xref: /netbsd-src/sys/dev/pci/if_bge.c (revision da9817918ec7e88db2912a2882967c7570a83f47)
1 /*	$NetBSD: if_bge.c,v 1.165 2009/05/05 18:08:28 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 1.165 2009/05/05 18:08:28 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 #include <dev/pci/if_bgevar.h>
137 
138 #include <uvm/uvm_extern.h>
139 #include <prop/proplib.h>
140 
141 #define ETHER_MIN_NOPAD (ETHER_MIN_LEN - ETHER_CRC_LEN) /* i.e., 60 */
142 
143 
144 /*
145  * Tunable thresholds for rx-side bge interrupt mitigation.
146  */
147 
148 /*
149  * The pairs of values below were obtained from empirical measurement
150  * on bcm5700 rev B2; they ar designed to give roughly 1 receive
151  * interrupt for every N packets received, where N is, approximately,
152  * the second value (rx_max_bds) in each pair.  The values are chosen
153  * such that moving from one pair to the succeeding pair was observed
154  * to roughly halve interrupt rate under sustained input packet load.
155  * The values were empirically chosen to avoid overflowing internal
156  * limits on the  bcm5700: inreasing rx_ticks much beyond 600
157  * results in internal wrapping and higher interrupt rates.
158  * The limit of 46 frames was chosen to match NFS workloads.
159  *
160  * These values also work well on bcm5701, bcm5704C, and (less
161  * tested) bcm5703.  On other chipsets, (including the Altima chip
162  * family), the larger values may overflow internal chip limits,
163  * leading to increasing interrupt rates rather than lower interrupt
164  * rates.
165  *
166  * Applications using heavy interrupt mitigation (interrupting every
167  * 32 or 46 frames) in both directions may need to increase the TCP
168  * windowsize to above 131072 bytes (e.g., to 199608 bytes) to sustain
169  * full link bandwidth, due to ACKs and window updates lingering
170  * in the RX queue during the 30-to-40-frame interrupt-mitigation window.
171  */
172 static const struct bge_load_rx_thresh {
173 	int rx_ticks;
174 	int rx_max_bds; }
175 bge_rx_threshes[] = {
176 	{ 32,   2 },
177 	{ 50,   4 },
178 	{ 100,  8 },
179 	{ 192, 16 },
180 	{ 416, 32 },
181 	{ 598, 46 }
182 };
183 #define NBGE_RX_THRESH (sizeof(bge_rx_threshes) / sizeof(bge_rx_threshes[0]))
184 
185 /* XXX patchable; should be sysctl'able */
186 static int	bge_auto_thresh = 1;
187 static int	bge_rx_thresh_lvl;
188 
189 static int	bge_rxthresh_nodenum;
190 
191 typedef int (*bge_eaddr_fcn_t)(struct bge_softc *, u_int8_t[]);
192 
193 static int	bge_probe(device_t, cfdata_t, void *);
194 static void	bge_attach(device_t, device_t, void *);
195 static void	bge_release_resources(struct bge_softc *);
196 static void	bge_txeof(struct bge_softc *);
197 static void	bge_rxeof(struct bge_softc *);
198 
199 static int 	bge_get_eaddr_mem(struct bge_softc *, u_int8_t[]);
200 static int 	bge_get_eaddr_nvram(struct bge_softc *, u_int8_t[]);
201 static int 	bge_get_eaddr_eeprom(struct bge_softc *, u_int8_t[]);
202 static int 	bge_get_eaddr(struct bge_softc *, u_int8_t[]);
203 
204 static void	bge_tick(void *);
205 static void	bge_stats_update(struct bge_softc *);
206 static int	bge_encap(struct bge_softc *, struct mbuf *, u_int32_t *);
207 
208 static int	bge_intr(void *);
209 static void	bge_start(struct ifnet *);
210 static int	bge_ioctl(struct ifnet *, u_long, void *);
211 static int	bge_init(struct ifnet *);
212 static void	bge_stop(struct ifnet *, int);
213 static void	bge_watchdog(struct ifnet *);
214 static int	bge_ifmedia_upd(struct ifnet *);
215 static void	bge_ifmedia_sts(struct ifnet *, struct ifmediareq *);
216 
217 static void	bge_setmulti(struct bge_softc *);
218 
219 static void	bge_handle_events(struct bge_softc *);
220 static int	bge_alloc_jumbo_mem(struct bge_softc *);
221 #if 0 /* XXX */
222 static void	bge_free_jumbo_mem(struct bge_softc *);
223 #endif
224 static void	*bge_jalloc(struct bge_softc *);
225 static void	bge_jfree(struct mbuf *, void *, size_t, void *);
226 static int	bge_newbuf_std(struct bge_softc *, int, struct mbuf *,
227 			       bus_dmamap_t);
228 static int	bge_newbuf_jumbo(struct bge_softc *, int, struct mbuf *);
229 static int	bge_init_rx_ring_std(struct bge_softc *);
230 static void	bge_free_rx_ring_std(struct bge_softc *);
231 static int	bge_init_rx_ring_jumbo(struct bge_softc *);
232 static void	bge_free_rx_ring_jumbo(struct bge_softc *);
233 static void	bge_free_tx_ring(struct bge_softc *);
234 static int	bge_init_tx_ring(struct bge_softc *);
235 
236 static int	bge_chipinit(struct bge_softc *);
237 static int	bge_blockinit(struct bge_softc *);
238 static int	bge_setpowerstate(struct bge_softc *, int);
239 
240 static void	bge_reset(struct bge_softc *);
241 static void	bge_link_upd(struct bge_softc *);
242 
243 #define BGE_DEBUG
244 #ifdef BGE_DEBUG
245 #define DPRINTF(x)	if (bgedebug) printf x
246 #define DPRINTFN(n,x)	if (bgedebug >= (n)) printf x
247 #define BGE_TSO_PRINTF(x)  do { if (bge_tso_debug) printf x ;} while (0)
248 int	bgedebug = 0;
249 int	bge_tso_debug = 0;
250 #else
251 #define DPRINTF(x)
252 #define DPRINTFN(n,x)
253 #define BGE_TSO_PRINTF(x)
254 #endif
255 
256 #ifdef BGE_EVENT_COUNTERS
257 #define	BGE_EVCNT_INCR(ev)	(ev).ev_count++
258 #define	BGE_EVCNT_ADD(ev, val)	(ev).ev_count += (val)
259 #define	BGE_EVCNT_UPD(ev, val)	(ev).ev_count = (val)
260 #else
261 #define	BGE_EVCNT_INCR(ev)	/* nothing */
262 #define	BGE_EVCNT_ADD(ev, val)	/* nothing */
263 #define	BGE_EVCNT_UPD(ev, val)	/* nothing */
264 #endif
265 
266 static const struct bge_product {
267 	pci_vendor_id_t		bp_vendor;
268 	pci_product_id_t	bp_product;
269 	const char		*bp_name;
270 } bge_products[] = {
271 	/*
272 	 * The BCM5700 documentation seems to indicate that the hardware
273 	 * still has the Alteon vendor ID burned into it, though it
274 	 * should always be overridden by the value in the EEPROM.  We'll
275 	 * check for it anyway.
276 	 */
277 	{ PCI_VENDOR_ALTEON,
278 	  PCI_PRODUCT_ALTEON_BCM5700,
279 	  "Broadcom BCM5700 Gigabit Ethernet",
280 	  },
281 	{ PCI_VENDOR_ALTEON,
282 	  PCI_PRODUCT_ALTEON_BCM5701,
283 	  "Broadcom BCM5701 Gigabit Ethernet",
284 	  },
285 	{ PCI_VENDOR_ALTIMA,
286 	  PCI_PRODUCT_ALTIMA_AC1000,
287 	  "Altima AC1000 Gigabit Ethernet",
288 	  },
289 	{ PCI_VENDOR_ALTIMA,
290 	  PCI_PRODUCT_ALTIMA_AC1001,
291 	  "Altima AC1001 Gigabit Ethernet",
292 	   },
293 	{ PCI_VENDOR_ALTIMA,
294 	  PCI_PRODUCT_ALTIMA_AC9100,
295 	  "Altima AC9100 Gigabit Ethernet",
296 	  },
297 	{ PCI_VENDOR_BROADCOM,
298 	  PCI_PRODUCT_BROADCOM_BCM5700,
299 	  "Broadcom BCM5700 Gigabit Ethernet",
300 	  },
301 	{ PCI_VENDOR_BROADCOM,
302 	  PCI_PRODUCT_BROADCOM_BCM5701,
303 	  "Broadcom BCM5701 Gigabit Ethernet",
304 	  },
305 	{ PCI_VENDOR_BROADCOM,
306 	  PCI_PRODUCT_BROADCOM_BCM5702,
307 	  "Broadcom BCM5702 Gigabit Ethernet",
308 	  },
309 	{ PCI_VENDOR_BROADCOM,
310 	  PCI_PRODUCT_BROADCOM_BCM5702X,
311 	  "Broadcom BCM5702X Gigabit Ethernet" },
312 	{ PCI_VENDOR_BROADCOM,
313 	  PCI_PRODUCT_BROADCOM_BCM5703,
314 	  "Broadcom BCM5703 Gigabit Ethernet",
315 	  },
316 	{ PCI_VENDOR_BROADCOM,
317 	  PCI_PRODUCT_BROADCOM_BCM5703X,
318 	  "Broadcom BCM5703X Gigabit Ethernet",
319 	  },
320 	{ PCI_VENDOR_BROADCOM,
321 	  PCI_PRODUCT_BROADCOM_BCM5703_ALT,
322 	  "Broadcom BCM5703 Gigabit Ethernet",
323 	  },
324    	{ PCI_VENDOR_BROADCOM,
325 	  PCI_PRODUCT_BROADCOM_BCM5704C,
326 	  "Broadcom BCM5704C Dual Gigabit Ethernet",
327 	  },
328    	{ PCI_VENDOR_BROADCOM,
329 	  PCI_PRODUCT_BROADCOM_BCM5704S,
330 	  "Broadcom BCM5704S Dual Gigabit Ethernet",
331 	  },
332    	{ PCI_VENDOR_BROADCOM,
333 	  PCI_PRODUCT_BROADCOM_BCM5705,
334 	  "Broadcom BCM5705 Gigabit Ethernet",
335 	  },
336    	{ PCI_VENDOR_BROADCOM,
337 	  PCI_PRODUCT_BROADCOM_BCM5705K,
338 	  "Broadcom BCM5705K Gigabit Ethernet",
339 	  },
340    	{ PCI_VENDOR_BROADCOM,
341 	  PCI_PRODUCT_BROADCOM_BCM5705M,
342 	  "Broadcom BCM5705M Gigabit Ethernet",
343 	  },
344    	{ PCI_VENDOR_BROADCOM,
345 	  PCI_PRODUCT_BROADCOM_BCM5705M_ALT,
346 	  "Broadcom BCM5705M Gigabit Ethernet",
347 	  },
348 	{ PCI_VENDOR_BROADCOM,
349 	  PCI_PRODUCT_BROADCOM_BCM5714,
350 	  "Broadcom BCM5714/5715 Gigabit Ethernet",
351 	  },
352 	{ PCI_VENDOR_BROADCOM,
353 	  PCI_PRODUCT_BROADCOM_BCM5715,
354 	  "Broadcom BCM5714/5715 Gigabit Ethernet",
355 	  },
356 	{ PCI_VENDOR_BROADCOM,
357 	  PCI_PRODUCT_BROADCOM_BCM5789,
358 	  "Broadcom BCM5789 Gigabit Ethernet",
359 	  },
360 	{ PCI_VENDOR_BROADCOM,
361 	  PCI_PRODUCT_BROADCOM_BCM5721,
362 	  "Broadcom BCM5721 Gigabit Ethernet",
363 	  },
364 	{ PCI_VENDOR_BROADCOM,
365 	  PCI_PRODUCT_BROADCOM_BCM5722,
366 	  "Broadcom BCM5722 Gigabit Ethernet",
367 	  },
368 	{ PCI_VENDOR_BROADCOM,
369 	  PCI_PRODUCT_BROADCOM_BCM5750,
370 	  "Broadcom BCM5750 Gigabit Ethernet",
371 	  },
372 	{ PCI_VENDOR_BROADCOM,
373 	  PCI_PRODUCT_BROADCOM_BCM5750M,
374 	  "Broadcom BCM5750M Gigabit Ethernet",
375 	  },
376 	{ PCI_VENDOR_BROADCOM,
377 	  PCI_PRODUCT_BROADCOM_BCM5751,
378 	  "Broadcom BCM5751 Gigabit Ethernet",
379 	  },
380 	{ PCI_VENDOR_BROADCOM,
381 	  PCI_PRODUCT_BROADCOM_BCM5751M,
382 	  "Broadcom BCM5751M Gigabit Ethernet",
383 	  },
384 	{ PCI_VENDOR_BROADCOM,
385 	  PCI_PRODUCT_BROADCOM_BCM5752,
386 	  "Broadcom BCM5752 Gigabit Ethernet",
387 	  },
388 	{ PCI_VENDOR_BROADCOM,
389 	  PCI_PRODUCT_BROADCOM_BCM5752M,
390 	  "Broadcom BCM5752M Gigabit Ethernet",
391 	  },
392 	{ PCI_VENDOR_BROADCOM,
393 	  PCI_PRODUCT_BROADCOM_BCM5753,
394 	  "Broadcom BCM5753 Gigabit Ethernet",
395 	  },
396 	{ PCI_VENDOR_BROADCOM,
397 	  PCI_PRODUCT_BROADCOM_BCM5753M,
398 	  "Broadcom BCM5753M Gigabit Ethernet",
399 	  },
400 	{ PCI_VENDOR_BROADCOM,
401 	  PCI_PRODUCT_BROADCOM_BCM5754,
402 	  "Broadcom BCM5754 Gigabit Ethernet",
403 	},
404 	{ PCI_VENDOR_BROADCOM,
405 	  PCI_PRODUCT_BROADCOM_BCM5754M,
406 	  "Broadcom BCM5754M Gigabit Ethernet",
407 	},
408 	{ PCI_VENDOR_BROADCOM,
409 	  PCI_PRODUCT_BROADCOM_BCM5755,
410 	  "Broadcom BCM5755 Gigabit Ethernet",
411 	},
412 	{ PCI_VENDOR_BROADCOM,
413 	  PCI_PRODUCT_BROADCOM_BCM5755M,
414 	  "Broadcom BCM5755M Gigabit Ethernet",
415 	},
416    	{ PCI_VENDOR_BROADCOM,
417 	  PCI_PRODUCT_BROADCOM_BCM5780,
418 	  "Broadcom BCM5780 Gigabit Ethernet",
419 	  },
420    	{ PCI_VENDOR_BROADCOM,
421 	  PCI_PRODUCT_BROADCOM_BCM5780S,
422 	  "Broadcom BCM5780S Gigabit Ethernet",
423 	  },
424    	{ PCI_VENDOR_BROADCOM,
425 	  PCI_PRODUCT_BROADCOM_BCM5782,
426 	  "Broadcom BCM5782 Gigabit Ethernet",
427 	},
428 	{ PCI_VENDOR_BROADCOM,
429 	  PCI_PRODUCT_BROADCOM_BCM5786,
430 	  "Broadcom BCM5786 Gigabit Ethernet",
431 	},
432 	{ PCI_VENDOR_BROADCOM,
433 	  PCI_PRODUCT_BROADCOM_BCM5787,
434 	  "Broadcom BCM5787 Gigabit Ethernet",
435 	},
436 	{ PCI_VENDOR_BROADCOM,
437 	  PCI_PRODUCT_BROADCOM_BCM5787M,
438 	  "Broadcom BCM5787M Gigabit Ethernet",
439 	},
440    	{ PCI_VENDOR_BROADCOM,
441 	  PCI_PRODUCT_BROADCOM_BCM5788,
442 	  "Broadcom BCM5788 Gigabit Ethernet",
443 	  },
444    	{ PCI_VENDOR_BROADCOM,
445 	  PCI_PRODUCT_BROADCOM_BCM5789,
446 	  "Broadcom BCM5789 Gigabit Ethernet",
447 	  },
448    	{ PCI_VENDOR_BROADCOM,
449 	  PCI_PRODUCT_BROADCOM_BCM5901,
450 	  "Broadcom BCM5901 Fast Ethernet",
451 	  },
452    	{ PCI_VENDOR_BROADCOM,
453 	  PCI_PRODUCT_BROADCOM_BCM5901A2,
454 	  "Broadcom BCM5901A2 Fast Ethernet",
455 	  },
456 	{ PCI_VENDOR_SCHNEIDERKOCH,
457 	  PCI_PRODUCT_SCHNEIDERKOCH_SK_9DX1,
458 	  "SysKonnect SK-9Dx1 Gigabit Ethernet",
459 	  },
460 	{ PCI_VENDOR_3COM,
461 	  PCI_PRODUCT_3COM_3C996,
462 	  "3Com 3c996 Gigabit Ethernet",
463 	  },
464 	{ PCI_VENDOR_BROADCOM,
465 	  PCI_PRODUCT_BROADCOM_BCM5906,
466 	  "Broadcom BCM5906 Fast Ethernet",
467 	  },
468 	{ PCI_VENDOR_BROADCOM,
469 	  PCI_PRODUCT_BROADCOM_BCM5906M,
470 	  "Broadcom BCM5906M Fast Ethernet",
471 	  },
472 	{ 0,
473 	  0,
474 	  NULL },
475 };
476 
477 /*
478  * XXX: how to handle variants based on 5750 and derivatives:
479  * 5750 5751, 5721, possibly 5714, 5752, and 5708?, which
480  * in general behave like a 5705, except with additional quirks.
481  * This driver's current handling of the 5721 is wrong;
482  * how we map ASIC revision to "quirks" needs more thought.
483  * (defined here until the thought is done).
484  */
485 #define BGE_IS_5714_FAMILY(sc) \
486 	(BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5714_A0 || \
487 	 BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5780 ||	\
488 	 BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5714 )
489 
490 #define BGE_IS_5750_OR_BEYOND(sc)  \
491 	(BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5750 || \
492 	 BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5752 || \
493 	 BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 || \
494 	 BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5787 || \
495 	 BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906 || \
496 	 BGE_IS_5714_FAMILY(sc) )
497 
498 #define BGE_IS_5705_OR_BEYOND(sc)  \
499 	(BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5705 || \
500 	    (BGE_IS_5750_OR_BEYOND(sc)))
501 
502 static const struct bge_revision {
503 	uint32_t		br_chipid;
504 	const char		*br_name;
505 } bge_revisions[] = {
506 	{ BGE_CHIPID_BCM5700_A0, "BCM5700 A0" },
507 	{ BGE_CHIPID_BCM5700_A1, "BCM5700 A1" },
508 	{ BGE_CHIPID_BCM5700_B0, "BCM5700 B0" },
509 	{ BGE_CHIPID_BCM5700_B1, "BCM5700 B1" },
510 	{ BGE_CHIPID_BCM5700_B2, "BCM5700 B2" },
511 	{ BGE_CHIPID_BCM5700_B3, "BCM5700 B3" },
512 	/* This is treated like a BCM5700 Bx */
513 	{ BGE_CHIPID_BCM5700_ALTIMA, "BCM5700 Altima" },
514 	{ BGE_CHIPID_BCM5700_C0, "BCM5700 C0" },
515 	{ BGE_CHIPID_BCM5701_A0, "BCM5701 A0" },
516 	{ BGE_CHIPID_BCM5701_B0, "BCM5701 B0" },
517 	{ BGE_CHIPID_BCM5701_B2, "BCM5701 B2" },
518 	{ BGE_CHIPID_BCM5701_B5, "BCM5701 B5" },
519 	{ BGE_CHIPID_BCM5703_A0, "BCM5703 A0" },
520 	{ BGE_CHIPID_BCM5703_A1, "BCM5703 A1" },
521 	{ BGE_CHIPID_BCM5703_A2, "BCM5703 A2" },
522 	{ BGE_CHIPID_BCM5703_A3, "BCM5703 A3" },
523 	{ BGE_CHIPID_BCM5703_B0, "BCM5703 B0" },
524 	{ BGE_CHIPID_BCM5704_A0, "BCM5704 A0" },
525 	{ BGE_CHIPID_BCM5704_A1, "BCM5704 A1" },
526 	{ BGE_CHIPID_BCM5704_A2, "BCM5704 A2" },
527 	{ BGE_CHIPID_BCM5704_A3, "BCM5704 A3" },
528 	{ BGE_CHIPID_BCM5704_B0, "BCM5704 B0" },
529 	{ BGE_CHIPID_BCM5705_A0, "BCM5705 A0" },
530 	{ BGE_CHIPID_BCM5705_A1, "BCM5705 A1" },
531 	{ BGE_CHIPID_BCM5705_A2, "BCM5705 A2" },
532 	{ BGE_CHIPID_BCM5705_A3, "BCM5705 A3" },
533 	{ BGE_CHIPID_BCM5750_A0, "BCM5750 A0" },
534 	{ BGE_CHIPID_BCM5750_A1, "BCM5750 A1" },
535 	{ BGE_CHIPID_BCM5750_A3, "BCM5750 A3" },
536 	{ BGE_CHIPID_BCM5750_B0, "BCM5750 B0" },
537 	{ BGE_CHIPID_BCM5750_B1, "BCM5750 B1" },
538 	{ BGE_CHIPID_BCM5750_C0, "BCM5750 C0" },
539 	{ BGE_CHIPID_BCM5750_C1, "BCM5750 C1" },
540 	{ BGE_CHIPID_BCM5750_C2, "BCM5750 C2" },
541 	{ BGE_CHIPID_BCM5752_A0, "BCM5752 A0" },
542 	{ BGE_CHIPID_BCM5752_A1, "BCM5752 A1" },
543 	{ BGE_CHIPID_BCM5752_A2, "BCM5752 A2" },
544 	{ BGE_CHIPID_BCM5714_A0, "BCM5714 A0" },
545 	{ BGE_CHIPID_BCM5714_B0, "BCM5714 B0" },
546 	{ BGE_CHIPID_BCM5714_B3, "BCM5714 B3" },
547 	{ BGE_CHIPID_BCM5715_A0, "BCM5715 A0" },
548 	{ BGE_CHIPID_BCM5715_A1, "BCM5715 A1" },
549 	{ BGE_CHIPID_BCM5715_A3, "BCM5715 A3" },
550 	{ BGE_CHIPID_BCM5755_A0, "BCM5755 A0" },
551 	{ BGE_CHIPID_BCM5755_A1, "BCM5755 A1" },
552 	{ BGE_CHIPID_BCM5755_A2, "BCM5755 A2" },
553 	{ BGE_CHIPID_BCM5755_C0, "BCM5755 C0" },
554 	{ BGE_CHIPID_BCM5787_A0, "BCM5754/5787 A0" },
555 	{ BGE_CHIPID_BCM5787_A1, "BCM5754/5787 A1" },
556 	{ BGE_CHIPID_BCM5787_A2, "BCM5754/5787 A2" },
557 	{ BGE_CHIPID_BCM5906_A1, "BCM5906 A1" },
558 	{ BGE_CHIPID_BCM5906_A2, "BCM5906 A2" },
559 	{ 0, NULL }
560 };
561 
562 /*
563  * Some defaults for major revisions, so that newer steppings
564  * that we don't know about have a shot at working.
