xref: /freebsd/sys/dev/bge/if_bge.c (revision 3cbb4cc200f8a0ad7ed08233425ea54524a21f1c)
1 /*-
2  * SPDX-License-Identifier: BSD-4-Clause
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 
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 /*
40  * Broadcom BCM57xx(x)/BCM590x NetXtreme and NetLink family Ethernet driver
41  *
42  * The Broadcom BCM5700 is based on technology originally developed by
43  * Alteon Networks as part of the Tigon I and Tigon II Gigabit Ethernet
44  * MAC chips. The BCM5700, sometimes referred to as the Tigon III, has
45  * two on-board MIPS R4000 CPUs and can have as much as 16MB of external
46  * SSRAM. The BCM5700 supports TCP, UDP and IP checksum offload, jumbo
47  * frames, highly configurable RX filtering, and 16 RX and TX queues
48  * (which, along with RX filter rules, can be used for QOS applications).
49  * Other features, such as TCP segmentation, may be available as part
50  * of value-added firmware updates. Unlike the Tigon I and Tigon II,
51  * firmware images can be stored in hardware and need not be compiled
52  * into the driver.
53  *
54  * The BCM5700 supports the PCI v2.2 and PCI-X v1.0 standards, and will
55  * function in a 32-bit/64-bit 33/66Mhz bus, or a 64-bit/133Mhz bus.
56  *
57  * The BCM5701 is a single-chip solution incorporating both the BCM5700
58  * MAC and a BCM5401 10/100/1000 PHY. Unlike the BCM5700, the BCM5701
59  * does not support external SSRAM.
60  *
61  * Broadcom also produces a variation of the BCM5700 under the "Altima"
62  * brand name, which is functionally similar but lacks PCI-X support.
63  *
64  * Without external SSRAM, you can only have at most 4 TX rings,
65  * and the use of the mini RX ring is disabled. This seems to imply
66  * that these features are simply not available on the BCM5701. As a
67  * result, this driver does not implement any support for the mini RX
68  * ring.
69  */
70 
71 #ifdef HAVE_KERNEL_OPTION_HEADERS
72 #include "opt_device_polling.h"
73 #endif
74 
75 #include <sys/param.h>
76 #include <sys/endian.h>
77 #include <sys/systm.h>
78 #include <sys/sockio.h>
79 #include <sys/mbuf.h>
80 #include <sys/malloc.h>
81 #include <sys/kernel.h>
82 #include <sys/module.h>
83 #include <sys/socket.h>
84 #include <sys/sysctl.h>
85 #include <sys/taskqueue.h>
86 
87 #include <net/debugnet.h>
88 #include <net/if.h>
89 #include <net/if_var.h>
90 #include <net/if_arp.h>
91 #include <net/ethernet.h>
92 #include <net/if_dl.h>
93 #include <net/if_media.h>
94 
95 #include <net/bpf.h>
96 
97 #include <net/if_types.h>
98 #include <net/if_vlan_var.h>
99 
100 #include <netinet/in_systm.h>
101 #include <netinet/in.h>
102 #include <netinet/ip.h>
103 #include <netinet/tcp.h>
104 
105 #include <machine/bus.h>
106 #include <machine/resource.h>
107 #include <sys/bus.h>
108 #include <sys/rman.h>
109 
110 #include <dev/mii/mii.h>
111 #include <dev/mii/miivar.h>
112 #include "miidevs.h"
113 #include <dev/mii/brgphyreg.h>
114 
115 #include <dev/pci/pcireg.h>
116 #include <dev/pci/pcivar.h>
117 
118 #include <dev/bge/if_bgereg.h>
119 
120 #define	BGE_CSUM_FEATURES	(CSUM_IP | CSUM_TCP)
121 #define	ETHER_MIN_NOPAD		(ETHER_MIN_LEN - ETHER_CRC_LEN) /* i.e., 60 */
122 
123 MODULE_DEPEND(bge, pci, 1, 1, 1);
124 MODULE_DEPEND(bge, ether, 1, 1, 1);
125 MODULE_DEPEND(bge, miibus, 1, 1, 1);
126 
127 /* "device miibus" required.  See GENERIC if you get errors here. */
128 #include "miibus_if.h"
129 
130 /*
131  * Various supported device vendors/types and their names. Note: the
132  * spec seems to indicate that the hardware still has Alteon's vendor
133  * ID burned into it, though it will always be overriden by the vendor
134  * ID in the EEPROM. Just to be safe, we cover all possibilities.
135  */
136 static const struct bge_type {
137 	uint16_t	bge_vid;
138 	uint16_t	bge_did;
139 } bge_devs[] = {
140 	{ ALTEON_VENDORID,	ALTEON_DEVICEID_BCM5700 },
141 	{ ALTEON_VENDORID,	ALTEON_DEVICEID_BCM5701 },
142 
143 	{ ALTIMA_VENDORID,	ALTIMA_DEVICE_AC1000 },
144 	{ ALTIMA_VENDORID,	ALTIMA_DEVICE_AC1002 },
145 	{ ALTIMA_VENDORID,	ALTIMA_DEVICE_AC9100 },
146 
147 	{ APPLE_VENDORID,	APPLE_DEVICE_BCM5701 },
148 
149 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5700 },
150 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5701 },
151 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5702 },
152 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5702_ALT },
153 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5702X },
154 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5703 },
155 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5703_ALT },
156 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5703X },
157 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5704C },
158 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5704S },
159 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5704S_ALT },
160 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5705 },
161 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5705F },
162 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5705K },
163 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5705M },
164 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5705M_ALT },
165 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5714C },
166 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5714S },
167 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5715 },
168 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5715S },
169 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5717 },
170 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5717C },
171 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5718 },
172 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5719 },
173 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5720 },
174 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5721 },
175 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5722 },
176 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5723 },
177 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5725 },
178 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5727 },
179 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5750 },
180 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5750M },
181 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5751 },
182 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5751F },
183 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5751M },
184 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5752 },
185 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5752M },
186 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5753 },
187 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5753F },
188 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5753M },
189 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5754 },
190 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5754M },
191 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5755 },
192 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5755M },
193 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5756 },
194 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5761 },
195 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5761E },
196 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5761S },
197 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5761SE },
198 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5762 },
199 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5764 },
200 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5780 },
201 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5780S },
202 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5781 },
203 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5782 },
204 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5784 },
205 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5785F },
206 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5785G },
207 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5786 },
208 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5787 },
209 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5787F },
210 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5787M },
211 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5788 },
212 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5789 },
213 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5901 },
214 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5901A2 },
215 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5903M },
216 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5906 },
217 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5906M },
218 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57760 },
219 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57761 },
220 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57762 },
221 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57764 },
222 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57765 },
223 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57766 },
224 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57767 },
225 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57780 },
226 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57781 },
227 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57782 },
228 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57785 },
229 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57786 },
230 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57787 },
231 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57788 },
232 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57790 },
233 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57791 },
234 	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57795 },
235 
236 	{ SK_VENDORID,		SK_DEVICEID_ALTIMA },
237 
238 	{ TC_VENDORID,		TC_DEVICEID_3C996 },
239 
240 	{ FJTSU_VENDORID,	FJTSU_DEVICEID_PW008GE4 },
241 	{ FJTSU_VENDORID,	FJTSU_DEVICEID_PW008GE5 },
242 	{ FJTSU_VENDORID,	FJTSU_DEVICEID_PP250450 },
243 	{ 0, 0 }
244 };
245 
246 static const struct bge_vendor {
247 	uint16_t	v_id;
248 	const char	*v_name;
249 } bge_vendors[] = {
250 	{ ALTEON_VENDORID,	"Alteon" },
251 	{ ALTIMA_VENDORID,	"Altima" },
252 	{ APPLE_VENDORID,	"Apple" },
253 	{ BCOM_VENDORID,	"Broadcom" },
254 	{ SK_VENDORID,		"SysKonnect" },
255 	{ TC_VENDORID,		"3Com" },
256 	{ FJTSU_VENDORID,	"Fujitsu" },
257 	{ 0, NULL }
258 };
259 
260 static const struct bge_revision {
261 	uint32_t	br_chipid;
262 	const char	*br_name;
263 } bge_revisions[] = {
264 	{ BGE_CHIPID_BCM5700_A0,	"BCM5700 A0" },
265 	{ BGE_CHIPID_BCM5700_A1,	"BCM5700 A1" },
266 	{ BGE_CHIPID_BCM5700_B0,	"BCM5700 B0" },
267 	{ BGE_CHIPID_BCM5700_B1,	"BCM5700 B1" },
268 	{ BGE_CHIPID_BCM5700_B2,	"BCM5700 B2" },
269 	{ BGE_CHIPID_BCM5700_B3,	"BCM5700 B3" },
270 	{ BGE_CHIPID_BCM5700_ALTIMA,	"BCM5700 Altima" },
271 	{ BGE_CHIPID_BCM5700_C0,	"BCM5700 C0" },
272 	{ BGE_CHIPID_BCM5701_A0,	"BCM5701 A0" },
273 	{ BGE_CHIPID_BCM5701_B0,	"BCM5701 B0" },
274 	{ BGE_CHIPID_BCM5701_B2,	"BCM5701 B2" },
275 	{ BGE_CHIPID_BCM5701_B5,	"BCM5701 B5" },
276 	{ BGE_CHIPID_BCM5703_A0,	"BCM5703 A0" },
277 	{ BGE_CHIPID_BCM5703_A1,	"BCM5703 A1" },
278 	{ BGE_CHIPID_BCM5703_A2,	"BCM5703 A2" },
279 	{ BGE_CHIPID_BCM5703_A3,	"BCM5703 A3" },
280 	{ BGE_CHIPID_BCM5703_B0,	"BCM5703 B0" },
281 	{ BGE_CHIPID_BCM5704_A0,	"BCM5704 A0" },
282 	{ BGE_CHIPID_BCM5704_A1,	"BCM5704 A1" },
283 	{ BGE_CHIPID_BCM5704_A2,	"BCM5704 A2" },
284 	{ BGE_CHIPID_BCM5704_A3,	"BCM5704 A3" },
285 	{ BGE_CHIPID_BCM5704_B0,	"BCM5704 B0" },
286 	{ BGE_CHIPID_BCM5705_A0,	"BCM5705 A0" },
287 	{ BGE_CHIPID_BCM5705_A1,	"BCM5705 A1" },
288 	{ BGE_CHIPID_BCM5705_A2,	"BCM5705 A2" },
289 	{ BGE_CHIPID_BCM5705_A3,	"BCM5705 A3" },
290 	{ BGE_CHIPID_BCM5750_A0,	"BCM5750 A0" },
291 	{ BGE_CHIPID_BCM5750_A1,	"BCM5750 A1" },
292 	{ BGE_CHIPID_BCM5750_A3,	"BCM5750 A3" },
293 	{ BGE_CHIPID_BCM5750_B0,	"BCM5750 B0" },
294 	{ BGE_CHIPID_BCM5750_B1,	"BCM5750 B1" },
295 	{ BGE_CHIPID_BCM5750_C0,	"BCM5750 C0" },
296 	{ BGE_CHIPID_BCM5750_C1,	"BCM5750 C1" },
297 	{ BGE_CHIPID_BCM5750_C2,	"BCM5750 C2" },
298 	{ BGE_CHIPID_BCM5714_A0,	"BCM5714 A0" },
299 	{ BGE_CHIPID_BCM5752_A0,	"BCM5752 A0" },
300 	{ BGE_CHIPID_BCM5752_A1,	"BCM5752 A1" },
301 	{ BGE_CHIPID_BCM5752_A2,	"BCM5752 A2" },
302 	{ BGE_CHIPID_BCM5714_B0,	"BCM5714 B0" },
303 	{ BGE_CHIPID_BCM5714_B3,	"BCM5714 B3" },
304 	{ BGE_CHIPID_BCM5715_A0,	"BCM5715 A0" },
305 	{ BGE_CHIPID_BCM5715_A1,	"BCM5715 A1" },
306 	{ BGE_CHIPID_BCM5715_A3,	"BCM5715 A3" },
307 	{ BGE_CHIPID_BCM5717_A0,	"BCM5717 A0" },
308 	{ BGE_CHIPID_BCM5717_B0,	"BCM5717 B0" },
309 	{ BGE_CHIPID_BCM5717_C0,	"BCM5717 C0" },
310 	{ BGE_CHIPID_BCM5719_A0,	"BCM5719 A0" },
311 	{ BGE_CHIPID_BCM5720_A0,	"BCM5720 A0" },
312 	{ BGE_CHIPID_BCM5755_A0,	"BCM5755 A0" },
313 	{ BGE_CHIPID_BCM5755_A1,	"BCM5755 A1" },
314 	{ BGE_CHIPID_BCM5755_A2,	"BCM5755 A2" },
315 	{ BGE_CHIPID_BCM5722_A0,	"BCM5722 A0" },
316 	{ BGE_CHIPID_BCM5761_A0,	"BCM5761 A0" },
317 	{ BGE_CHIPID_BCM5761_A1,	"BCM5761 A1" },
318 	{ BGE_CHIPID_BCM5762_A0,	"BCM5762 A0" },
319 	{ BGE_CHIPID_BCM5784_A0,	"BCM5784 A0" },
320 	{ BGE_CHIPID_BCM5784_A1,	"BCM5784 A1" },
321 	/* 5754 and 5787 share the same ASIC ID */
322 	{ BGE_CHIPID_BCM5787_A0,	"BCM5754/5787 A0" },
323 	{ BGE_CHIPID_BCM5787_A1,	"BCM5754/5787 A1" },
324 	{ BGE_CHIPID_BCM5787_A2,	"BCM5754/5787 A2" },
325 	{ BGE_CHIPID_BCM5906_A1,	"BCM5906 A1" },
326 	{ BGE_CHIPID_BCM5906_A2,	"BCM5906 A2" },
327 	{ BGE_CHIPID_BCM57765_A0,	"BCM57765 A0" },
328 	{ BGE_CHIPID_BCM57765_B0,	"BCM57765 B0" },
329 	{ BGE_CHIPID_BCM57780_A0,	"BCM57780 A0" },
330 	{ BGE_CHIPID_BCM57780_A1,	"BCM57780 A1" },
331 	{ 0, NULL }
332 };
333 
334 /*
335  * Some defaults for major revisions, so that newer steppings
336  * that we don't know about have a shot at working.
337  */
338 static const struct bge_revision bge_majorrevs[] = {
339 	{ BGE_ASICREV_BCM5700,		"unknown BCM5700" },
340 	{ BGE_ASICREV_BCM5701,		"unknown BCM5701" },
341 	{ BGE_ASICREV_BCM5703,		"unknown BCM5703" },
342 	{ BGE_ASICREV_BCM5704,		"unknown BCM5704" },
343 	{ BGE_ASICREV_BCM5705,		"unknown BCM5705" },
344 	{ BGE_ASICREV_BCM5750,		"unknown BCM5750" },
345 	{ BGE_ASICREV_BCM5714_A0,	"unknown BCM5714" },
346 	{ BGE_ASICREV_BCM5752,		"unknown BCM5752" },
347 	{ BGE_ASICREV_BCM5780,		"unknown BCM5780" },
348 	{ BGE_ASICREV_BCM5714,		"unknown BCM5714" },
349 	{ BGE_ASICREV_BCM5755,		"unknown BCM5755" },
350 	{ BGE_ASICREV_BCM5761,		"unknown BCM5761" },
351 	{ BGE_ASICREV_BCM5784,		"unknown BCM5784" },
352 	{ BGE_ASICREV_BCM5785,		"unknown BCM5785" },
353 	/* 5754 and 5787 share the same ASIC ID */
354 	{ BGE_ASICREV_BCM5787,		"unknown BCM5754/5787" },
355 	{ BGE_ASICREV_BCM5906,		"unknown BCM5906" },
356 	{ BGE_ASICREV_BCM57765,		"unknown BCM57765" },
357 	{ BGE_ASICREV_BCM57766,		"unknown BCM57766" },
358 	{ BGE_ASICREV_BCM57780,		"unknown BCM57780" },
359 	{ BGE_ASICREV_BCM5717,		"unknown BCM5717" },
360 	{ BGE_ASICREV_BCM5719,		"unknown BCM5719" },
361 	{ BGE_ASICREV_BCM5720,		"unknown BCM5720" },
362 	{ BGE_ASICREV_BCM5762,		"unknown BCM5762" },
363 	{ 0, NULL }
364 };
365 
366 #define	BGE_IS_JUMBO_CAPABLE(sc)	((sc)->bge_flags & BGE_FLAG_JUMBO)
367 #define	BGE_IS_5700_FAMILY(sc)		((sc)->bge_flags & BGE_FLAG_5700_FAMILY)
368 #define	BGE_IS_5705_PLUS(sc)		((sc)->bge_flags & BGE_FLAG_5705_PLUS)
369 #define	BGE_IS_5714_FAMILY(sc)		((sc)->bge_flags & BGE_FLAG_5714_FAMILY)
370 #define	BGE_IS_575X_PLUS(sc)		((sc)->bge_flags & BGE_FLAG_575X_PLUS)
371 #define	BGE_IS_5755_PLUS(sc)		((sc)->bge_flags & BGE_FLAG_5755_PLUS)
372 #define	BGE_IS_5717_PLUS(sc)		((sc)->bge_flags & BGE_FLAG_5717_PLUS)
373 #define	BGE_IS_57765_PLUS(sc)		((sc)->bge_flags & BGE_FLAG_57765_PLUS)
374 
375 static uint32_t bge_chipid(device_t);
376 static const struct bge_vendor * bge_lookup_vendor(uint16_t);
377 static const struct bge_revision * bge_lookup_rev(uint32_t);
378 
379 typedef int	(*bge_eaddr_fcn_t)(struct bge_softc *, uint8_t[]);
380 
381 static int bge_probe(device_t);
382 static int bge_attach(device_t);
383 static int bge_detach(device_t);
384 static int bge_suspend(device_t);
385 static int bge_resume(device_t);
386 static void bge_release_resources(struct bge_softc *);
387 static void bge_dma_map_addr(void *, bus_dma_segment_t *, int, int);
388 static int bge_dma_alloc(struct bge_softc *);
389 static void bge_dma_free(struct bge_softc *);
390 static int bge_dma_ring_alloc(struct bge_softc *, bus_size_t, bus_size_t,
391     bus_dma_tag_t *, uint8_t **, bus_dmamap_t *, bus_addr_t *, const char *);
392 
393 static void bge_devinfo(struct bge_softc *);
394 static int bge_mbox_reorder(struct bge_softc *);
395 
396 static int bge_get_eaddr_fw(struct bge_softc *sc, uint8_t ether_addr[]);
397 static int bge_get_eaddr_mem(struct bge_softc *, uint8_t[]);
398 static int bge_get_eaddr_nvram(struct bge_softc *, uint8_t[]);
399 static int bge_get_eaddr_eeprom(struct bge_softc *, uint8_t[]);
400 static int bge_get_eaddr(struct bge_softc *, uint8_t[]);
401 
402 static void bge_txeof(struct bge_softc *, uint16_t);
403 static void bge_rxcsum(struct bge_softc *, struct bge_rx_bd *, struct mbuf *);
404 static int bge_rxeof(struct bge_softc *, uint16_t, int);
405 
406 static void bge_asf_driver_up (struct bge_softc *);
407 static void bge_tick(void *);
408 static void bge_stats_clear_regs(struct bge_softc *);
409 static void bge_stats_update(struct bge_softc *);
410 static void bge_stats_update_regs(struct bge_softc *);
411 static struct mbuf *bge_check_short_dma(struct mbuf *);
412 static struct mbuf *bge_setup_tso(struct bge_softc *, struct mbuf *,
413     uint16_t *, uint16_t *);
414 static int bge_encap(struct bge_softc *, struct mbuf **, uint32_t *);
415 
416 static void bge_intr(void *);
417 static int bge_msi_intr(void *);
418 static void bge_intr_task(void *, int);
419 static void bge_start(if_t);
420 static void bge_start_locked(if_t);
421 static void bge_start_tx(struct bge_softc *, uint32_t);
422 static int bge_ioctl(if_t, u_long, caddr_t);
423 static void bge_init_locked(struct bge_softc *);
424 static void bge_init(void *);
425 static void bge_stop_block(struct bge_softc *, bus_size_t, uint32_t);
426 static void bge_stop(struct bge_softc *);
427 static void bge_watchdog(struct bge_softc *);
428 static int bge_shutdown(device_t);
429 static int bge_ifmedia_upd_locked(if_t);
430 static int bge_ifmedia_upd(if_t);
431 static void bge_ifmedia_sts(if_t, struct ifmediareq *);
432 static uint64_t bge_get_counter(if_t, ift_counter);
433 
434 static uint8_t bge_nvram_getbyte(struct bge_softc *, int, uint8_t *);
435 static int bge_read_nvram(struct bge_softc *, caddr_t, int, int);
436 
437 static uint8_t bge_eeprom_getbyte(struct bge_softc *, int, uint8_t *);
438 static int bge_read_eeprom(struct bge_softc *, caddr_t, int, int);
439 
440 static void bge_setpromisc(struct bge_softc *);
441 static void bge_setmulti(struct bge_softc *);
442 static void bge_setvlan(struct bge_softc *);
443 
444 static __inline void bge_rxreuse_std(struct bge_softc *, int);
445 static __inline void bge_rxreuse_jumbo(struct bge_softc *, int);
446 static int bge_newbuf_std(struct bge_softc *, int);
447 static int bge_newbuf_jumbo(struct bge_softc *, int);
448 static int bge_init_rx_ring_std(struct bge_softc *);
449 static void bge_free_rx_ring_std(struct bge_softc *);
450 static int bge_init_rx_ring_jumbo(struct bge_softc *);
451 static void bge_free_rx_ring_jumbo(struct bge_softc *);
452 static void bge_free_tx_ring(struct bge_softc *);
453 static int bge_init_tx_ring(struct bge_softc *);
454 
455 static int bge_chipinit(struct bge_softc *);
456 static int bge_blockinit(struct bge_softc *);
457 static uint32_t bge_dma_swap_options(struct bge_softc *);
458 
459 static int bge_has_eaddr(struct bge_softc *);
460 static uint32_t bge_readmem_ind(struct bge_softc *, int);
461 static void bge_writemem_ind(struct bge_softc *, int, int);
462 static void bge_writembx(struct bge_softc *, int, int);
463 #ifdef notdef
464 static uint32_t bge_readreg_ind(struct bge_softc *, int);
465 #endif
466 static void bge_writemem_direct(struct bge_softc *, int, int);
467 static void bge_writereg_ind(struct bge_softc *, int, int);
468 
469 static int bge_miibus_readreg(device_t, int, int);
470 static int bge_miibus_writereg(device_t, int, int, int);
471 static void bge_miibus_statchg(device_t);
472 #ifdef DEVICE_POLLING
473 static int bge_poll(if_t ifp, enum poll_cmd cmd, int count);
474 #endif
475 
476 #define	BGE_RESET_SHUTDOWN	0
477 #define	BGE_RESET_START		1
478 #define	BGE_RESET_SUSPEND	2
479 static void bge_sig_post_reset(struct bge_softc *, int);
480 static void bge_sig_legacy(struct bge_softc *, int);
481 static void bge_sig_pre_reset(struct bge_softc *, int);
482 static void bge_stop_fw(struct bge_softc *);
483 static int bge_reset(struct bge_softc *);
484 static void bge_link_upd(struct bge_softc *);
485 
486 static void bge_ape_lock_init(struct bge_softc *);
487 static void bge_ape_read_fw_ver(struct bge_softc *);
488 static int bge_ape_lock(struct bge_softc *, int);
489 static void bge_ape_unlock(struct bge_softc *, int);
490 static void bge_ape_send_event(struct bge_softc *, uint32_t);
491 static void bge_ape_driver_state_change(struct bge_softc *, int);
492 
493 /*
494  * The BGE_REGISTER_DEBUG option is only for low-level debugging.  It may
495  * leak information to untrusted users.  It is also known to cause alignment
496  * traps on certain architectures.
497  */
498 #ifdef BGE_REGISTER_DEBUG
499 static int bge_sysctl_debug_info(SYSCTL_HANDLER_ARGS);
500 static int bge_sysctl_reg_read(SYSCTL_HANDLER_ARGS);
501 static int bge_sysctl_ape_read(SYSCTL_HANDLER_ARGS);
502 static int bge_sysctl_mem_read(SYSCTL_HANDLER_ARGS);
503 #endif
504 static void bge_add_sysctls(struct bge_softc *);
505 static void bge_add_sysctl_stats_regs(struct bge_softc *,
506     struct sysctl_ctx_list *, struct sysctl_oid_list *);
507 static void bge_add_sysctl_stats(struct bge_softc *, struct sysctl_ctx_list *,
508     struct sysctl_oid_list *);
509 static int bge_sysctl_stats(SYSCTL_HANDLER_ARGS);
510 
511 DEBUGNET_DEFINE(bge);
512 
513 static device_method_t bge_methods[] = {
514 	/* Device interface */
515 	DEVMETHOD(device_probe,		bge_probe),
516 	DEVMETHOD(device_attach,	bge_attach),
517 	DEVMETHOD(device_detach,	bge_detach),
518 	DEVMETHOD(device_shutdown,	bge_shutdown),
519 	DEVMETHOD(device_suspend,	bge_suspend),
520 	DEVMETHOD(device_resume,	bge_resume),
521 
522 	/* MII interface */
523 	DEVMETHOD(miibus_readreg,	bge_miibus_readreg),
524 	DEVMETHOD(miibus_writereg,	bge_miibus_writereg),
525 	DEVMETHOD(miibus_statchg,	bge_miibus_statchg),
526 
527 	DEVMETHOD_END
528 };
529 
530 static driver_t bge_driver = {
531 	"bge",
532 	bge_methods,
533 	sizeof(struct bge_softc)
534 };
535 
536 static devclass_t bge_devclass;
537 
538 DRIVER_MODULE(bge, pci, bge_driver, bge_devclass, 0, 0);
539 MODULE_PNP_INFO("U16:vendor;U16:device", pci, bge, bge_devs,
540     nitems(bge_devs) - 1);
541 DRIVER_MODULE(miibus, bge, miibus_driver, miibus_devclass, 0, 0);
542 
543 static int bge_allow_asf = 1;
544 
545 static SYSCTL_NODE(_hw, OID_AUTO, bge, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
546     "BGE driver parameters");
547 SYSCTL_INT(_hw_bge, OID_AUTO, allow_asf, CTLFLAG_RDTUN, &bge_allow_asf, 0,
548 	"Allow ASF mode if available");
549 
550 static int
551 bge_has_eaddr(struct bge_softc *sc)
552 {
553 	return (1);
554 }
555 
556 static uint32_t
557 bge_readmem_ind(struct bge_softc *sc, int off)
558 {
559 	device_t dev;
560 	uint32_t val;
561 
562 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
563 	    off >= BGE_STATS_BLOCK && off < BGE_SEND_RING_1_TO_4)
564 		return (0);
565 
566 	dev = sc->bge_dev;
567 
568 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4);
569 	val = pci_read_config(dev, BGE_PCI_MEMWIN_DATA, 4);
570 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4);
571 	return (val);
572 }
573 
574 static void
575 bge_writemem_ind(struct bge_softc *sc, int off, int val)
576 {
577 	device_t dev;
578 
579 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
580 	    off >= BGE_STATS_BLOCK && off < BGE_SEND_RING_1_TO_4)
581 		return;
582 
583 	dev = sc->bge_dev;
584 
585 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4);
586 	pci_write_config(dev, BGE_PCI_MEMWIN_DATA, val, 4);
587 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4);
588 }
589 
590 #ifdef notdef
591 static uint32_t
592 bge_readreg_ind(struct bge_softc *sc, int off)
593 {
594 	device_t dev;
595 
596 	dev = sc->bge_dev;
597 
598 	pci_write_config(dev, BGE_PCI_REG_BASEADDR, off, 4);
599 	return (pci_read_config(dev, BGE_PCI_REG_DATA, 4));
600 }
601 #endif
602 
603 static void
604 bge_writereg_ind(struct bge_softc *sc, int off, int val)
605 {
606 	device_t dev;
607 
608 	dev = sc->bge_dev;
609 
610 	pci_write_config(dev, BGE_PCI_REG_BASEADDR, off, 4);
611 	pci_write_config(dev, BGE_PCI_REG_DATA, val, 4);
612 }
613 
614 static void
615 bge_writemem_direct(struct bge_softc *sc, int off, int val)
616 {
617 	CSR_WRITE_4(sc, off, val);
618 }
619 
620 static void
621 bge_writembx(struct bge_softc *sc, int off, int val)
622 {
623 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
624 		off += BGE_LPMBX_IRQ0_HI - BGE_MBX_IRQ0_HI;
625 
626 	CSR_WRITE_4(sc, off, val);
627 	if ((sc->bge_flags & BGE_FLAG_MBOX_REORDER) != 0)
628 		CSR_READ_4(sc, off);
629 }
630 
631 /*
632  * Clear all stale locks and select the lock for this driver instance.
