xref: /netbsd-src/sys/dev/pci/if_bge.c (revision 3816d47b2c42fcd6e549e3407f842a5b1a1d23ad)
1 /*	$NetBSD: if_bge.c,v 1.168 2009/09/05 14:09:55 tsutsui 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 1.168 2009/09/05 14:09:55 tsutsui 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 #define BGE_IS_JUMBO_CAPABLE(sc)  \
503 	(BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700    || \
504 	 BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5701    || \
505 	 BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703    || \
506 	 BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704)
507 
508 
509 static const struct bge_revision {
510 	uint32_t		br_chipid;
511 	const char		*br_name;
512 } bge_revisions[] = {
513 	{ BGE_CHIPID_BCM5700_A0, "BCM5700 A0" },
514 	{ BGE_CHIPID_BCM5700_A1, "BCM5700 A1" },
515 	{ BGE_CHIPID_BCM5700_B0, "BCM5700 B0" },
516 	{ BGE_CHIPID_BCM5700_B1, "BCM5700 B1" },
517 	{ BGE_CHIPID_BCM5700_B2, "BCM5700 B2" },
518 	{ BGE_CHIPID_BCM5700_B3, "BCM5700 B3" },
519 	/* This is treated like a BCM5700 Bx */
520 	{ BGE_CHIPID_BCM5700_ALTIMA, "BCM5700 Altima" },
521 	{ BGE_CHIPID_BCM5700_C0, "BCM5700 C0" },
522 	{ BGE_CHIPID_BCM5701_A0, "BCM5701 A0" },
523 	{ BGE_CHIPID_BCM5701_B0, "BCM5701 B0" },
524 	{ BGE_CHIPID_BCM5701_B2, "BCM5701 B2" },
525 	{ BGE_CHIPID_BCM5701_B5, "BCM5701 B5" },
526 	{ BGE_CHIPID_BCM5703_A0, "BCM5703 A0" },
527 	{ BGE_CHIPID_BCM5703_A1, "BCM5703 A1" },
528 	{ BGE_CHIPID_BCM5703_A2, "BCM5703 A2" },
529 	{ BGE_CHIPID_BCM5703_A3, "BCM5703 A3" },
530 	{ BGE_CHIPID_BCM5703_B0, "BCM5703 B0" },
531 	{ BGE_CHIPID_BCM5704_A0, "BCM5704 A0" },
532 	{ BGE_CHIPID_BCM5704_A1, "BCM5704 A1" },
533 	{ BGE_CHIPID_BCM5704_A2, "BCM5704 A2" },
534 	{ BGE_CHIPID_BCM5704_A3, "BCM5704 A3" },
535 	{ BGE_CHIPID_BCM5704_B0, "BCM5704 B0" },
536 	{ BGE_CHIPID_BCM5705_A0, "BCM5705 A0" },
537 	{ BGE_CHIPID_BCM5705_A1, "BCM5705 A1" },
538 	{ BGE_CHIPID_BCM5705_A2, "BCM5705 A2" },
539 	{ BGE_CHIPID_BCM5705_A3, "BCM5705 A3" },
540 	{ BGE_CHIPID_BCM5750_A0, "BCM5750 A0" },
541 	{ BGE_CHIPID_BCM5750_A1, "BCM5750 A1" },
542 	{ BGE_CHIPID_BCM5750_A3, "BCM5750 A3" },
543 	{ BGE_CHIPID_BCM5750_B0, "BCM5750 B0" },
544 	{ BGE_CHIPID_BCM5750_B1, "BCM5750 B1" },
545 	{ BGE_CHIPID_BCM5750_C0, "BCM5750 C0" },
546 	{ BGE_CHIPID_BCM5750_C1, "BCM5750 C1" },
547 	{ BGE_CHIPID_BCM5750_C2, "BCM5750 C2" },
548 	{ BGE_CHIPID_BCM5752_A0, "BCM5752 A0" },
549 	{ BGE_CHIPID_BCM5752_A1, "BCM5752 A1" },
550 	{ BGE_CHIPID_BCM5752_A2, "BCM5752 A2" },
551 	{ BGE_CHIPID_BCM5714_A0, "BCM5714 A0" },
552 	{ BGE_CHIPID_BCM5714_B0, "BCM5714 B0" },
553 	{ BGE_CHIPID_BCM5714_B3, "BCM5714 B3" },
554 	{ BGE_CHIPID_BCM5715_A0, "BCM5715 A0" },
555 	{ BGE_CHIPID_BCM5715_A1, "BCM5715 A1" },
556 	{ BGE_CHIPID_BCM5715_A3, "BCM5715 A3" },
557 	{ BGE_CHIPID_BCM5755_A0, "BCM5755 A0" },
558 	{ BGE_CHIPID_BCM5755_A1, "BCM5755 A1" },
559 	{ BGE_CHIPID_BCM5755_A2, "BCM5755 A2" },
560 	{ BGE_CHIPID_BCM5755_C0, "BCM5755 C0" },
561 	{ BGE_CHIPID_BCM5787_A0, "BCM5754/5787 A0" },
562 	{ BGE_CHIPID_BCM5787_A1, "BCM5754/5787 A1" },
563 	{ BGE_CHIPID_BCM5787_A2, "BCM5754/5787 A2" },
564 	{ BGE_CHIPID_BCM5906_A1, "BCM5906 A1" },
565 	{ BGE_CHIPID_BCM5906_A2, "BCM5906 A2" },
566 	{ 0, NULL }
567 };
568 
569 /*
570  * Some defaults for major revisions, so that newer steppings
571  * that we don't know about have a shot at working.
572  */
573 static const struct bge_revision bge_majorrevs[] = {
574 	{ BGE_ASICREV_BCM5700, "unknown BCM5700" },
575 	{ BGE_ASICREV_BCM5701, "unknown BCM5701" },
576 	{ BGE_ASICREV_BCM5703, "unknown BCM5703" },
577 	{ BGE_ASICREV_BCM5704, "unknown BCM5704" },
578 	{ BGE_ASICREV_BCM5705, "unknown BCM5705" },
579 	{ BGE_ASICREV_BCM5750, "unknown BCM5750" },
580 	{ BGE_ASICREV_BCM5714_A0, "unknown BCM5714" },
581 	{ BGE_ASICREV_BCM5714, "unknown BCM5714" },
582 	{ BGE_ASICREV_BCM5752, "unknown BCM5752 family" },
583 	{ BGE_ASICREV_BCM5755, "unknown BCM5755" },
584 	{ BGE_ASICREV_BCM5780, "unknown BCM5780" },
585 	/* 5754 and 5787 share the same ASIC ID */
586 	{ BGE_ASICREV_BCM5787, "unknown BCM5754/5787" },
587 	{ BGE_ASICREV_BCM5906, "unknown BCM5906" },
588 	{ 0, NULL }
589 };
590 
591 CFATTACH_DECL_NEW(bge, sizeof(struct bge_softc),
592     bge_probe, bge_attach, NULL, NULL);
593 
594 static u_int32_t
595 bge_readmem_ind(struct bge_softc *sc, int off)
596 {
597 	pcireg_t val;
598 
599 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, off);
600 	val = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_DATA);
601 	return val;
602 }
603 
604 static void
605 bge_writemem_ind(struct bge_softc *sc, int off, int val)
606 {
607 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, off);
608 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_DATA, val);
609 }
610 
611 #ifdef notdef
612 static u_int32_t
613 bge_readreg_ind(struct bge_softc *sc, int off)
614 {
615 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_BASEADDR, off);
616 	return (pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_DATA));
617 }
618 #endif
619 
620 static void
621 bge_writereg_ind(struct bge_softc *sc, int off, int val)
622 {
623 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_BASEADDR, off);
624 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_REG_DATA, val);
625 }
626 
627 static void
628 bge_writemem_direct(struct bge_softc *sc, int off, int val)
629 {
630 	CSR_WRITE_4(sc, off, val);
631 }
632 
633 static void
634 bge_writembx(struct bge_softc *sc, int off, int val)
635 {
636 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906)
637 		off += BGE_LPMBX_IRQ0_HI - BGE_MBX_IRQ0_HI;
638 
639 	CSR_WRITE_4(sc, off, val);
640 }
641 
642 static u_int8_t
643 bge_nvram_getbyte(struct bge_softc *sc, int addr, u_int8_t *dest)
644 {
645 	u_int32_t access, byte = 0;
646 	int i;
647 
648 	/* Lock. */
649 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_SET1);
650 	for (i = 0; i < 8000; i++) {
651 		if (CSR_READ_4(sc, BGE_NVRAM_SWARB) & BGE_NVRAMSWARB_GNT1)
652 			break;
653 		DELAY(20);
654 	}
655 	if (i == 8000)
656 		return (1);
657 
658 	/* Enable access. */
659 	access = CSR_READ_4(sc, BGE_NVRAM_ACCESS);
660 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access | BGE_NVRAMACC_ENABLE);
661 
662 	CSR_WRITE_4(sc, BGE_NVRAM_ADDR, addr & 0xfffffffc);
663 	CSR_WRITE_4(sc, BGE_NVRAM_CMD, BGE_NVRAM_READCMD);
664 	for (i = 0; i < BGE_TIMEOUT * 10; i++) {
665 		DELAY(10);
666 		if (CSR_READ_4(sc, BGE_NVRAM_CMD) & BGE_NVRAMCMD_DONE) {
667 			DELAY(10);
668 			break;
669 		}
670 	}
671 
672 	if (i == BGE_TIMEOUT * 10) {
673 		aprint_error_dev(sc->bge_dev, "nvram read timed out\n");
674 		return (1);
675 	}
676 
677 	/* Get result. */
678 	byte = CSR_READ_4(sc, BGE_NVRAM_RDDATA);
679 
680 	*dest = (bswap32(byte) >> ((addr % 4) * 8)) & 0xFF;
681 
682 	/* Disable access. */
683 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access);
684 
685 	/* Unlock. */
686 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_CLR1);
687 	CSR_READ_4(sc, BGE_NVRAM_SWARB);
688 
689 	return (0);
690 }
691 
692 /*
693  * Read a sequence of bytes from NVRAM.
694  */
695 static int
696 bge_read_nvram(struct bge_softc *sc, u_int8_t *dest, int off, int cnt)
697 {
698 	int err = 0, i;
699 	u_int8_t byte = 0;
700 
701 	if (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5906)
702 		return (1);
703 
704 	for (i = 0; i < cnt; i++) {
705 		err = bge_nvram_getbyte(sc, off + i, &byte);
706 		if (err)
707 			break;
708 		*(dest + i) = byte;
709 	}
710 
711 	return (err ? 1 : 0);
712 }
713 
714 /*
715  * Read a byte of data stored in the EEPROM at address 'addr.' The
716  * BCM570x supports both the traditional bitbang interface and an
717  * auto access interface for reading the EEPROM. We use the auto
718  * access method.
719  */
720 static u_int8_t
721 bge_eeprom_getbyte(struct bge_softc *sc, int addr, u_int8_t *dest)
722 {
723 	int i;
724 	u_int32_t byte = 0;
725 
726 	/*
727 	 * Enable use of auto EEPROM access so we can avoid
728 	 * having to use the bitbang method.
729 	 */
730 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_AUTO_EEPROM);
731 
732 	/* Reset the EEPROM, load the clock period. */
733 	CSR_WRITE_4(sc, BGE_EE_ADDR,
734 	    BGE_EEADDR_RESET | BGE_EEHALFCLK(BGE_HALFCLK_384SCL));
735 	DELAY(20);
736 
737 	/* Issue the read EEPROM command. */
738 	CSR_WRITE_4(sc, BGE_EE_ADDR, BGE_EE_READCMD | addr);
739 
740 	/* Wait for completion */
741 	for(i = 0; i < BGE_TIMEOUT * 10; i++) {
742 		DELAY(10);
743 		if (CSR_READ_4(sc, BGE_EE_ADDR) & BGE_EEADDR_DONE)
744 			break;
745 	}
746 
747 	if (i == BGE_TIMEOUT) {
748 		aprint_error_dev(sc->bge_dev, "eeprom read timed out\n");
749 		return (0);
750 	}
751 
752 	/* Get result. */
753 	byte = CSR_READ_4(sc, BGE_EE_DATA);
754 
755 	*dest = (byte >> ((addr % 4) * 8)) & 0xFF;
756 
757 	return (0);
758 }
759 
760 /*
761  * Read a sequence of bytes from the EEPROM.
762  */
763 static int
764 bge_read_eeprom(struct bge_softc *sc, void *destv, int off, int cnt)
765 {
766 	int err = 0, i;
767 	u_int8_t byte = 0;
768 	char *dest = destv;
769 
770 	for (i = 0; i < cnt; i++) {
771 		err = bge_eeprom_getbyte(sc, off + i, &byte);
772 		if (err)
773 			break;
774 		*(dest + i) = byte;
775 	}
776 
777 	return (err ? 1 : 0);
778 }
779 
780 static int
781 bge_miibus_readreg(device_t dev, int phy, int reg)
782 {
783 	struct bge_softc *sc = device_private(dev);
784 	u_int32_t val;
785 	u_int32_t saved_autopoll;
786 	int i;
787 
788 	/*
789 	 * Broadcom's own driver always assumes the internal
790 	 * PHY is at GMII address 1. On some chips, the PHY responds
791 	 * to accesses at all addresses, which could cause us to
792 	 * bogusly attach the PHY 32 times at probe type. Always
793 	 * restricting the lookup to address 1 is simpler than
794 	 * trying to figure out which chips revisions should be
795 	 * special-cased.
796 	 */
797 	if (phy != 1)
798 		return (0);
799 
800 	/* Reading with autopolling on may trigger PCI errors */
801 	saved_autopoll = CSR_READ_4(sc, BGE_MI_MODE);
802 	if (saved_autopoll & BGE_MIMODE_AUTOPOLL) {
803 		BGE_STS_CLRBIT(sc, BGE_STS_AUTOPOLL);
804 		CSR_WRITE_4(sc, BGE_MI_MODE,
805 		    saved_autopoll &~ BGE_MIMODE_AUTOPOLL);
806 		DELAY(40);
807 	}
808 
809 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_READ|BGE_MICOMM_BUSY|
810 	    BGE_MIPHY(phy)|BGE_MIREG(reg));
811 
812 	for (i = 0; i < BGE_TIMEOUT; i++) {
813 		val = CSR_READ_4(sc, BGE_MI_COMM);
814 		if (!(val & BGE_MICOMM_BUSY))
815 			break;
816 		delay(10);
817 	}
818 
819 	if (i == BGE_TIMEOUT) {
820 		aprint_error_dev(sc->bge_dev, "PHY read timed out\n");
821 		val = 0;
822 		goto done;
823 	}
824 
825 	val = CSR_READ_4(sc, BGE_MI_COMM);
826 
827 done:
828 	if (saved_autopoll & BGE_MIMODE_AUTOPOLL) {
829 		BGE_STS_SETBIT(sc, BGE_STS_AUTOPOLL);
830 		CSR_WRITE_4(sc, BGE_MI_MODE, saved_autopoll);
831 		DELAY(40);
832 	}
833 
834 	if (val & BGE_MICOMM_READFAIL)
835 		return (0);
836 
837 	return (val & 0xFFFF);
838 }
839 
840 static void
841 bge_miibus_writereg(device_t dev, int phy, int reg, int val)
842 {
843 	struct bge_softc *sc = device_private(dev);
844 	u_int32_t saved_autopoll;
845 	int i;
846 
847 	if (phy!=1) {
848 		return;
849 	}
850 
851 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906 &&
852 	    (reg == BRGPHY_MII_1000CTL || reg == BRGPHY_MII_AUXCTL)) {
853 		return;
854 	}
855 
856 	/* Reading with autopolling on may trigger PCI errors */
857 	saved_autopoll = CSR_READ_4(sc, BGE_MI_MODE);
858 	if (saved_autopoll & BGE_MIMODE_AUTOPOLL) {
859 		delay(40);
860 		BGE_STS_CLRBIT(sc, BGE_STS_AUTOPOLL);
861 		CSR_WRITE_4(sc, BGE_MI_MODE,
862 		    saved_autopoll & (~BGE_MIMODE_AUTOPOLL));
863 		delay(10); /* 40 usec is supposed to be adequate */
864 	}
865 
866 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_WRITE | BGE_MICOMM_BUSY |
867 	    BGE_MIPHY(phy) | BGE_MIREG(reg)|val);
868 
869 	for (i = 0; i < BGE_TIMEOUT; i++) {
870 		delay(10);
871 		if (!(CSR_READ_4(sc, BGE_MI_COMM) & BGE_MICOMM_BUSY)) {
872 			delay(5);
873 			CSR_READ_4(sc, BGE_MI_COMM);
874 			break;
875 		}
876 	}
877 
878 	if (saved_autopoll & BGE_MIMODE_AUTOPOLL) {
879 		BGE_STS_SETBIT(sc, BGE_STS_AUTOPOLL);
880 		CSR_WRITE_4(sc, BGE_MI_MODE, saved_autopoll);
881 		delay(40);
882 	}
883 
884 	if (i == BGE_TIMEOUT)
885 		aprint_error_dev(sc->bge_dev, "PHY read timed out\n");
886 }
887 
888 static void
889 bge_miibus_statchg(device_t dev)
890 {
891 	struct bge_softc *sc = device_private(dev);
892 	struct mii_data *mii = &sc->bge_mii;
893 
894 	/*
895 	 * Get flow control negotiation result.
896 	 */
897 	if (IFM_SUBTYPE(mii->mii_media.ifm_cur->ifm_media) == IFM_AUTO &&
898 	    (mii->mii_media_active & IFM_ETH_FMASK) != sc->bge_flowflags) {
899 		sc->bge_flowflags = mii->mii_media_active & IFM_ETH_FMASK;
900 		mii->mii_media_active &= ~IFM_ETH_FMASK;
901 	}
902 
903 	BGE_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_PORTMODE);
904 	if (IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_T ||
905 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_SX)
906 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_GMII);
907 	else
908 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_MII);
909 
910 	if ((mii->mii_media_active & IFM_GMASK) == IFM_FDX)
911 		BGE_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
912 	else
913 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
914 
915 	/*
916 	 * 802.3x flow control
917 	 */
918 	if (sc->bge_flowflags & IFM_ETH_RXPAUSE)
919 		BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_FLOWCTL_ENABLE);
920 	else
921 		BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_FLOWCTL_ENABLE);
922 
923 	if (sc->bge_flowflags & IFM_ETH_TXPAUSE)
924 		BGE_SETBIT(sc, BGE_TX_MODE, BGE_TXMODE_FLOWCTL_ENABLE);
925 	else
926 		BGE_CLRBIT(sc, BGE_TX_MODE, BGE_TXMODE_FLOWCTL_ENABLE);
927 }
928 
929 /*
930  * Update rx threshold levels to values in a particular slot
931  * of the interrupt-mitigation table bge_rx_threshes.
932  */
933 static void
934 bge_set_thresh(struct ifnet *ifp, int lvl)
935 {
936 	struct bge_softc *sc = ifp->if_softc;
937 	int s;
938 
939 	/* For now, just save the new Rx-intr thresholds and record
940 	 * that a threshold update is pending.  Updating the hardware
941 	 * registers here (even at splhigh()) is observed to
942 	 * occasionaly cause glitches where Rx-interrupts are not
943 	 * honoured for up to 10 seconds. jonathan@NetBSD.org, 2003-04-05
944 	 */
945 	s = splnet();
946 	sc->bge_rx_coal_ticks = bge_rx_threshes[lvl].rx_ticks;
947 	sc->bge_rx_max_coal_bds = bge_rx_threshes[lvl].rx_max_bds;
948 	sc->bge_pending_rxintr_change = 1;
949 	splx(s);
950 
951 	 return;
952 }
953 
954 
955 /*
956  * Update Rx thresholds of all bge devices
957  */
958 static void
959 bge_update_all_threshes(int lvl)
960 {
961 	struct ifnet *ifp;
962 	const char * const namebuf = "bge";
963 	int namelen;
964 
965 	if (lvl < 0)
966 		lvl = 0;
967 	else if( lvl >= NBGE_RX_THRESH)
968 		lvl = NBGE_RX_THRESH - 1;
969 
970 	namelen = strlen(namebuf);
971 	/*
972 	 * Now search all the interfaces for this name/number
973 	 */
974 	IFNET_FOREACH(ifp) {
975 		if (strncmp(ifp->if_xname, namebuf, namelen) != 0)
976 		      continue;
977 		/* We got a match: update if doing auto-threshold-tuning */
978 		if (bge_auto_thresh)
979 			bge_set_thresh(ifp, lvl);
980 	}
981 }
982 
983 /*
984  * Handle events that have triggered interrupts.
985  */
986 static void
987 bge_handle_events(struct bge_softc *sc)
988 {
989 
990 	return;
991 }
992 
993 /*
994  * Memory management for jumbo frames.