565  */
566 static const struct bge_revision bge_majorrevs[] = {
567 	{ BGE_ASICREV_BCM5700, "unknown BCM5700" },
568 	{ BGE_ASICREV_BCM5701, "unknown BCM5701" },
569 	{ BGE_ASICREV_BCM5703, "unknown BCM5703" },
570 	{ BGE_ASICREV_BCM5704, "unknown BCM5704" },
571 	{ BGE_ASICREV_BCM5705, "unknown BCM5705" },
572 	{ BGE_ASICREV_BCM5750, "unknown BCM5750" },
573 	{ BGE_ASICREV_BCM5714_A0, "unknown BCM5714" },
574 	{ BGE_ASICREV_BCM5714, "unknown BCM5714" },
575 	{ BGE_ASICREV_BCM5752, "unknown BCM5752 family" },
576 	{ BGE_ASICREV_BCM5755, "unknown BCM5755" },
577 	{ BGE_ASICREV_BCM5780, "unknown BCM5780" },
578 	/* 5754 and 5787 share the same ASIC ID */
579 	{ BGE_ASICREV_BCM5787, "unknown BCM5787/5787" },
580 	{ BGE_ASICREV_BCM5906, "unknown BCM5906" },
581 	{ 0, NULL }
582 };
583 
584 CFATTACH_DECL_NEW(bge, sizeof(struct bge_softc),
585     bge_probe, bge_attach, NULL, NULL);
586 
587 static u_int32_t
588 bge_readmem_ind(struct bge_softc *sc, int off)
589 {
590 	pcireg_t val;
591 
592 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, off);
593 	val = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_DATA);
594 	return val;
595 }
596 
597 static void
598 bge_writemem_ind(struct bge_softc *sc, int off, int val)
599 {
600 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, off);
601 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_DATA, val);
602 }
603 
604 #ifdef notdef
605 static u_int32_t
606 bge_readreg_ind(struct bge_softc *sc, int off)
607 {
608 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_BASEADDR, off);
609 	return (pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_DATA));
610 }
611 #endif
612 
613 static void
614 bge_writereg_ind(struct bge_softc *sc, int off, int val)
615 {
616 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_BASEADDR, off);
617 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_DATA, val);
618 }
619 
620 static void
621 bge_writemem_direct(struct bge_softc *sc, int off, int val)
622 {
623 	CSR_WRITE_4(sc, off, val);
624 }
625 
626 static void
627 bge_writembx(struct bge_softc *sc, int off, int val)
628 {
629 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906)
630 		off += BGE_LPMBX_IRQ0_HI - BGE_MBX_IRQ0_HI;
631 
632 	CSR_WRITE_4(sc, off, val);
633 }
634 
635 static u_int8_t
636 bge_nvram_getbyte(struct bge_softc *sc, int addr, u_int8_t *dest)
637 {
638 	u_int32_t access, byte = 0;
639 	int i;
640 
641 	/* Lock. */
642 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_SET1);
643 	for (i = 0; i < 8000; i++) {
644 		if (CSR_READ_4(sc, BGE_NVRAM_SWARB) & BGE_NVRAMSWARB_GNT1)
645 			break;
646 		DELAY(20);
647 	}
648 	if (i == 8000)
649 		return (1);
650 
651 	/* Enable access. */
652 	access = CSR_READ_4(sc, BGE_NVRAM_ACCESS);
653 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access | BGE_NVRAMACC_ENABLE);
654 
655 	CSR_WRITE_4(sc, BGE_NVRAM_ADDR, addr & 0xfffffffc);
656 	CSR_WRITE_4(sc, BGE_NVRAM_CMD, BGE_NVRAM_READCMD);
657 	for (i = 0; i < BGE_TIMEOUT * 10; i++) {
658 		DELAY(10);
659 		if (CSR_READ_4(sc, BGE_NVRAM_CMD) & BGE_NVRAMCMD_DONE) {
660 			DELAY(10);
661 			break;
662 		}
663 	}
664 
665 	if (i == BGE_TIMEOUT * 10) {
666 		aprint_error_dev(sc->bge_dev, "nvram read timed out\n");
667 		return (1);
668 	}
669 
670 	/* Get result. */
671 	byte = CSR_READ_4(sc, BGE_NVRAM_RDDATA);
672 
673 	*dest = (bswap32(byte) >> ((addr % 4) * 8)) & 0xFF;
674 
675 	/* Disable access. */
676 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access);
677 
678 	/* Unlock. */
679 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_CLR1);
680 	CSR_READ_4(sc, BGE_NVRAM_SWARB);
681 
682 	return (0);
683 }
684 
685 /*
686  * Read a sequence of bytes from NVRAM.
687  */
688 static int
689 bge_read_nvram(struct bge_softc *sc, u_int8_t *dest, int off, int cnt)
690 {
691 	int err = 0, i;
692 	u_int8_t byte = 0;
693 
694 	if (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5906)
695 		return (1);
696 
697 	for (i = 0; i < cnt; i++) {
698 		err = bge_nvram_getbyte(sc, off + i, &byte);
699 		if (err)
700 			break;
701 		*(dest + i) = byte;
702 	}
703 
704 	return (err ? 1 : 0);
705 }
706 
707 /*
708  * Read a byte of data stored in the EEPROM at address 'addr.' The
709  * BCM570x supports both the traditional bitbang interface and an
710  * auto access interface for reading the EEPROM. We use the auto
711  * access method.
712  */
713 static u_int8_t
714 bge_eeprom_getbyte(struct bge_softc *sc, int addr, u_int8_t *dest)
715 {
716 	int i;
717 	u_int32_t byte = 0;
718 
719 	/*
720 	 * Enable use of auto EEPROM access so we can avoid
721 	 * having to use the bitbang method.
722 	 */
723 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_AUTO_EEPROM);
724 
725 	/* Reset the EEPROM, load the clock period. */
726 	CSR_WRITE_4(sc, BGE_EE_ADDR,
727 	    BGE_EEADDR_RESET | BGE_EEHALFCLK(BGE_HALFCLK_384SCL));
728 	DELAY(20);
729 
730 	/* Issue the read EEPROM command. */
731 	CSR_WRITE_4(sc, BGE_EE_ADDR, BGE_EE_READCMD | addr);
732 
733 	/* Wait for completion */
734 	for(i = 0; i < BGE_TIMEOUT * 10; i++) {
735 		DELAY(10);
736 		if (CSR_READ_4(sc, BGE_EE_ADDR) & BGE_EEADDR_DONE)
737 			break;
738 	}
739 
740 	if (i == BGE_TIMEOUT) {
741 		aprint_error_dev(sc->bge_dev, "eeprom read timed out\n");
742 		return (0);
743 	}
744 
745 	/* Get result. */
746 	byte = CSR_READ_4(sc, BGE_EE_DATA);
747 
748 	*dest = (byte >> ((addr % 4) * 8)) & 0xFF;
749 
750 	return (0);
751 }
752 
753 /*
754  * Read a sequence of bytes from the EEPROM.
755  */
756 static int
757 bge_read_eeprom(struct bge_softc *sc, void *destv, int off, int cnt)
758 {
759 	int err = 0, i;
760 	u_int8_t byte = 0;
761 	char *dest = destv;
762 
763 	for (i = 0; i < cnt; i++) {
764 		err = bge_eeprom_getbyte(sc, off + i, &byte);
765 		if (err)
766 			break;
767 		*(dest + i) = byte;
768 	}
769 
770 	return (err ? 1 : 0);
771 }
772 
773 static int
774 bge_miibus_readreg(device_t dev, int phy, int reg)
775 {
776 	struct bge_softc *sc = device_private(dev);
777 	u_int32_t val;
778 	u_int32_t saved_autopoll;
779 	int i;
780 
781 	/*
782 	 * Broadcom's own driver always assumes the internal
783 	 * PHY is at GMII address 1. On some chips, the PHY responds
784 	 * to accesses at all addresses, which could cause us to
785 	 * bogusly attach the PHY 32 times at probe type. Always
786 	 * restricting the lookup to address 1 is simpler than
787 	 * trying to figure out which chips revisions should be
788 	 * special-cased.
789 	 */
790 	if (phy != 1)
791 		return (0);
792 
793 	/* Reading with autopolling on may trigger PCI errors */
794 	saved_autopoll = CSR_READ_4(sc, BGE_MI_MODE);
795 	if (saved_autopoll & BGE_MIMODE_AUTOPOLL) {
796 		BGE_STS_CLRBIT(sc, BGE_STS_AUTOPOLL);
797 		CSR_WRITE_4(sc, BGE_MI_MODE,
798 		    saved_autopoll &~ BGE_MIMODE_AUTOPOLL);
799 		DELAY(40);
800 	}
801 
802 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_READ|BGE_MICOMM_BUSY|
803 	    BGE_MIPHY(phy)|BGE_MIREG(reg));
804 
805 	for (i = 0; i < BGE_TIMEOUT; i++) {
806 		val = CSR_READ_4(sc, BGE_MI_COMM);
807 		if (!(val & BGE_MICOMM_BUSY))
808 			break;
809 		delay(10);
810 	}
811 
812 	if (i == BGE_TIMEOUT) {
813 		aprint_error_dev(sc->bge_dev, "PHY read timed out\n");
814 		val = 0;
815 		goto done;
816 	}
817 
818 	val = CSR_READ_4(sc, BGE_MI_COMM);
819 
820 done:
821 	if (saved_autopoll & BGE_MIMODE_AUTOPOLL) {
822 		BGE_STS_SETBIT(sc, BGE_STS_AUTOPOLL);
823 		CSR_WRITE_4(sc, BGE_MI_MODE, saved_autopoll);
824 		DELAY(40);
825 	}
826 
827 	if (val & BGE_MICOMM_READFAIL)
828 		return (0);
829 
830 	return (val & 0xFFFF);
831 }
832 
833 static void
834 bge_miibus_writereg(device_t dev, int phy, int reg, int val)
835 {
836 	struct bge_softc *sc = device_private(dev);
837 	u_int32_t saved_autopoll;
838 	int i;
839 
840 	if (phy!=1) {
841 		return;
842 	}
843 
844 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906 &&
845 	    (reg == BRGPHY_MII_1000CTL || reg == BRGPHY_MII_AUXCTL)) {
846 		return;
847 	}
848 
849 	/* Reading with autopolling on may trigger PCI errors */
850 	saved_autopoll = CSR_READ_4(sc, BGE_MI_MODE);
851 	if (saved_autopoll & BGE_MIMODE_AUTOPOLL) {
852 		delay(40);
853 		BGE_STS_CLRBIT(sc, BGE_STS_AUTOPOLL);
854 		CSR_WRITE_4(sc, BGE_MI_MODE,
855 		    saved_autopoll & (~BGE_MIMODE_AUTOPOLL));
856 		delay(10); /* 40 usec is supposed to be adequate */
857 	}
858 
859 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_WRITE | BGE_MICOMM_BUSY |
860 	    BGE_MIPHY(phy) | BGE_MIREG(reg)|val);
861 
862 	for (i = 0; i < BGE_TIMEOUT; i++) {
863 		delay(10);
864 		if (!(CSR_READ_4(sc, BGE_MI_COMM) & BGE_MICOMM_BUSY)) {
865 			delay(5);
866 			CSR_READ_4(sc, BGE_MI_COMM);
867 			break;
868 		}
869 	}
870 
871 	if (saved_autopoll & BGE_MIMODE_AUTOPOLL) {
872 		BGE_STS_SETBIT(sc, BGE_STS_AUTOPOLL);
873 		CSR_WRITE_4(sc, BGE_MI_MODE, saved_autopoll);
874 		delay(40);
875 	}
876 
877 	if (i == BGE_TIMEOUT)
878 		aprint_error_dev(sc->bge_dev, "PHY read timed out\n");
879 }
880 
881 static void
882 bge_miibus_statchg(device_t dev)
883 {
884 	struct bge_softc *sc = device_private(dev);
885 	struct mii_data *mii = &sc->bge_mii;
886 
887 	/*
888 	 * Get flow control negotiation result.
889 	 */
890 	if (IFM_SUBTYPE(mii->mii_media.ifm_cur->ifm_media) == IFM_AUTO &&
891 	    (mii->mii_media_active & IFM_ETH_FMASK) != sc->bge_flowflags) {
892 		sc->bge_flowflags = mii->mii_media_active & IFM_ETH_FMASK;
893 		mii->mii_media_active &= ~IFM_ETH_FMASK;
894 	}
895 
896 	BGE_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_PORTMODE);
897 	if (IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_T ||
898 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_SX)
899 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_GMII);
900 	else
901 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_MII);
902 
903 	if ((mii->mii_media_active & IFM_GMASK) == IFM_FDX)
904 		BGE_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
905 	else
906 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
907 
908 	/*
909 	 * 802.3x flow control
910 	 */
911 	if (sc->bge_flowflags & IFM_ETH_RXPAUSE)
912 		BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_FLOWCTL_ENABLE);
913 	else
914 		BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_FLOWCTL_ENABLE);
915 
916 	if (sc->bge_flowflags & IFM_ETH_TXPAUSE)
917 		BGE_SETBIT(sc, BGE_TX_MODE, BGE_TXMODE_FLOWCTL_ENABLE);
918 	else
919 		BGE_CLRBIT(sc, BGE_TX_MODE, BGE_TXMODE_FLOWCTL_ENABLE);
920 }
921 
922 /*
923  * Update rx threshold levels to values in a particular slot
924  * of the interrupt-mitigation table bge_rx_threshes.
925  */
926 static void
927 bge_set_thresh(struct ifnet *ifp, int lvl)
928 {
929 	struct bge_softc *sc = ifp->if_softc;
930 	int s;
931 
932 	/* For now, just save the new Rx-intr thresholds and record
933 	 * that a threshold update is pending.  Updating the hardware
934 	 * registers here (even at splhigh()) is observed to
935 	 * occasionaly cause glitches where Rx-interrupts are not
936 	 * honoured for up to 10 seconds. jonathan@NetBSD.org, 2003-04-05
937 	 */
938 	s = splnet();
939 	sc->bge_rx_coal_ticks = bge_rx_threshes[lvl].rx_ticks;
940 	sc->bge_rx_max_coal_bds = bge_rx_threshes[lvl].rx_max_bds;
941 	sc->bge_pending_rxintr_change = 1;
942 	splx(s);
943 
944 	 return;
945 }
946 
947 
948 /*
949  * Update Rx thresholds of all bge devices
950  */
951 static void
952 bge_update_all_threshes(int lvl)
953 {
954 	struct ifnet *ifp;
955 	const char * const namebuf = "bge";
956 	int namelen;
957 
958 	if (lvl < 0)
959 		lvl = 0;
960 	else if( lvl >= NBGE_RX_THRESH)
961 		lvl = NBGE_RX_THRESH - 1;
962 
963 	namelen = strlen(namebuf);
964 	/*
965 	 * Now search all the interfaces for this name/number
966 	 */
967 	IFNET_FOREACH(ifp) {
968 		if (strncmp(ifp->if_xname, namebuf, namelen) != 0)
969 		      continue;
970 		/* We got a match: update if doing auto-threshold-tuning */
971 		if (bge_auto_thresh)
972 			bge_set_thresh(ifp, lvl);
973 	}
974 }
975 
976 /*
977  * Handle events that have triggered interrupts.
978  */
979 static void
980 bge_handle_events(struct bge_softc *sc)
981 {
982 
983 	return;
984 }
985 
986 /*
987  * Memory management for jumbo frames.
988  */
989 
990 static int
991 bge_alloc_jumbo_mem(struct bge_softc *sc)
992 {
993 	char *ptr, *kva;
994 	bus_dma_segment_t	seg;
995 	int		i, rseg, state, error;
996 	struct bge_jpool_entry   *entry;
997 
998 	state = error = 0;
999 
1000 	/* Grab a big chunk o' storage. */
1001 	if (bus_dmamem_alloc(sc->bge_dmatag, BGE_JMEM, PAGE_SIZE, 0,
1002 	     &seg, 1, &rseg, BUS_DMA_NOWAIT)) {
1003 		aprint_error_dev(sc->bge_dev, "can't alloc rx buffers\n");
1004 		return ENOBUFS;
1005 	}
1006 
1007 	state = 1;
1008 	if (bus_dmamem_map(sc->bge_dmatag, &seg, rseg, BGE_JMEM, (void **)&kva,
1009 	    BUS_DMA_NOWAIT)) {
1010 		aprint_error_dev(sc->bge_dev,
1011 		    "can't map DMA buffers (%d bytes)\n", (int)BGE_JMEM);
1012 		error = ENOBUFS;
1013 		goto out;
1014 	}
1015 
1016 	state = 2;
1017 	if (bus_dmamap_create(sc->bge_dmatag, BGE_JMEM, 1, BGE_JMEM, 0,
1018 	    BUS_DMA_NOWAIT, &sc->bge_cdata.bge_rx_jumbo_map)) {
1019 		aprint_error_dev(sc->bge_dev, "can't create DMA map\n");
1020 		error = ENOBUFS;
1021 		goto out;
1022 	}
1023 
1024 	state = 3;
1025 	if (bus_dmamap_load(sc->bge_dmatag, sc->bge_cdata.bge_rx_jumbo_map,
1026 	    kva, BGE_JMEM, NULL, BUS_DMA_NOWAIT)) {
1027 		aprint_error_dev(sc->bge_dev, "can't load DMA map\n");
1028 		error = ENOBUFS;
1029 		goto out;
1030 	}
1031 
1032 	state = 4;
1033 	sc->bge_cdata.bge_jumbo_buf = (void *)kva;
1034 	DPRINTFN(1,("bge_jumbo_buf = %p\n", sc->bge_cdata.bge_jumbo_buf));
1035 
1036 	SLIST_INIT(&sc->bge_jfree_listhead);
1037 	SLIST_INIT(&sc->bge_jinuse_listhead);
1038 
1039 	/*
1040 	 * Now divide it up into 9K pieces and save the addresses
1041 	 * in an array.
1042 	 */
1043 	ptr = sc->bge_cdata.bge_jumbo_buf;
1044 	for (i = 0; i < BGE_JSLOTS; i++) {
1045 		sc->bge_cdata.bge_jslots[i] = ptr;
1046 		ptr += BGE_JLEN;
1047 		entry = malloc(sizeof(struct bge_jpool_entry),
1048 		    M_DEVBUF, M_NOWAIT);
1049 		if (entry == NULL) {
1050 			aprint_error_dev(sc->bge_dev,
1051 			    "no memory for jumbo buffer queue!\n");
1052 			error = ENOBUFS;
1053 			goto out;
1054 		}
1055 		entry->slot = i;
1056 		SLIST_INSERT_HEAD(&sc->bge_jfree_listhead,
1057 				 entry, jpool_entries);
1058 	}
1059 out:
1060 	if (error != 0) {
1061 		switch (state) {
1062 		case 4:
1063 			bus_dmamap_unload(sc->bge_dmatag,
1064 			    sc->bge_cdata.bge_rx_jumbo_map);
1065 		case 3:
1066 			bus_dmamap_destroy(sc->bge_dmatag,
1067 			    sc->bge_cdata.bge_rx_jumbo_map);
1068 		case 2:
1069 			bus_dmamem_unmap(sc->bge_dmatag, kva, BGE_JMEM);
1070 		case 1:
1071 			bus_dmamem_free(sc->bge_dmatag, &seg, rseg);
1072 			break;
1073 		default:
1074 			break;
1075 		}
1076 	}
1077 
1078 	return error;
1079 }
1080 
1081 /*
1082  * Allocate a jumbo buffer.
1083  */
1084 static void *
1085 bge_jalloc(struct bge_softc *sc)
1086 {
1087 	struct bge_jpool_entry   *entry;
1088 
1089 	entry = SLIST_FIRST(&sc->bge_jfree_listhead);
1090 
1091 	if (entry == NULL) {
1092 		aprint_error_dev(sc->bge_dev, "no free jumbo buffers\n");
1093 		return (NULL);
1094 	}
1095 
1096 	SLIST_REMOVE_HEAD(&sc->bge_jfree_listhead, jpool_entries);
1097 	SLIST_INSERT_HEAD(&sc->bge_jinuse_listhead, entry, jpool_entries);
1098 	return (sc->bge_cdata.bge_jslots[entry->slot]);
1099 }
1100 
1101 /*
1102  * Release a jumbo buffer.
1103  */
1104 static void
1105 bge_jfree(struct mbuf *m, void *buf, size_t size, void *arg)
1106 {
1107 	struct bge_jpool_entry *entry;
1108 	struct bge_softc *sc;
1109 	int i, s;
1110 
1111 	/* Extract the softc struct pointer. */
1112 	sc = (struct bge_softc *)arg;
1113 
1114 	if (sc == NULL)
1115 		panic("bge_jfree: can't find softc pointer!");
1116 
1117 	/* calculate the slot this buffer belongs to */
1118 
1119 	i = ((char *)buf
1120 	     - (char *)sc->bge_cdata.bge_jumbo_buf) / BGE_JLEN;
1121 
1122 	if ((i < 0) || (i >= BGE_JSLOTS))
1123 		panic("bge_jfree: asked to free buffer that we don't manage!");
1124 
1125 	s = splvm();
1126 	entry = SLIST_FIRST(&sc->bge_jinuse_listhead);
1127 	if (entry == NULL)
1128 		panic("bge_jfree: buffer not in use!");
1129 	entry->slot = i;
1130 	SLIST_REMOVE_HEAD(&sc->bge_jinuse_listhead, jpool_entries);
1131 	SLIST_INSERT_HEAD(&sc->bge_jfree_listhead, entry, jpool_entries);
1132 
1133 	if (__predict_true(m != NULL))
1134   		pool_cache_put(mb_cache, m);
1135 	splx(s);
1136 }
1137 
1138 
1139 /*
1140  * Intialize a standard receive ring descriptor.
1141  */
1142 static int
1143 bge_newbuf_std(struct bge_softc *sc, int i, struct mbuf *m, bus_dmamap_t dmamap)
1144 {
1145 	struct mbuf		*m_new = NULL;
1146 	struct bge_rx_bd	*r;
1147 	int			error;
1148 
1149 	if (dmamap == NULL) {
1150 		error = bus_dmamap_create(sc->bge_dmatag, MCLBYTES, 1,
1151 		    MCLBYTES, 0, BUS_DMA_NOWAIT, &dmamap);
1152 		if (error != 0)
1153 			return error;
1154 	}
1155 
1156 	sc->bge_cdata.bge_rx_std_map[i] = dmamap;
1157 
1158 	if (m == NULL) {
1159 		MGETHDR(m_new, M_DONTWAIT, MT_DATA);
1160 		if (m_new == NULL)
1161 			return (ENOBUFS);
1162 
1163 		MCLGET(m_new, M_DONTWAIT);
1164 		if (!(m_new->m_flags & M_EXT)) {
1165 			m_freem(m_new);
1166 			return (ENOBUFS);
1167 		}
1168 		m_new->m_len = m_new->m_pkthdr.len = MCLBYTES;
1169 
1170 	} else {
1171 		m_new = m;
1172 		m_new->m_len = m_new->m_pkthdr.len = MCLBYTES;
1173 		m_new->m_data = m_new->m_ext.ext_buf;
1174 	}
1175 	if (!(sc->bge_flags & BGE_RX_ALIGNBUG))
1176 	    m_adj(m_new, ETHER_ALIGN);
1177 	if (bus_dmamap_load_mbuf(sc->bge_dmatag, dmamap, m_new,
1178 	    BUS_DMA_READ|BUS_DMA_NOWAIT))
1179 		return (ENOBUFS);
1180 	bus_dmamap_sync(sc->bge_dmatag, dmamap, 0, dmamap->dm_mapsize,
1181 	    BUS_DMASYNC_PREREAD);
1182 
1183 	sc->bge_cdata.bge_rx_std_chain[i] = m_new;
1184 	r = &sc->bge_rdata->bge_rx_std_ring[i];
1185 	bge_set_hostaddr(&r->bge_addr,
1186 	    dmamap->dm_segs[0].ds_addr);
1187 	r->bge_flags = BGE_RXBDFLAG_END;
1188 	r->bge_len = m_new->m_len;
1189 	r->bge_idx = i;
1190 
1191 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
1192 	    offsetof(struct bge_ring_data, bge_rx_std_ring) +
1193 		i * sizeof (struct bge_rx_bd),
1194 	    sizeof (struct bge_rx_bd),
1195 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1196 
1197 	return (0);
1198 }
1199 
1200 /*
1201  * Initialize a jumbo receive ring descriptor. This allocates
1202  * a jumbo buffer from the pool managed internally by the driver.