633  */
634 static void
635 bge_ape_lock_init(struct bge_softc *sc)
636 {
637 	uint32_t bit, regbase;
638 	int i;
639 
640 	if (sc->bge_asicrev == BGE_ASICREV_BCM5761)
641 		regbase = BGE_APE_LOCK_GRANT;
642 	else
643 		regbase = BGE_APE_PER_LOCK_GRANT;
644 
645 	/* Clear any stale locks. */
646 	for (i = BGE_APE_LOCK_PHY0; i <= BGE_APE_LOCK_GPIO; i++) {
647 		switch (i) {
648 		case BGE_APE_LOCK_PHY0:
649 		case BGE_APE_LOCK_PHY1:
650 		case BGE_APE_LOCK_PHY2:
651 		case BGE_APE_LOCK_PHY3:
652 			bit = BGE_APE_LOCK_GRANT_DRIVER0;
653 			break;
654 		default:
655 			if (sc->bge_func_addr == 0)
656 				bit = BGE_APE_LOCK_GRANT_DRIVER0;
657 			else
658 				bit = (1 << sc->bge_func_addr);
659 		}
660 		APE_WRITE_4(sc, regbase + 4 * i, bit);
661 	}
662 
663 	/* Select the PHY lock based on the device's function number. */
664 	switch (sc->bge_func_addr) {
665 	case 0:
666 		sc->bge_phy_ape_lock = BGE_APE_LOCK_PHY0;
667 		break;
668 	case 1:
669 		sc->bge_phy_ape_lock = BGE_APE_LOCK_PHY1;
670 		break;
671 	case 2:
672 		sc->bge_phy_ape_lock = BGE_APE_LOCK_PHY2;
673 		break;
674 	case 3:
675 		sc->bge_phy_ape_lock = BGE_APE_LOCK_PHY3;
676 		break;
677 	default:
678 		device_printf(sc->bge_dev,
679 		    "PHY lock not supported on this function\n");
680 	}
681 }
682 
683 /*
684  * Check for APE firmware, set flags, and print version info.
685  */
686 static void
687 bge_ape_read_fw_ver(struct bge_softc *sc)
688 {
689 	const char *fwtype;
690 	uint32_t apedata, features;
691 
692 	/* Check for a valid APE signature in shared memory. */
693 	apedata = APE_READ_4(sc, BGE_APE_SEG_SIG);
694 	if (apedata != BGE_APE_SEG_SIG_MAGIC) {
695 		sc->bge_mfw_flags &= ~ BGE_MFW_ON_APE;
696 		return;
697 	}
698 
699 	/* Check if APE firmware is running. */
700 	apedata = APE_READ_4(sc, BGE_APE_FW_STATUS);
701 	if ((apedata & BGE_APE_FW_STATUS_READY) == 0) {
702 		device_printf(sc->bge_dev, "APE signature found "
703 		    "but FW status not ready! 0x%08x\n", apedata);
704 		return;
705 	}
706 
707 	sc->bge_mfw_flags |= BGE_MFW_ON_APE;
708 
709 	/* Fetch the APE firwmare type and version. */
710 	apedata = APE_READ_4(sc, BGE_APE_FW_VERSION);
711 	features = APE_READ_4(sc, BGE_APE_FW_FEATURES);
712 	if ((features & BGE_APE_FW_FEATURE_NCSI) != 0) {
713 		sc->bge_mfw_flags |= BGE_MFW_TYPE_NCSI;
714 		fwtype = "NCSI";
715 	} else if ((features & BGE_APE_FW_FEATURE_DASH) != 0) {
716 		sc->bge_mfw_flags |= BGE_MFW_TYPE_DASH;
717 		fwtype = "DASH";
718 	} else
719 		fwtype = "UNKN";
720 
721 	/* Print the APE firmware version. */
722 	device_printf(sc->bge_dev, "APE FW version: %s v%d.%d.%d.%d\n",
723 	    fwtype,
724 	    (apedata & BGE_APE_FW_VERSION_MAJMSK) >> BGE_APE_FW_VERSION_MAJSFT,
725 	    (apedata & BGE_APE_FW_VERSION_MINMSK) >> BGE_APE_FW_VERSION_MINSFT,
726 	    (apedata & BGE_APE_FW_VERSION_REVMSK) >> BGE_APE_FW_VERSION_REVSFT,
727 	    (apedata & BGE_APE_FW_VERSION_BLDMSK));
728 }
729 
730 static int
731 bge_ape_lock(struct bge_softc *sc, int locknum)
732 {
733 	uint32_t bit, gnt, req, status;
734 	int i, off;
735 
736 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) == 0)
737 		return (0);
738 
739 	/* Lock request/grant registers have different bases. */
740 	if (sc->bge_asicrev == BGE_ASICREV_BCM5761) {
741 		req = BGE_APE_LOCK_REQ;
742 		gnt = BGE_APE_LOCK_GRANT;
743 	} else {
744 		req = BGE_APE_PER_LOCK_REQ;
745 		gnt = BGE_APE_PER_LOCK_GRANT;
746 	}
747 
748 	off = 4 * locknum;
749 
750 	switch (locknum) {
751 	case BGE_APE_LOCK_GPIO:
752 		/* Lock required when using GPIO. */
753 		if (sc->bge_asicrev == BGE_ASICREV_BCM5761)
754 			return (0);
755 		if (sc->bge_func_addr == 0)
756 			bit = BGE_APE_LOCK_REQ_DRIVER0;
757 		else
758 			bit = (1 << sc->bge_func_addr);
759 		break;
760 	case BGE_APE_LOCK_GRC:
761 		/* Lock required to reset the device. */
762 		if (sc->bge_func_addr == 0)
763 			bit = BGE_APE_LOCK_REQ_DRIVER0;
764 		else
765 			bit = (1 << sc->bge_func_addr);
766 		break;
767 	case BGE_APE_LOCK_MEM:
768 		/* Lock required when accessing certain APE memory. */
769 		if (sc->bge_func_addr == 0)
770 			bit = BGE_APE_LOCK_REQ_DRIVER0;
771 		else
772 			bit = (1 << sc->bge_func_addr);
773 		break;
774 	case BGE_APE_LOCK_PHY0:
775 	case BGE_APE_LOCK_PHY1:
776 	case BGE_APE_LOCK_PHY2:
777 	case BGE_APE_LOCK_PHY3:
778 		/* Lock required when accessing PHYs. */
779 		bit = BGE_APE_LOCK_REQ_DRIVER0;
780 		break;
781 	default:
782 		return (EINVAL);
783 	}
784 
785 	/* Request a lock. */
786 	APE_WRITE_4(sc, req + off, bit);
787 
788 	/* Wait up to 1 second to acquire lock. */
789 	for (i = 0; i < 20000; i++) {
790 		status = APE_READ_4(sc, gnt + off);
791 		if (status == bit)
792 			break;
793 		DELAY(50);
794 	}
795 
796 	/* Handle any errors. */
797 	if (status != bit) {
798 		device_printf(sc->bge_dev, "APE lock %d request failed! "
799 		    "request = 0x%04x[0x%04x], status = 0x%04x[0x%04x]\n",
800 		    locknum, req + off, bit & 0xFFFF, gnt + off,
801 		    status & 0xFFFF);
802 		/* Revoke the lock request. */
803 		APE_WRITE_4(sc, gnt + off, bit);
804 		return (EBUSY);
805 	}
806 
807 	return (0);
808 }
809 
810 static void
811 bge_ape_unlock(struct bge_softc *sc, int locknum)
812 {
813 	uint32_t bit, gnt;
814 	int off;
815 
816 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) == 0)
817 		return;
818 
819 	if (sc->bge_asicrev == BGE_ASICREV_BCM5761)
820 		gnt = BGE_APE_LOCK_GRANT;
821 	else
822 		gnt = BGE_APE_PER_LOCK_GRANT;
823 
824 	off = 4 * locknum;
825 
826 	switch (locknum) {
827 	case BGE_APE_LOCK_GPIO:
828 		if (sc->bge_asicrev == BGE_ASICREV_BCM5761)
829 			return;
830 		if (sc->bge_func_addr == 0)
831 			bit = BGE_APE_LOCK_GRANT_DRIVER0;
832 		else
833 			bit = (1 << sc->bge_func_addr);
834 		break;
835 	case BGE_APE_LOCK_GRC:
836 		if (sc->bge_func_addr == 0)
837 			bit = BGE_APE_LOCK_GRANT_DRIVER0;
838 		else
839 			bit = (1 << sc->bge_func_addr);
840 		break;
841 	case BGE_APE_LOCK_MEM:
842 		if (sc->bge_func_addr == 0)
843 			bit = BGE_APE_LOCK_GRANT_DRIVER0;
844 		else
845 			bit = (1 << sc->bge_func_addr);
846 		break;
847 	case BGE_APE_LOCK_PHY0:
848 	case BGE_APE_LOCK_PHY1:
849 	case BGE_APE_LOCK_PHY2:
850 	case BGE_APE_LOCK_PHY3:
851 		bit = BGE_APE_LOCK_GRANT_DRIVER0;
852 		break;
853 	default:
854 		return;
855 	}
856 
857 	APE_WRITE_4(sc, gnt + off, bit);
858 }
859 
860 /*
861  * Send an event to the APE firmware.
862  */
863 static void
864 bge_ape_send_event(struct bge_softc *sc, uint32_t event)
865 {
866 	uint32_t apedata;
867 	int i;
868 
869 	/* NCSI does not support APE events. */
870 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) == 0)
871 		return;
872 
873 	/* Wait up to 1ms for APE to service previous event. */
874 	for (i = 10; i > 0; i--) {
875 		if (bge_ape_lock(sc, BGE_APE_LOCK_MEM) != 0)
876 			break;
877 		apedata = APE_READ_4(sc, BGE_APE_EVENT_STATUS);
878 		if ((apedata & BGE_APE_EVENT_STATUS_EVENT_PENDING) == 0) {
879 			APE_WRITE_4(sc, BGE_APE_EVENT_STATUS, event |
880 			    BGE_APE_EVENT_STATUS_EVENT_PENDING);
881 			bge_ape_unlock(sc, BGE_APE_LOCK_MEM);
882 			APE_WRITE_4(sc, BGE_APE_EVENT, BGE_APE_EVENT_1);
883 			break;
884 		}
885 		bge_ape_unlock(sc, BGE_APE_LOCK_MEM);
886 		DELAY(100);
887 	}
888 	if (i == 0)
889 		device_printf(sc->bge_dev, "APE event 0x%08x send timed out\n",
890 		    event);
891 }
892 
893 static void
894 bge_ape_driver_state_change(struct bge_softc *sc, int kind)
895 {
896 	uint32_t apedata, event;
897 
898 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) == 0)
899 		return;
900 
901 	switch (kind) {
902 	case BGE_RESET_START:
903 		/* If this is the first load, clear the load counter. */
904 		apedata = APE_READ_4(sc, BGE_APE_HOST_SEG_SIG);
905 		if (apedata != BGE_APE_HOST_SEG_SIG_MAGIC)
906 			APE_WRITE_4(sc, BGE_APE_HOST_INIT_COUNT, 0);
907 		else {
908 			apedata = APE_READ_4(sc, BGE_APE_HOST_INIT_COUNT);
909 			APE_WRITE_4(sc, BGE_APE_HOST_INIT_COUNT, ++apedata);
910 		}
911 		APE_WRITE_4(sc, BGE_APE_HOST_SEG_SIG,
912 		    BGE_APE_HOST_SEG_SIG_MAGIC);
913 		APE_WRITE_4(sc, BGE_APE_HOST_SEG_LEN,
914 		    BGE_APE_HOST_SEG_LEN_MAGIC);
915 
916 		/* Add some version info if bge(4) supports it. */
917 		APE_WRITE_4(sc, BGE_APE_HOST_DRIVER_ID,
918 		    BGE_APE_HOST_DRIVER_ID_MAGIC(1, 0));
919 		APE_WRITE_4(sc, BGE_APE_HOST_BEHAVIOR,
920 		    BGE_APE_HOST_BEHAV_NO_PHYLOCK);
921 		APE_WRITE_4(sc, BGE_APE_HOST_HEARTBEAT_INT_MS,
922 		    BGE_APE_HOST_HEARTBEAT_INT_DISABLE);
923 		APE_WRITE_4(sc, BGE_APE_HOST_DRVR_STATE,
924 		    BGE_APE_HOST_DRVR_STATE_START);
925 		event = BGE_APE_EVENT_STATUS_STATE_START;
926 		break;
927 	case BGE_RESET_SHUTDOWN:
928 		APE_WRITE_4(sc, BGE_APE_HOST_DRVR_STATE,
929 		    BGE_APE_HOST_DRVR_STATE_UNLOAD);
930 		event = BGE_APE_EVENT_STATUS_STATE_UNLOAD;
931 		break;
932 	case BGE_RESET_SUSPEND:
933 		event = BGE_APE_EVENT_STATUS_STATE_SUSPEND;
934 		break;
935 	default:
936 		return;
937 	}
938 
939 	bge_ape_send_event(sc, event | BGE_APE_EVENT_STATUS_DRIVER_EVNT |
940 	    BGE_APE_EVENT_STATUS_STATE_CHNGE);
941 }
942 
943 /*
944  * Map a single buffer address.
945  */
946 
947 static void
948 bge_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
949 {
950 	struct bge_dmamap_arg *ctx;
951 
952 	if (error)
953 		return;
954 
955 	KASSERT(nseg == 1, ("%s: %d segments returned!", __func__, nseg));
956 
957 	ctx = arg;
958 	ctx->bge_busaddr = segs->ds_addr;
959 }
960 
961 static uint8_t
962 bge_nvram_getbyte(struct bge_softc *sc, int addr, uint8_t *dest)
963 {
964 	uint32_t access, byte = 0;
965 	int i;
966 
967 	/* Lock. */
968 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_SET1);
969 	for (i = 0; i < 8000; i++) {
970 		if (CSR_READ_4(sc, BGE_NVRAM_SWARB) & BGE_NVRAMSWARB_GNT1)
971 			break;
972 		DELAY(20);
973 	}
974 	if (i == 8000)
975 		return (1);
976 
977 	/* Enable access. */
978 	access = CSR_READ_4(sc, BGE_NVRAM_ACCESS);
979 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access | BGE_NVRAMACC_ENABLE);
980 
981 	CSR_WRITE_4(sc, BGE_NVRAM_ADDR, addr & 0xfffffffc);
982 	CSR_WRITE_4(sc, BGE_NVRAM_CMD, BGE_NVRAM_READCMD);
983 	for (i = 0; i < BGE_TIMEOUT * 10; i++) {
984 		DELAY(10);
985 		if (CSR_READ_4(sc, BGE_NVRAM_CMD) & BGE_NVRAMCMD_DONE) {
986 			DELAY(10);
987 			break;
988 		}
989 	}
990 
991 	if (i == BGE_TIMEOUT * 10) {
992 		if_printf(sc->bge_ifp, "nvram read timed out\n");
993 		return (1);
994 	}
995 
996 	/* Get result. */
997 	byte = CSR_READ_4(sc, BGE_NVRAM_RDDATA);
998 
999 	*dest = (bswap32(byte) >> ((addr % 4) * 8)) & 0xFF;
1000 
1001 	/* Disable access. */
1002 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access);
1003 
1004 	/* Unlock. */
1005 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_CLR1);
1006 	CSR_READ_4(sc, BGE_NVRAM_SWARB);
1007 
1008 	return (0);
1009 }
1010 
1011 /*
1012  * Read a sequence of bytes from NVRAM.
1013  */
1014 static int
1015 bge_read_nvram(struct bge_softc *sc, caddr_t dest, int off, int cnt)
1016 {
1017 	int err = 0, i;
1018 	uint8_t byte = 0;
1019 
1020 	if (sc->bge_asicrev != BGE_ASICREV_BCM5906)
1021 		return (1);
1022 
1023 	for (i = 0; i < cnt; i++) {
1024 		err = bge_nvram_getbyte(sc, off + i, &byte);
1025 		if (err)
1026 			break;
1027 		*(dest + i) = byte;
1028 	}
1029 
1030 	return (err ? 1 : 0);
1031 }
1032 
1033 /*
1034  * Read a byte of data stored in the EEPROM at address 'addr.' The
1035  * BCM570x supports both the traditional bitbang interface and an
1036  * auto access interface for reading the EEPROM. We use the auto
1037  * access method.
1038  */
1039 static uint8_t
1040 bge_eeprom_getbyte(struct bge_softc *sc, int addr, uint8_t *dest)
1041 {
1042 	int i;
1043 	uint32_t byte = 0;
1044 
1045 	/*
1046 	 * Enable use of auto EEPROM access so we can avoid
1047 	 * having to use the bitbang method.
1048 	 */
1049 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_AUTO_EEPROM);
1050 
1051 	/* Reset the EEPROM, load the clock period. */
1052 	CSR_WRITE_4(sc, BGE_EE_ADDR,
1053 	    BGE_EEADDR_RESET | BGE_EEHALFCLK(BGE_HALFCLK_384SCL));
1054 	DELAY(20);
1055 
1056 	/* Issue the read EEPROM command. */
1057 	CSR_WRITE_4(sc, BGE_EE_ADDR, BGE_EE_READCMD | addr);
1058 
1059 	/* Wait for completion */
1060 	for(i = 0; i < BGE_TIMEOUT * 10; i++) {
1061 		DELAY(10);
1062 		if (CSR_READ_4(sc, BGE_EE_ADDR) & BGE_EEADDR_DONE)
1063 			break;
1064 	}
1065 
1066 	if (i == BGE_TIMEOUT * 10) {
1067 		device_printf(sc->bge_dev, "EEPROM read timed out\n");
1068 		return (1);
1069 	}
1070 
1071 	/* Get result. */
1072 	byte = CSR_READ_4(sc, BGE_EE_DATA);
1073 
1074 	*dest = (byte >> ((addr % 4) * 8)) & 0xFF;
1075 
1076 	return (0);
1077 }
1078 
1079 /*
1080  * Read a sequence of bytes from the EEPROM.
1081  */
1082 static int
1083 bge_read_eeprom(struct bge_softc *sc, caddr_t dest, int off, int cnt)
1084 {
1085 	int i, error = 0;
1086 	uint8_t byte = 0;
1087 
1088 	for (i = 0; i < cnt; i++) {
1089 		error = bge_eeprom_getbyte(sc, off + i, &byte);
1090 		if (error)
1091 			break;
1092 		*(dest + i) = byte;
1093 	}
1094 
1095 	return (error ? 1 : 0);
1096 }
1097 
1098 static int
1099 bge_miibus_readreg(device_t dev, int phy, int reg)
1100 {
1101 	struct bge_softc *sc;
1102 	uint32_t val;
1103 	int i;
1104 
1105 	sc = device_get_softc(dev);
1106 
1107 	if (bge_ape_lock(sc, sc->bge_phy_ape_lock) != 0)
1108 		return (0);
1109 
1110 	/* Clear the autopoll bit if set, otherwise may trigger PCI errors. */
1111 	if ((sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) != 0) {
1112 		CSR_WRITE_4(sc, BGE_MI_MODE,
1113 		    sc->bge_mi_mode & ~BGE_MIMODE_AUTOPOLL);
1114 		DELAY(80);
1115 	}
1116 
1117 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_READ | BGE_MICOMM_BUSY |
1118 	    BGE_MIPHY(phy) | BGE_MIREG(reg));
1119 
1120 	/* Poll for the PHY register access to complete. */
1121 	for (i = 0; i < BGE_TIMEOUT; i++) {
1122 		DELAY(10);
1123 		val = CSR_READ_4(sc, BGE_MI_COMM);
1124 		if ((val & BGE_MICOMM_BUSY) == 0) {
1125 			DELAY(5);
1126 			val = CSR_READ_4(sc, BGE_MI_COMM);
1127 			break;
1128 		}
1129 	}
1130 
1131 	if (i == BGE_TIMEOUT) {
1132 		device_printf(sc->bge_dev,
1133 		    "PHY read timed out (phy %d, reg %d, val 0x%08x)\n",
1134 		    phy, reg, val);
1135 		val = 0;
1136 	}
1137 
1138 	/* Restore the autopoll bit if necessary. */
1139 	if ((sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) != 0) {
1140 		CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode);
1141 		DELAY(80);
1142 	}
1143 
1144 	bge_ape_unlock(sc, sc->bge_phy_ape_lock);
1145 
1146 	if (val & BGE_MICOMM_READFAIL)
1147 		return (0);
1148 
1149 	return (val & 0xFFFF);
1150 }
1151 
1152 static int
1153 bge_miibus_writereg(device_t dev, int phy, int reg, int val)
1154 {
1155 	struct bge_softc *sc;
1156 	int i;
1157 
1158 	sc = device_get_softc(dev);
1159 
1160 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
1161 	    (reg == BRGPHY_MII_1000CTL || reg == BRGPHY_MII_AUXCTL))
1162 		return (0);
1163 
1164 	if (bge_ape_lock(sc, sc->bge_phy_ape_lock) != 0)
1165 		return (0);
1166 
1167 	/* Clear the autopoll bit if set, otherwise may trigger PCI errors. */
1168 	if ((sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) != 0) {
1169 		CSR_WRITE_4(sc, BGE_MI_MODE,
1170 		    sc->bge_mi_mode & ~BGE_MIMODE_AUTOPOLL);
1171 		DELAY(80);
1172 	}
1173 
1174 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_WRITE | BGE_MICOMM_BUSY |
1175 	    BGE_MIPHY(phy) | BGE_MIREG(reg) | val);
1176 
1177 	for (i = 0; i < BGE_TIMEOUT; i++) {
1178 		DELAY(10);
1179 		if (!(CSR_READ_4(sc, BGE_MI_COMM) & BGE_MICOMM_BUSY)) {
1180 			DELAY(5);
1181 			CSR_READ_4(sc, BGE_MI_COMM); /* dummy read */
1182 			break;
1183 		}
1184 	}
1185 
1186 	/* Restore the autopoll bit if necessary. */
1187 	if ((sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) != 0) {
1188 		CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode);
1189 		DELAY(80);
1190 	}
1191 
1192 	bge_ape_unlock(sc, sc->bge_phy_ape_lock);
1193 
1194 	if (i == BGE_TIMEOUT)
1195 		device_printf(sc->bge_dev,
1196 		    "PHY write timed out (phy %d, reg %d, val 0x%04x)\n",
1197 		    phy, reg, val);
1198 
1199 	return (0);
1200 }
1201 
1202 static void
1203 bge_miibus_statchg(device_t dev)
1204 {
1205 	struct bge_softc *sc;
1206 	struct mii_data *mii;
1207 	uint32_t mac_mode, rx_mode, tx_mode;
1208 
1209 	sc = device_get_softc(dev);
1210 	if ((if_getdrvflags(sc->bge_ifp) & IFF_DRV_RUNNING) == 0)
1211 		return;
1212 	mii = device_get_softc(sc->bge_miibus);
1213 
1214 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
1215 	    (IFM_ACTIVE | IFM_AVALID)) {
1216 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
1217 		case IFM_10_T:
1218 		case IFM_100_TX:
1219 			sc->bge_link = 1;
1220 			break;
1221 		case IFM_1000_T:
1222 		case IFM_1000_SX:
1223 		case IFM_2500_SX:
1224 			if (sc->bge_asicrev != BGE_ASICREV_BCM5906)
1225 				sc->bge_link = 1;
1226 			else
1227 				sc->bge_link = 0;
1228 			break;
1229 		default:
1230 			sc->bge_link = 0;
1231 			break;
1232 		}
1233 	} else
1234 		sc->bge_link = 0;
1235 	if (sc->bge_link == 0)
1236 		return;
1237 
1238 	/*
1239 	 * APE firmware touches these registers to keep the MAC
1240 	 * connected to the outside world.  Try to keep the
1241 	 * accesses atomic.
1242 	 */
1243 
1244 	/* Set the port mode (MII/GMII) to match the link speed. */
1245 	mac_mode = CSR_READ_4(sc, BGE_MAC_MODE) &
1246 	    ~(BGE_MACMODE_PORTMODE | BGE_MACMODE_HALF_DUPLEX);
1247 	tx_mode = CSR_READ_4(sc, BGE_TX_MODE);
1248 	rx_mode = CSR_READ_4(sc, BGE_RX_MODE);
1249 
1250 	if (IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_T ||
1251 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_SX)
1252 		mac_mode |= BGE_PORTMODE_GMII;
1253 	else
1254 		mac_mode |= BGE_PORTMODE_MII;
1255 
1256 	/* Set MAC flow control behavior to match link flow control settings. */
1257 	tx_mode &= ~BGE_TXMODE_FLOWCTL_ENABLE;
1258 	rx_mode &= ~BGE_RXMODE_FLOWCTL_ENABLE;
1259 	if ((IFM_OPTIONS(mii->mii_media_active) & IFM_FDX) != 0) {
1260 		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_TXPAUSE) != 0)
1261 			tx_mode |= BGE_TXMODE_FLOWCTL_ENABLE;
1262 		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_RXPAUSE) != 0)
1263 			rx_mode |= BGE_RXMODE_FLOWCTL_ENABLE;
1264 	} else
1265 		mac_mode |= BGE_MACMODE_HALF_DUPLEX;
1266 
1267 	CSR_WRITE_4(sc, BGE_MAC_MODE, mac_mode);
1268 	DELAY(40);
1269 	CSR_WRITE_4(sc, BGE_TX_MODE, tx_mode);
1270 	CSR_WRITE_4(sc, BGE_RX_MODE, rx_mode);
1271 }
1272 
1273 /*
1274  * Intialize a standard receive ring descriptor.