995  */
996 
997 static int
998 bge_alloc_jumbo_mem(struct bge_softc *sc)
999 {
1000 	char *ptr, *kva;
1001 	bus_dma_segment_t	seg;
1002 	int		i, rseg, state, error;
1003 	struct bge_jpool_entry   *entry;
1004 
1005 	state = error = 0;
1006 
1007 	/* Grab a big chunk o' storage. */
1008 	if (bus_dmamem_alloc(sc->bge_dmatag, BGE_JMEM, PAGE_SIZE, 0,
1009 	     &seg, 1, &rseg, BUS_DMA_NOWAIT)) {
1010 		aprint_error_dev(sc->bge_dev, "can't alloc rx buffers\n");
1011 		return ENOBUFS;
1012 	}
1013 
1014 	state = 1;
1015 	if (bus_dmamem_map(sc->bge_dmatag, &seg, rseg, BGE_JMEM, (void **)&kva,
1016 	    BUS_DMA_NOWAIT)) {
1017 		aprint_error_dev(sc->bge_dev,
1018 		    "can't map DMA buffers (%d bytes)\n", (int)BGE_JMEM);
1019 		error = ENOBUFS;
1020 		goto out;
1021 	}
1022 
1023 	state = 2;
1024 	if (bus_dmamap_create(sc->bge_dmatag, BGE_JMEM, 1, BGE_JMEM, 0,
1025 	    BUS_DMA_NOWAIT, &sc->bge_cdata.bge_rx_jumbo_map)) {
1026 		aprint_error_dev(sc->bge_dev, "can't create DMA map\n");
1027 		error = ENOBUFS;
1028 		goto out;
1029 	}
1030 
1031 	state = 3;
1032 	if (bus_dmamap_load(sc->bge_dmatag, sc->bge_cdata.bge_rx_jumbo_map,
1033 	    kva, BGE_JMEM, NULL, BUS_DMA_NOWAIT)) {
1034 		aprint_error_dev(sc->bge_dev, "can't load DMA map\n");
1035 		error = ENOBUFS;
1036 		goto out;
1037 	}
1038 
1039 	state = 4;
1040 	sc->bge_cdata.bge_jumbo_buf = (void *)kva;
1041 	DPRINTFN(1,("bge_jumbo_buf = %p\n", sc->bge_cdata.bge_jumbo_buf));
1042 
1043 	SLIST_INIT(&sc->bge_jfree_listhead);
1044 	SLIST_INIT(&sc->bge_jinuse_listhead);
1045 
1046 	/*
1047 	 * Now divide it up into 9K pieces and save the addresses
1048 	 * in an array.
1049 	 */
1050 	ptr = sc->bge_cdata.bge_jumbo_buf;
1051 	for (i = 0; i < BGE_JSLOTS; i++) {
1052 		sc->bge_cdata.bge_jslots[i] = ptr;
1053 		ptr += BGE_JLEN;
1054 		entry = malloc(sizeof(struct bge_jpool_entry),
1055 		    M_DEVBUF, M_NOWAIT);
1056 		if (entry == NULL) {
1057 			aprint_error_dev(sc->bge_dev,
1058 			    "no memory for jumbo buffer queue!\n");
1059 			error = ENOBUFS;
1060 			goto out;
1061 		}
1062 		entry->slot = i;
1063 		SLIST_INSERT_HEAD(&sc->bge_jfree_listhead,
1064 				 entry, jpool_entries);
1065 	}
1066 out:
1067 	if (error != 0) {
1068 		switch (state) {
1069 		case 4:
1070 			bus_dmamap_unload(sc->bge_dmatag,
1071 			    sc->bge_cdata.bge_rx_jumbo_map);
1072 		case 3:
1073 			bus_dmamap_destroy(sc->bge_dmatag,
1074 			    sc->bge_cdata.bge_rx_jumbo_map);
1075 		case 2:
1076 			bus_dmamem_unmap(sc->bge_dmatag, kva, BGE_JMEM);
1077 		case 1:
1078 			bus_dmamem_free(sc->bge_dmatag, &seg, rseg);
1079 			break;
1080 		default:
1081 			break;
1082 		}
1083 	}
1084 
1085 	return error;
1086 }
1087 
1088 /*
1089  * Allocate a jumbo buffer.
1090  */
1091 static void *
1092 bge_jalloc(struct bge_softc *sc)
1093 {
1094 	struct bge_jpool_entry   *entry;
1095 
1096 	entry = SLIST_FIRST(&sc->bge_jfree_listhead);
1097 
1098 	if (entry == NULL) {
1099 		aprint_error_dev(sc->bge_dev, "no free jumbo buffers\n");
1100 		return (NULL);
1101 	}
1102 
1103 	SLIST_REMOVE_HEAD(&sc->bge_jfree_listhead, jpool_entries);
1104 	SLIST_INSERT_HEAD(&sc->bge_jinuse_listhead, entry, jpool_entries);
1105 	return (sc->bge_cdata.bge_jslots[entry->slot]);
1106 }
1107 
1108 /*
1109  * Release a jumbo buffer.
1110  */
1111 static void
1112 bge_jfree(struct mbuf *m, void *buf, size_t size, void *arg)
1113 {
1114 	struct bge_jpool_entry *entry;
1115 	struct bge_softc *sc;
1116 	int i, s;
1117 
1118 	/* Extract the softc struct pointer. */
1119 	sc = (struct bge_softc *)arg;
1120 
1121 	if (sc == NULL)
1122 		panic("bge_jfree: can't find softc pointer!");
1123 
1124 	/* calculate the slot this buffer belongs to */
1125 
1126 	i = ((char *)buf
1127 	     - (char *)sc->bge_cdata.bge_jumbo_buf) / BGE_JLEN;
1128 
1129 	if ((i < 0) || (i >= BGE_JSLOTS))
1130 		panic("bge_jfree: asked to free buffer that we don't manage!");
1131 
1132 	s = splvm();
1133 	entry = SLIST_FIRST(&sc->bge_jinuse_listhead);
1134 	if (entry == NULL)
1135 		panic("bge_jfree: buffer not in use!");
1136 	entry->slot = i;
1137 	SLIST_REMOVE_HEAD(&sc->bge_jinuse_listhead, jpool_entries);
1138 	SLIST_INSERT_HEAD(&sc->bge_jfree_listhead, entry, jpool_entries);
1139 
1140 	if (__predict_true(m != NULL))
1141   		pool_cache_put(mb_cache, m);
1142 	splx(s);
1143 }
1144 
1145 
1146 /*
1147  * Intialize a standard receive ring descriptor.
1148  */
1149 static int
1150 bge_newbuf_std(struct bge_softc *sc, int i, struct mbuf *m, bus_dmamap_t dmamap)
1151 {
1152 	struct mbuf		*m_new = NULL;
1153 	struct bge_rx_bd	*r;
1154 	int			error;
1155 
1156 	if (dmamap == NULL) {
1157 		error = bus_dmamap_create(sc->bge_dmatag, MCLBYTES, 1,
1158 		    MCLBYTES, 0, BUS_DMA_NOWAIT, &dmamap);
1159 		if (error != 0)
1160 			return error;
1161 	}
1162 
1163 	sc->bge_cdata.bge_rx_std_map[i] = dmamap;
1164 
1165 	if (m == NULL) {
1166 		MGETHDR(m_new, M_DONTWAIT, MT_DATA);
1167 		if (m_new == NULL)
1168 			return (ENOBUFS);
1169 
1170 		MCLGET(m_new, M_DONTWAIT);
1171 		if (!(m_new->m_flags & M_EXT)) {
1172 			m_freem(m_new);
1173 			return (ENOBUFS);
1174 		}
1175 		m_new->m_len = m_new->m_pkthdr.len = MCLBYTES;
1176 
1177 	} else {
1178 		m_new = m;
1179 		m_new->m_len = m_new->m_pkthdr.len = MCLBYTES;
1180 		m_new->m_data = m_new->m_ext.ext_buf;
1181 	}
1182 	if (!(sc->bge_flags & BGE_RX_ALIGNBUG))
1183 	    m_adj(m_new, ETHER_ALIGN);
1184 	if (bus_dmamap_load_mbuf(sc->bge_dmatag, dmamap, m_new,
1185 	    BUS_DMA_READ|BUS_DMA_NOWAIT))
1186 		return (ENOBUFS);
1187 	bus_dmamap_sync(sc->bge_dmatag, dmamap, 0, dmamap->dm_mapsize,
1188 	    BUS_DMASYNC_PREREAD);
1189 
1190 	sc->bge_cdata.bge_rx_std_chain[i] = m_new;
1191 	r = &sc->bge_rdata->bge_rx_std_ring[i];
1192 	bge_set_hostaddr(&r->bge_addr,
1193 	    dmamap->dm_segs[0].ds_addr);
1194 	r->bge_flags = BGE_RXBDFLAG_END;
1195 	r->bge_len = m_new->m_len;
1196 	r->bge_idx = i;
1197 
1198 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
1199 	    offsetof(struct bge_ring_data, bge_rx_std_ring) +
1200 		i * sizeof (struct bge_rx_bd),
1201 	    sizeof (struct bge_rx_bd),
1202 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1203 
1204 	return (0);
1205 }
1206 
1207 /*
1208  * Initialize a jumbo receive ring descriptor. This allocates
1209  * a jumbo buffer from the pool managed internally by the driver.
1210  */
1211 static int
1212 bge_newbuf_jumbo(struct bge_softc *sc, int i, struct mbuf *m)
1213 {
1214 	struct mbuf *m_new = NULL;
1215 	struct bge_rx_bd *r;
1216 	void *buf = NULL;
1217 
1218 	if (m == NULL) {
1219 
1220 		/* Allocate the mbuf. */
1221 		MGETHDR(m_new, M_DONTWAIT, MT_DATA);
1222 		if (m_new == NULL)
1223 			return (ENOBUFS);
1224 
1225 		/* Allocate the jumbo buffer */
1226 		buf = bge_jalloc(sc);
1227 		if (buf == NULL) {
1228 			m_freem(m_new);
1229 			aprint_error_dev(sc->bge_dev,
1230 			    "jumbo allocation failed -- packet dropped!\n");
1231 			return (ENOBUFS);
1232 		}
1233 
1234 		/* Attach the buffer to the mbuf. */
1235 		m_new->m_len = m_new->m_pkthdr.len = BGE_JUMBO_FRAMELEN;
1236 		MEXTADD(m_new, buf, BGE_JUMBO_FRAMELEN, M_DEVBUF,
1237 		    bge_jfree, sc);
1238 		m_new->m_flags |= M_EXT_RW;
1239 	} else {
1240 		m_new = m;
1241 		buf = m_new->m_data = m_new->m_ext.ext_buf;
1242 		m_new->m_ext.ext_size = BGE_JUMBO_FRAMELEN;
1243 	}
1244 	if (!(sc->bge_flags & BGE_RX_ALIGNBUG))
1245 	    m_adj(m_new, ETHER_ALIGN);
1246 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_cdata.bge_rx_jumbo_map,
1247 	    mtod(m_new, char *) - (char *)sc->bge_cdata.bge_jumbo_buf, BGE_JLEN,
1248 	    BUS_DMASYNC_PREREAD);
1249 	/* Set up the descriptor. */
1250 	r = &sc->bge_rdata->bge_rx_jumbo_ring[i];
1251 	sc->bge_cdata.bge_rx_jumbo_chain[i] = m_new;
1252 	bge_set_hostaddr(&r->bge_addr, BGE_JUMBO_DMA_ADDR(sc, m_new));
1253 	r->bge_flags = BGE_RXBDFLAG_END|BGE_RXBDFLAG_JUMBO_RING;
1254 	r->bge_len = m_new->m_len;
1255 	r->bge_idx = i;
1256 
1257 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
1258 	    offsetof(struct bge_ring_data, bge_rx_jumbo_ring) +
1259 		i * sizeof (struct bge_rx_bd),
1260 	    sizeof (struct bge_rx_bd),
1261 	    BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
1262 
1263 	return (0);
1264 }
1265 
1266 /*
1267  * The standard receive ring has 512 entries in it. At 2K per mbuf cluster,
1268  * that's 1MB or memory, which is a lot. For now, we fill only the first
1269  * 256 ring entries and hope that our CPU is fast enough to keep up with
1270  * the NIC.
1271  */
1272 static int
1273 bge_init_rx_ring_std(struct bge_softc *sc)
1274 {
1275 	int i;
1276 
1277 	if (sc->bge_flags & BGE_RXRING_VALID)
1278 		return 0;
1279 
1280 	for (i = 0; i < BGE_SSLOTS; i++) {
1281 		if (bge_newbuf_std(sc, i, NULL, 0) == ENOBUFS)
1282 			return (ENOBUFS);
1283 	}
1284 
1285 	sc->bge_std = i - 1;
1286 	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bge_std);
1287 
1288 	sc->bge_flags |= BGE_RXRING_VALID;
1289 
1290 	return (0);
1291 }
1292 
1293 static void
1294 bge_free_rx_ring_std(struct bge_softc *sc)
1295 {
1296 	int i;
1297 
1298 	if (!(sc->bge_flags & BGE_RXRING_VALID))
1299 		return;
1300 
1301 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1302 		if (sc->bge_cdata.bge_rx_std_chain[i] != NULL) {
1303 			m_freem(sc->bge_cdata.bge_rx_std_chain[i]);
1304 			sc->bge_cdata.bge_rx_std_chain[i] = NULL;
1305 			bus_dmamap_destroy(sc->bge_dmatag,
1306 			    sc->bge_cdata.bge_rx_std_map[i]);
1307 		}
1308 		memset((char *)&sc->bge_rdata->bge_rx_std_ring[i], 0,
1309 		    sizeof(struct bge_rx_bd));
1310 	}
1311 
1312 	sc->bge_flags &= ~BGE_RXRING_VALID;
1313 }
1314 
1315 static int
1316 bge_init_rx_ring_jumbo(struct bge_softc *sc)
1317 {
1318 	int i;
1319 	volatile struct bge_rcb *rcb;
1320 
1321 	if (sc->bge_flags & BGE_JUMBO_RXRING_VALID)
1322 		return 0;
1323 
1324 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1325 		if (bge_newbuf_jumbo(sc, i, NULL) == ENOBUFS)
1326 			return (ENOBUFS);
1327 	};
1328 
1329 	sc->bge_jumbo = i - 1;
1330 	sc->bge_flags |= BGE_JUMBO_RXRING_VALID;
1331 
1332 	rcb = &sc->bge_rdata->bge_info.bge_jumbo_rx_rcb;
1333 	rcb->bge_maxlen_flags = 0;
1334 	CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1335 
1336 	bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
1337 
1338 	return (0);
1339 }
1340 
1341 static void
1342 bge_free_rx_ring_jumbo(struct bge_softc *sc)
1343 {
1344 	int i;
1345 
1346 	if (!(sc->bge_flags & BGE_JUMBO_RXRING_VALID))
1347 		return;
1348 
1349 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1350 		if (sc->bge_cdata.bge_rx_jumbo_chain[i] != NULL) {
1351 			m_freem(sc->bge_cdata.bge_rx_jumbo_chain[i]);
1352 			sc->bge_cdata.bge_rx_jumbo_chain[i] = NULL;
1353 		}
1354 		memset((char *)&sc->bge_rdata->bge_rx_jumbo_ring[i], 0,
1355 		    sizeof(struct bge_rx_bd));
1356 	}
1357 
1358 	sc->bge_flags &= ~BGE_JUMBO_RXRING_VALID;
1359 }
1360 
1361 static void
1362 bge_free_tx_ring(struct bge_softc *sc)
1363 {
1364 	int i, freed;
1365 	struct txdmamap_pool_entry *dma;
1366 
1367 	if (!(sc->bge_flags & BGE_TXRING_VALID))
1368 		return;
1369 
1370 	freed = 0;
1371 
1372 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
1373 		if (sc->bge_cdata.bge_tx_chain[i] != NULL) {
1374 			freed++;
1375 			m_freem(sc->bge_cdata.bge_tx_chain[i]);
1376 			sc->bge_cdata.bge_tx_chain[i] = NULL;
1377 			SLIST_INSERT_HEAD(&sc->txdma_list, sc->txdma[i],
1378 					    link);
1379 			sc->txdma[i] = 0;
1380 		}
1381 		memset((char *)&sc->bge_rdata->bge_tx_ring[i], 0,
1382 		    sizeof(struct bge_tx_bd));
1383 	}
1384 
1385 	while ((dma = SLIST_FIRST(&sc->txdma_list))) {
1386 		SLIST_REMOVE_HEAD(&sc->txdma_list, link);
1387 		bus_dmamap_destroy(sc->bge_dmatag, dma->dmamap);
1388 		free(dma, M_DEVBUF);
1389 	}
1390 
1391 	sc->bge_flags &= ~BGE_TXRING_VALID;
1392 }
1393 
1394 static int
1395 bge_init_tx_ring(struct bge_softc *sc)
1396 {
1397 	int i;
1398 	bus_dmamap_t dmamap;
1399 	struct txdmamap_pool_entry *dma;
1400 
1401 	if (sc->bge_flags & BGE_TXRING_VALID)
1402 		return 0;
1403 
1404 	sc->bge_txcnt = 0;
1405 	sc->bge_tx_saved_considx = 0;
1406 
1407 	/* Initialize transmit producer index for host-memory send ring. */
1408 	sc->bge_tx_prodidx = 0;
1409 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1410 	/* 5700 b2 errata */
1411 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX)
1412 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1413 
1414 	/* NIC-memory send ring not used; initialize to zero. */
1415 	bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1416 	/* 5700 b2 errata */
1417 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX)
1418 		bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1419 
1420 	SLIST_INIT(&sc->txdma_list);
1421 	for (i = 0; i < BGE_RSLOTS; i++) {
1422 		if (bus_dmamap_create(sc->bge_dmatag, BGE_TXDMA_MAX,
1423 		    BGE_NTXSEG, ETHER_MAX_LEN_JUMBO, 0, BUS_DMA_NOWAIT,
1424 		    &dmamap))
1425 			return (ENOBUFS);
1426 		if (dmamap == NULL)
1427 			panic("dmamap NULL in bge_init_tx_ring");
1428 		dma = malloc(sizeof(*dma), M_DEVBUF, M_NOWAIT);
1429 		if (dma == NULL) {
1430 			aprint_error_dev(sc->bge_dev,
1431 			    "can't alloc txdmamap_pool_entry\n");
1432 			bus_dmamap_destroy(sc->bge_dmatag, dmamap);
1433 			return (ENOMEM);
1434 		}
1435 		dma->dmamap = dmamap;
1436 		SLIST_INSERT_HEAD(&sc->txdma_list, dma, link);
1437 	}
1438 
1439 	sc->bge_flags |= BGE_TXRING_VALID;
1440 
1441 	return (0);
1442 }
1443 
1444 static void
1445 bge_setmulti(struct bge_softc *sc)
1446 {
1447 	struct ethercom		*ac = &sc->ethercom;
1448 	struct ifnet		*ifp = &ac->ec_if;
1449 	struct ether_multi	*enm;
1450 	struct ether_multistep  step;
1451 	u_int32_t		hashes[4] = { 0, 0, 0, 0 };
1452 	u_int32_t		h;
1453 	int			i;
1454 
1455 	if (ifp->if_flags & IFF_PROMISC)
1456 		goto allmulti;
1457 
1458 	/* Now program new ones. */
1459 	ETHER_FIRST_MULTI(step, ac, enm);
1460 	while (enm != NULL) {
1461 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
1462 			/*
1463 			 * We must listen to a range of multicast addresses.
1464 			 * For now, just accept all multicasts, rather than
1465 			 * trying to set only those filter bits needed to match
1466 			 * the range.  (At this time, the only use of address
1467 			 * ranges is for IP multicast routing, for which the
1468 			 * range is big enough to require all bits set.)
1469 			 */
1470 			goto allmulti;
1471 		}
1472 
1473 		h = ether_crc32_le(enm->enm_addrlo, ETHER_ADDR_LEN);
1474 
1475 		/* Just want the 7 least-significant bits. */
1476 		h &= 0x7f;
1477 
1478 		hashes[(h & 0x60) >> 5] |= 1 << (h & 0x1F);
1479 		ETHER_NEXT_MULTI(step, enm);
1480 	}
1481 
1482 	ifp->if_flags &= ~IFF_ALLMULTI;
1483 	goto setit;
1484 
1485  allmulti:
1486 	ifp->if_flags |= IFF_ALLMULTI;
1487 	hashes[0] = hashes[1] = hashes[2] = hashes[3] = 0xffffffff;
1488 
1489  setit:
1490 	for (i = 0; i < 4; i++)
1491 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), hashes[i]);
1492 }
1493 
1494 const int bge_swapbits[] = {
1495 	0,
1496 	BGE_MODECTL_BYTESWAP_DATA,
1497 	BGE_MODECTL_WORDSWAP_DATA,
1498 	BGE_MODECTL_BYTESWAP_NONFRAME,
1499 	BGE_MODECTL_WORDSWAP_NONFRAME,
1500 
1501 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_WORDSWAP_DATA,
1502 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_BYTESWAP_NONFRAME,
1503 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_WORDSWAP_NONFRAME,
1504 
1505 	BGE_MODECTL_WORDSWAP_DATA|BGE_MODECTL_BYTESWAP_NONFRAME,
1506 	BGE_MODECTL_WORDSWAP_DATA|BGE_MODECTL_WORDSWAP_NONFRAME,
1507 
1508 	BGE_MODECTL_BYTESWAP_NONFRAME|BGE_MODECTL_WORDSWAP_NONFRAME,
1509 
1510 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_WORDSWAP_DATA|
1511 	    BGE_MODECTL_BYTESWAP_NONFRAME,
1512 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_WORDSWAP_DATA|
1513 	    BGE_MODECTL_WORDSWAP_NONFRAME,
1514 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_BYTESWAP_NONFRAME|
1515 	    BGE_MODECTL_WORDSWAP_NONFRAME,
1516 	BGE_MODECTL_WORDSWAP_DATA|BGE_MODECTL_BYTESWAP_NONFRAME|
1517 	    BGE_MODECTL_WORDSWAP_NONFRAME,
1518 
1519 	BGE_MODECTL_BYTESWAP_DATA|BGE_MODECTL_WORDSWAP_DATA|
1520 	    BGE_MODECTL_BYTESWAP_NONFRAME|BGE_MODECTL_WORDSWAP_NONFRAME,
1521 };
1522 
1523 int bge_swapindex = 0;
1524 
1525 /*
1526  * Do endian, PCI and DMA initialization. Also check the on-board ROM
1527  * self-test results.