1203  */
1204 static int
1205 bge_newbuf_jumbo(struct bge_softc *sc, int i, struct mbuf *m)
1206 {
1207 	struct mbuf *m_new = NULL;
1208 	struct bge_rx_bd *r;
1209 	void *buf = NULL;
1210 
1211 	if (m == NULL) {
1212 
1213 		/* Allocate the mbuf. */
1214 		MGETHDR(m_new, M_DONTWAIT, MT_DATA);
1215 		if (m_new == NULL)
1216 			return (ENOBUFS);
1217 
1218 		/* Allocate the jumbo buffer */
1219 		buf = bge_jalloc(sc);
1220 		if (buf == NULL) {
1221 			m_freem(m_new);
1222 			aprint_error_dev(sc->bge_dev,
1223 			    "jumbo allocation failed -- packet dropped!\n");
1224 			return (ENOBUFS);
1225 		}
1226 
1227 		/* Attach the buffer to the mbuf. */
1228 		m_new->m_len = m_new->m_pkthdr.len = BGE_JUMBO_FRAMELEN;
1229 		MEXTADD(m_new, buf, BGE_JUMBO_FRAMELEN, M_DEVBUF,
1230 		    bge_jfree, sc);
1231 		m_new->m_flags |= M_EXT_RW;
1232 	} else {
1233 		m_new = m;
1234 		buf = m_new->m_data = m_new->m_ext.ext_buf;
1235 		m_new->m_ext.ext_size = BGE_JUMBO_FRAMELEN;
1236 	}
1237 	if (!(sc->bge_flags & BGE_RX_ALIGNBUG))
1238 	    m_adj(m_new, ETHER_ALIGN);
1239 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_cdata.bge_rx_jumbo_map,
1240 	    mtod(m_new, char *) - (char *)sc->bge_cdata.bge_jumbo_buf, BGE_JLEN,
1241 	    BUS_DMASYNC_PREREAD);
1242 	/* Set up the descriptor. */
1243 	r = &sc->bge_rdata->bge_rx_jumbo_ring[i];
1244 	sc->bge_cdata.bge_rx_jumbo_chain[i] = m_new;
1245 	bge_set_hostaddr(&r->bge_addr, BGE_JUMBO_DMA_ADDR(sc, m_new));
1246 	r->bge_flags = BGE_RXBDFLAG_END|BGE_RXBDFLAG_JUMBO_RING;
1247 	r->bge_len = m_new->m_len;
1248 	r->bge_idx = i;
1249 
1250 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
1251 	    offsetof(struct bge_ring_data, bge_rx_jumbo_ring) +
1252 		i * sizeof (struct bge_rx_bd),
1253 	    sizeof (struct bge_rx_bd),
1254 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1255 
1256 	return (0);
1257 }
1258 
1259 /*
1260  * The standard receive ring has 512 entries in it. At 2K per mbuf cluster,
1261  * that's 1MB or memory, which is a lot. For now, we fill only the first
1262  * 256 ring entries and hope that our CPU is fast enough to keep up with
1263  * the NIC.
1264  */
1265 static int
1266 bge_init_rx_ring_std(struct bge_softc *sc)
1267 {
1268 	int i;
1269 
1270 	if (sc->bge_flags & BGE_RXRING_VALID)
1271 		return 0;
1272 
1273 	for (i = 0; i < BGE_SSLOTS; i++) {
1274 		if (bge_newbuf_std(sc, i, NULL, 0) == ENOBUFS)
1275 			return (ENOBUFS);
1276 	}
1277 
1278 	sc->bge_std = i - 1;
1279 	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bge_std);
1280 
1281 	sc->bge_flags |= BGE_RXRING_VALID;
1282 
1283 	return (0);
1284 }
1285 
1286 static void
1287 bge_free_rx_ring_std(struct bge_softc *sc)
1288 {
1289 	int i;
1290 
1291 	if (!(sc->bge_flags & BGE_RXRING_VALID))
1292 		return;
1293 
1294 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1295 		if (sc->bge_cdata.bge_rx_std_chain[i] != NULL) {
1296 			m_freem(sc->bge_cdata.bge_rx_std_chain[i]);
1297 			sc->bge_cdata.bge_rx_std_chain[i] = NULL;
1298 			bus_dmamap_destroy(sc->bge_dmatag,
1299 			    sc->bge_cdata.bge_rx_std_map[i]);
1300 		}
1301 		memset((char *)&sc->bge_rdata->bge_rx_std_ring[i], 0,
1302 		    sizeof(struct bge_rx_bd));
1303 	}
1304 
1305 	sc->bge_flags &= ~BGE_RXRING_VALID;
1306 }
1307 
1308 static int
1309 bge_init_rx_ring_jumbo(struct bge_softc *sc)
1310 {
1311 	int i;
1312 	volatile struct bge_rcb *rcb;
1313 
1314 	if (sc->bge_flags & BGE_JUMBO_RXRING_VALID)
1315 		return 0;
1316 
1317 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1318 		if (bge_newbuf_jumbo(sc, i, NULL) == ENOBUFS)
1319 			return (ENOBUFS);
1320 	};
1321 
1322 	sc->bge_jumbo = i - 1;
1323 	sc->bge_flags |= BGE_JUMBO_RXRING_VALID;
1324 
1325 	rcb = &sc->bge_rdata->bge_info.bge_jumbo_rx_rcb;
1326 	rcb->bge_maxlen_flags = 0;
1327 	CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1328 
1329 	bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
1330 
1331 	return (0);
1332 }
1333 
1334 static void
1335 bge_free_rx_ring_jumbo(struct bge_softc *sc)
1336 {
1337 	int i;
1338 
1339 	if (!(sc->bge_flags & BGE_JUMBO_RXRING_VALID))
1340 		return;
1341 
1342 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1343 		if (sc->bge_cdata.bge_rx_jumbo_chain[i] != NULL) {
1344 			m_freem(sc->bge_cdata.bge_rx_jumbo_chain[i]);
1345 			sc->bge_cdata.bge_rx_jumbo_chain[i] = NULL;
1346 		}
1347 		memset((char *)&sc->bge_rdata->bge_rx_jumbo_ring[i], 0,
1348 		    sizeof(struct bge_rx_bd));
1349 	}
1350 
1351 	sc->bge_flags &= ~BGE_JUMBO_RXRING_VALID;
1352 }
1353 
1354 static void
1355 bge_free_tx_ring(struct bge_softc *sc)
1356 {
1357 	int i, freed;
1358 	struct txdmamap_pool_entry *dma;
1359 
1360 	if (!(sc->bge_flags & BGE_TXRING_VALID))
1361 		return;
1362 
1363 	freed = 0;
1364 
1365 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
1366 		if (sc->bge_cdata.bge_tx_chain[i] != NULL) {
1367 			freed++;
1368 			m_freem(sc->bge_cdata.bge_tx_chain[i]);
1369 			sc->bge_cdata.bge_tx_chain[i] = NULL;
1370 			SLIST_INSERT_HEAD(&sc->txdma_list, sc->txdma[i],
1371 					    link);
1372 			sc->txdma[i] = 0;
1373 		}
1374 		memset((char *)&sc->bge_rdata->bge_tx_ring[i], 0,
1375 		    sizeof(struct bge_tx_bd));
1376 	}
1377 
1378 	while ((dma = SLIST_FIRST(&sc->txdma_list))) {
1379 		SLIST_REMOVE_HEAD(&sc->txdma_list, link);
1380 		bus_dmamap_destroy(sc->bge_dmatag, dma->dmamap);
1381 		free(dma, M_DEVBUF);
1382 	}
1383 
1384 	sc->bge_flags &= ~BGE_TXRING_VALID;
1385 }
1386 
1387 static int
1388 bge_init_tx_ring(struct bge_softc *sc)
1389 {
1390 	int i;
1391 	bus_dmamap_t dmamap;
1392 	struct txdmamap_pool_entry *dma;
1393 
1394 	if (sc->bge_flags & BGE_TXRING_VALID)
1395 		return 0;
1396 
1397 	sc->bge_txcnt = 0;
1398 	sc->bge_tx_saved_considx = 0;
1399 
1400 	/* Initialize transmit producer index for host-memory send ring. */
1401 	sc->bge_tx_prodidx = 0;
1402 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1403 	/* 5700 b2 errata */
1404 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX)
1405 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1406 
1407 	/* NIC-memory send ring not used; initialize to zero. */
1408 	bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1409 	/* 5700 b2 errata */
1410 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX)
1411 		bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1412 
1413 	SLIST_INIT(&sc->txdma_list);
1414 	for (i = 0; i < BGE_RSLOTS; i++) {
1415 		if (bus_dmamap_create(sc->bge_dmatag, BGE_TXDMA_MAX,
1416 		    BGE_NTXSEG, ETHER_MAX_LEN_JUMBO, 0, BUS_DMA_NOWAIT,
1417 		    &dmamap))
1418 			return (ENOBUFS);
1419 		if (dmamap == NULL)
1420 			panic("dmamap NULL in bge_init_tx_ring");
1421 		dma = malloc(sizeof(*dma), M_DEVBUF, M_NOWAIT);
1422 		if (dma == NULL) {
1423 			aprint_error_dev(sc->bge_dev,
1424 			    "can't alloc txdmamap_pool_entry\n");
1425 			bus_dmamap_destroy(sc->bge_dmatag, dmamap);
1426 			return (ENOMEM);
1427 		}
1428 		dma->dmamap = dmamap;
1429 		SLIST_INSERT_HEAD(&sc->txdma_list, dma, link);
1430 	}
1431 
1432 	sc->bge_flags |= BGE_TXRING_VALID;
1433 
1434 	return (0);
1435 }
1436 
1437 static void
1438 bge_setmulti(struct bge_softc *sc)
1439 {
1440 	struct ethercom		*ac = &sc->ethercom;
1441 	struct ifnet		*ifp = &ac->ec_if;
1442 	struct ether_multi	*enm;
1443 	struct ether_multistep  step;
1444 	u_int32_t		hashes[4] = { 0, 0, 0, 0 };
1445 	u_int32_t		h;
1446 	int			i;
1447 
1448 	if (ifp->if_flags & IFF_PROMISC)
1449 		goto allmulti;
1450 
1451 	/* Now program new ones. */
1452 	ETHER_FIRST_MULTI(step, ac, enm);
1453 	while (enm != NULL) {
1454 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
1455 			/*
1456 			 * We must listen to a range of multicast addresses.
1457 			 * For now, just accept all multicasts, rather than
1458 			 * trying to set only those filter bits needed to match
1459 			 * the range.  (At this time, the only use of address
1460 			 * ranges is for IP multicast routing, for which the
1461 			 * range is big enough to require all bits set.)
1462 			 */
1463 			goto allmulti;
1464 		}
1465 
1466 		h = ether_crc32_le(enm->enm_addrlo, ETHER_ADDR_LEN);
1467 
1468 		/* Just want the 7 least-significant bits. */
1469 		h &= 0x7f;
1470 
1471 		hashes[(h & 0x60) >> 5] |= 1 << (h & 0x1F);
1472 		ETHER_NEXT_MULTI(step, enm);
1473 	}
1474 
1475 	ifp->if_flags &= ~IFF_ALLMULTI;
1476 	goto setit;
1477 
1478  allmulti:
1479 	ifp->if_flags |= IFF_ALLMULTI;
1480 	hashes[0] = hashes[1] = hashes[2] = hashes[3] = 0xffffffff;
1481 
1482  setit:
1483 	for (i = 0; i < 4; i++)
1484 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), hashes[i]);
1485 }
1486 
1487 const int bge_swapbits[] = {
1488 	0,
1489 	BGE_MODECTL_BYTESWAP_DATA,
1490 	BGE_MODECTL_WORDSWAP_DATA,
1491 	BGE_MODECTL_BYTESWAP_NONFRAME,
1492 	BGE_MODECTL_WORDSWAP_NONFRAME,
1493 
1494 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_WORDSWAP_DATA,
1495 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_BYTESWAP_NONFRAME,
1496 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_WORDSWAP_NONFRAME,
1497 
1498 	BGE_MODECTL_WORDSWAP_DATA|BGE_MODECTL_BYTESWAP_NONFRAME,
1499 	BGE_MODECTL_WORDSWAP_DATA|BGE_MODECTL_WORDSWAP_NONFRAME,
1500 
1501 	BGE_MODECTL_BYTESWAP_NONFRAME|BGE_MODECTL_WORDSWAP_NONFRAME,
1502 
1503 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_WORDSWAP_DATA|
1504 	    BGE_MODECTL_BYTESWAP_NONFRAME,
1505 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_WORDSWAP_DATA|
1506 	    BGE_MODECTL_WORDSWAP_NONFRAME,
1507 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_BYTESWAP_NONFRAME|
1508 	    BGE_MODECTL_WORDSWAP_NONFRAME,
1509 	BGE_MODECTL_WORDSWAP_DATA|BGE_MODECTL_BYTESWAP_NONFRAME|
1510 	    BGE_MODECTL_WORDSWAP_NONFRAME,
1511 
1512 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_WORDSWAP_DATA|
1513 	    BGE_MODECTL_BYTESWAP_NONFRAME|BGE_MODECTL_WORDSWAP_NONFRAME,
1514 };
1515 
1516 int bge_swapindex = 0;
1517 
1518 /*
1519  * Do endian, PCI and DMA initialization. Also check the on-board ROM
1520  * self-test results.
1521  */
1522 static int
1523 bge_chipinit(struct bge_softc *sc)
1524 {
1525 	int			i;
1526 	u_int32_t		dma_rw_ctl;
1527 
1528 
1529 	/* Set endianness before we access any non-PCI registers. */
1530 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL,
1531 	    BGE_INIT);
1532 
1533 	/* Set power state to D0. */
1534 	bge_setpowerstate(sc, 0);
1535 
1536 	/*
1537 	 * Check the 'ROM failed' bit on the RX CPU to see if
1538 	 * self-tests passed.
1539 	 */
1540 	if (CSR_READ_4(sc, BGE_RXCPU_MODE) & BGE_RXCPUMODE_ROMFAIL) {
1541 		aprint_error_dev(sc->bge_dev,
1542 		    "RX CPU self-diagnostics failed!\n");
1543 		return (ENODEV);
1544 	}
1545 
1546 	/* Clear the MAC control register */
1547 	CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
1548 
1549 	/*
1550 	 * Clear the MAC statistics block in the NIC's
1551 	 * internal memory.
1552 	 */
1553 	for (i = BGE_STATS_BLOCK;
1554 	    i < BGE_STATS_BLOCK_END + 1; i += sizeof(u_int32_t))
1555 		BGE_MEMWIN_WRITE(sc->sc_pc, sc->sc_pcitag, i, 0);
1556 
1557 	for (i = BGE_STATUS_BLOCK;
1558 	    i < BGE_STATUS_BLOCK_END + 1; i += sizeof(u_int32_t))
1559 		BGE_MEMWIN_WRITE(sc->sc_pc, sc->sc_pcitag, i, 0);
1560 
1561 	/* Set up the PCI DMA control register. */
1562 	if (sc->bge_flags & BGE_PCIE) {
1563 	  u_int32_t device_ctl;
1564 
1565 		/* From FreeBSD */
1566 		DPRINTFN(4, ("(%s: PCI-Express DMA setting)\n",
1567 		    device_xname(sc->bge_dev)));
1568 		dma_rw_ctl = (BGE_PCI_READ_CMD | BGE_PCI_WRITE_CMD |
1569 		    (0xf << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1570 		    (0x2 << BGE_PCIDMARWCTL_WR_WAT_SHIFT));
1571 
1572 		/* jonathan: alternative from Linux driver */
1573 #define DMA_CTRL_WRITE_PCIE_H20MARK_128         0x00180000
1574 #define DMA_CTRL_WRITE_PCIE_H20MARK_256         0x00380000
1575 
1576 		dma_rw_ctl =   0x76000000; /* XXX XXX XXX */;
1577 		device_ctl = pci_conf_read(sc->sc_pc, sc->sc_pcitag,
1578 					   BGE_PCI_CONF_DEV_CTRL);
1579 		aprint_debug_dev(sc->bge_dev, "pcie mode=0x%x\n", device_ctl);
1580 
1581 		if ((device_ctl & 0x00e0) && 0) {
1582 			/*
1583 			 * XXX jonathan@NetBSD.org:
1584 			 * This clause is exactly what the Broadcom-supplied
1585 			 * Linux does; but given overall register programming
1586 			 * by if_bge(4), this larger DMA-write watermark
1587 			 * value causes bcm5721 chips to totally wedge.
1588 			 */
1589 			dma_rw_ctl |= BGE_PCIDMA_RWCTL_PCIE_WRITE_WATRMARK_256;
1590 		} else {
1591 			dma_rw_ctl |= BGE_PCIDMA_RWCTL_PCIE_WRITE_WATRMARK_128;
1592 		}
1593 	} else if (sc->bge_flags & BGE_PCIX){
1594 	  	DPRINTFN(4, ("(:%s: PCI-X DMA setting)\n",
1595 		    device_xname(sc->bge_dev)));
1596 		/* PCI-X bus */
1597 		dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD |
1598 		    (0x3 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1599 		    (0x3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT) |
1600 		    (0x0F);
1601 		/*
1602 		 * 5703 and 5704 need ONEDMA_AT_ONCE as a workaround
1603 		 * for hardware bugs, which means we should also clear
1604 		 * the low-order MINDMA bits.  In addition, the 5704
1605 		 * uses a different encoding of read/write watermarks.
1606 		 */
1607 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704) {
1608 			dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD |
1609 			  /* should be 0x1f0000 */
1610 			  (0x7 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1611 			  (0x3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1612 			dma_rw_ctl |= BGE_PCIDMARWCTL_ONEDMA_ATONCE;
1613 		}
1614 		else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703) {
1615 			dma_rw_ctl &=  0xfffffff0;
1616 			dma_rw_ctl |= BGE_PCIDMARWCTL_ONEDMA_ATONCE;
1617 		}
1618 		else if (BGE_IS_5714_FAMILY(sc)) {
1619 			dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD;
1620 			dma_rw_ctl &= ~BGE_PCIDMARWCTL_ONEDMA_ATONCE; /* XXX */
1621 			/* XXX magic values, Broadcom-supplied Linux driver */
1622 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5780)
1623 				dma_rw_ctl |= (1 << 20) | (1 << 18) |
1624 				  BGE_PCIDMARWCTL_ONEDMA_ATONCE;
1625 			else
1626 				dma_rw_ctl |= (1<<20) | (1<<18) | (1 << 15);
1627 		}
1628 	} else {
1629 		/* Conventional PCI bus */
1630 	  	DPRINTFN(4, ("(%s: PCI 2.2 DMA setting)\n",
1631 		    device_xname(sc->bge_dev)));
1632 		dma_rw_ctl = (BGE_PCI_READ_CMD | BGE_PCI_WRITE_CMD |
1633 		   (0x7 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1634 		   (0x7 << BGE_PCIDMARWCTL_WR_WAT_SHIFT));
1635 		if (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5705 &&
1636 		    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5750)
1637 			dma_rw_ctl |= 0x0F;
1638 	}
1639 
1640 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
1641 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5701)
1642 		dma_rw_ctl |= BGE_PCIDMARWCTL_USE_MRM |
1643 		    BGE_PCIDMARWCTL_ASRT_ALL_BE;
1644 
1645 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703 ||
1646 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704)
1647 		dma_rw_ctl &= ~BGE_PCIDMARWCTL_MINDMA;
1648 
1649 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_DMA_RW_CTL, dma_rw_ctl);
1650 
1651 	/*
1652 	 * Set up general mode register.
1653 	 */
1654 	CSR_WRITE_4(sc, BGE_MODE_CTL, BGE_DMA_SWAP_OPTIONS |
1655 		    BGE_MODECTL_MAC_ATTN_INTR | BGE_MODECTL_HOST_SEND_BDS |
1656 		    BGE_MODECTL_TX_NO_PHDR_CSUM | BGE_MODECTL_RX_NO_PHDR_CSUM);
1657 
1658 	/*
1659 	 * Disable memory write invalidate.  Apparently it is not supported
1660 	 * properly by these devices.
1661 	 */
1662 	PCI_CLRBIT(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD, PCIM_CMD_MWIEN);
1663 
1664 
1665 #ifdef __brokenalpha__
1666 	/*
1667 	 * Must insure that we do not cross an 8K (bytes) boundary
1668 	 * for DMA reads.  Our highest limit is 1K bytes.  This is a
1669 	 * restriction on some ALPHA platforms with early revision
1670 	 * 21174 PCI chipsets, such as the AlphaPC 164lx
1671 	 */
1672 	PCI_SETBIT(sc, BGE_PCI_DMA_RW_CTL, BGE_PCI_READ_BNDRY_1024, 4);
1673 #endif
1674 
1675 	/* Set the timer prescaler (always 66MHz) */
1676 	CSR_WRITE_4(sc, BGE_MISC_CFG, 65 << 1/*BGE_32BITTIME_66MHZ*/);
1677 
1678 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
1679 		DELAY(40);	/* XXX */
1680 
1681 		/* Put PHY into ready state */
1682 		BGE_CLRBIT(sc, BGE_MISC_CFG, BGE_MISCCFG_EPHY_IDDQ);
1683 		CSR_READ_4(sc, BGE_MISC_CFG); /* Flush */
1684 		DELAY(40);
1685 	}
1686 
1687 	return (0);
1688 }
1689 
1690 static int
1691 bge_blockinit(struct bge_softc *sc)
1692 {
1693 	volatile struct bge_rcb		*rcb;
1694 	bus_size_t		rcb_addr;
1695 	int			i;
1696 	struct ifnet		*ifp = &sc->ethercom.ec_if;
1697 	bge_hostaddr		taddr;
1698 	u_int32_t		val;
1699 
1700 	/*
1701 	 * Initialize the memory window pointer register so that
1702 	 * we can access the first 32K of internal NIC RAM. This will
1703 	 * allow us to set up the TX send ring RCBs and the RX return
1704 	 * ring RCBs, plus other things which live in NIC memory.
1705 	 */
1706 
1707 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, 0);
1708 
1709 	/* Configure mbuf memory pool */
1710 	if (!(BGE_IS_5705_OR_BEYOND(sc))) {
1711 		if (sc->bge_extram) {
1712 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_BASEADDR,
1713 			    BGE_EXT_SSRAM);
1714 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704)
1715 				CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x10000);
1716 			else
1717 				CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x18000);
1718 		} else {
1719 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_BASEADDR,
1720 			    BGE_BUFFPOOL_1);
1721 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704)
1722 				CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x10000);
1723 			else
1724 				CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x18000);
1725 		}
1726 
1727 		/* Configure DMA resource pool */
1728 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_BASEADDR,
1729 		    BGE_DMA_DESCRIPTORS);
1730 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LEN, 0x2000);
1731 	}
1732 
1733 	/* Configure mbuf pool watermarks */
1734 #ifdef ORIG_WPAUL_VALUES
1735 	CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 24);
1736 	CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 24);
1737 	CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 48);
1738 #else
1739 
1740 	/* new broadcom docs strongly recommend these: */
1741 	if (!BGE_IS_5705_OR_BEYOND(sc)) {
1742 		if (ifp->if_mtu > ETHER_MAX_LEN) {
1743 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x50);
1744 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x20);
1745 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1746 		} else {
1747 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 304);
1748 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 152);
1749 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 380);
1750 		}
1751 	} else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
1752 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1753 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x04);
1754 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x10);
1755 	} else {
1756 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1757 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x10);
1758 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1759 	}
1760 #endif
1761 
1762 	/* Configure DMA resource watermarks */
1763 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LOWAT, 5);
1764 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_HIWAT, 10);
1765 
1766 	/* Enable buffer manager */
1767 	if (!BGE_IS_5705_OR_BEYOND(sc)) {
1768 		CSR_WRITE_4(sc, BGE_BMAN_MODE,
1769 		    BGE_BMANMODE_ENABLE | BGE_BMANMODE_LOMBUF_ATTN);
1770 
1771 		/* Poll for buffer manager start indication */
1772 		for (i = 0; i < BGE_TIMEOUT; i++) {
1773 			if (CSR_READ_4(sc, BGE_BMAN_MODE) & BGE_BMANMODE_ENABLE)
1774 				break;
1775 			DELAY(10);
1776 		}
1777 
1778 		if (i == BGE_TIMEOUT) {
1779 			aprint_error_dev(sc->bge_dev,
1780 			    "buffer manager failed to start\n");
1781 			return (ENXIO);
1782 		}
1783 	}
1784 
1785 	/* Enable flow-through queues */
1786 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
1787 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
1788 
1789 	/* Wait until queue initialization is complete */
1790 	for (i = 0; i < BGE_TIMEOUT; i++) {
1791 		if (CSR_READ_4(sc, BGE_FTQ_RESET) == 0)
1792 			break;
1793 		DELAY(10);
1794 	}
1795 
1796 	if (i == BGE_TIMEOUT) {
1797 		aprint_error_dev(sc->bge_dev,
1798 		    "flow-through queue init failed\n");
1799 		return (ENXIO);
1800 	}
1801 
1802 	/* Initialize the standard RX ring control block */
1803 	rcb = &sc->bge_rdata->bge_info.bge_std_rx_rcb;
1804 	bge_set_hostaddr(&rcb->bge_hostaddr,
1805 	    BGE_RING_DMA_ADDR(sc, bge_rx_std_ring));
1806 	if (BGE_IS_5705_OR_BEYOND(sc))
1807 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(512, 0);
1808 	else
1809 		rcb->bge_maxlen_flags =
1810 		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN, 0);
1811 	if (sc->bge_extram)
1812 		rcb->bge_nicaddr = BGE_EXT_STD_RX_RINGS;
1813 	else
1814 		rcb->bge_nicaddr = BGE_STD_RX_RINGS;
1815 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_HI, rcb->bge_hostaddr.bge_addr_hi);
1816 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_LO, rcb->bge_hostaddr.bge_addr_lo);
1817 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1818 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_NICADDR, rcb->bge_nicaddr);
1819 
1820 	if (BGE_IS_5705_OR_BEYOND(sc))
1821 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT_5705;
1822 	else
1823 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT;
1824 
1825 	/*
1826 	 * Initialize the jumbo RX ring control block
1827 	 * We set the 'ring disabled' bit in the flags
1828 	 * field until we're actually ready to start
1829 	 * using this ring (i.e. once we set the MTU
1830 	 * high enough to require it).
