved_considx = 0;
1397 
1398 	/* Initialize transmit producer index for host-memory send ring. */
1399 	sc->bge_tx_prodidx = 0;
1400 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1401 	/* 5700 b2 errata */
1402 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX)
1403 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1404 
1405 	/* NIC-memory send ring not used; initialize to zero. */
1406 	bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1407 	/* 5700 b2 errata */
1408 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX)
1409 		bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1410 
1411 	SLIST_INIT(&sc->txdma_list);
1412 	for (i = 0; i < BGE_RSLOTS; i++) {
1413 		if (bus_dmamap_create(sc->bge_dmatag, BGE_TXDMA_MAX,
1414 		    BGE_NTXSEG, ETHER_MAX_LEN_JUMBO, 0, BUS_DMA_NOWAIT,
1415 		    &dmamap))
1416 			return (ENOBUFS);
1417 		if (dmamap == NULL)
1418 			panic("dmamap NULL in bge_init_tx_ring");
1419 		dma = malloc(sizeof(*dma), M_DEVBUF, M_NOWAIT);
1420 		if (dma == NULL) {
1421 			aprint_error_dev(sc->bge_dev,
1422 			    "can't alloc txdmamap_pool_entry\n");
1423 			bus_dmamap_destroy(sc->bge_dmatag, dmamap);
1424 			return (ENOMEM);
1425 		}
1426 		dma->dmamap = dmamap;
1427 		SLIST_INSERT_HEAD(&sc->txdma_list, dma, link);
1428 	}
1429 
1430 	sc->bge_flags |= BGE_TXRING_VALID;
1431 
1432 	return (0);
1433 }
1434 
1435 static void
1436 bge_setmulti(struct bge_softc *sc)
1437 {
1438 	struct ethercom		*ac = &sc->ethercom;
1439 	struct ifnet		*ifp = &ac->ec_if;
1440 	struct ether_multi	*enm;
1441 	struct ether_multistep  step;
1442 	u_int32_t		hashes[4] = { 0, 0, 0, 0 };
1443 	u_int32_t		h;
1444 	int			i;
1445 
1446 	if (ifp->if_flags & IFF_PROMISC)
1447 		goto allmulti;
1448 
1449 	/* Now program new ones. */
1450 	ETHER_FIRST_MULTI(step, ac, enm);
1451 	while (enm != NULL) {
1452 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
1453 			/*
1454 			 * We must listen to a range of multicast addresses.
1455 			 * For now, just accept all multicasts, rather than
1456 			 * trying to set only those filter bits needed to match
1457 			 * the range.  (At this time, the only use of address
1458 			 * ranges is for IP multicast routing, for which the
1459 			 * range is big enough to require all bits set.)
1460 			 */
1461 			goto allmulti;
1462 		}
1463 
1464 		h = ether_crc32_le(enm->enm_addrlo, ETHER_ADDR_LEN);
1465 
1466 		/* Just want the 7 least-significant bits. */
1467 		h &= 0x7f;
1468 
1469 		hashes[(h & 0x60) >> 5] |= 1 << (h & 0x1F);
1470 		ETHER_NEXT_MULTI(step, enm);
1471 	}
1472 
1473 	ifp->if_flags &= ~IFF_ALLMULTI;
1474 	goto setit;
1475 
1476  allmulti:
1477 	ifp->if_flags |= IFF_ALLMULTI;
1478 	hashes[0] = hashes[1] = hashes[2] = hashes[3] = 0xffffffff;
1479 
1480  setit:
1481 	for (i = 0; i < 4; i++)
1482 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), hashes[i]);
1483 }
1484 
1485 const int bge_swapbits[] = {
1486 	0,
1487 	BGE_MODECTL_BYTESWAP_DATA,
1488 	BGE_MODECTL_WORDSWAP_DATA,
1489 	BGE_MODECTL_BYTESWAP_NONFRAME,
1490 	BGE_MODECTL_WORDSWAP_NONFRAME,
1491 
1492 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_WORDSWAP_DATA,
1493 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_BYTESWAP_NONFRAME,
1494 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_WORDSWAP_NONFRAME,
1495 
1496 	BGE_MODECTL_WORDSWAP_DATA|BGE_MODECTL_BYTESWAP_NONFRAME,
1497 	BGE_MODECTL_WORDSWAP_DATA|BGE_MODECTL_WORDSWAP_NONFRAME,
1498 
1499 	BGE_MODECTL_BYTESWAP_NONFRAME|BGE_MODECTL_WORDSWAP_NONFRAME,
1500 
1501 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_WORDSWAP_DATA|
1502 	    BGE_MODECTL_BYTESWAP_NONFRAME,
1503 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_WORDSWAP_DATA|
1504 	    BGE_MODECTL_WORDSWAP_NONFRAME,
1505 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_BYTESWAP_NONFRAME|
1506 	    BGE_MODECTL_WORDSWAP_NONFRAME,
1507 	BGE_MODECTL_WORDSWAP_DATA|BGE_MODECTL_BYTESWAP_NONFRAME|
1508 	    BGE_MODECTL_WORDSWAP_NONFRAME,
1509 
1510 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_WORDSWAP_DATA|
1511 	    BGE_MODECTL_BYTESWAP_NONFRAME|BGE_MODECTL_WORDSWAP_NONFRAME,
1512 };
1513 
1514 int bge_swapindex = 0;
1515 
1516 /*
1517  * Do endian, PCI and DMA initialization. Also check the on-board ROM
1518  * self-test results.
1519  */
1520 static int
1521 bge_chipinit(struct bge_softc *sc)
1522 {
1523 	u_int32_t		cachesize;
1524 	int			i;
1525 	u_int32_t		dma_rw_ctl;
1526 
1527 
1528 	/* Set endianness before we access any non-PCI registers. */
1529 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL,
1530 	    BGE_INIT);
1531 
1532 	/* Set power state to D0. */
1533 	bge_setpowerstate(sc, 0);
1534 
1535 	/*
1536 	 * Check the 'ROM failed' bit on the RX CPU to see if
1537 	 * self-tests passed.
1538 	 */
1539 	if (CSR_READ_4(sc, BGE_RXCPU_MODE) & BGE_RXCPUMODE_ROMFAIL) {
1540 		aprint_error_dev(sc->bge_dev,
1541 		    "RX CPU self-diagnostics failed!\n");
1542 		return (ENODEV);
1543 	}
1544 
1545 	/* Clear the MAC control register */
1546 	CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
1547 
1548 	/*
1549 	 * Clear the MAC statistics block in the NIC's
1550 	 * internal memory.
1551 	 */
1552 	for (i = BGE_STATS_BLOCK;
1553 	    i < BGE_STATS_BLOCK_END + 1; i += sizeof(u_int32_t))
1554 		BGE_MEMWIN_WRITE(sc->sc_pc, sc->sc_pcitag, i, 0);
1555 
1556 	for (i = BGE_STATUS_BLOCK;
1557 	    i < BGE_STATUS_BLOCK_END + 1; i += sizeof(u_int32_t))
1558 		BGE_MEMWIN_WRITE(sc->sc_pc, sc->sc_pcitag, i, 0);
1559 
1560 	/* Set up the PCI DMA control register. */
1561 	if (sc->bge_flags & BGE_PCIE) {
1562 	  u_int32_t device_ctl;
1563 
1564 		/* From FreeBSD */
1565 		DPRINTFN(4, ("(%s: PCI-Express DMA setting)\n",
1566 		    device_xname(sc->bge_dev)));
1567 		dma_rw_ctl = (BGE_PCI_READ_CMD | BGE_PCI_WRITE_CMD |
1568 		    (0xf << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1569 		    (0x2 << BGE_PCIDMARWCTL_WR_WAT_SHIFT));
1570 
1571 		/* jonathan: alternative from Linux driver */
1572 #define DMA_CTRL_WRITE_PCIE_H20MARK_128         0x00180000
1573 #define DMA_CTRL_WRITE_PCIE_H20MARK_256         0x00380000
1574 
1575 		dma_rw_ctl =   0x76000000; /* XXX XXX XXX */;
1576 		device_ctl = pci_conf_read(sc->sc_pc, sc->sc_pcitag,
1577 					   BGE_PCI_CONF_DEV_CTRL);
1578 		aprint_debug_dev(sc->bge_dev, "pcie mode=0x%x\n", device_ctl);
1579 
1580 		if ((device_ctl & 0x00e0) && 0) {
1581 			/*
1582 			 * XXX jonathan@NetBSD.org:
1583 			 * This clause is exactly what the Broadcom-supplied
1584 			 * Linux does; but given overall register programming
1585 			 * by if_bge(4), this larger DMA-write watermark
1586 			 * value causes bcm5721 chips to totally wedge.
1587 			 */
1588 			dma_rw_ctl |= BGE_PCIDMA_RWCTL_PCIE_WRITE_WATRMARK_256;
1589 		} else {
1590 			dma_rw_ctl |= BGE_PCIDMA_RWCTL_PCIE_WRITE_WATRMARK_128;
1591 		}
1592 	} else if (sc->bge_flags & BGE_PCIX){
1593 	  	DPRINTFN(4, ("(:%s: PCI-X DMA setting)\n",
1594 		    device_xname(sc->bge_dev)));
1595 		/* PCI-X bus */
1596 		dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD |
1597 		    (0x3 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1598 		    (0x3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT) |
1599 		    (0x0F);
1600 		/*
1601 		 * 5703 and 5704 need ONEDMA_AT_ONCE as a workaround
1602 		 * for hardware bugs, which means we should also clear
1603 		 * the low-order MINDMA bits.  In addition, the 5704
1604 		 * uses a different encoding of read/write watermarks.
1605 		 */
1606 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704) {
1607 			dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD |
1608 			  /* should be 0x1f0000 */
1609 			  (0x7 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1610 			  (0x3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1611 			dma_rw_ctl |= BGE_PCIDMARWCTL_ONEDMA_ATONCE;
1612 		}
1613 		else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703) {
1614 			dma_rw_ctl &=  0xfffffff0;
1615 			dma_rw_ctl |= BGE_PCIDMARWCTL_ONEDMA_ATONCE;
1616 		}
1617 		else if (BGE_IS_5714_FAMILY(sc)) {
1618 			dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD;
1619 			dma_rw_ctl &= ~BGE_PCIDMARWCTL_ONEDMA_ATONCE; /* XXX */
1620 			/* XXX magic values, Broadcom-supplied Linux driver */
1621 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5780)
1622 				dma_rw_ctl |= (1 << 20) | (1 << 18) |
1623 				  BGE_PCIDMARWCTL_ONEDMA_ATONCE;
1624 			else
1625 				dma_rw_ctl |= (1<<20) | (1<<18) | (1 << 15);
1626 		}
1627 	} else {
1628 		/* Conventional PCI bus */
1629 	  	DPRINTFN(4, ("(%s: PCI 2.2 DMA setting)\n",
1630 		    device_xname(sc->bge_dev)));
1631 		dma_rw_ctl = (BGE_PCI_READ_CMD | BGE_PCI_WRITE_CMD |
1632 		   (0x7 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1633 		   (0x7 << BGE_PCIDMARWCTL_WR_WAT_SHIFT));
1634 		if (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5705 &&
1635 		    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5750)
1636 			dma_rw_ctl |= 0x0F;
1637 	}
1638 
1639 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
1640 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5701)
1641 		dma_rw_ctl |= BGE_PCIDMARWCTL_USE_MRM |
1642 		    BGE_PCIDMARWCTL_ASRT_ALL_BE;
1643 
1644 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703 ||
1645 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704)
1646 		dma_rw_ctl &= ~BGE_PCIDMARWCTL_MINDMA;
1647 
1648 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_DMA_RW_CTL, dma_rw_ctl);
1649 
1650 	/*
1651 	 * Set up general mode register.
1652 	 */
1653 	CSR_WRITE_4(sc, BGE_MODE_CTL, BGE_DMA_SWAP_OPTIONS |
1654 		    BGE_MODECTL_MAC_ATTN_INTR | BGE_MODECTL_HOST_SEND_BDS |
1655 		    BGE_MODECTL_TX_NO_PHDR_CSUM | BGE_MODECTL_RX_NO_PHDR_CSUM);
1656 
1657 	/* Get cache line size. */
1658 	cachesize = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CACHESZ);
1659 
1660 	/*
1661 	 * Avoid violating PCI spec on certain chip revs.
1662 	 */
1663 	if (pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD) &
1664 	    PCIM_CMD_MWIEN) {
1665 		switch(cachesize) {
1666 		case 1:
1667 			PCI_SETBIT(sc->sc_pc, sc->sc_pcitag, BGE_PCI_DMA_RW_CTL,
1668 				   BGE_PCI_WRITE_BNDRY_16BYTES);
1669 			break;
1670 		case 2:
1671 			PCI_SETBIT(sc->sc_pc, sc->sc_pcitag, BGE_PCI_DMA_RW_CTL,
1672 				   BGE_PCI_WRITE_BNDRY_32BYTES);
1673 			break;
1674 		case 4:
1675 			PCI_SETBIT(sc->sc_pc, sc->sc_pcitag, BGE_PCI_DMA_RW_CTL,
1676 				   BGE_PCI_WRITE_BNDRY_64BYTES);
1677 			break;
1678 		case 8:
1679 			PCI_SETBIT(sc->sc_pc, sc->sc_pcitag, BGE_PCI_DMA_RW_CTL,
1680 				   BGE_PCI_WRITE_BNDRY_128BYTES);
1681 			break;
1682 		case 16:
1683 			PCI_SETBIT(sc->sc_pc, sc->sc_pcitag, BGE_PCI_DMA_RW_CTL,
1684 				   BGE_PCI_WRITE_BNDRY_256BYTES);
1685 			break;
1686 		case 32:
1687 			PCI_SETBIT(sc->sc_pc, sc->sc_pcitag, BGE_PCI_DMA_RW_CTL,
1688 				   BGE_PCI_WRITE_BNDRY_512BYTES);
1689 			break;
1690 		case 64:
1691 			PCI_SETBIT(sc->sc_pc, sc->sc_pcitag, BGE_PCI_DMA_RW_CTL,
1692 				   BGE_PCI_WRITE_BNDRY_1024BYTES);
1693 			break;
1694 		default:
1695 		/* Disable PCI memory write and invalidate. */
1696 #if 0
1697 			if (bootverbose)
1698 				aprint_error_dev(sc->bge_dev,
1699 				    "cache line size %d not supported "
1700 				    "disabling PCI MWI\n",
1701 #endif
1702 			PCI_CLRBIT(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD,
1703 			    PCIM_CMD_MWIEN);
1704 			break;
1705 		}
1706 	}
1707 
1708 	/*
1709 	 * Disable memory write invalidate.  Apparently it is not supported
1710 	 * properly by these devices.
1711 	 */
1712 	PCI_CLRBIT(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD, PCIM_CMD_MWIEN);
1713 
1714 
1715 #ifdef __brokenalpha__
1716 	/*
1717 	 * Must insure that we do not cross an 8K (bytes) boundary
1718 	 * for DMA reads.  Our highest limit is 1K bytes.  This is a
1719 	 * restriction on some ALPHA platforms with early revision
1720 	 * 21174 PCI chipsets, such as the AlphaPC 164lx
1721 	 */
1722 	PCI_SETBIT(sc, BGE_PCI_DMA_RW_CTL, BGE_PCI_READ_BNDRY_1024, 4);
1723 #endif
1724 
1725 	/* Set the timer prescaler (always 66MHz) */
1726 	CSR_WRITE_4(sc, BGE_MISC_CFG, 65 << 1/*BGE_32BITTIME_66MHZ*/);
1727 
1728 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
1729 		DELAY(40);	/* XXX */
1730 
1731 		/* Put PHY into ready state */
1732 		BGE_CLRBIT(sc, BGE_MISC_CFG, BGE_MISCCFG_EPHY_IDDQ);
1733 		CSR_READ_4(sc, BGE_MISC_CFG); /* Flush */
1734 		DELAY(40);
1735 	}
1736 
1737 	return (0);
1738 }
1739 
1740 static int
1741 bge_blockinit(struct bge_softc *sc)
1742 {
1743 	volatile struct bge_rcb		*rcb;
1744 	bus_size_t		rcb_addr;
1745 	int			i;
1746 	struct ifnet		*ifp = &sc->ethercom.ec_if;
1747 	bge_hostaddr		taddr;
1748 	u_int32_t		val;
1749 
1750 	/*
1751 	 * Initialize the memory window pointer register so that
1752 	 * we can access the first 32K of internal NIC RAM. This will
1753 	 * allow us to set up the TX send ring RCBs and the RX return
1754 	 * ring RCBs, plus other things which live in NIC memory.