1275  */
1276 static int
1277 bge_newbuf_std(struct bge_softc *sc, int i)
1278 {
1279 	struct mbuf *m;
1280 	struct bge_rx_bd *r;
1281 	bus_dma_segment_t segs[1];
1282 	bus_dmamap_t map;
1283 	int error, nsegs;
1284 
1285 	if (sc->bge_flags & BGE_FLAG_JUMBO_STD &&
1286 	    (if_getmtu(sc->bge_ifp) + ETHER_HDR_LEN + ETHER_CRC_LEN +
1287 	    ETHER_VLAN_ENCAP_LEN > (MCLBYTES - ETHER_ALIGN))) {
1288 		m = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUM9BYTES);
1289 		if (m == NULL)
1290 			return (ENOBUFS);
1291 		m->m_len = m->m_pkthdr.len = MJUM9BYTES;
1292 	} else {
1293 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
1294 		if (m == NULL)
1295 			return (ENOBUFS);
1296 		m->m_len = m->m_pkthdr.len = MCLBYTES;
1297 	}
1298 	if ((sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) == 0)
1299 		m_adj(m, ETHER_ALIGN);
1300 
1301 	error = bus_dmamap_load_mbuf_sg(sc->bge_cdata.bge_rx_mtag,
1302 	    sc->bge_cdata.bge_rx_std_sparemap, m, segs, &nsegs, 0);
1303 	if (error != 0) {
1304 		m_freem(m);
1305 		return (error);
1306 	}
1307 	if (sc->bge_cdata.bge_rx_std_chain[i] != NULL) {
1308 		bus_dmamap_sync(sc->bge_cdata.bge_rx_mtag,
1309 		    sc->bge_cdata.bge_rx_std_dmamap[i], BUS_DMASYNC_POSTREAD);
1310 		bus_dmamap_unload(sc->bge_cdata.bge_rx_mtag,
1311 		    sc->bge_cdata.bge_rx_std_dmamap[i]);
1312 	}
1313 	map = sc->bge_cdata.bge_rx_std_dmamap[i];
1314 	sc->bge_cdata.bge_rx_std_dmamap[i] = sc->bge_cdata.bge_rx_std_sparemap;
1315 	sc->bge_cdata.bge_rx_std_sparemap = map;
1316 	sc->bge_cdata.bge_rx_std_chain[i] = m;
1317 	sc->bge_cdata.bge_rx_std_seglen[i] = segs[0].ds_len;
1318 	r = &sc->bge_ldata.bge_rx_std_ring[sc->bge_std];
1319 	r->bge_addr.bge_addr_lo = BGE_ADDR_LO(segs[0].ds_addr);
1320 	r->bge_addr.bge_addr_hi = BGE_ADDR_HI(segs[0].ds_addr);
1321 	r->bge_flags = BGE_RXBDFLAG_END;
1322 	r->bge_len = segs[0].ds_len;
1323 	r->bge_idx = i;
1324 
1325 	bus_dmamap_sync(sc->bge_cdata.bge_rx_mtag,
1326 	    sc->bge_cdata.bge_rx_std_dmamap[i], BUS_DMASYNC_PREREAD);
1327 
1328 	return (0);
1329 }
1330 
1331 /*
1332  * Initialize a jumbo receive ring descriptor. This allocates
1333  * a jumbo buffer from the pool managed internally by the driver.
1334  */
1335 static int
1336 bge_newbuf_jumbo(struct bge_softc *sc, int i)
1337 {
1338 	bus_dma_segment_t segs[BGE_NSEG_JUMBO];
1339 	bus_dmamap_t map;
1340 	struct bge_extrx_bd *r;
1341 	struct mbuf *m;
1342 	int error, nsegs;
1343 
1344 	MGETHDR(m, M_NOWAIT, MT_DATA);
1345 	if (m == NULL)
1346 		return (ENOBUFS);
1347 
1348 	if (m_cljget(m, M_NOWAIT, MJUM9BYTES) == NULL) {
1349 		m_freem(m);
1350 		return (ENOBUFS);
1351 	}
1352 	m->m_len = m->m_pkthdr.len = MJUM9BYTES;
1353 	if ((sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) == 0)
1354 		m_adj(m, ETHER_ALIGN);
1355 
1356 	error = bus_dmamap_load_mbuf_sg(sc->bge_cdata.bge_mtag_jumbo,
1357 	    sc->bge_cdata.bge_rx_jumbo_sparemap, m, segs, &nsegs, 0);
1358 	if (error != 0) {
1359 		m_freem(m);
1360 		return (error);
1361 	}
1362 
1363 	if (sc->bge_cdata.bge_rx_jumbo_chain[i] != NULL) {
1364 		bus_dmamap_sync(sc->bge_cdata.bge_mtag_jumbo,
1365 		    sc->bge_cdata.bge_rx_jumbo_dmamap[i], BUS_DMASYNC_POSTREAD);
1366 		bus_dmamap_unload(sc->bge_cdata.bge_mtag_jumbo,
1367 		    sc->bge_cdata.bge_rx_jumbo_dmamap[i]);
1368 	}
1369 	map = sc->bge_cdata.bge_rx_jumbo_dmamap[i];
1370 	sc->bge_cdata.bge_rx_jumbo_dmamap[i] =
1371 	    sc->bge_cdata.bge_rx_jumbo_sparemap;
1372 	sc->bge_cdata.bge_rx_jumbo_sparemap = map;
1373 	sc->bge_cdata.bge_rx_jumbo_chain[i] = m;
1374 	sc->bge_cdata.bge_rx_jumbo_seglen[i][0] = 0;
1375 	sc->bge_cdata.bge_rx_jumbo_seglen[i][1] = 0;
1376 	sc->bge_cdata.bge_rx_jumbo_seglen[i][2] = 0;
1377 	sc->bge_cdata.bge_rx_jumbo_seglen[i][3] = 0;
1378 
1379 	/*
1380 	 * Fill in the extended RX buffer descriptor.
1381 	 */
1382 	r = &sc->bge_ldata.bge_rx_jumbo_ring[sc->bge_jumbo];
1383 	r->bge_flags = BGE_RXBDFLAG_JUMBO_RING | BGE_RXBDFLAG_END;
1384 	r->bge_idx = i;
1385 	r->bge_len3 = r->bge_len2 = r->bge_len1 = 0;
1386 	switch (nsegs) {
1387 	case 4:
1388 		r->bge_addr3.bge_addr_lo = BGE_ADDR_LO(segs[3].ds_addr);
1389 		r->bge_addr3.bge_addr_hi = BGE_ADDR_HI(segs[3].ds_addr);
1390 		r->bge_len3 = segs[3].ds_len;
1391 		sc->bge_cdata.bge_rx_jumbo_seglen[i][3] = segs[3].ds_len;
1392 	case 3:
1393 		r->bge_addr2.bge_addr_lo = BGE_ADDR_LO(segs[2].ds_addr);
1394 		r->bge_addr2.bge_addr_hi = BGE_ADDR_HI(segs[2].ds_addr);
1395 		r->bge_len2 = segs[2].ds_len;
1396 		sc->bge_cdata.bge_rx_jumbo_seglen[i][2] = segs[2].ds_len;
1397 	case 2:
1398 		r->bge_addr1.bge_addr_lo = BGE_ADDR_LO(segs[1].ds_addr);
1399 		r->bge_addr1.bge_addr_hi = BGE_ADDR_HI(segs[1].ds_addr);
1400 		r->bge_len1 = segs[1].ds_len;
1401 		sc->bge_cdata.bge_rx_jumbo_seglen[i][1] = segs[1].ds_len;
1402 	case 1:
1403 		r->bge_addr0.bge_addr_lo = BGE_ADDR_LO(segs[0].ds_addr);
1404 		r->bge_addr0.bge_addr_hi = BGE_ADDR_HI(segs[0].ds_addr);
1405 		r->bge_len0 = segs[0].ds_len;
1406 		sc->bge_cdata.bge_rx_jumbo_seglen[i][0] = segs[0].ds_len;
1407 		break;
1408 	default:
1409 		panic("%s: %d segments\n", __func__, nsegs);
1410 	}
1411 
1412 	bus_dmamap_sync(sc->bge_cdata.bge_mtag_jumbo,
1413 	    sc->bge_cdata.bge_rx_jumbo_dmamap[i], BUS_DMASYNC_PREREAD);
1414 
1415 	return (0);
1416 }
1417 
1418 static int
1419 bge_init_rx_ring_std(struct bge_softc *sc)
1420 {
1421 	int error, i;
1422 
1423 	bzero(sc->bge_ldata.bge_rx_std_ring, BGE_STD_RX_RING_SZ);
1424 	sc->bge_std = 0;
1425 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1426 		if ((error = bge_newbuf_std(sc, i)) != 0)
1427 			return (error);
1428 		BGE_INC(sc->bge_std, BGE_STD_RX_RING_CNT);
1429 	}
1430 
1431 	bus_dmamap_sync(sc->bge_cdata.bge_rx_std_ring_tag,
1432 	    sc->bge_cdata.bge_rx_std_ring_map, BUS_DMASYNC_PREWRITE);
1433 
1434 	sc->bge_std = 0;
1435 	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, BGE_STD_RX_RING_CNT - 1);
1436 
1437 	return (0);
1438 }
1439 
1440 static void
1441 bge_free_rx_ring_std(struct bge_softc *sc)
1442 {
1443 	int i;
1444 
1445 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1446 		if (sc->bge_cdata.bge_rx_std_chain[i] != NULL) {
1447 			bus_dmamap_sync(sc->bge_cdata.bge_rx_mtag,
1448 			    sc->bge_cdata.bge_rx_std_dmamap[i],
1449 			    BUS_DMASYNC_POSTREAD);
1450 			bus_dmamap_unload(sc->bge_cdata.bge_rx_mtag,
1451 			    sc->bge_cdata.bge_rx_std_dmamap[i]);
1452 			m_freem(sc->bge_cdata.bge_rx_std_chain[i]);
1453 			sc->bge_cdata.bge_rx_std_chain[i] = NULL;
1454 		}
1455 		bzero((char *)&sc->bge_ldata.bge_rx_std_ring[i],
1456 		    sizeof(struct bge_rx_bd));
1457 	}
1458 }
1459 
1460 static int
1461 bge_init_rx_ring_jumbo(struct bge_softc *sc)
1462 {
1463 	struct bge_rcb *rcb;
1464 	int error, i;
1465 
1466 	bzero(sc->bge_ldata.bge_rx_jumbo_ring, BGE_JUMBO_RX_RING_SZ);
1467 	sc->bge_jumbo = 0;
1468 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1469 		if ((error = bge_newbuf_jumbo(sc, i)) != 0)
1470 			return (error);
1471 		BGE_INC(sc->bge_jumbo, BGE_JUMBO_RX_RING_CNT);
1472 	}
1473 
1474 	bus_dmamap_sync(sc->bge_cdata.bge_rx_jumbo_ring_tag,
1475 	    sc->bge_cdata.bge_rx_jumbo_ring_map, BUS_DMASYNC_PREWRITE);
1476 
1477 	sc->bge_jumbo = 0;
1478 
1479 	/* Enable the jumbo receive producer ring. */
1480 	rcb = &sc->bge_ldata.bge_info.bge_jumbo_rx_rcb;
1481 	rcb->bge_maxlen_flags =
1482 	    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_USE_EXT_RX_BD);
1483 	CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1484 
1485 	bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, BGE_JUMBO_RX_RING_CNT - 1);
1486 
1487 	return (0);
1488 }
1489 
1490 static void
1491 bge_free_rx_ring_jumbo(struct bge_softc *sc)
1492 {
1493 	int i;
1494 
1495 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1496 		if (sc->bge_cdata.bge_rx_jumbo_chain[i] != NULL) {
1497 			bus_dmamap_sync(sc->bge_cdata.bge_mtag_jumbo,
1498 			    sc->bge_cdata.bge_rx_jumbo_dmamap[i],
1499 			    BUS_DMASYNC_POSTREAD);
1500 			bus_dmamap_unload(sc->bge_cdata.bge_mtag_jumbo,
1501 			    sc->bge_cdata.bge_rx_jumbo_dmamap[i]);
1502 			m_freem(sc->bge_cdata.bge_rx_jumbo_chain[i]);
1503 			sc->bge_cdata.bge_rx_jumbo_chain[i] = NULL;
1504 		}
1505 		bzero((char *)&sc->bge_ldata.bge_rx_jumbo_ring[i],
1506 		    sizeof(struct bge_extrx_bd));
1507 	}
1508 }
1509 
1510 static void
1511 bge_free_tx_ring(struct bge_softc *sc)
1512 {
1513 	int i;
1514 
1515 	if (sc->bge_ldata.bge_tx_ring == NULL)
1516 		return;
1517 
1518 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
1519 		if (sc->bge_cdata.bge_tx_chain[i] != NULL) {
1520 			bus_dmamap_sync(sc->bge_cdata.bge_tx_mtag,
1521 			    sc->bge_cdata.bge_tx_dmamap[i],
1522 			    BUS_DMASYNC_POSTWRITE);
1523 			bus_dmamap_unload(sc->bge_cdata.bge_tx_mtag,
1524 			    sc->bge_cdata.bge_tx_dmamap[i]);
1525 			m_freem(sc->bge_cdata.bge_tx_chain[i]);
1526 			sc->bge_cdata.bge_tx_chain[i] = NULL;
1527 		}
1528 		bzero((char *)&sc->bge_ldata.bge_tx_ring[i],
1529 		    sizeof(struct bge_tx_bd));
1530 	}
1531 }
1532 
1533 static int
1534 bge_init_tx_ring(struct bge_softc *sc)
1535 {
1536 	sc->bge_txcnt = 0;
1537 	sc->bge_tx_saved_considx = 0;
1538 
1539 	bzero(sc->bge_ldata.bge_tx_ring, BGE_TX_RING_SZ);
1540 	bus_dmamap_sync(sc->bge_cdata.bge_tx_ring_tag,
1541 	    sc->bge_cdata.bge_tx_ring_map, BUS_DMASYNC_PREWRITE);
1542 
1543 	/* Initialize transmit producer index for host-memory send ring. */
1544 	sc->bge_tx_prodidx = 0;
1545 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1546 
1547 	/* 5700 b2 errata */
1548 	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
1549 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1550 
1551 	/* NIC-memory send ring not used; initialize to zero. */
1552 	bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1553 	/* 5700 b2 errata */
1554 	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
1555 		bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1556 
1557 	return (0);
1558 }
1559 
1560 static void
1561 bge_setpromisc(struct bge_softc *sc)
1562 {
1563 	if_t ifp;
1564 
1565 	BGE_LOCK_ASSERT(sc);
1566 
1567 	ifp = sc->bge_ifp;
1568 
1569 	/* Enable or disable promiscuous mode as needed. */
1570 	if (if_getflags(ifp) & IFF_PROMISC)
1571 		BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
1572 	else
1573 		BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
1574 }
1575 
1576 static u_int
1577 bge_hash_maddr(void *arg, struct sockaddr_dl *sdl, u_int cnt)
1578 {
1579 	uint32_t *hashes = arg;
1580 	int h;
1581 
1582 	h = ether_crc32_le(LLADDR(sdl), ETHER_ADDR_LEN) & 0x7F;
1583 	hashes[(h & 0x60) >> 5] |= 1 << (h & 0x1F);
1584 
1585 	return (1);
1586 }
1587 
1588 static void
1589 bge_setmulti(struct bge_softc *sc)
1590 {
1591 	if_t ifp;
1592 	uint32_t hashes[4] = { 0, 0, 0, 0 };
1593 	int i;
1594 
1595 	BGE_LOCK_ASSERT(sc);
1596 
1597 	ifp = sc->bge_ifp;
1598 
1599 	if (if_getflags(ifp) & IFF_ALLMULTI || if_getflags(ifp) & IFF_PROMISC) {
1600 		for (i = 0; i < 4; i++)
1601 			CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0xFFFFFFFF);
1602 		return;
1603 	}
1604 
1605 	/* First, zot all the existing filters. */
1606 	for (i = 0; i < 4; i++)
1607 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0);
1608 
1609 	if_foreach_llmaddr(ifp, bge_hash_maddr, hashes);
1610 
1611 	for (i = 0; i < 4; i++)
1612 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), hashes[i]);
1613 }
1614 
1615 static void
1616 bge_setvlan(struct bge_softc *sc)
1617 {
1618 	if_t ifp;
1619 
1620 	BGE_LOCK_ASSERT(sc);
1621 
1622 	ifp = sc->bge_ifp;
1623 
1624 	/* Enable or disable VLAN tag stripping as needed. */
1625 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING)
1626 		BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_KEEP_VLAN_DIAG);
1627 	else
1628 		BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_KEEP_VLAN_DIAG);
1629 }
1630 
1631 static void
1632 bge_sig_pre_reset(struct bge_softc *sc, int type)
1633 {
1634 
1635 	/*
1636 	 * Some chips don't like this so only do this if ASF is enabled
1637 	 */
1638 	if (sc->bge_asf_mode)
1639 		bge_writemem_ind(sc, BGE_SRAM_FW_MB, BGE_SRAM_FW_MB_MAGIC);
1640 
1641 	if (sc->bge_asf_mode & ASF_NEW_HANDSHAKE) {
1642 		switch (type) {
1643 		case BGE_RESET_START:
1644 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1645 			    BGE_FW_DRV_STATE_START);
1646 			break;
1647 		case BGE_RESET_SHUTDOWN:
1648 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1649 			    BGE_FW_DRV_STATE_UNLOAD);
1650 			break;
1651 		case BGE_RESET_SUSPEND:
1652 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1653 			    BGE_FW_DRV_STATE_SUSPEND);
1654 			break;
1655 		}
1656 	}
1657 
1658 	if (type == BGE_RESET_START || type == BGE_RESET_SUSPEND)
1659 		bge_ape_driver_state_change(sc, type);
1660 }
1661 
1662 static void
1663 bge_sig_post_reset(struct bge_softc *sc, int type)
1664 {
1665 
1666 	if (sc->bge_asf_mode & ASF_NEW_HANDSHAKE) {
1667 		switch (type) {
1668 		case BGE_RESET_START:
1669 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1670 			    BGE_FW_DRV_STATE_START_DONE);
1671 			/* START DONE */
1672 			break;
1673 		case BGE_RESET_SHUTDOWN:
1674 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1675 			    BGE_FW_DRV_STATE_UNLOAD_DONE);
1676 			break;
1677 		}
1678 	}
1679 	if (type == BGE_RESET_SHUTDOWN)
1680 		bge_ape_driver_state_change(sc, type);
1681 }
1682 
1683 static void
1684 bge_sig_legacy(struct bge_softc *sc, int type)
1685 {
1686 
1687 	if (sc->bge_asf_mode) {
1688 		switch (type) {
1689 		case BGE_RESET_START:
1690 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1691 			    BGE_FW_DRV_STATE_START);
1692 			break;
1693 		case BGE_RESET_SHUTDOWN:
1694 			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1695 			    BGE_FW_DRV_STATE_UNLOAD);
1696 			break;
1697 		}
1698 	}
1699 }
1700 
1701 static void
1702 bge_stop_fw(struct bge_softc *sc)
1703 {
1704 	int i;
1705 
1706 	if (sc->bge_asf_mode) {
1707 		bge_writemem_ind(sc, BGE_SRAM_FW_CMD_MB, BGE_FW_CMD_PAUSE);
1708 		CSR_WRITE_4(sc, BGE_RX_CPU_EVENT,
1709 		    CSR_READ_4(sc, BGE_RX_CPU_EVENT) | BGE_RX_CPU_DRV_EVENT);
1710 
1711 		for (i = 0; i < 100; i++ ) {
1712 			if (!(CSR_READ_4(sc, BGE_RX_CPU_EVENT) &
1713 			    BGE_RX_CPU_DRV_EVENT))
1714 				break;
1715 			DELAY(10);
1716 		}
1717 	}
1718 }
1719 
1720 static uint32_t
1721 bge_dma_swap_options(struct bge_softc *sc)
1722 {
1723 	uint32_t dma_options;
1724 
1725 	dma_options = BGE_MODECTL_WORDSWAP_NONFRAME |
1726 	    BGE_MODECTL_BYTESWAP_DATA | BGE_MODECTL_WORDSWAP_DATA;
1727 #if BYTE_ORDER == BIG_ENDIAN
1728 	dma_options |= BGE_MODECTL_BYTESWAP_NONFRAME;
1729 #endif
1730 	return (dma_options);
1731 }
1732 
1733 /*
1734  * Do endian, PCI and DMA initialization.
1735  */
1736 static int
1737 bge_chipinit(struct bge_softc *sc)
1738 {
1739 	uint32_t dma_rw_ctl, misc_ctl, mode_ctl;
1740 	uint16_t val;
1741 	int i;
1742 
1743 	/* Set endianness before we access any non-PCI registers. */
1744 	misc_ctl = BGE_INIT;
1745 	if (sc->bge_flags & BGE_FLAG_TAGGED_STATUS)
1746 		misc_ctl |= BGE_PCIMISCCTL_TAGGED_STATUS;
1747 	pci_write_config(sc->bge_dev, BGE_PCI_MISC_CTL, misc_ctl, 4);
1748 
1749 	/*
1750 	 * Clear the MAC statistics block in the NIC's
1751 	 * internal memory.
1752 	 */
1753 	for (i = BGE_STATS_BLOCK;
1754 	    i < BGE_STATS_BLOCK_END + 1; i += sizeof(uint32_t))
1755 		BGE_MEMWIN_WRITE(sc, i, 0);
1756 
1757 	for (i = BGE_STATUS_BLOCK;
1758 	    i < BGE_STATUS_BLOCK_END + 1; i += sizeof(uint32_t))
1759 		BGE_MEMWIN_WRITE(sc, i, 0);
1760 
1761 	if (sc->bge_chiprev == BGE_CHIPREV_5704_BX) {
1762 		/*
1763 		 *  Fix data corruption caused by non-qword write with WB.
1764 		 *  Fix master abort in PCI mode.
1765 		 *  Fix PCI latency timer.
1766 		 */
1767 		val = pci_read_config(sc->bge_dev, BGE_PCI_MSI_DATA + 2, 2);
1768 		val |= (1 << 10) | (1 << 12) | (1 << 13);
1769 		pci_write_config(sc->bge_dev, BGE_PCI_MSI_DATA + 2, val, 2);
1770 	}
1771 
1772 	if (sc->bge_asicrev == BGE_ASICREV_BCM57765 ||
1773 	    sc->bge_asicrev == BGE_ASICREV_BCM57766) {
1774 		/*
1775 		 * For the 57766 and non Ax versions of 57765, bootcode
1776 		 * needs to setup the PCIE Fast Training Sequence (FTS)
1777 		 * value to prevent transmit hangs.
1778 		 */
1779 		if (sc->bge_chiprev != BGE_CHIPREV_57765_AX) {
1780 			CSR_WRITE_4(sc, BGE_CPMU_PADRNG_CTL,
1781 			    CSR_READ_4(sc, BGE_CPMU_PADRNG_CTL) |
1782 			    BGE_CPMU_PADRNG_CTL_RDIV2);
1783 		}
1784 	}
1785 
1786 	/*
1787 	 * Set up the PCI DMA control register.
1788 	 */
1789 	dma_rw_ctl = BGE_PCIDMARWCTL_RD_CMD_SHIFT(6) |
1790 	    BGE_PCIDMARWCTL_WR_CMD_SHIFT(7);
1791 	if (sc->bge_flags & BGE_FLAG_PCIE) {
1792 		if (sc->bge_mps >= 256)
1793 			dma_rw_ctl |= BGE_PCIDMARWCTL_WR_WAT_SHIFT(7);
1794 		else
1795 			dma_rw_ctl |= BGE_PCIDMARWCTL_WR_WAT_SHIFT(3);
1796 	} else if (sc->bge_flags & BGE_FLAG_PCIX) {
1797 		if (BGE_IS_5714_FAMILY(sc)) {
1798 			/* 256 bytes for read and write. */
1799 			dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(2) |
1800 			    BGE_PCIDMARWCTL_WR_WAT_SHIFT(2);
1801 			dma_rw_ctl |= (sc->bge_asicrev == BGE_ASICREV_BCM5780) ?
1802 			    BGE_PCIDMARWCTL_ONEDMA_ATONCE_GLOBAL :
1803 			    BGE_PCIDMARWCTL_ONEDMA_ATONCE_LOCAL;
1804 		} else if (sc->bge_asicrev == BGE_ASICREV_BCM5703) {
1805 			/*
1806 			 * In the BCM5703, the DMA read watermark should
1807 			 * be set to less than or equal to the maximum
1808 			 * memory read byte count of the PCI-X command
1809 			 * register.
1810 			 */
1811 			dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(4) |
1812 			    BGE_PCIDMARWCTL_WR_WAT_SHIFT(3);
1813 		} else if (sc->bge_asicrev == BGE_ASICREV_BCM5704) {
1814 			/* 1536 bytes for read, 384 bytes for write. */
1815 			dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(7) |
1816 			    BGE_PCIDMARWCTL_WR_WAT_SHIFT(3);
1817 		} else {
1818 			/* 384 bytes for read and write. */
1819 			dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(3) |
1820 			    BGE_PCIDMARWCTL_WR_WAT_SHIFT(3) |
1821 			    0x0F;
1822 		}
1823 		if (sc->bge_asicrev == BGE_ASICREV_BCM5703 ||
1824 		    sc->bge_asicrev == BGE_ASICREV_BCM5704) {
1825 			uint32_t tmp;
1826 
1827 			/* Set ONE_DMA_AT_ONCE for hardware workaround. */
1828 			tmp = CSR_READ_4(sc, BGE_PCI_CLKCTL) & 0x1F;
1829 			if (tmp == 6 || tmp == 7)
1830 				dma_rw_ctl |=
1831 				    BGE_PCIDMARWCTL_ONEDMA_ATONCE_GLOBAL;
1832 
1833 			/* Set PCI-X DMA write workaround. */
1834 			dma_rw_ctl |= BGE_PCIDMARWCTL_ASRT_ALL_BE;
1835 		}
1836 	} else {
1837 		/* Conventional PCI bus: 256 bytes for read and write. */
1838 		dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(7) |
1839 		    BGE_PCIDMARWCTL_WR_WAT_SHIFT(7);
1840 
1841 		if (sc->bge_asicrev != BGE_ASICREV_BCM5705 &&
1842 		    sc->bge_asicrev != BGE_ASICREV_BCM5750)
1843 			dma_rw_ctl |= 0x0F;
1844 	}
1845 	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 ||
1846 	    sc->bge_asicrev == BGE_ASICREV_BCM5701)
1847 		dma_rw_ctl |= BGE_PCIDMARWCTL_USE_MRM |
1848 		    BGE_PCIDMARWCTL_ASRT_ALL_BE;
1849 	if (sc->bge_asicrev == BGE_ASICREV_BCM5703 ||
1850 	    sc->bge_asicrev == BGE_ASICREV_BCM5704)
1851 		dma_rw_ctl &= ~BGE_PCIDMARWCTL_MINDMA;
1852 	if (BGE_IS_5717_PLUS(sc)) {
1853 		dma_rw_ctl &= ~BGE_PCIDMARWCTL_DIS_CACHE_ALIGNMENT;
1854 		if (sc->bge_chipid == BGE_CHIPID_BCM57765_A0)
1855 			dma_rw_ctl &= ~BGE_PCIDMARWCTL_CRDRDR_RDMA_MRRS_MSK;
1856 		/*
1857 		 * Enable HW workaround for controllers that misinterpret
1858 		 * a status tag update and leave interrupts permanently
1859 		 * disabled.
1860 		 */
1861 		if (!BGE_IS_57765_PLUS(sc) &&
1862 		    sc->bge_asicrev != BGE_ASICREV_BCM5717 &&
1863 		    sc->bge_asicrev != BGE_ASICREV_BCM5762)
1864 			dma_rw_ctl |= BGE_PCIDMARWCTL_TAGGED_STATUS_WA;
1865 	}
1866 	pci_write_config(sc->bge_dev, BGE_PCI_DMA_RW_CTL, dma_rw_ctl, 4);
1867 
1868 	/*
1869 	 * Set up general mode register.
1870 	 */
1871 	mode_ctl = bge_dma_swap_options(sc);
1872 	if (sc->bge_asicrev == BGE_ASICREV_BCM5720 ||
1873 	    sc->bge_asicrev == BGE_ASICREV_BCM5762) {
1874 		/* Retain Host-2-BMC settings written by APE firmware. */
1875 		mode_ctl |= CSR_READ_4(sc, BGE_MODE_CTL) &
1876 		    (BGE_MODECTL_BYTESWAP_B2HRX_DATA |
1877 		    BGE_MODECTL_WORDSWAP_B2HRX_DATA |
1878 		    BGE_MODECTL_B2HRX_ENABLE | BGE_MODECTL_HTX2B_ENABLE);
1879 	}
1880 	mode_ctl |= BGE_MODECTL_MAC_ATTN_INTR | BGE_MODECTL_HOST_SEND_BDS |
1881 	    BGE_MODECTL_TX_NO_PHDR_CSUM;
1882 
1883 	/*
1884 	 * BCM5701 B5 have a bug causing data corruption when using
1885 	 * 64-bit DMA reads, which can be terminated early and then
1886 	 * completed later as 32-bit accesses, in combination with
1887 	 * certain bridges.