1528  */
1529 static int
1530 bge_chipinit(struct bge_softc *sc)
1531 {
1532 	int			i;
1533 	u_int32_t		dma_rw_ctl;
1534 
1535 
1536 	/* Set endianness before we access any non-PCI registers. */
1537 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL,
1538 	    BGE_INIT);
1539 
1540 	/* Set power state to D0. */
1541 	bge_setpowerstate(sc, 0);
1542 
1543 	/*
1544 	 * Check the 'ROM failed' bit on the RX CPU to see if
1545 	 * self-tests passed.
1546 	 */
1547 	if (CSR_READ_4(sc, BGE_RXCPU_MODE) & BGE_RXCPUMODE_ROMFAIL) {
1548 		aprint_error_dev(sc->bge_dev,
1549 		    "RX CPU self-diagnostics failed!\n");
1550 		return (ENODEV);
1551 	}
1552 
1553 	/* Clear the MAC control register */
1554 	CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
1555 
1556 	/*
1557 	 * Clear the MAC statistics block in the NIC's
1558 	 * internal memory.
1559 	 */
1560 	for (i = BGE_STATS_BLOCK;
1561 	    i < BGE_STATS_BLOCK_END + 1; i += sizeof(u_int32_t))
1562 		BGE_MEMWIN_WRITE(sc->sc_pc, sc->sc_pcitag, i, 0);
1563 
1564 	for (i = BGE_STATUS_BLOCK;
1565 	    i < BGE_STATUS_BLOCK_END + 1; i += sizeof(u_int32_t))
1566 		BGE_MEMWIN_WRITE(sc->sc_pc, sc->sc_pcitag, i, 0);
1567 
1568 	/* Set up the PCI DMA control register. */
1569 	dma_rw_ctl = BGE_PCI_READ_CMD | BGE_PCI_WRITE_CMD;
1570 	if (sc->bge_flags & BGE_PCIE) {
1571 		/* Read watermark not used, 128 bytes for write. */
1572 		DPRINTFN(4, ("(%s: PCI-Express DMA setting)\n",
1573 		    device_xname(sc->bge_dev)));
1574 		dma_rw_ctl |= (0x3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1575 	} else if (sc->bge_flags & BGE_PCIX){
1576 	  	DPRINTFN(4, ("(:%s: PCI-X DMA setting)\n",
1577 		    device_xname(sc->bge_dev)));
1578 		/* PCI-X bus */
1579 		dma_rw_ctl |= (0x3 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1580 		    (0x3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT) |
1581 		    (0x0F);
1582 		/*
1583 		 * 5703 and 5704 need ONEDMA_AT_ONCE as a workaround
1584 		 * for hardware bugs, which means we should also clear
1585 		 * the low-order MINDMA bits.  In addition, the 5704
1586 		 * uses a different encoding of read/write watermarks.
1587 		 */
1588 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704) {
1589 			dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD |
1590 			  /* should be 0x1f0000 */
1591 			  (0x7 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1592 			  (0x3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1593 			dma_rw_ctl |= BGE_PCIDMARWCTL_ONEDMA_ATONCE;
1594 		}
1595 		else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703) {
1596 			dma_rw_ctl &=  0xfffffff0;
1597 			dma_rw_ctl |= BGE_PCIDMARWCTL_ONEDMA_ATONCE;
1598 		}
1599 		else if (BGE_IS_5714_FAMILY(sc)) {
1600 			dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD;
1601 			dma_rw_ctl &= ~BGE_PCIDMARWCTL_ONEDMA_ATONCE; /* XXX */
1602 			/* XXX magic values, Broadcom-supplied Linux driver */
1603 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5780)
1604 				dma_rw_ctl |= (1 << 20) | (1 << 18) |
1605 				  BGE_PCIDMARWCTL_ONEDMA_ATONCE;
1606 			else
1607 				dma_rw_ctl |= (1<<20) | (1<<18) | (1 << 15);
1608 		}
1609 	} else {
1610 		/* Conventional PCI bus */
1611 	  	DPRINTFN(4, ("(%s: PCI 2.2 DMA setting)\n",
1612 		    device_xname(sc->bge_dev)));
1613 		dma_rw_ctl = (0x7 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1614 		   (0x7 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1615 		if (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5705 &&
1616 		    BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5750)
1617 			dma_rw_ctl |= 0x0F;
1618 	}
1619 
1620 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
1621 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5701)
1622 		dma_rw_ctl |= BGE_PCIDMARWCTL_USE_MRM |
1623 		    BGE_PCIDMARWCTL_ASRT_ALL_BE;
1624 
1625 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5703 ||
1626 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704)
1627 		dma_rw_ctl &= ~BGE_PCIDMARWCTL_MINDMA;
1628 
1629 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_DMA_RW_CTL, dma_rw_ctl);
1630 
1631 	/*
1632 	 * Set up general mode register.
1633 	 */
1634 	CSR_WRITE_4(sc, BGE_MODE_CTL, BGE_DMA_SWAP_OPTIONS |
1635 		    BGE_MODECTL_MAC_ATTN_INTR | BGE_MODECTL_HOST_SEND_BDS |
1636 		    BGE_MODECTL_TX_NO_PHDR_CSUM | BGE_MODECTL_RX_NO_PHDR_CSUM);
1637 
1638 	/*
1639 	 * Disable memory write invalidate.  Apparently it is not supported
1640 	 * properly by these devices.
1641 	 */
1642 	PCI_CLRBIT(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD, PCIM_CMD_MWIEN);
1643 
1644 
1645 #ifdef __brokenalpha__
1646 	/*
1647 	 * Must insure that we do not cross an 8K (bytes) boundary
1648 	 * for DMA reads.  Our highest limit is 1K bytes.  This is a
1649 	 * restriction on some ALPHA platforms with early revision
1650 	 * 21174 PCI chipsets, such as the AlphaPC 164lx
1651 	 */
1652 	PCI_SETBIT(sc, BGE_PCI_DMA_RW_CTL, BGE_PCI_READ_BNDRY_1024, 4);
1653 #endif
1654 
1655 	/* Set the timer prescaler (always 66MHz) */
1656 	CSR_WRITE_4(sc, BGE_MISC_CFG, 65 << 1/*BGE_32BITTIME_66MHZ*/);
1657 
1658 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
1659 		DELAY(40);	/* XXX */
1660 
1661 		/* Put PHY into ready state */
1662 		BGE_CLRBIT(sc, BGE_MISC_CFG, BGE_MISCCFG_EPHY_IDDQ);
1663 		CSR_READ_4(sc, BGE_MISC_CFG); /* Flush */
1664 		DELAY(40);
1665 	}
1666 
1667 	return (0);
1668 }
1669 
1670 static int
1671 bge_blockinit(struct bge_softc *sc)
1672 {
1673 	volatile struct bge_rcb		*rcb;
1674 	bus_size_t		rcb_addr;
1675 	int			i;
1676 	struct ifnet		*ifp = &sc->ethercom.ec_if;
1677 	bge_hostaddr		taddr;
1678 	u_int32_t		val;
1679 
1680 	/*
1681 	 * Initialize the memory window pointer register so that
1682 	 * we can access the first 32K of internal NIC RAM. This will
1683 	 * allow us to set up the TX send ring RCBs and the RX return
1684 	 * ring RCBs, plus other things which live in NIC memory.
1685 	 */
1686 
1687 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MEMWIN_BASEADDR, 0);
1688 
1689 	/* Configure mbuf memory pool */
1690 	if (!(BGE_IS_5705_OR_BEYOND(sc))) {
1691 		if (sc->bge_extram) {
1692 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_BASEADDR,
1693 			    BGE_EXT_SSRAM);
1694 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704)
1695 				CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x10000);
1696 			else
1697 				CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x18000);
1698 		} else {
1699 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_BASEADDR,
1700 			    BGE_BUFFPOOL_1);
1701 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704)
1702 				CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x10000);
1703 			else
1704 				CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x18000);
1705 		}
1706 
1707 		/* Configure DMA resource pool */
1708 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_BASEADDR,
1709 		    BGE_DMA_DESCRIPTORS);
1710 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LEN, 0x2000);
1711 	}
1712 
1713 	/* Configure mbuf pool watermarks */
1714 #ifdef ORIG_WPAUL_VALUES
1715 	CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 24);
1716 	CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 24);
1717 	CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 48);
1718 #else
1719 
1720 	/* new broadcom docs strongly recommend these: */
1721 	if (!BGE_IS_5705_OR_BEYOND(sc)) {
1722 		if (ifp->if_mtu > ETHER_MAX_LEN) {
1723 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x50);
1724 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x20);
1725 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1726 		} else {
1727 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 304);
1728 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 152);
1729 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 380);
1730 		}
1731 	} else if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
1732 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1733 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x04);
1734 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x10);
1735 	} else {
1736 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1737 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x10);
1738 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1739 	}
1740 #endif
1741 
1742 	/* Configure DMA resource watermarks */
1743 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LOWAT, 5);
1744 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_HIWAT, 10);
1745 
1746 	/* Enable buffer manager */
1747 	if (!BGE_IS_5705_OR_BEYOND(sc)) {
1748 		CSR_WRITE_4(sc, BGE_BMAN_MODE,
1749 		    BGE_BMANMODE_ENABLE | BGE_BMANMODE_LOMBUF_ATTN);
1750 
1751 		/* Poll for buffer manager start indication */
1752 		for (i = 0; i < BGE_TIMEOUT; i++) {
1753 			if (CSR_READ_4(sc, BGE_BMAN_MODE) & BGE_BMANMODE_ENABLE)
1754 				break;
1755 			DELAY(10);
1756 		}
1757 
1758 		if (i == BGE_TIMEOUT) {
1759 			aprint_error_dev(sc->bge_dev,
1760 			    "buffer manager failed to start\n");
1761 			return (ENXIO);
1762 		}
1763 	}
1764 
1765 	/* Enable flow-through queues */
1766 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
1767 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
1768 
1769 	/* Wait until queue initialization is complete */
1770 	for (i = 0; i < BGE_TIMEOUT; i++) {
1771 		if (CSR_READ_4(sc, BGE_FTQ_RESET) == 0)
1772 			break;
1773 		DELAY(10);
1774 	}
1775 
1776 	if (i == BGE_TIMEOUT) {
1777 		aprint_error_dev(sc->bge_dev,
1778 		    "flow-through queue init failed\n");
1779 		return (ENXIO);
1780 	}
1781 
1782 	/* Initialize the standard RX ring control block */
1783 	rcb = &sc->bge_rdata->bge_info.bge_std_rx_rcb;
1784 	bge_set_hostaddr(&rcb->bge_hostaddr,
1785 	    BGE_RING_DMA_ADDR(sc, bge_rx_std_ring));
1786 	if (BGE_IS_5705_OR_BEYOND(sc))
1787 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(512, 0);
1788 	else
1789 		rcb->bge_maxlen_flags =
1790 		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN, 0);
1791 	if (sc->bge_extram)
1792 		rcb->bge_nicaddr = BGE_EXT_STD_RX_RINGS;
1793 	else
1794 		rcb->bge_nicaddr = BGE_STD_RX_RINGS;
1795 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_HI, rcb->bge_hostaddr.bge_addr_hi);
1796 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_LO, rcb->bge_hostaddr.bge_addr_lo);
1797 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1798 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_NICADDR, rcb->bge_nicaddr);
1799 
1800 	if (BGE_IS_5705_OR_BEYOND(sc))
1801 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT_5705;
1802 	else
1803 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT;
1804 
1805 	/*
1806 	 * Initialize the jumbo RX ring control block
1807 	 * We set the 'ring disabled' bit in the flags
1808 	 * field until we're actually ready to start
1809 	 * using this ring (i.e. once we set the MTU
1810 	 * high enough to require it).
1811 	 */
1812 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
1813 		rcb = &sc->bge_rdata->bge_info.bge_jumbo_rx_rcb;
1814 		bge_set_hostaddr(&rcb->bge_hostaddr,
1815 		    BGE_RING_DMA_ADDR(sc, bge_rx_jumbo_ring));
1816 		rcb->bge_maxlen_flags =
1817 		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN,
1818 			BGE_RCB_FLAG_RING_DISABLED);
1819 		if (sc->bge_extram)
1820 			rcb->bge_nicaddr = BGE_EXT_JUMBO_RX_RINGS;
1821 		else
1822 			rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS;
1823 
1824 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_HI,
1825 		    rcb->bge_hostaddr.bge_addr_hi);
1826 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_LO,
1827 		    rcb->bge_hostaddr.bge_addr_lo);
1828 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS,
1829 		    rcb->bge_maxlen_flags);
1830 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_NICADDR, rcb->bge_nicaddr);
1831 
1832 		/* Set up dummy disabled mini ring RCB */
1833 		rcb = &sc->bge_rdata->bge_info.bge_mini_rx_rcb;
1834 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(0,
1835 		    BGE_RCB_FLAG_RING_DISABLED);
1836 		CSR_WRITE_4(sc, BGE_RX_MINI_RCB_MAXLEN_FLAGS,
1837 		    rcb->bge_maxlen_flags);
1838 
1839 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
1840 		    offsetof(struct bge_ring_data, bge_info),
1841 		    sizeof (struct bge_gib),
1842 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1843 	}
1844 
1845 	/*
1846 	 * Set the BD ring replenish thresholds. The recommended
1847 	 * values are 1/8th the number of descriptors allocated to
1848 	 * each ring.
1849 	 */
1850 	i = BGE_STD_RX_RING_CNT / 8;
1851 
1852 	/*
1853 	 * Use a value of 8 for the following chips to workaround HW errata.
1854 	 * Some of these chips have been added based on empirical
1855 	 * evidence (they don't work unless this is done).
1856 	 */
1857 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5750 ||
1858 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5752 ||
1859 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 ||
1860 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5787 ||
1861 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906)
1862 		i = 8;
1863 
1864 	CSR_WRITE_4(sc, BGE_RBDI_STD_REPL_THRESH, i);
1865 	CSR_WRITE_4(sc, BGE_RBDI_JUMBO_REPL_THRESH, BGE_JUMBO_RX_RING_CNT / 8);
1866 
1867 	/*
1868 	 * Disable all unused send rings by setting the 'ring disabled'
1869 	 * bit in the flags field of all the TX send ring control blocks.
1870 	 * These are located in NIC memory.
1871 	 */
1872 	rcb_addr = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1873 	for (i = 0; i < BGE_TX_RINGS_EXTSSRAM_MAX; i++) {
1874 		RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
1875 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED));
1876 		RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, 0);
1877 		rcb_addr += sizeof(struct bge_rcb);
1878 	}
1879 
1880 	/* Configure TX RCB 0 (we use only the first ring) */
1881 	rcb_addr = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1882 	bge_set_hostaddr(&taddr, BGE_RING_DMA_ADDR(sc, bge_tx_ring));
1883 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1884 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1885 	RCB_WRITE_4(sc, rcb_addr, bge_nicaddr,
1886 		    BGE_NIC_TXRING_ADDR(0, BGE_TX_RING_CNT));
1887 	if (!(BGE_IS_5705_OR_BEYOND(sc)))
1888 		RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
1889 		    BGE_RCB_MAXLEN_FLAGS(BGE_TX_RING_CNT, 0));
1890 
1891 	/* Disable all unused RX return rings */
1892 	rcb_addr = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1893 	for (i = 0; i < BGE_RX_RINGS_MAX; i++) {
1894 		RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_hi, 0);
1895 		RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_lo, 0);
1896 		RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
1897 			    BGE_RCB_MAXLEN_FLAGS(sc->bge_return_ring_cnt,
1898                                      BGE_RCB_FLAG_RING_DISABLED));
1899 		RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, 0);
1900 		bge_writembx(sc, BGE_MBX_RX_CONS0_LO +
1901 		    (i * (sizeof(u_int64_t))), 0);
1902 		rcb_addr += sizeof(struct bge_rcb);
1903 	}
1904 
1905 	/* Initialize RX ring indexes */
1906 	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, 0);
1907 	bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, 0);
1908 	bge_writembx(sc, BGE_MBX_RX_MINI_PROD_LO, 0);
1909 
1910 	/*
1911 	 * Set up RX return ring 0
1912 	 * Note that the NIC address for RX return rings is 0x00000000.
1913 	 * The return rings live entirely within the host, so the
1914 	 * nicaddr field in the RCB isn't used.
1915 	 */
1916 	rcb_addr = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1917 	bge_set_hostaddr(&taddr, BGE_RING_DMA_ADDR(sc, bge_rx_return_ring));
1918 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1919 	RCB_WRITE_4(sc, rcb_addr, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1920 	RCB_WRITE_4(sc, rcb_addr, bge_nicaddr, 0x00000000);
1921 	RCB_WRITE_4(sc, rcb_addr, bge_maxlen_flags,
1922 	    BGE_RCB_MAXLEN_FLAGS(sc->bge_return_ring_cnt, 0));
1923 
1924 	/* Set random backoff seed for TX */
1925 	CSR_WRITE_4(sc, BGE_TX_RANDOM_BACKOFF,
1926 	    CLLADDR(ifp->if_sadl)[0] + CLLADDR(ifp->if_sadl)[1] +
1927 	    CLLADDR(ifp->if_sadl)[2] + CLLADDR(ifp->if_sadl)[3] +
1928 	    CLLADDR(ifp->if_sadl)[4] + CLLADDR(ifp->if_sadl)[5] +
1929 	    BGE_TX_BACKOFF_SEED_MASK);
1930 
1931 	/* Set inter-packet gap */
1932 	CSR_WRITE_4(sc, BGE_TX_LENGTHS, 0x2620);
1933 
1934 	/*
1935 	 * Specify which ring to use for packets that don't match
1936 	 * any RX rules.
1937 	 */
1938 	CSR_WRITE_4(sc, BGE_RX_RULES_CFG, 0x08);
1939 
1940 	/*
1941 	 * Configure number of RX lists. One interrupt distribution
1942 	 * list, sixteen active lists, one bad frames class.
1943 	 */
1944 	CSR_WRITE_4(sc, BGE_RXLP_CFG, 0x181);
1945 
1946 	/* Inialize RX list placement stats mask. */
1947 	CSR_WRITE_4(sc, BGE_RXLP_STATS_ENABLE_MASK, 0x007FFFFF);
1948 	CSR_WRITE_4(sc, BGE_RXLP_STATS_CTL, 0x1);
1949 
1950 	/* Disable host coalescing until we get it set up */
1951 	CSR_WRITE_4(sc, BGE_HCC_MODE, 0x00000000);
1952 
1953 	/* Poll to make sure it's shut down. */
1954 	for (i = 0; i < BGE_TIMEOUT; i++) {
1955 		if (!(CSR_READ_4(sc, BGE_HCC_MODE) & BGE_HCCMODE_ENABLE))
1956 			break;
1957 		DELAY(10);
1958 	}
1959 
1960 	if (i == BGE_TIMEOUT) {
1961 		aprint_error_dev(sc->bge_dev,
1962 		    "host coalescing engine failed to idle\n");
1963 		return (ENXIO);
1964 	}
1965 
1966 	/* Set up host coalescing defaults */
1967 	CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, sc->bge_rx_coal_ticks);
1968 	CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS, sc->bge_tx_coal_ticks);
1969 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, sc->bge_rx_max_coal_bds);
1970 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS, sc->bge_tx_max_coal_bds);
1971 	if (!(BGE_IS_5705_OR_BEYOND(sc))) {
1972 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS_INT, 0);
1973 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS_INT, 0);
1974 	}
1975 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT, 0);
1976 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT, 0);
1977 
1978 	/* Set up address of statistics block */
1979 	if (!(BGE_IS_5705_OR_BEYOND(sc))) {
1980 		bge_set_hostaddr(&taddr,
1981 		    BGE_RING_DMA_ADDR(sc, bge_info.bge_stats));
1982 		CSR_WRITE_4(sc, BGE_HCC_STATS_TICKS, sc->bge_stat_ticks);
1983 		CSR_WRITE_4(sc, BGE_HCC_STATS_BASEADDR, BGE_STATS_BLOCK);
1984 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_HI, taddr.bge_addr_hi);
1985 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_LO, taddr.bge_addr_lo);
1986 	}
1987 
1988 	/* Set up address of status block */
1989 	bge_set_hostaddr(&taddr, BGE_RING_DMA_ADDR(sc, bge_status_block));
1990 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_BASEADDR, BGE_STATUS_BLOCK);
1991 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_HI, taddr.bge_addr_hi);
1992 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_LO, taddr.bge_addr_lo);
1993 	sc->bge_rdata->bge_status_block.bge_idx[0].bge_rx_prod_idx = 0;
1994 	sc->bge_rdata->bge_status_block.bge_idx[0].bge_tx_cons_idx = 0;
1995 
1996 	/* Turn on host coalescing state machine */
1997 	CSR_WRITE_4(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE);
1998 
1999 	/* Turn on RX BD completion state machine and enable attentions */
2000 	CSR_WRITE_4(sc, BGE_RBDC_MODE,
2001 	    BGE_RBDCMODE_ENABLE | BGE_RBDCMODE_ATTN);
2002 
2003 	/* Turn on RX list placement state machine */
2004 	CSR_WRITE_4(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
2005 
2006 	/* Turn on RX list selector state machine. */
2007 	if (!(BGE_IS_5705_OR_BEYOND(sc)))
2008 		CSR_WRITE_4(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
2009 
2010 	val = BGE_MACMODE_TXDMA_ENB | BGE_MACMODE_RXDMA_ENB |
2011 	    BGE_MACMODE_RX_STATS_CLEAR | BGE_MACMODE_TX_STATS_CLEAR |
2012 	    BGE_MACMODE_RX_STATS_ENB | BGE_MACMODE_TX_STATS_ENB |
2013 	    BGE_MACMODE_FRMHDR_DMA_ENB;
2014 
2015 	if (sc->bge_flags & BGE_PHY_FIBER_TBI)
2016 	    val |= BGE_PORTMODE_TBI;
2017 	else if (sc->bge_flags & BGE_PHY_FIBER_MII)
2018 	    val |= BGE_PORTMODE_GMII;
2019 	else
2020 	    val |= BGE_PORTMODE_MII;
2021 
2022 	/* Turn on DMA, clear stats */
2023 	CSR_WRITE_4(sc, BGE_MAC_MODE, val);
2024 
2025 
2026 	/* Set misc. local control, enable interrupts on attentions */
2027 	sc->bge_local_ctrl_reg = BGE_MLC_INTR_ONATTN | BGE_MLC_AUTO_EEPROM;
2028 
2029 #ifdef notdef
2030 	/* Assert GPIO pins for PHY reset */
2031 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUT0|
2032 	    BGE_MLC_MISCIO_OUT1|BGE_MLC_MISCIO_OUT2);
2033 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUTEN0|
2034 	    BGE_MLC_MISCIO_OUTEN1|BGE_MLC_MISCIO_OUTEN2);
2035 #endif
2036 
2037 #if defined(not_quite_yet)
2038 	/* Linux driver enables enable gpio pin #1 on 5700s */
2039 	if (sc->bge_chipid == BGE_CHIPID_BCM5700) {
2040 		sc->bge_local_ctrl_reg |=
2041 		  (BGE_MLC_MISCIO_OUT1|BGE_MLC_MISCIO_OUTEN1);
2042 	}
2043 #endif
2044 	CSR_WRITE_4(sc, BGE_MISC_LOCAL_CTL, sc->bge_local_ctrl_reg);
2045 
2046 	/* Turn on DMA completion state machine */
2047 	if (!(BGE_IS_5705_OR_BEYOND(sc)))
2048 		CSR_WRITE_4(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
2049 
2050 	/* Turn on write DMA state machine */
2051 	{
2052 		uint32_t bge_wdma_mode =
2053 			BGE_WDMAMODE_ENABLE|BGE_WDMAMODE_ALL_ATTNS;
2054 
2055 		/* Enable host coalescing bug fix; see Linux tg3.c */
2056 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 ||
2057 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5787)
2058 			bge_wdma_mode |= (1 << 29);
2059 
2060 		CSR_WRITE_4(sc, BGE_WDMA_MODE, bge_wdma_mode);
2061         }
2062 
2063 	/* Turn on read DMA state machine */
2064 	{
2065 		uint32_t dma_read_modebits;
2066 
2067 		dma_read_modebits =
2068 		  BGE_RDMAMODE_ENABLE | BGE_RDMAMODE_ALL_ATTNS;
2069 
2070 		if ((sc->bge_flags & BGE_PCIE) && 0) {
2071 			dma_read_modebits |= BGE_RDMA_MODE_FIFO_LONG_BURST;
2072 		} else if (BGE_IS_5705_OR_BEYOND(sc)) {
2073 			dma_read_modebits |= BGE_RDMA_MODE_FIFO_SIZE_128;
2074 		}
2075 
2076 		/* XXX broadcom-supplied linux driver; undocumented */
2077 		if (BGE_IS_5750_OR_BEYOND(sc)) {
2078  			/*
2079 			 * XXX: magic values.