1755 	 */
1756 
1757 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, 0);
1758 
1759 	/* Configure mbuf memory pool */
1760 	if (!(BGE_IS_5705_OR_BEYOND(sc))) {
1761 		if (sc->bge_extram) {
1762 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_BASEADDR,
1763 			    BGE_EXT_SSRAM);
1764 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704)
1765 				CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x10000);
1766 			else
1767 				CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x18000);
1768 		} else {
1769 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_BASEADDR,
1770 			    BGE_BUFFPOOL_1);
1771 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704)
1772 				CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x10000);
1773 			else
1774 				CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x18000);
1775 		}
1776 
1777 		/* Configure DMA resource pool */
1778 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_BASEADDR,
1779 		    BGE_DMA_DESCRIPTORS);
1780 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LEN, 0x2000);
1781 	}
1782 
1783 	/* Configure mbuf pool watermarks */
1784 #ifdef ORIG_WPAUL_VALUES
1785 	CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 24);
1786 	CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 24);
1787 	CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 48);
1788 #else
1789 
1790 	/* new broadcom docs strongly recommend these: */
1791 	if (!BGE_IS_5705_OR_BEYOND(sc)) {
1792 		if (ifp->if_mtu > ETHER_MAX_LEN) {
1793 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x50);
1794 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x20);
1795 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1796 		} else {
1797 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 304);
1798 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 152);
1799 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 380);
1800 		}
1801 	} else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
1802 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1803 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x04);
1804 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x10);
1805 	} else {
1806 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1807 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x10);
1808 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1809 	}
1810 #endif
1811 
1812 	/* Configure DMA resource watermarks */
1813 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LOWAT, 5);
1814 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_HIWAT, 10);
1815 
1816 	/* Enable buffer manager */
1817 	if (!BGE_IS_5705_OR_BEYOND(sc)) {
1818 		CSR_WRITE_4(sc, BGE_BMAN_MODE,
1819 		    BGE_BMANMODE_ENABLE | BGE_BMANMODE_LOMBUF_ATTN);
1820 
1821 		/* Poll for buffer manager start indication */
1822 		for (i = 0; i < BGE_TIMEOUT; i++) {
1823 			if (CSR_READ_4(sc, BGE_BMAN_MODE) & BGE_BMANMODE_ENABLE)
1824 				break;
1825 			DELAY(10);
1826 		}
1827 
1828 		if (i == BGE_TIMEOUT) {
1829 			aprint_error_dev(sc->bge_dev,
1830 			    "buffer manager failed to start\n");
1831 			return (ENXIO);
1832 		}
1833 	}
1834 
1835 	/* Enable flow-through queues */
1836 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
1837 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
1838 
1839 	/* Wait until queue initialization is complete */
1840 	for (i = 0; i < BGE_TIMEOUT; i++) {
1841 		if (CSR_READ_4(sc, BGE_FTQ_RESET) == 0)
1842 			break;
1843 		DELAY(10);
1844 	}
1845 
1846 	if (i == BGE_TIMEOUT) {
1847 		aprint_error_dev(sc->bge_dev,
1848 		    "flow-through queue init failed\n");
1849 		return (ENXIO);
1850 	}
1851 
1852 	/* Initialize the standard RX ring control block */
1853 	rcb = &sc->bge_rdata->bge_info.bge_std_rx_rcb;
1854 	bge_set_hostaddr(&rcb->bge_hostaddr,
1855 	    BGE_RING_DMA_ADDR(sc, bge_rx_std_ring));
1856 	if (BGE_IS_5705_OR_BEYOND(sc))
1857 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(512, 0);
1858 	else
1859 		rcb->bge_maxlen_flags =
1860 		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN, 0);
1861 	if (sc->bge_extram)
1862 		rcb->bge_nicaddr = BGE_EXT_STD_RX_RINGS;
1863 	else
1864 		rcb->bge_nicaddr = BGE_STD_RX_RINGS;
1865 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_HI, rcb->bge_hostaddr.bge_addr_hi);
1866 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_LO, rcb->bge_hostaddr.bge_addr_lo);
1867 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1868 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_NICADDR, rcb->bge_nicaddr);
1869 
1870 	if (BGE_IS_5705_OR_BEYOND(sc))
1871 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT_5705;
1872 	else
1873 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT;
1874 
1875 	/*
1876 	 * Initialize the jumbo RX ring control block
1877 	 * We set the 'ring disabled' bit in the flags
1878 	 * field until we're actually ready to start
1879 	 * using this ring (i.e. once we set the MTU
1880 	 * high enough to require it).
1881 	 */
1882 	if (!BGE_IS_5705_OR_BEYOND(sc)) {
1883 		rcb = &sc->bge_rdata->bge_info.bge_jumbo_rx_rcb;
1884 		bge_set_hostaddr(&rcb->bge_hostaddr,
1885 		    BGE_RING_DMA_ADDR(sc, bge_rx_jumbo_ring));
1886 		rcb->bge_maxlen_flags =
1887 		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN,
1888 			BGE_RCB_FLAG_RING_DISABLED);
1889 		if (sc->bge_extram)
1890 			rcb->bge_nicaddr = BGE_EXT_JUMBO_RX_RINGS;
1891 		else
1892 			rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS;
1893 
1894 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_HI,
1895 		    rcb->bge_hostaddr.bge_addr_hi);
1896 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_LO,
1897 		    rcb->bge_hostaddr.bge_addr_lo);
1898 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS,
1899 		    rcb->bge_maxlen_flags);
1900 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_NICADDR, rcb->bge_nicaddr);
1901 
1902 		/* Set up dummy disabled mini ring RCB */
1903 		rcb = &sc->bge_rdata->bge_info.bge_mini_rx_rcb;
1904 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(0,
1905 		    BGE_RCB_FLAG_RING_DISABLED);
1906 		CSR_WRITE_4(sc, BGE_RX_MINI_RCB_MAXLEN_FLAGS,
1907 		    rcb->bge_maxlen_flags);
1908 
1909 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
1910 		    offsetof(struct bge_ring_data, bge_info),
1911 		    sizeof (struct bge_gib),
1912 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1913 	}
1914 
1915 	/*
1916 	 * Set the BD ring replenish thresholds. The recommended
1917 	 * values are 1/8th the number of descriptors allocated to
1918 	 * each ring.
1919 	 */
1920 	i = BGE_STD_RX_RING_CNT / 8;
1921 
1922 	/*
1923 	 * Use a value of 8 for the following chips to workaround HW errata.
1924 	 * Some of these chips have been added based on empirical
1925 	 * evidence (they don't work unless this is done).
1926 	 */
1927 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5750 ||
1928 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5752 ||
1929 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 ||
1930 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5787 ||
1931 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906)
1932 		i = 8;
1933 
1934 	CSR_WRITE_4(sc, BGE_RBDI_STD_REPL_THRESH, i);
1935 	CSR_WRITE_4(sc, BGE_RBDI_JUMBO_REPL_THRESH, BGE_JUMBO_RX_RING_CNT / 8);
1936 
1937 	/*
1938 	 * Disable all unused send rings by setting the 'ring disabled'
1939 	 * bit in the flags field of all the TX send ring control blocks.
1940 	 * These are located in NIC memory.
1941 	 */
1942 	rcb_addr = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1943 	for (i = 0; i < BGE_TX_RINGS_EXTSSRAM_MAX; i++) {
1944 		RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
1945 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED));
1946 		RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, 0);
1947 		rcb_addr += sizeof(struct bge_rcb);
1948 	}
1949 
1950 	/* Configure TX RCB 0 (we use only the first ring) */
1951 	rcb_addr = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1952 	bge_set_hostaddr(&taddr, BGE_RING_DMA_ADDR(sc, bge_tx_ring));
1953 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1954 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1955 	RCB_WRITE_4(sc, rcb_addr, bge_nicaddr,
1956 		    BGE_NIC_TXRING_ADDR(0, BGE_TX_RING_CNT));
1957 	if (!(BGE_IS_5705_OR_BEYOND(sc)))
1958 		RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
1959 		    BGE_RCB_MAXLEN_FLAGS(BGE_TX_RING_CNT, 0));
1960 
1961 	/* Disable all unused RX return rings */
1962 	rcb_addr = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1963 	for (i = 0; i < BGE_RX_RINGS_MAX; i++) {
1964 		RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_hi, 0);
1965 		RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_lo, 0);
1966 		RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
1967 			    BGE_RCB_MAXLEN_FLAGS(sc->bge_return_ring_cnt,
1968                                      BGE_RCB_FLAG_RING_DISABLED));
1969 		RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, 0);
1970 		bge_writembx(sc, BGE_MBX_RX_CONS0_LO +
1971 		    (i * (sizeof(u_int64_t))), 0);
1972 		rcb_addr += sizeof(struct bge_rcb);
1973 	}
1974 
1975 	/* Initialize RX ring indexes */
1976 	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, 0);
1977 	bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, 0);
1978 	bge_writembx(sc, BGE_MBX_RX_MINI_PROD_LO, 0);
1979 
1980 	/*
1981 	 * Set up RX return ring 0
1982 	 * Note that the NIC address for RX return rings is 0x00000000.
1983 	 * The return rings live entirely within the host, so the
1984 	 * nicaddr field in the RCB isn't used.
1985 	 */
1986 	rcb_addr = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1987 	bge_set_hostaddr(&taddr, BGE_RING_DMA_ADDR(sc, bge_rx_return_ring));
1988 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1989 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1990 	RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, 0x00000000);
1991 	RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
1992 	    BGE_RCB_MAXLEN_FLAGS(sc->bge_return_ring_cnt, 0));
1993 
1994 	/* Set random backoff seed for TX */
1995 	CSR_WRITE_4(sc, BGE_TX_RANDOM_BACKOFF,
1996 	    CLLADDR(ifp->if_sadl)[0] + CLLADDR(ifp->if_sadl)[1] +
1997 	    CLLADDR(ifp->if_sadl)[2] + CLLADDR(ifp->if_sadl)[3] +
1998 	    CLLADDR(ifp->if_sadl)[4] + CLLADDR(ifp->if_sadl)[5] +
1999 	    BGE_TX_BACKOFF_SEED_MASK);
2000 
2001 	/* Set inter-packet gap */
2002 	CSR_WRITE_4(sc, BGE_TX_LENGTHS, 0x2620);
2003 
2004 	/*
2005 	 * Specify which ring to use for packets that don't match
2006 	 * any RX rules.
2007 	 */
2008 	CSR_WRITE_4(sc, BGE_RX_RULES_CFG, 0x08);
2009 
2010 	/*
2011 	 * Configure number of RX lists. One interrupt distribution
2012 	 * list, sixteen active lists, one bad frames class.
2013 	 */
2014 	CSR_WRITE_4(sc, BGE_RXLP_CFG, 0x181);
2015 
2016 	/* Inialize RX list placement stats mask. */
2017 	CSR_WRITE_4(sc, BGE_RXLP_STATS_ENABLE_MASK, 0x007FFFFF);
2018 	CSR_WRITE_4(sc, BGE_RXLP_STATS_CTL, 0x1);
2019 
2020 	/* Disable host coalescing until we get it set up */
2021 	CSR_WRITE_4(sc, BGE_HCC_MODE, 0x00000000);
2022 
2023 	/* Poll to make sure it's shut down. */
2024 	for (i = 0; i < BGE_TIMEOUT; i++) {
2025 		if (!(CSR_READ_4(sc, BGE_HCC_MODE) & BGE_HCCMODE_ENABLE))
2026 			break;
2027 		DELAY(10);
2028 	}
2029 
2030 	if (i == BGE_TIMEOUT) {
2031 		aprint_error_dev(sc->bge_dev,
2032 		    "host coalescing engine failed to idle\n");
2033 		return (ENXIO);
2034 	}
2035 
2036 	/* Set up host coalescing defaults */
2037 	CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, sc->bge_rx_coal_ticks);
2038 	CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS, sc->bge_tx_coal_ticks);
2039 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, sc->bge_rx_max_coal_bds);
2040 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS, sc->bge_tx_max_coal_bds);
2041 	if (!(BGE_IS_5705_OR_BEYOND(sc))) {
2042 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS_INT, 0);
2043 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS_INT, 0);
2044 	}
2045 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT, 0);
2046 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT, 0);
2047 
2048 	/* Set up address of statistics block */
2049 	if (!(BGE_IS_5705_OR_BEYOND(sc))) {
2050 		bge_set_hostaddr(&taddr,
2051 		    BGE_RING_DMA_ADDR(sc, bge_info.bge_stats));
2052 		CSR_WRITE_4(sc, BGE_HCC_STATS_TICKS, sc->bge_stat_ticks);
2053 		CSR_WRITE_4(sc, BGE_HCC_STATS_BASEADDR, BGE_STATS_BLOCK);
2054 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_HI, taddr.bge_addr_hi);
2055 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_LO, taddr.bge_addr_lo);
2056 	}
2057 
2058 	/* Set up address of status block */
2059 	bge_set_hostaddr(&taddr, BGE_RING_DMA_ADDR(sc, bge_status_block));
2060 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_BASEADDR, BGE_STATUS_BLOCK);
2061 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_HI, taddr.bge_addr_hi);
2062 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_LO, taddr.bge_addr_lo);
2063 	sc->bge_rdata->bge_status_block.bge_idx[0].bge_rx_prod_idx = 0;
2064 	sc->bge_rdata->bge_status_block.bge_idx[0].bge_tx_cons_idx = 0;
2065 
2066 	/* Turn on host coalescing state machine */
2067 	CSR_WRITE_4(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE);
2068 
2069 	/* Turn on RX BD completion state machine and enable attentions */
2070 	CSR_WRITE_4(sc, BGE_RBDC_MODE,
2071 	    BGE_RBDCMODE_ENABLE | BGE_RBDCMODE_ATTN);
2072 
2073 	/* Turn on RX list placement state machine */
2074 	CSR_WRITE_4(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
2075 
2076 	/* Turn on RX list selector state machine. */
2077 	if (!(BGE_IS_5705_OR_BEYOND(sc)))
2078 		CSR_WRITE_4(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
2079 
2080 	val = BGE_MACMODE_TXDMA_ENB | BGE_MACMODE_RXDMA_ENB |
2081 	    BGE_MACMODE_RX_STATS_CLEAR | BGE_MACMODE_TX_STATS_CLEAR |
2082 	    BGE_MACMODE_RX_STATS_ENB | BGE_MACMODE_TX_STATS_ENB |
2083 	    BGE_MACMODE_FRMHDR_DMA_ENB;
2084 
2085 	if (sc->bge_flags & BGE_PHY_FIBER_TBI)
2086 	    val |= BGE_PORTMODE_TBI;
2087 	else if (sc->bge_flags & BGE_PHY_FIBER_MII)
2088 	    val |= BGE_PORTMODE_GMII;
2089 	else
2090 	    val |= BGE_PORTMODE_MII;
2091 
2092 	/* Turn on DMA, clear stats */
2093 	CSR_WRITE_4(sc, BGE_MAC_MODE, val);
2094 
2095 
2096 	/* Set misc. local control, enable interrupts on attentions */
2097 	sc->bge_local_ctrl_reg = BGE_MLC_INTR_ONATTN | BGE_MLC_AUTO_EEPROM;
2098 
2099 #ifdef notdef
2100 	/* Assert GPIO pins for PHY reset */
2101 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUT0|
2102 	    BGE_MLC_MISCIO_OUT1|BGE_MLC_MISCIO_OUT2);
2103 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUTEN0|
2104 	    BGE_MLC_MISCIO_OUTEN1|BGE_MLC_MISCIO_OUTEN2);
2105 #endif
2106 
2107 #if defined(not_quite_yet)
2108 	/* Linux driver enables enable gpio pin #1 on 5700s */
2109 	if (sc->bge_chipid == BGE_CHIPID_BCM5700) {
2110 		sc->bge_local_ctrl_reg |=
2111 		  (BGE_MLC_MISCIO_OUT1|BGE_MLC_MISCIO_OUTEN1);
2112 	}
2113 #endif
2114 	CSR_WRITE_4(sc, BGE_MISC_LOCAL_CTL, sc->bge_local_ctrl_reg);
2115 
2116 	/* Turn on DMA completion state machine */
2117 	if (!(BGE_IS_5705_OR_BEYOND(sc)))
2118 		CSR_WRITE_4(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
2119 
2120 	/* Turn on write DMA state machine */
2121 	{
2122 		uint32_t bge_wdma_mode =
2123 			BGE_WDMAMODE_ENABLE|BGE_WDMAMODE_ALL_ATTNS;
2124 
2125 		/* Enable host coalescing bug fix; see Linux tg3.c */
2126 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 ||
2127 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5787)
2128 			bge_wdma_mode |= (1 << 29);
2129 
2130 		CSR_WRITE_4(sc, BGE_WDMA_MODE, bge_wdma_mode);
2131         }
2132 
2133 	/* Turn on read DMA state machine */
2134 	{
2135 		uint32_t dma_read_modebits;
2136 
2137 		dma_read_modebits =
2138 		  BGE_RDMAMODE_ENABLE | BGE_RDMAMODE_ALL_ATTNS;
2139 
2140 		if ((sc->bge_flags & BGE_PCIE) && 0) {
2141 			dma_read_modebits |= BGE_RDMA_MODE_FIFO_LONG_BURST;
2142 		} else if (BGE_IS_5705_OR_BEYOND(sc)) {
2143 			dma_read_modebits |= BGE_RDMA_MODE_FIFO_SIZE_128;
2144 		}
2145 
2146 		/* XXX broadcom-supplied linux driver; undocumented */
2147 		if (BGE_IS_5750_OR_BEYOND(sc)) {
2148  			/*
2149 			 * XXX: magic values.
2150 			 * From Broadcom-supplied Linux driver;  apparently
2151 			 * required to workaround a DMA bug affecting TSO
2152 			 * on bcm575x/bcm5721?