1888 	 */
1889 	if (sc->bge_asicrev == BGE_ASICREV_BCM5701 &&
1890 	    sc->bge_chipid == BGE_CHIPID_BCM5701_B5)
1891 		mode_ctl |= BGE_MODECTL_FORCE_PCI32;
1892 
1893 	/*
1894 	 * Tell the firmware the driver is running
1895 	 */
1896 	if (sc->bge_asf_mode & ASF_STACKUP)
1897 		mode_ctl |= BGE_MODECTL_STACKUP;
1898 
1899 	CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
1900 
1901 	/*
1902 	 * Disable memory write invalidate.  Apparently it is not supported
1903 	 * properly by these devices.
1904 	 */
1905 	PCI_CLRBIT(sc->bge_dev, BGE_PCI_CMD, PCIM_CMD_MWIEN, 4);
1906 
1907 	/* Set the timer prescaler (always 66 MHz). */
1908 	CSR_WRITE_4(sc, BGE_MISC_CFG, BGE_32BITTIME_66MHZ);
1909 
1910 	/* XXX: The Linux tg3 driver does this at the start of brgphy_reset. */
1911 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
1912 		DELAY(40);	/* XXX */
1913 
1914 		/* Put PHY into ready state */
1915 		BGE_CLRBIT(sc, BGE_MISC_CFG, BGE_MISCCFG_EPHY_IDDQ);
1916 		CSR_READ_4(sc, BGE_MISC_CFG); /* Flush */
1917 		DELAY(40);
1918 	}
1919 
1920 	return (0);
1921 }
1922 
1923 static int
1924 bge_blockinit(struct bge_softc *sc)
1925 {
1926 	struct bge_rcb *rcb;
1927 	bus_size_t vrcb;
1928 	bge_hostaddr taddr;
1929 	uint32_t dmactl, rdmareg, val;
1930 	int i, limit;
1931 
1932 	/*
1933 	 * Initialize the memory window pointer register so that
1934 	 * we can access the first 32K of internal NIC RAM. This will
1935 	 * allow us to set up the TX send ring RCBs and the RX return
1936 	 * ring RCBs, plus other things which live in NIC memory.
1937 	 */
1938 	CSR_WRITE_4(sc, BGE_PCI_MEMWIN_BASEADDR, 0);
1939 
1940 	/* Note: the BCM5704 has a smaller mbuf space than other chips. */
1941 
1942 	if (!(BGE_IS_5705_PLUS(sc))) {
1943 		/* Configure mbuf memory pool */
1944 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_BASEADDR, BGE_BUFFPOOL_1);
1945 		if (sc->bge_asicrev == BGE_ASICREV_BCM5704)
1946 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x10000);
1947 		else
1948 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x18000);
1949 
1950 		/* Configure DMA resource pool */
1951 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_BASEADDR,
1952 		    BGE_DMA_DESCRIPTORS);
1953 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LEN, 0x2000);
1954 	}
1955 
1956 	/* Configure mbuf pool watermarks */
1957 	if (BGE_IS_5717_PLUS(sc)) {
1958 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1959 		if (if_getmtu(sc->bge_ifp) > ETHERMTU) {
1960 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x7e);
1961 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0xea);
1962 		} else {
1963 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x2a);
1964 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0xa0);
1965 		}
1966 	} else if (!BGE_IS_5705_PLUS(sc)) {
1967 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x50);
1968 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x20);
1969 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1970 	} else if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
1971 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1972 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x04);
1973 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x10);
1974 	} else {
1975 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1976 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x10);
1977 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1978 	}
1979 
1980 	/* Configure DMA resource watermarks */
1981 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LOWAT, 5);
1982 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_HIWAT, 10);
1983 
1984 	/* Enable buffer manager */
1985 	val = BGE_BMANMODE_ENABLE | BGE_BMANMODE_LOMBUF_ATTN;
1986 	/*
1987 	 * Change the arbitration algorithm of TXMBUF read request to
1988 	 * round-robin instead of priority based for BCM5719.  When
1989 	 * TXFIFO is almost empty, RDMA will hold its request until
1990 	 * TXFIFO is not almost empty.
1991 	 */
1992 	if (sc->bge_asicrev == BGE_ASICREV_BCM5719)
1993 		val |= BGE_BMANMODE_NO_TX_UNDERRUN;
1994 	CSR_WRITE_4(sc, BGE_BMAN_MODE, val);
1995 
1996 	/* Poll for buffer manager start indication */
1997 	for (i = 0; i < BGE_TIMEOUT; i++) {
1998 		DELAY(10);
1999 		if (CSR_READ_4(sc, BGE_BMAN_MODE) & BGE_BMANMODE_ENABLE)
2000 			break;
2001 	}
2002 
2003 	if (i == BGE_TIMEOUT) {
2004 		device_printf(sc->bge_dev, "buffer manager failed to start\n");
2005 		return (ENXIO);
2006 	}
2007 
2008 	/* Enable flow-through queues */
2009 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
2010 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
2011 
2012 	/* Wait until queue initialization is complete */
2013 	for (i = 0; i < BGE_TIMEOUT; i++) {
2014 		DELAY(10);
2015 		if (CSR_READ_4(sc, BGE_FTQ_RESET) == 0)
2016 			break;
2017 	}
2018 
2019 	if (i == BGE_TIMEOUT) {
2020 		device_printf(sc->bge_dev, "flow-through queue init failed\n");
2021 		return (ENXIO);
2022 	}
2023 
2024 	/*
2025 	 * Summary of rings supported by the controller:
2026 	 *
2027 	 * Standard Receive Producer Ring
2028 	 * - This ring is used to feed receive buffers for "standard"
2029 	 *   sized frames (typically 1536 bytes) to the controller.
2030 	 *
2031 	 * Jumbo Receive Producer Ring
2032 	 * - This ring is used to feed receive buffers for jumbo sized
2033 	 *   frames (i.e. anything bigger than the "standard" frames)
2034 	 *   to the controller.
2035 	 *
2036 	 * Mini Receive Producer Ring
2037 	 * - This ring is used to feed receive buffers for "mini"
2038 	 *   sized frames to the controller.
2039 	 * - This feature required external memory for the controller
2040 	 *   but was never used in a production system.  Should always
2041 	 *   be disabled.
2042 	 *
2043 	 * Receive Return Ring
2044 	 * - After the controller has placed an incoming frame into a
2045 	 *   receive buffer that buffer is moved into a receive return
2046 	 *   ring.  The driver is then responsible to passing the
2047 	 *   buffer up to the stack.  Many versions of the controller
2048 	 *   support multiple RR rings.
2049 	 *
2050 	 * Send Ring
2051 	 * - This ring is used for outgoing frames.  Many versions of
2052 	 *   the controller support multiple send rings.
2053 	 */
2054 
2055 	/* Initialize the standard receive producer ring control block. */
2056 	rcb = &sc->bge_ldata.bge_info.bge_std_rx_rcb;
2057 	rcb->bge_hostaddr.bge_addr_lo =
2058 	    BGE_ADDR_LO(sc->bge_ldata.bge_rx_std_ring_paddr);
2059 	rcb->bge_hostaddr.bge_addr_hi =
2060 	    BGE_ADDR_HI(sc->bge_ldata.bge_rx_std_ring_paddr);
2061 	bus_dmamap_sync(sc->bge_cdata.bge_rx_std_ring_tag,
2062 	    sc->bge_cdata.bge_rx_std_ring_map, BUS_DMASYNC_PREREAD);
2063 	if (BGE_IS_5717_PLUS(sc)) {
2064 		/*
2065 		 * Bits 31-16: Programmable ring size (2048, 1024, 512, .., 32)
2066 		 * Bits 15-2 : Maximum RX frame size
2067 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring ENabled
2068 		 * Bit 0     : Reserved
2069 		 */
2070 		rcb->bge_maxlen_flags =
2071 		    BGE_RCB_MAXLEN_FLAGS(512, BGE_MAX_FRAMELEN << 2);
2072 	} else if (BGE_IS_5705_PLUS(sc)) {
2073 		/*
2074 		 * Bits 31-16: Programmable ring size (512, 256, 128, 64, 32)
2075 		 * Bits 15-2 : Reserved (should be 0)
2076 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
2077 		 * Bit 0     : Reserved
2078 		 */
2079 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(512, 0);
2080 	} else {
2081 		/*
2082 		 * Ring size is always XXX entries
2083 		 * Bits 31-16: Maximum RX frame size
2084 		 * Bits 15-2 : Reserved (should be 0)
2085 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
2086 		 * Bit 0     : Reserved
2087 		 */
2088 		rcb->bge_maxlen_flags =
2089 		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN, 0);
2090 	}
2091 	if (sc->bge_asicrev == BGE_ASICREV_BCM5717 ||
2092 	    sc->bge_asicrev == BGE_ASICREV_BCM5719 ||
2093 	    sc->bge_asicrev == BGE_ASICREV_BCM5720)
2094 		rcb->bge_nicaddr = BGE_STD_RX_RINGS_5717;
2095 	else
2096 		rcb->bge_nicaddr = BGE_STD_RX_RINGS;
2097 	/* Write the standard receive producer ring control block. */
2098 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_HI, rcb->bge_hostaddr.bge_addr_hi);
2099 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_LO, rcb->bge_hostaddr.bge_addr_lo);
2100 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
2101 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_NICADDR, rcb->bge_nicaddr);
2102 
2103 	/* Reset the standard receive producer ring producer index. */
2104 	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, 0);
2105 
2106 	/*
2107 	 * Initialize the jumbo RX producer ring control
2108 	 * block.  We set the 'ring disabled' bit in the
2109 	 * flags field until we're actually ready to start
2110 	 * using this ring (i.e. once we set the MTU
2111 	 * high enough to require it).
2112 	 */
2113 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
2114 		rcb = &sc->bge_ldata.bge_info.bge_jumbo_rx_rcb;
2115 		/* Get the jumbo receive producer ring RCB parameters. */
2116 		rcb->bge_hostaddr.bge_addr_lo =
2117 		    BGE_ADDR_LO(sc->bge_ldata.bge_rx_jumbo_ring_paddr);
2118 		rcb->bge_hostaddr.bge_addr_hi =
2119 		    BGE_ADDR_HI(sc->bge_ldata.bge_rx_jumbo_ring_paddr);
2120 		bus_dmamap_sync(sc->bge_cdata.bge_rx_jumbo_ring_tag,
2121 		    sc->bge_cdata.bge_rx_jumbo_ring_map,
2122 		    BUS_DMASYNC_PREREAD);
2123 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(0,
2124 		    BGE_RCB_FLAG_USE_EXT_RX_BD | BGE_RCB_FLAG_RING_DISABLED);
2125 		if (sc->bge_asicrev == BGE_ASICREV_BCM5717 ||
2126 		    sc->bge_asicrev == BGE_ASICREV_BCM5719 ||
2127 		    sc->bge_asicrev == BGE_ASICREV_BCM5720)
2128 			rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS_5717;
2129 		else
2130 			rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS;
2131 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_HI,
2132 		    rcb->bge_hostaddr.bge_addr_hi);
2133 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_LO,
2134 		    rcb->bge_hostaddr.bge_addr_lo);
2135 		/* Program the jumbo receive producer ring RCB parameters. */
2136 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS,
2137 		    rcb->bge_maxlen_flags);
2138 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_NICADDR, rcb->bge_nicaddr);
2139 		/* Reset the jumbo receive producer ring producer index. */
2140 		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, 0);
2141 	}
2142 
2143 	/* Disable the mini receive producer ring RCB. */
2144 	if (BGE_IS_5700_FAMILY(sc)) {
2145 		rcb = &sc->bge_ldata.bge_info.bge_mini_rx_rcb;
2146 		rcb->bge_maxlen_flags =
2147 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED);
2148 		CSR_WRITE_4(sc, BGE_RX_MINI_RCB_MAXLEN_FLAGS,
2149 		    rcb->bge_maxlen_flags);
2150 		/* Reset the mini receive producer ring producer index. */
2151 		bge_writembx(sc, BGE_MBX_RX_MINI_PROD_LO, 0);
2152 	}
2153 
2154 	/* Choose de-pipeline mode for BCM5906 A0, A1 and A2. */
2155 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
2156 		if (sc->bge_chipid == BGE_CHIPID_BCM5906_A0 ||
2157 		    sc->bge_chipid == BGE_CHIPID_BCM5906_A1 ||
2158 		    sc->bge_chipid == BGE_CHIPID_BCM5906_A2)
2159 			CSR_WRITE_4(sc, BGE_ISO_PKT_TX,
2160 			    (CSR_READ_4(sc, BGE_ISO_PKT_TX) & ~3) | 2);
2161 	}
2162 	/*
2163 	 * The BD ring replenish thresholds control how often the
2164 	 * hardware fetches new BD's from the producer rings in host
2165 	 * memory.  Setting the value too low on a busy system can
2166 	 * starve the hardware and recue the throughpout.
2167 	 *
2168 	 * Set the BD ring replentish thresholds. The recommended
2169 	 * values are 1/8th the number of descriptors allocated to
2170 	 * each ring.
2171 	 * XXX The 5754 requires a lower threshold, so it might be a
2172 	 * requirement of all 575x family chips.  The Linux driver sets
2173 	 * the lower threshold for all 5705 family chips as well, but there
2174 	 * are reports that it might not need to be so strict.
2175 	 *
2176 	 * XXX Linux does some extra fiddling here for the 5906 parts as
2177 	 * well.
2178 	 */
2179 	if (BGE_IS_5705_PLUS(sc))
2180 		val = 8;
2181 	else
2182 		val = BGE_STD_RX_RING_CNT / 8;
2183 	CSR_WRITE_4(sc, BGE_RBDI_STD_REPL_THRESH, val);
2184 	if (BGE_IS_JUMBO_CAPABLE(sc))
2185 		CSR_WRITE_4(sc, BGE_RBDI_JUMBO_REPL_THRESH,
2186 		    BGE_JUMBO_RX_RING_CNT/8);
2187 	if (BGE_IS_5717_PLUS(sc)) {
2188 		CSR_WRITE_4(sc, BGE_STD_REPLENISH_LWM, 32);
2189 		CSR_WRITE_4(sc, BGE_JMB_REPLENISH_LWM, 16);
2190 	}
2191 
2192 	/*
2193 	 * Disable all send rings by setting the 'ring disabled' bit
2194 	 * in the flags field of all the TX send ring control blocks,
2195 	 * located in NIC memory.
2196 	 */
2197 	if (!BGE_IS_5705_PLUS(sc))
2198 		/* 5700 to 5704 had 16 send rings. */
2199 		limit = BGE_TX_RINGS_EXTSSRAM_MAX;
2200 	else if (BGE_IS_57765_PLUS(sc) ||
2201 	    sc->bge_asicrev == BGE_ASICREV_BCM5762)
2202 		limit = 2;
2203 	else if (BGE_IS_5717_PLUS(sc))
2204 		limit = 4;
2205 	else
2206 		limit = 1;
2207 	vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
2208 	for (i = 0; i < limit; i++) {
2209 		RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
2210 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED));
2211 		RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
2212 		vrcb += sizeof(struct bge_rcb);
2213 	}
2214 
2215 	/* Configure send ring RCB 0 (we use only the first ring) */
2216 	vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
2217 	BGE_HOSTADDR(taddr, sc->bge_ldata.bge_tx_ring_paddr);
2218 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
2219 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
2220 	if (sc->bge_asicrev == BGE_ASICREV_BCM5717 ||
2221 	    sc->bge_asicrev == BGE_ASICREV_BCM5719 ||
2222 	    sc->bge_asicrev == BGE_ASICREV_BCM5720)
2223 		RCB_WRITE_4(sc, vrcb, bge_nicaddr, BGE_SEND_RING_5717);
2224 	else
2225 		RCB_WRITE_4(sc, vrcb, bge_nicaddr,
2226 		    BGE_NIC_TXRING_ADDR(0, BGE_TX_RING_CNT));
2227 	RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
2228 	    BGE_RCB_MAXLEN_FLAGS(BGE_TX_RING_CNT, 0));
2229 
2230 	/*
2231 	 * Disable all receive return rings by setting the
2232 	 * 'ring diabled' bit in the flags field of all the receive
2233 	 * return ring control blocks, located in NIC memory.
2234 	 */
2235 	if (sc->bge_asicrev == BGE_ASICREV_BCM5717 ||
2236 	    sc->bge_asicrev == BGE_ASICREV_BCM5719 ||
2237 	    sc->bge_asicrev == BGE_ASICREV_BCM5720) {
2238 		/* Should be 17, use 16 until we get an SRAM map. */
2239 		limit = 16;
2240 	} else if (!BGE_IS_5705_PLUS(sc))
2241 		limit = BGE_RX_RINGS_MAX;
2242 	else if (sc->bge_asicrev == BGE_ASICREV_BCM5755 ||
2243 	    sc->bge_asicrev == BGE_ASICREV_BCM5762 ||
2244 	    BGE_IS_57765_PLUS(sc))
2245 		limit = 4;
2246 	else
2247 		limit = 1;
2248 	/* Disable all receive return rings. */
2249 	vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
2250 	for (i = 0; i < limit; i++) {
2251 		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, 0);
2252 		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, 0);
2253 		RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
2254 		    BGE_RCB_FLAG_RING_DISABLED);
2255 		RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
2256 		bge_writembx(sc, BGE_MBX_RX_CONS0_LO +
2257 		    (i * (sizeof(uint64_t))), 0);
2258 		vrcb += sizeof(struct bge_rcb);
2259 	}
2260 
2261 	/*
2262 	 * Set up receive return ring 0.  Note that the NIC address
2263 	 * for RX return rings is 0x0.  The return rings live entirely
2264 	 * within the host, so the nicaddr field in the RCB isn't used.
2265 	 */
2266 	vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
2267 	BGE_HOSTADDR(taddr, sc->bge_ldata.bge_rx_return_ring_paddr);
2268 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
2269 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
2270 	RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
2271 	RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
2272 	    BGE_RCB_MAXLEN_FLAGS(sc->bge_return_ring_cnt, 0));
2273 
2274 	/* Set random backoff seed for TX */
2275 	CSR_WRITE_4(sc, BGE_TX_RANDOM_BACKOFF,
2276 	    (IF_LLADDR(sc->bge_ifp)[0] + IF_LLADDR(sc->bge_ifp)[1] +
2277 	    IF_LLADDR(sc->bge_ifp)[2] + IF_LLADDR(sc->bge_ifp)[3] +
2278 	    IF_LLADDR(sc->bge_ifp)[4] + IF_LLADDR(sc->bge_ifp)[5]) &
2279 	    BGE_TX_BACKOFF_SEED_MASK);
2280 
2281 	/* Set inter-packet gap */
2282 	val = 0x2620;
2283 	if (sc->bge_asicrev == BGE_ASICREV_BCM5720 ||
2284 	    sc->bge_asicrev == BGE_ASICREV_BCM5762)
2285 		val |= CSR_READ_4(sc, BGE_TX_LENGTHS) &
2286 		    (BGE_TXLEN_JMB_FRM_LEN_MSK | BGE_TXLEN_CNT_DN_VAL_MSK);
2287 	CSR_WRITE_4(sc, BGE_TX_LENGTHS, val);
2288 
2289 	/*
2290 	 * Specify which ring to use for packets that don't match
2291 	 * any RX rules.
2292 	 */
2293 	CSR_WRITE_4(sc, BGE_RX_RULES_CFG, 0x08);
2294 
2295 	/*
2296 	 * Configure number of RX lists. One interrupt distribution
2297 	 * list, sixteen active lists, one bad frames class.
2298 	 */
2299 	CSR_WRITE_4(sc, BGE_RXLP_CFG, 0x181);
2300 
2301 	/* Inialize RX list placement stats mask. */
2302 	CSR_WRITE_4(sc, BGE_RXLP_STATS_ENABLE_MASK, 0x007FFFFF);
2303 	CSR_WRITE_4(sc, BGE_RXLP_STATS_CTL, 0x1);
2304 
2305 	/* Disable host coalescing until we get it set up */
2306 	CSR_WRITE_4(sc, BGE_HCC_MODE, 0x00000000);
2307 
2308 	/* Poll to make sure it's shut down. */
2309 	for (i = 0; i < BGE_TIMEOUT; i++) {
2310 		DELAY(10);
2311 		if (!(CSR_READ_4(sc, BGE_HCC_MODE) & BGE_HCCMODE_ENABLE))
2312 			break;
2313 	}
2314 
2315 	if (i == BGE_TIMEOUT) {
2316 		device_printf(sc->bge_dev,
2317 		    "host coalescing engine failed to idle\n");
2318 		return (ENXIO);
2319 	}
2320 
2321 	/* Set up host coalescing defaults */
2322 	CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, sc->bge_rx_coal_ticks);
2323 	CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS, sc->bge_tx_coal_ticks);
2324 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, sc->bge_rx_max_coal_bds);
2325 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS, sc->bge_tx_max_coal_bds);
2326 	if (!(BGE_IS_5705_PLUS(sc))) {
2327 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS_INT, 0);
2328 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS_INT, 0);
2329 	}
2330 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT, 1);
2331 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT, 1);
2332 
2333 	/* Set up address of statistics block */
2334 	if (!(BGE_IS_5705_PLUS(sc))) {
2335 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_HI,
2336 		    BGE_ADDR_HI(sc->bge_ldata.bge_stats_paddr));
2337 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_LO,
2338 		    BGE_ADDR_LO(sc->bge_ldata.bge_stats_paddr));
2339 		CSR_WRITE_4(sc, BGE_HCC_STATS_BASEADDR, BGE_STATS_BLOCK);
2340 		CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_BASEADDR, BGE_STATUS_BLOCK);
2341 		CSR_WRITE_4(sc, BGE_HCC_STATS_TICKS, sc->bge_stat_ticks);
2342 	}
2343 
2344 	/* Set up address of status block */
2345 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_HI,
2346 	    BGE_ADDR_HI(sc->bge_ldata.bge_status_block_paddr));
2347 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_LO,
2348 	    BGE_ADDR_LO(sc->bge_ldata.bge_status_block_paddr));
2349 
2350 	/* Set up status block size. */
2351 	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
2352 	    sc->bge_chipid != BGE_CHIPID_BCM5700_C0) {
2353 		val = BGE_STATBLKSZ_FULL;
2354 		bzero(sc->bge_ldata.bge_status_block, BGE_STATUS_BLK_SZ);
2355 	} else {
2356 		val = BGE_STATBLKSZ_32BYTE;
2357 		bzero(sc->bge_ldata.bge_status_block, 32);
2358 	}
2359 	bus_dmamap_sync(sc->bge_cdata.bge_status_tag,
2360 	    sc->bge_cdata.bge_status_map,
2361 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2362 
2363 	/* Turn on host coalescing state machine */
2364 	CSR_WRITE_4(sc, BGE_HCC_MODE, val | BGE_HCCMODE_ENABLE);
2365 
2366 	/* Turn on RX BD completion state machine and enable attentions */
2367 	CSR_WRITE_4(sc, BGE_RBDC_MODE,
2368 	    BGE_RBDCMODE_ENABLE | BGE_RBDCMODE_ATTN);
2369 
2370 	/* Turn on RX list placement state machine */
2371 	CSR_WRITE_4(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
2372 
2373 	/* Turn on RX list selector state machine. */
2374 	if (!(BGE_IS_5705_PLUS(sc)))
2375 		CSR_WRITE_4(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
2376 
2377 	/* Turn on DMA, clear stats. */
2378 	val = BGE_MACMODE_TXDMA_ENB | BGE_MACMODE_RXDMA_ENB |
2379 	    BGE_MACMODE_RX_STATS_CLEAR | BGE_MACMODE_TX_STATS_CLEAR |
2380 	    BGE_MACMODE_RX_STATS_ENB | BGE_MACMODE_TX_STATS_ENB |
2381 	    BGE_MACMODE_FRMHDR_DMA_ENB;
2382 
2383 	if (sc->bge_flags & BGE_FLAG_TBI)
2384 		val |= BGE_PORTMODE_TBI;
2385 	else if (sc->bge_flags & BGE_FLAG_MII_SERDES)
2386 		val |= BGE_PORTMODE_GMII;
2387 	else
2388 		val |= BGE_PORTMODE_MII;
2389 
2390 	/* Allow APE to send/receive frames. */
2391 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) != 0)
2392 		val |= BGE_MACMODE_APE_RX_EN | BGE_MACMODE_APE_TX_EN;
2393 
2394 	CSR_WRITE_4(sc, BGE_MAC_MODE, val);
2395 	DELAY(40);
2396 
2397 	/* Set misc. local control, enable interrupts on attentions */
2398 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_ONATTN);
2399 
2400 #ifdef notdef
2401 	/* Assert GPIO pins for PHY reset */
2402 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUT0 |
2403 	    BGE_MLC_MISCIO_OUT1 | BGE_MLC_MISCIO_OUT2);
2404 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUTEN0 |
2405 	    BGE_MLC_MISCIO_OUTEN1 | BGE_MLC_MISCIO_OUTEN2);
2406 #endif
2407 
2408 	/* Turn on DMA completion state machine */
2409 	if (!(BGE_IS_5705_PLUS(sc)))
2410 		CSR_WRITE_4(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
2411 
2412 	val = BGE_WDMAMODE_ENABLE | BGE_WDMAMODE_ALL_ATTNS;
2413 
2414 	/* Enable host coalescing bug fix. */
2415 	if (BGE_IS_5755_PLUS(sc))
2416 		val |= BGE_WDMAMODE_STATUS_TAG_FIX;
2417 
2418 	/* Request larger DMA burst size to get better performance. */
2419 	if (sc->bge_asicrev == BGE_ASICREV_BCM5785)
2420 		val |= BGE_WDMAMODE_BURST_ALL_DATA;
2421 
2422 	/* Turn on write DMA state machine */
2423 	CSR_WRITE_4(sc, BGE_WDMA_MODE, val);
2424 	DELAY(40);
2425 
2426 	/* Turn on read DMA state machine */
2427 	val = BGE_RDMAMODE_ENABLE | BGE_RDMAMODE_ALL_ATTNS;
2428 
2429 	if (sc->bge_asicrev == BGE_ASICREV_BCM5717)
2430 		val |= BGE_RDMAMODE_MULT_DMA_RD_DIS;
2431 
2432 	if (sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
2433 	    sc->bge_asicrev == BGE_ASICREV_BCM5785 ||
2434 	    sc->bge_asicrev == BGE_ASICREV_BCM57780)
2435 		val |= BGE_RDMAMODE_BD_SBD_CRPT_ATTN |
2436 		    BGE_RDMAMODE_MBUF_RBD_CRPT_ATTN |
2437 		    BGE_RDMAMODE_MBUF_SBD_CRPT_ATTN;
2438 	if (sc->bge_flags & BGE_FLAG_PCIE)
2439 		val |= BGE_RDMAMODE_FIFO_LONG_BURST;
2440 	if (sc->bge_flags & (BGE_FLAG_TSO | BGE_FLAG_TSO3)) {
2441 		val |= BGE_RDMAMODE_TSO4_ENABLE;
2442 		if (sc->bge_flags & BGE_FLAG_TSO3 ||
2443 		    sc->bge_asicrev == BGE_ASICREV_BCM5785 ||
2444 		    sc->bge_asicrev == BGE_ASICREV_BCM57780)
2445 			val |= BGE_RDMAMODE_TSO6_ENABLE;
2446 	}
2447 
2448 	if (sc->bge_asicrev == BGE_ASICREV_BCM5720 ||
2449 	    sc->bge_asicrev == BGE_ASICREV_BCM5762) {
2450 		val |= CSR_READ_4(sc, BGE_RDMA_MODE) &
2451 			BGE_RDMAMODE_H2BNC_VLAN_DET;
2452 		/*
2453 		 * Allow multiple outstanding read requests from
2454 		 * non-LSO read DMA engine.