1831 	 */
1832 	if (!BGE_IS_5705_OR_BEYOND(sc)) {
1833 		rcb = &sc->bge_rdata->bge_info.bge_jumbo_rx_rcb;
1834 		bge_set_hostaddr(&rcb->bge_hostaddr,
1835 		    BGE_RING_DMA_ADDR(sc, bge_rx_jumbo_ring));
1836 		rcb->bge_maxlen_flags =
1837 		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN,
1838 			BGE_RCB_FLAG_RING_DISABLED);
1839 		if (sc->bge_extram)
1840 			rcb->bge_nicaddr = BGE_EXT_JUMBO_RX_RINGS;
1841 		else
1842 			rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS;
1843 
1844 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_HI,
1845 		    rcb->bge_hostaddr.bge_addr_hi);
1846 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_LO,
1847 		    rcb->bge_hostaddr.bge_addr_lo);
1848 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS,
1849 		    rcb->bge_maxlen_flags);
1850 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_NICADDR, rcb->bge_nicaddr);
1851 
1852 		/* Set up dummy disabled mini ring RCB */
1853 		rcb = &sc->bge_rdata->bge_info.bge_mini_rx_rcb;
1854 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(0,
1855 		    BGE_RCB_FLAG_RING_DISABLED);
1856 		CSR_WRITE_4(sc, BGE_RX_MINI_RCB_MAXLEN_FLAGS,
1857 		    rcb->bge_maxlen_flags);
1858 
1859 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
1860 		    offsetof(struct bge_ring_data, bge_info),
1861 		    sizeof (struct bge_gib),
1862 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1863 	}
1864 
1865 	/*
1866 	 * Set the BD ring replenish thresholds. The recommended
1867 	 * values are 1/8th the number of descriptors allocated to
1868 	 * each ring.
1869 	 */
1870 	i = BGE_STD_RX_RING_CNT / 8;
1871 
1872 	/*
1873 	 * Use a value of 8 for the following chips to workaround HW errata.
1874 	 * Some of these chips have been added based on empirical
1875 	 * evidence (they don't work unless this is done).
1876 	 */
1877 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5750 ||
1878 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5752 ||
1879 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 ||
1880 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5787 ||
1881 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906)
1882 		i = 8;
1883 
1884 	CSR_WRITE_4(sc, BGE_RBDI_STD_REPL_THRESH, i);
1885 	CSR_WRITE_4(sc, BGE_RBDI_JUMBO_REPL_THRESH, BGE_JUMBO_RX_RING_CNT / 8);
1886 
1887 	/*
1888 	 * Disable all unused send rings by setting the 'ring disabled'
1889 	 * bit in the flags field of all the TX send ring control blocks.
1890 	 * These are located in NIC memory.
1891 	 */
1892 	rcb_addr = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1893 	for (i = 0; i < BGE_TX_RINGS_EXTSSRAM_MAX; i++) {
1894 		RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
1895 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED));
1896 		RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, 0);
1897 		rcb_addr += sizeof(struct bge_rcb);
1898 	}
1899 
1900 	/* Configure TX RCB 0 (we use only the first ring) */
1901 	rcb_addr = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1902 	bge_set_hostaddr(&taddr, BGE_RING_DMA_ADDR(sc, bge_tx_ring));
1903 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1904 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1905 	RCB_WRITE_4(sc, rcb_addr, bge_nicaddr,
1906 		    BGE_NIC_TXRING_ADDR(0, BGE_TX_RING_CNT));
1907 	if (!(BGE_IS_5705_OR_BEYOND(sc)))
1908 		RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
1909 		    BGE_RCB_MAXLEN_FLAGS(BGE_TX_RING_CNT, 0));
1910 
1911 	/* Disable all unused RX return rings */
1912 	rcb_addr = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1913 	for (i = 0; i < BGE_RX_RINGS_MAX; i++) {
1914 		RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_hi, 0);
1915 		RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_lo, 0);
1916 		RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
1917 			    BGE_RCB_MAXLEN_FLAGS(sc->bge_return_ring_cnt,
1918                                      BGE_RCB_FLAG_RING_DISABLED));
1919 		RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, 0);
1920 		bge_writembx(sc, BGE_MBX_RX_CONS0_LO +
1921 		    (i * (sizeof(u_int64_t))), 0);
1922 		rcb_addr += sizeof(struct bge_rcb);
1923 	}
1924 
1925 	/* Initialize RX ring indexes */
1926 	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, 0);
1927 	bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, 0);
1928 	bge_writembx(sc, BGE_MBX_RX_MINI_PROD_LO, 0);
1929 
1930 	/*
1931 	 * Set up RX return ring 0
1932 	 * Note that the NIC address for RX return rings is 0x00000000.
1933 	 * The return rings live entirely within the host, so the
1934 	 * nicaddr field in the RCB isn't used.
1935 	 */
1936 	rcb_addr = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1937 	bge_set_hostaddr(&taddr, BGE_RING_DMA_ADDR(sc, bge_rx_return_ring));
1938 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1939 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1940 	RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, 0x00000000);
1941 	RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
1942 	    BGE_RCB_MAXLEN_FLAGS(sc->bge_return_ring_cnt, 0));
1943 
1944 	/* Set random backoff seed for TX */
1945 	CSR_WRITE_4(sc, BGE_TX_RANDOM_BACKOFF,
1946 	    CLLADDR(ifp->if_sadl)[0] + CLLADDR(ifp->if_sadl)[1] +
1947 	    CLLADDR(ifp->if_sadl)[2] + CLLADDR(ifp->if_sadl)[3] +
1948 	    CLLADDR(ifp->if_sadl)[4] + CLLADDR(ifp->if_sadl)[5] +
1949 	    BGE_TX_BACKOFF_SEED_MASK);
1950 
1951 	/* Set inter-packet gap */
1952 	CSR_WRITE_4(sc, BGE_TX_LENGTHS, 0x2620);
1953 
1954 	/*
1955 	 * Specify which ring to use for packets that don't match
1956 	 * any RX rules.
1957 	 */
1958 	CSR_WRITE_4(sc, BGE_RX_RULES_CFG, 0x08);
1959 
1960 	/*
1961 	 * Configure number of RX lists. One interrupt distribution
1962 	 * list, sixteen active lists, one bad frames class.
1963 	 */
1964 	CSR_WRITE_4(sc, BGE_RXLP_CFG, 0x181);
1965 
1966 	/* Inialize RX list placement stats mask. */
1967 	CSR_WRITE_4(sc, BGE_RXLP_STATS_ENABLE_MASK, 0x007FFFFF);
1968 	CSR_WRITE_4(sc, BGE_RXLP_STATS_CTL, 0x1);
1969 
1970 	/* Disable host coalescing until we get it set up */
1971 	CSR_WRITE_4(sc, BGE_HCC_MODE, 0x00000000);
1972 
1973 	/* Poll to make sure it's shut down. */
1974 	for (i = 0; i < BGE_TIMEOUT; i++) {
1975 		if (!(CSR_READ_4(sc, BGE_HCC_MODE) & BGE_HCCMODE_ENABLE))
1976 			break;
1977 		DELAY(10);
1978 	}
1979 
1980 	if (i == BGE_TIMEOUT) {
1981 		aprint_error_dev(sc->bge_dev,
1982 		    "host coalescing engine failed to idle\n");
1983 		return (ENXIO);
1984 	}
1985 
1986 	/* Set up host coalescing defaults */
1987 	CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, sc->bge_rx_coal_ticks);
1988 	CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS, sc->bge_tx_coal_ticks);
1989 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, sc->bge_rx_max_coal_bds);
1990 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS, sc->bge_tx_max_coal_bds);
1991 	if (!(BGE_IS_5705_OR_BEYOND(sc))) {
1992 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS_INT, 0);
1993 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS_INT, 0);
1994 	}
1995 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT, 0);
1996 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT, 0);
1997 
1998 	/* Set up address of statistics block */
1999 	if (!(BGE_IS_5705_OR_BEYOND(sc))) {
2000 		bge_set_hostaddr(&taddr,
2001 		    BGE_RING_DMA_ADDR(sc, bge_info.bge_stats));
2002 		CSR_WRITE_4(sc, BGE_HCC_STATS_TICKS, sc->bge_stat_ticks);
2003 		CSR_WRITE_4(sc, BGE_HCC_STATS_BASEADDR, BGE_STATS_BLOCK);
2004 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_HI, taddr.bge_addr_hi);
2005 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_LO, taddr.bge_addr_lo);
2006 	}
2007 
2008 	/* Set up address of status block */
2009 	bge_set_hostaddr(&taddr, BGE_RING_DMA_ADDR(sc, bge_status_block));
2010 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_BASEADDR, BGE_STATUS_BLOCK);
2011 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_HI, taddr.bge_addr_hi);
2012 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_LO, taddr.bge_addr_lo);
2013 	sc->bge_rdata->bge_status_block.bge_idx[0].bge_rx_prod_idx = 0;
2014 	sc->bge_rdata->bge_status_block.bge_idx[0].bge_tx_cons_idx = 0;
2015 
2016 	/* Turn on host coalescing state machine */
2017 	CSR_WRITE_4(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE);
2018 
2019 	/* Turn on RX BD completion state machine and enable attentions */
2020 	CSR_WRITE_4(sc, BGE_RBDC_MODE,
2021 	    BGE_RBDCMODE_ENABLE | BGE_RBDCMODE_ATTN);
2022 
2023 	/* Turn on RX list placement state machine */
2024 	CSR_WRITE_4(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
2025 
2026 	/* Turn on RX list selector state machine. */
2027 	if (!(BGE_IS_5705_OR_BEYOND(sc)))
2028 		CSR_WRITE_4(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
2029 
2030 	val = BGE_MACMODE_TXDMA_ENB | BGE_MACMODE_RXDMA_ENB |
2031 	    BGE_MACMODE_RX_STATS_CLEAR | BGE_MACMODE_TX_STATS_CLEAR |
2032 	    BGE_MACMODE_RX_STATS_ENB | BGE_MACMODE_TX_STATS_ENB |
2033 	    BGE_MACMODE_FRMHDR_DMA_ENB;
2034 
2035 	if (sc->bge_flags & BGE_PHY_FIBER_TBI)
2036 	    val |= BGE_PORTMODE_TBI;
2037 	else if (sc->bge_flags & BGE_PHY_FIBER_MII)
2038 	    val |= BGE_PORTMODE_GMII;
2039 	else
2040 	    val |= BGE_PORTMODE_MII;
2041 
2042 	/* Turn on DMA, clear stats */
2043 	CSR_WRITE_4(sc, BGE_MAC_MODE, val);
2044 
2045 
2046 	/* Set misc. local control, enable interrupts on attentions */
2047 	sc->bge_local_ctrl_reg = BGE_MLC_INTR_ONATTN | BGE_MLC_AUTO_EEPROM;
2048 
2049 #ifdef notdef
2050 	/* Assert GPIO pins for PHY reset */
2051 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUT0|
2052 	    BGE_MLC_MISCIO_OUT1|BGE_MLC_MISCIO_OUT2);
2053 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUTEN0|
2054 	    BGE_MLC_MISCIO_OUTEN1|BGE_MLC_MISCIO_OUTEN2);
2055 #endif
2056 
2057 #if defined(not_quite_yet)
2058 	/* Linux driver enables enable gpio pin #1 on 5700s */
2059 	if (sc->bge_chipid == BGE_CHIPID_BCM5700) {
2060 		sc->bge_local_ctrl_reg |=
2061 		  (BGE_MLC_MISCIO_OUT1|BGE_MLC_MISCIO_OUTEN1);
2062 	}
2063 #endif
2064 	CSR_WRITE_4(sc, BGE_MISC_LOCAL_CTL, sc->bge_local_ctrl_reg);
2065 
2066 	/* Turn on DMA completion state machine */
2067 	if (!(BGE_IS_5705_OR_BEYOND(sc)))
2068 		CSR_WRITE_4(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
2069 
2070 	/* Turn on write DMA state machine */
2071 	{
2072 		uint32_t bge_wdma_mode =
2073 			BGE_WDMAMODE_ENABLE|BGE_WDMAMODE_ALL_ATTNS;
2074 
2075 		/* Enable host coalescing bug fix; see Linux tg3.c */
2076 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 ||
2077 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5787)
2078 			bge_wdma_mode |= (1 << 29);
2079 
2080 		CSR_WRITE_4(sc, BGE_WDMA_MODE, bge_wdma_mode);
2081         }
2082 
2083 	/* Turn on read DMA state machine */
2084 	{
2085 		uint32_t dma_read_modebits;
2086 
2087 		dma_read_modebits =
2088 		  BGE_RDMAMODE_ENABLE | BGE_RDMAMODE_ALL_ATTNS;
2089 
2090 		if ((sc->bge_flags & BGE_PCIE) && 0) {
2091 			dma_read_modebits |= BGE_RDMA_MODE_FIFO_LONG_BURST;
2092 		} else if (BGE_IS_5705_OR_BEYOND(sc)) {
2093 			dma_read_modebits |= BGE_RDMA_MODE_FIFO_SIZE_128;
2094 		}
2095 
2096 		/* XXX broadcom-supplied linux driver; undocumented */
2097 		if (BGE_IS_5750_OR_BEYOND(sc)) {
2098  			/*
2099 			 * XXX: magic values.
2100 			 * From Broadcom-supplied Linux driver;  apparently
2101 			 * required to workaround a DMA bug affecting TSO
2102 			 * on bcm575x/bcm5721?
2103 			 */
2104 			dma_read_modebits |= (1 << 27);
2105 		}
2106 		CSR_WRITE_4(sc, BGE_RDMA_MODE, dma_read_modebits);
2107 	}
2108 
2109 	/* Turn on RX data completion state machine */
2110 	CSR_WRITE_4(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
2111 
2112 	/* Turn on RX BD initiator state machine */
2113 	CSR_WRITE_4(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
2114 
2115 	/* Turn on RX data and RX BD initiator state machine */
2116 	CSR_WRITE_4(sc, BGE_RDBDI_MODE, BGE_RDBDIMODE_ENABLE);
2117 
2118 	/* Turn on Mbuf cluster free state machine */
2119 	if (!(BGE_IS_5705_OR_BEYOND(sc)))
2120 		CSR_WRITE_4(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
2121 
2122 	/* Turn on send BD completion state machine */
2123 	CSR_WRITE_4(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
2124 
2125 	/* Turn on send data completion state machine */
2126 	CSR_WRITE_4(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE);
2127 
2128 	/* Turn on send data initiator state machine */
2129 	if (BGE_IS_5750_OR_BEYOND(sc)) {
2130 		/* XXX: magic value from Linux driver */
2131 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE | 0x08);
2132 	} else {
2133 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
2134 	}
2135 
2136 	/* Turn on send BD initiator state machine */
2137 	CSR_WRITE_4(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
2138 
2139 	/* Turn on send BD selector state machine */
2140 	CSR_WRITE_4(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
2141 
2142 	CSR_WRITE_4(sc, BGE_SDI_STATS_ENABLE_MASK, 0x007FFFFF);
2143 	CSR_WRITE_4(sc, BGE_SDI_STATS_CTL,
2144 	    BGE_SDISTATSCTL_ENABLE | BGE_SDISTATSCTL_FASTER);
2145 
2146 	/* ack/clear link change events */
2147 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
2148 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
2149 	    BGE_MACSTAT_CFG_CHANGED);
2150 	CSR_WRITE_4(sc, BGE_MI_STS, 0);
2151 
2152 	/* Enable PHY auto polling (for MII/GMII only) */
2153 	if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
2154 		CSR_WRITE_4(sc, BGE_MI_STS, BGE_MISTS_LINK);
2155  	} else {
2156 		BGE_STS_SETBIT(sc, BGE_STS_AUTOPOLL);
2157 		BGE_SETBIT(sc, BGE_MI_MODE, BGE_MIMODE_AUTOPOLL | (10 << 16));
2158 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700)
2159 			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
2160 			    BGE_EVTENB_MI_INTERRUPT);
2161 	}
2162 
2163 	/*
2164 	 * Clear any pending link state attention.
2165 	 * Otherwise some link state change events may be lost until attention
2166 	 * is cleared by bge_intr() -> bge_link_upd() sequence.
2167 	 * It's not necessary on newer BCM chips - perhaps enabling link
2168 	 * state change attentions implies clearing pending attention.
2169 	 */
2170 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
2171 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
2172 	    BGE_MACSTAT_LINK_CHANGED);
2173 
2174 	/* Enable link state change attentions. */
2175 	BGE_SETBIT(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_LINK_CHANGED);
2176 
2177 	return (0);
2178 }
2179 
2180 static const struct bge_revision *
2181 bge_lookup_rev(uint32_t chipid)
2182 {
2183 	const struct bge_revision *br;
2184 
2185 	for (br = bge_revisions; br->br_name != NULL; br++) {
2186 		if (br->br_chipid == chipid)
2187 			return (br);
2188 	}
2189 
2190 	for (br = bge_majorrevs; br->br_name != NULL; br++) {
2191 		if (br->br_chipid == BGE_ASICREV(chipid))
2192 			return (br);
2193 	}
2194 
2195 	return (NULL);
2196 }
2197 
2198 static const struct bge_product *
2199 bge_lookup(const struct pci_attach_args *pa)
2200 {
2201 	const struct bge_product *bp;
2202 
2203 	for (bp = bge_products; bp->bp_name != NULL; bp++) {
2204 		if (PCI_VENDOR(pa->pa_id) == bp->bp_vendor &&
2205 		    PCI_PRODUCT(pa->pa_id) == bp->bp_product)
2206 			return (bp);
2207 	}
2208 
2209 	return (NULL);
2210 }
2211 
2212 static int
2213 bge_setpowerstate(struct bge_softc *sc, int powerlevel)
2214 {
2215 #ifdef NOTYET
2216 	u_int32_t pm_ctl = 0;
2217 
2218 	/* XXX FIXME: make sure indirect accesses enabled? */
2219 	pm_ctl = pci_conf_read(sc->bge_dev, BGE_PCI_MISC_CTL, 4);
2220 	pm_ctl |= BGE_PCIMISCCTL_INDIRECT_ACCESS;
2221 	pci_write_config(sc->bge_dev, BGE_PCI_MISC_CTL, pm_ctl, 4);
2222 
2223 	/* clear the PME_assert bit and power state bits, enable PME */
2224 	pm_ctl = pci_conf_read(sc->bge_dev, BGE_PCI_PWRMGMT_CMD, 2);
2225 	pm_ctl &= ~PCIM_PSTAT_DMASK;
2226 	pm_ctl |= (1 << 8);
2227 
2228 	if (powerlevel == 0) {
2229 		pm_ctl |= PCIM_PSTAT_D0;
2230 		pci_write_config(sc->bge_dev, BGE_PCI_PWRMGMT_CMD,
2231 		    pm_ctl, 2);
2232 		DELAY(10000);
2233 		CSR_WRITE_4(sc, BGE_MISC_LOCAL_CTL, sc->bge_local_ctrl_reg);
2234 		DELAY(10000);
2235 
2236 #ifdef NOTYET
2237 		/* XXX FIXME: write 0x02 to phy aux_Ctrl reg */
2238 		bge_miibus_writereg(sc->bge_dev, 1, 0x18, 0x02);
2239 #endif
2240 		DELAY(40); DELAY(40); DELAY(40);
2241 		DELAY(10000);	/* above not quite adequate on 5700 */
2242 		return 0;
2243 	}
2244 
2245 
2246 	/*
2247 	 * Entering ACPI power states D1-D3 is achieved by wiggling
2248 	 * GMII gpio pins. Example code assumes all hardware vendors
2249 	 * followed Broadom's sample pcb layout. Until we verify that
2250 	 * for all supported OEM cards, states D1-D3 are  unsupported.
2251 	 */
2252 	aprint_error_dev(sc->bge_dev,
2253 	    "power state %d unimplemented; check GPIO pins\n",
2254 	    powerlevel);
2255 #endif
2256 	return EOPNOTSUPP;
2257 }
2258 
2259 
2260 /*
2261  * Probe for a Broadcom chip. Check the PCI vendor and device IDs
2262  * against our list and return its name if we find a match. Note
2263  * that since the Broadcom controller contains VPD support, we
2264  * can get the device name string from the controller itself instead
2265  * of the compiled-in string. This is a little slow, but it guarantees
2266  * we'll always announce the right product name.
2267  */
2268 static int
2269 bge_probe(device_t parent, cfdata_t match, void *aux)
2270 {
2271 	struct pci_attach_args *pa = (struct pci_attach_args *)aux;
2272 
2273 	if (bge_lookup(pa) != NULL)
2274 		return (1);
2275 
2276 	return (0);
2277 }
2278 
2279 static void
2280 bge_attach(device_t parent, device_t self, void *aux)
2281 {
2282 	struct bge_softc	*sc = device_private(self);
2283 	struct pci_attach_args	*pa = aux;
2284 	prop_dictionary_t dict;
2285 	const struct bge_product *bp;
2286 	const struct bge_revision *br;
2287 	pci_chipset_tag_t	pc;
2288 	pci_intr_handle_t	ih;
2289 	const char		*intrstr = NULL;
2290 	bus_dma_segment_t	seg;
2291 	int			rseg;
2292 	u_int32_t		hwcfg = 0;
2293 	u_int32_t		command;
2294 	struct ifnet		*ifp;
2295 	u_int32_t		misccfg;
2296 	void *			kva;
2297 	u_char			eaddr[ETHER_ADDR_LEN];
2298 	pcireg_t		memtype;
2299 	bus_addr_t		memaddr;
2300 	bus_size_t		memsize;
2301 	u_int32_t		pm_ctl;
2302 
2303 	bp = bge_lookup(pa);
2304 	KASSERT(bp != NULL);
2305 
2306 	sc->sc_pc = pa->pa_pc;
2307 	sc->sc_pcitag = pa->pa_tag;
2308 	sc->bge_dev = self;
2309 
2310 	aprint_naive(": Ethernet controller\n");
2311 	aprint_normal(": %s\n", bp->bp_name);
2312 
2313 	/*
2314 	 * Map control/status registers.