2153 			 */
2154 			dma_read_modebits |= (1 << 27);
2155 		}
2156 		CSR_WRITE_4(sc, BGE_RDMA_MODE, dma_read_modebits);
2157 	}
2158 
2159 	/* Turn on RX data completion state machine */
2160 	CSR_WRITE_4(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
2161 
2162 	/* Turn on RX BD initiator state machine */
2163 	CSR_WRITE_4(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
2164 
2165 	/* Turn on RX data and RX BD initiator state machine */
2166 	CSR_WRITE_4(sc, BGE_RDBDI_MODE, BGE_RDBDIMODE_ENABLE);
2167 
2168 	/* Turn on Mbuf cluster free state machine */
2169 	if (!(BGE_IS_5705_OR_BEYOND(sc)))
2170 		CSR_WRITE_4(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
2171 
2172 	/* Turn on send BD completion state machine */
2173 	CSR_WRITE_4(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
2174 
2175 	/* Turn on send data completion state machine */
2176 	CSR_WRITE_4(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE);
2177 
2178 	/* Turn on send data initiator state machine */
2179 	if (BGE_IS_5750_OR_BEYOND(sc)) {
2180 		/* XXX: magic value from Linux driver */
2181 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE | 0x08);
2182 	} else {
2183 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
2184 	}
2185 
2186 	/* Turn on send BD initiator state machine */
2187 	CSR_WRITE_4(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
2188 
2189 	/* Turn on send BD selector state machine */
2190 	CSR_WRITE_4(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
2191 
2192 	CSR_WRITE_4(sc, BGE_SDI_STATS_ENABLE_MASK, 0x007FFFFF);
2193 	CSR_WRITE_4(sc, BGE_SDI_STATS_CTL,
2194 	    BGE_SDISTATSCTL_ENABLE | BGE_SDISTATSCTL_FASTER);
2195 
2196 	/* ack/clear link change events */
2197 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
2198 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
2199 	    BGE_MACSTAT_CFG_CHANGED);
2200 	CSR_WRITE_4(sc, BGE_MI_STS, 0);
2201 
2202 	/* Enable PHY auto polling (for MII/GMII only) */
2203 	if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
2204 		CSR_WRITE_4(sc, BGE_MI_STS, BGE_MISTS_LINK);
2205  	} else {
2206 		BGE_STS_SETBIT(sc, BGE_STS_AUTOPOLL);
2207 		BGE_SETBIT(sc, BGE_MI_MODE, BGE_MIMODE_AUTOPOLL | (10 << 16));
2208 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700)
2209 			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
2210 			    BGE_EVTENB_MI_INTERRUPT);
2211 	}
2212 
2213 	/*
2214 	 * Clear any pending link state attention.
2215 	 * Otherwise some link state change events may be lost until attention
2216 	 * is cleared by bge_intr() -> bge_link_upd() sequence.
2217 	 * It's not necessary on newer BCM chips - perhaps enabling link
2218 	 * state change attentions implies clearing pending attention.
2219 	 */
2220 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
2221 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
2222 	    BGE_MACSTAT_LINK_CHANGED);
2223 
2224 	/* Enable link state change attentions. */
2225 	BGE_SETBIT(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_LINK_CHANGED);
2226 
2227 	return (0);
2228 }
2229 
2230 static const struct bge_revision *
2231 bge_lookup_rev(uint32_t chipid)
2232 {
2233 	const struct bge_revision *br;
2234 
2235 	for (br = bge_revisions; br->br_name != NULL; br++) {
2236 		if (br->br_chipid == chipid)
2237 			return (br);
2238 	}
2239 
2240 	for (br = bge_majorrevs; br->br_name != NULL; br++) {
2241 		if (br->br_chipid == BGE_ASICREV(chipid))
2242 			return (br);
2243 	}
2244 
2245 	return (NULL);
2246 }
2247 
2248 static const struct bge_product *
2249 bge_lookup(const struct pci_attach_args *pa)
2250 {
2251 	const struct bge_product *bp;
2252 
2253 	for (bp = bge_products; bp->bp_name != NULL; bp++) {
2254 		if (PCI_VENDOR(pa->pa_id) == bp->bp_vendor &&
2255 		    PCI_PRODUCT(pa->pa_id) == bp->bp_product)
2256 			return (bp);
2257 	}
2258 
2259 	return (NULL);
2260 }
2261 
2262 static int
2263 bge_setpowerstate(struct bge_softc *sc, int powerlevel)
2264 {
2265 #ifdef NOTYET
2266 	u_int32_t pm_ctl = 0;
2267 
2268 	/* XXX FIXME: make sure indirect accesses enabled? */
2269 	pm_ctl = pci_conf_read(sc->bge_dev, BGE_PCI_MISC_CTL, 4);
2270 	pm_ctl |= BGE_PCIMISCCTL_INDIRECT_ACCESS;
2271 	pci_write_config(sc->bge_dev, BGE_PCI_MISC_CTL, pm_ctl, 4);
2272 
2273 	/* clear the PME_assert bit and power state bits, enable PME */
2274 	pm_ctl = pci_conf_read(sc->bge_dev, BGE_PCI_PWRMGMT_CMD, 2);
2275 	pm_ctl &= ~PCIM_PSTAT_DMASK;
2276 	pm_ctl |= (1 << 8);
2277 
2278 	if (powerlevel == 0) {
2279 		pm_ctl |= PCIM_PSTAT_D0;
2280 		pci_write_config(sc->bge_dev, BGE_PCI_PWRMGMT_CMD,
2281 		    pm_ctl, 2);
2282 		DELAY(10000);
2283 		CSR_WRITE_4(sc, BGE_MISC_LOCAL_CTL, sc->bge_local_ctrl_reg);
2284 		DELAY(10000);
2285 
2286 #ifdef NOTYET
2287 		/* XXX FIXME: write 0x02 to phy aux_Ctrl reg */
2288 		bge_miibus_writereg(sc->bge_dev, 1, 0x18, 0x02);
2289 #endif
2290 		DELAY(40); DELAY(40); DELAY(40);
2291 		DELAY(10000);	/* above not quite adequate on 5700 */
2292 		return 0;
2293 	}
2294 
2295 
2296 	/*
2297 	 * Entering ACPI power states D1-D3 is achieved by wiggling
2298 	 * GMII gpio pins. Example code assumes all hardware vendors
2299 	 * followed Broadom's sample pcb layout. Until we verify that
2300 	 * for all supported OEM cards, states D1-D3 are  unsupported.
2301 	 */
2302 	aprint_error_dev(sc->bge_dev,
2303 	    "power state %d unimplemented; check GPIO pins\n",
2304 	    powerlevel);
2305 #endif
2306 	return EOPNOTSUPP;
2307 }
2308 
2309 
2310 /*
2311  * Probe for a Broadcom chip. Check the PCI vendor and device IDs
2312  * against our list and return its name if we find a match. Note
2313  * that since the Broadcom controller contains VPD support, we
2314  * can get the device name string from the controller itself instead
2315  * of the compiled-in string. This is a little slow, but it guarantees
2316  * we'll always announce the right product name.
2317  */
2318 static int
2319 bge_probe(device_t parent, cfdata_t match, void *aux)
2320 {
2321 	struct pci_attach_args *pa = (struct pci_attach_args *)aux;
2322 
2323 	if (bge_lookup(pa) != NULL)
2324 		return (1);
2325 
2326 	return (0);
2327 }
2328 
2329 static void
2330 bge_attach(device_t parent, device_t self, void *aux)
2331 {
2332 	struct bge_softc	*sc = device_private(self);
2333 	struct pci_attach_args	*pa = aux;
2334 	const struct bge_product *bp;
2335 	const struct bge_revision *br;
2336 	pci_chipset_tag_t	pc;
2337 	pci_intr_handle_t	ih;
2338 	const char		*intrstr = NULL;
2339 	bus_dma_segment_t	seg;
2340 	int			rseg;
2341 	u_int32_t		hwcfg = 0;
2342 	u_int32_t		command;
2343 	struct ifnet		*ifp;
2344 	u_int32_t		misccfg;
2345 	void *			kva;
2346 	u_char			eaddr[ETHER_ADDR_LEN];
2347 	pcireg_t		memtype;
2348 	bus_addr_t		memaddr;
2349 	bus_size_t		memsize;
2350 	u_int32_t		pm_ctl;
2351 
2352 	bp = bge_lookup(pa);
2353 	KASSERT(bp != NULL);
2354 
2355 	sc->sc_pc = pa->pa_pc;
2356 	sc->sc_pcitag = pa->pa_tag;
2357 	sc->bge_dev = self;
2358 
2359 	aprint_naive(": Ethernet controller\n");
2360 	aprint_normal(": %s\n", bp->bp_name);
2361 
2362 	/*
2363 	 * Map control/status registers.
2364 	 */
2365 	DPRINTFN(5, ("Map control/status regs\n"));
2366 	pc = sc->sc_pc;
2367 	command = pci_conf_read(pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG);
2368 	command |= PCI_COMMAND_MEM_ENABLE | PCI_COMMAND_MASTER_ENABLE;
2369 	pci_conf_write(pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG, command);
2370 	command = pci_conf_read(pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG);
2371 
2372 	if (!(command & PCI_COMMAND_MEM_ENABLE)) {
2373 		aprint_error_dev(sc->bge_dev,
2374 		    "failed to enable memory mapping!\n");
2375 		return;
2376 	}
2377 
2378 	DPRINTFN(5, ("pci_mem_find\n"));
2379 	memtype = pci_mapreg_type(sc->sc_pc, sc->sc_pcitag, BGE_PCI_BAR0);
2380  	switch (memtype) {
2381 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_32BIT:
2382 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_64BIT:
2383 		if (pci_mapreg_map(pa, BGE_PCI_BAR0,
2384 		    memtype, 0, &sc->bge_btag, &sc->bge_bhandle,
2385 		    &memaddr, &memsize) == 0)
2386 			break;
2387 	default:
2388 		aprint_error_dev(sc->bge_dev, "can't find mem space\n");
2389 		return;
2390 	}
2391 
2392 	DPRINTFN(5, ("pci_intr_map\n"));
2393 	if (pci_intr_map(pa, &ih)) {
2394 		aprint_error_dev(sc->bge_dev, "couldn't map interrupt\n");
2395 		return;
2396 	}
2397 
2398 	DPRINTFN(5, ("pci_intr_string\n"));
2399 	intrstr = pci_intr_string(pc, ih);
2400 
2401 	DPRINTFN(5, ("pci_intr_establish\n"));
2402 	sc->bge_intrhand = pci_intr_establish(pc, ih, IPL_NET, bge_intr, sc);
2403 
2404 	if (sc->bge_intrhand == NULL) {
2405 		aprint_error_dev(sc->bge_dev,
2406 		    "couldn't establish interrupt%s%s\n",
2407 		    intrstr ? " at " : "", intrstr ? intrstr : "");
2408 		return;
2409 	}
2410 	aprint_normal_dev(sc->bge_dev, "interrupting at %s\n", intrstr);
2411 
2412 	/*
2413 	 * Kludge for 5700 Bx bug: a hardware bug (PCIX byte enable?)
2414 	 * can clobber the chip's PCI config-space power control registers,
2415 	 * leaving the card in D3 powersave state.
2416 	 * We do not have memory-mapped registers in this state,
2417 	 * so force device into D0 state before starting initialization.
2418 	 */
2419 	pm_ctl = pci_conf_read(pc, sc->sc_pcitag, BGE_PCI_PWRMGMT_CMD);
2420 	pm_ctl &= ~(PCI_PWR_D0|PCI_PWR_D1|PCI_PWR_D2|PCI_PWR_D3);
2421 	pm_ctl |= (1 << 8) | PCI_PWR_D0 ; /* D0 state */
2422 	pci_conf_write(pc, sc->sc_pcitag, BGE_PCI_PWRMGMT_CMD, pm_ctl);
2423 	DELAY(1000);	/* 27 usec is allegedly sufficent */
2424 
2425 	/*
2426 	 * Save ASIC rev.
2427 	 */
2428 	sc->bge_chipid =
2429 	    pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL) &
2430 	    BGE_PCIMISCCTL_ASICREV;
2431 
2432 	/*
2433 	 * Detect PCI-Express devices
2434 	 * XXX: guessed from Linux/FreeBSD; no documentation
2435 	 */
2436 	if (pci_get_capability(sc->sc_pc, sc->sc_pcitag, PCI_CAP_PCIEXPRESS,
2437 	        NULL, NULL) != 0)
2438 		sc->bge_flags |= BGE_PCIE;
2439 
2440 	/*
2441 	 * PCI-X check.
2442 	 */
2443 	if ((pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_PCISTATE) &
2444 		BGE_PCISTATE_PCI_BUSMODE) == 0)
2445 		sc->bge_flags |= BGE_PCIX;
2446 
2447 	if (sc->bge_chipid == BGE_CHIPID_BCM5701_A0 ||
2448 	    sc->bge_chipid == BGE_CHIPID_BCM5701_B0)
2449 		sc->bge_flags |= BGE_PHY_CRC_BUG;
2450 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5703_AX ||
2451 	    BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5704_AX)
2452 		sc->bge_flags |= BGE_PHY_ADC_BUG;
2453 	if (sc->bge_chipid == BGE_CHIPID_BCM5704_A0)
2454 		sc->bge_flags |= BGE_PHY_5704_A0_BUG;
2455 
2456 	if (BGE_IS_5705_OR_BEYOND(sc)) {
2457 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 ||
2458 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5787) {
2459 			if (PCI_PRODUCT(pa->pa_id) != PCI_PRODUCT_BROADCOM_BCM5722 &&
2460 			    PCI_PRODUCT(pa->pa_id) != PCI_PRODUCT_BROADCOM_BCM5756)
2461 				sc->bge_flags |= BGE_PHY_JITTER_BUG;
2462 			if (PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5755M)
2463 				sc->bge_flags |= BGE_PHY_ADJUST_TRIM;
2464 		} else if (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5906)
2465 			sc->bge_flags |= BGE_PHY_BER_BUG;
2466 	}
2467 
2468 	/* Try to reset the chip. */
2469 	DPRINTFN(5, ("bge_reset\n"));
2470 	bge_reset(sc);
2471 
2472 	if (bge_chipinit(sc)) {
2473 		aprint_error_dev(sc->bge_dev, "chip initialization failed\n");
2474 		bge_release_resources(sc);
2475 		return;
2476 	}
2477 
2478 	/*
2479 	 * Get station address from the EEPROM.
2480 	 */
2481 	if (bge_get_eaddr(sc, eaddr)) {
2482 		aprint_error_dev(sc->bge_dev,
2483 		"failed to reade station address\n");
2484 		bge_release_resources(sc);
2485 		return;
2486 	}
2487 
2488 	br = bge_lookup_rev(sc->bge_chipid);
2489 
2490 	if (br == NULL) {
2491 		aprint_normal_dev(sc->bge_dev, "unknown ASIC (0x%04x)",
2492 		    sc->bge_chipid >> 16);
2493 	} else {
2494 		aprint_normal_dev(sc->bge_dev, "ASIC %s (0x%04x)",
2495 		    br->br_name, sc->bge_chipid >> 16);
2496 	}
2497 	aprint_normal(", Ethernet address %s\n", ether_sprintf(eaddr));
2498 
2499 	/* Allocate the general information block and ring buffers. */
2500 	if (pci_dma64_available(pa))
2501 		sc->bge_dmatag = pa->pa_dmat64;
2502 	else
2503 		sc->bge_dmatag = pa->pa_dmat;
2504 	DPRINTFN(5, ("bus_dmamem_alloc\n"));
2505 	if (bus_dmamem_alloc(sc->bge_dmatag, sizeof(struct bge_ring_data),
2506 			     PAGE_SIZE, 0, &seg, 1, &rseg, BUS_DMA_NOWAIT)) {
2507 		aprint_error_dev(sc->bge_dev, "can't alloc rx buffers\n");
2508 		return;
2509 	}
2510 	DPRINTFN(5, ("bus_dmamem_map\n"));
2511 	if (bus_dmamem_map(sc->bge_dmatag, &seg, rseg,
2512 			   sizeof(struct bge_ring_data), &kva,
2513 			   BUS_DMA_NOWAIT)) {
2514 		aprint_error_dev(sc->bge_dev,
2515 		    "can't map DMA buffers (%zu bytes)\n",
2516 		    sizeof(struct bge_ring_data));
2517 		bus_dmamem_free(sc->bge_dmatag, &seg, rseg);
2518 		return;
2519 	}
2520 	DPRINTFN(5, ("bus_dmamem_create\n"));
2521 	if (bus_dmamap_create(sc->bge_dmatag, sizeof(struct bge_ring_data), 1,
2522 	    sizeof(struct bge_ring_data), 0,
2523 	    BUS_DMA_NOWAIT, &sc->bge_ring_map)) {
2524 		aprint_error_dev(sc->bge_dev, "can't create DMA map\n");
2525 		bus_dmamem_unmap(sc->bge_dmatag, kva,
2526 				 sizeof(struct bge_ring_data));
2527 		bus_dmamem_free(sc->bge_dmatag, &seg, rseg);
2528 		return;
2529 	}
2530 	DPRINTFN(5, ("bus_dmamem_load\n"));
2531 	if (bus_dmamap_load(sc->bge_dmatag, sc->bge_ring_map, kva,
2532 			    sizeof(struct bge_ring_data), NULL,
2533 			    BUS_DMA_NOWAIT)) {
2534 		bus_dmamap_destroy(sc->bge_dmatag, sc->bge_ring_map);
2535 		bus_dmamem_unmap(sc->bge_dmatag, kva,
2536 				 sizeof(struct bge_ring_data));
2537 		bus_dmamem_free(sc->bge_dmatag, &seg, rseg);
2538 		return;
2539 	}
2540 
2541 	DPRINTFN(5, ("bzero\n"));
2542 	sc->bge_rdata = (struct bge_ring_data *)kva;
2543 
2544 	memset(sc->bge_rdata, 0, sizeof(struct bge_ring_data));
2545 
2546 	/* Try to allocate memory for jumbo buffers. */
2547 	if (!(BGE_IS_5705_OR_BEYOND(sc))) {
2548 		if (bge_alloc_jumbo_mem(sc)) {
2549 			aprint_error_dev(sc->bge_dev,
2550 			    "jumbo buffer allocation failed\n");
2551 		} else
2552 			sc->ethercom.ec_capabilities |= ETHERCAP_JUMBO_MTU;
2553 	}
2554 
2555 	/* Set default tuneable values. */
2556 	sc->bge_stat_ticks = BGE_TICKS_PER_SEC;
2557 	sc->bge_rx_coal_ticks = 150;
2558 	sc->bge_rx_max_coal_bds = 64;
2559 #ifdef ORIG_WPAUL_VALUES
2560 	sc->bge_tx_coal_ticks = 150;
2561 	sc->bge_tx_max_coal_bds = 128;
2562 #else
2563 	sc->bge_tx_coal_ticks = 300;
2564 	sc->bge_tx_max_coal_bds = 400;
2565 #endif
2566 	if (BGE_IS_5705_OR_BEYOND(sc)) {
2567 		sc->bge_tx_coal_ticks = (12 * 5);
2568 		sc->bge_tx_max_coal_bds = (12 * 5);
2569 			aprint_verbose_dev(sc->bge_dev,
2570 			    "setting short Tx thresholds\n");
2571 	}
2572 
2573 	/* Set up ifnet structure */
2574 	ifp = &sc->ethercom.ec_if;
2575 	ifp->if_softc = sc;
2576 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2577 	ifp->if_ioctl = bge_ioctl;
2578 	ifp->if_stop = bge_stop;
2579 	ifp->if_start = bge_start;
2580 	ifp->if_init = bge_init;
2581 	ifp->if_watchdog = bge_watchdog;
2582 	IFQ_SET_MAXLEN(&ifp->if_snd, max(BGE_TX_RING_CNT - 1, IFQ_MAXLEN));
2583 	IFQ_SET_READY(&ifp->if_snd);
2584 	DPRINTFN(5, ("strcpy if_xname\n"));
2585 	strcpy(ifp->if_xname, device_xname(sc->bge_dev));
2586 
2587 	if (sc->bge_chipid != BGE_CHIPID_BCM5700_B0)
2588 		sc->ethercom.ec_if.if_capabilities |=
2589 		    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
2590 		    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
2591 		    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
2592 	sc->ethercom.ec_capabilities |=
2593 	    ETHERCAP_VLAN_HWTAGGING | ETHERCAP_VLAN_MTU;
2594 
2595 	if (sc->bge_flags & BGE_PCIE)
2596 		sc->ethercom.ec_if.if_capabilities |= IFCAP_TSOv4;
2597 
2598 	/*
2599 	 * Do MII setup.