2455 		 */
2456 		val &= ~BGE_RDMAMODE_MULT_DMA_RD_DIS;
2457 	}
2458 
2459 	if (sc->bge_asicrev == BGE_ASICREV_BCM5761 ||
2460 	    sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
2461 	    sc->bge_asicrev == BGE_ASICREV_BCM5785 ||
2462 	    sc->bge_asicrev == BGE_ASICREV_BCM57780 ||
2463 	    BGE_IS_5717_PLUS(sc) || BGE_IS_57765_PLUS(sc)) {
2464 		if (sc->bge_asicrev == BGE_ASICREV_BCM5762)
2465 			rdmareg = BGE_RDMA_RSRVCTRL_REG2;
2466 		else
2467 			rdmareg = BGE_RDMA_RSRVCTRL;
2468 		dmactl = CSR_READ_4(sc, rdmareg);
2469 		/*
2470 		 * Adjust tx margin to prevent TX data corruption and
2471 		 * fix internal FIFO overflow.
2472 		 */
2473 		if (sc->bge_chipid == BGE_CHIPID_BCM5719_A0 ||
2474 		    sc->bge_asicrev == BGE_ASICREV_BCM5762) {
2475 			dmactl &= ~(BGE_RDMA_RSRVCTRL_FIFO_LWM_MASK |
2476 			    BGE_RDMA_RSRVCTRL_FIFO_HWM_MASK |
2477 			    BGE_RDMA_RSRVCTRL_TXMRGN_MASK);
2478 			dmactl |= BGE_RDMA_RSRVCTRL_FIFO_LWM_1_5K |
2479 			    BGE_RDMA_RSRVCTRL_FIFO_HWM_1_5K |
2480 			    BGE_RDMA_RSRVCTRL_TXMRGN_320B;
2481 		}
2482 		/*
2483 		 * Enable fix for read DMA FIFO overruns.
2484 		 * The fix is to limit the number of RX BDs
2485 		 * the hardware would fetch at a fime.
2486 		 */
2487 		CSR_WRITE_4(sc, rdmareg, dmactl |
2488 		    BGE_RDMA_RSRVCTRL_FIFO_OFLW_FIX);
2489 	}
2490 
2491 	if (sc->bge_asicrev == BGE_ASICREV_BCM5719) {
2492 		CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL,
2493 		    CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL) |
2494 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_4K |
2495 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
2496 	} else if (sc->bge_asicrev == BGE_ASICREV_BCM5720) {
2497 		/*
2498 		 * Allow 4KB burst length reads for non-LSO frames.
2499 		 * Enable 512B burst length reads for buffer descriptors.
2500 		 */
2501 		CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL,
2502 		    CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL) |
2503 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_512 |
2504 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
2505 	} else if (sc->bge_asicrev == BGE_ASICREV_BCM5762) {
2506 		CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL_REG2,
2507 		    CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL_REG2) |
2508 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_4K |
2509 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
2510 	}
2511 
2512 	CSR_WRITE_4(sc, BGE_RDMA_MODE, val);
2513 	DELAY(40);
2514 
2515 	if (sc->bge_flags & BGE_FLAG_RDMA_BUG) {
2516 		for (i = 0; i < BGE_NUM_RDMA_CHANNELS / 2; i++) {
2517 			val = CSR_READ_4(sc, BGE_RDMA_LENGTH + i * 4);
2518 			if ((val & 0xFFFF) > BGE_FRAMELEN)
2519 				break;
2520 			if (((val >> 16) & 0xFFFF) > BGE_FRAMELEN)
2521 				break;
2522 		}
2523 		if (i != BGE_NUM_RDMA_CHANNELS / 2) {
2524 			val = CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL);
2525 			if (sc->bge_asicrev == BGE_ASICREV_BCM5719)
2526 				val |= BGE_RDMA_TX_LENGTH_WA_5719;
2527 			else
2528 				val |= BGE_RDMA_TX_LENGTH_WA_5720;
2529 			CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL, val);
2530 		}
2531 	}
2532 
2533 	/* Turn on RX data completion state machine */
2534 	CSR_WRITE_4(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
2535 
2536 	/* Turn on RX BD initiator state machine */
2537 	CSR_WRITE_4(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
2538 
2539 	/* Turn on RX data and RX BD initiator state machine */
2540 	CSR_WRITE_4(sc, BGE_RDBDI_MODE, BGE_RDBDIMODE_ENABLE);
2541 
2542 	/* Turn on Mbuf cluster free state machine */
2543 	if (!(BGE_IS_5705_PLUS(sc)))
2544 		CSR_WRITE_4(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
2545 
2546 	/* Turn on send BD completion state machine */
2547 	CSR_WRITE_4(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
2548 
2549 	/* Turn on send data completion state machine */
2550 	val = BGE_SDCMODE_ENABLE;
2551 	if (sc->bge_asicrev == BGE_ASICREV_BCM5761)
2552 		val |= BGE_SDCMODE_CDELAY;
2553 	CSR_WRITE_4(sc, BGE_SDC_MODE, val);
2554 
2555 	/* Turn on send data initiator state machine */
2556 	if (sc->bge_flags & (BGE_FLAG_TSO | BGE_FLAG_TSO3))
2557 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE |
2558 		    BGE_SDIMODE_HW_LSO_PRE_DMA);
2559 	else
2560 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
2561 
2562 	/* Turn on send BD initiator state machine */
2563 	CSR_WRITE_4(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
2564 
2565 	/* Turn on send BD selector state machine */
2566 	CSR_WRITE_4(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
2567 
2568 	CSR_WRITE_4(sc, BGE_SDI_STATS_ENABLE_MASK, 0x007FFFFF);
2569 	CSR_WRITE_4(sc, BGE_SDI_STATS_CTL,
2570 	    BGE_SDISTATSCTL_ENABLE | BGE_SDISTATSCTL_FASTER);
2571 
2572 	/* ack/clear link change events */
2573 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
2574 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
2575 	    BGE_MACSTAT_LINK_CHANGED);
2576 	CSR_WRITE_4(sc, BGE_MI_STS, 0);
2577 
2578 	/*
2579 	 * Enable attention when the link has changed state for
2580 	 * devices that use auto polling.
2581 	 */
2582 	if (sc->bge_flags & BGE_FLAG_TBI) {
2583 		CSR_WRITE_4(sc, BGE_MI_STS, BGE_MISTS_LINK);
2584 	} else {
2585 		if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
2586 			CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode);
2587 			DELAY(80);
2588 		}
2589 		if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
2590 		    sc->bge_chipid != BGE_CHIPID_BCM5700_B2)
2591 			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
2592 			    BGE_EVTENB_MI_INTERRUPT);
2593 	}
2594 
2595 	/*
2596 	 * Clear any pending link state attention.
2597 	 * Otherwise some link state change events may be lost until attention
2598 	 * is cleared by bge_intr() -> bge_link_upd() sequence.
2599 	 * It's not necessary on newer BCM chips - perhaps enabling link
2600 	 * state change attentions implies clearing pending attention.
2601 	 */
2602 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
2603 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
2604 	    BGE_MACSTAT_LINK_CHANGED);
2605 
2606 	/* Enable link state change attentions. */
2607 	BGE_SETBIT(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_LINK_CHANGED);
2608 
2609 	return (0);
2610 }
2611 
2612 static const struct bge_revision *
2613 bge_lookup_rev(uint32_t chipid)
2614 {
2615 	const struct bge_revision *br;
2616 
2617 	for (br = bge_revisions; br->br_name != NULL; br++) {
2618 		if (br->br_chipid == chipid)
2619 			return (br);
2620 	}
2621 
2622 	for (br = bge_majorrevs; br->br_name != NULL; br++) {
2623 		if (br->br_chipid == BGE_ASICREV(chipid))
2624 			return (br);
2625 	}
2626 
2627 	return (NULL);
2628 }
2629 
2630 static const struct bge_vendor *
2631 bge_lookup_vendor(uint16_t vid)
2632 {
2633 	const struct bge_vendor *v;
2634 
2635 	for (v = bge_vendors; v->v_name != NULL; v++)
2636 		if (v->v_id == vid)
2637 			return (v);
2638 
2639 	return (NULL);
2640 }
2641 
2642 static uint32_t
2643 bge_chipid(device_t dev)
2644 {
2645 	uint32_t id;
2646 
2647 	id = pci_read_config(dev, BGE_PCI_MISC_CTL, 4) >>
2648 	    BGE_PCIMISCCTL_ASICREV_SHIFT;
2649 	if (BGE_ASICREV(id) == BGE_ASICREV_USE_PRODID_REG) {
2650 		/*
2651 		 * Find the ASCI revision.  Different chips use different
2652 		 * registers.
2653 		 */
2654 		switch (pci_get_device(dev)) {
2655 		case BCOM_DEVICEID_BCM5717C:
2656 			/* 5717 C0 seems to belong to 5720 line. */
2657 			id = BGE_CHIPID_BCM5720_A0;
2658 			break;
2659 		case BCOM_DEVICEID_BCM5717:
2660 		case BCOM_DEVICEID_BCM5718:
2661 		case BCOM_DEVICEID_BCM5719:
2662 		case BCOM_DEVICEID_BCM5720:
2663 		case BCOM_DEVICEID_BCM5725:
2664 		case BCOM_DEVICEID_BCM5727:
2665 		case BCOM_DEVICEID_BCM5762:
2666 		case BCOM_DEVICEID_BCM57764:
2667 		case BCOM_DEVICEID_BCM57767:
2668 		case BCOM_DEVICEID_BCM57787:
2669 			id = pci_read_config(dev,
2670 			    BGE_PCI_GEN2_PRODID_ASICREV, 4);
2671 			break;
2672 		case BCOM_DEVICEID_BCM57761:
2673 		case BCOM_DEVICEID_BCM57762:
2674 		case BCOM_DEVICEID_BCM57765:
2675 		case BCOM_DEVICEID_BCM57766:
2676 		case BCOM_DEVICEID_BCM57781:
2677 		case BCOM_DEVICEID_BCM57782:
2678 		case BCOM_DEVICEID_BCM57785:
2679 		case BCOM_DEVICEID_BCM57786:
2680 		case BCOM_DEVICEID_BCM57791:
2681 		case BCOM_DEVICEID_BCM57795:
2682 			id = pci_read_config(dev,
2683 			    BGE_PCI_GEN15_PRODID_ASICREV, 4);
2684 			break;
2685 		default:
2686 			id = pci_read_config(dev, BGE_PCI_PRODID_ASICREV, 4);
2687 		}
2688 	}
2689 	return (id);
2690 }
2691 
2692 /*
2693  * Probe for a Broadcom chip. Check the PCI vendor and device IDs
2694  * against our list and return its name if we find a match.
2695  *
2696  * Note that since the Broadcom controller contains VPD support, we
2697  * try to get the device name string from the controller itself instead
2698  * of the compiled-in string. It guarantees we'll always announce the
2699  * right product name. We fall back to the compiled-in string when
2700  * VPD is unavailable or corrupt.
2701  */
2702 static int
2703 bge_probe(device_t dev)
2704 {
2705 	char buf[96];
2706 	char model[64];
2707 	const struct bge_revision *br;
2708 	const char *pname;
2709 	struct bge_softc *sc;
2710 	const struct bge_type *t = bge_devs;
2711 	const struct bge_vendor *v;
2712 	uint32_t id;
2713 	uint16_t did, vid;
2714 
2715 	sc = device_get_softc(dev);
2716 	sc->bge_dev = dev;
2717 	vid = pci_get_vendor(dev);
2718 	did = pci_get_device(dev);
2719 	while(t->bge_vid != 0) {
2720 		if ((vid == t->bge_vid) && (did == t->bge_did)) {
2721 			id = bge_chipid(dev);
2722 			br = bge_lookup_rev(id);
2723 			if (bge_has_eaddr(sc) &&
2724 			    pci_get_vpd_ident(dev, &pname) == 0)
2725 				snprintf(model, sizeof(model), "%s", pname);
2726 			else {
2727 				v = bge_lookup_vendor(vid);
2728 				snprintf(model, sizeof(model), "%s %s",
2729 				    v != NULL ? v->v_name : "Unknown",
2730 				    br != NULL ? br->br_name :
2731 				    "NetXtreme/NetLink Ethernet Controller");
2732 			}
2733 			snprintf(buf, sizeof(buf), "%s, %sASIC rev. %#08x",
2734 			    model, br != NULL ? "" : "unknown ", id);
2735 			device_set_desc_copy(dev, buf);
2736 			return (BUS_PROBE_DEFAULT);
2737 		}
2738 		t++;
2739 	}
2740 
2741 	return (ENXIO);
2742 }
2743 
2744 static void
2745 bge_dma_free(struct bge_softc *sc)
2746 {
2747 	int i;
2748 
2749 	/* Destroy DMA maps for RX buffers. */
2750 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
2751 		if (sc->bge_cdata.bge_rx_std_dmamap[i])
2752 			bus_dmamap_destroy(sc->bge_cdata.bge_rx_mtag,
2753 			    sc->bge_cdata.bge_rx_std_dmamap[i]);
2754 	}
2755 	if (sc->bge_cdata.bge_rx_std_sparemap)
2756 		bus_dmamap_destroy(sc->bge_cdata.bge_rx_mtag,
2757 		    sc->bge_cdata.bge_rx_std_sparemap);
2758 
2759 	/* Destroy DMA maps for jumbo RX buffers. */
2760 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
2761 		if (sc->bge_cdata.bge_rx_jumbo_dmamap[i])
2762 			bus_dmamap_destroy(sc->bge_cdata.bge_mtag_jumbo,
2763 			    sc->bge_cdata.bge_rx_jumbo_dmamap[i]);
2764 	}
2765 	if (sc->bge_cdata.bge_rx_jumbo_sparemap)
2766 		bus_dmamap_destroy(sc->bge_cdata.bge_mtag_jumbo,
2767 		    sc->bge_cdata.bge_rx_jumbo_sparemap);
2768 
2769 	/* Destroy DMA maps for TX buffers. */
2770 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
2771 		if (sc->bge_cdata.bge_tx_dmamap[i])
2772 			bus_dmamap_destroy(sc->bge_cdata.bge_tx_mtag,
2773 			    sc->bge_cdata.bge_tx_dmamap[i]);
2774 	}
2775 
2776 	if (sc->bge_cdata.bge_rx_mtag)
2777 		bus_dma_tag_destroy(sc->bge_cdata.bge_rx_mtag);
2778 	if (sc->bge_cdata.bge_mtag_jumbo)
2779 		bus_dma_tag_destroy(sc->bge_cdata.bge_mtag_jumbo);
2780 	if (sc->bge_cdata.bge_tx_mtag)
2781 		bus_dma_tag_destroy(sc->bge_cdata.bge_tx_mtag);
2782 
2783 	/* Destroy standard RX ring. */
2784 	if (sc->bge_ldata.bge_rx_std_ring_paddr)
2785 		bus_dmamap_unload(sc->bge_cdata.bge_rx_std_ring_tag,
2786 		    sc->bge_cdata.bge_rx_std_ring_map);
2787 	if (sc->bge_ldata.bge_rx_std_ring)
2788 		bus_dmamem_free(sc->bge_cdata.bge_rx_std_ring_tag,
2789 		    sc->bge_ldata.bge_rx_std_ring,
2790 		    sc->bge_cdata.bge_rx_std_ring_map);
2791 
2792 	if (sc->bge_cdata.bge_rx_std_ring_tag)
2793 		bus_dma_tag_destroy(sc->bge_cdata.bge_rx_std_ring_tag);
2794 
2795 	/* Destroy jumbo RX ring. */
2796 	if (sc->bge_ldata.bge_rx_jumbo_ring_paddr)
2797 		bus_dmamap_unload(sc->bge_cdata.bge_rx_jumbo_ring_tag,
2798 		    sc->bge_cdata.bge_rx_jumbo_ring_map);
2799 
2800 	if (sc->bge_ldata.bge_rx_jumbo_ring)
2801 		bus_dmamem_free(sc->bge_cdata.bge_rx_jumbo_ring_tag,
2802 		    sc->bge_ldata.bge_rx_jumbo_ring,
2803 		    sc->bge_cdata.bge_rx_jumbo_ring_map);
2804 
2805 	if (sc->bge_cdata.bge_rx_jumbo_ring_tag)
2806 		bus_dma_tag_destroy(sc->bge_cdata.bge_rx_jumbo_ring_tag);
2807 
2808 	/* Destroy RX return ring. */
2809 	if (sc->bge_ldata.bge_rx_return_ring_paddr)
2810 		bus_dmamap_unload(sc->bge_cdata.bge_rx_return_ring_tag,
2811 		    sc->bge_cdata.bge_rx_return_ring_map);
2812 
2813 	if (sc->bge_ldata.bge_rx_return_ring)
2814 		bus_dmamem_free(sc->bge_cdata.bge_rx_return_ring_tag,
2815 		    sc->bge_ldata.bge_rx_return_ring,
2816 		    sc->bge_cdata.bge_rx_return_ring_map);
2817 
2818 	if (sc->bge_cdata.bge_rx_return_ring_tag)
2819 		bus_dma_tag_destroy(sc->bge_cdata.bge_rx_return_ring_tag);
2820 
2821 	/* Destroy TX ring. */
2822 	if (sc->bge_ldata.bge_tx_ring_paddr)
2823 		bus_dmamap_unload(sc->bge_cdata.bge_tx_ring_tag,
2824 		    sc->bge_cdata.bge_tx_ring_map);
2825 
2826 	if (sc->bge_ldata.bge_tx_ring)
2827 		bus_dmamem_free(sc->bge_cdata.bge_tx_ring_tag,
2828 		    sc->bge_ldata.bge_tx_ring,
2829 		    sc->bge_cdata.bge_tx_ring_map);
2830 
2831 	if (sc->bge_cdata.bge_tx_ring_tag)
2832 		bus_dma_tag_destroy(sc->bge_cdata.bge_tx_ring_tag);
2833 
2834 	/* Destroy status block. */
2835 	if (sc->bge_ldata.bge_status_block_paddr)
2836 		bus_dmamap_unload(sc->bge_cdata.bge_status_tag,
2837 		    sc->bge_cdata.bge_status_map);
2838 
2839 	if (sc->bge_ldata.bge_status_block)
2840 		bus_dmamem_free(sc->bge_cdata.bge_status_tag,
2841 		    sc->bge_ldata.bge_status_block,
2842 		    sc->bge_cdata.bge_status_map);
2843 
2844 	if (sc->bge_cdata.bge_status_tag)
2845 		bus_dma_tag_destroy(sc->bge_cdata.bge_status_tag);
2846 
2847 	/* Destroy statistics block. */
2848 	if (sc->bge_ldata.bge_stats_paddr)
2849 		bus_dmamap_unload(sc->bge_cdata.bge_stats_tag,
2850 		    sc->bge_cdata.bge_stats_map);
2851 
2852 	if (sc->bge_ldata.bge_stats)
2853 		bus_dmamem_free(sc->bge_cdata.bge_stats_tag,
2854 		    sc->bge_ldata.bge_stats,
2855 		    sc->bge_cdata.bge_stats_map);
2856 
2857 	if (sc->bge_cdata.bge_stats_tag)
2858 		bus_dma_tag_destroy(sc->bge_cdata.bge_stats_tag);
2859 
2860 	if (sc->bge_cdata.bge_buffer_tag)
2861 		bus_dma_tag_destroy(sc->bge_cdata.bge_buffer_tag);
2862 
2863 	/* Destroy the parent tag. */
2864 	if (sc->bge_cdata.bge_parent_tag)
2865 		bus_dma_tag_destroy(sc->bge_cdata.bge_parent_tag);
2866 }
2867 
2868 static int
2869 bge_dma_ring_alloc(struct bge_softc *sc, bus_size_t alignment,
2870     bus_size_t maxsize, bus_dma_tag_t *tag, uint8_t **ring, bus_dmamap_t *map,
2871     bus_addr_t *paddr, const char *msg)
2872 {
2873 	struct bge_dmamap_arg ctx;
2874 	bus_addr_t lowaddr;
2875 	bus_size_t ring_end;
2876 	int error;
2877 
2878 	lowaddr = BUS_SPACE_MAXADDR;
2879 again:
2880 	error = bus_dma_tag_create(sc->bge_cdata.bge_parent_tag,
2881 	    alignment, 0, lowaddr, BUS_SPACE_MAXADDR, NULL,
2882 	    NULL, maxsize, 1, maxsize, 0, NULL, NULL, tag);
2883 	if (error != 0) {
2884 		device_printf(sc->bge_dev,
2885 		    "could not create %s dma tag\n", msg);
2886 		return (ENOMEM);
2887 	}
2888 	/* Allocate DMA'able memory for ring. */
2889 	error = bus_dmamem_alloc(*tag, (void **)ring,
2890 	    BUS_DMA_NOWAIT | BUS_DMA_ZERO | BUS_DMA_COHERENT, map);
2891 	if (error != 0) {
2892 		device_printf(sc->bge_dev,
2893 		    "could not allocate DMA'able memory for %s\n", msg);
2894 		return (ENOMEM);
2895 	}
2896 	/* Load the address of the ring. */
2897 	ctx.bge_busaddr = 0;
2898 	error = bus_dmamap_load(*tag, *map, *ring, maxsize, bge_dma_map_addr,
2899 	    &ctx, BUS_DMA_NOWAIT);
2900 	if (error != 0) {
2901 		device_printf(sc->bge_dev,
2902 		    "could not load DMA'able memory for %s\n", msg);
2903 		return (ENOMEM);
2904 	}
2905 	*paddr = ctx.bge_busaddr;
2906 	ring_end = *paddr + maxsize;
2907 	if ((sc->bge_flags & BGE_FLAG_4G_BNDRY_BUG) != 0 &&
2908 	    BGE_ADDR_HI(*paddr) != BGE_ADDR_HI(ring_end)) {
2909 		/*
2910 		 * 4GB boundary crossed.  Limit maximum allowable DMA
2911 		 * address space to 32bit and try again.
2912 		 */
2913 		bus_dmamap_unload(*tag, *map);
2914 		bus_dmamem_free(*tag, *ring, *map);
2915 		bus_dma_tag_destroy(*tag);
2916 		if (bootverbose)
2917 			device_printf(sc->bge_dev, "4GB boundary crossed, "
2918 			    "limit DMA address space to 32bit for %s\n", msg);
2919 		*ring = NULL;
2920 		*tag = NULL;
2921 		*map = NULL;
2922 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
2923 		goto again;
2924 	}
2925 	return (0);
2926 }
2927 
2928 static int
2929 bge_dma_alloc(struct bge_softc *sc)
2930 {
2931 	bus_addr_t lowaddr;
2932 	bus_size_t boundary, sbsz, rxmaxsegsz, txsegsz, txmaxsegsz;
2933 	int i, error;
2934 
2935 	lowaddr = BUS_SPACE_MAXADDR;
2936 	if ((sc->bge_flags & BGE_FLAG_40BIT_BUG) != 0)
2937 		lowaddr = BGE_DMA_MAXADDR;
2938 	/*
2939 	 * Allocate the parent bus DMA tag appropriate for PCI.
2940 	 */
2941 	error = bus_dma_tag_create(bus_get_dma_tag(sc->bge_dev),
2942 	    1, 0, lowaddr, BUS_SPACE_MAXADDR, NULL,
2943 	    NULL, BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT,
2944 	    0, NULL, NULL, &sc->bge_cdata.bge_parent_tag);
2945 	if (error != 0) {
2946 		device_printf(sc->bge_dev,
2947 		    "could not allocate parent dma tag\n");
2948 		return (ENOMEM);
2949 	}
2950 
2951 	/* Create tag for standard RX ring. */
2952 	error = bge_dma_ring_alloc(sc, PAGE_SIZE, BGE_STD_RX_RING_SZ,
2953 	    &sc->bge_cdata.bge_rx_std_ring_tag,
2954 	    (uint8_t **)&sc->bge_ldata.bge_rx_std_ring,
2955 	    &sc->bge_cdata.bge_rx_std_ring_map,
2956 	    &sc->bge_ldata.bge_rx_std_ring_paddr, "RX ring");
2957 	if (error)
2958 		return (error);
2959 
2960 	/* Create tag for RX return ring. */
2961 	error = bge_dma_ring_alloc(sc, PAGE_SIZE, BGE_RX_RTN_RING_SZ(sc),
2962 	    &sc->bge_cdata.bge_rx_return_ring_tag,
2963 	    (uint8_t **)&sc->bge_ldata.bge_rx_return_ring,
2964 	    &sc->bge_cdata.bge_rx_return_ring_map,
2965 	    &sc->bge_ldata.bge_rx_return_ring_paddr, "RX return ring");
2966 	if (error)
2967 		return (error);
2968 
2969 	/* Create tag for TX ring. */
2970 	error = bge_dma_ring_alloc(sc, PAGE_SIZE, BGE_TX_RING_SZ,
2971 	    &sc->bge_cdata.bge_tx_ring_tag,
2972 	    (uint8_t **)&sc->bge_ldata.bge_tx_ring,
2973 	    &sc->bge_cdata.bge_tx_ring_map,
2974 	    &sc->bge_ldata.bge_tx_ring_paddr, "TX ring");
2975 	if (error)
2976 		return (error);
2977 
2978 	/*
2979 	 * Create tag for status block.
2980 	 * Because we only use single Tx/Rx/Rx return ring, use
2981 	 * minimum status block size except BCM5700 AX/BX which
2982 	 * seems to want to see full status block size regardless
2983 	 * of configured number of ring.
2984 	 */
2985 	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
2986 	    sc->bge_chipid != BGE_CHIPID_BCM5700_C0)
2987 		sbsz = BGE_STATUS_BLK_SZ;
2988 	else
2989 		sbsz = 32;
2990 	error = bge_dma_ring_alloc(sc, PAGE_SIZE, sbsz,
2991 	    &sc->bge_cdata.bge_status_tag,
2992 	    (uint8_t **)&sc->bge_ldata.bge_status_block,
2993 	    &sc->bge_cdata.bge_status_map,
2994 	    &sc->bge_ldata.bge_status_block_paddr, "status block");
2995 	if (error)
2996 		return (error);
2997 
2998 	/* Create tag for statistics block. */
2999 	error = bge_dma_ring_alloc(sc, PAGE_SIZE, BGE_STATS_SZ,
3000 	    &sc->bge_cdata.bge_stats_tag,
3001 	    (uint8_t **)&sc->bge_ldata.bge_stats,
3002 	    &sc->bge_cdata.bge_stats_map,
3003 	    &sc->bge_ldata.bge_stats_paddr, "statistics block");
3004 	if (error)
3005 		return (error);
3006 
3007 	/* Create tag for jumbo RX ring. */
3008 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
3009 		error = bge_dma_ring_alloc(sc, PAGE_SIZE, BGE_JUMBO_RX_RING_SZ,
3010 		    &sc->bge_cdata.bge_rx_jumbo_ring_tag,
3011 		    (uint8_t **)&sc->bge_ldata.bge_rx_jumbo_ring,
3012 		    &sc->bge_cdata.bge_rx_jumbo_ring_map,
3013 		    &sc->bge_ldata.bge_rx_jumbo_ring_paddr, "jumbo RX ring");
3014 		if (error)
3015 			return (error);
3016 	}
3017 
3018 	/* Create parent tag for buffers. */
3019 	boundary = 0;
3020 	if ((sc->bge_flags & BGE_FLAG_4G_BNDRY_BUG) != 0) {
3021 		boundary = BGE_DMA_BNDRY;
3022 		/*
3023 		 * XXX
3024 		 * watchdog timeout issue was observed on BCM5704 which
3025 		 * lives behind PCI-X bridge(e.g AMD 8131 PCI-X bridge).