2315 	 */
2316 	DPRINTFN(5, ("Map control/status regs\n"));
2317 	pc = sc->sc_pc;
2318 	command = pci_conf_read(pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG);
2319 	command |= PCI_COMMAND_MEM_ENABLE | PCI_COMMAND_MASTER_ENABLE;
2320 	pci_conf_write(pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG, command);
2321 	command = pci_conf_read(pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG);
2322 
2323 	if (!(command & PCI_COMMAND_MEM_ENABLE)) {
2324 		aprint_error_dev(sc->bge_dev,
2325 		    "failed to enable memory mapping!\n");
2326 		return;
2327 	}
2328 
2329 	DPRINTFN(5, ("pci_mem_find\n"));
2330 	memtype = pci_mapreg_type(sc->sc_pc, sc->sc_pcitag, BGE_PCI_BAR0);
2331  	switch (memtype) {
2332 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_32BIT:
2333 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_64BIT:
2334 		if (pci_mapreg_map(pa, BGE_PCI_BAR0,
2335 		    memtype, 0, &sc->bge_btag, &sc->bge_bhandle,
2336 		    &memaddr, &memsize) == 0)
2337 			break;
2338 	default:
2339 		aprint_error_dev(sc->bge_dev, "can't find mem space\n");
2340 		return;
2341 	}
2342 
2343 	DPRINTFN(5, ("pci_intr_map\n"));
2344 	if (pci_intr_map(pa, &ih)) {
2345 		aprint_error_dev(sc->bge_dev, "couldn't map interrupt\n");
2346 		return;
2347 	}
2348 
2349 	DPRINTFN(5, ("pci_intr_string\n"));
2350 	intrstr = pci_intr_string(pc, ih);
2351 
2352 	DPRINTFN(5, ("pci_intr_establish\n"));
2353 	sc->bge_intrhand = pci_intr_establish(pc, ih, IPL_NET, bge_intr, sc);
2354 
2355 	if (sc->bge_intrhand == NULL) {
2356 		aprint_error_dev(sc->bge_dev,
2357 		    "couldn't establish interrupt%s%s\n",
2358 		    intrstr ? " at " : "", intrstr ? intrstr : "");
2359 		return;
2360 	}
2361 	aprint_normal_dev(sc->bge_dev, "interrupting at %s\n", intrstr);
2362 
2363 	/*
2364 	 * Kludge for 5700 Bx bug: a hardware bug (PCIX byte enable?)
2365 	 * can clobber the chip's PCI config-space power control registers,
2366 	 * leaving the card in D3 powersave state.
2367 	 * We do not have memory-mapped registers in this state,
2368 	 * so force device into D0 state before starting initialization.
2369 	 */
2370 	pm_ctl = pci_conf_read(pc, sc->sc_pcitag, BGE_PCI_PWRMGMT_CMD);
2371 	pm_ctl &= ~(PCI_PWR_D0|PCI_PWR_D1|PCI_PWR_D2|PCI_PWR_D3);
2372 	pm_ctl |= (1 << 8) | PCI_PWR_D0 ; /* D0 state */
2373 	pci_conf_write(pc, sc->sc_pcitag, BGE_PCI_PWRMGMT_CMD, pm_ctl);
2374 	DELAY(1000);	/* 27 usec is allegedly sufficent */
2375 
2376 	/*
2377 	 * Save ASIC rev.
2378 	 */
2379 	sc->bge_chipid =
2380 	    pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL) &
2381 	    BGE_PCIMISCCTL_ASICREV;
2382 
2383 	/*
2384 	 * Detect PCI-Express devices
2385 	 * XXX: guessed from Linux/FreeBSD; no documentation
2386 	 */
2387 	if (pci_get_capability(sc->sc_pc, sc->sc_pcitag, PCI_CAP_PCIEXPRESS,
2388 	        NULL, NULL) != 0)
2389 		sc->bge_flags |= BGE_PCIE;
2390 
2391 	/*
2392 	 * PCI-X check.
2393 	 */
2394 	if ((pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_PCISTATE) &
2395 		BGE_PCISTATE_PCI_BUSMODE) == 0)
2396 		sc->bge_flags |= BGE_PCIX;
2397 
2398 	if (sc->bge_chipid == BGE_CHIPID_BCM5701_A0 ||
2399 	    sc->bge_chipid == BGE_CHIPID_BCM5701_B0)
2400 		sc->bge_flags |= BGE_PHY_CRC_BUG;
2401 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5703_AX ||
2402 	    BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5704_AX)
2403 		sc->bge_flags |= BGE_PHY_ADC_BUG;
2404 	if (sc->bge_chipid == BGE_CHIPID_BCM5704_A0)
2405 		sc->bge_flags |= BGE_PHY_5704_A0_BUG;
2406 
2407 	if (BGE_IS_5705_OR_BEYOND(sc)) {
2408 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 ||
2409 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5787) {
2410 			if (PCI_PRODUCT(pa->pa_id) != PCI_PRODUCT_BROADCOM_BCM5722 &&
2411 			    PCI_PRODUCT(pa->pa_id) != PCI_PRODUCT_BROADCOM_BCM5756)
2412 				sc->bge_flags |= BGE_PHY_JITTER_BUG;
2413 			if (PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5755M)
2414 				sc->bge_flags |= BGE_PHY_ADJUST_TRIM;
2415 		} else if (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5906)
2416 			sc->bge_flags |= BGE_PHY_BER_BUG;
2417 	}
2418 
2419 	/* Try to reset the chip. */
2420 	DPRINTFN(5, ("bge_reset\n"));
2421 	bge_reset(sc);
2422 
2423 	if (bge_chipinit(sc)) {
2424 		aprint_error_dev(sc->bge_dev, "chip initialization failed\n");
2425 		bge_release_resources(sc);
2426 		return;
2427 	}
2428 
2429 	/*
2430 	 * Get station address from the EEPROM.
2431 	 */
2432 	if (bge_get_eaddr(sc, eaddr)) {
2433 		aprint_error_dev(sc->bge_dev,
2434 		"failed to reade station address\n");
2435 		bge_release_resources(sc);
2436 		return;
2437 	}
2438 
2439 	br = bge_lookup_rev(sc->bge_chipid);
2440 
2441 	if (br == NULL) {
2442 		aprint_normal_dev(sc->bge_dev, "unknown ASIC (0x%04x)",
2443 		    sc->bge_chipid >> 16);
2444 	} else {
2445 		aprint_normal_dev(sc->bge_dev, "ASIC %s (0x%04x)",
2446 		    br->br_name, sc->bge_chipid >> 16);
2447 	}
2448 	aprint_normal(", Ethernet address %s\n", ether_sprintf(eaddr));
2449 
2450 	/* Allocate the general information block and ring buffers. */
2451 	if (pci_dma64_available(pa))
2452 		sc->bge_dmatag = pa->pa_dmat64;
2453 	else
2454 		sc->bge_dmatag = pa->pa_dmat;
2455 	DPRINTFN(5, ("bus_dmamem_alloc\n"));
2456 	if (bus_dmamem_alloc(sc->bge_dmatag, sizeof(struct bge_ring_data),
2457 			     PAGE_SIZE, 0, &seg, 1, &rseg, BUS_DMA_NOWAIT)) {
2458 		aprint_error_dev(sc->bge_dev, "can't alloc rx buffers\n");
2459 		return;
2460 	}
2461 	DPRINTFN(5, ("bus_dmamem_map\n"));
2462 	if (bus_dmamem_map(sc->bge_dmatag, &seg, rseg,
2463 			   sizeof(struct bge_ring_data), &kva,
2464 			   BUS_DMA_NOWAIT)) {
2465 		aprint_error_dev(sc->bge_dev,
2466 		    "can't map DMA buffers (%zu bytes)\n",
2467 		    sizeof(struct bge_ring_data));
2468 		bus_dmamem_free(sc->bge_dmatag, &seg, rseg);
2469 		return;
2470 	}
2471 	DPRINTFN(5, ("bus_dmamem_create\n"));
2472 	if (bus_dmamap_create(sc->bge_dmatag, sizeof(struct bge_ring_data), 1,
2473 	    sizeof(struct bge_ring_data), 0,
2474 	    BUS_DMA_NOWAIT, &sc->bge_ring_map)) {
2475 		aprint_error_dev(sc->bge_dev, "can't create DMA map\n");
2476 		bus_dmamem_unmap(sc->bge_dmatag, kva,
2477 				 sizeof(struct bge_ring_data));
2478 		bus_dmamem_free(sc->bge_dmatag, &seg, rseg);
2479 		return;
2480 	}
2481 	DPRINTFN(5, ("bus_dmamem_load\n"));
2482 	if (bus_dmamap_load(sc->bge_dmatag, sc->bge_ring_map, kva,
2483 			    sizeof(struct bge_ring_data), NULL,
2484 			    BUS_DMA_NOWAIT)) {
2485 		bus_dmamap_destroy(sc->bge_dmatag, sc->bge_ring_map);
2486 		bus_dmamem_unmap(sc->bge_dmatag, kva,
2487 				 sizeof(struct bge_ring_data));
2488 		bus_dmamem_free(sc->bge_dmatag, &seg, rseg);
2489 		return;
2490 	}
2491 
2492 	DPRINTFN(5, ("bzero\n"));
2493 	sc->bge_rdata = (struct bge_ring_data *)kva;
2494 
2495 	memset(sc->bge_rdata, 0, sizeof(struct bge_ring_data));
2496 
2497 	/* Try to allocate memory for jumbo buffers. */
2498 	if (!(BGE_IS_5705_OR_BEYOND(sc))) {
2499 		if (bge_alloc_jumbo_mem(sc)) {
2500 			aprint_error_dev(sc->bge_dev,
2501 			    "jumbo buffer allocation failed\n");
2502 		} else
2503 			sc->ethercom.ec_capabilities |= ETHERCAP_JUMBO_MTU;
2504 	}
2505 
2506 	/* Set default tuneable values. */
2507 	sc->bge_stat_ticks = BGE_TICKS_PER_SEC;
2508 	sc->bge_rx_coal_ticks = 150;
2509 	sc->bge_rx_max_coal_bds = 64;
2510 #ifdef ORIG_WPAUL_VALUES
2511 	sc->bge_tx_coal_ticks = 150;
2512 	sc->bge_tx_max_coal_bds = 128;
2513 #else
2514 	sc->bge_tx_coal_ticks = 300;
2515 	sc->bge_tx_max_coal_bds = 400;
2516 #endif
2517 	if (BGE_IS_5705_OR_BEYOND(sc)) {
2518 		sc->bge_tx_coal_ticks = (12 * 5);
2519 		sc->bge_tx_max_coal_bds = (12 * 5);
2520 			aprint_verbose_dev(sc->bge_dev,
2521 			    "setting short Tx thresholds\n");
2522 	}
2523 
2524 	/* Set up ifnet structure */
2525 	ifp = &sc->ethercom.ec_if;
2526 	ifp->if_softc = sc;
2527 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2528 	ifp->if_ioctl = bge_ioctl;
2529 	ifp->if_stop = bge_stop;
2530 	ifp->if_start = bge_start;
2531 	ifp->if_init = bge_init;
2532 	ifp->if_watchdog = bge_watchdog;
2533 	IFQ_SET_MAXLEN(&ifp->if_snd, max(BGE_TX_RING_CNT - 1, IFQ_MAXLEN));
2534 	IFQ_SET_READY(&ifp->if_snd);
2535 	DPRINTFN(5, ("strcpy if_xname\n"));
2536 	strcpy(ifp->if_xname, device_xname(sc->bge_dev));
2537 
2538 	if (sc->bge_chipid != BGE_CHIPID_BCM5700_B0)
2539 		sc->ethercom.ec_if.if_capabilities |=
2540 		    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
2541 		    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
2542 		    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
2543 	sc->ethercom.ec_capabilities |=
2544 	    ETHERCAP_VLAN_HWTAGGING | ETHERCAP_VLAN_MTU;
2545 
2546 	if (sc->bge_flags & BGE_PCIE)
2547 		sc->ethercom.ec_if.if_capabilities |= IFCAP_TSOv4;
2548 
2549 	/*
2550 	 * Do MII setup.
2551 	 */
2552 	DPRINTFN(5, ("mii setup\n"));
2553 	sc->bge_mii.mii_ifp = ifp;
2554 	sc->bge_mii.mii_readreg = bge_miibus_readreg;
2555 	sc->bge_mii.mii_writereg = bge_miibus_writereg;
2556 	sc->bge_mii.mii_statchg = bge_miibus_statchg;
2557 
2558 	/*
2559 	 * Figure out what sort of media we have by checking the
2560 	 * hardware config word in the first 32k of NIC internal memory,
2561 	 * or fall back to the config word in the EEPROM. Note: on some BCM5700
2562 	 * cards, this value appears to be unset. If that's the
2563 	 * case, we have to rely on identifying the NIC by its PCI
2564 	 * subsystem ID, as we do below for the SysKonnect SK-9D41.
2565 	 */
2566 	if (bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_SIG) == BGE_MAGIC_NUMBER) {
2567 		hwcfg = bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_NICCFG);
2568 	} else {
2569 		bge_read_eeprom(sc, (void *)&hwcfg,
2570 		    BGE_EE_HWCFG_OFFSET, sizeof(hwcfg));
2571 		hwcfg = be32toh(hwcfg);
2572 	}
2573 	/* The SysKonnect SK-9D41 is a 1000baseSX card. */
2574 	if (PCI_PRODUCT(pa->pa_id) == SK_SUBSYSID_9D41 ||
2575 	    (hwcfg & BGE_HWCFG_MEDIA) == BGE_MEDIA_FIBER) {
2576 		if (BGE_IS_5714_FAMILY(sc))
2577 		    sc->bge_flags |= BGE_PHY_FIBER_MII;
2578 		else
2579 		    sc->bge_flags |= BGE_PHY_FIBER_TBI;
2580 	}
2581 
2582 	if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
2583 		ifmedia_init(&sc->bge_ifmedia, IFM_IMASK, bge_ifmedia_upd,
2584 		    bge_ifmedia_sts);
2585 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER|IFM_1000_SX, 0, NULL);
2586 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER|IFM_1000_SX|IFM_FDX,
2587 			    0, NULL);
2588 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER|IFM_AUTO, 0, NULL);
2589 		ifmedia_set(&sc->bge_ifmedia, IFM_ETHER|IFM_AUTO);
2590 		/* Pretend the user requested this setting */
2591 		sc->bge_ifmedia.ifm_media = sc->bge_ifmedia.ifm_cur->ifm_media;
2592 	} else {
2593 		/*
2594 		 * Do transceiver setup.
2595 		 */
2596 		ifmedia_init(&sc->bge_mii.mii_media, 0, bge_ifmedia_upd,
2597 			     bge_ifmedia_sts);
2598 		mii_attach(sc->bge_dev, &sc->bge_mii, 0xffffffff,
2599 			   MII_PHY_ANY, MII_OFFSET_ANY,
2600 			   MIIF_FORCEANEG|MIIF_DOPAUSE);
2601 
2602 		if (LIST_EMPTY(&sc->bge_mii.mii_phys)) {
2603 			aprint_error_dev(sc->bge_dev, "no PHY found!\n");
2604 			ifmedia_add(&sc->bge_mii.mii_media,
2605 				    IFM_ETHER|IFM_MANUAL, 0, NULL);
2606 			ifmedia_set(&sc->bge_mii.mii_media,
2607 				    IFM_ETHER|IFM_MANUAL);
2608 		} else
2609 			ifmedia_set(&sc->bge_mii.mii_media,
2610 				    IFM_ETHER|IFM_AUTO);
2611 	}
2612 
2613 	/*
2614 	 * When using the BCM5701 in PCI-X mode, data corruption has
2615 	 * been observed in the first few bytes of some received packets.
2616 	 * Aligning the packet buffer in memory eliminates the corruption.
2617 	 * Unfortunately, this misaligns the packet payloads.  On platforms
2618 	 * which do not support unaligned accesses, we will realign the
2619 	 * payloads by copying the received packets.
2620 	 */
2621 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5701 &&
2622 		sc->bge_flags & BGE_PCIX)
2623 		sc->bge_flags |= BGE_RX_ALIGNBUG;
2624 
2625 	misccfg = CSR_READ_4(sc, BGE_MISC_CFG);
2626 	misccfg &= BGE_MISCCFG_BOARD_ID_MASK;
2627 
2628 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5705 &&
2629 	    (misccfg == BGE_MISCCFG_BOARD_ID_5788 ||
2630 	     misccfg == BGE_MISCCFG_BOARD_ID_5788M))
2631 		sc->bge_flags |= BGE_IS_5788;
2632 
2633 	/*
2634 	 * Call MI attach routine.
2635 	 */
2636 	DPRINTFN(5, ("if_attach\n"));
2637 	if_attach(ifp);
2638 	DPRINTFN(5, ("ether_ifattach\n"));
2639 	ether_ifattach(ifp, eaddr);
2640 #if NRND > 0
2641 	rnd_attach_source(&sc->rnd_source, device_xname(sc->bge_dev),
2642 		RND_TYPE_NET, 0);
2643 #endif
2644 #ifdef BGE_EVENT_COUNTERS
2645 	/*
2646 	 * Attach event counters.
2647 	 */
2648 	evcnt_attach_dynamic(&sc->bge_ev_intr, EVCNT_TYPE_INTR,
2649 	    NULL, device_xname(sc->bge_dev), "intr");
2650 	evcnt_attach_dynamic(&sc->bge_ev_tx_xoff, EVCNT_TYPE_MISC,
2651 	    NULL, device_xname(sc->bge_dev), "tx_xoff");
2652 	evcnt_attach_dynamic(&sc->bge_ev_tx_xon, EVCNT_TYPE_MISC,
2653 	    NULL, device_xname(sc->bge_dev), "tx_xon");
2654 	evcnt_attach_dynamic(&sc->bge_ev_rx_xoff, EVCNT_TYPE_MISC,
2655 	    NULL, device_xname(sc->bge_dev), "rx_xoff");
2656 	evcnt_attach_dynamic(&sc->bge_ev_rx_xon, EVCNT_TYPE_MISC,
2657 	    NULL, device_xname(sc->bge_dev), "rx_xon");
2658 	evcnt_attach_dynamic(&sc->bge_ev_rx_macctl, EVCNT_TYPE_MISC,
2659 	    NULL, device_xname(sc->bge_dev), "rx_macctl");
2660 	evcnt_attach_dynamic(&sc->bge_ev_xoffentered, EVCNT_TYPE_MISC,
2661 	    NULL, device_xname(sc->bge_dev), "xoffentered");
2662 #endif /* BGE_EVENT_COUNTERS */
2663 	DPRINTFN(5, ("callout_init\n"));
2664 	callout_init(&sc->bge_timeout, 0);
2665 
2666 	if (!pmf_device_register(self, NULL, NULL))
2667 		aprint_error_dev(self, "couldn't establish power handler\n");
2668 	else
2669 		pmf_class_network_register(self, ifp);
2670 
2671 	dict = device_properties(self);
2672 	prop_dictionary_set_uint32(dict, "phyflags", sc->bge_flags);
2673 }
2674 
2675 static void
2676 bge_release_resources(struct bge_softc *sc)
2677 {
2678 	if (sc->bge_vpd_prodname != NULL)
2679 		free(sc->bge_vpd_prodname, M_DEVBUF);
2680 
2681 	if (sc->bge_vpd_readonly != NULL)
2682 		free(sc->bge_vpd_readonly, M_DEVBUF);
2683 }
2684 
2685 static void
2686 bge_reset(struct bge_softc *sc)
2687 {
2688 	u_int32_t cachesize, command, pcistate, new_pcistate;
2689 	int i, val;
2690 	void (*write_op)(struct bge_softc *, int, int);
2691 
2692 	if (BGE_IS_5750_OR_BEYOND(sc) && !BGE_IS_5714_FAMILY(sc) &&
2693 	    (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5906)) {
2694 	    	if (sc->bge_flags & BGE_PCIE) {
2695 			write_op = bge_writemem_direct;
2696 		} else {
2697 			write_op = bge_writemem_ind;
2698 		}
2699 	} else {
2700 		write_op = bge_writereg_ind;
2701 	}
2702 
2703 
2704 	/* Save some important PCI state. */
2705 	cachesize = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CACHESZ);
2706 	command = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD);
2707 	pcistate = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_PCISTATE);
2708 
2709 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL,
2710 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2711 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW);
2712 
2713 	/* Disable fastboot on controllers that support it. */
2714 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5752 ||
2715 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 ||
2716 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5787)
2717 		CSR_WRITE_4(sc, BGE_FASTBOOT_PC, 0);
2718 
2719 	val = BGE_MISCCFG_RESET_CORE_CLOCKS | (65<<1);
2720 	/*
2721 	 * XXX: from FreeBSD/Linux; no documentation
2722 	 */
2723 	if (sc->bge_flags & BGE_PCIE) {
2724 		if (CSR_READ_4(sc, BGE_PCIE_CTL1) == 0x60)
2725 			/* PCI Express 1.0 system */
2726 			CSR_WRITE_4(sc, BGE_PCIE_CTL1, 0x20);
2727 		if (sc->bge_chipid != BGE_CHIPID_BCM5750_A0) {
2728 			/*
2729 			 * Prevent PCI Express link training
2730 			 * during global reset.
2731 			 */
2732 			CSR_WRITE_4(sc, BGE_MISC_CFG, 1 << 29);
2733 			val |= (1<<29);
2734 		}
2735 	}
2736 
2737 	/*
2738 	 * Set GPHY Power Down Override to leave GPHY
2739 	 * powered up in D0 uninitialized.
2740 	 */
2741 	if (BGE_IS_5705_OR_BEYOND(sc))
2742 		val |= BGE_MISCCFG_KEEP_GPHY_POWER;
2743 
2744 	/* Issue global reset */
2745 	write_op(sc, BGE_MISC_CFG, val);
2746 
2747 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
2748 		i = CSR_READ_4(sc, BGE_VCPU_STATUS);
2749 		CSR_WRITE_4(sc, BGE_VCPU_STATUS,
2750 		    i | BGE_VCPU_STATUS_DRV_RESET);
2751 		i = CSR_READ_4(sc, BGE_VCPU_EXT_CTRL);
2752 		CSR_WRITE_4(sc, BGE_VCPU_EXT_CTRL,
2753 		    i & ~BGE_VCPU_EXT_CTRL_HALT_CPU);
2754 	}
2755 
2756 	DELAY(1000);
2757 
2758 	/*
2759 	 * XXX: from FreeBSD/Linux; no documentation
2760 	 */
2761 	if (sc->bge_flags & BGE_PCIE) {
2762 		if (sc->bge_chipid == BGE_CHIPID_BCM5750_A0) {
2763 			pcireg_t reg;
2764 
2765 			DELAY(500000);
2766 			/* XXX: Magic Numbers */
2767 			reg = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_UNKNOWN0);
2768 			pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_UNKNOWN0,
2769 			    reg | (1 << 15));
2770 		}
2771 		/*
2772 		 * XXX: Magic Numbers.
2773 		 * Sets maximal PCI-e payload and clears any PCI-e errors.
2774 		 * Should be replaced with references to PCI config-space
2775 		 * capability block for PCI-Express.
2776 		 */
2777 		pci_conf_write(sc->sc_pc, sc->sc_pcitag,
2778 		    BGE_PCI_CONF_DEV_CTRL, 0xf5000);
2779 
2780 	}
2781 
2782 	/* Reset some of the PCI state that got zapped by reset */
2783 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL,
2784 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2785 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW);
2786 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD, command);
2787 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CACHESZ, cachesize);
2788 	write_op(sc, BGE_MISC_CFG, (65 << 1));
2789 
2790 	/* Enable memory arbiter. */
2791 	{
2792 		uint32_t marbmode = 0;
2793 		if (BGE_IS_5714_FAMILY(sc)) {
2794 			marbmode = CSR_READ_4(sc, BGE_MARB_MODE);
2795 		}
2796  		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE | marbmode);
2797 	}
2798 
2799 
2800 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
2801 		for (i = 0; i < BGE_TIMEOUT; i++) {
2802 			val = CSR_READ_4(sc, BGE_VCPU_STATUS);
2803 			if (val & BGE_VCPU_STATUS_INIT_DONE)
2804 				break;
2805 			DELAY(100);
2806 		}
2807 		if (i == BGE_TIMEOUT) {
2808 			aprint_error_dev(sc->bge_dev, "reset timed out\n");
2809 			return;
2810 		}
2811 	} else {
2812 		/*
2813 		 * Write the magic number to the firmware mailbox at 0xb50
2814 		 * so that the driver can synchronize with the firmware.