2600 	 */
2601 	DPRINTFN(5, ("mii setup\n"));
2602 	sc->bge_mii.mii_ifp = ifp;
2603 	sc->bge_mii.mii_readreg = bge_miibus_readreg;
2604 	sc->bge_mii.mii_writereg = bge_miibus_writereg;
2605 	sc->bge_mii.mii_statchg = bge_miibus_statchg;
2606 
2607 	/*
2608 	 * Figure out what sort of media we have by checking the
2609 	 * hardware config word in the first 32k of NIC internal memory,
2610 	 * or fall back to the config word in the EEPROM. Note: on some BCM5700
2611 	 * cards, this value appears to be unset. If that's the
2612 	 * case, we have to rely on identifying the NIC by its PCI
2613 	 * subsystem ID, as we do below for the SysKonnect SK-9D41.
2614 	 */
2615 	if (bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_SIG) == BGE_MAGIC_NUMBER) {
2616 		hwcfg = bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_NICCFG);
2617 	} else {
2618 		bge_read_eeprom(sc, (void *)&hwcfg,
2619 		    BGE_EE_HWCFG_OFFSET, sizeof(hwcfg));
2620 		hwcfg = be32toh(hwcfg);
2621 	}
2622 	/* The SysKonnect SK-9D41 is a 1000baseSX card. */
2623 	if (PCI_PRODUCT(pa->pa_id) == SK_SUBSYSID_9D41 ||
2624 	    (hwcfg & BGE_HWCFG_MEDIA) == BGE_MEDIA_FIBER) {
2625 		if (BGE_IS_5714_FAMILY(sc))
2626 		    sc->bge_flags |= BGE_PHY_FIBER_MII;
2627 		else
2628 		    sc->bge_flags |= BGE_PHY_FIBER_TBI;
2629 	}
2630 
2631 	if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
2632 		ifmedia_init(&sc->bge_ifmedia, IFM_IMASK, bge_ifmedia_upd,
2633 		    bge_ifmedia_sts);
2634 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER|IFM_1000_SX, 0, NULL);
2635 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER|IFM_1000_SX|IFM_FDX,
2636 			    0, NULL);
2637 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER|IFM_AUTO, 0, NULL);
2638 		ifmedia_set(&sc->bge_ifmedia, IFM_ETHER|IFM_AUTO);
2639 		/* Pretend the user requested this setting */
2640 		sc->bge_ifmedia.ifm_media = sc->bge_ifmedia.ifm_cur->ifm_media;
2641 	} else {
2642 		/*
2643 		 * Do transceiver setup.
2644 		 */
2645 		ifmedia_init(&sc->bge_mii.mii_media, 0, bge_ifmedia_upd,
2646 			     bge_ifmedia_sts);
2647 		mii_attach(sc->bge_dev, &sc->bge_mii, 0xffffffff,
2648 			   MII_PHY_ANY, MII_OFFSET_ANY,
2649 			   MIIF_FORCEANEG|MIIF_DOPAUSE);
2650 
2651 		if (LIST_EMPTY(&sc->bge_mii.mii_phys)) {
2652 			aprint_error_dev(sc->bge_dev, "no PHY found!\n");
2653 			ifmedia_add(&sc->bge_mii.mii_media,
2654 				    IFM_ETHER|IFM_MANUAL, 0, NULL);
2655 			ifmedia_set(&sc->bge_mii.mii_media,
2656 				    IFM_ETHER|IFM_MANUAL);
2657 		} else
2658 			ifmedia_set(&sc->bge_mii.mii_media,
2659 				    IFM_ETHER|IFM_AUTO);
2660 	}
2661 
2662 	/*
2663 	 * When using the BCM5701 in PCI-X mode, data corruption has
2664 	 * been observed in the first few bytes of some received packets.
2665 	 * Aligning the packet buffer in memory eliminates the corruption.
2666 	 * Unfortunately, this misaligns the packet payloads.  On platforms
2667 	 * which do not support unaligned accesses, we will realign the
2668 	 * payloads by copying the received packets.
2669 	 */
2670 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5701 &&
2671 		sc->bge_flags & BGE_PCIX)
2672 		sc->bge_flags |= BGE_RX_ALIGNBUG;
2673 
2674 	misccfg = CSR_READ_4(sc, BGE_MISC_CFG);
2675 	misccfg &= BGE_MISCCFG_BOARD_ID_MASK;
2676 
2677 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5705 &&
2678 	    (misccfg == BGE_MISCCFG_BOARD_ID_5788 ||
2679 	     misccfg == BGE_MISCCFG_BOARD_ID_5788M))
2680 		sc->bge_flags |= BGE_IS_5788;
2681 
2682 	/*
2683 	 * Call MI attach routine.
2684 	 */
2685 	DPRINTFN(5, ("if_attach\n"));
2686 	if_attach(ifp);
2687 	DPRINTFN(5, ("ether_ifattach\n"));
2688 	ether_ifattach(ifp, eaddr);
2689 #if NRND > 0
2690 	rnd_attach_source(&sc->rnd_source, device_xname(sc->bge_dev),
2691 		RND_TYPE_NET, 0);
2692 #endif
2693 #ifdef BGE_EVENT_COUNTERS
2694 	/*
2695 	 * Attach event counters.
2696 	 */
2697 	evcnt_attach_dynamic(&sc->bge_ev_intr, EVCNT_TYPE_INTR,
2698 	    NULL, device_xname(sc->bge_dev), "intr");
2699 	evcnt_attach_dynamic(&sc->bge_ev_tx_xoff, EVCNT_TYPE_MISC,
2700 	    NULL, device_xname(sc->bge_dev), "tx_xoff");
2701 	evcnt_attach_dynamic(&sc->bge_ev_tx_xon, EVCNT_TYPE_MISC,
2702 	    NULL, device_xname(sc->bge_dev), "tx_xon");
2703 	evcnt_attach_dynamic(&sc->bge_ev_rx_xoff, EVCNT_TYPE_MISC,
2704 	    NULL, device_xname(sc->bge_dev), "rx_xoff");
2705 	evcnt_attach_dynamic(&sc->bge_ev_rx_xon, EVCNT_TYPE_MISC,
2706 	    NULL, device_xname(sc->bge_dev), "rx_xon");
2707 	evcnt_attach_dynamic(&sc->bge_ev_rx_macctl, EVCNT_TYPE_MISC,
2708 	    NULL, device_xname(sc->bge_dev), "rx_macctl");
2709 	evcnt_attach_dynamic(&sc->bge_ev_xoffentered, EVCNT_TYPE_MISC,
2710 	    NULL, device_xname(sc->bge_dev), "xoffentered");
2711 #endif /* BGE_EVENT_COUNTERS */
2712 	DPRINTFN(5, ("callout_init\n"));
2713 	callout_init(&sc->bge_timeout, 0);
2714 
2715 	if (!pmf_device_register(self, NULL, NULL))
2716 		aprint_error_dev(self, "couldn't establish power handler\n");
2717 	else
2718 		pmf_class_network_register(self, ifp);
2719 }
2720 
2721 static void
2722 bge_release_resources(struct bge_softc *sc)
2723 {
2724 	if (sc->bge_vpd_prodname != NULL)
2725 		free(sc->bge_vpd_prodname, M_DEVBUF);
2726 
2727 	if (sc->bge_vpd_readonly != NULL)
2728 		free(sc->bge_vpd_readonly, M_DEVBUF);
2729 }
2730 
2731 static void
2732 bge_reset(struct bge_softc *sc)
2733 {
2734 	u_int32_t cachesize, command, pcistate, new_pcistate;
2735 	int i, val;
2736 	void (*write_op)(struct bge_softc *, int, int);
2737 
2738 	if (BGE_IS_5750_OR_BEYOND(sc) && !BGE_IS_5714_FAMILY(sc) &&
2739 	    (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5906)) {
2740 	    	if (sc->bge_flags & BGE_PCIE) {
2741 			write_op = bge_writemem_direct;
2742 		} else {
2743 			write_op = bge_writemem_ind;
2744 		}
2745 	} else {
2746 		write_op = bge_writereg_ind;
2747 	}
2748 
2749 
2750 	/* Save some important PCI state. */
2751 	cachesize = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CACHESZ);
2752 	command = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD);
2753 	pcistate = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_PCISTATE);
2754 
2755 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL,
2756 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2757 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW);
2758 
2759 	/* Disable fastboot on controllers that support it. */
2760 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5752 ||
2761 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 ||
2762 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5787)
2763 		CSR_WRITE_4(sc, BGE_FASTBOOT_PC, 0);
2764 
2765 	val = BGE_MISCCFG_RESET_CORE_CLOCKS | (65<<1);
2766 	/*
2767 	 * XXX: from FreeBSD/Linux; no documentation
2768 	 */
2769 	if (sc->bge_flags & BGE_PCIE) {
2770 		if (CSR_READ_4(sc, BGE_PCIE_CTL1) == 0x60)
2771 			/* PCI Express 1.0 system */
2772 			CSR_WRITE_4(sc, BGE_PCIE_CTL1, 0x20);
2773 		if (sc->bge_chipid != BGE_CHIPID_BCM5750_A0) {
2774 			/*
2775 			 * Prevent PCI Express link training
2776 			 * during global reset.
2777 			 */
2778 			CSR_WRITE_4(sc, BGE_MISC_CFG, 1 << 29);
2779 			val |= (1<<29);
2780 		}
2781 	}
2782 
2783 	/*
2784 	 * Set GPHY Power Down Override to leave GPHY
2785 	 * powered up in D0 uninitialized.
2786 	 */
2787 	if (BGE_IS_5705_OR_BEYOND(sc))
2788 		val |= BGE_MISCCFG_KEEP_GPHY_POWER;
2789 
2790 	/* Issue global reset */
2791 	write_op(sc, BGE_MISC_CFG, val);
2792 
2793 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
2794 		i = CSR_READ_4(sc, BGE_VCPU_STATUS);
2795 		CSR_WRITE_4(sc, BGE_VCPU_STATUS,
2796 		    i | BGE_VCPU_STATUS_DRV_RESET);
2797 		i = CSR_READ_4(sc, BGE_VCPU_EXT_CTRL);
2798 		CSR_WRITE_4(sc, BGE_VCPU_EXT_CTRL,
2799 		    i & ~BGE_VCPU_EXT_CTRL_HALT_CPU);
2800 	}
2801 
2802 	DELAY(1000);
2803 
2804 	/*
2805 	 * XXX: from FreeBSD/Linux; no documentation
2806 	 */
2807 	if (sc->bge_flags & BGE_PCIE) {
2808 		if (sc->bge_chipid == BGE_CHIPID_BCM5750_A0) {
2809 			pcireg_t reg;
2810 
2811 			DELAY(500000);
2812 			/* XXX: Magic Numbers */
2813 			reg = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_UNKNOWN0);
2814 			pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_UNKNOWN0,
2815 			    reg | (1 << 15));
2816 		}
2817 		/*
2818 		 * XXX: Magic Numbers.
2819 		 * Sets maximal PCI-e payload and clears any PCI-e errors.
2820 		 * Should be replaced with references to PCI config-space
2821 		 * capability block for PCI-Express.
2822 		 */
2823 		pci_conf_write(sc->sc_pc, sc->sc_pcitag,
2824 		    BGE_PCI_CONF_DEV_CTRL, 0xf5000);
2825 
2826 	}
2827 
2828 	/* Reset some of the PCI state that got zapped by reset */
2829 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL,
2830 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2831 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW);
2832 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD, command);
2833 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CACHESZ, cachesize);
2834 	write_op(sc, BGE_MISC_CFG, (65 << 1));
2835 
2836 	/* Enable memory arbiter. */
2837 	{
2838 		uint32_t marbmode = 0;
2839 		if (BGE_IS_5714_FAMILY(sc)) {
2840 			marbmode = CSR_READ_4(sc, BGE_MARB_MODE);
2841 		}
2842  		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE | marbmode);
2843 	}
2844 
2845 
2846 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
2847 		for (i = 0; i < BGE_TIMEOUT; i++) {
2848 			val = CSR_READ_4(sc, BGE_VCPU_STATUS);
2849 			if (val & BGE_VCPU_STATUS_INIT_DONE)
2850 				break;
2851 			DELAY(100);
2852 		}
2853 		if (i == BGE_TIMEOUT) {
2854 			aprint_error_dev(sc->bge_dev, "reset timed out\n");
2855 			return;
2856 		}
2857 	} else {
2858 		/*
2859 		 * Write the magic number to the firmware mailbox at 0xb50
2860 		 * so that the driver can synchronize with the firmware.
2861 		 */
2862 		bge_writemem_ind(sc, BGE_SOFTWARE_GENCOMM, BGE_MAGIC_NUMBER);
2863 
2864 		/*
2865 		 * Poll the value location we just wrote until
2866 		 * we see the 1's complement of the magic number.
2867 		 * This indicates that the firmware initialization
2868 		 * is complete.
2869 		 */
2870 		for (i = 0; i < BGE_TIMEOUT; i++) {
2871 			val = bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM);
2872 			if (val == ~BGE_MAGIC_NUMBER)
2873 				break;
2874 			DELAY(1000);
2875 		}
2876 
2877 		if (i >= BGE_TIMEOUT) {
2878 			aprint_error_dev(sc->bge_dev,
2879 			    "firmware handshake timed out, val = %x\n", val);
2880 			/*
2881 			 * XXX: occasionally fired on bcm5721, but without
2882 			 * apparent harm.  For now, keep going if we timeout
2883 			 * against PCI-E devices.
2884 			 */
2885 			if ((sc->bge_flags & BGE_PCIE) == 0)
2886 				  return;
2887 		}
2888 	}
2889 
2890 	/*
2891 	 * XXX Wait for the value of the PCISTATE register to
2892 	 * return to its original pre-reset state. This is a
2893 	 * fairly good indicator of reset completion. If we don't
2894 	 * wait for the reset to fully complete, trying to read
2895 	 * from the device's non-PCI registers may yield garbage
2896 	 * results.
2897 	 */
2898 	for (i = 0; i < 10000; i++) {
2899 		new_pcistate = pci_conf_read(sc->sc_pc, sc->sc_pcitag,
2900 		    BGE_PCI_PCISTATE);
2901 		if ((new_pcistate & ~BGE_PCISTATE_RESERVED) ==
2902 		    (pcistate & ~BGE_PCISTATE_RESERVED))
2903 			break;
2904 		DELAY(10);
2905 	}
2906 	if ((new_pcistate & ~BGE_PCISTATE_RESERVED) !=
2907 	    (pcistate & ~BGE_PCISTATE_RESERVED)) {
2908 		aprint_error_dev(sc->bge_dev, "pcistate failed to revert\n");
2909 	}
2910 
2911 	/* Enable memory arbiter. */
2912 	/* XXX why do this twice? */
2913 	{
2914 		uint32_t marbmode = 0;
2915 		if (BGE_IS_5714_FAMILY(sc)) {
2916 			marbmode = CSR_READ_4(sc, BGE_MARB_MODE);
2917 		}
2918  		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE | marbmode);
2919 	}
2920 
2921 	/* Fix up byte swapping */
2922 	CSR_WRITE_4(sc, BGE_MODE_CTL, BGE_DMA_SWAP_OPTIONS);
2923 
2924 	CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
2925 
2926 	/*
2927 	 * The 5704 in TBI mode apparently needs some special
2928 	 * adjustment to insure the SERDES drive level is set
2929 	 * to 1.2V.