2080 			 * From Broadcom-supplied Linux driver;  apparently
2081 			 * required to workaround a DMA bug affecting TSO
2082 			 * on bcm575x/bcm5721?
2083 			 */
2084 			dma_read_modebits |= (1 << 27);
2085 		}
2086 		CSR_WRITE_4(sc, BGE_RDMA_MODE, dma_read_modebits);
2087 	}
2088 
2089 	/* Turn on RX data completion state machine */
2090 	CSR_WRITE_4(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
2091 
2092 	/* Turn on RX BD initiator state machine */
2093 	CSR_WRITE_4(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
2094 
2095 	/* Turn on RX data and RX BD initiator state machine */
2096 	CSR_WRITE_4(sc, BGE_RDBDI_MODE, BGE_RDBDIMODE_ENABLE);
2097 
2098 	/* Turn on Mbuf cluster free state machine */
2099 	if (!(BGE_IS_5705_OR_BEYOND(sc)))
2100 		CSR_WRITE_4(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
2101 
2102 	/* Turn on send BD completion state machine */
2103 	CSR_WRITE_4(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
2104 
2105 	/* Turn on send data completion state machine */
2106 	CSR_WRITE_4(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE);
2107 
2108 	/* Turn on send data initiator state machine */
2109 	if (BGE_IS_5750_OR_BEYOND(sc)) {
2110 		/* XXX: magic value from Linux driver */
2111 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE | 0x08);
2112 	} else {
2113 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
2114 	}
2115 
2116 	/* Turn on send BD initiator state machine */
2117 	CSR_WRITE_4(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
2118 
2119 	/* Turn on send BD selector state machine */
2120 	CSR_WRITE_4(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
2121 
2122 	CSR_WRITE_4(sc, BGE_SDI_STATS_ENABLE_MASK, 0x007FFFFF);
2123 	CSR_WRITE_4(sc, BGE_SDI_STATS_CTL,
2124 	    BGE_SDISTATSCTL_ENABLE | BGE_SDISTATSCTL_FASTER);
2125 
2126 	/* ack/clear link change events */
2127 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
2128 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
2129 	    BGE_MACSTAT_CFG_CHANGED);
2130 	CSR_WRITE_4(sc, BGE_MI_STS, 0);
2131 
2132 	/* Enable PHY auto polling (for MII/GMII only) */
2133 	if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
2134 		CSR_WRITE_4(sc, BGE_MI_STS, BGE_MISTS_LINK);
2135  	} else {
2136 		BGE_STS_SETBIT(sc, BGE_STS_AUTOPOLL);
2137 		BGE_SETBIT(sc, BGE_MI_MODE, BGE_MIMODE_AUTOPOLL | (10 << 16));
2138 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700)
2139 			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
2140 			    BGE_EVTENB_MI_INTERRUPT);
2141 	}
2142 
2143 	/*
2144 	 * Clear any pending link state attention.
2145 	 * Otherwise some link state change events may be lost until attention
2146 	 * is cleared by bge_intr() -> bge_link_upd() sequence.
2147 	 * It's not necessary on newer BCM chips - perhaps enabling link
2148 	 * state change attentions implies clearing pending attention.
2149 	 */
2150 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
2151 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
2152 	    BGE_MACSTAT_LINK_CHANGED);
2153 
2154 	/* Enable link state change attentions. */
2155 	BGE_SETBIT(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_LINK_CHANGED);
2156 
2157 	return (0);
2158 }
2159 
2160 static const struct bge_revision *
2161 bge_lookup_rev(uint32_t chipid)
2162 {
2163 	const struct bge_revision *br;
2164 
2165 	for (br = bge_revisions; br->br_name != NULL; br++) {
2166 		if (br->br_chipid == chipid)
2167 			return (br);
2168 	}
2169 
2170 	for (br = bge_majorrevs; br->br_name != NULL; br++) {
2171 		if (br->br_chipid == BGE_ASICREV(chipid))
2172 			return (br);
2173 	}
2174 
2175 	return (NULL);
2176 }
2177 
2178 static const struct bge_product *
2179 bge_lookup(const struct pci_attach_args *pa)
2180 {
2181 	const struct bge_product *bp;
2182 
2183 	for (bp = bge_products; bp->bp_name != NULL; bp++) {
2184 		if (PCI_VENDOR(pa->pa_id) == bp->bp_vendor &&
2185 		    PCI_PRODUCT(pa->pa_id) == bp->bp_product)
2186 			return (bp);
2187 	}
2188 
2189 	return (NULL);
2190 }
2191 
2192 static int
2193 bge_setpowerstate(struct bge_softc *sc, int powerlevel)
2194 {
2195 #ifdef NOTYET
2196 	u_int32_t pm_ctl = 0;
2197 
2198 	/* XXX FIXME: make sure indirect accesses enabled? */
2199 	pm_ctl = pci_conf_read(sc->bge_dev, BGE_PCI_MISC_CTL, 4);
2200 	pm_ctl |= BGE_PCIMISCCTL_INDIRECT_ACCESS;
2201 	pci_write_config(sc->bge_dev, BGE_PCI_MISC_CTL, pm_ctl, 4);
2202 
2203 	/* clear the PME_assert bit and power state bits, enable PME */
2204 	pm_ctl = pci_conf_read(sc->bge_dev, BGE_PCI_PWRMGMT_CMD, 2);
2205 	pm_ctl &= ~PCIM_PSTAT_DMASK;
2206 	pm_ctl |= (1 << 8);
2207 
2208 	if (powerlevel == 0) {
2209 		pm_ctl |= PCIM_PSTAT_D0;
2210 		pci_write_config(sc->bge_dev, BGE_PCI_PWRMGMT_CMD,
2211 		    pm_ctl, 2);
2212 		DELAY(10000);
2213 		CSR_WRITE_4(sc, BGE_MISC_LOCAL_CTL, sc->bge_local_ctrl_reg);
2214 		DELAY(10000);
2215 
2216 #ifdef NOTYET
2217 		/* XXX FIXME: write 0x02 to phy aux_Ctrl reg */
2218 		bge_miibus_writereg(sc->bge_dev, 1, 0x18, 0x02);
2219 #endif
2220 		DELAY(40); DELAY(40); DELAY(40);
2221 		DELAY(10000);	/* above not quite adequate on 5700 */
2222 		return 0;
2223 	}
2224 
2225 
2226 	/*
2227 	 * Entering ACPI power states D1-D3 is achieved by wiggling
2228 	 * GMII gpio pins. Example code assumes all hardware vendors
2229 	 * followed Broadom's sample pcb layout. Until we verify that
2230 	 * for all supported OEM cards, states D1-D3 are  unsupported.
2231 	 */
2232 	aprint_error_dev(sc->bge_dev,
2233 	    "power state %d unimplemented; check GPIO pins\n",
2234 	    powerlevel);
2235 #endif
2236 	return EOPNOTSUPP;
2237 }
2238 
2239 
2240 /*
2241  * Probe for a Broadcom chip. Check the PCI vendor and device IDs
2242  * against our list and return its name if we find a match. Note
2243  * that since the Broadcom controller contains VPD support, we
2244  * can get the device name string from the controller itself instead
2245  * of the compiled-in string. This is a little slow, but it guarantees
2246  * we'll always announce the right product name.
2247  */
2248 static int
2249 bge_probe(device_t parent, cfdata_t match, void *aux)
2250 {
2251 	struct pci_attach_args *pa = (struct pci_attach_args *)aux;
2252 
2253 	if (bge_lookup(pa) != NULL)
2254 		return (1);
2255 
2256 	return (0);
2257 }
2258 
2259 static void
2260 bge_attach(device_t parent, device_t self, void *aux)
2261 {
2262 	struct bge_softc	*sc = device_private(self);
2263 	struct pci_attach_args	*pa = aux;
2264 	prop_dictionary_t dict;
2265 	const struct bge_product *bp;
2266 	const struct bge_revision *br;
2267 	pci_chipset_tag_t	pc;
2268 	pci_intr_handle_t	ih;
2269 	const char		*intrstr = NULL;
2270 	bus_dma_segment_t	seg;
2271 	int			rseg;
2272 	u_int32_t		hwcfg = 0;
2273 	u_int32_t		command;
2274 	struct ifnet		*ifp;
2275 	u_int32_t		misccfg;
2276 	void *			kva;
2277 	u_char			eaddr[ETHER_ADDR_LEN];
2278 	pcireg_t		memtype;
2279 	bus_addr_t		memaddr;
2280 	bus_size_t		memsize;
2281 	u_int32_t		pm_ctl;
2282 
2283 	bp = bge_lookup(pa);
2284 	KASSERT(bp != NULL);
2285 
2286 	sc->sc_pc = pa->pa_pc;
2287 	sc->sc_pcitag = pa->pa_tag;
2288 	sc->bge_dev = self;
2289 
2290 	aprint_naive(": Ethernet controller\n");
2291 	aprint_normal(": %s\n", bp->bp_name);
2292 
2293 	/*
2294 	 * Map control/status registers.
2295 	 */
2296 	DPRINTFN(5, ("Map control/status regs\n"));
2297 	pc = sc->sc_pc;
2298 	command = pci_conf_read(pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG);
2299 	command |= PCI_COMMAND_MEM_ENABLE | PCI_COMMAND_MASTER_ENABLE;
2300 	pci_conf_write(pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG, command);
2301 	command = pci_conf_read(pc, sc->sc_pcitag, PCI_COMMAND_STATUS_REG);
2302 
2303 	if (!(command & PCI_COMMAND_MEM_ENABLE)) {
2304 		aprint_error_dev(sc->bge_dev,
2305 		    "failed to enable memory mapping!\n");
2306 		return;
2307 	}
2308 
2309 	DPRINTFN(5, ("pci_mem_find\n"));
2310 	memtype = pci_mapreg_type(sc->sc_pc, sc->sc_pcitag, BGE_PCI_BAR0);
2311  	switch (memtype) {
2312 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_32BIT:
2313 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_64BIT:
2314 		if (pci_mapreg_map(pa, BGE_PCI_BAR0,
2315 		    memtype, 0, &sc->bge_btag, &sc->bge_bhandle,
2316 		    &memaddr, &memsize) == 0)
2317 			break;
2318 	default:
2319 		aprint_error_dev(sc->bge_dev, "can't find mem space\n");
2320 		return;
2321 	}
2322 
2323 	DPRINTFN(5, ("pci_intr_map\n"));
2324 	if (pci_intr_map(pa, &ih)) {
2325 		aprint_error_dev(sc->bge_dev, "couldn't map interrupt\n");
2326 		return;
2327 	}
2328 
2329 	DPRINTFN(5, ("pci_intr_string\n"));
2330 	intrstr = pci_intr_string(pc, ih);
2331 
2332 	DPRINTFN(5, ("pci_intr_establish\n"));
2333 	sc->bge_intrhand = pci_intr_establish(pc, ih, IPL_NET, bge_intr, sc);
2334 
2335 	if (sc->bge_intrhand == NULL) {
2336 		aprint_error_dev(sc->bge_dev,
2337 		    "couldn't establish interrupt%s%s\n",
2338 		    intrstr ? " at " : "", intrstr ? intrstr : "");
2339 		return;
2340 	}
2341 	aprint_normal_dev(sc->bge_dev, "interrupting at %s\n", intrstr);
2342 
2343 	/*
2344 	 * Kludge for 5700 Bx bug: a hardware bug (PCIX byte enable?)
2345 	 * can clobber the chip's PCI config-space power control registers,
2346 	 * leaving the card in D3 powersave state.
2347 	 * We do not have memory-mapped registers in this state,
2348 	 * so force device into D0 state before starting initialization.
2349 	 */
2350 	pm_ctl = pci_conf_read(pc, sc->sc_pcitag, BGE_PCI_PWRMGMT_CMD);
2351 	pm_ctl &= ~(PCI_PWR_D0|PCI_PWR_D1|PCI_PWR_D2|PCI_PWR_D3);
2352 	pm_ctl |= (1 << 8) | PCI_PWR_D0 ; /* D0 state */
2353 	pci_conf_write(pc, sc->sc_pcitag, BGE_PCI_PWRMGMT_CMD, pm_ctl);
2354 	DELAY(1000);	/* 27 usec is allegedly sufficent */
2355 
2356 	/*
2357 	 * Save ASIC rev.
2358 	 */
2359 	sc->bge_chipid =
2360 	    pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL) &
2361 	    BGE_PCIMISCCTL_ASICREV;
2362 
2363 	/*
2364 	 * Detect PCI-Express devices
2365 	 * XXX: guessed from Linux/FreeBSD; no documentation
2366 	 */
2367 	if (pci_get_capability(sc->sc_pc, sc->sc_pcitag, PCI_CAP_PCIEXPRESS,
2368 	        NULL, NULL) != 0)
2369 		sc->bge_flags |= BGE_PCIE;
2370 
2371 	/*
2372 	 * PCI-X check.
2373 	 */
2374 	if ((pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_PCISTATE) &
2375 		BGE_PCISTATE_PCI_BUSMODE) == 0)
2376 		sc->bge_flags |= BGE_PCIX;
2377 
2378 	if (sc->bge_chipid == BGE_CHIPID_BCM5701_A0 ||
2379 	    sc->bge_chipid == BGE_CHIPID_BCM5701_B0)
2380 		sc->bge_flags |= BGE_PHY_CRC_BUG;
2381 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5703_AX ||
2382 	    BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5704_AX)
2383 		sc->bge_flags |= BGE_PHY_ADC_BUG;
2384 	if (sc->bge_chipid == BGE_CHIPID_BCM5704_A0)
2385 		sc->bge_flags |= BGE_PHY_5704_A0_BUG;
2386 
2387 	if (BGE_IS_5705_OR_BEYOND(sc)) {
2388 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 ||
2389 		    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5787) {
2390 			if (PCI_PRODUCT(pa->pa_id) != PCI_PRODUCT_BROADCOM_BCM5722 &&
2391 			    PCI_PRODUCT(pa->pa_id) != PCI_PRODUCT_BROADCOM_BCM5756)
2392 				sc->bge_flags |= BGE_PHY_JITTER_BUG;
2393 			if (PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_BROADCOM_BCM5755M)
2394 				sc->bge_flags |= BGE_PHY_ADJUST_TRIM;
2395 		} else if (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5906)
2396 			sc->bge_flags |= BGE_PHY_BER_BUG;
2397 	}
2398 
2399 	/* Try to reset the chip. */
2400 	DPRINTFN(5, ("bge_reset\n"));
2401 	bge_reset(sc);
2402 
2403 	if (bge_chipinit(sc)) {
2404 		aprint_error_dev(sc->bge_dev, "chip initialization failed\n");
2405 		bge_release_resources(sc);
2406 		return;
2407 	}
2408 
2409 	/*
2410 	 * Get station address from the EEPROM.
2411 	 */
2412 	if (bge_get_eaddr(sc, eaddr)) {
2413 		aprint_error_dev(sc->bge_dev,
2414 		"failed to reade station address\n");
2415 		bge_release_resources(sc);
2416 		return;
2417 	}
2418 
2419 	br = bge_lookup_rev(sc->bge_chipid);
2420 
2421 	if (br == NULL) {
2422 		aprint_normal_dev(sc->bge_dev, "unknown ASIC (0x%04x)",
2423 		    sc->bge_chipid >> 16);
2424 	} else {
2425 		aprint_normal_dev(sc->bge_dev, "ASIC %s (0x%04x)",
2426 		    br->br_name, sc->bge_chipid >> 16);
2427 	}
2428 	aprint_normal(", Ethernet address %s\n", ether_sprintf(eaddr));
2429 
2430 	/* Allocate the general information block and ring buffers. */
2431 	if (pci_dma64_available(pa))
2432 		sc->bge_dmatag = pa->pa_dmat64;
2433 	else
2434 		sc->bge_dmatag = pa->pa_dmat;
2435 	DPRINTFN(5, ("bus_dmamem_alloc\n"));
2436 	if (bus_dmamem_alloc(sc->bge_dmatag, sizeof(struct bge_ring_data),
2437 			     PAGE_SIZE, 0, &seg, 1, &rseg, BUS_DMA_NOWAIT)) {
2438 		aprint_error_dev(sc->bge_dev, "can't alloc rx buffers\n");
2439 		return;
2440 	}
2441 	DPRINTFN(5, ("bus_dmamem_map\n"));
2442 	if (bus_dmamem_map(sc->bge_dmatag, &seg, rseg,
2443 			   sizeof(struct bge_ring_data), &kva,
2444 			   BUS_DMA_NOWAIT)) {
2445 		aprint_error_dev(sc->bge_dev,
2446 		    "can't map DMA buffers (%zu bytes)\n",
2447 		    sizeof(struct bge_ring_data));
2448 		bus_dmamem_free(sc->bge_dmatag, &seg, rseg);
2449 		return;
2450 	}
2451 	DPRINTFN(5, ("bus_dmamem_create\n"));
2452 	if (bus_dmamap_create(sc->bge_dmatag, sizeof(struct bge_ring_data), 1,
2453 	    sizeof(struct bge_ring_data), 0,
2454 	    BUS_DMA_NOWAIT, &sc->bge_ring_map)) {
2455 		aprint_error_dev(sc->bge_dev, "can't create DMA map\n");
2456 		bus_dmamem_unmap(sc->bge_dmatag, kva,
2457 				 sizeof(struct bge_ring_data));
2458 		bus_dmamem_free(sc->bge_dmatag, &seg, rseg);
2459 		return;
2460 	}
2461 	DPRINTFN(5, ("bus_dmamem_load\n"));
2462 	if (bus_dmamap_load(sc->bge_dmatag, sc->bge_ring_map, kva,
2463 			    sizeof(struct bge_ring_data), NULL,
2464 			    BUS_DMA_NOWAIT)) {
2465 		bus_dmamap_destroy(sc->bge_dmatag, sc->bge_ring_map);
2466 		bus_dmamem_unmap(sc->bge_dmatag, kva,
2467 				 sizeof(struct bge_ring_data));
2468 		bus_dmamem_free(sc->bge_dmatag, &seg, rseg);
2469 		return;
2470 	}
2471 
2472 	DPRINTFN(5, ("bzero\n"));
2473 	sc->bge_rdata = (struct bge_ring_data *)kva;
2474 
2475 	memset(sc->bge_rdata, 0, sizeof(struct bge_ring_data));
2476 
2477 	/* Try to allocate memory for jumbo buffers. */
2478 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
2479 		if (bge_alloc_jumbo_mem(sc)) {
2480 			aprint_error_dev(sc->bge_dev,
2481 			    "jumbo buffer allocation failed\n");
2482 		} else
2483 			sc->ethercom.ec_capabilities |= ETHERCAP_JUMBO_MTU;
2484 	}
2485 
2486 	/* Set default tuneable values. */
2487 	sc->bge_stat_ticks = BGE_TICKS_PER_SEC;
2488 	sc->bge_rx_coal_ticks = 150;
2489 	sc->bge_rx_max_coal_bds = 64;
2490 #ifdef ORIG_WPAUL_VALUES
2491 	sc->bge_tx_coal_ticks = 150;
2492 	sc->bge_tx_max_coal_bds = 128;
2493 #else
2494 	sc->bge_tx_coal_ticks = 300;
2495 	sc->bge_tx_max_coal_bds = 400;
2496 #endif
2497 	if (BGE_IS_5705_OR_BEYOND(sc)) {
2498 		sc->bge_tx_coal_ticks = (12 * 5);
2499 		sc->bge_tx_max_coal_bds = (12 * 5);
2500 			aprint_verbose_dev(sc->bge_dev,
2501 			    "setting short Tx thresholds\n");
2502 	}
2503 
2504 	/* Set up ifnet structure */
2505 	ifp = &sc->ethercom.ec_if;
2506 	ifp->if_softc = sc;
2507 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2508 	ifp->if_ioctl = bge_ioctl;
2509 	ifp->if_stop = bge_stop;
2510 	ifp->if_start = bge_start;
2511 	ifp->if_init = bge_init;
2512 	ifp->if_watchdog = bge_watchdog;
2513 	IFQ_SET_MAXLEN(&ifp->if_snd, max(BGE_TX_RING_CNT - 1, IFQ_MAXLEN));
2514 	IFQ_SET_READY(&ifp->if_snd);
2515 	DPRINTFN(5, ("strcpy if_xname\n"));
2516 	strcpy(ifp->if_xname, device_xname(sc->bge_dev));
2517 
2518 	if (sc->bge_chipid != BGE_CHIPID_BCM5700_B0)
2519 		sc->ethercom.ec_if.if_capabilities |=
2520 		    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
2521 		    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
2522 		    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
2523 	sc->ethercom.ec_capabilities |=
2524 	    ETHERCAP_VLAN_HWTAGGING | ETHERCAP_VLAN_MTU;
2525 
2526 	if (sc->bge_flags & BGE_PCIE)
2527 		sc->ethercom.ec_if.if_capabilities |= IFCAP_TSOv4;
2528 
2529 	/*
2530 	 * Do MII setup.