2815 		 */
2816 		bge_writemem_ind(sc, BGE_SOFTWARE_GENCOMM, BGE_MAGIC_NUMBER);
2817 
2818 		/*
2819 		 * Poll the value location we just wrote until
2820 		 * we see the 1's complement of the magic number.
2821 		 * This indicates that the firmware initialization
2822 		 * is complete.
2823 		 */
2824 		for (i = 0; i < BGE_TIMEOUT; i++) {
2825 			val = bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM);
2826 			if (val == ~BGE_MAGIC_NUMBER)
2827 				break;
2828 			DELAY(1000);
2829 		}
2830 
2831 		if (i >= BGE_TIMEOUT) {
2832 			aprint_error_dev(sc->bge_dev,
2833 			    "firmware handshake timed out, val = %x\n", val);
2834 			/*
2835 			 * XXX: occasionally fired on bcm5721, but without
2836 			 * apparent harm.  For now, keep going if we timeout
2837 			 * against PCI-E devices.
2838 			 */
2839 			if ((sc->bge_flags & BGE_PCIE) == 0)
2840 				  return;
2841 		}
2842 	}
2843 
2844 	/*
2845 	 * XXX Wait for the value of the PCISTATE register to
2846 	 * return to its original pre-reset state. This is a
2847 	 * fairly good indicator of reset completion. If we don't
2848 	 * wait for the reset to fully complete, trying to read
2849 	 * from the device's non-PCI registers may yield garbage
2850 	 * results.
2851 	 */
2852 	for (i = 0; i < 10000; i++) {
2853 		new_pcistate = pci_conf_read(sc->sc_pc, sc->sc_pcitag,
2854 		    BGE_PCI_PCISTATE);
2855 		if ((new_pcistate & ~BGE_PCISTATE_RESERVED) ==
2856 		    (pcistate & ~BGE_PCISTATE_RESERVED))
2857 			break;
2858 		DELAY(10);
2859 	}
2860 	if ((new_pcistate & ~BGE_PCISTATE_RESERVED) !=
2861 	    (pcistate & ~BGE_PCISTATE_RESERVED)) {
2862 		aprint_error_dev(sc->bge_dev, "pcistate failed to revert\n");
2863 	}
2864 
2865 	/* Enable memory arbiter. */
2866 	/* XXX why do this twice? */
2867 	{
2868 		uint32_t marbmode = 0;
2869 		if (BGE_IS_5714_FAMILY(sc)) {
2870 			marbmode = CSR_READ_4(sc, BGE_MARB_MODE);
2871 		}
2872  		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE | marbmode);
2873 	}
2874 
2875 	/* Fix up byte swapping */
2876 	CSR_WRITE_4(sc, BGE_MODE_CTL, BGE_DMA_SWAP_OPTIONS);
2877 
2878 	CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
2879 
2880 	/*
2881 	 * The 5704 in TBI mode apparently needs some special
2882 	 * adjustment to insure the SERDES drive level is set
2883 	 * to 1.2V.
2884 	 */
2885 	if (sc->bge_flags & BGE_PHY_FIBER_TBI &&
2886 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704) {
2887 		u_int32_t serdescfg;
2888 
2889 		serdescfg = CSR_READ_4(sc, BGE_SERDES_CFG);
2890 		serdescfg = (serdescfg & ~0xFFF) | 0x880;
2891 		CSR_WRITE_4(sc, BGE_SERDES_CFG, serdescfg);
2892 	}
2893 
2894 	/* XXX: from FreeBSD/Linux; no documentation */
2895 	if (sc->bge_flags & BGE_PCIE &&
2896 	    sc->bge_chipid != BGE_CHIPID_BCM5750_A0)
2897 		CSR_WRITE_4(sc, BGE_PCIE_CTL0, CSR_READ_4(sc, BGE_PCIE_CTL0) | (1<<25));
2898 	DELAY(10000);
2899 }
2900 
2901 /*
2902  * Frame reception handling. This is called if there's a frame
2903  * on the receive return list.
2904  *
2905  * Note: we have to be able to handle two possibilities here:
2906  * 1) the frame is from the jumbo recieve ring
2907  * 2) the frame is from the standard receive ring
2908  */
2909 
2910 static void
2911 bge_rxeof(struct bge_softc *sc)
2912 {
2913 	struct ifnet *ifp;
2914 	int stdcnt = 0, jumbocnt = 0;
2915 	bus_dmamap_t dmamap;
2916 	bus_addr_t offset, toff;
2917 	bus_size_t tlen;
2918 	int tosync;
2919 
2920 	ifp = &sc->ethercom.ec_if;
2921 
2922 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
2923 	    offsetof(struct bge_ring_data, bge_status_block),
2924 	    sizeof (struct bge_status_block),
2925 	    BUS_DMASYNC_POSTREAD);
2926 
2927 	offset = offsetof(struct bge_ring_data, bge_rx_return_ring);
2928 	tosync = sc->bge_rdata->bge_status_block.bge_idx[0].bge_rx_prod_idx -
2929 	    sc->bge_rx_saved_considx;
2930 
2931 #if NRND > 0
2932 	if (tosync != 0 && RND_ENABLED(&sc->rnd_source))
2933 		rnd_add_uint32(&sc->rnd_source, tosync);
2934 #endif
2935 
2936 	toff = offset + (sc->bge_rx_saved_considx * sizeof (struct bge_rx_bd));
2937 
2938 	if (tosync < 0) {
2939 		tlen = (sc->bge_return_ring_cnt - sc->bge_rx_saved_considx) *
2940 		    sizeof (struct bge_rx_bd);
2941 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
2942 		    toff, tlen, BUS_DMASYNC_POSTREAD);
2943 		tosync = -tosync;
2944 	}
2945 
2946 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
2947 	    offset, tosync * sizeof (struct bge_rx_bd),
2948 	    BUS_DMASYNC_POSTREAD);
2949 
2950 	while(sc->bge_rx_saved_considx !=
2951 	    sc->bge_rdata->bge_status_block.bge_idx[0].bge_rx_prod_idx) {
2952 		struct bge_rx_bd	*cur_rx;
2953 		u_int32_t		rxidx;
2954 		struct mbuf		*m = NULL;
2955 
2956 		cur_rx = &sc->bge_rdata->
2957 			bge_rx_return_ring[sc->bge_rx_saved_considx];
2958 
2959 		rxidx = cur_rx->bge_idx;
2960 		BGE_INC(sc->bge_rx_saved_considx, sc->bge_return_ring_cnt);
2961 
2962 		if (cur_rx->bge_flags & BGE_RXBDFLAG_JUMBO_RING) {
2963 			BGE_INC(sc->bge_jumbo, BGE_JUMBO_RX_RING_CNT);
2964 			m = sc->bge_cdata.bge_rx_jumbo_chain[rxidx];
2965 			sc->bge_cdata.bge_rx_jumbo_chain[rxidx] = NULL;
2966 			jumbocnt++;
2967 			bus_dmamap_sync(sc->bge_dmatag,
2968 			    sc->bge_cdata.bge_rx_jumbo_map,
2969 			    mtod(m, char *) - (char *)sc->bge_cdata.bge_jumbo_buf,
2970 			    BGE_JLEN, BUS_DMASYNC_POSTREAD);
2971 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
2972 				ifp->if_ierrors++;
2973 				bge_newbuf_jumbo(sc, sc->bge_jumbo, m);
2974 				continue;
2975 			}
2976 			if (bge_newbuf_jumbo(sc, sc->bge_jumbo,
2977 					     NULL)== ENOBUFS) {
2978 				ifp->if_ierrors++;
2979 				bge_newbuf_jumbo(sc, sc->bge_jumbo, m);
2980 				continue;
2981 			}
2982 		} else {
2983 			BGE_INC(sc->bge_std, BGE_STD_RX_RING_CNT);
2984 			m = sc->bge_cdata.bge_rx_std_chain[rxidx];
2985 
2986 			sc->bge_cdata.bge_rx_std_chain[rxidx] = NULL;
2987 			stdcnt++;
2988 			dmamap = sc->bge_cdata.bge_rx_std_map[rxidx];
2989 			sc->bge_cdata.bge_rx_std_map[rxidx] = 0;
2990 			bus_dmamap_sync(sc->bge_dmatag, dmamap, 0,
2991 			    dmamap->dm_mapsize, BUS_DMASYNC_POSTREAD);
2992 			bus_dmamap_unload(sc->bge_dmatag, dmamap);
2993 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
2994 				ifp->if_ierrors++;
2995 				bge_newbuf_std(sc, sc->bge_std, m, dmamap);
2996 				continue;
2997 			}
2998 			if (bge_newbuf_std(sc, sc->bge_std,
2999 			    NULL, dmamap) == ENOBUFS) {
3000 				ifp->if_ierrors++;
3001 				bge_newbuf_std(sc, sc->bge_std, m, dmamap);
3002 				continue;
3003 			}
3004 		}
3005 
3006 		ifp->if_ipackets++;
3007 #ifndef __NO_STRICT_ALIGNMENT
3008                 /*
3009                  * XXX: if the 5701 PCIX-Rx-DMA workaround is in effect,
3010                  * the Rx buffer has the layer-2 header unaligned.
3011                  * If our CPU requires alignment, re-align by copying.
3012                  */
3013 		if (sc->bge_flags & BGE_RX_ALIGNBUG) {
3014 			memmove(mtod(m, char *) + ETHER_ALIGN, m->m_data,
3015                                 cur_rx->bge_len);
3016 			m->m_data += ETHER_ALIGN;
3017 		}
3018 #endif
3019 
3020 		m->m_pkthdr.len = m->m_len = cur_rx->bge_len - ETHER_CRC_LEN;
3021 		m->m_pkthdr.rcvif = ifp;
3022 
3023 #if NBPFILTER > 0
3024 		/*
3025 		 * Handle BPF listeners. Let the BPF user see the packet.
3026 		 */
3027 		if (ifp->if_bpf)
3028 			bpf_mtap(ifp->if_bpf, m);
3029 #endif
3030 
3031 		m->m_pkthdr.csum_flags = M_CSUM_IPv4;
3032 
3033 		if ((cur_rx->bge_ip_csum ^ 0xffff) != 0)
3034 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
3035 		/*
3036 		 * Rx transport checksum-offload may also
3037 		 * have bugs with packets which, when transmitted,
3038 		 * were `runts' requiring padding.
3039 		 */
3040 		if (cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM &&
3041 		    (/* (sc->_bge_quirks & BGE_QUIRK_SHORT_CKSUM_BUG) == 0 ||*/
3042 		     m->m_pkthdr.len >= ETHER_MIN_NOPAD)) {
3043 			m->m_pkthdr.csum_data =
3044 			    cur_rx->bge_tcp_udp_csum;
3045 			m->m_pkthdr.csum_flags |=
3046 			    (M_CSUM_TCPv4|M_CSUM_UDPv4|
3047 			     M_CSUM_DATA|M_CSUM_NO_PSEUDOHDR);
3048 		}
3049 
3050 		/*
3051 		 * If we received a packet with a vlan tag, pass it
3052 		 * to vlan_input() instead of ether_input().
3053 		 */
3054 		if (cur_rx->bge_flags & BGE_RXBDFLAG_VLAN_TAG) {
3055 			VLAN_INPUT_TAG(ifp, m, cur_rx->bge_vlan_tag, continue);
3056 		}
3057 
3058 		(*ifp->if_input)(ifp, m);
3059 	}
3060 
3061 	bge_writembx(sc, BGE_MBX_RX_CONS0_LO, sc->bge_rx_saved_considx);
3062 	if (stdcnt)
3063 		bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bge_std);
3064 	if (jumbocnt)
3065 		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
3066 }
3067 
3068 static void
3069 bge_txeof(struct bge_softc *sc)
3070 {
3071 	struct bge_tx_bd *cur_tx = NULL;
3072 	struct ifnet *ifp;
3073 	struct txdmamap_pool_entry *dma;
3074 	bus_addr_t offset, toff;
3075 	bus_size_t tlen;
3076 	int tosync;
3077 	struct mbuf *m;
3078 
3079 	ifp = &sc->ethercom.ec_if;
3080 
3081 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
3082 	    offsetof(struct bge_ring_data, bge_status_block),
3083 	    sizeof (struct bge_status_block),
3084 	    BUS_DMASYNC_POSTREAD);
3085 
3086 	offset = offsetof(struct bge_ring_data, bge_tx_ring);
3087 	tosync = sc->bge_rdata->bge_status_block.bge_idx[0].bge_tx_cons_idx -
3088 	    sc->bge_tx_saved_considx;
3089 
3090 #if NRND > 0
3091 	if (tosync != 0 && RND_ENABLED(&sc->rnd_source))
3092 		rnd_add_uint32(&sc->rnd_source, tosync);
3093 #endif
3094 
3095 	toff = offset + (sc->bge_tx_saved_considx * sizeof (struct bge_tx_bd));
3096 
3097 	if (tosync < 0) {
3098 		tlen = (BGE_TX_RING_CNT - sc->bge_tx_saved_considx) *
3099 		    sizeof (struct bge_tx_bd);
3100 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
3101 		    toff, tlen, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
3102 		tosync = -tosync;
3103 	}
3104 
3105 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
3106 	    offset, tosync * sizeof (struct bge_tx_bd),
3107 	    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
3108 
3109 	/*
3110 	 * Go through our tx ring and free mbufs for those
3111 	 * frames that have been sent.
3112 	 */
3113 	while (sc->bge_tx_saved_considx !=
3114 	    sc->bge_rdata->bge_status_block.bge_idx[0].bge_tx_cons_idx) {
3115 		u_int32_t		idx = 0;
3116 
3117 		idx = sc->bge_tx_saved_considx;
3118 		cur_tx = &sc->bge_rdata->bge_tx_ring[idx];
3119 		if (cur_tx->bge_flags & BGE_TXBDFLAG_END)
3120 			ifp->if_opackets++;
3121 		m = sc->bge_cdata.bge_tx_chain[idx];
3122 		if (m != NULL) {
3123 			sc->bge_cdata.bge_tx_chain[idx] = NULL;
3124 			dma = sc->txdma[idx];
3125 			bus_dmamap_sync(sc->bge_dmatag, dma->dmamap, 0,
3126 			    dma->dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
3127 			bus_dmamap_unload(sc->bge_dmatag, dma->dmamap);
3128 			SLIST_INSERT_HEAD(&sc->txdma_list, dma, link);
3129 			sc->txdma[idx] = NULL;
3130 
3131 			m_freem(m);
3132 		}
3133 		sc->bge_txcnt--;
3134 		BGE_INC(sc->bge_tx_saved_considx, BGE_TX_RING_CNT);
3135 		ifp->if_timer = 0;
3136 	}
3137 
3138 	if (cur_tx != NULL)
3139 		ifp->if_flags &= ~IFF_OACTIVE;
3140 }
3141 
3142 static int
3143 bge_intr(void *xsc)
3144 {
3145 	struct bge_softc *sc;
3146 	struct ifnet *ifp;
3147 	uint32_t statusword;
3148 
3149 	sc = xsc;
3150 	ifp = &sc->ethercom.ec_if;
3151 
3152 	/* It is possible for the interrupt to arrive before
3153 	 * the status block is updated prior to the interrupt.
3154 	 * Reading the PCI State register will confirm whether the
3155 	 * interrupt is ours and will flush the status block.
3156 	 */
3157 
3158 	/* read status word from status block */
3159 	statusword = sc->bge_rdata->bge_status_block.bge_status;
3160 
3161 	if ((statusword & BGE_STATFLAG_UPDATED) ||
3162 	    (!(CSR_READ_4(sc, BGE_PCI_PCISTATE) & BGE_PCISTATE_INTR_NOT_ACTIVE))) {
3163 		/* Ack interrupt and stop others from occuring. */
3164 		bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
3165 
3166 		BGE_EVCNT_INCR(sc->bge_ev_intr);
3167 
3168 		/* clear status word */
3169 		sc->bge_rdata->bge_status_block.bge_status = 0;
3170 
3171 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
3172 		    statusword & BGE_STATFLAG_LINKSTATE_CHANGED ||
3173 		    BGE_STS_BIT(sc, BGE_STS_LINK_EVT))
3174 			bge_link_upd(sc);
3175 
3176 		if (ifp->if_flags & IFF_RUNNING) {
3177 			/* Check RX return ring producer/consumer */
3178 			bge_rxeof(sc);
3179 
3180 			/* Check TX ring producer/consumer */
3181 			bge_txeof(sc);
3182 		}
3183 
3184 		if (sc->bge_pending_rxintr_change) {
3185 			uint32_t rx_ticks = sc->bge_rx_coal_ticks;
3186 			uint32_t rx_bds = sc->bge_rx_max_coal_bds;
3187 			uint32_t junk;
3188 
3189 			CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, rx_ticks);
3190 			DELAY(10);
3191 			junk = CSR_READ_4(sc, BGE_HCC_RX_COAL_TICKS);
3192 
3193 			CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, rx_bds);
3194 			DELAY(10);
3195 			junk = CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS);
3196 
3197 			sc->bge_pending_rxintr_change = 0;
3198 		}
3199 		bge_handle_events(sc);
3200 
3201 		/* Re-enable interrupts. */
3202 		bge_writembx(sc, BGE_MBX_IRQ0_LO, 0);
3203 
3204 		if (ifp->if_flags & IFF_RUNNING && !IFQ_IS_EMPTY(&ifp->if_snd))
3205 			bge_start(ifp);
3206 
3207 		return (1);
3208 	} else
3209 		return (0);
3210 }
3211 
3212 static void
3213 bge_tick(void *xsc)
3214 {
3215 	struct bge_softc *sc = xsc;
3216 	struct mii_data *mii = &sc->bge_mii;
3217 	int s;
3218 
3219 	s = splnet();
3220 
3221 	bge_stats_update(sc);
3222 
3223 	if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
3224 		/*
3225 		 * Since in TBI mode auto-polling can't be used we should poll
3226 		 * link status manually. Here we register pending link event
3227 		 * and trigger interrupt.
3228 		 */
3229 		BGE_STS_SETBIT(sc, BGE_STS_LINK_EVT);
3230 		BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
3231 	} else {
3232 		/*
3233 		 * Do not touch PHY if we have link up. This could break
3234 		 * IPMI/ASF mode or produce extra input errors.
3235 		 * (extra input errors was reported for bcm5701 & bcm5704).
3236 		 */
3237 		if (!BGE_STS_BIT(sc, BGE_STS_LINK))
3238 			mii_tick(mii);
3239 	}
3240 
3241 	callout_reset(&sc->bge_timeout, hz, bge_tick, sc);
3242 
3243 	splx(s);
3244 }
3245 
3246 static void
3247 bge_stats_update(struct bge_softc *sc)
3248 {
3249 	struct ifnet *ifp = &sc->ethercom.ec_if;
3250 	bus_size_t stats = BGE_MEMWIN_START + BGE_STATS_BLOCK;
3251 	bus_size_t rstats = BGE_RX_STATS;
3252 
3253 #define READ_RSTAT(sc, stats, stat) \
3254 	  CSR_READ_4(sc, stats + offsetof(struct bge_mac_stats_regs, stat))
3255 
3256 	if (BGE_IS_5705_OR_BEYOND(sc)) {
3257 		ifp->if_collisions +=
3258 		    READ_RSTAT(sc, rstats, dot3StatsSingleCollisionFrames) +
3259 		    READ_RSTAT(sc, rstats, dot3StatsMultipleCollisionFrames) +
3260 		    READ_RSTAT(sc, rstats, dot3StatsExcessiveCollisions) +
3261 		    READ_RSTAT(sc, rstats, dot3StatsLateCollisions);
3262 
3263 		BGE_EVCNT_ADD(sc->bge_ev_tx_xoff,
3264 			      READ_RSTAT(sc, rstats, outXoffSent));
3265 		BGE_EVCNT_ADD(sc->bge_ev_tx_xon,
3266 			      READ_RSTAT(sc, rstats, outXonSent));
3267 		BGE_EVCNT_ADD(sc->bge_ev_rx_xoff,
3268 			      READ_RSTAT(sc, rstats, xoffPauseFramesReceived));
3269 		BGE_EVCNT_ADD(sc->bge_ev_rx_xon,
3270 			      READ_RSTAT(sc, rstats, xonPauseFramesReceived));
3271 		BGE_EVCNT_ADD(sc->bge_ev_rx_macctl,
3272 			      READ_RSTAT(sc, rstats, macControlFramesReceived));
3273 		BGE_EVCNT_ADD(sc->bge_ev_xoffentered,
3274 			      READ_RSTAT(sc, rstats, xoffStateEntered));
3275 		return;
3276 	}
3277 
3278 #undef READ_RSTAT
3279 #define READ_STAT(sc, stats, stat) \
3280 	  CSR_READ_4(sc, stats + offsetof(struct bge_stats, stat))
3281 
3282 	ifp->if_collisions +=
3283 	  (READ_STAT(sc, stats, dot3StatsSingleCollisionFrames.bge_addr_lo) +
3284 	   READ_STAT(sc, stats, dot3StatsMultipleCollisionFrames.bge_addr_lo) +
3285 	   READ_STAT(sc, stats, dot3StatsExcessiveCollisions.bge_addr_lo) +
3286 	   READ_STAT(sc, stats, dot3StatsLateCollisions.bge_addr_lo)) -
3287 	  ifp->if_collisions;
3288 
3289 	BGE_EVCNT_UPD(sc->bge_ev_tx_xoff,
3290 		      READ_STAT(sc, stats, outXoffSent.bge_addr_lo));
3291 	BGE_EVCNT_UPD(sc->bge_ev_tx_xon,
3292 		      READ_STAT(sc, stats, outXonSent.bge_addr_lo));
3293 	BGE_EVCNT_UPD(sc->bge_ev_rx_xoff,
3294 		      READ_STAT(sc, stats,
3295 		      		xoffPauseFramesReceived.bge_addr_lo));
3296 	BGE_EVCNT_UPD(sc->bge_ev_rx_xon,
3297 		      READ_STAT(sc, stats, xonPauseFramesReceived.bge_addr_lo));
3298 	BGE_EVCNT_UPD(sc->bge_ev_rx_macctl,
3299 		      READ_STAT(sc, stats,
3300 		      		macControlFramesReceived.bge_addr_lo));
3301 	BGE_EVCNT_UPD(sc->bge_ev_xoffentered,
3302 		      READ_STAT(sc, stats, xoffStateEntered.bge_addr_lo));
3303 
3304 #undef READ_STAT
3305 
3306 #ifdef notdef
3307 	ifp->if_collisions +=
3308 	   (sc->bge_rdata->bge_info.bge_stats.dot3StatsSingleCollisionFrames +
3309 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsMultipleCollisionFrames +
3310 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsExcessiveCollisions +
3311 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsLateCollisions) -
3312 	   ifp->if_collisions;
3313 #endif
3314 }
3315 
3316 /*
3317  * Pad outbound frame to ETHER_MIN_NOPAD for an unusual reason.
3318  * The bge hardware will pad out Tx runts to ETHER_MIN_NOPAD,
3319  * but when such padded frames employ the  bge IP/TCP checksum offload,
3320  * the hardware checksum assist gives incorrect results (possibly
3321  * from incorporating its own padding into the UDP/TCP checksum; who knows).
3322  * If we pad such runts with zeros, the onboard checksum comes out correct.
3323  */
3324 static inline int
3325 bge_cksum_pad(struct mbuf *pkt)
3326 {
3327 	struct mbuf *last = NULL;
3328 	int padlen;
3329 
3330 	padlen = ETHER_MIN_NOPAD - pkt->m_pkthdr.len;
3331 
3332 	/* if there's only the packet-header and we can pad there, use it. */
3333 	if (pkt->m_pkthdr.len == pkt->m_len &&
3334 	    M_TRAILINGSPACE(pkt) >= padlen) {
3335 		last = pkt;
3336 	} else {
3337 		/*
3338 		 * Walk packet chain to find last mbuf. We will either
3339 		 * pad there, or append a new mbuf and pad it
3340 		 * (thus perhaps avoiding the bcm5700 dma-min bug).