2930 	 */
2931 	if (sc->bge_flags & BGE_PHY_FIBER_TBI &&
2932 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704) {
2933 		u_int32_t serdescfg;
2934 
2935 		serdescfg = CSR_READ_4(sc, BGE_SERDES_CFG);
2936 		serdescfg = (serdescfg & ~0xFFF) | 0x880;
2937 		CSR_WRITE_4(sc, BGE_SERDES_CFG, serdescfg);
2938 	}
2939 
2940 	/* XXX: from FreeBSD/Linux; no documentation */
2941 	if (sc->bge_flags & BGE_PCIE &&
2942 	    sc->bge_chipid != BGE_CHIPID_BCM5750_A0)
2943 		CSR_WRITE_4(sc, BGE_PCIE_CTL0, CSR_READ_4(sc, BGE_PCIE_CTL0) | (1<<25));
2944 	DELAY(10000);
2945 }
2946 
2947 /*
2948  * Frame reception handling. This is called if there's a frame
2949  * on the receive return list.
2950  *
2951  * Note: we have to be able to handle two possibilities here:
2952  * 1) the frame is from the jumbo recieve ring
2953  * 2) the frame is from the standard receive ring
2954  */
2955 
2956 static void
2957 bge_rxeof(struct bge_softc *sc)
2958 {
2959 	struct ifnet *ifp;
2960 	int stdcnt = 0, jumbocnt = 0;
2961 	bus_dmamap_t dmamap;
2962 	bus_addr_t offset, toff;
2963 	bus_size_t tlen;
2964 	int tosync;
2965 
2966 	ifp = &sc->ethercom.ec_if;
2967 
2968 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
2969 	    offsetof(struct bge_ring_data, bge_status_block),
2970 	    sizeof (struct bge_status_block),
2971 	    BUS_DMASYNC_POSTREAD);
2972 
2973 	offset = offsetof(struct bge_ring_data, bge_rx_return_ring);
2974 	tosync = sc->bge_rdata->bge_status_block.bge_idx[0].bge_rx_prod_idx -
2975 	    sc->bge_rx_saved_considx;
2976 
2977 #if NRND > 0
2978 	if (tosync != 0 && RND_ENABLED(&sc->rnd_source))
2979 		rnd_add_uint32(&sc->rnd_source, tosync);
2980 #endif
2981 
2982 	toff = offset + (sc->bge_rx_saved_considx * sizeof (struct bge_rx_bd));
2983 
2984 	if (tosync < 0) {
2985 		tlen = (sc->bge_return_ring_cnt - sc->bge_rx_saved_considx) *
2986 		    sizeof (struct bge_rx_bd);
2987 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
2988 		    toff, tlen, BUS_DMASYNC_POSTREAD);
2989 		tosync = -tosync;
2990 	}
2991 
2992 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
2993 	    offset, tosync * sizeof (struct bge_rx_bd),
2994 	    BUS_DMASYNC_POSTREAD);
2995 
2996 	while(sc->bge_rx_saved_considx !=
2997 	    sc->bge_rdata->bge_status_block.bge_idx[0].bge_rx_prod_idx) {
2998 		struct bge_rx_bd	*cur_rx;
2999 		u_int32_t		rxidx;
3000 		struct mbuf		*m = NULL;
3001 
3002 		cur_rx = &sc->bge_rdata->
3003 			bge_rx_return_ring[sc->bge_rx_saved_considx];
3004 
3005 		rxidx = cur_rx->bge_idx;
3006 		BGE_INC(sc->bge_rx_saved_considx, sc->bge_return_ring_cnt);
3007 
3008 		if (cur_rx->bge_flags & BGE_RXBDFLAG_JUMBO_RING) {
3009 			BGE_INC(sc->bge_jumbo, BGE_JUMBO_RX_RING_CNT);
3010 			m = sc->bge_cdata.bge_rx_jumbo_chain[rxidx];
3011 			sc->bge_cdata.bge_rx_jumbo_chain[rxidx] = NULL;
3012 			jumbocnt++;
3013 			bus_dmamap_sync(sc->bge_dmatag,
3014 			    sc->bge_cdata.bge_rx_jumbo_map,
3015 			    mtod(m, char *) - (char *)sc->bge_cdata.bge_jumbo_buf,
3016 			    BGE_JLEN, BUS_DMASYNC_POSTREAD);
3017 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
3018 				ifp->if_ierrors++;
3019 				bge_newbuf_jumbo(sc, sc->bge_jumbo, m);
3020 				continue;
3021 			}
3022 			if (bge_newbuf_jumbo(sc, sc->bge_jumbo,
3023 					     NULL)== ENOBUFS) {
3024 				ifp->if_ierrors++;
3025 				bge_newbuf_jumbo(sc, sc->bge_jumbo, m);
3026 				continue;
3027 			}
3028 		} else {
3029 			BGE_INC(sc->bge_std, BGE_STD_RX_RING_CNT);
3030 			m = sc->bge_cdata.bge_rx_std_chain[rxidx];
3031 
3032 			sc->bge_cdata.bge_rx_std_chain[rxidx] = NULL;
3033 			stdcnt++;
3034 			dmamap = sc->bge_cdata.bge_rx_std_map[rxidx];
3035 			sc->bge_cdata.bge_rx_std_map[rxidx] = 0;
3036 			bus_dmamap_sync(sc->bge_dmatag, dmamap, 0,
3037 			    dmamap->dm_mapsize, BUS_DMASYNC_POSTREAD);
3038 			bus_dmamap_unload(sc->bge_dmatag, dmamap);
3039 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
3040 				ifp->if_ierrors++;
3041 				bge_newbuf_std(sc, sc->bge_std, m, dmamap);
3042 				continue;
3043 			}
3044 			if (bge_newbuf_std(sc, sc->bge_std,
3045 			    NULL, dmamap) == ENOBUFS) {
3046 				ifp->if_ierrors++;
3047 				bge_newbuf_std(sc, sc->bge_std, m, dmamap);
3048 				continue;
3049 			}
3050 		}
3051 
3052 		ifp->if_ipackets++;
3053 #ifndef __NO_STRICT_ALIGNMENT
3054                 /*
3055                  * XXX: if the 5701 PCIX-Rx-DMA workaround is in effect,
3056                  * the Rx buffer has the layer-2 header unaligned.
3057                  * If our CPU requires alignment, re-align by copying.
3058                  */
3059 		if (sc->bge_flags & BGE_RX_ALIGNBUG) {
3060 			memmove(mtod(m, char *) + ETHER_ALIGN, m->m_data,
3061                                 cur_rx->bge_len);
3062 			m->m_data += ETHER_ALIGN;
3063 		}
3064 #endif
3065 
3066 		m->m_pkthdr.len = m->m_len = cur_rx->bge_len - ETHER_CRC_LEN;
3067 		m->m_pkthdr.rcvif = ifp;
3068 
3069 #if NBPFILTER > 0
3070 		/*
3071 		 * Handle BPF listeners. Let the BPF user see the packet.
3072 		 */
3073 		if (ifp->if_bpf)
3074 			bpf_mtap(ifp->if_bpf, m);
3075 #endif
3076 
3077 		m->m_pkthdr.csum_flags = M_CSUM_IPv4;
3078 
3079 		if ((cur_rx->bge_ip_csum ^ 0xffff) != 0)
3080 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
3081 		/*
3082 		 * Rx transport checksum-offload may also
3083 		 * have bugs with packets which, when transmitted,
3084 		 * were `runts' requiring padding.
3085 		 */
3086 		if (cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM &&
3087 		    (/* (sc->_bge_quirks & BGE_QUIRK_SHORT_CKSUM_BUG) == 0 ||*/
3088 		     m->m_pkthdr.len >= ETHER_MIN_NOPAD)) {
3089 			m->m_pkthdr.csum_data =
3090 			    cur_rx->bge_tcp_udp_csum;
3091 			m->m_pkthdr.csum_flags |=
3092 			    (M_CSUM_TCPv4|M_CSUM_UDPv4|
3093 			     M_CSUM_DATA|M_CSUM_NO_PSEUDOHDR);
3094 		}
3095 
3096 		/*
3097 		 * If we received a packet with a vlan tag, pass it
3098 		 * to vlan_input() instead of ether_input().
3099 		 */
3100 		if (cur_rx->bge_flags & BGE_RXBDFLAG_VLAN_TAG) {
3101 			VLAN_INPUT_TAG(ifp, m, cur_rx->bge_vlan_tag, continue);
3102 		}
3103 
3104 		(*ifp->if_input)(ifp, m);
3105 	}
3106 
3107 	bge_writembx(sc, BGE_MBX_RX_CONS0_LO, sc->bge_rx_saved_considx);
3108 	if (stdcnt)
3109 		bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bge_std);
3110 	if (jumbocnt)
3111 		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
3112 }
3113 
3114 static void
3115 bge_txeof(struct bge_softc *sc)
3116 {
3117 	struct bge_tx_bd *cur_tx = NULL;
3118 	struct ifnet *ifp;
3119 	struct txdmamap_pool_entry *dma;
3120 	bus_addr_t offset, toff;
3121 	bus_size_t tlen;
3122 	int tosync;
3123 	struct mbuf *m;
3124 
3125 	ifp = &sc->ethercom.ec_if;
3126 
3127 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
3128 	    offsetof(struct bge_ring_data, bge_status_block),
3129 	    sizeof (struct bge_status_block),
3130 	    BUS_DMASYNC_POSTREAD);
3131 
3132 	offset = offsetof(struct bge_ring_data, bge_tx_ring);
3133 	tosync = sc->bge_rdata->bge_status_block.bge_idx[0].bge_tx_cons_idx -
3134 	    sc->bge_tx_saved_considx;
3135 
3136 #if NRND > 0
3137 	if (tosync != 0 && RND_ENABLED(&sc->rnd_source))
3138 		rnd_add_uint32(&sc->rnd_source, tosync);
3139 #endif
3140 
3141 	toff = offset + (sc->bge_tx_saved_considx * sizeof (struct bge_tx_bd));
3142 
3143 	if (tosync < 0) {
3144 		tlen = (BGE_TX_RING_CNT - sc->bge_tx_saved_considx) *
3145 		    sizeof (struct bge_tx_bd);
3146 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
3147 		    toff, tlen, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
3148 		tosync = -tosync;
3149 	}
3150 
3151 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
3152 	    offset, tosync * sizeof (struct bge_tx_bd),
3153 	    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
3154 
3155 	/*
3156 	 * Go through our tx ring and free mbufs for those
3157 	 * frames that have been sent.
3158 	 */
3159 	while (sc->bge_tx_saved_considx !=
3160 	    sc->bge_rdata->bge_status_block.bge_idx[0].bge_tx_cons_idx) {
3161 		u_int32_t		idx = 0;
3162 
3163 		idx = sc->bge_tx_saved_considx;
3164 		cur_tx = &sc->bge_rdata->bge_tx_ring[idx];
3165 		if (cur_tx->bge_flags & BGE_TXBDFLAG_END)
3166 			ifp->if_opackets++;
3167 		m = sc->bge_cdata.bge_tx_chain[idx];
3168 		if (m != NULL) {
3169 			sc->bge_cdata.bge_tx_chain[idx] = NULL;
3170 			dma = sc->txdma[idx];
3171 			bus_dmamap_sync(sc->bge_dmatag, dma->dmamap, 0,
3172 			    dma->dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
3173 			bus_dmamap_unload(sc->bge_dmatag, dma->dmamap);
3174 			SLIST_INSERT_HEAD(&sc->txdma_list, dma, link);
3175 			sc->txdma[idx] = NULL;
3176 
3177 			m_freem(m);
3178 		}
3179 		sc->bge_txcnt--;
3180 		BGE_INC(sc->bge_tx_saved_considx, BGE_TX_RING_CNT);
3181 		ifp->if_timer = 0;
3182 	}
3183 
3184 	if (cur_tx != NULL)
3185 		ifp->if_flags &= ~IFF_OACTIVE;
3186 }
3187 
3188 static int
3189 bge_intr(void *xsc)
3190 {
3191 	struct bge_softc *sc;
3192 	struct ifnet *ifp;
3193 	uint32_t statusword;
3194 
3195 	sc = xsc;
3196 	ifp = &sc->ethercom.ec_if;
3197 
3198 	/* It is possible for the interrupt to arrive before
3199 	 * the status block is updated prior to the interrupt.
3200 	 * Reading the PCI State register will confirm whether the
3201 	 * interrupt is ours and will flush the status block.
3202 	 */
3203 
3204 	/* read status word from status block */
3205 	statusword = sc->bge_rdata->bge_status_block.bge_status;
3206 
3207 	if ((statusword & BGE_STATFLAG_UPDATED) ||
3208 	    (!(CSR_READ_4(sc, BGE_PCI_PCISTATE) & BGE_PCISTATE_INTR_NOT_ACTIVE))) {
3209 		/* Ack interrupt and stop others from occuring. */
3210 		bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
3211 
3212 		BGE_EVCNT_INCR(sc->bge_ev_intr);
3213 
3214 		/* clear status word */
3215 		sc->bge_rdata->bge_status_block.bge_status = 0;
3216 
3217 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
3218 		    statusword & BGE_STATFLAG_LINKSTATE_CHANGED ||
3219 		    BGE_STS_BIT(sc, BGE_STS_LINK_EVT))
3220 			bge_link_upd(sc);
3221 
3222 		if (ifp->if_flags & IFF_RUNNING) {
3223 			/* Check RX return ring producer/consumer */
3224 			bge_rxeof(sc);
3225 
3226 			/* Check TX ring producer/consumer */
3227 			bge_txeof(sc);
3228 		}
3229 
3230 		if (sc->bge_pending_rxintr_change) {
3231 			uint32_t rx_ticks = sc->bge_rx_coal_ticks;
3232 			uint32_t rx_bds = sc->bge_rx_max_coal_bds;
3233 			uint32_t junk;
3234 
3235 			CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, rx_ticks);
3236 			DELAY(10);
3237 			junk = CSR_READ_4(sc, BGE_HCC_RX_COAL_TICKS);
3238 
3239 			CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, rx_bds);
3240 			DELAY(10);
3241 			junk = CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS);
3242 
3243 			sc->bge_pending_rxintr_change = 0;
3244 		}
3245 		bge_handle_events(sc);
3246 
3247 		/* Re-enable interrupts. */
3248 		bge_writembx(sc, BGE_MBX_IRQ0_LO, 0);
3249 
3250 		if (ifp->if_flags & IFF_RUNNING && !IFQ_IS_EMPTY(&ifp->if_snd))
3251 			bge_start(ifp);
3252 
3253 		return (1);
3254 	} else
3255 		return (0);
3256 }
3257 
3258 static void
3259 bge_tick(void *xsc)
3260 {
3261 	struct bge_softc *sc = xsc;
3262 	struct mii_data *mii = &sc->bge_mii;
3263 	int s;
3264 
3265 	s = splnet();
3266 
3267 	bge_stats_update(sc);
3268 
3269 	if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
3270 		/*
3271 		 * Since in TBI mode auto-polling can't be used we should poll
3272 		 * link status manually. Here we register pending link event
3273 		 * and trigger interrupt.
3274 		 */
3275 		BGE_STS_SETBIT(sc, BGE_STS_LINK_EVT);
3276 		BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
3277 	} else {
3278 		/*
3279 		 * Do not touch PHY if we have link up. This could break
3280 		 * IPMI/ASF mode or produce extra input errors.
3281 		 * (extra input errors was reported for bcm5701 & bcm5704).