2531 	 */
2532 	DPRINTFN(5, ("mii setup\n"));
2533 	sc->bge_mii.mii_ifp = ifp;
2534 	sc->bge_mii.mii_readreg = bge_miibus_readreg;
2535 	sc->bge_mii.mii_writereg = bge_miibus_writereg;
2536 	sc->bge_mii.mii_statchg = bge_miibus_statchg;
2537 
2538 	/*
2539 	 * Figure out what sort of media we have by checking the
2540 	 * hardware config word in the first 32k of NIC internal memory,
2541 	 * or fall back to the config word in the EEPROM. Note: on some BCM5700
2542 	 * cards, this value appears to be unset. If that's the
2543 	 * case, we have to rely on identifying the NIC by its PCI
2544 	 * subsystem ID, as we do below for the SysKonnect SK-9D41.
2545 	 */
2546 	if (bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_SIG) == BGE_MAGIC_NUMBER) {
2547 		hwcfg = bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_NICCFG);
2548 	} else {
2549 		bge_read_eeprom(sc, (void *)&hwcfg,
2550 		    BGE_EE_HWCFG_OFFSET, sizeof(hwcfg));
2551 		hwcfg = be32toh(hwcfg);
2552 	}
2553 	/* The SysKonnect SK-9D41 is a 1000baseSX card. */
2554 	if (PCI_PRODUCT(pa->pa_id) == SK_SUBSYSID_9D41 ||
2555 	    (hwcfg & BGE_HWCFG_MEDIA) == BGE_MEDIA_FIBER) {
2556 		if (BGE_IS_5714_FAMILY(sc))
2557 		    sc->bge_flags |= BGE_PHY_FIBER_MII;
2558 		else
2559 		    sc->bge_flags |= BGE_PHY_FIBER_TBI;
2560 	}
2561 
2562 	/* set phyflags before mii_attach() */
2563 	dict = device_properties(self);
2564 	prop_dictionary_set_uint32(dict, "phyflags", sc->bge_flags);
2565 
2566 	if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
2567 		ifmedia_init(&sc->bge_ifmedia, IFM_IMASK, bge_ifmedia_upd,
2568 		    bge_ifmedia_sts);
2569 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER|IFM_1000_SX, 0, NULL);
2570 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER|IFM_1000_SX|IFM_FDX,
2571 			    0, NULL);
2572 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER|IFM_AUTO, 0, NULL);
2573 		ifmedia_set(&sc->bge_ifmedia, IFM_ETHER|IFM_AUTO);
2574 		/* Pretend the user requested this setting */
2575 		sc->bge_ifmedia.ifm_media = sc->bge_ifmedia.ifm_cur->ifm_media;
2576 	} else {
2577 		/*
2578 		 * Do transceiver setup.
2579 		 */
2580 		ifmedia_init(&sc->bge_mii.mii_media, 0, bge_ifmedia_upd,
2581 			     bge_ifmedia_sts);
2582 		mii_attach(sc->bge_dev, &sc->bge_mii, 0xffffffff,
2583 			   MII_PHY_ANY, MII_OFFSET_ANY,
2584 			   MIIF_FORCEANEG|MIIF_DOPAUSE);
2585 
2586 		if (LIST_EMPTY(&sc->bge_mii.mii_phys)) {
2587 			aprint_error_dev(sc->bge_dev, "no PHY found!\n");
2588 			ifmedia_add(&sc->bge_mii.mii_media,
2589 				    IFM_ETHER|IFM_MANUAL, 0, NULL);
2590 			ifmedia_set(&sc->bge_mii.mii_media,
2591 				    IFM_ETHER|IFM_MANUAL);
2592 		} else
2593 			ifmedia_set(&sc->bge_mii.mii_media,
2594 				    IFM_ETHER|IFM_AUTO);
2595 	}
2596 
2597 	/*
2598 	 * When using the BCM5701 in PCI-X mode, data corruption has
2599 	 * been observed in the first few bytes of some received packets.
2600 	 * Aligning the packet buffer in memory eliminates the corruption.
2601 	 * Unfortunately, this misaligns the packet payloads.  On platforms
2602 	 * which do not support unaligned accesses, we will realign the
2603 	 * payloads by copying the received packets.
2604 	 */
2605 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5701 &&
2606 		sc->bge_flags & BGE_PCIX)
2607 		sc->bge_flags |= BGE_RX_ALIGNBUG;
2608 
2609 	misccfg = CSR_READ_4(sc, BGE_MISC_CFG);
2610 	misccfg &= BGE_MISCCFG_BOARD_ID_MASK;
2611 
2612 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5705 &&
2613 	    (misccfg == BGE_MISCCFG_BOARD_ID_5788 ||
2614 	     misccfg == BGE_MISCCFG_BOARD_ID_5788M))
2615 		sc->bge_flags |= BGE_IS_5788;
2616 
2617 	/*
2618 	 * Call MI attach routine.
2619 	 */
2620 	DPRINTFN(5, ("if_attach\n"));
2621 	if_attach(ifp);
2622 	DPRINTFN(5, ("ether_ifattach\n"));
2623 	ether_ifattach(ifp, eaddr);
2624 #if NRND > 0
2625 	rnd_attach_source(&sc->rnd_source, device_xname(sc->bge_dev),
2626 		RND_TYPE_NET, 0);
2627 #endif
2628 #ifdef BGE_EVENT_COUNTERS
2629 	/*
2630 	 * Attach event counters.
2631 	 */
2632 	evcnt_attach_dynamic(&sc->bge_ev_intr, EVCNT_TYPE_INTR,
2633 	    NULL, device_xname(sc->bge_dev), "intr");
2634 	evcnt_attach_dynamic(&sc->bge_ev_tx_xoff, EVCNT_TYPE_MISC,
2635 	    NULL, device_xname(sc->bge_dev), "tx_xoff");
2636 	evcnt_attach_dynamic(&sc->bge_ev_tx_xon, EVCNT_TYPE_MISC,
2637 	    NULL, device_xname(sc->bge_dev), "tx_xon");
2638 	evcnt_attach_dynamic(&sc->bge_ev_rx_xoff, EVCNT_TYPE_MISC,
2639 	    NULL, device_xname(sc->bge_dev), "rx_xoff");
2640 	evcnt_attach_dynamic(&sc->bge_ev_rx_xon, EVCNT_TYPE_MISC,
2641 	    NULL, device_xname(sc->bge_dev), "rx_xon");
2642 	evcnt_attach_dynamic(&sc->bge_ev_rx_macctl, EVCNT_TYPE_MISC,
2643 	    NULL, device_xname(sc->bge_dev), "rx_macctl");
2644 	evcnt_attach_dynamic(&sc->bge_ev_xoffentered, EVCNT_TYPE_MISC,
2645 	    NULL, device_xname(sc->bge_dev), "xoffentered");
2646 #endif /* BGE_EVENT_COUNTERS */
2647 	DPRINTFN(5, ("callout_init\n"));
2648 	callout_init(&sc->bge_timeout, 0);
2649 
2650 	if (pmf_device_register(self, NULL, NULL))
2651 		pmf_class_network_register(self, ifp);
2652 	else
2653 		aprint_error_dev(self, "couldn't establish power handler\n");
2654 }
2655 
2656 static void
2657 bge_release_resources(struct bge_softc *sc)
2658 {
2659 	if (sc->bge_vpd_prodname != NULL)
2660 		free(sc->bge_vpd_prodname, M_DEVBUF);
2661 
2662 	if (sc->bge_vpd_readonly != NULL)
2663 		free(sc->bge_vpd_readonly, M_DEVBUF);
2664 }
2665 
2666 static void
2667 bge_reset(struct bge_softc *sc)
2668 {
2669 	u_int32_t cachesize, command, pcistate, new_pcistate;
2670 	int i, val;
2671 	void (*write_op)(struct bge_softc *, int, int);
2672 
2673 	if (BGE_IS_5750_OR_BEYOND(sc) && !BGE_IS_5714_FAMILY(sc) &&
2674 	    (BGE_ASICREV(sc->bge_chipid) != BGE_ASICREV_BCM5906)) {
2675 	    	if (sc->bge_flags & BGE_PCIE) {
2676 			write_op = bge_writemem_direct;
2677 		} else {
2678 			write_op = bge_writemem_ind;
2679 		}
2680 	} else {
2681 		write_op = bge_writereg_ind;
2682 	}
2683 
2684 
2685 	/* Save some important PCI state. */
2686 	cachesize = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CACHESZ);
2687 	command = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD);
2688 	pcistate = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_PCISTATE);
2689 
2690 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL,
2691 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2692 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW);
2693 
2694 	/* Disable fastboot on controllers that support it. */
2695 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5752 ||
2696 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5755 ||
2697 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5787)
2698 		CSR_WRITE_4(sc, BGE_FASTBOOT_PC, 0);
2699 
2700 	val = BGE_MISCCFG_RESET_CORE_CLOCKS | (65<<1);
2701 	/*
2702 	 * XXX: from FreeBSD/Linux; no documentation
2703 	 */
2704 	if (sc->bge_flags & BGE_PCIE) {
2705 		if (CSR_READ_4(sc, BGE_PCIE_CTL1) == 0x60)
2706 			/* PCI Express 1.0 system */
2707 			CSR_WRITE_4(sc, BGE_PCIE_CTL1, 0x20);
2708 		if (sc->bge_chipid != BGE_CHIPID_BCM5750_A0) {
2709 			/*
2710 			 * Prevent PCI Express link training
2711 			 * during global reset.
2712 			 */
2713 			CSR_WRITE_4(sc, BGE_MISC_CFG, 1 << 29);
2714 			val |= (1<<29);
2715 		}
2716 	}
2717 
2718 	/*
2719 	 * Set GPHY Power Down Override to leave GPHY
2720 	 * powered up in D0 uninitialized.
2721 	 */
2722 	if (BGE_IS_5705_OR_BEYOND(sc))
2723 		val |= BGE_MISCCFG_KEEP_GPHY_POWER;
2724 
2725 	/* Issue global reset */
2726 	write_op(sc, BGE_MISC_CFG, val);
2727 
2728 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
2729 		i = CSR_READ_4(sc, BGE_VCPU_STATUS);
2730 		CSR_WRITE_4(sc, BGE_VCPU_STATUS,
2731 		    i | BGE_VCPU_STATUS_DRV_RESET);
2732 		i = CSR_READ_4(sc, BGE_VCPU_EXT_CTRL);
2733 		CSR_WRITE_4(sc, BGE_VCPU_EXT_CTRL,
2734 		    i & ~BGE_VCPU_EXT_CTRL_HALT_CPU);
2735 	}
2736 
2737 	DELAY(1000);
2738 
2739 	/*
2740 	 * XXX: from FreeBSD/Linux; no documentation
2741 	 */
2742 	if (sc->bge_flags & BGE_PCIE) {
2743 		if (sc->bge_chipid == BGE_CHIPID_BCM5750_A0) {
2744 			pcireg_t reg;
2745 
2746 			DELAY(500000);
2747 			/* XXX: Magic Numbers */
2748 			reg = pci_conf_read(sc->sc_pc, sc->sc_pcitag, BGE_PCI_UNKNOWN0);
2749 			pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_UNKNOWN0,
2750 			    reg | (1 << 15));
2751 		}
2752 		/*
2753 		 * XXX: Magic Numbers.
2754 		 * Sets maximal PCI-e payload and clears any PCI-e errors.
2755 		 * Should be replaced with references to PCI config-space
2756 		 * capability block for PCI-Express.
2757 		 */
2758 		pci_conf_write(sc->sc_pc, sc->sc_pcitag,
2759 		    BGE_PCI_CONF_DEV_CTRL, 0xf5000);
2760 
2761 	}
2762 
2763 	/* Reset some of the PCI state that got zapped by reset */
2764 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_MISC_CTL,
2765 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2766 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW);
2767 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CMD, command);
2768 	pci_conf_write(sc->sc_pc, sc->sc_pcitag, BGE_PCI_CACHESZ, cachesize);
2769 	write_op(sc, BGE_MISC_CFG, (65 << 1));
2770 
2771 	/* Enable memory arbiter. */
2772 	{
2773 		uint32_t marbmode = 0;
2774 		if (BGE_IS_5714_FAMILY(sc)) {
2775 			marbmode = CSR_READ_4(sc, BGE_MARB_MODE);
2776 		}
2777  		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE | marbmode);
2778 	}
2779 
2780 
2781 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
2782 		for (i = 0; i < BGE_TIMEOUT; i++) {
2783 			val = CSR_READ_4(sc, BGE_VCPU_STATUS);
2784 			if (val & BGE_VCPU_STATUS_INIT_DONE)
2785 				break;
2786 			DELAY(100);
2787 		}
2788 		if (i == BGE_TIMEOUT) {
2789 			aprint_error_dev(sc->bge_dev, "reset timed out\n");
2790 			return;
2791 		}
2792 	} else {
2793 		/*
2794 		 * Write the magic number to the firmware mailbox at 0xb50
2795 		 * so that the driver can synchronize with the firmware.
2796 		 */
2797 		bge_writemem_ind(sc, BGE_SOFTWARE_GENCOMM, BGE_MAGIC_NUMBER);
2798 
2799 		/*
2800 		 * Poll the value location we just wrote until
2801 		 * we see the 1's complement of the magic number.
2802 		 * This indicates that the firmware initialization
2803 		 * is complete.
2804 		 */
2805 		for (i = 0; i < BGE_TIMEOUT; i++) {
2806 			val = bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM);
2807 			if (val == ~BGE_MAGIC_NUMBER)
2808 				break;
2809 			DELAY(1000);
2810 		}
2811 
2812 		if (i >= BGE_TIMEOUT) {
2813 			aprint_error_dev(sc->bge_dev,
2814 			    "firmware handshake timed out, val = %x\n", val);
2815 			/*
2816 			 * XXX: occasionally fired on bcm5721, but without
2817 			 * apparent harm.  For now, keep going if we timeout
2818 			 * against PCI-E devices.
2819 			 */
2820 			if ((sc->bge_flags & BGE_PCIE) == 0)
2821 				  return;
2822 		}
2823 	}
2824 
2825 	/*
2826 	 * XXX Wait for the value of the PCISTATE register to
2827 	 * return to its original pre-reset state. This is a
2828 	 * fairly good indicator of reset completion. If we don't
2829 	 * wait for the reset to fully complete, trying to read
2830 	 * from the device's non-PCI registers may yield garbage
2831 	 * results.
2832 	 */
2833 	for (i = 0; i < 10000; i++) {
2834 		new_pcistate = pci_conf_read(sc->sc_pc, sc->sc_pcitag,
2835 		    BGE_PCI_PCISTATE);
2836 		if ((new_pcistate & ~BGE_PCISTATE_RESERVED) ==
2837 		    (pcistate & ~BGE_PCISTATE_RESERVED))
2838 			break;
2839 		DELAY(10);
2840 	}
2841 	if ((new_pcistate & ~BGE_PCISTATE_RESERVED) !=
2842 	    (pcistate & ~BGE_PCISTATE_RESERVED)) {
2843 		aprint_error_dev(sc->bge_dev, "pcistate failed to revert\n");
2844 	}
2845 
2846 	/* Enable memory arbiter. */
2847 	/* XXX why do this twice? */
2848 	{
2849 		uint32_t marbmode = 0;
2850 		if (BGE_IS_5714_FAMILY(sc)) {
2851 			marbmode = CSR_READ_4(sc, BGE_MARB_MODE);
2852 		}
2853  		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE | marbmode);
2854 	}
2855 
2856 	/* Fix up byte swapping */
2857 	CSR_WRITE_4(sc, BGE_MODE_CTL, BGE_DMA_SWAP_OPTIONS);
2858 
2859 	CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
2860 
2861 	/*
2862 	 * The 5704 in TBI mode apparently needs some special
2863 	 * adjustment to insure the SERDES drive level is set
2864 	 * to 1.2V.
2865 	 */
2866 	if (sc->bge_flags & BGE_PHY_FIBER_TBI &&
2867 	    BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704) {
2868 		u_int32_t serdescfg;
2869 
2870 		serdescfg = CSR_READ_4(sc, BGE_SERDES_CFG);
2871 		serdescfg = (serdescfg & ~0xFFF) | 0x880;
2872 		CSR_WRITE_4(sc, BGE_SERDES_CFG, serdescfg);
2873 	}
2874 
2875 	/* XXX: from FreeBSD/Linux; no documentation */
2876 	if (sc->bge_flags & BGE_PCIE &&
2877 	    sc->bge_chipid != BGE_CHIPID_BCM5750_A0)
2878 		CSR_WRITE_4(sc, BGE_PCIE_CTL0, CSR_READ_4(sc, BGE_PCIE_CTL0) | (1<<25));
2879 	DELAY(10000);
2880 }
2881 
2882 /*
2883  * Frame reception handling. This is called if there's a frame
2884  * on the receive return list.
2885  *
2886  * Note: we have to be able to handle two possibilities here:
2887  * 1) the frame is from the jumbo recieve ring
2888  * 2) the frame is from the standard receive ring
2889  */
2890 
2891 static void
2892 bge_rxeof(struct bge_softc *sc)
2893 {
2894 	struct ifnet *ifp;
2895 	int stdcnt = 0, jumbocnt = 0;
2896 	bus_dmamap_t dmamap;
2897 	bus_addr_t offset, toff;
2898 	bus_size_t tlen;
2899 	int tosync;
2900 
2901 	ifp = &sc->ethercom.ec_if;
2902 
2903 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
2904 	    offsetof(struct bge_ring_data, bge_status_block),
2905 	    sizeof (struct bge_status_block),
2906 	    BUS_DMASYNC_POSTREAD);
2907 
2908 	offset = offsetof(struct bge_ring_data, bge_rx_return_ring);
2909 	tosync = sc->bge_rdata->bge_status_block.bge_idx[0].bge_rx_prod_idx -
2910 	    sc->bge_rx_saved_considx;
2911 
2912 #if NRND > 0
2913 	if (tosync != 0 && RND_ENABLED(&sc->rnd_source))
2914 		rnd_add_uint32(&sc->rnd_source, tosync);
2915 #endif
2916 
2917 	toff = offset + (sc->bge_rx_saved_considx * sizeof (struct bge_rx_bd));
2918 
2919 	if (tosync < 0) {
2920 		tlen = (sc->bge_return_ring_cnt - sc->bge_rx_saved_considx) *
2921 		    sizeof (struct bge_rx_bd);
2922 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
2923 		    toff, tlen, BUS_DMASYNC_POSTREAD);
2924 		tosync = -tosync;
2925 	}
2926 
2927 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
2928 	    offset, tosync * sizeof (struct bge_rx_bd),
2929 	    BUS_DMASYNC_POSTREAD);
2930 
2931 	while(sc->bge_rx_saved_considx !=
2932 	    sc->bge_rdata->bge_status_block.bge_idx[0].bge_rx_prod_idx) {
2933 		struct bge_rx_bd	*cur_rx;
2934 		u_int32_t		rxidx;
2935 		struct mbuf		*m = NULL;
2936 
2937 		cur_rx = &sc->bge_rdata->
2938 			bge_rx_return_ring[sc->bge_rx_saved_considx];
2939 
2940 		rxidx = cur_rx->bge_idx;
2941 		BGE_INC(sc->bge_rx_saved_considx, sc->bge_return_ring_cnt);
2942 
2943 		if (cur_rx->bge_flags & BGE_RXBDFLAG_JUMBO_RING) {
2944 			BGE_INC(sc->bge_jumbo, BGE_JUMBO_RX_RING_CNT);
2945 			m = sc->bge_cdata.bge_rx_jumbo_chain[rxidx];
2946 			sc->bge_cdata.bge_rx_jumbo_chain[rxidx] = NULL;
2947 			jumbocnt++;
2948 			bus_dmamap_sync(sc->bge_dmatag,
2949 			    sc->bge_cdata.bge_rx_jumbo_map,
2950 			    mtod(m, char *) - (char *)sc->bge_cdata.bge_jumbo_buf,
2951 			    BGE_JLEN, BUS_DMASYNC_POSTREAD);
2952 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
2953 				ifp->if_ierrors++;
2954 				bge_newbuf_jumbo(sc, sc->bge_jumbo, m);
2955 				continue;
2956 			}
2957 			if (bge_newbuf_jumbo(sc, sc->bge_jumbo,
2958 					     NULL)== ENOBUFS) {
2959 				ifp->if_ierrors++;
2960 				bge_newbuf_jumbo(sc, sc->bge_jumbo, m);
2961 				continue;
2962 			}
2963 		} else {
2964 			BGE_INC(sc->bge_std, BGE_STD_RX_RING_CNT);
2965 			m = sc->bge_cdata.bge_rx_std_chain[rxidx];
2966 
2967 			sc->bge_cdata.bge_rx_std_chain[rxidx] = NULL;
2968 			stdcnt++;
2969 			dmamap = sc->bge_cdata.bge_rx_std_map[rxidx];
2970 			sc->bge_cdata.bge_rx_std_map[rxidx] = 0;
2971 			bus_dmamap_sync(sc->bge_dmatag, dmamap, 0,
2972 			    dmamap->dm_mapsize, BUS_DMASYNC_POSTREAD);
2973 			bus_dmamap_unload(sc->bge_dmatag, dmamap);
2974 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
2975 				ifp->if_ierrors++;
2976 				bge_newbuf_std(sc, sc->bge_std, m, dmamap);
2977 				continue;
2978 			}
2979 			if (bge_newbuf_std(sc, sc->bge_std,
2980 			    NULL, dmamap) == ENOBUFS) {
2981 				ifp->if_ierrors++;
2982 				bge_newbuf_std(sc, sc->bge_std, m, dmamap);
2983 				continue;
2984 			}
2985 		}
2986 
2987 		ifp->if_ipackets++;
2988 #ifndef __NO_STRICT_ALIGNMENT
2989                 /*
2990                  * XXX: if the 5701 PCIX-Rx-DMA workaround is in effect,
2991                  * the Rx buffer has the layer-2 header unaligned.