3341 		 */
3342 		for (last = pkt; last->m_next != NULL; last = last->m_next) {
3343 	      	       continue; /* do nothing */
3344 		}
3345 
3346 		/* `last' now points to last in chain. */
3347 		if (M_TRAILINGSPACE(last) < padlen) {
3348 			/* Allocate new empty mbuf, pad it. Compact later. */
3349 			struct mbuf *n;
3350 			MGET(n, M_DONTWAIT, MT_DATA);
3351 			if (n == NULL)
3352 				return ENOBUFS;
3353 			n->m_len = 0;
3354 			last->m_next = n;
3355 			last = n;
3356 		}
3357 	}
3358 
3359 	KDASSERT(!M_READONLY(last));
3360 	KDASSERT(M_TRAILINGSPACE(last) >= padlen);
3361 
3362 	/* Now zero the pad area, to avoid the bge cksum-assist bug */
3363 	memset(mtod(last, char *) + last->m_len, 0, padlen);
3364 	last->m_len += padlen;
3365 	pkt->m_pkthdr.len += padlen;
3366 	return 0;
3367 }
3368 
3369 /*
3370  * Compact outbound packets to avoid bug with DMA segments less than 8 bytes.
3371  */
3372 static inline int
3373 bge_compact_dma_runt(struct mbuf *pkt)
3374 {
3375 	struct mbuf	*m, *prev;
3376 	int 		totlen, prevlen;
3377 
3378 	prev = NULL;
3379 	totlen = 0;
3380 	prevlen = -1;
3381 
3382 	for (m = pkt; m != NULL; prev = m,m = m->m_next) {
3383 		int mlen = m->m_len;
3384 		int shortfall = 8 - mlen ;
3385 
3386 		totlen += mlen;
3387 		if (mlen == 0) {
3388 			continue;
3389 		}
3390 		if (mlen >= 8)
3391 			continue;
3392 
3393 		/* If we get here, mbuf data is too small for DMA engine.
3394 		 * Try to fix by shuffling data to prev or next in chain.
3395 		 * If that fails, do a compacting deep-copy of the whole chain.
3396 		 */
3397 
3398 		/* Internal frag. If fits in prev, copy it there. */
3399 		if (prev && M_TRAILINGSPACE(prev) >= m->m_len) {
3400 		  	memcpy(prev->m_data + prev->m_len, m->m_data, mlen);
3401 			prev->m_len += mlen;
3402 			m->m_len = 0;
3403 			/* XXX stitch chain */
3404 			prev->m_next = m_free(m);
3405 			m = prev;
3406 			continue;
3407 		}
3408 		else if (m->m_next != NULL &&
3409 			     M_TRAILINGSPACE(m) >= shortfall &&
3410 			     m->m_next->m_len >= (8 + shortfall)) {
3411 		    /* m is writable and have enough data in next, pull up. */
3412 
3413 		  	memcpy(m->m_data + m->m_len, m->m_next->m_data,
3414 			    shortfall);
3415 			m->m_len += shortfall;
3416 			m->m_next->m_len -= shortfall;
3417 			m->m_next->m_data += shortfall;
3418 		}
3419 		else if (m->m_next == NULL || 1) {
3420 		  	/* Got a runt at the very end of the packet.
3421 			 * borrow data from the tail of the preceding mbuf and
3422 			 * update its length in-place. (The original data is still
3423 			 * valid, so we can do this even if prev is not writable.)
3424 			 */
3425 
3426 			/* if we'd make prev a runt, just move all of its data. */
3427 			KASSERT(prev != NULL /*, ("runt but null PREV")*/);
3428 			KASSERT(prev->m_len >= 8 /*, ("runt prev")*/);
3429 
3430 			if ((prev->m_len - shortfall) < 8)
3431 				shortfall = prev->m_len;
3432 
3433 #ifdef notyet	/* just do the safe slow thing for now */
3434 			if (!M_READONLY(m)) {
3435 				if (M_LEADINGSPACE(m) < shorfall) {
3436 					void *m_dat;
3437 					m_dat = (m->m_flags & M_PKTHDR) ?
3438 					  m->m_pktdat : m->dat;
3439 					memmove(m_dat, mtod(m, void*), m->m_len);
3440 					m->m_data = m_dat;
3441 				    }
3442 			} else
3443 #endif	/* just do the safe slow thing */
3444 			{
3445 				struct mbuf * n = NULL;
3446 				int newprevlen = prev->m_len - shortfall;
3447 
3448 				MGET(n, M_NOWAIT, MT_DATA);
3449 				if (n == NULL)
3450 				   return ENOBUFS;
3451 				KASSERT(m->m_len + shortfall < MLEN
3452 					/*,
3453 					  ("runt %d +prev %d too big\n", m->m_len, shortfall)*/);
3454 
3455 				/* first copy the data we're stealing from prev */
3456 				memcpy(n->m_data, prev->m_data + newprevlen,
3457 				    shortfall);
3458 
3459 				/* update prev->m_len accordingly */
3460 				prev->m_len -= shortfall;
3461 
3462 				/* copy data from runt m */
3463 				memcpy(n->m_data + shortfall, m->m_data,
3464 				    m->m_len);
3465 
3466 				/* n holds what we stole from prev, plus m */
3467 				n->m_len = shortfall + m->m_len;
3468 
3469 				/* stitch n into chain and free m */
3470 				n->m_next = m->m_next;
3471 				prev->m_next = n;
3472 				/* KASSERT(m->m_next == NULL); */
3473 				m->m_next = NULL;
3474 				m_free(m);
3475 				m = n;	/* for continuing loop */
3476 			}
3477 		}
3478 		prevlen = m->m_len;
3479 	}
3480 	return 0;
3481 }
3482 
3483 /*
3484  * Encapsulate an mbuf chain in the tx ring  by coupling the mbuf data
3485  * pointers to descriptors.
3486  */
3487 static int
3488 bge_encap(struct bge_softc *sc, struct mbuf *m_head, u_int32_t *txidx)
3489 {
3490 	struct bge_tx_bd	*f = NULL;
3491 	u_int32_t		frag, cur;
3492 	u_int16_t		csum_flags = 0;
3493 	u_int16_t		txbd_tso_flags = 0;
3494 	struct txdmamap_pool_entry *dma;
3495 	bus_dmamap_t dmamap;
3496 	int			i = 0;
3497 	struct m_tag		*mtag;
3498 	int			use_tso, maxsegsize, error;
3499 
3500 	cur = frag = *txidx;
3501 
3502 	if (m_head->m_pkthdr.csum_flags) {
3503 		if (m_head->m_pkthdr.csum_flags & M_CSUM_IPv4)
3504 			csum_flags |= BGE_TXBDFLAG_IP_CSUM;
3505 		if (m_head->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4))
3506 			csum_flags |= BGE_TXBDFLAG_TCP_UDP_CSUM;
3507 	}
3508 
3509 	/*
3510 	 * If we were asked to do an outboard checksum, and the NIC
3511 	 * has the bug where it sometimes adds in the Ethernet padding,
3512 	 * explicitly pad with zeros so the cksum will be correct either way.
3513 	 * (For now, do this for all chip versions, until newer
3514 	 * are confirmed to not require the workaround.)
3515 	 */
3516 	if ((csum_flags & BGE_TXBDFLAG_TCP_UDP_CSUM) == 0 ||
3517 #ifdef notyet
3518 	    (sc->bge_quirks & BGE_QUIRK_SHORT_CKSUM_BUG) == 0 ||
3519 #endif
3520 	    m_head->m_pkthdr.len >= ETHER_MIN_NOPAD)
3521 		goto check_dma_bug;
3522 
3523 	if (bge_cksum_pad(m_head) != 0) {
3524 	    return ENOBUFS;
3525 	}
3526 
3527 check_dma_bug:
3528 	if (!(BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX))
3529 		goto doit;
3530 
3531 	/*
3532 	 * bcm5700 Revision B silicon cannot handle DMA descriptors with
3533 	 * less than eight bytes.  If we encounter a teeny mbuf
3534 	 * at the end of a chain, we can pad.  Otherwise, copy.
3535 	 */
3536 	if (bge_compact_dma_runt(m_head) != 0)
3537 		return ENOBUFS;
3538 
3539 doit:
3540 	dma = SLIST_FIRST(&sc->txdma_list);
3541 	if (dma == NULL)
3542 		return ENOBUFS;
3543 	dmamap = dma->dmamap;
3544 
3545 	/*
3546 	 * Set up any necessary TSO state before we start packing...
3547 	 */
3548 	use_tso = (m_head->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0;
3549 	if (!use_tso) {
3550 		maxsegsize = 0;
3551 	} else {	/* TSO setup */
3552 		unsigned  mss;
3553 		struct ether_header *eh;
3554 		unsigned ip_tcp_hlen, iptcp_opt_words, tcp_seg_flags, offset;
3555 		struct mbuf * m0 = m_head;
3556 		struct ip *ip;
3557 		struct tcphdr *th;
3558 		int iphl, hlen;
3559 
3560 		/*
3561 		 * XXX It would be nice if the mbuf pkthdr had offset
3562 		 * fields for the protocol headers.
3563 		 */
3564 
3565 		eh = mtod(m0, struct ether_header *);
3566 		switch (htons(eh->ether_type)) {
3567 		case ETHERTYPE_IP:
3568 			offset = ETHER_HDR_LEN;
3569 			break;
3570 
3571 		case ETHERTYPE_VLAN:
3572 			offset = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
3573 			break;
3574 
3575 		default:
3576 			/*
3577 			 * Don't support this protocol or encapsulation.
3578 			 */
3579 			return (ENOBUFS);
3580 		}
3581 
3582 		/*
3583 		 * TCP/IP headers are in the first mbuf; we can do
3584 		 * this the easy way.
3585 		 */
3586 		iphl = M_CSUM_DATA_IPv4_IPHL(m0->m_pkthdr.csum_data);
3587 		hlen = iphl + offset;
3588 		if (__predict_false(m0->m_len <
3589 				    (hlen + sizeof(struct tcphdr)))) {
3590 
3591 			aprint_debug_dev(sc->bge_dev,
3592 			    "TSO: hard case m0->m_len == %d < ip/tcp hlen %zd,"
3593 			    "not handled yet\n",
3594 			     m0->m_len, hlen+ sizeof(struct tcphdr));
3595 #ifdef NOTYET
3596 			/*
3597 			 * XXX jonathan@NetBSD.org: untested.
3598 			 * how to force  this branch to be taken?
3599 			 */
3600 			BGE_EVCNT_INCR(&sc->sc_ev_txtsopain);
3601 
3602 			m_copydata(m0, offset, sizeof(ip), &ip);
3603 			m_copydata(m0, hlen, sizeof(th), &th);
3604 
3605 			ip.ip_len = 0;
3606 
3607 			m_copyback(m0, hlen + offsetof(struct ip, ip_len),
3608 			    sizeof(ip.ip_len), &ip.ip_len);
3609 
3610 			th.th_sum = in_cksum_phdr(ip.ip_src.s_addr,
3611 			    ip.ip_dst.s_addr, htons(IPPROTO_TCP));
3612 
3613 			m_copyback(m0, hlen + offsetof(struct tcphdr, th_sum),
3614 			    sizeof(th.th_sum), &th.th_sum);
3615 
3616 			hlen += th.th_off << 2;
3617 			iptcp_opt_words	= hlen;
3618 #else
3619 			/*
3620 			 * if_wm "hard" case not yet supported, can we not
3621 			 * mandate it out of existence?
3622 			 */
3623 			(void) ip; (void)th; (void) ip_tcp_hlen;
3624 
3625 			return ENOBUFS;
3626 #endif
3627 		} else {
3628 			ip = (struct ip *) (mtod(m0, char *) + offset);
3629 			th = (struct tcphdr *) (mtod(m0, char *) + hlen);
3630 			ip_tcp_hlen = iphl +  (th->th_off << 2);
3631 
3632 			/* Total IP/TCP options, in 32-bit words */
3633 			iptcp_opt_words = (ip_tcp_hlen
3634 					   - sizeof(struct tcphdr)
3635 					   - sizeof(struct ip)) >> 2;
3636 		}
3637 		if (BGE_IS_5750_OR_BEYOND(sc)) {
3638 			th->th_sum = 0;
3639 			csum_flags &= ~(BGE_TXBDFLAG_TCP_UDP_CSUM);
3640 		} else {
3641 			/*
3642 			 * XXX jonathan@NetBSD.org: 5705 untested.
3643 			 * Requires TSO firmware patch for 5701/5703/5704.
3644 			 */
3645 			th->th_sum = in_cksum_phdr(ip->ip_src.s_addr,
3646 			    ip->ip_dst.s_addr, htons(IPPROTO_TCP));
3647 		}
3648 
3649 		mss = m_head->m_pkthdr.segsz;
3650 		txbd_tso_flags |=
3651 		    BGE_TXBDFLAG_CPU_PRE_DMA |
3652 		    BGE_TXBDFLAG_CPU_POST_DMA;
3653 
3654 		/*
3655 		 * Our NIC TSO-assist assumes TSO has standard, optionless
3656 		 * IPv4 and TCP headers, which total 40 bytes. By default,
3657 		 * the NIC copies 40 bytes of IP/TCP header from the
3658 		 * supplied header into the IP/TCP header portion of
3659 		 * each post-TSO-segment. If the supplied packet has IP or
3660 		 * TCP options, we need to tell the NIC to copy those extra
3661 		 * bytes into each  post-TSO header, in addition to the normal
3662 		 * 40-byte IP/TCP header (and to leave space accordingly).
3663 		 * Unfortunately, the driver encoding of option length
3664 		 * varies across different ASIC families.
3665 		 */
3666 		tcp_seg_flags = 0;
3667 		if (iptcp_opt_words) {
3668 			if ( BGE_IS_5705_OR_BEYOND(sc)) {
3669 				tcp_seg_flags =
3670 					iptcp_opt_words << 11;
3671 			} else {
3672 				txbd_tso_flags |=
3673 					iptcp_opt_words << 12;
3674 			}
3675 		}
3676 		maxsegsize = mss | tcp_seg_flags;
3677 		ip->ip_len = htons(mss + ip_tcp_hlen);
3678 
3679 	}	/* TSO setup */
3680 
3681 	/*
3682 	 * Start packing the mbufs in this chain into
3683 	 * the fragment pointers. Stop when we run out
3684 	 * of fragments or hit the end of the mbuf chain.
3685 	 */
3686 	error = bus_dmamap_load_mbuf(sc->bge_dmatag, dmamap, m_head,
3687 	    BUS_DMA_NOWAIT);
3688 	if (error) {
3689 		return (ENOBUFS);
3690 	}
3691 	/*
3692 	 * Sanity check: avoid coming within 16 descriptors
3693 	 * of the end of the ring.
3694 	 */
3695 	if (dmamap->dm_nsegs > (BGE_TX_RING_CNT - sc->bge_txcnt - 16)) {
3696 		BGE_TSO_PRINTF(("%s: "
3697 		    " dmamap_load_mbuf too close to ring wrap\n",
3698 		    device_xname(sc->bge_dev)));
3699 		goto fail_unload;
3700 	}
3701 
3702 	mtag = sc->ethercom.ec_nvlans ?
3703 	    m_tag_find(m_head, PACKET_TAG_VLAN, NULL) : NULL;
3704 
3705 
3706 	/* Iterate over dmap-map fragments. */
3707 	for (i = 0; i < dmamap->dm_nsegs; i++) {
3708 		f = &sc->bge_rdata->bge_tx_ring[frag];
3709 		if (sc->bge_cdata.bge_tx_chain[frag] != NULL)
3710 			break;
3711 
3712 		bge_set_hostaddr(&f->bge_addr, dmamap->dm_segs[i].ds_addr);
3713 		f->bge_len = dmamap->dm_segs[i].ds_len;
3714 
3715 		/*
3716 		 * For 5751 and follow-ons, for TSO we must turn
3717 		 * off checksum-assist flag in the tx-descr, and
3718 		 * supply the ASIC-revision-specific encoding
3719 		 * of TSO flags and segsize.
3720 		 */
3721 		if (use_tso) {
3722 			if (BGE_IS_5750_OR_BEYOND(sc) || i == 0) {
3723 				f->bge_rsvd = maxsegsize;
3724 				f->bge_flags = csum_flags | txbd_tso_flags;
3725 			} else {
3726 				f->bge_rsvd = 0;
3727 				f->bge_flags =
3728 				  (csum_flags | txbd_tso_flags) & 0x0fff;
3729 			}
3730 		} else {
3731 			f->bge_rsvd = 0;
3732 			f->bge_flags = csum_flags;
3733 		}
3734 
3735 		if (mtag != NULL) {
3736 			f->bge_flags |= BGE_TXBDFLAG_VLAN_TAG;
3737 			f->bge_vlan_tag = VLAN_TAG_VALUE(mtag);
3738 		} else {
3739 			f->bge_vlan_tag = 0;
3740 		}
3741 		cur = frag;
3742 		BGE_INC(frag, BGE_TX_RING_CNT);
3743 	}
3744 
3745 	if (i < dmamap->dm_nsegs) {
3746 		BGE_TSO_PRINTF(("%s: reached %d < dm_nsegs %d\n",
3747 		    device_xname(sc->bge_dev), i, dmamap->dm_nsegs));
3748 		goto fail_unload;
3749 	}
3750 
3751 	bus_dmamap_sync(sc->bge_dmatag, dmamap, 0, dmamap->dm_mapsize,
3752 	    BUS_DMASYNC_PREWRITE);
3753 
3754 	if (frag == sc->bge_tx_saved_considx) {
3755 		BGE_TSO_PRINTF(("%s: frag %d = wrapped id %d?\n",
3756 		    device_xname(sc->bge_dev), frag, sc->bge_tx_saved_considx));
3757 
3758 		goto fail_unload;
3759 	}
3760 
3761 	sc->bge_rdata->bge_tx_ring[cur].bge_flags |= BGE_TXBDFLAG_END;
3762 	sc->bge_cdata.bge_tx_chain[cur] = m_head;
3763 	SLIST_REMOVE_HEAD(&sc->txdma_list, link);
3764 	sc->txdma[cur] = dma;
3765 	sc->bge_txcnt += dmamap->dm_nsegs;
3766 
3767 	*txidx = frag;
3768 
3769 	return (0);
3770 
3771 fail_unload:
3772 	bus_dmamap_unload(sc->bge_dmatag, dmamap);
3773 
3774 	return ENOBUFS;
3775 }
3776 
3777 /*
3778  * Main transmit routine. To avoid having to do mbuf copies, we put pointers
3779  * to the mbuf data regions directly in the transmit descriptors.
3780  */
3781 static void
3782 bge_start(struct ifnet *ifp)
3783 {
3784 	struct bge_softc *sc;
3785 	struct mbuf *m_head = NULL;
3786 	u_int32_t prodidx;
3787 	int pkts = 0;
3788 
3789 	sc = ifp->if_softc;
3790 
3791 	if ((ifp->if_flags & (IFF_RUNNING|IFF_OACTIVE)) != IFF_RUNNING)
3792 		return;
3793 
3794 	prodidx = sc->bge_tx_prodidx;
3795 
3796 	while(sc->bge_cdata.bge_tx_chain[prodidx] == NULL) {
3797 		IFQ_POLL(&ifp->if_snd, m_head);
3798 		if (m_head == NULL)
3799 			break;
3800 
3801 #if 0
3802 		/*
3803 		 * XXX
3804 		 * safety overkill.  If this is a fragmented packet chain
3805 		 * with delayed TCP/UDP checksums, then only encapsulate
3806 		 * it if we have enough descriptors to handle the entire
3807 		 * chain at once.
3808 		 * (paranoia -- may not actually be needed)
3809 		 */
3810 		if (m_head->m_flags & M_FIRSTFRAG &&
3811 		    m_head->m_pkthdr.csum_flags & (CSUM_DELAY_DATA)) {
3812 			if ((BGE_TX_RING_CNT - sc->bge_txcnt) <
3813 			    M_CSUM_DATA_IPv4_OFFSET(m_head->m_pkthdr.csum_data) + 16) {
3814 				ifp->if_flags |= IFF_OACTIVE;
3815 				break;
3816 			}
3817 		}
3818 #endif
3819 
3820 		/*
3821 		 * Pack the data into the transmit ring. If we
3822 		 * don't have room, set the OACTIVE flag and wait
3823 		 * for the NIC to drain the ring.
3824 		 */
3825 		if (bge_encap(sc, m_head, &prodidx)) {
3826 			ifp->if_flags |= IFF_OACTIVE;
3827 			break;
3828 		}
3829 
3830 		/* now we are committed to transmit the packet */
3831 		IFQ_DEQUEUE(&ifp->if_snd, m_head);
3832 		pkts++;
3833 
3834 #if NBPFILTER > 0
3835 		/*
3836 		 * If there's a BPF listener, bounce a copy of this frame
3837 		 * to him.
3838 		 */
3839 		if (ifp->if_bpf)
3840 			bpf_mtap(ifp->if_bpf, m_head);
3841 #endif
3842 	}
3843 	if (pkts == 0)
3844 		return;
3845 
3846 	/* Transmit */
3847 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
3848 	/* 5700 b2 errata */
3849 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX)
3850 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
3851 
3852 	sc->bge_tx_prodidx = prodidx;
3853 
3854 	/*
3855 	 * Set a timeout in case the chip goes out to lunch.
3856 	 */
3857 	ifp->if_timer = 5;
3858 }
3859 
3860 static int
3861 bge_init(struct ifnet *ifp)
3862 {
3863 	struct bge_softc *sc = ifp->if_softc;
3864 	const u_int16_t *m;
3865 	int s, error = 0;
3866 
3867 	s = splnet();
3868 
3869 	ifp = &sc->ethercom.ec_if;
3870 
3871 	/* Cancel pending I/O and flush buffers. */
3872 	bge_stop(ifp, 0);
3873 	bge_reset(sc);
3874 	bge_chipinit(sc);
3875 
3876 	/*
3877 	 * Init the various state machines, ring
3878 	 * control blocks and firmware.
3879 	 */
3880 	error = bge_blockinit(sc);
3881 	if (error != 0) {
3882 		aprint_error_dev(sc->bge_dev, "initialization error %d\n",
3883 		    error);
3884 		splx(s);
3885 		return error;
3886 	}
3887 
3888 	ifp = &sc->ethercom.ec_if;
3889 
3890 	/* Specify MTU. */
3891 	CSR_WRITE_4(sc, BGE_RX_MTU, ifp->if_mtu +
3892 	    ETHER_HDR_LEN + ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN);
3893 
3894 	/* Load our MAC address. */
3895 	m = (const u_int16_t *)&(CLLADDR(ifp->if_sadl)[0]);
3896 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_LO, htons(m[0]));
3897 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_HI, (htons(m[1]) << 16) | htons(m[2]));
3898 
3899 	/* Enable or disable promiscuous mode as needed. */
3900 	if (ifp->if_flags & IFF_PROMISC) {
3901 		BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
3902 	} else {
3903 		BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
3904 	}
3905 
3906 	/* Program multicast filter. */
3907 	bge_setmulti(sc);
3908 
3909 	/* Init RX ring. */
3910 	bge_init_rx_ring_std(sc);
3911 
3912 	/*
3913 	 * Workaround for a bug in 5705 ASIC rev A0. Poll the NIC's
3914 	 * memory to insure that the chip has in fact read the first
3915 	 * entry of the ring.