3282 		 */
3283 		if (!BGE_STS_BIT(sc, BGE_STS_LINK))
3284 			mii_tick(mii);
3285 	}
3286 
3287 	callout_reset(&sc->bge_timeout, hz, bge_tick, sc);
3288 
3289 	splx(s);
3290 }
3291 
3292 static void
3293 bge_stats_update(struct bge_softc *sc)
3294 {
3295 	struct ifnet *ifp = &sc->ethercom.ec_if;
3296 	bus_size_t stats = BGE_MEMWIN_START + BGE_STATS_BLOCK;
3297 	bus_size_t rstats = BGE_RX_STATS;
3298 
3299 #define READ_RSTAT(sc, stats, stat) \
3300 	  CSR_READ_4(sc, stats + offsetof(struct bge_mac_stats_regs, stat))
3301 
3302 	if (BGE_IS_5705_OR_BEYOND(sc)) {
3303 		ifp->if_collisions +=
3304 		    READ_RSTAT(sc, rstats, dot3StatsSingleCollisionFrames) +
3305 		    READ_RSTAT(sc, rstats, dot3StatsMultipleCollisionFrames) +
3306 		    READ_RSTAT(sc, rstats, dot3StatsExcessiveCollisions) +
3307 		    READ_RSTAT(sc, rstats, dot3StatsLateCollisions);
3308 
3309 		BGE_EVCNT_ADD(sc->bge_ev_tx_xoff,
3310 			      READ_RSTAT(sc, rstats, outXoffSent));
3311 		BGE_EVCNT_ADD(sc->bge_ev_tx_xon,
3312 			      READ_RSTAT(sc, rstats, outXonSent));
3313 		BGE_EVCNT_ADD(sc->bge_ev_rx_xoff,
3314 			      READ_RSTAT(sc, rstats, xoffPauseFramesReceived));
3315 		BGE_EVCNT_ADD(sc->bge_ev_rx_xon,
3316 			      READ_RSTAT(sc, rstats, xonPauseFramesReceived));
3317 		BGE_EVCNT_ADD(sc->bge_ev_rx_macctl,
3318 			      READ_RSTAT(sc, rstats, macControlFramesReceived));
3319 		BGE_EVCNT_ADD(sc->bge_ev_xoffentered,
3320 			      READ_RSTAT(sc, rstats, xoffStateEntered));
3321 		return;
3322 	}
3323 
3324 #undef READ_RSTAT
3325 #define READ_STAT(sc, stats, stat) \
3326 	  CSR_READ_4(sc, stats + offsetof(struct bge_stats, stat))
3327 
3328 	ifp->if_collisions +=
3329 	  (READ_STAT(sc, stats, dot3StatsSingleCollisionFrames.bge_addr_lo) +
3330 	   READ_STAT(sc, stats, dot3StatsMultipleCollisionFrames.bge_addr_lo) +
3331 	   READ_STAT(sc, stats, dot3StatsExcessiveCollisions.bge_addr_lo) +
3332 	   READ_STAT(sc, stats, dot3StatsLateCollisions.bge_addr_lo)) -
3333 	  ifp->if_collisions;
3334 
3335 	BGE_EVCNT_UPD(sc->bge_ev_tx_xoff,
3336 		      READ_STAT(sc, stats, outXoffSent.bge_addr_lo));
3337 	BGE_EVCNT_UPD(sc->bge_ev_tx_xon,
3338 		      READ_STAT(sc, stats, outXonSent.bge_addr_lo));
3339 	BGE_EVCNT_UPD(sc->bge_ev_rx_xoff,
3340 		      READ_STAT(sc, stats,
3341 		      		xoffPauseFramesReceived.bge_addr_lo));
3342 	BGE_EVCNT_UPD(sc->bge_ev_rx_xon,
3343 		      READ_STAT(sc, stats, xonPauseFramesReceived.bge_addr_lo));
3344 	BGE_EVCNT_UPD(sc->bge_ev_rx_macctl,
3345 		      READ_STAT(sc, stats,
3346 		      		macControlFramesReceived.bge_addr_lo));
3347 	BGE_EVCNT_UPD(sc->bge_ev_xoffentered,
3348 		      READ_STAT(sc, stats, xoffStateEntered.bge_addr_lo));
3349 
3350 #undef READ_STAT
3351 
3352 #ifdef notdef
3353 	ifp->if_collisions +=
3354 	   (sc->bge_rdata->bge_info.bge_stats.dot3StatsSingleCollisionFrames +
3355 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsMultipleCollisionFrames +
3356 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsExcessiveCollisions +
3357 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsLateCollisions) -
3358 	   ifp->if_collisions;
3359 #endif
3360 }
3361 
3362 /*
3363  * Pad outbound frame to ETHER_MIN_NOPAD for an unusual reason.
3364  * The bge hardware will pad out Tx runts to ETHER_MIN_NOPAD,
3365  * but when such padded frames employ the  bge IP/TCP checksum offload,
3366  * the hardware checksum assist gives incorrect results (possibly
3367  * from incorporating its own padding into the UDP/TCP checksum; who knows).
3368  * If we pad such runts with zeros, the onboard checksum comes out correct.
3369  */
3370 static inline int
3371 bge_cksum_pad(struct mbuf *pkt)
3372 {
3373 	struct mbuf *last = NULL;
3374 	int padlen;
3375 
3376 	padlen = ETHER_MIN_NOPAD - pkt->m_pkthdr.len;
3377 
3378 	/* if there's only the packet-header and we can pad there, use it. */
3379 	if (pkt->m_pkthdr.len == pkt->m_len &&
3380 	    M_TRAILINGSPACE(pkt) >= padlen) {
3381 		last = pkt;
3382 	} else {
3383 		/*
3384 		 * Walk packet chain to find last mbuf. We will either
3385 		 * pad there, or append a new mbuf and pad it
3386 		 * (thus perhaps avoiding the bcm5700 dma-min bug).
3387 		 */
3388 		for (last = pkt; last->m_next != NULL; last = last->m_next) {
3389 	      	       continue; /* do nothing */
3390 		}
3391 
3392 		/* `last' now points to last in chain. */
3393 		if (M_TRAILINGSPACE(last) < padlen) {
3394 			/* Allocate new empty mbuf, pad it. Compact later. */
3395 			struct mbuf *n;
3396 			MGET(n, M_DONTWAIT, MT_DATA);
3397 			if (n == NULL)
3398 				return ENOBUFS;
3399 			n->m_len = 0;
3400 			last->m_next = n;
3401 			last = n;
3402 		}
3403 	}
3404 
3405 	KDASSERT(!M_READONLY(last));
3406 	KDASSERT(M_TRAILINGSPACE(last) >= padlen);
3407 
3408 	/* Now zero the pad area, to avoid the bge cksum-assist bug */
3409 	memset(mtod(last, char *) + last->m_len, 0, padlen);
3410 	last->m_len += padlen;
3411 	pkt->m_pkthdr.len += padlen;
3412 	return 0;
3413 }
3414 
3415 /*
3416  * Compact outbound packets to avoid bug with DMA segments less than 8 bytes.
3417  */
3418 static inline int
3419 bge_compact_dma_runt(struct mbuf *pkt)
3420 {
3421 	struct mbuf	*m, *prev;
3422 	int 		totlen, prevlen;
3423 
3424 	prev = NULL;
3425 	totlen = 0;
3426 	prevlen = -1;
3427 
3428 	for (m = pkt; m != NULL; prev = m,m = m->m_next) {
3429 		int mlen = m->m_len;
3430 		int shortfall = 8 - mlen ;
3431 
3432 		totlen += mlen;
3433 		if (mlen == 0) {
3434 			continue;
3435 		}
3436 		if (mlen >= 8)
3437 			continue;
3438 
3439 		/* If we get here, mbuf data is too small for DMA engine.
3440 		 * Try to fix by shuffling data to prev or next in chain.
3441 		 * If that fails, do a compacting deep-copy of the whole chain.
3442 		 */
3443 
3444 		/* Internal frag. If fits in prev, copy it there. */
3445 		if (prev && M_TRAILINGSPACE(prev) >= m->m_len) {
3446 		  	memcpy(prev->m_data + prev->m_len, m->m_data, mlen);
3447 			prev->m_len += mlen;
3448 			m->m_len = 0;
3449 			/* XXX stitch chain */
3450 			prev->m_next = m_free(m);
3451 			m = prev;
3452 			continue;
3453 		}
3454 		else if (m->m_next != NULL &&
3455 			     M_TRAILINGSPACE(m) >= shortfall &&
3456 			     m->m_next->m_len >= (8 + shortfall)) {
3457 		    /* m is writable and have enough data in next, pull up. */
3458 
3459 		  	memcpy(m->m_data + m->m_len, m->m_next->m_data,
3460 			    shortfall);
3461 			m->m_len += shortfall;
3462 			m->m_next->m_len -= shortfall;
3463 			m->m_next->m_data += shortfall;
3464 		}
3465 		else if (m->m_next == NULL || 1) {
3466 		  	/* Got a runt at the very end of the packet.
3467 			 * borrow data from the tail of the preceding mbuf and
3468 			 * update its length in-place. (The original data is still
3469 			 * valid, so we can do this even if prev is not writable.)
3470 			 */
3471 
3472 			/* if we'd make prev a runt, just move all of its data. */
3473 			KASSERT(prev != NULL /*, ("runt but null PREV")*/);
3474 			KASSERT(prev->m_len >= 8 /*, ("runt prev")*/);
3475 
3476 			if ((prev->m_len - shortfall) < 8)
3477 				shortfall = prev->m_len;
3478 
3479 #ifdef notyet	/* just do the safe slow thing for now */
3480 			if (!M_READONLY(m)) {
3481 				if (M_LEADINGSPACE(m) < shorfall) {
3482 					void *m_dat;
3483 					m_dat = (m->m_flags & M_PKTHDR) ?
3484 					  m->m_pktdat : m->dat;
3485 					memmove(m_dat, mtod(m, void*), m->m_len);
3486 					m->m_data = m_dat;
3487 				    }
3488 			} else
3489 #endif	/* just do the safe slow thing */
3490 			{
3491 				struct mbuf * n = NULL;
3492 				int newprevlen = prev->m_len - shortfall;
3493 
3494 				MGET(n, M_NOWAIT, MT_DATA);
3495 				if (n == NULL)
3496 				   return ENOBUFS;
3497 				KASSERT(m->m_len + shortfall < MLEN
3498 					/*,
3499 					  ("runt %d +prev %d too big\n", m->m_len, shortfall)*/);
3500 
3501 				/* first copy the data we're stealing from prev */
3502 				memcpy(n->m_data, prev->m_data + newprevlen,
3503 				    shortfall);
3504 
3505 				/* update prev->m_len accordingly */
3506 				prev->m_len -= shortfall;
3507 
3508 				/* copy data from runt m */
3509 				memcpy(n->m_data + shortfall, m->m_data,
3510 				    m->m_len);
3511 
3512 				/* n holds what we stole from prev, plus m */
3513 				n->m_len = shortfall + m->m_len;
3514 
3515 				/* stitch n into chain and free m */
3516 				n->m_next = m->m_next;
3517 				prev->m_next = n;
3518 				/* KASSERT(m->m_next == NULL); */
3519 				m->m_next = NULL;
3520 				m_free(m);
3521 				m = n;	/* for continuing loop */
3522 			}
3523 		}
3524 		prevlen = m->m_len;
3525 	}
3526 	return 0;
3527 }
3528 
3529 /*
3530  * Encapsulate an mbuf chain in the tx ring  by coupling the mbuf data
3531  * pointers to descriptors.
3532  */
3533 static int
3534 bge_encap(struct bge_softc *sc, struct mbuf *m_head, u_int32_t *txidx)
3535 {
3536 	struct bge_tx_bd	*f = NULL;
3537 	u_int32_t		frag, cur;
3538 	u_int16_t		csum_flags = 0;
3539 	u_int16_t		txbd_tso_flags = 0;
3540 	struct txdmamap_pool_entry *dma;
3541 	bus_dmamap_t dmamap;
3542 	int			i = 0;
3543 	struct m_tag		*mtag;
3544 	int			use_tso, maxsegsize, error;
3545 
3546 	cur = frag = *txidx;
3547 
3548 	if (m_head->m_pkthdr.csum_flags) {
3549 		if (m_head->m_pkthdr.csum_flags & M_CSUM_IPv4)
3550 			csum_flags |= BGE_TXBDFLAG_IP_CSUM;
3551 		if (m_head->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4))
3552 			csum_flags |= BGE_TXBDFLAG_TCP_UDP_CSUM;
3553 	}
3554 
3555 	/*
3556 	 * If we were asked to do an outboard checksum, and the NIC
3557 	 * has the bug where it sometimes adds in the Ethernet padding,
3558 	 * explicitly pad with zeros so the cksum will be correct either way.
3559 	 * (For now, do this for all chip versions, until newer
3560 	 * are confirmed to not require the workaround.)
3561 	 */
3562 	if ((csum_flags & BGE_TXBDFLAG_TCP_UDP_CSUM) == 0 ||
3563 #ifdef notyet
3564 	    (sc->bge_quirks & BGE_QUIRK_SHORT_CKSUM_BUG) == 0 ||
3565 #endif
3566 	    m_head->m_pkthdr.len >= ETHER_MIN_NOPAD)
3567 		goto check_dma_bug;
3568 
3569 	if (bge_cksum_pad(m_head) != 0) {
3570 	    return ENOBUFS;
3571 	}
3572 
3573 check_dma_bug:
3574 	if (!(BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX))
3575 		goto doit;
3576 
3577 	/*
3578 	 * bcm5700 Revision B silicon cannot handle DMA descriptors with
3579 	 * less than eight bytes.  If we encounter a teeny mbuf
3580 	 * at the end of a chain, we can pad.  Otherwise, copy.
3581 	 */
3582 	if (bge_compact_dma_runt(m_head) != 0)
3583 		return ENOBUFS;
3584 
3585 doit:
3586 	dma = SLIST_FIRST(&sc->txdma_list);
3587 	if (dma == NULL)
3588 		return ENOBUFS;
3589 	dmamap = dma->dmamap;
3590 
3591 	/*
3592 	 * Set up any necessary TSO state before we start packing...
3593 	 */
3594 	use_tso = (m_head->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0;
3595 	if (!use_tso) {
3596 		maxsegsize = 0;
3597 	} else {	/* TSO setup */
3598 		unsigned  mss;
3599 		struct ether_header *eh;
3600 		unsigned ip_tcp_hlen, iptcp_opt_words, tcp_seg_flags, offset;
3601 		struct mbuf * m0 = m_head;
3602 		struct ip *ip;
3603 		struct tcphdr *th;
3604 		int iphl, hlen;
3605 
3606 		/*
3607 		 * XXX It would be nice if the mbuf pkthdr had offset
3608 		 * fields for the protocol headers.
3609 		 */
3610 
3611 		eh = mtod(m0, struct ether_header *);
3612 		switch (htons(eh->ether_type)) {
3613 		case ETHERTYPE_IP:
3614 			offset = ETHER_HDR_LEN;
3615 			break;
3616 
3617 		case ETHERTYPE_VLAN:
3618 			offset = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
3619 			break;
3620 
3621 		default:
3622 			/*
3623 			 * Don't support this protocol or encapsulation.
3624 			 */
3625 			return (ENOBUFS);
3626 		}
3627 
3628 		/*
3629 		 * TCP/IP headers are in the first mbuf; we can do
3630 		 * this the easy way.
3631 		 */
3632 		iphl = M_CSUM_DATA_IPv4_IPHL(m0->m_pkthdr.csum_data);
3633 		hlen = iphl + offset;
3634 		if (__predict_false(m0->m_len <
3635 				    (hlen + sizeof(struct tcphdr)))) {
3636 
3637 			aprint_debug_dev(sc->bge_dev,
3638 			    "TSO: hard case m0->m_len == %d < ip/tcp hlen %zd,"
3639 			    "not handled yet\n",
3640 			     m0->m_len, hlen+ sizeof(struct tcphdr));
3641 #ifdef NOTYET
3642 			/*
3643 			 * XXX jonathan@NetBSD.org: untested.
3644 			 * how to force  this branch to be taken?
3645 			 */
3646 			BGE_EVCNT_INCR(&sc->sc_ev_txtsopain);
3647 
3648 			m_copydata(m0, offset, sizeof(ip), &ip);
3649 			m_copydata(m0, hlen, sizeof(th), &th);
3650 
3651 			ip.ip_len = 0;
3652 
3653 			m_copyback(m0, hlen + offsetof(struct ip, ip_len),
3654 			    sizeof(ip.ip_len), &ip.ip_len);
3655 
3656 			th.th_sum = in_cksum_phdr(ip.ip_src.s_addr,
3657 			    ip.ip_dst.s_addr, htons(IPPROTO_TCP));
3658 
3659 			m_copyback(m0, hlen + offsetof(struct tcphdr, th_sum),
3660 			    sizeof(th.th_sum), &th.th_sum);
3661 
3662 			hlen += th.th_off << 2;
3663 			iptcp_opt_words	= hlen;
3664 #else
3665 			/*
3666 			 * if_wm "hard" case not yet supported, can we not
3667 			 * mandate it out of existence?
3668 			 */
3669 			(void) ip; (void)th; (void) ip_tcp_hlen;
3670 
3671 			return ENOBUFS;
3672 #endif
3673 		} else {
3674 			ip = (struct ip *) (mtod(m0, char *) + offset);
3675 			th = (struct tcphdr *) (mtod(m0, char *) + hlen);
3676 			ip_tcp_hlen = iphl +  (th->th_off << 2);
3677 
3678 			/* Total IP/TCP options, in 32-bit words */
3679 			iptcp_opt_words = (ip_tcp_hlen
3680 					   - sizeof(struct tcphdr)
3681 					   - sizeof(struct ip)) >> 2;
3682 		}
3683 		if (BGE_IS_5750_OR_BEYOND(sc)) {
3684 			th->th_sum = 0;
3685 			csum_flags &= ~(BGE_TXBDFLAG_TCP_UDP_CSUM);
3686 		} else {
3687 			/*
3688 			 * XXX jonathan@NetBSD.org: 5705 untested.
3689 			 * Requires TSO firmware patch for 5701/5703/5704.
3690 			 */
3691 			th->th_sum = in_cksum_phdr(ip->ip_src.s_addr,
3692 			    ip->ip_dst.s_addr, htons(IPPROTO_TCP));
3693 		}
3694 
3695 		mss = m_head->m_pkthdr.segsz;
3696 		txbd_tso_flags |=
3697 		    BGE_TXBDFLAG_CPU_PRE_DMA |
3698 		    BGE_TXBDFLAG_CPU_POST_DMA;
3699 
3700 		/*
3701 		 * Our NIC TSO-assist assumes TSO has standard, optionless
3702 		 * IPv4 and TCP headers, which total 40 bytes. By default,
3703 		 * the NIC copies 40 bytes of IP/TCP header from the
3704 		 * supplied header into the IP/TCP header portion of
3705 		 * each post-TSO-segment. If the supplied packet has IP or
3706 		 * TCP options, we need to tell the NIC to copy those extra
3707 		 * bytes into each  post-TSO header, in addition to the normal
3708 		 * 40-byte IP/TCP header (and to leave space accordingly).