3026 		 * Both limiting DMA address space to 32bits and flushing
3027 		 * mailbox write seem to address the issue.
3028 		 */
3029 		if (sc->bge_pcixcap != 0)
3030 			lowaddr = BUS_SPACE_MAXADDR_32BIT;
3031 	}
3032 	error = bus_dma_tag_create(bus_get_dma_tag(sc->bge_dev),
3033 	    1, boundary, lowaddr, BUS_SPACE_MAXADDR, NULL,
3034 	    NULL, BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT,
3035 	    0, NULL, NULL, &sc->bge_cdata.bge_buffer_tag);
3036 	if (error != 0) {
3037 		device_printf(sc->bge_dev,
3038 		    "could not allocate buffer dma tag\n");
3039 		return (ENOMEM);
3040 	}
3041 	/* Create tag for Tx mbufs. */
3042 	if (sc->bge_flags & (BGE_FLAG_TSO | BGE_FLAG_TSO3)) {
3043 		txsegsz = BGE_TSOSEG_SZ;
3044 		txmaxsegsz = 65535 + sizeof(struct ether_vlan_header);
3045 	} else {
3046 		txsegsz = MCLBYTES;
3047 		txmaxsegsz = MCLBYTES * BGE_NSEG_NEW;
3048 	}
3049 	error = bus_dma_tag_create(sc->bge_cdata.bge_buffer_tag, 1,
3050 	    0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
3051 	    txmaxsegsz, BGE_NSEG_NEW, txsegsz, 0, NULL, NULL,
3052 	    &sc->bge_cdata.bge_tx_mtag);
3053 
3054 	if (error) {
3055 		device_printf(sc->bge_dev, "could not allocate TX dma tag\n");
3056 		return (ENOMEM);
3057 	}
3058 
3059 	/* Create tag for Rx mbufs. */
3060 	if (sc->bge_flags & BGE_FLAG_JUMBO_STD)
3061 		rxmaxsegsz = MJUM9BYTES;
3062 	else
3063 		rxmaxsegsz = MCLBYTES;
3064 	error = bus_dma_tag_create(sc->bge_cdata.bge_buffer_tag, 1, 0,
3065 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, rxmaxsegsz, 1,
3066 	    rxmaxsegsz, 0, NULL, NULL, &sc->bge_cdata.bge_rx_mtag);
3067 
3068 	if (error) {
3069 		device_printf(sc->bge_dev, "could not allocate RX dma tag\n");
3070 		return (ENOMEM);
3071 	}
3072 
3073 	/* Create DMA maps for RX buffers. */
3074 	error = bus_dmamap_create(sc->bge_cdata.bge_rx_mtag, 0,
3075 	    &sc->bge_cdata.bge_rx_std_sparemap);
3076 	if (error) {
3077 		device_printf(sc->bge_dev,
3078 		    "can't create spare DMA map for RX\n");
3079 		return (ENOMEM);
3080 	}
3081 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
3082 		error = bus_dmamap_create(sc->bge_cdata.bge_rx_mtag, 0,
3083 			    &sc->bge_cdata.bge_rx_std_dmamap[i]);
3084 		if (error) {
3085 			device_printf(sc->bge_dev,
3086 			    "can't create DMA map for RX\n");
3087 			return (ENOMEM);
3088 		}
3089 	}
3090 
3091 	/* Create DMA maps for TX buffers. */
3092 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
3093 		error = bus_dmamap_create(sc->bge_cdata.bge_tx_mtag, 0,
3094 			    &sc->bge_cdata.bge_tx_dmamap[i]);
3095 		if (error) {
3096 			device_printf(sc->bge_dev,
3097 			    "can't create DMA map for TX\n");
3098 			return (ENOMEM);
3099 		}
3100 	}
3101 
3102 	/* Create tags for jumbo RX buffers. */
3103 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
3104 		error = bus_dma_tag_create(sc->bge_cdata.bge_buffer_tag,
3105 		    1, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
3106 		    NULL, MJUM9BYTES, BGE_NSEG_JUMBO, PAGE_SIZE,
3107 		    0, NULL, NULL, &sc->bge_cdata.bge_mtag_jumbo);
3108 		if (error) {
3109 			device_printf(sc->bge_dev,
3110 			    "could not allocate jumbo dma tag\n");
3111 			return (ENOMEM);
3112 		}
3113 		/* Create DMA maps for jumbo RX buffers. */
3114 		error = bus_dmamap_create(sc->bge_cdata.bge_mtag_jumbo,
3115 		    0, &sc->bge_cdata.bge_rx_jumbo_sparemap);
3116 		if (error) {
3117 			device_printf(sc->bge_dev,
3118 			    "can't create spare DMA map for jumbo RX\n");
3119 			return (ENOMEM);
3120 		}
3121 		for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
3122 			error = bus_dmamap_create(sc->bge_cdata.bge_mtag_jumbo,
3123 				    0, &sc->bge_cdata.bge_rx_jumbo_dmamap[i]);
3124 			if (error) {
3125 				device_printf(sc->bge_dev,
3126 				    "can't create DMA map for jumbo RX\n");
3127 				return (ENOMEM);
3128 			}
3129 		}
3130 	}
3131 
3132 	return (0);
3133 }
3134 
3135 /*
3136  * Return true if this device has more than one port.
3137  */
3138 static int
3139 bge_has_multiple_ports(struct bge_softc *sc)
3140 {
3141 	device_t dev = sc->bge_dev;
3142 	u_int b, d, f, fscan, s;
3143 
3144 	d = pci_get_domain(dev);
3145 	b = pci_get_bus(dev);
3146 	s = pci_get_slot(dev);
3147 	f = pci_get_function(dev);
3148 	for (fscan = 0; fscan <= PCI_FUNCMAX; fscan++)
3149 		if (fscan != f && pci_find_dbsf(d, b, s, fscan) != NULL)
3150 			return (1);
3151 	return (0);
3152 }
3153 
3154 /*
3155  * Return true if MSI can be used with this device.
3156  */
3157 static int
3158 bge_can_use_msi(struct bge_softc *sc)
3159 {
3160 	int can_use_msi = 0;
3161 
3162 	if (sc->bge_msi == 0)
3163 		return (0);
3164 
3165 	/* Disable MSI for polling(4). */
3166 #ifdef DEVICE_POLLING
3167 	return (0);
3168 #endif
3169 	switch (sc->bge_asicrev) {
3170 	case BGE_ASICREV_BCM5714_A0:
3171 	case BGE_ASICREV_BCM5714:
3172 		/*
3173 		 * Apparently, MSI doesn't work when these chips are
3174 		 * configured in single-port mode.
3175 		 */
3176 		if (bge_has_multiple_ports(sc))
3177 			can_use_msi = 1;
3178 		break;
3179 	case BGE_ASICREV_BCM5750:
3180 		if (sc->bge_chiprev != BGE_CHIPREV_5750_AX &&
3181 		    sc->bge_chiprev != BGE_CHIPREV_5750_BX)
3182 			can_use_msi = 1;
3183 		break;
3184 	case BGE_ASICREV_BCM5784:
3185 		/*
3186 		 * Prevent infinite "watchdog timeout" errors
3187 		 * in some MacBook Pro and make it work out-of-the-box.
3188 		 */
3189 		if (sc->bge_chiprev == BGE_CHIPREV_5784_AX)
3190 			break;
3191 		/* FALLTHROUGH */
3192 	default:
3193 		if (BGE_IS_575X_PLUS(sc))
3194 			can_use_msi = 1;
3195 	}
3196 	return (can_use_msi);
3197 }
3198 
3199 static int
3200 bge_mbox_reorder(struct bge_softc *sc)
3201 {
3202 	/* Lists of PCI bridges that are known to reorder mailbox writes. */
3203 	static const struct mbox_reorder {
3204 		const uint16_t vendor;
3205 		const uint16_t device;
3206 		const char *desc;
3207 	} mbox_reorder_lists[] = {
3208 		{ 0x1022, 0x7450, "AMD-8131 PCI-X Bridge" },
3209 	};
3210 	devclass_t pci, pcib;
3211 	device_t bus, dev;
3212 	int i;
3213 
3214 	pci = devclass_find("pci");
3215 	pcib = devclass_find("pcib");
3216 	dev = sc->bge_dev;
3217 	bus = device_get_parent(dev);
3218 	for (;;) {
3219 		dev = device_get_parent(bus);
3220 		bus = device_get_parent(dev);
3221 		if (device_get_devclass(dev) != pcib)
3222 			break;
3223 		if (device_get_devclass(bus) != pci)
3224 			break;
3225 		for (i = 0; i < nitems(mbox_reorder_lists); i++) {
3226 			if (pci_get_vendor(dev) ==
3227 			    mbox_reorder_lists[i].vendor &&
3228 			    pci_get_device(dev) ==
3229 			    mbox_reorder_lists[i].device) {
3230 				device_printf(sc->bge_dev,
3231 				    "enabling MBOX workaround for %s\n",
3232 				    mbox_reorder_lists[i].desc);
3233 				return (1);
3234 			}
3235 		}
3236 	}
3237 	return (0);
3238 }
3239 
3240 static void
3241 bge_devinfo(struct bge_softc *sc)
3242 {
3243 	uint32_t cfg, clk;
3244 
3245 	device_printf(sc->bge_dev,
3246 	    "CHIP ID 0x%08x; ASIC REV 0x%02x; CHIP REV 0x%02x; ",
3247 	    sc->bge_chipid, sc->bge_asicrev, sc->bge_chiprev);
3248 	if (sc->bge_flags & BGE_FLAG_PCIE)
3249 		printf("PCI-E\n");
3250 	else if (sc->bge_flags & BGE_FLAG_PCIX) {
3251 		printf("PCI-X ");
3252 		cfg = CSR_READ_4(sc, BGE_MISC_CFG) & BGE_MISCCFG_BOARD_ID_MASK;
3253 		if (cfg == BGE_MISCCFG_BOARD_ID_5704CIOBE)
3254 			clk = 133;
3255 		else {
3256 			clk = CSR_READ_4(sc, BGE_PCI_CLKCTL) & 0x1F;
3257 			switch (clk) {
3258 			case 0:
3259 				clk = 33;
3260 				break;
3261 			case 2:
3262 				clk = 50;
3263 				break;
3264 			case 4:
3265 				clk = 66;
3266 				break;
3267 			case 6:
3268 				clk = 100;
3269 				break;
3270 			case 7:
3271 				clk = 133;
3272 				break;
3273 			}
3274 		}
3275 		printf("%u MHz\n", clk);
3276 	} else {
3277 		if (sc->bge_pcixcap != 0)
3278 			printf("PCI on PCI-X ");
3279 		else
3280 			printf("PCI ");
3281 		cfg = pci_read_config(sc->bge_dev, BGE_PCI_PCISTATE, 4);
3282 		if (cfg & BGE_PCISTATE_PCI_BUSSPEED)
3283 			clk = 66;
3284 		else
3285 			clk = 33;
3286 		if (cfg & BGE_PCISTATE_32BIT_BUS)
3287 			printf("%u MHz; 32bit\n", clk);
3288 		else
3289 			printf("%u MHz; 64bit\n", clk);
3290 	}
3291 }
3292 
3293 static int
3294 bge_attach(device_t dev)
3295 {
3296 	if_t ifp;
3297 	struct bge_softc *sc;
3298 	uint32_t hwcfg = 0, misccfg, pcistate;
3299 	u_char eaddr[ETHER_ADDR_LEN];
3300 	int capmask, error, reg, rid, trys;
3301 
3302 	sc = device_get_softc(dev);
3303 	sc->bge_dev = dev;
3304 
3305 	BGE_LOCK_INIT(sc, device_get_nameunit(dev));
3306 	NET_TASK_INIT(&sc->bge_intr_task, 0, bge_intr_task, sc);
3307 	callout_init_mtx(&sc->bge_stat_ch, &sc->bge_mtx, 0);
3308 
3309 	pci_enable_busmaster(dev);
3310 
3311 	/*
3312 	 * Allocate control/status registers.
3313 	 */
3314 	rid = PCIR_BAR(0);
3315 	sc->bge_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
3316 	    RF_ACTIVE);
3317 
3318 	if (sc->bge_res == NULL) {
3319 		device_printf (sc->bge_dev, "couldn't map BAR0 memory\n");
3320 		error = ENXIO;
3321 		goto fail;
3322 	}
3323 
3324 	/* Save various chip information. */
3325 	sc->bge_func_addr = pci_get_function(dev);
3326 	sc->bge_chipid = bge_chipid(dev);
3327 	sc->bge_asicrev = BGE_ASICREV(sc->bge_chipid);
3328 	sc->bge_chiprev = BGE_CHIPREV(sc->bge_chipid);
3329 
3330 	/* Set default PHY address. */
3331 	sc->bge_phy_addr = 1;
3332 	 /*
3333 	  * PHY address mapping for various devices.
3334 	  *
3335 	  *          | F0 Cu | F0 Sr | F1 Cu | F1 Sr |
3336 	  * ---------+-------+-------+-------+-------+
3337 	  * BCM57XX  |   1   |   X   |   X   |   X   |
3338 	  * BCM5704  |   1   |   X   |   1   |   X   |
3339 	  * BCM5717  |   1   |   8   |   2   |   9   |
3340 	  * BCM5719  |   1   |   8   |   2   |   9   |
3341 	  * BCM5720  |   1   |   8   |   2   |   9   |
3342 	  *
3343 	  *          | F2 Cu | F2 Sr | F3 Cu | F3 Sr |
3344 	  * ---------+-------+-------+-------+-------+
3345 	  * BCM57XX  |   X   |   X   |   X   |   X   |
3346 	  * BCM5704  |   X   |   X   |   X   |   X   |
3347 	  * BCM5717  |   X   |   X   |   X   |   X   |
3348 	  * BCM5719  |   3   |   10  |   4   |   11  |
3349 	  * BCM5720  |   X   |   X   |   X   |   X   |
3350 	  *
3351 	  * Other addresses may respond but they are not
3352 	  * IEEE compliant PHYs and should be ignored.
3353 	  */
3354 	if (sc->bge_asicrev == BGE_ASICREV_BCM5717 ||
3355 	    sc->bge_asicrev == BGE_ASICREV_BCM5719 ||
3356 	    sc->bge_asicrev == BGE_ASICREV_BCM5720) {
3357 		if (sc->bge_chipid != BGE_CHIPID_BCM5717_A0) {
3358 			if (CSR_READ_4(sc, BGE_SGDIG_STS) &
3359 			    BGE_SGDIGSTS_IS_SERDES)
3360 				sc->bge_phy_addr = sc->bge_func_addr + 8;
3361 			else
3362 				sc->bge_phy_addr = sc->bge_func_addr + 1;
3363 		} else {
3364 			if (CSR_READ_4(sc, BGE_CPMU_PHY_STRAP) &
3365 			    BGE_CPMU_PHY_STRAP_IS_SERDES)
3366 				sc->bge_phy_addr = sc->bge_func_addr + 8;
3367 			else
3368 				sc->bge_phy_addr = sc->bge_func_addr + 1;
3369 		}
3370 	}
3371 
3372 	if (bge_has_eaddr(sc))
3373 		sc->bge_flags |= BGE_FLAG_EADDR;
3374 
3375 	/* Save chipset family. */
3376 	switch (sc->bge_asicrev) {
3377 	case BGE_ASICREV_BCM5762:
3378 	case BGE_ASICREV_BCM57765:
3379 	case BGE_ASICREV_BCM57766:
3380 		sc->bge_flags |= BGE_FLAG_57765_PLUS;
3381 		/* FALLTHROUGH */
3382 	case BGE_ASICREV_BCM5717:
3383 	case BGE_ASICREV_BCM5719:
3384 	case BGE_ASICREV_BCM5720:
3385 		sc->bge_flags |= BGE_FLAG_5717_PLUS | BGE_FLAG_5755_PLUS |
3386 		    BGE_FLAG_575X_PLUS | BGE_FLAG_5705_PLUS | BGE_FLAG_JUMBO |
3387 		    BGE_FLAG_JUMBO_FRAME;
3388 		if (sc->bge_asicrev == BGE_ASICREV_BCM5719 ||
3389 		    sc->bge_asicrev == BGE_ASICREV_BCM5720) {
3390 			/*
3391 			 * Enable work around for DMA engine miscalculation
3392 			 * of TXMBUF available space.
3393 			 */
3394 			sc->bge_flags |= BGE_FLAG_RDMA_BUG;
3395 			if (sc->bge_asicrev == BGE_ASICREV_BCM5719 &&
3396 			    sc->bge_chipid == BGE_CHIPID_BCM5719_A0) {
3397 				/* Jumbo frame on BCM5719 A0 does not work. */
3398 				sc->bge_flags &= ~BGE_FLAG_JUMBO;
3399 			}
3400 		}
3401 		break;
3402 	case BGE_ASICREV_BCM5755:
3403 	case BGE_ASICREV_BCM5761:
3404 	case BGE_ASICREV_BCM5784:
3405 	case BGE_ASICREV_BCM5785:
3406 	case BGE_ASICREV_BCM5787:
3407 	case BGE_ASICREV_BCM57780:
3408 		sc->bge_flags |= BGE_FLAG_5755_PLUS | BGE_FLAG_575X_PLUS |
3409 		    BGE_FLAG_5705_PLUS;
3410 		break;
3411 	case BGE_ASICREV_BCM5700:
3412 	case BGE_ASICREV_BCM5701:
3413 	case BGE_ASICREV_BCM5703:
3414 	case BGE_ASICREV_BCM5704:
3415 		sc->bge_flags |= BGE_FLAG_5700_FAMILY | BGE_FLAG_JUMBO;
3416 		break;
3417 	case BGE_ASICREV_BCM5714_A0:
3418 	case BGE_ASICREV_BCM5780:
3419 	case BGE_ASICREV_BCM5714:
3420 		sc->bge_flags |= BGE_FLAG_5714_FAMILY | BGE_FLAG_JUMBO_STD;
3421 		/* FALLTHROUGH */
3422 	case BGE_ASICREV_BCM5750:
3423 	case BGE_ASICREV_BCM5752:
3424 	case BGE_ASICREV_BCM5906:
3425 		sc->bge_flags |= BGE_FLAG_575X_PLUS;
3426 		/* FALLTHROUGH */
3427 	case BGE_ASICREV_BCM5705:
3428 		sc->bge_flags |= BGE_FLAG_5705_PLUS;
3429 		break;
3430 	}
3431 
3432 	/* Identify chips with APE processor. */
3433 	switch (sc->bge_asicrev) {
3434 	case BGE_ASICREV_BCM5717:
3435 	case BGE_ASICREV_BCM5719:
3436 	case BGE_ASICREV_BCM5720:
3437 	case BGE_ASICREV_BCM5761:
3438 	case BGE_ASICREV_BCM5762:
3439 		sc->bge_flags |= BGE_FLAG_APE;
3440 		break;
3441 	}
3442 
3443 	/* Chips with APE need BAR2 access for APE registers/memory. */
3444 	if ((sc->bge_flags & BGE_FLAG_APE) != 0) {
3445 		rid = PCIR_BAR(2);
3446 		sc->bge_res2 = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
3447 		    RF_ACTIVE);
3448 		if (sc->bge_res2 == NULL) {
3449 			device_printf (sc->bge_dev,
3450 			    "couldn't map BAR2 memory\n");
3451 			error = ENXIO;
3452 			goto fail;
3453 		}
3454 
3455 		/* Enable APE register/memory access by host driver. */
3456 		pcistate = pci_read_config(dev, BGE_PCI_PCISTATE, 4);
3457 		pcistate |= BGE_PCISTATE_ALLOW_APE_CTLSPC_WR |
3458 		    BGE_PCISTATE_ALLOW_APE_SHMEM_WR |
3459 		    BGE_PCISTATE_ALLOW_APE_PSPACE_WR;
3460 		pci_write_config(dev, BGE_PCI_PCISTATE, pcistate, 4);
3461 
3462 		bge_ape_lock_init(sc);
3463 		bge_ape_read_fw_ver(sc);
3464 	}
3465 
3466 	/* Add SYSCTLs, requires the chipset family to be set. */
3467 	bge_add_sysctls(sc);
3468 
3469 	/* Identify the chips that use an CPMU. */
3470 	if (BGE_IS_5717_PLUS(sc) ||
3471 	    sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
3472 	    sc->bge_asicrev == BGE_ASICREV_BCM5761 ||
3473 	    sc->bge_asicrev == BGE_ASICREV_BCM5785 ||
3474 	    sc->bge_asicrev == BGE_ASICREV_BCM57780)
3475 		sc->bge_flags |= BGE_FLAG_CPMU_PRESENT;
3476 	if ((sc->bge_flags & BGE_FLAG_CPMU_PRESENT) != 0)
3477 		sc->bge_mi_mode = BGE_MIMODE_500KHZ_CONST;
3478 	else
3479 		sc->bge_mi_mode = BGE_MIMODE_BASE;
3480 	/* Enable auto polling for BCM570[0-5]. */
3481 	if (BGE_IS_5700_FAMILY(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5705)
3482 		sc->bge_mi_mode |= BGE_MIMODE_AUTOPOLL;
3483 
3484 	/*
3485 	 * All Broadcom controllers have 4GB boundary DMA bug.
3486 	 * Whenever an address crosses a multiple of the 4GB boundary
3487 	 * (including 4GB, 8Gb, 12Gb, etc.) and makes the transition
3488 	 * from 0xX_FFFF_FFFF to 0x(X+1)_0000_0000 an internal DMA
3489 	 * state machine will lockup and cause the device to hang.
3490 	 */
3491 	sc->bge_flags |= BGE_FLAG_4G_BNDRY_BUG;
3492 
3493 	/* BCM5755 or higher and BCM5906 have short DMA bug. */
3494 	if (BGE_IS_5755_PLUS(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5906)
3495 		sc->bge_flags |= BGE_FLAG_SHORT_DMA_BUG;
3496 
3497 	/*
3498 	 * BCM5719 cannot handle DMA requests for DMA segments that
3499 	 * have larger than 4KB in size.  However the maximum DMA
3500 	 * segment size created in DMA tag is 4KB for TSO, so we
3501 	 * wouldn't encounter the issue here.
3502 	 */
3503 	if (sc->bge_asicrev == BGE_ASICREV_BCM5719)
3504 		sc->bge_flags |= BGE_FLAG_4K_RDMA_BUG;
3505 
3506 	misccfg = CSR_READ_4(sc, BGE_MISC_CFG) & BGE_MISCCFG_BOARD_ID_MASK;
3507 	if (sc->bge_asicrev == BGE_ASICREV_BCM5705) {
3508 		if (misccfg == BGE_MISCCFG_BOARD_ID_5788 ||
3509 		    misccfg == BGE_MISCCFG_BOARD_ID_5788M)
3510 			sc->bge_flags |= BGE_FLAG_5788;
3511 	}
3512 
3513 	capmask = BMSR_DEFCAPMASK;
3514 	if ((sc->bge_asicrev == BGE_ASICREV_BCM5703 &&
3515 	    (misccfg == 0x4000 || misccfg == 0x8000)) ||
3516 	    (sc->bge_asicrev == BGE_ASICREV_BCM5705 &&
3517 	    pci_get_vendor(dev) == BCOM_VENDORID &&
3518 	    (pci_get_device(dev) == BCOM_DEVICEID_BCM5901 ||
3519 	    pci_get_device(dev) == BCOM_DEVICEID_BCM5901A2 ||
3520 	    pci_get_device(dev) == BCOM_DEVICEID_BCM5705F)) ||
3521 	    (pci_get_vendor(dev) == BCOM_VENDORID &&
3522 	    (pci_get_device(dev) == BCOM_DEVICEID_BCM5751F ||
3523 	    pci_get_device(dev) == BCOM_DEVICEID_BCM5753F ||
3524 	    pci_get_device(dev) == BCOM_DEVICEID_BCM5787F)) ||
3525 	    pci_get_device(dev) == BCOM_DEVICEID_BCM57790 ||
3526 	    pci_get_device(dev) == BCOM_DEVICEID_BCM57791 ||
3527 	    pci_get_device(dev) == BCOM_DEVICEID_BCM57795 ||
3528 	    sc->bge_asicrev == BGE_ASICREV_BCM5906) {
3529 		/* These chips are 10/100 only. */
3530 		capmask &= ~BMSR_EXTSTAT;
3531 		sc->bge_phy_flags |= BGE_PHY_NO_WIRESPEED;
3532 	}
3533 
3534 	/*
3535 	 * Some controllers seem to require a special firmware to use
3536 	 * TSO. But the firmware is not available to FreeBSD and Linux
3537 	 * claims that the TSO performed by the firmware is slower than
3538 	 * hardware based TSO. Moreover the firmware based TSO has one
3539 	 * known bug which can't handle TSO if Ethernet header + IP/TCP
3540 	 * header is greater than 80 bytes. A workaround for the TSO
3541 	 * bug exist but it seems it's too expensive than not using
3542 	 * TSO at all. Some hardwares also have the TSO bug so limit
3543 	 * the TSO to the controllers that are not affected TSO issues
3544 	 * (e.g. 5755 or higher).
3545 	 */
3546 	if (BGE_IS_5717_PLUS(sc)) {
3547 		/* BCM5717 requires different TSO configuration. */
3548 		sc->bge_flags |= BGE_FLAG_TSO3;
3549 		if (sc->bge_asicrev == BGE_ASICREV_BCM5719 &&
3550 		    sc->bge_chipid == BGE_CHIPID_BCM5719_A0) {
3551 			/* TSO on BCM5719 A0 does not work. */
3552 			sc->bge_flags &= ~BGE_FLAG_TSO3;
3553 		}
3554 	} else if (BGE_IS_5755_PLUS(sc)) {
3555 		/*
3556 		 * BCM5754 and BCM5787 shares the same ASIC id so
3557 		 * explicit device id check is required.
3558 		 * Due to unknown reason TSO does not work on BCM5755M.
3559 		 */
3560 		if (pci_get_device(dev) != BCOM_DEVICEID_BCM5754 &&
3561 		    pci_get_device(dev) != BCOM_DEVICEID_BCM5754M &&
3562 		    pci_get_device(dev) != BCOM_DEVICEID_BCM5755M)
3563 			sc->bge_flags |= BGE_FLAG_TSO;
3564 	}
3565 
3566 	/*
3567 	 * Check if this is a PCI-X or PCI Express device.
3568 	 */
3569 	if (pci_find_cap(dev, PCIY_EXPRESS, &reg) == 0) {
3570 		/*
3571 		 * Found a PCI Express capabilities register, this
3572 		 * must be a PCI Express device.
3573 		 */
3574 		sc->bge_flags |= BGE_FLAG_PCIE;
3575 		sc->bge_expcap = reg;
3576 		/* Extract supported maximum payload size. */
3577 		sc->bge_mps = pci_read_config(dev, sc->bge_expcap +
3578 		    PCIER_DEVICE_CAP, 2);
3579 		sc->bge_mps = 128 << (sc->bge_mps & PCIEM_CAP_MAX_PAYLOAD);
3580 		if (sc->bge_asicrev == BGE_ASICREV_BCM5719 ||
3581 		    sc->bge_asicrev == BGE_ASICREV_BCM5720)
3582 			sc->bge_expmrq = 2048;
3583 		else
3584 			sc->bge_expmrq = 4096;
3585 		pci_set_max_read_req(dev, sc->bge_expmrq);
3586 	} else {
3587 		/*
3588 		 * Check if the device is in PCI-X Mode.
3589 		 * (This bit is not valid on PCI Express controllers.)