3916 	 */
3917 	if (sc->bge_chipid == BGE_CHIPID_BCM5705_A0) {
3918 		u_int32_t		v, i;
3919 		for (i = 0; i < 10; i++) {
3920 			DELAY(20);
3921 			v = bge_readmem_ind(sc, BGE_STD_RX_RINGS + 8);
3922 			if (v == (MCLBYTES - ETHER_ALIGN))
3923 				break;
3924 		}
3925 		if (i == 10)
3926 			aprint_error_dev(sc->bge_dev,
3927 			    "5705 A0 chip failed to load RX ring\n");
3928 	}
3929 
3930 	/* Init jumbo RX ring. */
3931 	if (ifp->if_mtu > (ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN))
3932 		bge_init_rx_ring_jumbo(sc);
3933 
3934 	/* Init our RX return ring index */
3935 	sc->bge_rx_saved_considx = 0;
3936 
3937 	/* Init TX ring. */
3938 	bge_init_tx_ring(sc);
3939 
3940 	/* Turn on transmitter */
3941 	BGE_SETBIT(sc, BGE_TX_MODE, BGE_TXMODE_ENABLE);
3942 
3943 	/* Turn on receiver */
3944 	BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
3945 
3946 	CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 2);
3947 
3948 	/* Tell firmware we're alive. */
3949 	BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3950 
3951 	/* Enable host interrupts. */
3952 	BGE_SETBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_CLEAR_INTA);
3953 	BGE_CLRBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR);
3954 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 0);
3955 
3956 	if ((error = bge_ifmedia_upd(ifp)) != 0)
3957 		goto out;
3958 
3959 	ifp->if_flags |= IFF_RUNNING;
3960 	ifp->if_flags &= ~IFF_OACTIVE;
3961 
3962 	callout_reset(&sc->bge_timeout, hz, bge_tick, sc);
3963 
3964 out:
3965 	splx(s);
3966 
3967 	return error;
3968 }
3969 
3970 /*
3971  * Set media options.
3972  */
3973 static int
3974 bge_ifmedia_upd(struct ifnet *ifp)
3975 {
3976 	struct bge_softc *sc = ifp->if_softc;
3977 	struct mii_data *mii = &sc->bge_mii;
3978 	struct ifmedia *ifm = &sc->bge_ifmedia;
3979 	int rc;
3980 
3981 	/* If this is a 1000baseX NIC, enable the TBI port. */
3982 	if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
3983 		if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
3984 			return (EINVAL);
3985 		switch(IFM_SUBTYPE(ifm->ifm_media)) {
3986 		case IFM_AUTO:
3987 			/*
3988 			 * The BCM5704 ASIC appears to have a special
3989 			 * mechanism for programming the autoneg
3990 			 * advertisement registers in TBI mode.
3991 			 */
3992 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704) {
3993 				u_int32_t sgdig;
3994 				sgdig = CSR_READ_4(sc, BGE_SGDIG_STS);
3995 				if (sgdig & BGE_SGDIGSTS_DONE) {
3996 					CSR_WRITE_4(sc, BGE_TX_TBI_AUTONEG, 0);
3997 					sgdig = CSR_READ_4(sc, BGE_SGDIG_CFG);
3998 					sgdig |= BGE_SGDIGCFG_AUTO |
3999 					    BGE_SGDIGCFG_PAUSE_CAP |
4000 					    BGE_SGDIGCFG_ASYM_PAUSE;
4001 					CSR_WRITE_4(sc, BGE_SGDIG_CFG,
4002 					    sgdig | BGE_SGDIGCFG_SEND);
4003 					DELAY(5);
4004 					CSR_WRITE_4(sc, BGE_SGDIG_CFG, sgdig);
4005 				}
4006 			}
4007 			break;
4008 		case IFM_1000_SX:
4009 			if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX) {
4010 				BGE_CLRBIT(sc, BGE_MAC_MODE,
4011 				    BGE_MACMODE_HALF_DUPLEX);
4012 			} else {
4013 				BGE_SETBIT(sc, BGE_MAC_MODE,
4014 				    BGE_MACMODE_HALF_DUPLEX);
4015 			}
4016 			break;
4017 		default:
4018 			return (EINVAL);
4019 		}
4020 		/* XXX 802.3x flow control for 1000BASE-SX */
4021 		return (0);
4022 	}
4023 
4024 	BGE_STS_SETBIT(sc, BGE_STS_LINK_EVT);
4025 	if ((rc = mii_mediachg(mii)) == ENXIO)
4026 		return 0;
4027 
4028 	/*
4029 	 * Force an interrupt so that we will call bge_link_upd
4030 	 * if needed and clear any pending link state attention.
4031 	 * Without this we are not getting any further interrupts
4032 	 * for link state changes and thus will not UP the link and
4033 	 * not be able to send in bge_start. The only way to get
4034 	 * things working was to receive a packet and get a RX intr.
4035 	 */
4036 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
4037 	    sc->bge_flags & BGE_IS_5788)
4038 		BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
4039 	else
4040 		BGE_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
4041 
4042 	return rc;
4043 }
4044 
4045 /*
4046  * Report current media status.
4047  */
4048 static void
4049 bge_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
4050 {
4051 	struct bge_softc *sc = ifp->if_softc;
4052 	struct mii_data *mii = &sc->bge_mii;
4053 
4054 	if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
4055 		ifmr->ifm_status = IFM_AVALID;
4056 		ifmr->ifm_active = IFM_ETHER;
4057 		if (CSR_READ_4(sc, BGE_MAC_STS) &
4058 		    BGE_MACSTAT_TBI_PCS_SYNCHED)
4059 			ifmr->ifm_status |= IFM_ACTIVE;
4060 		ifmr->ifm_active |= IFM_1000_SX;
4061 		if (CSR_READ_4(sc, BGE_MAC_MODE) & BGE_MACMODE_HALF_DUPLEX)
4062 			ifmr->ifm_active |= IFM_HDX;
4063 		else
4064 			ifmr->ifm_active |= IFM_FDX;
4065 		return;
4066 	}
4067 
4068 	mii_pollstat(mii);
4069 	ifmr->ifm_status = mii->mii_media_status;
4070 	ifmr->ifm_active = (mii->mii_media_active & ~IFM_ETH_FMASK) |
4071 	    sc->bge_flowflags;
4072 }
4073 
4074 static int
4075 bge_ioctl(struct ifnet *ifp, u_long command, void *data)
4076 {
4077 	struct bge_softc *sc = ifp->if_softc;
4078 	struct ifreq *ifr = (struct ifreq *) data;
4079 	int s, error = 0;
4080 	struct mii_data *mii;
4081 
4082 	s = splnet();
4083 
4084 	switch(command) {
4085 	case SIOCSIFFLAGS:
4086 		if ((error = ifioctl_common(ifp, command, data)) != 0)
4087 			break;
4088 		if (ifp->if_flags & IFF_UP) {
4089 			/*
4090 			 * If only the state of the PROMISC flag changed,
4091 			 * then just use the 'set promisc mode' command
4092 			 * instead of reinitializing the entire NIC. Doing
4093 			 * a full re-init means reloading the firmware and
4094 			 * waiting for it to start up, which may take a
4095 			 * second or two.
4096 			 */
4097 			if (ifp->if_flags & IFF_RUNNING &&
4098 			    ifp->if_flags & IFF_PROMISC &&
4099 			    !(sc->bge_if_flags & IFF_PROMISC)) {
4100 				BGE_SETBIT(sc, BGE_RX_MODE,
4101 				    BGE_RXMODE_RX_PROMISC);
4102 			} else if (ifp->if_flags & IFF_RUNNING &&
4103 			    !(ifp->if_flags & IFF_PROMISC) &&
4104 			    sc->bge_if_flags & IFF_PROMISC) {
4105 				BGE_CLRBIT(sc, BGE_RX_MODE,
4106 				    BGE_RXMODE_RX_PROMISC);
4107 			} else if (!(sc->bge_if_flags & IFF_UP))
4108 				bge_init(ifp);
4109 		} else {
4110 			if (ifp->if_flags & IFF_RUNNING)
4111 				bge_stop(ifp, 1);
4112 		}
4113 		sc->bge_if_flags = ifp->if_flags;
4114 		error = 0;
4115 		break;
4116 	case SIOCSIFMEDIA:
4117 		/* XXX Flow control is not supported for 1000BASE-SX */
4118 		if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
4119 			ifr->ifr_media &= ~IFM_ETH_FMASK;
4120 			sc->bge_flowflags = 0;
4121 		}
4122 
4123 		/* Flow control requires full-duplex mode. */
4124 		if (IFM_SUBTYPE(ifr->ifr_media) == IFM_AUTO ||
4125 		    (ifr->ifr_media & IFM_FDX) == 0) {
4126 		    	ifr->ifr_media &= ~IFM_ETH_FMASK;
4127 		}
4128 		if (IFM_SUBTYPE(ifr->ifr_media) != IFM_AUTO) {
4129 			if ((ifr->ifr_media & IFM_ETH_FMASK) == IFM_FLOW) {
4130 				/* We can do both TXPAUSE and RXPAUSE. */
4131 				ifr->ifr_media |=
4132 				    IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE;
4133 			}
4134 			sc->bge_flowflags = ifr->ifr_media & IFM_ETH_FMASK;
4135 		}
4136 		/* FALLTHROUGH */
4137 	case SIOCGIFMEDIA:
4138 		if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
4139 			error = ifmedia_ioctl(ifp, ifr, &sc->bge_ifmedia,
4140 			    command);
4141 		} else {
4142 			mii = &sc->bge_mii;
4143 			error = ifmedia_ioctl(ifp, ifr, &mii->mii_media,
4144 			    command);
4145 		}
4146 		break;
4147 	default:
4148 		if ((error = ether_ioctl(ifp, command, data)) != ENETRESET)
4149 			break;
4150 
4151 		error = 0;
4152 
4153 		if (command != SIOCADDMULTI && command != SIOCDELMULTI)
4154 			;
4155 		else if (ifp->if_flags & IFF_RUNNING)
4156 			bge_setmulti(sc);
4157 		break;
4158 	}
4159 
4160 	splx(s);
4161 
4162 	return (error);
4163 }
4164 
4165 static void
4166 bge_watchdog(struct ifnet *ifp)
4167 {
4168 	struct bge_softc *sc;
4169 
4170 	sc = ifp->if_softc;
4171 
4172 	aprint_error_dev(sc->bge_dev, "watchdog timeout -- resetting\n");
4173 
4174 	ifp->if_flags &= ~IFF_RUNNING;
4175 	bge_init(ifp);
4176 
4177 	ifp->if_oerrors++;
4178 }
4179 
4180 static void
4181 bge_stop_block(struct bge_softc *sc, bus_addr_t reg, uint32_t bit)
4182 {
4183 	int i;
4184 
4185 	BGE_CLRBIT(sc, reg, bit);
4186 
4187 	for (i = 0; i < BGE_TIMEOUT; i++) {
4188 		if ((CSR_READ_4(sc, reg) & bit) == 0)
4189 			return;
4190 		delay(100);
4191 		if (sc->bge_flags & BGE_PCIE)
4192 		  DELAY(1000);
4193 	}
4194 
4195 	/*
4196 	 * Doesn't print only when the register is BGE_SRS_MODE. It occurs
4197 	 * on some environment (and once after boot?)
4198 	 */
4199 	if (reg != BGE_SRS_MODE)
4200 		aprint_error_dev(sc->bge_dev,
4201 		    "block failed to stop: reg 0x%lx, bit 0x%08x\n",
4202 		    (u_long)reg, bit);
4203 }
4204 
4205 /*
4206  * Stop the adapter and free any mbufs allocated to the
4207  * RX and TX lists.
4208  */
4209 static void
4210 bge_stop(struct ifnet *ifp, int disable)
4211 {
4212 	struct bge_softc *sc = ifp->if_softc;
4213 
4214 	callout_stop(&sc->bge_timeout);
4215 
4216 	/*
4217 	 * Disable all of the receiver blocks
4218 	 */
4219 	bge_stop_block(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
4220 	bge_stop_block(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
4221 	bge_stop_block(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
4222 	if (!(BGE_IS_5705_OR_BEYOND(sc)))
4223 		bge_stop_block(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
4224 	bge_stop_block(sc, BGE_RDBDI_MODE, BGE_RBDIMODE_ENABLE);
4225 	bge_stop_block(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
4226 	bge_stop_block(sc, BGE_RBDC_MODE, BGE_RBDCMODE_ENABLE);
4227 
4228 	/*
4229 	 * Disable all of the transmit blocks
4230 	 */
4231 	bge_stop_block(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
4232 	bge_stop_block(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
4233 	bge_stop_block(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
4234 	bge_stop_block(sc, BGE_RDMA_MODE, BGE_RDMAMODE_ENABLE);
4235 	bge_stop_block(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE);
4236 	if (!(BGE_IS_5705_OR_BEYOND(sc)))
4237 		bge_stop_block(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
4238 	bge_stop_block(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
4239 
4240 	/*
4241 	 * Shut down all of the memory managers and related
4242 	 * state machines.
4243 	 */
4244 	bge_stop_block(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE);
4245 	bge_stop_block(sc, BGE_WDMA_MODE, BGE_WDMAMODE_ENABLE);
4246 	if (!(BGE_IS_5705_OR_BEYOND(sc)))
4247 		bge_stop_block(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
4248 
4249 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
4250 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
4251 
4252 	if (!(BGE_IS_5705_OR_BEYOND(sc))) {
4253 		bge_stop_block(sc, BGE_BMAN_MODE, BGE_BMANMODE_ENABLE);
4254 		bge_stop_block(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
4255 	}
4256 
4257 	/* Disable host interrupts. */
4258 	BGE_SETBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR);
4259 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
4260 
4261 	/*
4262 	 * Tell firmware we're shutting down.
4263 	 */
4264 	BGE_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
4265 
4266 	/* Free the RX lists. */
4267 	bge_free_rx_ring_std(sc);
4268 
4269 	/* Free jumbo RX list. */
4270 	bge_free_rx_ring_jumbo(sc);
4271 
4272 	/* Free TX buffers. */
4273 	bge_free_tx_ring(sc);
4274 
4275 	/*
4276 	 * Isolate/power down the PHY.
4277 	 */
4278 	if (!(sc->bge_flags & BGE_PHY_FIBER_TBI))
4279 		mii_down(&sc->bge_mii);
4280 
4281 	sc->bge_tx_saved_considx = BGE_TXCONS_UNSET;
4282 
4283 	/* Clear MAC's link state (PHY may still have link UP). */
4284 	BGE_STS_CLRBIT(sc, BGE_STS_LINK);
4285 
4286 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
4287 }
4288 
4289 static void
4290 bge_link_upd(struct bge_softc *sc)
4291 {
4292 	struct ifnet *ifp = &sc->ethercom.ec_if;
4293 	struct mii_data *mii = &sc->bge_mii;
4294 	u_int32_t status;
4295 	int link;
4296 
4297 	/* Clear 'pending link event' flag */
4298 	BGE_STS_CLRBIT(sc, BGE_STS_LINK_EVT);
4299 
4300 	/*
4301 	 * Process link state changes.
4302 	 * Grrr. The link status word in the status block does
4303 	 * not work correctly on the BCM5700 rev AX and BX chips,
4304 	 * according to all available information. Hence, we have
4305 	 * to enable MII interrupts in order to properly obtain
4306 	 * async link changes. Unfortunately, this also means that
4307 	 * we have to read the MAC status register to detect link
4308 	 * changes, thereby adding an additional register access to
4309 	 * the interrupt handler.
4310 	 */
4311 
4312 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700) {
4313 		status = CSR_READ_4(sc, BGE_MAC_STS);
4314 		if (status & BGE_MACSTAT_MI_INTERRUPT) {
4315 			mii_pollstat(mii);
4316 
4317 			if (!BGE_STS_BIT(sc, BGE_STS_LINK) &&
4318 			    mii->mii_media_status & IFM_ACTIVE &&
4319 			    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE)
4320 				BGE_STS_SETBIT(sc, BGE_STS_LINK);
4321 			else if (BGE_STS_BIT(sc, BGE_STS_LINK) &&
4322 			    (!(mii->mii_media_status & IFM_ACTIVE) ||
4323 			    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE))
4324 				BGE_STS_CLRBIT(sc, BGE_STS_LINK);
4325 
4326 			/* Clear the interrupt */
4327 			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
4328 			    BGE_EVTENB_MI_INTERRUPT);
4329 			bge_miibus_readreg(sc->bge_dev, 1, BRGPHY_MII_ISR);
4330 			bge_miibus_writereg(sc->bge_dev, 1, BRGPHY_MII_IMR,
4331 			    BRGPHY_INTRS);
4332 		}
4333 		return;
4334 	}
4335 
4336 	if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
4337 		status = CSR_READ_4(sc, BGE_MAC_STS);
4338 		if (status & BGE_MACSTAT_TBI_PCS_SYNCHED) {
4339 			if (!BGE_STS_BIT(sc, BGE_STS_LINK)) {
4340 				BGE_STS_SETBIT(sc, BGE_STS_LINK);
4341 				if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704)
4342 					BGE_CLRBIT(sc, BGE_MAC_MODE,
4343 					    BGE_MACMODE_TBI_SEND_CFGS);
4344 				CSR_WRITE_4(sc, BGE_MAC_STS, 0xFFFFFFFF);
4345 				if_link_state_change(ifp, LINK_STATE_UP);
4346 			}
4347 		} else if (BGE_STS_BIT(sc, BGE_STS_LINK)) {
4348 			BGE_STS_CLRBIT(sc, BGE_STS_LINK);
4349 			if_link_state_change(ifp, LINK_STATE_DOWN);
4350 		}
4351         /*
4352 	 * Discard link events for MII/GMII cards if MI auto-polling disabled.
4353 	 * This should not happen since mii callouts are locked now, but
4354 	 * we keep this check for debug.
4355 	 */
4356 	} else if (BGE_STS_BIT(sc, BGE_STS_AUTOPOLL)) {
4357 		/*
4358 		 * Some broken BCM chips have BGE_STATFLAG_LINKSTATE_CHANGED
4359 		 * bit in status word always set. Workaround this bug by
4360 		 * reading PHY link status directly.
4361 		 */
4362 		link = (CSR_READ_4(sc, BGE_MI_STS) & BGE_MISTS_LINK)?
4363 		    BGE_STS_LINK : 0;
4364 
4365 		if (BGE_STS_BIT(sc, BGE_STS_LINK) != link) {
4366 			mii_pollstat(mii);
4367 
4368 			if (!BGE_STS_BIT(sc, BGE_STS_LINK) &&
4369 			    mii->mii_media_status & IFM_ACTIVE &&
4370 			    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE)
4371 				BGE_STS_SETBIT(sc, BGE_STS_LINK);
4372 			else if (BGE_STS_BIT(sc, BGE_STS_LINK) &&
4373 			    (!(mii->mii_media_status & IFM_ACTIVE) ||
4374 			    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE))
4375 				BGE_STS_CLRBIT(sc, BGE_STS_LINK);
4376 		}
4377 	}
4378 
4379 	/* Clear the attention */
4380 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
4381 	    BGE_MACSTAT_CFG_CHANGED|BGE_MACSTAT_MI_COMPLETE|
4382 	    BGE_MACSTAT_LINK_CHANGED);
4383 }
4384 
4385 static int
4386 sysctl_bge_verify(SYSCTLFN_ARGS)
4387 {
4388 	int error, t;
4389 	struct sysctlnode node;
4390 
4391 	node = *rnode;
4392 	t = *(int*)rnode->sysctl_data;
4393 	node.sysctl_data = &t;
4394 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
4395 	if (error || newp == NULL)
4396 		return (error);
4397 
4398 #if 0
4399 	DPRINTF2(("%s: t = %d, nodenum = %d, rnodenum = %d\n", __func__, t,
4400 	    node.sysctl_num, rnode->sysctl_num));
4401 #endif
4402 
4403 	if (node.sysctl_num == bge_rxthresh_nodenum) {
4404 		if (t < 0 || t >= NBGE_RX_THRESH)
4405 			return (EINVAL);
4406 		bge_update_all_threshes(t);
4407 	} else
4408 		return (EINVAL);
4409 
4410 	*(int*)rnode->sysctl_data = t;
4411 
4412 	return (0);
4413 }
4414 
4415 /*
4416  * Set up sysctl(3) MIB, hw.bge.*.
4417  *
4418  * TBD condition SYSCTL_PERMANENT on being an LKM or not
4419  */
4420 SYSCTL_SETUP(sysctl_bge, "sysctl bge subtree setup")
4421 {
4422 	int rc, bge_root_num;
4423 	const struct sysctlnode *node;
4424 
4425 	if ((rc = sysctl_createv(clog, 0, NULL, NULL,
4426 	    CTLFLAG_PERMANENT, CTLTYPE_NODE, "hw", NULL,
4427 	    NULL, 0, NULL, 0, CTL_HW, CTL_EOL)) != 0) {
4428 		goto err;
4429 	}
4430 
4431 	if ((rc = sysctl_createv(clog, 0, NULL, &node,
4432 	    CTLFLAG_PERMANENT, CTLTYPE_NODE, "bge",
4433 	    SYSCTL_DESCR("BGE interface controls"),
4434 	    NULL, 0, NULL, 0, CTL_HW, CTL_CREATE, CTL_EOL)) != 0) {
4435 		goto err;
4436 	}
4437 
4438 	bge_root_num = node->sysctl_num;
4439 
4440 	/* BGE Rx interrupt mitigation level */
4441 	if ((rc = sysctl_createv(clog, 0, NULL, &node,
4442 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
4443 	    CTLTYPE_INT, "rx_lvl",
4444 	    SYSCTL_DESCR("BGE receive interrupt mitigation level"),
4445 	    sysctl_bge_verify, 0,
4446 	    &bge_rx_thresh_lvl,
4447 	    0, CTL_HW, bge_root_num, CTL_CREATE,
4448 	    CTL_EOL)) != 0) {
4449 		goto err;
4450 	}
4451 
4452 	bge_rxthresh_nodenum = node->sysctl_num;
4453 
4454 	return;
4455 
4456 err:
4457 	aprint_error("%s: sysctl_createv failed (rc = %d)\n", __func__, rc);
4458 }
4459 
4460 static int
4461 bge_get_eaddr_mem(struct bge_softc *sc, u_int8_t ether_addr[])
4462 {
4463 	u_int32_t mac_addr;
4464 
4465 	mac_addr = bge_readmem_ind(sc, 0x0c14);
4466 	if ((mac_addr >> 16) == 0x484b) {
4467 		ether_addr[0] = (uint8_t)(mac_addr >> 8);
4468 		ether_addr[1] = (uint8_t)mac_addr;
4469 		mac_addr = bge_readmem_ind(sc, 0x0c18);
4470 		ether_addr[2] = (uint8_t)(mac_addr >> 24);
4471 		ether_addr[3] = (uint8_t)(mac_addr >> 16);
4472 		ether_addr[4] = (uint8_t)(mac_addr >> 8);
4473 		ether_addr[5] = (uint8_t)mac_addr;
4474 		return (0);
4475 	}
4476 	return (1);
4477 }
4478 
4479 static int
4480 bge_get_eaddr_nvram(struct bge_softc *sc, u_int8_t ether_addr[])
4481 {
4482 	int mac_offset = BGE_EE_MAC_OFFSET;
4483 
4484 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
4485 		mac_offset = BGE_EE_MAC_OFFSET_5906;
4486 	}
4487 
4488 	return (bge_read_nvram(sc, ether_addr, mac_offset + 2,
4489 	    ETHER_ADDR_LEN));
4490 }
4491 
4492 static int
4493 bge_get_eaddr_eeprom(struct bge_softc *sc, u_int8_t ether_addr[])
4494 {
4495 
4496 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
4497 		return (1);
4498 	}
4499 
4500 	return (bge_read_eeprom(sc, ether_addr, BGE_EE_MAC_OFFSET + 2,
4501 	   ETHER_ADDR_LEN));
4502 }
4503 
4504 static int
4505 bge_get_eaddr(struct bge_softc *sc, u_int8_t eaddr[])
4506 {
4507 	static const bge_eaddr_fcn_t bge_eaddr_funcs[] = {
4508 		/* NOTE: Order is critical */
4509 		bge_get_eaddr_mem,
4510 		bge_get_eaddr_nvram,
4511 		bge_get_eaddr_eeprom,
4512 		NULL
4513 	};
4514 	const bge_eaddr_fcn_t *func;
4515 
4516 	for (func = bge_eaddr_funcs; *func != NULL; ++func) {
4517 		if ((*func)(sc, eaddr) == 0)
4518 			break;
4519 	}
4520 	return (*func == NULL ? ENXIO : 0);
4521 }
4522