3709 		 * Unfortunately, the driver encoding of option length
3710 		 * varies across different ASIC families.
3711 		 */
3712 		tcp_seg_flags = 0;
3713 		if (iptcp_opt_words) {
3714 			if ( BGE_IS_5705_OR_BEYOND(sc)) {
3715 				tcp_seg_flags =
3716 					iptcp_opt_words << 11;
3717 			} else {
3718 				txbd_tso_flags |=
3719 					iptcp_opt_words << 12;
3720 			}
3721 		}
3722 		maxsegsize = mss | tcp_seg_flags;
3723 		ip->ip_len = htons(mss + ip_tcp_hlen);
3724 
3725 	}	/* TSO setup */
3726 
3727 	/*
3728 	 * Start packing the mbufs in this chain into
3729 	 * the fragment pointers. Stop when we run out
3730 	 * of fragments or hit the end of the mbuf chain.
3731 	 */
3732 	error = bus_dmamap_load_mbuf(sc->bge_dmatag, dmamap, m_head,
3733 	    BUS_DMA_NOWAIT);
3734 	if (error) {
3735 		return (ENOBUFS);
3736 	}
3737 	/*
3738 	 * Sanity check: avoid coming within 16 descriptors
3739 	 * of the end of the ring.
3740 	 */
3741 	if (dmamap->dm_nsegs > (BGE_TX_RING_CNT - sc->bge_txcnt - 16)) {
3742 		BGE_TSO_PRINTF(("%s: "
3743 		    " dmamap_load_mbuf too close to ring wrap\n",
3744 		    device_xname(sc->bge_dev)));
3745 		goto fail_unload;
3746 	}
3747 
3748 	mtag = sc->ethercom.ec_nvlans ?
3749 	    m_tag_find(m_head, PACKET_TAG_VLAN, NULL) : NULL;
3750 
3751 
3752 	/* Iterate over dmap-map fragments. */
3753 	for (i = 0; i < dmamap->dm_nsegs; i++) {
3754 		f = &sc->bge_rdata->bge_tx_ring[frag];
3755 		if (sc->bge_cdata.bge_tx_chain[frag] != NULL)
3756 			break;
3757 
3758 		bge_set_hostaddr(&f->bge_addr, dmamap->dm_segs[i].ds_addr);
3759 		f->bge_len = dmamap->dm_segs[i].ds_len;
3760 
3761 		/*
3762 		 * For 5751 and follow-ons, for TSO we must turn
3763 		 * off checksum-assist flag in the tx-descr, and
3764 		 * supply the ASIC-revision-specific encoding
3765 		 * of TSO flags and segsize.
3766 		 */
3767 		if (use_tso) {
3768 			if (BGE_IS_5750_OR_BEYOND(sc) || i == 0) {
3769 				f->bge_rsvd = maxsegsize;
3770 				f->bge_flags = csum_flags | txbd_tso_flags;
3771 			} else {
3772 				f->bge_rsvd = 0;
3773 				f->bge_flags =
3774 				  (csum_flags | txbd_tso_flags) & 0x0fff;
3775 			}
3776 		} else {
3777 			f->bge_rsvd = 0;
3778 			f->bge_flags = csum_flags;
3779 		}
3780 
3781 		if (mtag != NULL) {
3782 			f->bge_flags |= BGE_TXBDFLAG_VLAN_TAG;
3783 			f->bge_vlan_tag = VLAN_TAG_VALUE(mtag);
3784 		} else {
3785 			f->bge_vlan_tag = 0;
3786 		}
3787 		cur = frag;
3788 		BGE_INC(frag, BGE_TX_RING_CNT);
3789 	}
3790 
3791 	if (i < dmamap->dm_nsegs) {
3792 		BGE_TSO_PRINTF(("%s: reached %d < dm_nsegs %d\n",
3793 		    device_xname(sc->bge_dev), i, dmamap->dm_nsegs));
3794 		goto fail_unload;
3795 	}
3796 
3797 	bus_dmamap_sync(sc->bge_dmatag, dmamap, 0, dmamap->dm_mapsize,
3798 	    BUS_DMASYNC_PREWRITE);
3799 
3800 	if (frag == sc->bge_tx_saved_considx) {
3801 		BGE_TSO_PRINTF(("%s: frag %d = wrapped id %d?\n",
3802 		    device_xname(sc->bge_dev), frag, sc->bge_tx_saved_considx));
3803 
3804 		goto fail_unload;
3805 	}
3806 
3807 	sc->bge_rdata->bge_tx_ring[cur].bge_flags |= BGE_TXBDFLAG_END;
3808 	sc->bge_cdata.bge_tx_chain[cur] = m_head;
3809 	SLIST_REMOVE_HEAD(&sc->txdma_list, link);
3810 	sc->txdma[cur] = dma;
3811 	sc->bge_txcnt += dmamap->dm_nsegs;
3812 
3813 	*txidx = frag;
3814 
3815 	return (0);
3816 
3817 fail_unload:
3818 	bus_dmamap_unload(sc->bge_dmatag, dmamap);
3819 
3820 	return ENOBUFS;
3821 }
3822 
3823 /*
3824  * Main transmit routine. To avoid having to do mbuf copies, we put pointers
3825  * to the mbuf data regions directly in the transmit descriptors.
3826  */
3827 static void
3828 bge_start(struct ifnet *ifp)
3829 {
3830 	struct bge_softc *sc;
3831 	struct mbuf *m_head = NULL;
3832 	u_int32_t prodidx;
3833 	int pkts = 0;
3834 
3835 	sc = ifp->if_softc;
3836 
3837 	if ((ifp->if_flags & (IFF_RUNNING|IFF_OACTIVE)) != IFF_RUNNING)
3838 		return;
3839 
3840 	prodidx = sc->bge_tx_prodidx;
3841 
3842 	while(sc->bge_cdata.bge_tx_chain[prodidx] == NULL) {
3843 		IFQ_POLL(&ifp->if_snd, m_head);
3844 		if (m_head == NULL)
3845 			break;
3846 
3847 #if 0
3848 		/*
3849 		 * XXX
3850 		 * safety overkill.  If this is a fragmented packet chain
3851 		 * with delayed TCP/UDP checksums, then only encapsulate
3852 		 * it if we have enough descriptors to handle the entire
3853 		 * chain at once.
3854 		 * (paranoia -- may not actually be needed)
3855 		 */
3856 		if (m_head->m_flags & M_FIRSTFRAG &&
3857 		    m_head->m_pkthdr.csum_flags & (CSUM_DELAY_DATA)) {
3858 			if ((BGE_TX_RING_CNT - sc->bge_txcnt) <
3859 			    M_CSUM_DATA_IPv4_OFFSET(m_head->m_pkthdr.csum_data) + 16) {
3860 				ifp->if_flags |= IFF_OACTIVE;
3861 				break;
3862 			}
3863 		}
3864 #endif
3865 
3866 		/*
3867 		 * Pack the data into the transmit ring. If we
3868 		 * don't have room, set the OACTIVE flag and wait
3869 		 * for the NIC to drain the ring.
3870 		 */
3871 		if (bge_encap(sc, m_head, &prodidx)) {
3872 			ifp->if_flags |= IFF_OACTIVE;
3873 			break;
3874 		}
3875 
3876 		/* now we are committed to transmit the packet */
3877 		IFQ_DEQUEUE(&ifp->if_snd, m_head);
3878 		pkts++;
3879 
3880 #if NBPFILTER > 0
3881 		/*
3882 		 * If there's a BPF listener, bounce a copy of this frame
3883 		 * to him.
3884 		 */
3885 		if (ifp->if_bpf)
3886 			bpf_mtap(ifp->if_bpf, m_head);
3887 #endif
3888 	}
3889 	if (pkts == 0)
3890 		return;
3891 
3892 	/* Transmit */
3893 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
3894 	/* 5700 b2 errata */
3895 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX)
3896 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
3897 
3898 	sc->bge_tx_prodidx = prodidx;
3899 
3900 	/*
3901 	 * Set a timeout in case the chip goes out to lunch.
3902 	 */
3903 	ifp->if_timer = 5;
3904 }
3905 
3906 static int
3907 bge_init(struct ifnet *ifp)
3908 {
3909 	struct bge_softc *sc = ifp->if_softc;
3910 	const u_int16_t *m;
3911 	int s, error = 0;
3912 
3913 	s = splnet();
3914 
3915 	ifp = &sc->ethercom.ec_if;
3916 
3917 	/* Cancel pending I/O and flush buffers. */
3918 	bge_stop(ifp, 0);
3919 	bge_reset(sc);
3920 	bge_chipinit(sc);
3921 
3922 	/*
3923 	 * Init the various state machines, ring
3924 	 * control blocks and firmware.
3925 	 */
3926 	error = bge_blockinit(sc);
3927 	if (error != 0) {
3928 		aprint_error_dev(sc->bge_dev, "initialization error %d\n",
3929 		    error);
3930 		splx(s);
3931 		return error;
3932 	}
3933 
3934 	ifp = &sc->ethercom.ec_if;
3935 
3936 	/* Specify MTU. */
3937 	CSR_WRITE_4(sc, BGE_RX_MTU, ifp->if_mtu +
3938 	    ETHER_HDR_LEN + ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN);
3939 
3940 	/* Load our MAC address. */
3941 	m = (const u_int16_t *)&(CLLADDR(ifp->if_sadl)[0]);
3942 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_LO, htons(m[0]));
3943 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_HI, (htons(m[1]) << 16) | htons(m[2]));
3944 
3945 	/* Enable or disable promiscuous mode as needed. */
3946 	if (ifp->if_flags & IFF_PROMISC) {
3947 		BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
3948 	} else {
3949 		BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
3950 	}
3951 
3952 	/* Program multicast filter. */
3953 	bge_setmulti(sc);
3954 
3955 	/* Init RX ring. */
3956 	bge_init_rx_ring_std(sc);
3957 
3958 	/*
3959 	 * Workaround for a bug in 5705 ASIC rev A0. Poll the NIC's
3960 	 * memory to insure that the chip has in fact read the first
3961 	 * entry of the ring.
3962 	 */
3963 	if (sc->bge_chipid == BGE_CHIPID_BCM5705_A0) {
3964 		u_int32_t		v, i;
3965 		for (i = 0; i < 10; i++) {
3966 			DELAY(20);
3967 			v = bge_readmem_ind(sc, BGE_STD_RX_RINGS + 8);
3968 			if (v == (MCLBYTES - ETHER_ALIGN))
3969 				break;
3970 		}
3971 		if (i == 10)
3972 			aprint_error_dev(sc->bge_dev,
3973 			    "5705 A0 chip failed to load RX ring\n");
3974 	}
3975 
3976 	/* Init jumbo RX ring. */
3977 	if (ifp->if_mtu > (ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN))
3978 		bge_init_rx_ring_jumbo(sc);
3979 
3980 	/* Init our RX return ring index */
3981 	sc->bge_rx_saved_considx = 0;
3982 
3983 	/* Init TX ring. */
3984 	bge_init_tx_ring(sc);
3985 
3986 	/* Turn on transmitter */
3987 	BGE_SETBIT(sc, BGE_TX_MODE, BGE_TXMODE_ENABLE);
3988 
3989 	/* Turn on receiver */
3990 	BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
3991 
3992 	CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 2);
3993 
3994 	/* Tell firmware we're alive. */
3995 	BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3996 
3997 	/* Enable host interrupts. */
3998 	BGE_SETBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_CLEAR_INTA);
3999 	BGE_CLRBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR);
4000 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 0);
4001 
4002 	if ((error = bge_ifmedia_upd(ifp)) != 0)
4003 		goto out;
4004 
4005 	ifp->if_flags |= IFF_RUNNING;
4006 	ifp->if_flags &= ~IFF_OACTIVE;
4007 
4008 	callout_reset(&sc->bge_timeout, hz, bge_tick, sc);
4009 
4010 out:
4011 	splx(s);
4012 
4013 	return error;
4014 }
4015 
4016 /*
4017  * Set media options.
4018  */
4019 static int
4020 bge_ifmedia_upd(struct ifnet *ifp)
4021 {
4022 	struct bge_softc *sc = ifp->if_softc;
4023 	struct mii_data *mii = &sc->bge_mii;
4024 	struct ifmedia *ifm = &sc->bge_ifmedia;
4025 	int rc;
4026 
4027 	/* If this is a 1000baseX NIC, enable the TBI port. */
4028 	if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
4029 		if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
4030 			return (EINVAL);
4031 		switch(IFM_SUBTYPE(ifm->ifm_media)) {
4032 		case IFM_AUTO:
4033 			/*
4034 			 * The BCM5704 ASIC appears to have a special
4035 			 * mechanism for programming the autoneg
4036 			 * advertisement registers in TBI mode.
4037 			 */
4038 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704) {
4039 				u_int32_t sgdig;
4040 				sgdig = CSR_READ_4(sc, BGE_SGDIG_STS);
4041 				if (sgdig & BGE_SGDIGSTS_DONE) {
4042 					CSR_WRITE_4(sc, BGE_TX_TBI_AUTONEG, 0);
4043 					sgdig = CSR_READ_4(sc, BGE_SGDIG_CFG);
4044 					sgdig |= BGE_SGDIGCFG_AUTO |
4045 					    BGE_SGDIGCFG_PAUSE_CAP |
4046 					    BGE_SGDIGCFG_ASYM_PAUSE;
4047 					CSR_WRITE_4(sc, BGE_SGDIG_CFG,
4048 					    sgdig | BGE_SGDIGCFG_SEND);
4049 					DELAY(5);
4050 					CSR_WRITE_4(sc, BGE_SGDIG_CFG, sgdig);
4051 				}
4052 			}
4053 			break;
4054 		case IFM_1000_SX:
4055 			if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX) {
4056 				BGE_CLRBIT(sc, BGE_MAC_MODE,
4057 				    BGE_MACMODE_HALF_DUPLEX);
4058 			} else {
4059 				BGE_SETBIT(sc, BGE_MAC_MODE,
4060 				    BGE_MACMODE_HALF_DUPLEX);
4061 			}
4062 			break;
4063 		default:
4064 			return (EINVAL);
4065 		}
4066 		/* XXX 802.3x flow control for 1000BASE-SX */
4067 		return (0);
4068 	}
4069 
4070 	BGE_STS_SETBIT(sc, BGE_STS_LINK_EVT);
4071 	if ((rc = mii_mediachg(mii)) == ENXIO)
4072 		return 0;
4073 
4074 	/*
4075 	 * Force an interrupt so that we will call bge_link_upd
4076 	 * if needed and clear any pending link state attention.
4077 	 * Without this we are not getting any further interrupts
4078 	 * for link state changes and thus will not UP the link and
4079 	 * not be able to send in bge_start. The only way to get
4080 	 * things working was to receive a packet and get a RX intr.
4081 	 */
4082 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
4083 	    sc->bge_flags & BGE_IS_5788)
4084 		BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
4085 	else
4086 		BGE_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
4087 
4088 	return rc;
4089 }
4090 
4091 /*
4092  * Report current media status.
4093  */
4094 static void
4095 bge_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
4096 {
4097 	struct bge_softc *sc = ifp->if_softc;
4098 	struct mii_data *mii = &sc->bge_mii;
4099 
4100 	if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
4101 		ifmr->ifm_status = IFM_AVALID;
4102 		ifmr->ifm_active = IFM_ETHER;
4103 		if (CSR_READ_4(sc, BGE_MAC_STS) &
4104 		    BGE_MACSTAT_TBI_PCS_SYNCHED)
4105 			ifmr->ifm_status |= IFM_ACTIVE;
4106 		ifmr->ifm_active |= IFM_1000_SX;
4107 		if (CSR_READ_4(sc, BGE_MAC_MODE) & BGE_MACMODE_HALF_DUPLEX)
4108 			ifmr->ifm_active |= IFM_HDX;
4109 		else
4110 			ifmr->ifm_active |= IFM_FDX;
4111 		return;
4112 	}
4113 
4114 	mii_pollstat(mii);
4115 	ifmr->ifm_status = mii->mii_media_status;
4116 	ifmr->ifm_active = (mii->mii_media_active & ~IFM_ETH_FMASK) |
4117 	    sc->bge_flowflags;
4118 }
4119 
4120 static int
4121 bge_ioctl(struct ifnet *ifp, u_long command, void *data)
4122 {
4123 	struct bge_softc *sc = ifp->if_softc;
4124 	struct ifreq *ifr = (struct ifreq *) data;
4125 	int s, error = 0;
4126 	struct mii_data *mii;
4127 
4128 	s = splnet();
4129 
4130 	switch(command) {
4131 	case SIOCSIFFLAGS:
4132 		if ((error = ifioctl_common(ifp, command, data)) != 0)
4133 			break;
4134 		if (ifp->if_flags & IFF_UP) {
4135 			/*
4136 			 * If only the state of the PROMISC flag changed,
4137 			 * then just use the 'set promisc mode' command
4138 			 * instead of reinitializing the entire NIC. Doing
4139 			 * a full re-init means reloading the firmware and
4140 			 * waiting for it to start up, which may take a
4141 			 * second or two.