2992                  * If our CPU requires alignment, re-align by copying.
2993                  */
2994 		if (sc->bge_flags & BGE_RX_ALIGNBUG) {
2995 			memmove(mtod(m, char *) + ETHER_ALIGN, m->m_data,
2996                                 cur_rx->bge_len);
2997 			m->m_data += ETHER_ALIGN;
2998 		}
2999 #endif
3000 
3001 		m->m_pkthdr.len = m->m_len = cur_rx->bge_len - ETHER_CRC_LEN;
3002 		m->m_pkthdr.rcvif = ifp;
3003 
3004 #if NBPFILTER > 0
3005 		/*
3006 		 * Handle BPF listeners. Let the BPF user see the packet.
3007 		 */
3008 		if (ifp->if_bpf)
3009 			bpf_mtap(ifp->if_bpf, m);
3010 #endif
3011 
3012 		m->m_pkthdr.csum_flags = M_CSUM_IPv4;
3013 
3014 		if ((cur_rx->bge_ip_csum ^ 0xffff) != 0)
3015 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
3016 		/*
3017 		 * Rx transport checksum-offload may also
3018 		 * have bugs with packets which, when transmitted,
3019 		 * were `runts' requiring padding.
3020 		 */
3021 		if (cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM &&
3022 		    (/* (sc->_bge_quirks & BGE_QUIRK_SHORT_CKSUM_BUG) == 0 ||*/
3023 		     m->m_pkthdr.len >= ETHER_MIN_NOPAD)) {
3024 			m->m_pkthdr.csum_data =
3025 			    cur_rx->bge_tcp_udp_csum;
3026 			m->m_pkthdr.csum_flags |=
3027 			    (M_CSUM_TCPv4|M_CSUM_UDPv4|
3028 			     M_CSUM_DATA|M_CSUM_NO_PSEUDOHDR);
3029 		}
3030 
3031 		/*
3032 		 * If we received a packet with a vlan tag, pass it
3033 		 * to vlan_input() instead of ether_input().
3034 		 */
3035 		if (cur_rx->bge_flags & BGE_RXBDFLAG_VLAN_TAG) {
3036 			VLAN_INPUT_TAG(ifp, m, cur_rx->bge_vlan_tag, continue);
3037 		}
3038 
3039 		(*ifp->if_input)(ifp, m);
3040 	}
3041 
3042 	bge_writembx(sc, BGE_MBX_RX_CONS0_LO, sc->bge_rx_saved_considx);
3043 	if (stdcnt)
3044 		bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bge_std);
3045 	if (jumbocnt)
3046 		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
3047 }
3048 
3049 static void
3050 bge_txeof(struct bge_softc *sc)
3051 {
3052 	struct bge_tx_bd *cur_tx = NULL;
3053 	struct ifnet *ifp;
3054 	struct txdmamap_pool_entry *dma;
3055 	bus_addr_t offset, toff;
3056 	bus_size_t tlen;
3057 	int tosync;
3058 	struct mbuf *m;
3059 
3060 	ifp = &sc->ethercom.ec_if;
3061 
3062 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
3063 	    offsetof(struct bge_ring_data, bge_status_block),
3064 	    sizeof (struct bge_status_block),
3065 	    BUS_DMASYNC_POSTREAD);
3066 
3067 	offset = offsetof(struct bge_ring_data, bge_tx_ring);
3068 	tosync = sc->bge_rdata->bge_status_block.bge_idx[0].bge_tx_cons_idx -
3069 	    sc->bge_tx_saved_considx;
3070 
3071 #if NRND > 0
3072 	if (tosync != 0 && RND_ENABLED(&sc->rnd_source))
3073 		rnd_add_uint32(&sc->rnd_source, tosync);
3074 #endif
3075 
3076 	toff = offset + (sc->bge_tx_saved_considx * sizeof (struct bge_tx_bd));
3077 
3078 	if (tosync < 0) {
3079 		tlen = (BGE_TX_RING_CNT - sc->bge_tx_saved_considx) *
3080 		    sizeof (struct bge_tx_bd);
3081 		bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
3082 		    toff, tlen, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
3083 		tosync = -tosync;
3084 	}
3085 
3086 	bus_dmamap_sync(sc->bge_dmatag, sc->bge_ring_map,
3087 	    offset, tosync * sizeof (struct bge_tx_bd),
3088 	    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
3089 
3090 	/*
3091 	 * Go through our tx ring and free mbufs for those
3092 	 * frames that have been sent.
3093 	 */
3094 	while (sc->bge_tx_saved_considx !=
3095 	    sc->bge_rdata->bge_status_block.bge_idx[0].bge_tx_cons_idx) {
3096 		u_int32_t		idx = 0;
3097 
3098 		idx = sc->bge_tx_saved_considx;
3099 		cur_tx = &sc->bge_rdata->bge_tx_ring[idx];
3100 		if (cur_tx->bge_flags & BGE_TXBDFLAG_END)
3101 			ifp->if_opackets++;
3102 		m = sc->bge_cdata.bge_tx_chain[idx];
3103 		if (m != NULL) {
3104 			sc->bge_cdata.bge_tx_chain[idx] = NULL;
3105 			dma = sc->txdma[idx];
3106 			bus_dmamap_sync(sc->bge_dmatag, dma->dmamap, 0,
3107 			    dma->dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
3108 			bus_dmamap_unload(sc->bge_dmatag, dma->dmamap);
3109 			SLIST_INSERT_HEAD(&sc->txdma_list, dma, link);
3110 			sc->txdma[idx] = NULL;
3111 
3112 			m_freem(m);
3113 		}
3114 		sc->bge_txcnt--;
3115 		BGE_INC(sc->bge_tx_saved_considx, BGE_TX_RING_CNT);
3116 		ifp->if_timer = 0;
3117 	}
3118 
3119 	if (cur_tx != NULL)
3120 		ifp->if_flags &= ~IFF_OACTIVE;
3121 }
3122 
3123 static int
3124 bge_intr(void *xsc)
3125 {
3126 	struct bge_softc *sc;
3127 	struct ifnet *ifp;
3128 	uint32_t statusword;
3129 
3130 	sc = xsc;
3131 	ifp = &sc->ethercom.ec_if;
3132 
3133 	/* It is possible for the interrupt to arrive before
3134 	 * the status block is updated prior to the interrupt.
3135 	 * Reading the PCI State register will confirm whether the
3136 	 * interrupt is ours and will flush the status block.
3137 	 */
3138 
3139 	/* read status word from status block */
3140 	statusword = sc->bge_rdata->bge_status_block.bge_status;
3141 
3142 	if ((statusword & BGE_STATFLAG_UPDATED) ||
3143 	    (!(CSR_READ_4(sc, BGE_PCI_PCISTATE) & BGE_PCISTATE_INTR_NOT_ACTIVE))) {
3144 		/* Ack interrupt and stop others from occuring. */
3145 		bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
3146 
3147 		BGE_EVCNT_INCR(sc->bge_ev_intr);
3148 
3149 		/* clear status word */
3150 		sc->bge_rdata->bge_status_block.bge_status = 0;
3151 
3152 		if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
3153 		    statusword & BGE_STATFLAG_LINKSTATE_CHANGED ||
3154 		    BGE_STS_BIT(sc, BGE_STS_LINK_EVT))
3155 			bge_link_upd(sc);
3156 
3157 		if (ifp->if_flags & IFF_RUNNING) {
3158 			/* Check RX return ring producer/consumer */
3159 			bge_rxeof(sc);
3160 
3161 			/* Check TX ring producer/consumer */
3162 			bge_txeof(sc);
3163 		}
3164 
3165 		if (sc->bge_pending_rxintr_change) {
3166 			uint32_t rx_ticks = sc->bge_rx_coal_ticks;
3167 			uint32_t rx_bds = sc->bge_rx_max_coal_bds;
3168 			uint32_t junk;
3169 
3170 			CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, rx_ticks);
3171 			DELAY(10);
3172 			junk = CSR_READ_4(sc, BGE_HCC_RX_COAL_TICKS);
3173 
3174 			CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, rx_bds);
3175 			DELAY(10);
3176 			junk = CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS);
3177 
3178 			sc->bge_pending_rxintr_change = 0;
3179 		}
3180 		bge_handle_events(sc);
3181 
3182 		/* Re-enable interrupts. */
3183 		bge_writembx(sc, BGE_MBX_IRQ0_LO, 0);
3184 
3185 		if (ifp->if_flags & IFF_RUNNING && !IFQ_IS_EMPTY(&ifp->if_snd))
3186 			bge_start(ifp);
3187 
3188 		return (1);
3189 	} else
3190 		return (0);
3191 }
3192 
3193 static void
3194 bge_tick(void *xsc)
3195 {
3196 	struct bge_softc *sc = xsc;
3197 	struct mii_data *mii = &sc->bge_mii;
3198 	int s;
3199 
3200 	s = splnet();
3201 
3202 	bge_stats_update(sc);
3203 
3204 	if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
3205 		/*
3206 		 * Since in TBI mode auto-polling can't be used we should poll
3207 		 * link status manually. Here we register pending link event
3208 		 * and trigger interrupt.
3209 		 */
3210 		BGE_STS_SETBIT(sc, BGE_STS_LINK_EVT);
3211 		BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
3212 	} else {
3213 		/*
3214 		 * Do not touch PHY if we have link up. This could break
3215 		 * IPMI/ASF mode or produce extra input errors.
3216 		 * (extra input errors was reported for bcm5701 & bcm5704).
3217 		 */
3218 		if (!BGE_STS_BIT(sc, BGE_STS_LINK))
3219 			mii_tick(mii);
3220 	}
3221 
3222 	callout_reset(&sc->bge_timeout, hz, bge_tick, sc);
3223 
3224 	splx(s);
3225 }
3226 
3227 static void
3228 bge_stats_update(struct bge_softc *sc)
3229 {
3230 	struct ifnet *ifp = &sc->ethercom.ec_if;
3231 	bus_size_t stats = BGE_MEMWIN_START + BGE_STATS_BLOCK;
3232 	bus_size_t rstats = BGE_RX_STATS;
3233 
3234 #define READ_RSTAT(sc, stats, stat) \
3235 	  CSR_READ_4(sc, stats + offsetof(struct bge_mac_stats_regs, stat))
3236 
3237 	if (BGE_IS_5705_OR_BEYOND(sc)) {
3238 		ifp->if_collisions +=
3239 		    READ_RSTAT(sc, rstats, dot3StatsSingleCollisionFrames) +
3240 		    READ_RSTAT(sc, rstats, dot3StatsMultipleCollisionFrames) +
3241 		    READ_RSTAT(sc, rstats, dot3StatsExcessiveCollisions) +
3242 		    READ_RSTAT(sc, rstats, dot3StatsLateCollisions);
3243 
3244 		BGE_EVCNT_ADD(sc->bge_ev_tx_xoff,
3245 			      READ_RSTAT(sc, rstats, outXoffSent));
3246 		BGE_EVCNT_ADD(sc->bge_ev_tx_xon,
3247 			      READ_RSTAT(sc, rstats, outXonSent));
3248 		BGE_EVCNT_ADD(sc->bge_ev_rx_xoff,
3249 			      READ_RSTAT(sc, rstats, xoffPauseFramesReceived));
3250 		BGE_EVCNT_ADD(sc->bge_ev_rx_xon,
3251 			      READ_RSTAT(sc, rstats, xonPauseFramesReceived));
3252 		BGE_EVCNT_ADD(sc->bge_ev_rx_macctl,
3253 			      READ_RSTAT(sc, rstats, macControlFramesReceived));
3254 		BGE_EVCNT_ADD(sc->bge_ev_xoffentered,
3255 			      READ_RSTAT(sc, rstats, xoffStateEntered));
3256 		return;
3257 	}
3258 
3259 #undef READ_RSTAT
3260 #define READ_STAT(sc, stats, stat) \
3261 	  CSR_READ_4(sc, stats + offsetof(struct bge_stats, stat))
3262 
3263 	ifp->if_collisions +=
3264 	  (READ_STAT(sc, stats, dot3StatsSingleCollisionFrames.bge_addr_lo) +
3265 	   READ_STAT(sc, stats, dot3StatsMultipleCollisionFrames.bge_addr_lo) +
3266 	   READ_STAT(sc, stats, dot3StatsExcessiveCollisions.bge_addr_lo) +
3267 	   READ_STAT(sc, stats, dot3StatsLateCollisions.bge_addr_lo)) -
3268 	  ifp->if_collisions;
3269 
3270 	BGE_EVCNT_UPD(sc->bge_ev_tx_xoff,
3271 		      READ_STAT(sc, stats, outXoffSent.bge_addr_lo));
3272 	BGE_EVCNT_UPD(sc->bge_ev_tx_xon,
3273 		      READ_STAT(sc, stats, outXonSent.bge_addr_lo));
3274 	BGE_EVCNT_UPD(sc->bge_ev_rx_xoff,
3275 		      READ_STAT(sc, stats,
3276 		      		xoffPauseFramesReceived.bge_addr_lo));
3277 	BGE_EVCNT_UPD(sc->bge_ev_rx_xon,
3278 		      READ_STAT(sc, stats, xonPauseFramesReceived.bge_addr_lo));
3279 	BGE_EVCNT_UPD(sc->bge_ev_rx_macctl,
3280 		      READ_STAT(sc, stats,
3281 		      		macControlFramesReceived.bge_addr_lo));
3282 	BGE_EVCNT_UPD(sc->bge_ev_xoffentered,
3283 		      READ_STAT(sc, stats, xoffStateEntered.bge_addr_lo));
3284 
3285 #undef READ_STAT
3286 
3287 #ifdef notdef
3288 	ifp->if_collisions +=
3289 	   (sc->bge_rdata->bge_info.bge_stats.dot3StatsSingleCollisionFrames +
3290 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsMultipleCollisionFrames +
3291 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsExcessiveCollisions +
3292 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsLateCollisions) -
3293 	   ifp->if_collisions;
3294 #endif
3295 }
3296 
3297 /*
3298  * Pad outbound frame to ETHER_MIN_NOPAD for an unusual reason.
3299  * The bge hardware will pad out Tx runts to ETHER_MIN_NOPAD,
3300  * but when such padded frames employ the  bge IP/TCP checksum offload,
3301  * the hardware checksum assist gives incorrect results (possibly
3302  * from incorporating its own padding into the UDP/TCP checksum; who knows).
3303  * If we pad such runts with zeros, the onboard checksum comes out correct.
3304  */
3305 static inline int
3306 bge_cksum_pad(struct mbuf *pkt)
3307 {
3308 	struct mbuf *last = NULL;
3309 	int padlen;
3310 
3311 	padlen = ETHER_MIN_NOPAD - pkt->m_pkthdr.len;
3312 
3313 	/* if there's only the packet-header and we can pad there, use it. */
3314 	if (pkt->m_pkthdr.len == pkt->m_len &&
3315 	    M_TRAILINGSPACE(pkt) >= padlen) {
3316 		last = pkt;
3317 	} else {
3318 		/*
3319 		 * Walk packet chain to find last mbuf. We will either
3320 		 * pad there, or append a new mbuf and pad it
3321 		 * (thus perhaps avoiding the bcm5700 dma-min bug).
3322 		 */
3323 		for (last = pkt; last->m_next != NULL; last = last->m_next) {
3324 	      	       continue; /* do nothing */
3325 		}
3326 
3327 		/* `last' now points to last in chain. */
3328 		if (M_TRAILINGSPACE(last) < padlen) {
3329 			/* Allocate new empty mbuf, pad it. Compact later. */
3330 			struct mbuf *n;
3331 			MGET(n, M_DONTWAIT, MT_DATA);
3332 			if (n == NULL)
3333 				return ENOBUFS;
3334 			n->m_len = 0;
3335 			last->m_next = n;
3336 			last = n;
3337 		}
3338 	}
3339 
3340 	KDASSERT(!M_READONLY(last));
3341 	KDASSERT(M_TRAILINGSPACE(last) >= padlen);
3342 
3343 	/* Now zero the pad area, to avoid the bge cksum-assist bug */
3344 	memset(mtod(last, char *) + last->m_len, 0, padlen);
3345 	last->m_len += padlen;
3346 	pkt->m_pkthdr.len += padlen;
3347 	return 0;
3348 }
3349 
3350 /*
3351  * Compact outbound packets to avoid bug with DMA segments less than 8 bytes.
3352  */
3353 static inline int
3354 bge_compact_dma_runt(struct mbuf *pkt)
3355 {
3356 	struct mbuf	*m, *prev;
3357 	int 		totlen, prevlen;
3358 
3359 	prev = NULL;
3360 	totlen = 0;
3361 	prevlen = -1;
3362 
3363 	for (m = pkt; m != NULL; prev = m,m = m->m_next) {
3364 		int mlen = m->m_len;
3365 		int shortfall = 8 - mlen ;
3366 
3367 		totlen += mlen;
3368 		if (mlen == 0) {
3369 			continue;
3370 		}
3371 		if (mlen >= 8)
3372 			continue;
3373 
3374 		/* If we get here, mbuf data is too small for DMA engine.
3375 		 * Try to fix by shuffling data to prev or next in chain.
3376 		 * If that fails, do a compacting deep-copy of the whole chain.
3377 		 */
3378 
3379 		/* Internal frag. If fits in prev, copy it there. */
3380 		if (prev && M_TRAILINGSPACE(prev) >= m->m_len) {
3381 		  	memcpy(prev->m_data + prev->m_len, m->m_data, mlen);
3382 			prev->m_len += mlen;
3383 			m->m_len = 0;
3384 			/* XXX stitch chain */
3385 			prev->m_next = m_free(m);
3386 			m = prev;
3387 			continue;
3388 		}
3389 		else if (m->m_next != NULL &&
3390 			     M_TRAILINGSPACE(m) >= shortfall &&
3391 			     m->m_next->m_len >= (8 + shortfall)) {
3392 		    /* m is writable and have enough data in next, pull up. */
3393 
3394 		  	memcpy(m->m_data + m->m_len, m->m_next->m_data,
3395 			    shortfall);
3396 			m->m_len += shortfall;
3397 			m->m_next->m_len -= shortfall;
3398 			m->m_next->m_data += shortfall;
3399 		}
3400 		else if (m->m_next == NULL || 1) {
3401 		  	/* Got a runt at the very end of the packet.
3402 			 * borrow data from the tail of the preceding mbuf and
3403 			 * update its length in-place. (The original data is still
3404 			 * valid, so we can do this even if prev is not writable.)
3405 			 */
3406 
3407 			/* if we'd make prev a runt, just move all of its data. */
3408 			KASSERT(prev != NULL /*, ("runt but null PREV")*/);
3409 			KASSERT(prev->m_len >= 8 /*, ("runt prev")*/);
3410 
3411 			if ((prev->m_len - shortfall) < 8)
3412 				shortfall = prev->m_len;
3413 
3414 #ifdef notyet	/* just do the safe slow thing for now */
3415 			if (!M_READONLY(m)) {
3416 				if (M_LEADINGSPACE(m) < shorfall) {
3417 					void *m_dat;
3418 					m_dat = (m->m_flags & M_PKTHDR) ?