3590 		 */
3591 		if (pci_find_cap(dev, PCIY_PCIX, &reg) == 0)
3592 			sc->bge_pcixcap = reg;
3593 		if ((pci_read_config(dev, BGE_PCI_PCISTATE, 4) &
3594 		    BGE_PCISTATE_PCI_BUSMODE) == 0)
3595 			sc->bge_flags |= BGE_FLAG_PCIX;
3596 	}
3597 
3598 	/*
3599 	 * The 40bit DMA bug applies to the 5714/5715 controllers and is
3600 	 * not actually a MAC controller bug but an issue with the embedded
3601 	 * PCIe to PCI-X bridge in the device. Use 40bit DMA workaround.
3602 	 */
3603 	if (BGE_IS_5714_FAMILY(sc) && (sc->bge_flags & BGE_FLAG_PCIX))
3604 		sc->bge_flags |= BGE_FLAG_40BIT_BUG;
3605 	/*
3606 	 * Some PCI-X bridges are known to trigger write reordering to
3607 	 * the mailbox registers. Typical phenomena is watchdog timeouts
3608 	 * caused by out-of-order TX completions.  Enable workaround for
3609 	 * PCI-X devices that live behind these bridges.
3610 	 * Note, PCI-X controllers can run in PCI mode so we can't use
3611 	 * BGE_FLAG_PCIX flag to detect PCI-X controllers.
3612 	 */
3613 	if (sc->bge_pcixcap != 0 && bge_mbox_reorder(sc) != 0)
3614 		sc->bge_flags |= BGE_FLAG_MBOX_REORDER;
3615 	/*
3616 	 * Allocate the interrupt, using MSI if possible.  These devices
3617 	 * support 8 MSI messages, but only the first one is used in
3618 	 * normal operation.
3619 	 */
3620 	rid = 0;
3621 	if (pci_find_cap(sc->bge_dev, PCIY_MSI, &reg) == 0) {
3622 		sc->bge_msicap = reg;
3623 		reg = 1;
3624 		if (bge_can_use_msi(sc) && pci_alloc_msi(dev, &reg) == 0) {
3625 			rid = 1;
3626 			sc->bge_flags |= BGE_FLAG_MSI;
3627 		}
3628 	}
3629 
3630 	/*
3631 	 * All controllers except BCM5700 supports tagged status but
3632 	 * we use tagged status only for MSI case on BCM5717. Otherwise
3633 	 * MSI on BCM5717 does not work.
3634 	 */
3635 #ifndef DEVICE_POLLING
3636 	if (sc->bge_flags & BGE_FLAG_MSI && BGE_IS_5717_PLUS(sc))
3637 		sc->bge_flags |= BGE_FLAG_TAGGED_STATUS;
3638 #endif
3639 
3640 	sc->bge_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
3641 	    RF_ACTIVE | (rid != 0 ? 0 : RF_SHAREABLE));
3642 
3643 	if (sc->bge_irq == NULL) {
3644 		device_printf(sc->bge_dev, "couldn't map interrupt\n");
3645 		error = ENXIO;
3646 		goto fail;
3647 	}
3648 
3649 	bge_devinfo(sc);
3650 
3651 	sc->bge_asf_mode = 0;
3652 	/* No ASF if APE present. */
3653 	if ((sc->bge_flags & BGE_FLAG_APE) == 0) {
3654 		if (bge_allow_asf && (bge_readmem_ind(sc, BGE_SRAM_DATA_SIG) ==
3655 		    BGE_SRAM_DATA_SIG_MAGIC)) {
3656 			if (bge_readmem_ind(sc, BGE_SRAM_DATA_CFG) &
3657 			    BGE_HWCFG_ASF) {
3658 				sc->bge_asf_mode |= ASF_ENABLE;
3659 				sc->bge_asf_mode |= ASF_STACKUP;
3660 				if (BGE_IS_575X_PLUS(sc))
3661 					sc->bge_asf_mode |= ASF_NEW_HANDSHAKE;
3662 			}
3663 		}
3664 	}
3665 
3666 	bge_stop_fw(sc);
3667 	bge_sig_pre_reset(sc, BGE_RESET_SHUTDOWN);
3668 	if (bge_reset(sc)) {
3669 		device_printf(sc->bge_dev, "chip reset failed\n");
3670 		error = ENXIO;
3671 		goto fail;
3672 	}
3673 
3674 	bge_sig_legacy(sc, BGE_RESET_SHUTDOWN);
3675 	bge_sig_post_reset(sc, BGE_RESET_SHUTDOWN);
3676 
3677 	if (bge_chipinit(sc)) {
3678 		device_printf(sc->bge_dev, "chip initialization failed\n");
3679 		error = ENXIO;
3680 		goto fail;
3681 	}
3682 
3683 	error = bge_get_eaddr(sc, eaddr);
3684 	if (error) {
3685 		device_printf(sc->bge_dev,
3686 		    "failed to read station address\n");
3687 		error = ENXIO;
3688 		goto fail;
3689 	}
3690 
3691 	/* 5705 limits RX return ring to 512 entries. */
3692 	if (BGE_IS_5717_PLUS(sc))
3693 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT;
3694 	else if (BGE_IS_5705_PLUS(sc))
3695 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT_5705;
3696 	else
3697 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT;
3698 
3699 	if (bge_dma_alloc(sc)) {
3700 		device_printf(sc->bge_dev,
3701 		    "failed to allocate DMA resources\n");
3702 		error = ENXIO;
3703 		goto fail;
3704 	}
3705 
3706 	/* Set default tuneable values. */
3707 	sc->bge_stat_ticks = BGE_TICKS_PER_SEC;
3708 	sc->bge_rx_coal_ticks = 150;
3709 	sc->bge_tx_coal_ticks = 150;
3710 	sc->bge_rx_max_coal_bds = 10;
3711 	sc->bge_tx_max_coal_bds = 10;
3712 
3713 	/* Initialize checksum features to use. */
3714 	sc->bge_csum_features = BGE_CSUM_FEATURES;
3715 	if (sc->bge_forced_udpcsum != 0)
3716 		sc->bge_csum_features |= CSUM_UDP;
3717 
3718 	/* Set up ifnet structure */
3719 	ifp = sc->bge_ifp = if_alloc(IFT_ETHER);
3720 	if (ifp == NULL) {
3721 		device_printf(sc->bge_dev, "failed to if_alloc()\n");
3722 		error = ENXIO;
3723 		goto fail;
3724 	}
3725 	if_setsoftc(ifp, sc);
3726 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
3727 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
3728 	if_setioctlfn(ifp, bge_ioctl);
3729 	if_setstartfn(ifp, bge_start);
3730 	if_setinitfn(ifp, bge_init);
3731 	if_setgetcounterfn(ifp, bge_get_counter);
3732 	if_setsendqlen(ifp, BGE_TX_RING_CNT - 1);
3733 	if_setsendqready(ifp);
3734 	if_sethwassist(ifp, sc->bge_csum_features);
3735 	if_setcapabilities(ifp, IFCAP_HWCSUM | IFCAP_VLAN_HWTAGGING |
3736 	    IFCAP_VLAN_MTU);
3737 	if ((sc->bge_flags & (BGE_FLAG_TSO | BGE_FLAG_TSO3)) != 0) {
3738 		if_sethwassistbits(ifp, CSUM_TSO, 0);
3739 		if_setcapabilitiesbit(ifp, IFCAP_TSO4 | IFCAP_VLAN_HWTSO, 0);
3740 	}
3741 #ifdef IFCAP_VLAN_HWCSUM
3742 	if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWCSUM, 0);
3743 #endif
3744 	if_setcapenable(ifp, if_getcapabilities(ifp));
3745 #ifdef DEVICE_POLLING
3746 	if_setcapabilitiesbit(ifp, IFCAP_POLLING, 0);
3747 #endif
3748 
3749 	/*
3750 	 * 5700 B0 chips do not support checksumming correctly due
3751 	 * to hardware bugs.
3752 	 */
3753 	if (sc->bge_chipid == BGE_CHIPID_BCM5700_B0) {
3754 		if_setcapabilitiesbit(ifp, 0, IFCAP_HWCSUM);
3755 		if_setcapenablebit(ifp, 0, IFCAP_HWCSUM);
3756 		if_sethwassist(ifp, 0);
3757 	}
3758 
3759 	/*
3760 	 * Figure out what sort of media we have by checking the
3761 	 * hardware config word in the first 32k of NIC internal memory,
3762 	 * or fall back to examining the EEPROM if necessary.
3763 	 * Note: on some BCM5700 cards, this value appears to be unset.
3764 	 * If that's the case, we have to rely on identifying the NIC
3765 	 * by its PCI subsystem ID, as we do below for the SysKonnect
3766 	 * SK-9D41.
3767 	 */
3768 	if (bge_readmem_ind(sc, BGE_SRAM_DATA_SIG) == BGE_SRAM_DATA_SIG_MAGIC)
3769 		hwcfg = bge_readmem_ind(sc, BGE_SRAM_DATA_CFG);
3770 	else if ((sc->bge_flags & BGE_FLAG_EADDR) &&
3771 	    (sc->bge_asicrev != BGE_ASICREV_BCM5906)) {
3772 		if (bge_read_eeprom(sc, (caddr_t)&hwcfg, BGE_EE_HWCFG_OFFSET,
3773 		    sizeof(hwcfg))) {
3774 			device_printf(sc->bge_dev, "failed to read EEPROM\n");
3775 			error = ENXIO;
3776 			goto fail;
3777 		}
3778 		hwcfg = ntohl(hwcfg);
3779 	}
3780 
3781 	/* The SysKonnect SK-9D41 is a 1000baseSX card. */
3782 	if ((pci_read_config(dev, BGE_PCI_SUBSYS, 4) >> 16) ==
3783 	    SK_SUBSYSID_9D41 || (hwcfg & BGE_HWCFG_MEDIA) == BGE_MEDIA_FIBER) {
3784 		if (BGE_IS_5705_PLUS(sc)) {
3785 			sc->bge_flags |= BGE_FLAG_MII_SERDES;
3786 			sc->bge_phy_flags |= BGE_PHY_NO_WIRESPEED;
3787 		} else
3788 			sc->bge_flags |= BGE_FLAG_TBI;
3789 	}
3790 
3791 	/* Set various PHY bug flags. */
3792 	if (sc->bge_chipid == BGE_CHIPID_BCM5701_A0 ||
3793 	    sc->bge_chipid == BGE_CHIPID_BCM5701_B0)
3794 		sc->bge_phy_flags |= BGE_PHY_CRC_BUG;
3795 	if (sc->bge_chiprev == BGE_CHIPREV_5703_AX ||
3796 	    sc->bge_chiprev == BGE_CHIPREV_5704_AX)
3797 		sc->bge_phy_flags |= BGE_PHY_ADC_BUG;
3798 	if (sc->bge_chipid == BGE_CHIPID_BCM5704_A0)
3799 		sc->bge_phy_flags |= BGE_PHY_5704_A0_BUG;
3800 	if (pci_get_subvendor(dev) == DELL_VENDORID)
3801 		sc->bge_phy_flags |= BGE_PHY_NO_3LED;
3802 	if ((BGE_IS_5705_PLUS(sc)) &&
3803 	    sc->bge_asicrev != BGE_ASICREV_BCM5906 &&
3804 	    sc->bge_asicrev != BGE_ASICREV_BCM5785 &&
3805 	    sc->bge_asicrev != BGE_ASICREV_BCM57780 &&
3806 	    !BGE_IS_5717_PLUS(sc)) {
3807 		if (sc->bge_asicrev == BGE_ASICREV_BCM5755 ||
3808 		    sc->bge_asicrev == BGE_ASICREV_BCM5761 ||
3809 		    sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
3810 		    sc->bge_asicrev == BGE_ASICREV_BCM5787) {
3811 			if (pci_get_device(dev) != BCOM_DEVICEID_BCM5722 &&
3812 			    pci_get_device(dev) != BCOM_DEVICEID_BCM5756)
3813 				sc->bge_phy_flags |= BGE_PHY_JITTER_BUG;
3814 			if (pci_get_device(dev) == BCOM_DEVICEID_BCM5755M)
3815 				sc->bge_phy_flags |= BGE_PHY_ADJUST_TRIM;
3816 		} else
3817 			sc->bge_phy_flags |= BGE_PHY_BER_BUG;
3818 	}
3819 
3820 	/*
3821 	 * Don't enable Ethernet@WireSpeed for the 5700 or the
3822 	 * 5705 A0 and A1 chips.
3823 	 */
3824 	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 ||
3825 	    (sc->bge_asicrev == BGE_ASICREV_BCM5705 &&
3826 	    (sc->bge_chipid != BGE_CHIPID_BCM5705_A0 &&
3827 	    sc->bge_chipid != BGE_CHIPID_BCM5705_A1)))
3828 		sc->bge_phy_flags |= BGE_PHY_NO_WIRESPEED;
3829 
3830 	if (sc->bge_flags & BGE_FLAG_TBI) {
3831 		ifmedia_init(&sc->bge_ifmedia, IFM_IMASK, bge_ifmedia_upd,
3832 		    bge_ifmedia_sts);
3833 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER | IFM_1000_SX, 0, NULL);
3834 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER | IFM_1000_SX | IFM_FDX,
3835 		    0, NULL);
3836 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER | IFM_AUTO, 0, NULL);
3837 		ifmedia_set(&sc->bge_ifmedia, IFM_ETHER | IFM_AUTO);
3838 		sc->bge_ifmedia.ifm_media = sc->bge_ifmedia.ifm_cur->ifm_media;
3839 	} else {
3840 		/*
3841 		 * Do transceiver setup and tell the firmware the
3842 		 * driver is down so we can try to get access the
3843 		 * probe if ASF is running.  Retry a couple of times
3844 		 * if we get a conflict with the ASF firmware accessing
3845 		 * the PHY.
3846 		 */
3847 		trys = 0;
3848 		BGE_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3849 again:
3850 		bge_asf_driver_up(sc);
3851 
3852 		error = mii_attach(dev, &sc->bge_miibus, ifp,
3853 		    (ifm_change_cb_t)bge_ifmedia_upd,
3854 		    (ifm_stat_cb_t)bge_ifmedia_sts, capmask, sc->bge_phy_addr,
3855 		    MII_OFFSET_ANY, MIIF_DOPAUSE);
3856 		if (error != 0) {
3857 			if (trys++ < 4) {
3858 				device_printf(sc->bge_dev, "Try again\n");
3859 				bge_miibus_writereg(sc->bge_dev,
3860 				    sc->bge_phy_addr, MII_BMCR, BMCR_RESET);
3861 				goto again;
3862 			}
3863 			device_printf(sc->bge_dev, "attaching PHYs failed\n");
3864 			goto fail;
3865 		}
3866 
3867 		/*
3868 		 * Now tell the firmware we are going up after probing the PHY
3869 		 */
3870 		if (sc->bge_asf_mode & ASF_STACKUP)
3871 			BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3872 	}
3873 
3874 	/*
3875 	 * When using the BCM5701 in PCI-X mode, data corruption has
3876 	 * been observed in the first few bytes of some received packets.
3877 	 * Aligning the packet buffer in memory eliminates the corruption.
3878 	 * Unfortunately, this misaligns the packet payloads.  On platforms
3879 	 * which do not support unaligned accesses, we will realign the
3880 	 * payloads by copying the received packets.
3881 	 */
3882 	if (sc->bge_asicrev == BGE_ASICREV_BCM5701 &&
3883 	    sc->bge_flags & BGE_FLAG_PCIX)
3884                 sc->bge_flags |= BGE_FLAG_RX_ALIGNBUG;
3885 
3886 	/*
3887 	 * Call MI attach routine.
3888 	 */
3889 	ether_ifattach(ifp, eaddr);
3890 
3891 	/* Tell upper layer we support long frames. */
3892 	if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
3893 
3894 	/*
3895 	 * Hookup IRQ last.
3896 	 */
3897 	if (BGE_IS_5755_PLUS(sc) && sc->bge_flags & BGE_FLAG_MSI) {
3898 		/* Take advantage of single-shot MSI. */
3899 		CSR_WRITE_4(sc, BGE_MSI_MODE, CSR_READ_4(sc, BGE_MSI_MODE) &
3900 		    ~BGE_MSIMODE_ONE_SHOT_DISABLE);
3901 		sc->bge_tq = taskqueue_create_fast("bge_taskq", M_WAITOK,
3902 		    taskqueue_thread_enqueue, &sc->bge_tq);
3903 		if (sc->bge_tq == NULL) {
3904 			device_printf(dev, "could not create taskqueue.\n");
3905 			ether_ifdetach(ifp);
3906 			error = ENOMEM;
3907 			goto fail;
3908 		}
3909 		error = taskqueue_start_threads(&sc->bge_tq, 1, PI_NET,
3910 		    "%s taskq", device_get_nameunit(sc->bge_dev));
3911 		if (error != 0) {
3912 			device_printf(dev, "could not start threads.\n");
3913 			ether_ifdetach(ifp);
3914 			goto fail;
3915 		}
3916 		error = bus_setup_intr(dev, sc->bge_irq,
3917 		    INTR_TYPE_NET | INTR_MPSAFE, bge_msi_intr, NULL, sc,
3918 		    &sc->bge_intrhand);
3919 	} else
3920 		error = bus_setup_intr(dev, sc->bge_irq,
3921 		    INTR_TYPE_NET | INTR_MPSAFE, NULL, bge_intr, sc,
3922 		    &sc->bge_intrhand);
3923 
3924 	if (error) {
3925 		ether_ifdetach(ifp);
3926 		device_printf(sc->bge_dev, "couldn't set up irq\n");
3927 		goto fail;
3928 	}
3929 
3930 	/* Attach driver debugnet methods. */
3931 	DEBUGNET_SET(ifp, bge);
3932 
3933 fail:
3934 	if (error)
3935 		bge_detach(dev);
3936 	return (error);
3937 }
3938 
3939 static int
3940 bge_detach(device_t dev)
3941 {
3942 	struct bge_softc *sc;
3943 	if_t ifp;
3944 
3945 	sc = device_get_softc(dev);
3946 	ifp = sc->bge_ifp;
3947 
3948 #ifdef DEVICE_POLLING
3949 	if (if_getcapenable(ifp) & IFCAP_POLLING)
3950 		ether_poll_deregister(ifp);
3951 #endif
3952 
3953 	if (device_is_attached(dev)) {
3954 		ether_ifdetach(ifp);
3955 		BGE_LOCK(sc);
3956 		bge_stop(sc);
3957 		BGE_UNLOCK(sc);
3958 		callout_drain(&sc->bge_stat_ch);
3959 	}
3960 
3961 	if (sc->bge_tq)
3962 		taskqueue_drain(sc->bge_tq, &sc->bge_intr_task);
3963 
3964 	if (sc->bge_flags & BGE_FLAG_TBI)
3965 		ifmedia_removeall(&sc->bge_ifmedia);
3966 	else if (sc->bge_miibus != NULL) {
3967 		bus_generic_detach(dev);
3968 		device_delete_child(dev, sc->bge_miibus);
3969 	}
3970 
3971 	bge_release_resources(sc);
3972 
3973 	return (0);
3974 }
3975 
3976 static void
3977 bge_release_resources(struct bge_softc *sc)
3978 {
3979 	device_t dev;
3980 
3981 	dev = sc->bge_dev;
3982 
3983 	if (sc->bge_tq != NULL)
3984 		taskqueue_free(sc->bge_tq);
3985 
3986 	if (sc->bge_intrhand != NULL)
3987 		bus_teardown_intr(dev, sc->bge_irq, sc->bge_intrhand);
3988 
3989 	if (sc->bge_irq != NULL) {
3990 		bus_release_resource(dev, SYS_RES_IRQ,
3991 		    rman_get_rid(sc->bge_irq), sc->bge_irq);
3992 		pci_release_msi(dev);
3993 	}
3994 
3995 	if (sc->bge_res != NULL)
3996 		bus_release_resource(dev, SYS_RES_MEMORY,
3997 		    rman_get_rid(sc->bge_res), sc->bge_res);
3998 
3999 	if (sc->bge_res2 != NULL)
4000 		bus_release_resource(dev, SYS_RES_MEMORY,
4001 		    rman_get_rid(sc->bge_res2), sc->bge_res2);
4002 
4003 	if (sc->bge_ifp != NULL)
4004 		if_free(sc->bge_ifp);
4005 
4006 	bge_dma_free(sc);
4007 
4008 	if (mtx_initialized(&sc->bge_mtx))	/* XXX */
4009 		BGE_LOCK_DESTROY(sc);
4010 }
4011 
4012 static int
4013 bge_reset(struct bge_softc *sc)
4014 {
4015 	device_t dev;
4016 	uint32_t cachesize, command, mac_mode, mac_mode_mask, reset, val;
4017 	void (*write_op)(struct bge_softc *, int, int);
4018 	uint16_t devctl;
4019 	int i;
4020 
4021 	dev = sc->bge_dev;
4022 
4023 	mac_mode_mask = BGE_MACMODE_HALF_DUPLEX | BGE_MACMODE_PORTMODE;
4024 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) != 0)
4025 		mac_mode_mask |= BGE_MACMODE_APE_RX_EN | BGE_MACMODE_APE_TX_EN;
4026 	mac_mode = CSR_READ_4(sc, BGE_MAC_MODE) & mac_mode_mask;
4027 
4028 	if (BGE_IS_575X_PLUS(sc) && !BGE_IS_5714_FAMILY(sc) &&
4029 	    (sc->bge_asicrev != BGE_ASICREV_BCM5906)) {
4030 		if (sc->bge_flags & BGE_FLAG_PCIE)
4031 			write_op = bge_writemem_direct;
4032 		else
4033 			write_op = bge_writemem_ind;
4034 	} else
4035 		write_op = bge_writereg_ind;
4036 
4037 	if (sc->bge_asicrev != BGE_ASICREV_BCM5700 &&
4038 	    sc->bge_asicrev != BGE_ASICREV_BCM5701) {
4039 		CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_SET1);
4040 		for (i = 0; i < 8000; i++) {
4041 			if (CSR_READ_4(sc, BGE_NVRAM_SWARB) &
4042 			    BGE_NVRAMSWARB_GNT1)
4043 				break;
4044 			DELAY(20);
4045 		}
4046 		if (i == 8000) {
4047 			if (bootverbose)
4048 				device_printf(dev, "NVRAM lock timedout!\n");
4049 		}
4050 	}
4051 	/* Take APE lock when performing reset. */
4052 	bge_ape_lock(sc, BGE_APE_LOCK_GRC);
4053 
4054 	/* Save some important PCI state. */
4055 	cachesize = pci_read_config(dev, BGE_PCI_CACHESZ, 4);
4056 	command = pci_read_config(dev, BGE_PCI_CMD, 4);
4057 
4058 	pci_write_config(dev, BGE_PCI_MISC_CTL,
4059 	    BGE_PCIMISCCTL_INDIRECT_ACCESS | BGE_PCIMISCCTL_MASK_PCI_INTR |
4060 	    BGE_HIF_SWAP_OPTIONS | BGE_PCIMISCCTL_PCISTATE_RW, 4);
4061 
4062 	/* Disable fastboot on controllers that support it. */
4063 	if (sc->bge_asicrev == BGE_ASICREV_BCM5752 ||
4064 	    BGE_IS_5755_PLUS(sc)) {
4065 		if (bootverbose)
4066 			device_printf(dev, "Disabling fastboot\n");
4067 		CSR_WRITE_4(sc, BGE_FASTBOOT_PC, 0x0);
4068 	}
4069 
4070 	/*
4071 	 * Write the magic number to SRAM at offset 0xB50.
4072 	 * When firmware finishes its initialization it will
4073 	 * write ~BGE_SRAM_FW_MB_MAGIC to the same location.
4074 	 */
4075 	bge_writemem_ind(sc, BGE_SRAM_FW_MB, BGE_SRAM_FW_MB_MAGIC);
4076 
4077 	reset = BGE_MISCCFG_RESET_CORE_CLOCKS | BGE_32BITTIME_66MHZ;
4078 
4079 	/* XXX: Broadcom Linux driver. */
4080 	if (sc->bge_flags & BGE_FLAG_PCIE) {
4081 		if (sc->bge_asicrev != BGE_ASICREV_BCM5785 &&
4082 		    (sc->bge_flags & BGE_FLAG_5717_PLUS) == 0) {
4083 			if (CSR_READ_4(sc, 0x7E2C) == 0x60)	/* PCIE 1.0 */
4084 				CSR_WRITE_4(sc, 0x7E2C, 0x20);
4085 		}
4086 		if (sc->bge_chipid != BGE_CHIPID_BCM5750_A0) {
4087 			/* Prevent PCIE link training during global reset */
4088 			CSR_WRITE_4(sc, BGE_MISC_CFG, 1 << 29);
4089 			reset |= 1 << 29;
4090 		}
4091 	}
4092 
4093 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
4094 		val = CSR_READ_4(sc, BGE_VCPU_STATUS);
4095 		CSR_WRITE_4(sc, BGE_VCPU_STATUS,
4096 		    val | BGE_VCPU_STATUS_DRV_RESET);
4097 		val = CSR_READ_4(sc, BGE_VCPU_EXT_CTRL);
4098 		CSR_WRITE_4(sc, BGE_VCPU_EXT_CTRL,
4099 		    val & ~BGE_VCPU_EXT_CTRL_HALT_CPU);
4100 	}
4101 
4102 	/*
4103 	 * Set GPHY Power Down Override to leave GPHY
4104 	 * powered up in D0 uninitialized.
4105 	 */
4106 	if (BGE_IS_5705_PLUS(sc) &&
4107 	    (sc->bge_flags & BGE_FLAG_CPMU_PRESENT) == 0)
4108 		reset |= BGE_MISCCFG_GPHY_PD_OVERRIDE;
4109 
4110 	/* Issue global reset */
4111 	write_op(sc, BGE_MISC_CFG, reset);
4112 
4113 	if (sc->bge_flags & BGE_FLAG_PCIE)
4114 		DELAY(100 * 1000);
4115 	else
4116 		DELAY(1000);
4117 
4118 	/* XXX: Broadcom Linux driver. */
4119 	if (sc->bge_flags & BGE_FLAG_PCIE) {
4120 		if (sc->bge_chipid == BGE_CHIPID_BCM5750_A0) {
4121 			DELAY(500000); /* wait for link training to complete */
4122 			val = pci_read_config(dev, 0xC4, 4);
4123 			pci_write_config(dev, 0xC4, val | (1 << 15), 4);
4124 		}
4125 		devctl = pci_read_config(dev,
4126 		    sc->bge_expcap + PCIER_DEVICE_CTL, 2);
4127 		/* Clear enable no snoop and disable relaxed ordering. */
4128 		devctl &= ~(PCIEM_CTL_RELAXED_ORD_ENABLE |
4129 		    PCIEM_CTL_NOSNOOP_ENABLE);
4130 		pci_write_config(dev, sc->bge_expcap + PCIER_DEVICE_CTL,
4131 		    devctl, 2);
4132 		pci_set_max_read_req(dev, sc->bge_expmrq);
4133 		/* Clear error status. */
4134 		pci_write_config(dev, sc->bge_expcap + PCIER_DEVICE_STA,
4135 		    PCIEM_STA_CORRECTABLE_ERROR |
4136 		    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
4137 		    PCIEM_STA_UNSUPPORTED_REQ, 2);
4138 	}
4139 
4140 	/* Reset some of the PCI state that got zapped by reset. */
4141 	pci_write_config(dev, BGE_PCI_MISC_CTL,
4142 	    BGE_PCIMISCCTL_INDIRECT_ACCESS | BGE_PCIMISCCTL_MASK_PCI_INTR |
4143 	    BGE_HIF_SWAP_OPTIONS | BGE_PCIMISCCTL_PCISTATE_RW, 4);
4144 	val = BGE_PCISTATE_ROM_ENABLE | BGE_PCISTATE_ROM_RETRY_ENABLE;
4145 	if (sc->bge_chipid == BGE_CHIPID_BCM5704_A0 &&
4146 	    (sc->bge_flags & BGE_FLAG_PCIX) != 0)
4147 		val |= BGE_PCISTATE_RETRY_SAME_DMA;
4148 	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) != 0)
4149 		val |= BGE_PCISTATE_ALLOW_APE_CTLSPC_WR |
4150 		    BGE_PCISTATE_ALLOW_APE_SHMEM_WR |
4151 		    BGE_PCISTATE_ALLOW_APE_PSPACE_WR;
4152 	pci_write_config(dev, BGE_PCI_PCISTATE, val, 4);
4153 	pci_write_config(dev, BGE_PCI_CACHESZ, cachesize, 4);
4154 	pci_write_config(dev, BGE_PCI_CMD, command, 4);
4155 	/*
4156 	 * Disable PCI-X relaxed ordering to ensure status block update
4157 	 * comes first then packet buffer DMA. Otherwise driver may
4158 	 * read stale status block.