4142 			 */
4143 			if (ifp->if_flags & IFF_RUNNING &&
4144 			    ifp->if_flags & IFF_PROMISC &&
4145 			    !(sc->bge_if_flags & IFF_PROMISC)) {
4146 				BGE_SETBIT(sc, BGE_RX_MODE,
4147 				    BGE_RXMODE_RX_PROMISC);
4148 			} else if (ifp->if_flags & IFF_RUNNING &&
4149 			    !(ifp->if_flags & IFF_PROMISC) &&
4150 			    sc->bge_if_flags & IFF_PROMISC) {
4151 				BGE_CLRBIT(sc, BGE_RX_MODE,
4152 				    BGE_RXMODE_RX_PROMISC);
4153 			} else if (!(sc->bge_if_flags & IFF_UP))
4154 				bge_init(ifp);
4155 		} else {
4156 			if (ifp->if_flags & IFF_RUNNING)
4157 				bge_stop(ifp, 1);
4158 		}
4159 		sc->bge_if_flags = ifp->if_flags;
4160 		error = 0;
4161 		break;
4162 	case SIOCSIFMEDIA:
4163 		/* XXX Flow control is not supported for 1000BASE-SX */
4164 		if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
4165 			ifr->ifr_media &= ~IFM_ETH_FMASK;
4166 			sc->bge_flowflags = 0;
4167 		}
4168 
4169 		/* Flow control requires full-duplex mode. */
4170 		if (IFM_SUBTYPE(ifr->ifr_media) == IFM_AUTO ||
4171 		    (ifr->ifr_media & IFM_FDX) == 0) {
4172 		    	ifr->ifr_media &= ~IFM_ETH_FMASK;
4173 		}
4174 		if (IFM_SUBTYPE(ifr->ifr_media) != IFM_AUTO) {
4175 			if ((ifr->ifr_media & IFM_ETH_FMASK) == IFM_FLOW) {
4176 				/* We can do both TXPAUSE and RXPAUSE. */
4177 				ifr->ifr_media |=
4178 				    IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE;
4179 			}
4180 			sc->bge_flowflags = ifr->ifr_media & IFM_ETH_FMASK;
4181 		}
4182 		/* FALLTHROUGH */
4183 	case SIOCGIFMEDIA:
4184 		if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
4185 			error = ifmedia_ioctl(ifp, ifr, &sc->bge_ifmedia,
4186 			    command);
4187 		} else {
4188 			mii = &sc->bge_mii;
4189 			error = ifmedia_ioctl(ifp, ifr, &mii->mii_media,
4190 			    command);
4191 		}
4192 		break;
4193 	default:
4194 		if ((error = ether_ioctl(ifp, command, data)) != ENETRESET)
4195 			break;
4196 
4197 		error = 0;
4198 
4199 		if (command != SIOCADDMULTI && command != SIOCDELMULTI)
4200 			;
4201 		else if (ifp->if_flags & IFF_RUNNING)
4202 			bge_setmulti(sc);
4203 		break;
4204 	}
4205 
4206 	splx(s);
4207 
4208 	return (error);
4209 }
4210 
4211 static void
4212 bge_watchdog(struct ifnet *ifp)
4213 {
4214 	struct bge_softc *sc;
4215 
4216 	sc = ifp->if_softc;
4217 
4218 	aprint_error_dev(sc->bge_dev, "watchdog timeout -- resetting\n");
4219 
4220 	ifp->if_flags &= ~IFF_RUNNING;
4221 	bge_init(ifp);
4222 
4223 	ifp->if_oerrors++;
4224 }
4225 
4226 static void
4227 bge_stop_block(struct bge_softc *sc, bus_addr_t reg, uint32_t bit)
4228 {
4229 	int i;
4230 
4231 	BGE_CLRBIT(sc, reg, bit);
4232 
4233 	for (i = 0; i < BGE_TIMEOUT; i++) {
4234 		if ((CSR_READ_4(sc, reg) & bit) == 0)
4235 			return;
4236 		delay(100);
4237 		if (sc->bge_flags & BGE_PCIE)
4238 		  DELAY(1000);
4239 	}
4240 
4241 	aprint_error_dev(sc->bge_dev,
4242 	    "block failed to stop: reg 0x%lx, bit 0x%08x\n", (u_long)reg, bit);
4243 }
4244 
4245 /*
4246  * Stop the adapter and free any mbufs allocated to the
4247  * RX and TX lists.
4248  */
4249 static void
4250 bge_stop(struct ifnet *ifp, int disable)
4251 {
4252 	struct bge_softc *sc = ifp->if_softc;
4253 
4254 	callout_stop(&sc->bge_timeout);
4255 
4256 	/*
4257 	 * Disable all of the receiver blocks
4258 	 */
4259 	bge_stop_block(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
4260 	bge_stop_block(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
4261 	bge_stop_block(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
4262 	if (!(BGE_IS_5705_OR_BEYOND(sc)))
4263 		bge_stop_block(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
4264 	bge_stop_block(sc, BGE_RDBDI_MODE, BGE_RBDIMODE_ENABLE);
4265 	bge_stop_block(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
4266 	bge_stop_block(sc, BGE_RBDC_MODE, BGE_RBDCMODE_ENABLE);
4267 
4268 	/*
4269 	 * Disable all of the transmit blocks
4270 	 */
4271 	bge_stop_block(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
4272 	bge_stop_block(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
4273 	bge_stop_block(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
4274 	bge_stop_block(sc, BGE_RDMA_MODE, BGE_RDMAMODE_ENABLE);
4275 	bge_stop_block(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE);
4276 	if (!(BGE_IS_5705_OR_BEYOND(sc)))
4277 		bge_stop_block(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
4278 	bge_stop_block(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
4279 
4280 	/*
4281 	 * Shut down all of the memory managers and related
4282 	 * state machines.
4283 	 */
4284 	bge_stop_block(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE);
4285 	bge_stop_block(sc, BGE_WDMA_MODE, BGE_WDMAMODE_ENABLE);
4286 	if (!(BGE_IS_5705_OR_BEYOND(sc)))
4287 		bge_stop_block(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
4288 
4289 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
4290 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
4291 
4292 	if (!(BGE_IS_5705_OR_BEYOND(sc))) {
4293 		bge_stop_block(sc, BGE_BMAN_MODE, BGE_BMANMODE_ENABLE);
4294 		bge_stop_block(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
4295 	}
4296 
4297 	/* Disable host interrupts. */
4298 	BGE_SETBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR);
4299 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
4300 
4301 	/*
4302 	 * Tell firmware we're shutting down.
4303 	 */
4304 	BGE_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
4305 
4306 	/* Free the RX lists. */
4307 	bge_free_rx_ring_std(sc);
4308 
4309 	/* Free jumbo RX list. */
4310 	bge_free_rx_ring_jumbo(sc);
4311 
4312 	/* Free TX buffers. */
4313 	bge_free_tx_ring(sc);
4314 
4315 	/*
4316 	 * Isolate/power down the PHY.
4317 	 */
4318 	if (!(sc->bge_flags & BGE_PHY_FIBER_TBI))
4319 		mii_down(&sc->bge_mii);
4320 
4321 	sc->bge_tx_saved_considx = BGE_TXCONS_UNSET;
4322 
4323 	/* Clear MAC's link state (PHY may still have link UP). */
4324 	BGE_STS_CLRBIT(sc, BGE_STS_LINK);
4325 
4326 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
4327 }
4328 
4329 static void
4330 bge_link_upd(struct bge_softc *sc)
4331 {
4332 	struct ifnet *ifp = &sc->ethercom.ec_if;
4333 	struct mii_data *mii = &sc->bge_mii;
4334 	u_int32_t status;
4335 	int link;
4336 
4337 	/* Clear 'pending link event' flag */
4338 	BGE_STS_CLRBIT(sc, BGE_STS_LINK_EVT);
4339 
4340 	/*
4341 	 * Process link state changes.
4342 	 * Grrr. The link status word in the status block does
4343 	 * not work correctly on the BCM5700 rev AX and BX chips,
4344 	 * according to all available information. Hence, we have
4345 	 * to enable MII interrupts in order to properly obtain
4346 	 * async link changes. Unfortunately, this also means that
4347 	 * we have to read the MAC status register to detect link
4348 	 * changes, thereby adding an additional register access to
4349 	 * the interrupt handler.
4350 	 */
4351 
4352 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700) {
4353 		status = CSR_READ_4(sc, BGE_MAC_STS);
4354 		if (status & BGE_MACSTAT_MI_INTERRUPT) {
4355 			mii_pollstat(mii);
4356 
4357 			if (!BGE_STS_BIT(sc, BGE_STS_LINK) &&
4358 			    mii->mii_media_status & IFM_ACTIVE &&
4359 			    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE)
4360 				BGE_STS_SETBIT(sc, BGE_STS_LINK);
4361 			else if (BGE_STS_BIT(sc, BGE_STS_LINK) &&
4362 			    (!(mii->mii_media_status & IFM_ACTIVE) ||
4363 			    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE))
4364 				BGE_STS_CLRBIT(sc, BGE_STS_LINK);
4365 
4366 			/* Clear the interrupt */
4367 			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
4368 			    BGE_EVTENB_MI_INTERRUPT);
4369 			bge_miibus_readreg(sc->bge_dev, 1, BRGPHY_MII_ISR);
4370 			bge_miibus_writereg(sc->bge_dev, 1, BRGPHY_MII_IMR,
4371 			    BRGPHY_INTRS);
4372 		}
4373 		return;
4374 	}
4375 
4376 	if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
4377 		status = CSR_READ_4(sc, BGE_MAC_STS);
4378 		if (status & BGE_MACSTAT_TBI_PCS_SYNCHED) {
4379 			if (!BGE_STS_BIT(sc, BGE_STS_LINK)) {
4380 				BGE_STS_SETBIT(sc, BGE_STS_LINK);
4381 				if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704)
4382 					BGE_CLRBIT(sc, BGE_MAC_MODE,
4383 					    BGE_MACMODE_TBI_SEND_CFGS);
4384 				CSR_WRITE_4(sc, BGE_MAC_STS, 0xFFFFFFFF);
4385 				if_link_state_change(ifp, LINK_STATE_UP);
4386 			}
4387 		} else if (BGE_STS_BIT(sc, BGE_STS_LINK)) {
4388 			BGE_STS_CLRBIT(sc, BGE_STS_LINK);
4389 			if_link_state_change(ifp, LINK_STATE_DOWN);
4390 		}
4391         /*
4392 	 * Discard link events for MII/GMII cards if MI auto-polling disabled.
4393 	 * This should not happen since mii callouts are locked now, but
4394 	 * we keep this check for debug.
4395 	 */
4396 	} else if (BGE_STS_BIT(sc, BGE_STS_AUTOPOLL)) {
4397 		/*
4398 		 * Some broken BCM chips have BGE_STATFLAG_LINKSTATE_CHANGED
4399 		 * bit in status word always set. Workaround this bug by
4400 		 * reading PHY link status directly.
4401 		 */
4402 		link = (CSR_READ_4(sc, BGE_MI_STS) & BGE_MISTS_LINK)?
4403 		    BGE_STS_LINK : 0;
4404 
4405 		if (BGE_STS_BIT(sc, BGE_STS_LINK) != link) {
4406 			mii_pollstat(mii);
4407 
4408 			if (!BGE_STS_BIT(sc, BGE_STS_LINK) &&
4409 			    mii->mii_media_status & IFM_ACTIVE &&
4410 			    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE)
4411 				BGE_STS_SETBIT(sc, BGE_STS_LINK);
4412 			else if (BGE_STS_BIT(sc, BGE_STS_LINK) &&
4413 			    (!(mii->mii_media_status & IFM_ACTIVE) ||
4414 			    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE))
4415 				BGE_STS_CLRBIT(sc, BGE_STS_LINK);
4416 		}
4417 	}
4418 
4419 	/* Clear the attention */
4420 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
4421 	    BGE_MACSTAT_CFG_CHANGED|BGE_MACSTAT_MI_COMPLETE|
4422 	    BGE_MACSTAT_LINK_CHANGED);
4423 }
4424 
4425 static int
4426 sysctl_bge_verify(SYSCTLFN_ARGS)
4427 {
4428 	int error, t;
4429 	struct sysctlnode node;
4430 
4431 	node = *rnode;
4432 	t = *(int*)rnode->sysctl_data;
4433 	node.sysctl_data = &t;
4434 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
4435 	if (error || newp == NULL)
4436 		return (error);
4437 
4438 #if 0
4439 	DPRINTF2(("%s: t = %d, nodenum = %d, rnodenum = %d\n", __func__, t,
4440 	    node.sysctl_num, rnode->sysctl_num));
4441 #endif
4442 
4443 	if (node.sysctl_num == bge_rxthresh_nodenum) {
4444 		if (t < 0 || t >= NBGE_RX_THRESH)
4445 			return (EINVAL);
4446 		bge_update_all_threshes(t);
4447 	} else
4448 		return (EINVAL);
4449 
4450 	*(int*)rnode->sysctl_data = t;
4451 
4452 	return (0);
4453 }
4454 
4455 /*
4456  * Set up sysctl(3) MIB, hw.bge.*.
4457  *
4458  * TBD condition SYSCTL_PERMANENT on being an LKM or not
4459  */
4460 SYSCTL_SETUP(sysctl_bge, "sysctl bge subtree setup")
4461 {
4462 	int rc, bge_root_num;
4463 	const struct sysctlnode *node;
4464 
4465 	if ((rc = sysctl_createv(clog, 0, NULL, NULL,
4466 	    CTLFLAG_PERMANENT, CTLTYPE_NODE, "hw", NULL,
4467 	    NULL, 0, NULL, 0, CTL_HW, CTL_EOL)) != 0) {
4468 		goto err;
4469 	}
4470 
4471 	if ((rc = sysctl_createv(clog, 0, NULL, &node,
4472 	    CTLFLAG_PERMANENT, CTLTYPE_NODE, "bge",
4473 	    SYSCTL_DESCR("BGE interface controls"),
4474 	    NULL, 0, NULL, 0, CTL_HW, CTL_CREATE, CTL_EOL)) != 0) {
4475 		goto err;
4476 	}
4477 
4478 	bge_root_num = node->sysctl_num;
4479 
4480 	/* BGE Rx interrupt mitigation level */
4481 	if ((rc = sysctl_createv(clog, 0, NULL, &node,
4482 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
4483 	    CTLTYPE_INT, "rx_lvl",
4484 	    SYSCTL_DESCR("BGE receive interrupt mitigation level"),
4485 	    sysctl_bge_verify, 0,
4486 	    &bge_rx_thresh_lvl,
4487 	    0, CTL_HW, bge_root_num, CTL_CREATE,
4488 	    CTL_EOL)) != 0) {
4489 		goto err;
4490 	}
4491 
4492 	bge_rxthresh_nodenum = node->sysctl_num;
4493 
4494 	return;
4495 
4496 err:
4497 	aprint_error("%s: sysctl_createv failed (rc = %d)\n", __func__, rc);
4498 }
4499 
4500 static int
4501 bge_get_eaddr_mem(struct bge_softc *sc, u_int8_t ether_addr[])
4502 {
4503 	u_int32_t mac_addr;
4504 
4505 	mac_addr = bge_readmem_ind(sc, 0x0c14);
4506 	if ((mac_addr >> 16) == 0x484b) {
4507 		ether_addr[0] = (uint8_t)(mac_addr >> 8);
4508 		ether_addr[1] = (uint8_t)mac_addr;
4509 		mac_addr = bge_readmem_ind(sc, 0x0c18);
4510 		ether_addr[2] = (uint8_t)(mac_addr >> 24);
4511 		ether_addr[3] = (uint8_t)(mac_addr >> 16);
4512 		ether_addr[4] = (uint8_t)(mac_addr >> 8);
4513 		ether_addr[5] = (uint8_t)mac_addr;
4514 		return (0);
4515 	}
4516 	return (1);
4517 }
4518 
4519 static int
4520 bge_get_eaddr_nvram(struct bge_softc *sc, u_int8_t ether_addr[])
4521 {
4522 	int mac_offset = BGE_EE_MAC_OFFSET;
4523 
4524 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
4525 		mac_offset = BGE_EE_MAC_OFFSET_5906;
4526 	}
4527 
4528 	return (bge_read_nvram(sc, ether_addr, mac_offset + 2,
4529 	    ETHER_ADDR_LEN));
4530 }
4531 
4532 static int
4533 bge_get_eaddr_eeprom(struct bge_softc *sc, u_int8_t ether_addr[])
4534 {
4535 
4536 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
4537 		return (1);
4538 	}
4539 
4540 	return (bge_read_eeprom(sc, ether_addr, BGE_EE_MAC_OFFSET + 2,
4541 	   ETHER_ADDR_LEN));
4542 }
4543 
4544 static int
4545 bge_get_eaddr(struct bge_softc *sc, u_int8_t eaddr[])
4546 {
4547 	static const bge_eaddr_fcn_t bge_eaddr_funcs[] = {
4548 		/* NOTE: Order is critical */
4549 		bge_get_eaddr_mem,
4550 		bge_get_eaddr_nvram,
4551 		bge_get_eaddr_eeprom,
4552 		NULL
4553 	};
4554 	const bge_eaddr_fcn_t *func;
4555 
4556 	for (func = bge_eaddr_funcs; *func != NULL; ++func) {
4557 		if ((*func)(sc, eaddr) == 0)
4558 			break;
4559 	}
4560 	return (*func == NULL ? ENXIO : 0);
4561 }
4562