3419 					  m->m_pktdat : m->dat;
3420 					memmove(m_dat, mtod(m, void*), m->m_len);
3421 					m->m_data = m_dat;
3422 				    }
3423 			} else
3424 #endif	/* just do the safe slow thing */
3425 			{
3426 				struct mbuf * n = NULL;
3427 				int newprevlen = prev->m_len - shortfall;
3428 
3429 				MGET(n, M_NOWAIT, MT_DATA);
3430 				if (n == NULL)
3431 				   return ENOBUFS;
3432 				KASSERT(m->m_len + shortfall < MLEN
3433 					/*,
3434 					  ("runt %d +prev %d too big\n", m->m_len, shortfall)*/);
3435 
3436 				/* first copy the data we're stealing from prev */
3437 				memcpy(n->m_data, prev->m_data + newprevlen,
3438 				    shortfall);
3439 
3440 				/* update prev->m_len accordingly */
3441 				prev->m_len -= shortfall;
3442 
3443 				/* copy data from runt m */
3444 				memcpy(n->m_data + shortfall, m->m_data,
3445 				    m->m_len);
3446 
3447 				/* n holds what we stole from prev, plus m */
3448 				n->m_len = shortfall + m->m_len;
3449 
3450 				/* stitch n into chain and free m */
3451 				n->m_next = m->m_next;
3452 				prev->m_next = n;
3453 				/* KASSERT(m->m_next == NULL); */
3454 				m->m_next = NULL;
3455 				m_free(m);
3456 				m = n;	/* for continuing loop */
3457 			}
3458 		}
3459 		prevlen = m->m_len;
3460 	}
3461 	return 0;
3462 }
3463 
3464 /*
3465  * Encapsulate an mbuf chain in the tx ring  by coupling the mbuf data
3466  * pointers to descriptors.
3467  */
3468 static int
3469 bge_encap(struct bge_softc *sc, struct mbuf *m_head, u_int32_t *txidx)
3470 {
3471 	struct bge_tx_bd	*f = NULL;
3472 	u_int32_t		frag, cur;
3473 	u_int16_t		csum_flags = 0;
3474 	u_int16_t		txbd_tso_flags = 0;
3475 	struct txdmamap_pool_entry *dma;
3476 	bus_dmamap_t dmamap;
3477 	int			i = 0;
3478 	struct m_tag		*mtag;
3479 	int			use_tso, maxsegsize, error;
3480 
3481 	cur = frag = *txidx;
3482 
3483 	if (m_head->m_pkthdr.csum_flags) {
3484 		if (m_head->m_pkthdr.csum_flags & M_CSUM_IPv4)
3485 			csum_flags |= BGE_TXBDFLAG_IP_CSUM;
3486 		if (m_head->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4))
3487 			csum_flags |= BGE_TXBDFLAG_TCP_UDP_CSUM;
3488 	}
3489 
3490 	/*
3491 	 * If we were asked to do an outboard checksum, and the NIC
3492 	 * has the bug where it sometimes adds in the Ethernet padding,
3493 	 * explicitly pad with zeros so the cksum will be correct either way.
3494 	 * (For now, do this for all chip versions, until newer
3495 	 * are confirmed to not require the workaround.)
3496 	 */
3497 	if ((csum_flags & BGE_TXBDFLAG_TCP_UDP_CSUM) == 0 ||
3498 #ifdef notyet
3499 	    (sc->bge_quirks & BGE_QUIRK_SHORT_CKSUM_BUG) == 0 ||
3500 #endif
3501 	    m_head->m_pkthdr.len >= ETHER_MIN_NOPAD)
3502 		goto check_dma_bug;
3503 
3504 	if (bge_cksum_pad(m_head) != 0) {
3505 	    return ENOBUFS;
3506 	}
3507 
3508 check_dma_bug:
3509 	if (!(BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX))
3510 		goto doit;
3511 
3512 	/*
3513 	 * bcm5700 Revision B silicon cannot handle DMA descriptors with
3514 	 * less than eight bytes.  If we encounter a teeny mbuf
3515 	 * at the end of a chain, we can pad.  Otherwise, copy.
3516 	 */
3517 	if (bge_compact_dma_runt(m_head) != 0)
3518 		return ENOBUFS;
3519 
3520 doit:
3521 	dma = SLIST_FIRST(&sc->txdma_list);
3522 	if (dma == NULL)
3523 		return ENOBUFS;
3524 	dmamap = dma->dmamap;
3525 
3526 	/*
3527 	 * Set up any necessary TSO state before we start packing...
3528 	 */
3529 	use_tso = (m_head->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0;
3530 	if (!use_tso) {
3531 		maxsegsize = 0;
3532 	} else {	/* TSO setup */
3533 		unsigned  mss;
3534 		struct ether_header *eh;
3535 		unsigned ip_tcp_hlen, iptcp_opt_words, tcp_seg_flags, offset;
3536 		struct mbuf * m0 = m_head;
3537 		struct ip *ip;
3538 		struct tcphdr *th;
3539 		int iphl, hlen;
3540 
3541 		/*
3542 		 * XXX It would be nice if the mbuf pkthdr had offset
3543 		 * fields for the protocol headers.
3544 		 */
3545 
3546 		eh = mtod(m0, struct ether_header *);
3547 		switch (htons(eh->ether_type)) {
3548 		case ETHERTYPE_IP:
3549 			offset = ETHER_HDR_LEN;
3550 			break;
3551 
3552 		case ETHERTYPE_VLAN:
3553 			offset = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
3554 			break;
3555 
3556 		default:
3557 			/*
3558 			 * Don't support this protocol or encapsulation.
3559 			 */
3560 			return (ENOBUFS);
3561 		}
3562 
3563 		/*
3564 		 * TCP/IP headers are in the first mbuf; we can do
3565 		 * this the easy way.
3566 		 */
3567 		iphl = M_CSUM_DATA_IPv4_IPHL(m0->m_pkthdr.csum_data);
3568 		hlen = iphl + offset;
3569 		if (__predict_false(m0->m_len <
3570 				    (hlen + sizeof(struct tcphdr)))) {
3571 
3572 			aprint_debug_dev(sc->bge_dev,
3573 			    "TSO: hard case m0->m_len == %d < ip/tcp hlen %zd,"
3574 			    "not handled yet\n",
3575 			     m0->m_len, hlen+ sizeof(struct tcphdr));
3576 #ifdef NOTYET
3577 			/*
3578 			 * XXX jonathan@NetBSD.org: untested.
3579 			 * how to force  this branch to be taken?
3580 			 */
3581 			BGE_EVCNT_INCR(&sc->sc_ev_txtsopain);
3582 
3583 			m_copydata(m0, offset, sizeof(ip), &ip);
3584 			m_copydata(m0, hlen, sizeof(th), &th);
3585 
3586 			ip.ip_len = 0;
3587 
3588 			m_copyback(m0, hlen + offsetof(struct ip, ip_len),
3589 			    sizeof(ip.ip_len), &ip.ip_len);
3590 
3591 			th.th_sum = in_cksum_phdr(ip.ip_src.s_addr,
3592 			    ip.ip_dst.s_addr, htons(IPPROTO_TCP));
3593 
3594 			m_copyback(m0, hlen + offsetof(struct tcphdr, th_sum),
3595 			    sizeof(th.th_sum), &th.th_sum);
3596 
3597 			hlen += th.th_off << 2;
3598 			iptcp_opt_words	= hlen;
3599 #else
3600 			/*
3601 			 * if_wm "hard" case not yet supported, can we not
3602 			 * mandate it out of existence?
3603 			 */
3604 			(void) ip; (void)th; (void) ip_tcp_hlen;
3605 
3606 			return ENOBUFS;
3607 #endif
3608 		} else {
3609 			ip = (struct ip *) (mtod(m0, char *) + offset);
3610 			th = (struct tcphdr *) (mtod(m0, char *) + hlen);
3611 			ip_tcp_hlen = iphl +  (th->th_off << 2);
3612 
3613 			/* Total IP/TCP options, in 32-bit words */
3614 			iptcp_opt_words = (ip_tcp_hlen
3615 					   - sizeof(struct tcphdr)
3616 					   - sizeof(struct ip)) >> 2;
3617 		}
3618 		if (BGE_IS_5750_OR_BEYOND(sc)) {
3619 			th->th_sum = 0;
3620 			csum_flags &= ~(BGE_TXBDFLAG_TCP_UDP_CSUM);
3621 		} else {
3622 			/*
3623 			 * XXX jonathan@NetBSD.org: 5705 untested.
3624 			 * Requires TSO firmware patch for 5701/5703/5704.
3625 			 */
3626 			th->th_sum = in_cksum_phdr(ip->ip_src.s_addr,
3627 			    ip->ip_dst.s_addr, htons(IPPROTO_TCP));
3628 		}
3629 
3630 		mss = m_head->m_pkthdr.segsz;
3631 		txbd_tso_flags |=
3632 		    BGE_TXBDFLAG_CPU_PRE_DMA |
3633 		    BGE_TXBDFLAG_CPU_POST_DMA;
3634 
3635 		/*
3636 		 * Our NIC TSO-assist assumes TSO has standard, optionless
3637 		 * IPv4 and TCP headers, which total 40 bytes. By default,
3638 		 * the NIC copies 40 bytes of IP/TCP header from the
3639 		 * supplied header into the IP/TCP header portion of
3640 		 * each post-TSO-segment. If the supplied packet has IP or
3641 		 * TCP options, we need to tell the NIC to copy those extra
3642 		 * bytes into each  post-TSO header, in addition to the normal
3643 		 * 40-byte IP/TCP header (and to leave space accordingly).
3644 		 * Unfortunately, the driver encoding of option length
3645 		 * varies across different ASIC families.
3646 		 */
3647 		tcp_seg_flags = 0;
3648 		if (iptcp_opt_words) {
3649 			if ( BGE_IS_5705_OR_BEYOND(sc)) {
3650 				tcp_seg_flags =
3651 					iptcp_opt_words << 11;
3652 			} else {
3653 				txbd_tso_flags |=
3654 					iptcp_opt_words << 12;
3655 			}
3656 		}
3657 		maxsegsize = mss | tcp_seg_flags;
3658 		ip->ip_len = htons(mss + ip_tcp_hlen);
3659 
3660 	}	/* TSO setup */
3661 
3662 	/*
3663 	 * Start packing the mbufs in this chain into
3664 	 * the fragment pointers. Stop when we run out
3665 	 * of fragments or hit the end of the mbuf chain.
3666 	 */
3667 	error = bus_dmamap_load_mbuf(sc->bge_dmatag, dmamap, m_head,
3668 	    BUS_DMA_NOWAIT);
3669 	if (error) {
3670 		return (ENOBUFS);
3671 	}
3672 	/*
3673 	 * Sanity check: avoid coming within 16 descriptors
3674 	 * of the end of the ring.
3675 	 */
3676 	if (dmamap->dm_nsegs > (BGE_TX_RING_CNT - sc->bge_txcnt - 16)) {
3677 		BGE_TSO_PRINTF(("%s: "
3678 		    " dmamap_load_mbuf too close to ring wrap\n",
3679 		    device_xname(sc->bge_dev)));
3680 		goto fail_unload;
3681 	}
3682 
3683 	mtag = sc->ethercom.ec_nvlans ?
3684 	    m_tag_find(m_head, PACKET_TAG_VLAN, NULL) : NULL;
3685 
3686 
3687 	/* Iterate over dmap-map fragments. */
3688 	for (i = 0; i < dmamap->dm_nsegs; i++) {
3689 		f = &sc->bge_rdata->bge_tx_ring[frag];
3690 		if (sc->bge_cdata.bge_tx_chain[frag] != NULL)
3691 			break;
3692 
3693 		bge_set_hostaddr(&f->bge_addr, dmamap->dm_segs[i].ds_addr);
3694 		f->bge_len = dmamap->dm_segs[i].ds_len;
3695 
3696 		/*
3697 		 * For 5751 and follow-ons, for TSO we must turn
3698 		 * off checksum-assist flag in the tx-descr, and
3699 		 * supply the ASIC-revision-specific encoding
3700 		 * of TSO flags and segsize.
3701 		 */
3702 		if (use_tso) {
3703 			if (BGE_IS_5750_OR_BEYOND(sc) || i == 0) {
3704 				f->bge_rsvd = maxsegsize;
3705 				f->bge_flags = csum_flags | txbd_tso_flags;
3706 			} else {
3707 				f->bge_rsvd = 0;
3708 				f->bge_flags =
3709 				  (csum_flags | txbd_tso_flags) & 0x0fff;
3710 			}
3711 		} else {
3712 			f->bge_rsvd = 0;
3713 			f->bge_flags = csum_flags;
3714 		}
3715 
3716 		if (mtag != NULL) {
3717 			f->bge_flags |= BGE_TXBDFLAG_VLAN_TAG;
3718 			f->bge_vlan_tag = VLAN_TAG_VALUE(mtag);
3719 		} else {
3720 			f->bge_vlan_tag = 0;
3721 		}
3722 		cur = frag;
3723 		BGE_INC(frag, BGE_TX_RING_CNT);
3724 	}
3725 
3726 	if (i < dmamap->dm_nsegs) {
3727 		BGE_TSO_PRINTF(("%s: reached %d < dm_nsegs %d\n",
3728 		    device_xname(sc->bge_dev), i, dmamap->dm_nsegs));
3729 		goto fail_unload;
3730 	}
3731 
3732 	bus_dmamap_sync(sc->bge_dmatag, dmamap, 0, dmamap->dm_mapsize,
3733 	    BUS_DMASYNC_PREWRITE);
3734 
3735 	if (frag == sc->bge_tx_saved_considx) {
3736 		BGE_TSO_PRINTF(("%s: frag %d = wrapped id %d?\n",
3737 		    device_xname(sc->bge_dev), frag, sc->bge_tx_saved_considx));
3738 
3739 		goto fail_unload;
3740 	}
3741 
3742 	sc->bge_rdata->bge_tx_ring[cur].bge_flags |= BGE_TXBDFLAG_END;
3743 	sc->bge_cdata.bge_tx_chain[cur] = m_head;
3744 	SLIST_REMOVE_HEAD(&sc->txdma_list, link);
3745 	sc->txdma[cur] = dma;
3746 	sc->bge_txcnt += dmamap->dm_nsegs;
3747 
3748 	*txidx = frag;
3749 
3750 	return (0);
3751 
3752 fail_unload:
3753 	bus_dmamap_unload(sc->bge_dmatag, dmamap);
3754 
3755 	return ENOBUFS;
3756 }
3757 
3758 /*
3759  * Main transmit routine. To avoid having to do mbuf copies, we put pointers
3760  * to the mbuf data regions directly in the transmit descriptors.
3761  */
3762 static void
3763 bge_start(struct ifnet *ifp)
3764 {
3765 	struct bge_softc *sc;
3766 	struct mbuf *m_head = NULL;
3767 	u_int32_t prodidx;
3768 	int pkts = 0;
3769 
3770 	sc = ifp->if_softc;
3771 
3772 	if ((ifp->if_flags & (IFF_RUNNING|IFF_OACTIVE)) != IFF_RUNNING)
3773 		return;
3774 
3775 	prodidx = sc->bge_tx_prodidx;
3776 
3777 	while(sc->bge_cdata.bge_tx_chain[prodidx] == NULL) {
3778 		IFQ_POLL(&ifp->if_snd, m_head);
3779 		if (m_head == NULL)
3780 			break;
3781 
3782 #if 0
3783 		/*
3784 		 * XXX
3785 		 * safety overkill.  If this is a fragmented packet chain
3786 		 * with delayed TCP/UDP checksums, then only encapsulate
3787 		 * it if we have enough descriptors to handle the entire
3788 		 * chain at once.
3789 		 * (paranoia -- may not actually be needed)
3790 		 */
3791 		if (m_head->m_flags & M_FIRSTFRAG &&
3792 		    m_head->m_pkthdr.csum_flags & (CSUM_DELAY_DATA)) {
3793 			if ((BGE_TX_RING_CNT - sc->bge_txcnt) <
3794 			    M_CSUM_DATA_IPv4_OFFSET(m_head->m_pkthdr.csum_data) + 16) {
3795 				ifp->if_flags |= IFF_OACTIVE;
3796 				break;
3797 			}
3798 		}
3799 #endif
3800 
3801 		/*
3802 		 * Pack the data into the transmit ring. If we
3803 		 * don't have room, set the OACTIVE flag and wait
3804 		 * for the NIC to drain the ring.
3805 		 */
3806 		if (bge_encap(sc, m_head, &prodidx)) {
3807 			ifp->if_flags |= IFF_OACTIVE;
3808 			break;
3809 		}
3810 
3811 		/* now we are committed to transmit the packet */
3812 		IFQ_DEQUEUE(&ifp->if_snd, m_head);
3813 		pkts++;
3814 
3815 #if NBPFILTER > 0
3816 		/*
3817 		 * If there's a BPF listener, bounce a copy of this frame
3818 		 * to him.
3819 		 */
3820 		if (ifp->if_bpf)
3821 			bpf_mtap(ifp->if_bpf, m_head);
3822 #endif
3823 	}
3824 	if (pkts == 0)
3825 		return;
3826 
3827 	/* Transmit */
3828 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
3829 	/* 5700 b2 errata */
3830 	if (BGE_CHIPREV(sc->bge_chipid) == BGE_CHIPREV_5700_BX)
3831 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
3832 
3833 	sc->bge_tx_prodidx = prodidx;
3834 
3835 	/*
3836 	 * Set a timeout in case the chip goes out to lunch.
3837 	 */
3838 	ifp->if_timer = 5;
3839 }
3840 
3841 static int
3842 bge_init(struct ifnet *ifp)
3843 {
3844 	struct bge_softc *sc = ifp->if_softc;
3845 	const u_int16_t *m;
3846 	int s, error = 0;
3847 
3848 	s = splnet();
3849 
3850 	ifp = &sc->ethercom.ec_if;
3851 
3852 	/* Cancel pending I/O and flush buffers. */
3853 	bge_stop(ifp, 0);
3854 	bge_reset(sc);
3855 	bge_chipinit(sc);
3856 
3857 	/*
3858 	 * Init the various state machines, ring
3859 	 * control blocks and firmware.
3860 	 */
3861 	error = bge_blockinit(sc);
3862 	if (error != 0) {
3863 		aprint_error_dev(sc->bge_dev, "initialization error %d\n",
3864 		    error);
3865 		splx(s);
3866 		return error;
3867 	}
3868 
3869 	ifp = &sc->ethercom.ec_if;
3870 
3871 	/* Specify MTU. */
3872 	CSR_WRITE_4(sc, BGE_RX_MTU, ifp->if_mtu +
3873 	    ETHER_HDR_LEN + ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN);
3874 
3875 	/* Load our MAC address. */
3876 	m = (const u_int16_t *)&(CLLADDR(ifp->if_sadl)[0]);
3877 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_LO, htons(m[0]));
3878 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_HI, (htons(m[1]) << 16) | htons(m[2]));
3879 
3880 	/* Enable or disable promiscuous mode as needed. */
3881 	if (ifp->if_flags & IFF_PROMISC) {
3882 		BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
3883 	} else {
3884 		BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
3885 	}
3886 
3887 	/* Program multicast filter. */
3888 	bge_setmulti(sc);
3889 
3890 	/* Init RX ring. */
3891 	bge_init_rx_ring_std(sc);
3892 
3893 	/*
3894 	 * Workaround for a bug in 5705 ASIC rev A0. Poll the NIC's
3895 	 * memory to insure that the chip has in fact read the first
3896 	 * entry of the ring.
3897 	 */
3898 	if (sc->bge_chipid == BGE_CHIPID_BCM5705_A0) {
3899 		u_int32_t		v, i;
3900 		for (i = 0; i < 10; i++) {
3901 			DELAY(20);
3902 			v = bge_readmem_ind(sc, BGE_STD_RX_RINGS + 8);
3903 			if (v == (MCLBYTES - ETHER_ALIGN))
3904 				break;
3905 		}
3906 		if (i == 10)
3907 			aprint_error_dev(sc->bge_dev,
3908 			    "5705 A0 chip failed to load RX ring\n");
3909 	}
3910 
3911 	/* Init jumbo RX ring. */
3912 	if (ifp->if_mtu > (ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN))
3913 		bge_init_rx_ring_jumbo(sc);
3914 
3915 	/* Init our RX return ring index */
3916 	sc->bge_rx_saved_considx = 0;
3917 
3918 	/* Init TX ring. */
3919 	bge_init_tx_ring(sc);
3920 
3921 	/* Turn on transmitter */
3922 	BGE_SETBIT(sc, BGE_TX_MODE, BGE_TXMODE_ENABLE);
3923 
3924 	/* Turn on receiver */
3925 	BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
3926 
3927 	CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 2);
3928 
3929 	/* Tell firmware we're alive. */
3930 	BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3931 
3932 	/* Enable host interrupts. */
3933 	BGE_SETBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_CLEAR_INTA);
3934 	BGE_CLRBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR);
3935 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 0);
3936 
3937 	if ((error = bge_ifmedia_upd(ifp)) != 0)
3938 		goto out;
3939 
3940 	ifp->if_flags |= IFF_RUNNING;
3941 	ifp->if_flags &= ~IFF_OACTIVE;
3942 
3943 	callout_reset(&sc->bge_timeout, hz, bge_tick, sc);
3944 
3945 out:
3946 	splx(s);
3947 
3948 	return error;
3949 }
3950 
3951 /*
3952  * Set media options.
3953  */
3954 static int
3955 bge_ifmedia_upd(struct ifnet *ifp)
3956 {
3957 	struct bge_softc *sc = ifp->if_softc;
3958 	struct mii_data *mii = &sc->bge_mii;
3959 	struct ifmedia *ifm = &sc->bge_ifmedia;
3960 	int rc;
3961 
3962 	/* If this is a 1000baseX NIC, enable the TBI port. */
3963 	if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
3964 		if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
3965 			return (EINVAL);
3966 		switch(IFM_SUBTYPE(ifm->ifm_media)) {
3967 		case IFM_AUTO:
3968 			/*
3969 			 * The BCM5704 ASIC appears to have a special
3970 			 * mechanism for programming the autoneg
3971 			 * advertisement registers in TBI mode.
3972 			 */
3973 			if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704) {
3974 				u_int32_t sgdig;
3975 				sgdig = CSR_READ_4(sc, BGE_SGDIG_STS);
3976 				if (sgdig & BGE_SGDIGSTS_DONE) {
3977 					CSR_WRITE_4(sc, BGE_TX_TBI_AUTONEG, 0);
3978 					sgdig = CSR_READ_4(sc, BGE_SGDIG_CFG);
3979 					sgdig |= BGE_SGDIGCFG_AUTO |
3980 					    BGE_SGDIGCFG_PAUSE_CAP |
3981 					    BGE_SGDIGCFG_ASYM_PAUSE;
3982 					CSR_WRITE_4(sc, BGE_SGDIG_CFG,
3983 					    sgdig | BGE_SGDIGCFG_SEND);
3984 					DELAY(5);
3985 					CSR_WRITE_4(sc, BGE_SGDIG_CFG, sgdig);
3986 				}
3987 			}
3988 			break;
3989 		case IFM_1000_SX:
3990 			if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX) {
3991 				BGE_CLRBIT(sc, BGE_MAC_MODE,
3992 				    BGE_MACMODE_HALF_DUPLEX);
3993 			} else {
3994 				BGE_SETBIT(sc, BGE_MAC_MODE,
3995 				    BGE_MACMODE_HALF_DUPLEX);
3996 			}
3997 			break;
3998 		default:
3999 			return (EINVAL);
4000 		}
4001 		/* XXX 802.3x flow control for 1000BASE-SX */
4002 		return (0);
4003 	}
4004 
4005 	BGE_STS_SETBIT(sc, BGE_STS_LINK_EVT);
4006 	if ((rc = mii_mediachg(mii)) == ENXIO)
4007 		return 0;
4008 
4009 	/*
4010 	 * Force an interrupt so that we will call bge_link_upd
4011 	 * if needed and clear any pending link state attention.
4012 	 * Without this we are not getting any further interrupts
4013 	 * for link state changes and thus will not UP the link and
4014 	 * not be able to send in bge_start. The only way to get
4015 	 * things working was to receive a packet and get a RX intr.
4016 	 */
4017 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700 ||
4018 	    sc->bge_flags & BGE_IS_5788)
4019 		BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
4020 	else
4021 		BGE_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
4022 
4023 	return rc;
4024 }
4025 
4026 /*
4027  * Report current media status.
4028  */
4029 static void
4030 bge_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
4031 {
4032 	struct bge_softc *sc = ifp->if_softc;
4033 	struct mii_data *mii = &sc->bge_mii;
4034 
4035 	if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
4036 		ifmr->ifm_status = IFM_AVALID;
4037 		ifmr->ifm_active = IFM_ETHER;
4038 		if (CSR_READ_4(sc, BGE_MAC_STS) &
4039 		    BGE_MACSTAT_TBI_PCS_SYNCHED)
4040 			ifmr->ifm_status |= IFM_ACTIVE;
4041 		ifmr->ifm_active |= IFM_1000_SX;
4042 		if (CSR_READ_4(sc, BGE_MAC_MODE) & BGE_MACMODE_HALF_DUPLEX)
4043 			ifmr->ifm_active |= IFM_HDX;
4044 		else
4045 			ifmr->ifm_active |= IFM_FDX;
4046 		return;
4047 	}
4048 
4049 	mii_pollstat(mii);
4050 	ifmr->ifm_status = mii->mii_media_status;
4051 	ifmr->ifm_active = (mii->mii_media_active & ~IFM_ETH_FMASK) |
4052 	    sc->bge_flowflags;
4053 }
4054 
4055 static int
4056 bge_ioctl(struct ifnet *ifp, u_long command, void *data)
4057 {
4058 	struct bge_softc *sc = ifp->if_softc;
4059 	struct ifreq *ifr = (struct ifreq *) data;
4060 	int s, error = 0;
4061 	struct mii_data *mii;
4062 
4063 	s = splnet();
4064 
4065 	switch(command) {
4066 	case SIOCSIFFLAGS:
4067 		if ((error = ifioctl_common(ifp, command, data)) != 0)
4068 			break;
4069 		if (ifp->if_flags & IFF_UP) {
4070 			/*
4071 			 * If only the state of the PROMISC flag changed,
4072 			 * then just use the 'set promisc mode' command
4073 			 * instead of reinitializing the entire NIC. Doing
4074 			 * a full re-init means reloading the firmware and
4075 			 * waiting for it to start up, which may take a
4076 			 * second or two.
4077 			 */
4078 			if (ifp->if_flags & IFF_RUNNING &&
4079 			    ifp->if_flags & IFF_PROMISC &&
4080 			    !(sc->bge_if_flags & IFF_PROMISC)) {
4081 				BGE_SETBIT(sc, BGE_RX_MODE,
4082 				    BGE_RXMODE_RX_PROMISC);
4083 			} else if (ifp->if_flags & IFF_RUNNING &&
4084 			    !(ifp->if_flags & IFF_PROMISC) &&
4085 			    sc->bge_if_flags & IFF_PROMISC) {
4086 				BGE_CLRBIT(sc, BGE_RX_MODE,
4087 				    BGE_RXMODE_RX_PROMISC);
4088 			} else if (!(sc->bge_if_flags & IFF_UP))
4089 				bge_init(ifp);
4090 		} else {
4091 			if (ifp->if_flags & IFF_RUNNING)
4092 				bge_stop(ifp, 1);
4093 		}
4094 		sc->bge_if_flags = ifp->if_flags;
4095 		error = 0;
4096 		break;
4097 	case SIOCSIFMEDIA:
4098 		/* XXX Flow control is not supported for 1000BASE-SX */
4099 		if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
4100 			ifr->ifr_media &= ~IFM_ETH_FMASK;
4101 			sc->bge_flowflags = 0;
4102 		}
4103 
4104 		/* Flow control requires full-duplex mode. */
4105 		if (IFM_SUBTYPE(ifr->ifr_media) == IFM_AUTO ||
4106 		    (ifr->ifr_media & IFM_FDX) == 0) {
4107 		    	ifr->ifr_media &= ~IFM_ETH_FMASK;
4108 		}
4109 		if (IFM_SUBTYPE(ifr->ifr_media) != IFM_AUTO) {
4110 			if ((ifr->ifr_media & IFM_ETH_FMASK) == IFM_FLOW) {
4111 				/* We can do both TXPAUSE and RXPAUSE. */
4112 				ifr->ifr_media |=
4113 				    IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE;
4114 			}
4115 			sc->bge_flowflags = ifr->ifr_media & IFM_ETH_FMASK;
4116 		}
4117 		/* FALLTHROUGH */
4118 	case SIOCGIFMEDIA:
4119 		if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
4120 			error = ifmedia_ioctl(ifp, ifr, &sc->bge_ifmedia,
4121 			    command);
4122 		} else {
4123 			mii = &sc->bge_mii;
4124 			error = ifmedia_ioctl(ifp, ifr, &mii->mii_media,
4125 			    command);
4126 		}
4127 		break;
4128 	default:
4129 		if ((error = ether_ioctl(ifp, command, data)) != ENETRESET)
4130 			break;
4131 
4132 		error = 0;
4133 
4134 		if (command != SIOCADDMULTI && command != SIOCDELMULTI)
4135 			;
4136 		else if (ifp->if_flags & IFF_RUNNING)
4137 			bge_setmulti(sc);
4138 		break;
4139 	}
4140 
4141 	splx(s);
4142 
4143 	return (error);
4144 }
4145 
4146 static void
4147 bge_watchdog(struct ifnet *ifp)
4148 {
4149 	struct bge_softc *sc;
4150 
4151 	sc = ifp->if_softc;
4152 
4153 	aprint_error_dev(sc->bge_dev, "watchdog timeout -- resetting\n");
4154 
4155 	ifp->if_flags &= ~IFF_RUNNING;
4156 	bge_init(ifp);
4157 
4158 	ifp->if_oerrors++;
4159 }
4160 
4161 static void
4162 bge_stop_block(struct bge_softc *sc, bus_addr_t reg, uint32_t bit)
4163 {
4164 	int i;
4165 
4166 	BGE_CLRBIT(sc, reg, bit);
4167 
4168 	for (i = 0; i < BGE_TIMEOUT; i++) {
4169 		if ((CSR_READ_4(sc, reg) & bit) == 0)
4170 			return;
4171 		delay(100);
4172 		if (sc->bge_flags & BGE_PCIE)
4173 		  DELAY(1000);
4174 	}
4175 
4176 	/*
4177 	 * Doesn't print only when the register is BGE_SRS_MODE. It occurs
4178 	 * on some environment (and once after boot?)
4179 	 */
4180 	if (reg != BGE_SRS_MODE)
4181 		aprint_error_dev(sc->bge_dev,
4182 		    "block failed to stop: reg 0x%lx, bit 0x%08x\n",
4183 		    (u_long)reg, bit);
4184 }
4185 
4186 /*
4187  * Stop the adapter and free any mbufs allocated to the
4188  * RX and TX lists.
4189  */
4190 static void
4191 bge_stop(struct ifnet *ifp, int disable)
4192 {
4193 	struct bge_softc *sc = ifp->if_softc;
4194 
4195 	callout_stop(&sc->bge_timeout);
4196 
4197 	/*
4198 	 * Disable all of the receiver blocks
4199 	 */
4200 	bge_stop_block(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
4201 	bge_stop_block(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
4202 	bge_stop_block(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
4203 	if (!(BGE_IS_5705_OR_BEYOND(sc)))
4204 		bge_stop_block(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
4205 	bge_stop_block(sc, BGE_RDBDI_MODE, BGE_RBDIMODE_ENABLE);
4206 	bge_stop_block(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
4207 	bge_stop_block(sc, BGE_RBDC_MODE, BGE_RBDCMODE_ENABLE);
4208 
4209 	/*
4210 	 * Disable all of the transmit blocks
4211 	 */
4212 	bge_stop_block(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
4213 	bge_stop_block(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
4214 	bge_stop_block(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
4215 	bge_stop_block(sc, BGE_RDMA_MODE, BGE_RDMAMODE_ENABLE);
4216 	bge_stop_block(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE);
4217 	if (!(BGE_IS_5705_OR_BEYOND(sc)))
4218 		bge_stop_block(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
4219 	bge_stop_block(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
4220 
4221 	/*
4222 	 * Shut down all of the memory managers and related
4223 	 * state machines.
4224 	 */
4225 	bge_stop_block(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE);
4226 	bge_stop_block(sc, BGE_WDMA_MODE, BGE_WDMAMODE_ENABLE);
4227 	if (!(BGE_IS_5705_OR_BEYOND(sc)))
4228 		bge_stop_block(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
4229 
4230 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
4231 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
4232 
4233 	if (!(BGE_IS_5705_OR_BEYOND(sc))) {
4234 		bge_stop_block(sc, BGE_BMAN_MODE, BGE_BMANMODE_ENABLE);
4235 		bge_stop_block(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
4236 	}
4237 
4238 	/* Disable host interrupts. */
4239 	BGE_SETBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR);
4240 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
4241 
4242 	/*
4243 	 * Tell firmware we're shutting down.
4244 	 */
4245 	BGE_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
4246 
4247 	/* Free the RX lists. */
4248 	bge_free_rx_ring_std(sc);
4249 
4250 	/* Free jumbo RX list. */
4251 	bge_free_rx_ring_jumbo(sc);
4252 
4253 	/* Free TX buffers. */
4254 	bge_free_tx_ring(sc);
4255 
4256 	/*
4257 	 * Isolate/power down the PHY.
4258 	 */
4259 	if (!(sc->bge_flags & BGE_PHY_FIBER_TBI))
4260 		mii_down(&sc->bge_mii);
4261 
4262 	sc->bge_tx_saved_considx = BGE_TXCONS_UNSET;
4263 
4264 	/* Clear MAC's link state (PHY may still have link UP). */
4265 	BGE_STS_CLRBIT(sc, BGE_STS_LINK);
4266 
4267 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
4268 }
4269 
4270 static void
4271 bge_link_upd(struct bge_softc *sc)
4272 {
4273 	struct ifnet *ifp = &sc->ethercom.ec_if;
4274 	struct mii_data *mii = &sc->bge_mii;
4275 	u_int32_t status;
4276 	int link;
4277 
4278 	/* Clear 'pending link event' flag */
4279 	BGE_STS_CLRBIT(sc, BGE_STS_LINK_EVT);
4280 
4281 	/*
4282 	 * Process link state changes.
4283 	 * Grrr. The link status word in the status block does
4284 	 * not work correctly on the BCM5700 rev AX and BX chips,
4285 	 * according to all available information. Hence, we have
4286 	 * to enable MII interrupts in order to properly obtain
4287 	 * async link changes. Unfortunately, this also means that
4288 	 * we have to read the MAC status register to detect link
4289 	 * changes, thereby adding an additional register access to
4290 	 * the interrupt handler.
4291 	 */
4292 
4293 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5700) {
4294 		status = CSR_READ_4(sc, BGE_MAC_STS);
4295 		if (status & BGE_MACSTAT_MI_INTERRUPT) {
4296 			mii_pollstat(mii);
4297 
4298 			if (!BGE_STS_BIT(sc, BGE_STS_LINK) &&
4299 			    mii->mii_media_status & IFM_ACTIVE &&
4300 			    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE)
4301 				BGE_STS_SETBIT(sc, BGE_STS_LINK);
4302 			else if (BGE_STS_BIT(sc, BGE_STS_LINK) &&
4303 			    (!(mii->mii_media_status & IFM_ACTIVE) ||
4304 			    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE))
4305 				BGE_STS_CLRBIT(sc, BGE_STS_LINK);
4306 
4307 			/* Clear the interrupt */
4308 			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
4309 			    BGE_EVTENB_MI_INTERRUPT);
4310 			bge_miibus_readreg(sc->bge_dev, 1, BRGPHY_MII_ISR);
4311 			bge_miibus_writereg(sc->bge_dev, 1, BRGPHY_MII_IMR,
4312 			    BRGPHY_INTRS);
4313 		}
4314 		return;
4315 	}
4316 
4317 	if (sc->bge_flags & BGE_PHY_FIBER_TBI) {
4318 		status = CSR_READ_4(sc, BGE_MAC_STS);
4319 		if (status & BGE_MACSTAT_TBI_PCS_SYNCHED) {
4320 			if (!BGE_STS_BIT(sc, BGE_STS_LINK)) {
4321 				BGE_STS_SETBIT(sc, BGE_STS_LINK);
4322 				if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5704)
4323 					BGE_CLRBIT(sc, BGE_MAC_MODE,
4324 					    BGE_MACMODE_TBI_SEND_CFGS);
4325 				CSR_WRITE_4(sc, BGE_MAC_STS, 0xFFFFFFFF);
4326 				if_link_state_change(ifp, LINK_STATE_UP);
4327 			}
4328 		} else if (BGE_STS_BIT(sc, BGE_STS_LINK)) {
4329 			BGE_STS_CLRBIT(sc, BGE_STS_LINK);
4330 			if_link_state_change(ifp, LINK_STATE_DOWN);
4331 		}
4332         /*
4333 	 * Discard link events for MII/GMII cards if MI auto-polling disabled.
4334 	 * This should not happen since mii callouts are locked now, but
4335 	 * we keep this check for debug.
4336 	 */
4337 	} else if (BGE_STS_BIT(sc, BGE_STS_AUTOPOLL)) {
4338 		/*
4339 		 * Some broken BCM chips have BGE_STATFLAG_LINKSTATE_CHANGED
4340 		 * bit in status word always set. Workaround this bug by
4341 		 * reading PHY link status directly.
4342 		 */
4343 		link = (CSR_READ_4(sc, BGE_MI_STS) & BGE_MISTS_LINK)?
4344 		    BGE_STS_LINK : 0;
4345 
4346 		if (BGE_STS_BIT(sc, BGE_STS_LINK) != link) {
4347 			mii_pollstat(mii);
4348 
4349 			if (!BGE_STS_BIT(sc, BGE_STS_LINK) &&
4350 			    mii->mii_media_status & IFM_ACTIVE &&
4351 			    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE)
4352 				BGE_STS_SETBIT(sc, BGE_STS_LINK);
4353 			else if (BGE_STS_BIT(sc, BGE_STS_LINK) &&
4354 			    (!(mii->mii_media_status & IFM_ACTIVE) ||
4355 			    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE))
4356 				BGE_STS_CLRBIT(sc, BGE_STS_LINK);
4357 		}
4358 	}
4359 
4360 	/* Clear the attention */
4361 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
4362 	    BGE_MACSTAT_CFG_CHANGED|BGE_MACSTAT_MI_COMPLETE|
4363 	    BGE_MACSTAT_LINK_CHANGED);
4364 }
4365 
4366 static int
4367 sysctl_bge_verify(SYSCTLFN_ARGS)
4368 {
4369 	int error, t;
4370 	struct sysctlnode node;
4371 
4372 	node = *rnode;
4373 	t = *(int*)rnode->sysctl_data;
4374 	node.sysctl_data = &t;
4375 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
4376 	if (error || newp == NULL)
4377 		return (error);
4378 
4379 #if 0
4380 	DPRINTF2(("%s: t = %d, nodenum = %d, rnodenum = %d\n", __func__, t,
4381 	    node.sysctl_num, rnode->sysctl_num));
4382 #endif
4383 
4384 	if (node.sysctl_num == bge_rxthresh_nodenum) {
4385 		if (t < 0 || t >= NBGE_RX_THRESH)
4386 			return (EINVAL);
4387 		bge_update_all_threshes(t);
4388 	} else
4389 		return (EINVAL);
4390 
4391 	*(int*)rnode->sysctl_data = t;
4392 
4393 	return (0);
4394 }
4395 
4396 /*
4397  * Set up sysctl(3) MIB, hw.bge.*.
4398  *
4399  * TBD condition SYSCTL_PERMANENT on being an LKM or not
4400  */
4401 SYSCTL_SETUP(sysctl_bge, "sysctl bge subtree setup")
4402 {
4403 	int rc, bge_root_num;
4404 	const struct sysctlnode *node;
4405 
4406 	if ((rc = sysctl_createv(clog, 0, NULL, NULL,
4407 	    CTLFLAG_PERMANENT, CTLTYPE_NODE, "hw", NULL,
4408 	    NULL, 0, NULL, 0, CTL_HW, CTL_EOL)) != 0) {
4409 		goto err;
4410 	}
4411 
4412 	if ((rc = sysctl_createv(clog, 0, NULL, &node,
4413 	    CTLFLAG_PERMANENT, CTLTYPE_NODE, "bge",
4414 	    SYSCTL_DESCR("BGE interface controls"),
4415 	    NULL, 0, NULL, 0, CTL_HW, CTL_CREATE, CTL_EOL)) != 0) {
4416 		goto err;
4417 	}
4418 
4419 	bge_root_num = node->sysctl_num;
4420 
4421 	/* BGE Rx interrupt mitigation level */
4422 	if ((rc = sysctl_createv(clog, 0, NULL, &node,
4423 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
4424 	    CTLTYPE_INT, "rx_lvl",
4425 	    SYSCTL_DESCR("BGE receive interrupt mitigation level"),
4426 	    sysctl_bge_verify, 0,
4427 	    &bge_rx_thresh_lvl,
4428 	    0, CTL_HW, bge_root_num, CTL_CREATE,
4429 	    CTL_EOL)) != 0) {
4430 		goto err;
4431 	}
4432 
4433 	bge_rxthresh_nodenum = node->sysctl_num;
4434 
4435 	return;
4436 
4437 err:
4438 	aprint_error("%s: sysctl_createv failed (rc = %d)\n", __func__, rc);
4439 }
4440 
4441 static int
4442 bge_get_eaddr_mem(struct bge_softc *sc, u_int8_t ether_addr[])
4443 {
4444 	u_int32_t mac_addr;
4445 
4446 	mac_addr = bge_readmem_ind(sc, 0x0c14);
4447 	if ((mac_addr >> 16) == 0x484b) {
4448 		ether_addr[0] = (uint8_t)(mac_addr >> 8);
4449 		ether_addr[1] = (uint8_t)mac_addr;
4450 		mac_addr = bge_readmem_ind(sc, 0x0c18);
4451 		ether_addr[2] = (uint8_t)(mac_addr >> 24);
4452 		ether_addr[3] = (uint8_t)(mac_addr >> 16);
4453 		ether_addr[4] = (uint8_t)(mac_addr >> 8);
4454 		ether_addr[5] = (uint8_t)mac_addr;
4455 		return (0);
4456 	}
4457 	return (1);
4458 }
4459 
4460 static int
4461 bge_get_eaddr_nvram(struct bge_softc *sc, u_int8_t ether_addr[])
4462 {
4463 	int mac_offset = BGE_EE_MAC_OFFSET;
4464 
4465 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
4466 		mac_offset = BGE_EE_MAC_OFFSET_5906;
4467 	}
4468 
4469 	return (bge_read_nvram(sc, ether_addr, mac_offset + 2,
4470 	    ETHER_ADDR_LEN));
4471 }
4472 
4473 static int
4474 bge_get_eaddr_eeprom(struct bge_softc *sc, u_int8_t ether_addr[])
4475 {
4476 
4477 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_BCM5906) {
4478 		return (1);
4479 	}
4480 
4481 	return (bge_read_eeprom(sc, ether_addr, BGE_EE_MAC_OFFSET + 2,
4482 	   ETHER_ADDR_LEN));
4483 }
4484 
4485 static int
4486 bge_get_eaddr(struct bge_softc *sc, u_int8_t eaddr[])
4487 {
4488 	static const bge_eaddr_fcn_t bge_eaddr_funcs[] = {
4489 		/* NOTE: Order is critical */
4490 		bge_get_eaddr_mem,
4491 		bge_get_eaddr_nvram,
4492 		bge_get_eaddr_eeprom,
4493 		NULL
4494 	};
4495 	const bge_eaddr_fcn_t *func;
4496 
4497 	for (func = bge_eaddr_funcs; *func != NULL; ++func) {
4498 		if ((*func)(sc, eaddr) == 0)
4499 			break;
4500 	}
4501 	return (*func == NULL ? ENXIO : 0);
4502 }
4503