4159 	 */
4160 	if (sc->bge_flags & BGE_FLAG_PCIX) {
4161 		devctl = pci_read_config(dev,
4162 		    sc->bge_pcixcap + PCIXR_COMMAND, 2);
4163 		devctl &= ~PCIXM_COMMAND_ERO;
4164 		if (sc->bge_asicrev == BGE_ASICREV_BCM5703) {
4165 			devctl &= ~PCIXM_COMMAND_MAX_READ;
4166 			devctl |= PCIXM_COMMAND_MAX_READ_2048;
4167 		} else if (sc->bge_asicrev == BGE_ASICREV_BCM5704) {
4168 			devctl &= ~(PCIXM_COMMAND_MAX_SPLITS |
4169 			    PCIXM_COMMAND_MAX_READ);
4170 			devctl |= PCIXM_COMMAND_MAX_READ_2048;
4171 		}
4172 		pci_write_config(dev, sc->bge_pcixcap + PCIXR_COMMAND,
4173 		    devctl, 2);
4174 	}
4175 	/* Re-enable MSI, if necessary, and enable the memory arbiter. */
4176 	if (BGE_IS_5714_FAMILY(sc)) {
4177 		/* This chip disables MSI on reset. */
4178 		if (sc->bge_flags & BGE_FLAG_MSI) {
4179 			val = pci_read_config(dev,
4180 			    sc->bge_msicap + PCIR_MSI_CTRL, 2);
4181 			pci_write_config(dev,
4182 			    sc->bge_msicap + PCIR_MSI_CTRL,
4183 			    val | PCIM_MSICTRL_MSI_ENABLE, 2);
4184 			val = CSR_READ_4(sc, BGE_MSI_MODE);
4185 			CSR_WRITE_4(sc, BGE_MSI_MODE,
4186 			    val | BGE_MSIMODE_ENABLE);
4187 		}
4188 		val = CSR_READ_4(sc, BGE_MARB_MODE);
4189 		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE | val);
4190 	} else
4191 		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
4192 
4193 	/* Fix up byte swapping. */
4194 	CSR_WRITE_4(sc, BGE_MODE_CTL, bge_dma_swap_options(sc));
4195 
4196 	val = CSR_READ_4(sc, BGE_MAC_MODE);
4197 	val = (val & ~mac_mode_mask) | mac_mode;
4198 	CSR_WRITE_4(sc, BGE_MAC_MODE, val);
4199 	DELAY(40);
4200 
4201 	bge_ape_unlock(sc, BGE_APE_LOCK_GRC);
4202 
4203 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
4204 		for (i = 0; i < BGE_TIMEOUT; i++) {
4205 			val = CSR_READ_4(sc, BGE_VCPU_STATUS);
4206 			if (val & BGE_VCPU_STATUS_INIT_DONE)
4207 				break;
4208 			DELAY(100);
4209 		}
4210 		if (i == BGE_TIMEOUT) {
4211 			device_printf(dev, "reset timed out\n");
4212 			return (1);
4213 		}
4214 	} else {
4215 		/*
4216 		 * Poll until we see the 1's complement of the magic number.
4217 		 * This indicates that the firmware initialization is complete.
4218 		 * We expect this to fail if no chip containing the Ethernet
4219 		 * address is fitted though.
4220 		 */
4221 		for (i = 0; i < BGE_TIMEOUT; i++) {
4222 			DELAY(10);
4223 			val = bge_readmem_ind(sc, BGE_SRAM_FW_MB);
4224 			if (val == ~BGE_SRAM_FW_MB_MAGIC)
4225 				break;
4226 		}
4227 
4228 		if ((sc->bge_flags & BGE_FLAG_EADDR) && i == BGE_TIMEOUT)
4229 			device_printf(dev,
4230 			    "firmware handshake timed out, found 0x%08x\n",
4231 			    val);
4232 		/* BCM57765 A0 needs additional time before accessing. */
4233 		if (sc->bge_chipid == BGE_CHIPID_BCM57765_A0)
4234 			DELAY(10 * 1000);	/* XXX */
4235 	}
4236 
4237 	/*
4238 	 * The 5704 in TBI mode apparently needs some special
4239 	 * adjustment to insure the SERDES drive level is set
4240 	 * to 1.2V.
4241 	 */
4242 	if (sc->bge_asicrev == BGE_ASICREV_BCM5704 &&
4243 	    sc->bge_flags & BGE_FLAG_TBI) {
4244 		val = CSR_READ_4(sc, BGE_SERDES_CFG);
4245 		val = (val & ~0xFFF) | 0x880;
4246 		CSR_WRITE_4(sc, BGE_SERDES_CFG, val);
4247 	}
4248 
4249 	/* XXX: Broadcom Linux driver. */
4250 	if (sc->bge_flags & BGE_FLAG_PCIE &&
4251 	    !BGE_IS_5717_PLUS(sc) &&
4252 	    sc->bge_chipid != BGE_CHIPID_BCM5750_A0 &&
4253 	    sc->bge_asicrev != BGE_ASICREV_BCM5785) {
4254 		/* Enable Data FIFO protection. */
4255 		val = CSR_READ_4(sc, 0x7C00);
4256 		CSR_WRITE_4(sc, 0x7C00, val | (1 << 25));
4257 	}
4258 
4259 	if (sc->bge_asicrev == BGE_ASICREV_BCM5720)
4260 		BGE_CLRBIT(sc, BGE_CPMU_CLCK_ORIDE,
4261 		    CPMU_CLCK_ORIDE_MAC_ORIDE_EN);
4262 
4263 	return (0);
4264 }
4265 
4266 static __inline void
4267 bge_rxreuse_std(struct bge_softc *sc, int i)
4268 {
4269 	struct bge_rx_bd *r;
4270 
4271 	r = &sc->bge_ldata.bge_rx_std_ring[sc->bge_std];
4272 	r->bge_flags = BGE_RXBDFLAG_END;
4273 	r->bge_len = sc->bge_cdata.bge_rx_std_seglen[i];
4274 	r->bge_idx = i;
4275 	BGE_INC(sc->bge_std, BGE_STD_RX_RING_CNT);
4276 }
4277 
4278 static __inline void
4279 bge_rxreuse_jumbo(struct bge_softc *sc, int i)
4280 {
4281 	struct bge_extrx_bd *r;
4282 
4283 	r = &sc->bge_ldata.bge_rx_jumbo_ring[sc->bge_jumbo];
4284 	r->bge_flags = BGE_RXBDFLAG_JUMBO_RING | BGE_RXBDFLAG_END;
4285 	r->bge_len0 = sc->bge_cdata.bge_rx_jumbo_seglen[i][0];
4286 	r->bge_len1 = sc->bge_cdata.bge_rx_jumbo_seglen[i][1];
4287 	r->bge_len2 = sc->bge_cdata.bge_rx_jumbo_seglen[i][2];
4288 	r->bge_len3 = sc->bge_cdata.bge_rx_jumbo_seglen[i][3];
4289 	r->bge_idx = i;
4290 	BGE_INC(sc->bge_jumbo, BGE_JUMBO_RX_RING_CNT);
4291 }
4292 
4293 /*
4294  * Frame reception handling. This is called if there's a frame
4295  * on the receive return list.
4296  *
4297  * Note: we have to be able to handle two possibilities here:
4298  * 1) the frame is from the jumbo receive ring
4299  * 2) the frame is from the standard receive ring
4300  */
4301 
4302 static int
4303 bge_rxeof(struct bge_softc *sc, uint16_t rx_prod, int holdlck)
4304 {
4305 	if_t ifp;
4306 	int rx_npkts = 0, stdcnt = 0, jumbocnt = 0;
4307 	uint16_t rx_cons;
4308 
4309 	rx_cons = sc->bge_rx_saved_considx;
4310 
4311 	/* Nothing to do. */
4312 	if (rx_cons == rx_prod)
4313 		return (rx_npkts);
4314 
4315 	ifp = sc->bge_ifp;
4316 
4317 	bus_dmamap_sync(sc->bge_cdata.bge_rx_return_ring_tag,
4318 	    sc->bge_cdata.bge_rx_return_ring_map, BUS_DMASYNC_POSTREAD);
4319 	bus_dmamap_sync(sc->bge_cdata.bge_rx_std_ring_tag,
4320 	    sc->bge_cdata.bge_rx_std_ring_map, BUS_DMASYNC_POSTWRITE);
4321 	if (BGE_IS_JUMBO_CAPABLE(sc) &&
4322 	    if_getmtu(ifp) + ETHER_HDR_LEN + ETHER_CRC_LEN +
4323 	    ETHER_VLAN_ENCAP_LEN > (MCLBYTES - ETHER_ALIGN))
4324 		bus_dmamap_sync(sc->bge_cdata.bge_rx_jumbo_ring_tag,
4325 		    sc->bge_cdata.bge_rx_jumbo_ring_map, BUS_DMASYNC_POSTWRITE);
4326 
4327 	while (rx_cons != rx_prod) {
4328 		struct bge_rx_bd	*cur_rx;
4329 		uint32_t		rxidx;
4330 		struct mbuf		*m = NULL;
4331 		uint16_t		vlan_tag = 0;
4332 		int			have_tag = 0;
4333 
4334 #ifdef DEVICE_POLLING
4335 		if (if_getcapenable(ifp) & IFCAP_POLLING) {
4336 			if (sc->rxcycles <= 0)
4337 				break;
4338 			sc->rxcycles--;
4339 		}
4340 #endif
4341 
4342 		cur_rx = &sc->bge_ldata.bge_rx_return_ring[rx_cons];
4343 
4344 		rxidx = cur_rx->bge_idx;
4345 		BGE_INC(rx_cons, sc->bge_return_ring_cnt);
4346 
4347 		if (if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING &&
4348 		    cur_rx->bge_flags & BGE_RXBDFLAG_VLAN_TAG) {
4349 			have_tag = 1;
4350 			vlan_tag = cur_rx->bge_vlan_tag;
4351 		}
4352 
4353 		if (cur_rx->bge_flags & BGE_RXBDFLAG_JUMBO_RING) {
4354 			jumbocnt++;
4355 			m = sc->bge_cdata.bge_rx_jumbo_chain[rxidx];
4356 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
4357 				bge_rxreuse_jumbo(sc, rxidx);
4358 				continue;
4359 			}
4360 			if (bge_newbuf_jumbo(sc, rxidx) != 0) {
4361 				bge_rxreuse_jumbo(sc, rxidx);
4362 				if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
4363 				continue;
4364 			}
4365 			BGE_INC(sc->bge_jumbo, BGE_JUMBO_RX_RING_CNT);
4366 		} else {
4367 			stdcnt++;
4368 			m = sc->bge_cdata.bge_rx_std_chain[rxidx];
4369 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
4370 				bge_rxreuse_std(sc, rxidx);
4371 				continue;
4372 			}
4373 			if (bge_newbuf_std(sc, rxidx) != 0) {
4374 				bge_rxreuse_std(sc, rxidx);
4375 				if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
4376 				continue;
4377 			}
4378 			BGE_INC(sc->bge_std, BGE_STD_RX_RING_CNT);
4379 		}
4380 
4381 		if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
4382 #ifndef __NO_STRICT_ALIGNMENT
4383 		/*
4384 		 * For architectures with strict alignment we must make sure
4385 		 * the payload is aligned.
4386 		 */
4387 		if (sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) {
4388 			bcopy(m->m_data, m->m_data + ETHER_ALIGN,
4389 			    cur_rx->bge_len);
4390 			m->m_data += ETHER_ALIGN;
4391 		}
4392 #endif
4393 		m->m_pkthdr.len = m->m_len = cur_rx->bge_len - ETHER_CRC_LEN;
4394 		m->m_pkthdr.rcvif = ifp;
4395 
4396 		if (if_getcapenable(ifp) & IFCAP_RXCSUM)
4397 			bge_rxcsum(sc, cur_rx, m);
4398 
4399 		/*
4400 		 * If we received a packet with a vlan tag,
4401 		 * attach that information to the packet.
4402 		 */
4403 		if (have_tag) {
4404 			m->m_pkthdr.ether_vtag = vlan_tag;
4405 			m->m_flags |= M_VLANTAG;
4406 		}
4407 
4408 		if (holdlck != 0) {
4409 			BGE_UNLOCK(sc);
4410 			if_input(ifp, m);
4411 			BGE_LOCK(sc);
4412 		} else
4413 			if_input(ifp, m);
4414 		rx_npkts++;
4415 
4416 		if (!(if_getdrvflags(ifp) & IFF_DRV_RUNNING))
4417 			return (rx_npkts);
4418 	}
4419 
4420 	bus_dmamap_sync(sc->bge_cdata.bge_rx_return_ring_tag,
4421 	    sc->bge_cdata.bge_rx_return_ring_map, BUS_DMASYNC_PREREAD);
4422 	if (stdcnt > 0)
4423 		bus_dmamap_sync(sc->bge_cdata.bge_rx_std_ring_tag,
4424 		    sc->bge_cdata.bge_rx_std_ring_map, BUS_DMASYNC_PREWRITE);
4425 
4426 	if (jumbocnt > 0)
4427 		bus_dmamap_sync(sc->bge_cdata.bge_rx_jumbo_ring_tag,
4428 		    sc->bge_cdata.bge_rx_jumbo_ring_map, BUS_DMASYNC_PREWRITE);
4429 
4430 	sc->bge_rx_saved_considx = rx_cons;
4431 	bge_writembx(sc, BGE_MBX_RX_CONS0_LO, sc->bge_rx_saved_considx);
4432 	if (stdcnt)
4433 		bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, (sc->bge_std +
4434 		    BGE_STD_RX_RING_CNT - 1) % BGE_STD_RX_RING_CNT);
4435 	if (jumbocnt)
4436 		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, (sc->bge_jumbo +
4437 		    BGE_JUMBO_RX_RING_CNT - 1) % BGE_JUMBO_RX_RING_CNT);
4438 #ifdef notyet
4439 	/*
4440 	 * This register wraps very quickly under heavy packet drops.
4441 	 * If you need correct statistics, you can enable this check.
4442 	 */
4443 	if (BGE_IS_5705_PLUS(sc))
4444 		if_incierrors(ifp, CSR_READ_4(sc, BGE_RXLP_LOCSTAT_IFIN_DROPS));
4445 #endif
4446 	return (rx_npkts);
4447 }
4448 
4449 static void
4450 bge_rxcsum(struct bge_softc *sc, struct bge_rx_bd *cur_rx, struct mbuf *m)
4451 {
4452 
4453 	if (BGE_IS_5717_PLUS(sc)) {
4454 		if ((cur_rx->bge_flags & BGE_RXBDFLAG_IPV6) == 0) {
4455 			if (cur_rx->bge_flags & BGE_RXBDFLAG_IP_CSUM) {
4456 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
4457 				if ((cur_rx->bge_error_flag &
4458 				    BGE_RXERRFLAG_IP_CSUM_NOK) == 0)
4459 					m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
4460 			}
4461 			if (cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM) {
4462 				m->m_pkthdr.csum_data =
4463 				    cur_rx->bge_tcp_udp_csum;
4464 				m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
4465 				    CSUM_PSEUDO_HDR;
4466 			}
4467 		}
4468 	} else {
4469 		if (cur_rx->bge_flags & BGE_RXBDFLAG_IP_CSUM) {
4470 			m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
4471 			if ((cur_rx->bge_ip_csum ^ 0xFFFF) == 0)
4472 				m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
4473 		}
4474 		if (cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM &&
4475 		    m->m_pkthdr.len >= ETHER_MIN_NOPAD) {
4476 			m->m_pkthdr.csum_data =
4477 			    cur_rx->bge_tcp_udp_csum;
4478 			m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
4479 			    CSUM_PSEUDO_HDR;
4480 		}
4481 	}
4482 }
4483 
4484 static void
4485 bge_txeof(struct bge_softc *sc, uint16_t tx_cons)
4486 {
4487 	struct bge_tx_bd *cur_tx;
4488 	if_t ifp;
4489 
4490 	BGE_LOCK_ASSERT(sc);
4491 
4492 	/* Nothing to do. */
4493 	if (sc->bge_tx_saved_considx == tx_cons)
4494 		return;
4495 
4496 	ifp = sc->bge_ifp;
4497 
4498 	bus_dmamap_sync(sc->bge_cdata.bge_tx_ring_tag,
4499 	    sc->bge_cdata.bge_tx_ring_map, BUS_DMASYNC_POSTWRITE);
4500 	/*
4501 	 * Go through our tx ring and free mbufs for those
4502 	 * frames that have been sent.
4503 	 */
4504 	while (sc->bge_tx_saved_considx != tx_cons) {
4505 		uint32_t		idx;
4506 
4507 		idx = sc->bge_tx_saved_considx;
4508 		cur_tx = &sc->bge_ldata.bge_tx_ring[idx];
4509 		if (cur_tx->bge_flags & BGE_TXBDFLAG_END)
4510 			if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
4511 		if (sc->bge_cdata.bge_tx_chain[idx] != NULL) {
4512 			bus_dmamap_sync(sc->bge_cdata.bge_tx_mtag,
4513 			    sc->bge_cdata.bge_tx_dmamap[idx],
4514 			    BUS_DMASYNC_POSTWRITE);
4515 			bus_dmamap_unload(sc->bge_cdata.bge_tx_mtag,
4516 			    sc->bge_cdata.bge_tx_dmamap[idx]);
4517 			m_freem(sc->bge_cdata.bge_tx_chain[idx]);
4518 			sc->bge_cdata.bge_tx_chain[idx] = NULL;
4519 		}
4520 		sc->bge_txcnt--;
4521 		BGE_INC(sc->bge_tx_saved_considx, BGE_TX_RING_CNT);
4522 	}
4523 
4524 	if_setdrvflagbits(ifp, 0, IFF_DRV_OACTIVE);
4525 	if (sc->bge_txcnt == 0)
4526 		sc->bge_timer = 0;
4527 }
4528 
4529 #ifdef DEVICE_POLLING
4530 static int
4531 bge_poll(if_t ifp, enum poll_cmd cmd, int count)
4532 {
4533 	struct bge_softc *sc = if_getsoftc(ifp);
4534 	uint16_t rx_prod, tx_cons;
4535 	uint32_t statusword;
4536 	int rx_npkts = 0;
4537 
4538 	BGE_LOCK(sc);
4539 	if (!(if_getdrvflags(ifp) & IFF_DRV_RUNNING)) {
4540 		BGE_UNLOCK(sc);
4541 		return (rx_npkts);
4542 	}
4543 
4544 	bus_dmamap_sync(sc->bge_cdata.bge_status_tag,
4545 	    sc->bge_cdata.bge_status_map,
4546 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
4547 	/* Fetch updates from the status block. */
4548 	rx_prod = sc->bge_ldata.bge_status_block->bge_idx[0].bge_rx_prod_idx;
4549 	tx_cons = sc->bge_ldata.bge_status_block->bge_idx[0].bge_tx_cons_idx;
4550 
4551 	statusword = sc->bge_ldata.bge_status_block->bge_status;
4552 	/* Clear the status so the next pass only sees the changes. */
4553 	sc->bge_ldata.bge_status_block->bge_status = 0;
4554 
4555 	bus_dmamap_sync(sc->bge_cdata.bge_status_tag,
4556 	    sc->bge_cdata.bge_status_map,
4557 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
4558 
4559 	/* Note link event. It will be processed by POLL_AND_CHECK_STATUS. */
4560 	if (statusword & BGE_STATFLAG_LINKSTATE_CHANGED)
4561 		sc->bge_link_evt++;
4562 
4563 	if (cmd == POLL_AND_CHECK_STATUS)
4564 		if ((sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
4565 		    sc->bge_chipid != BGE_CHIPID_BCM5700_B2) ||
4566 		    sc->bge_link_evt || (sc->bge_flags & BGE_FLAG_TBI))
4567 			bge_link_upd(sc);
4568 
4569 	sc->rxcycles = count;
4570 	rx_npkts = bge_rxeof(sc, rx_prod, 1);
4571 	if (!(if_getdrvflags(ifp) & IFF_DRV_RUNNING)) {
4572 		BGE_UNLOCK(sc);
4573 		return (rx_npkts);
4574 	}
4575 	bge_txeof(sc, tx_cons);
4576 	if (!if_sendq_empty(ifp))
4577 		bge_start_locked(ifp);
4578 
4579 	BGE_UNLOCK(sc);
4580 	return (rx_npkts);
4581 }
4582 #endif /* DEVICE_POLLING */
4583 
4584 static int
4585 bge_msi_intr(void *arg)
4586 {
4587 	struct bge_softc *sc;
4588 
4589 	sc = (struct bge_softc *)arg;
4590 	/*
4591 	 * This interrupt is not shared and controller already
4592 	 * disabled further interrupt.
4593 	 */
4594 	taskqueue_enqueue(sc->bge_tq, &sc->bge_intr_task);
4595 	return (FILTER_HANDLED);
4596 }
4597 
4598 static void
4599 bge_intr_task(void *arg, int pending)
4600 {
4601 	struct bge_softc *sc;
4602 	if_t ifp;
4603 	uint32_t status, status_tag;
4604 	uint16_t rx_prod, tx_cons;
4605 
4606 	sc = (struct bge_softc *)arg;
4607 	ifp = sc->bge_ifp;
4608 
4609 	BGE_LOCK(sc);
4610 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
4611 		BGE_UNLOCK(sc);
4612 		return;
4613 	}
4614 
4615 	/* Get updated status block. */
4616 	bus_dmamap_sync(sc->bge_cdata.bge_status_tag,
4617 	    sc->bge_cdata.bge_status_map,
4618 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
4619 
4620 	/* Save producer/consumer indices. */
4621 	rx_prod = sc->bge_ldata.bge_status_block->bge_idx[0].bge_rx_prod_idx;
4622 	tx_cons = sc->bge_ldata.bge_status_block->bge_idx[0].bge_tx_cons_idx;
4623 	status = sc->bge_ldata.bge_status_block->bge_status;
4624 	status_tag = sc->bge_ldata.bge_status_block->bge_status_tag << 24;
4625 	/* Dirty the status flag. */
4626 	sc->bge_ldata.bge_status_block->bge_status = 0;
4627 	bus_dmamap_sync(sc->bge_cdata.bge_status_tag,
4628 	    sc->bge_cdata.bge_status_map,
4629 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
4630 	if ((sc->bge_flags & BGE_FLAG_TAGGED_STATUS) == 0)
4631 		status_tag = 0;
4632 
4633 	if ((status & BGE_STATFLAG_LINKSTATE_CHANGED) != 0)
4634 		bge_link_upd(sc);
4635 
4636 	/* Let controller work. */
4637 	bge_writembx(sc, BGE_MBX_IRQ0_LO, status_tag);
4638 
4639 	if (if_getdrvflags(ifp) & IFF_DRV_RUNNING &&
4640 	    sc->bge_rx_saved_considx != rx_prod) {
4641 		/* Check RX return ring producer/consumer. */
4642 		BGE_UNLOCK(sc);
4643 		bge_rxeof(sc, rx_prod, 0);
4644 		BGE_LOCK(sc);
4645 	}
4646 	if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
4647 		/* Check TX ring producer/consumer. */
4648 		bge_txeof(sc, tx_cons);
4649 		if (!if_sendq_empty(ifp))
4650 			bge_start_locked(ifp);
4651 	}
4652 	BGE_UNLOCK(sc);
4653 }
4654 
4655 static void
4656 bge_intr(void *xsc)
4657 {
4658 	struct bge_softc *sc;
4659 	if_t ifp;
4660 	uint32_t statusword;
4661 	uint16_t rx_prod, tx_cons;
4662 
4663 	sc = xsc;
4664 
4665 	BGE_LOCK(sc);
4666 
4667 	ifp = sc->bge_ifp;
4668 
4669 #ifdef DEVICE_POLLING
4670 	if (if_getcapenable(ifp) & IFCAP_POLLING) {
4671 		BGE_UNLOCK(sc);
4672 		return;
4673 	}
4674 #endif
4675 
4676 	/*
4677 	 * Ack the interrupt by writing something to BGE_MBX_IRQ0_LO.  Don't
4678 	 * disable interrupts by writing nonzero like we used to, since with
4679 	 * our current organization this just gives complications and
4680 	 * pessimizations for re-enabling interrupts.  We used to have races
4681 	 * instead of the necessary complications.  Disabling interrupts
4682 	 * would just reduce the chance of a status update while we are
4683 	 * running (by switching to the interrupt-mode coalescence
4684 	 * parameters), but this chance is already very low so it is more
4685 	 * efficient to get another interrupt than prevent it.
4686 	 *
4687 	 * We do the ack first to ensure another interrupt if there is a
4688 	 * status update after the ack.  We don't check for the status
4689 	 * changing later because it is more efficient to get another
4690 	 * interrupt than prevent it, not quite as above (not checking is
4691 	 * a smaller optimization than not toggling the interrupt enable,
4692 	 * since checking doesn't involve PCI accesses and toggling require
4693 	 * the status check).  So toggling would probably be a pessimization
4694 	 * even with MSI.  It would only be needed for using a task queue.
4695 	 */
4696 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 0);
4697 
4698 	/*
4699 	 * Do the mandatory PCI flush as well as get the link status.
4700 	 */
4701 	statusword = CSR_READ_4(sc, BGE_MAC_STS) & BGE_MACSTAT_LINK_CHANGED;
4702 
4703 	/* Make sure the descriptor ring indexes are coherent. */
4704 	bus_dmamap_sync(sc->bge_cdata.bge_status_tag,
4705 	    sc->bge_cdata.bge_status_map,
4706 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
4707 	rx_prod = sc->bge_ldata.bge_status_block->bge_idx[0].bge_rx_prod_idx;
4708 	tx_cons = sc->bge_ldata.bge_status_block->bge_idx[0].bge_tx_cons_idx;
4709 	sc->bge_ldata.bge_status_block->bge_status = 0;
4710 	bus_dmamap_sync(sc->bge_cdata.bge_status_tag,
4711 	    sc->bge_cdata.bge_status_map,
4712 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
4713 
4714 	if ((sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
4715 	    sc->bge_chipid != BGE_CHIPID_BCM5700_B2) ||
4716 	    statusword || sc->bge_link_evt)
4717 		bge_link_upd(sc);
4718 
4719 	if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
4720 		/* Check RX return ring producer/consumer. */
4721 		bge_rxeof(sc, rx_prod, 1);
4722 	}
4723 
4724 	if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
4725 		/* Check TX ring producer/consumer. */
4726 		bge_txeof(sc, tx_cons);
4727 	}
4728 
4729 	if (if_getdrvflags(ifp) & IFF_DRV_RUNNING &&
4730 	    !if_sendq_empty(ifp))
4731 		bge_start_locked(ifp);
4732 
4733 	BGE_UNLOCK(sc);
4734 }
4735 
4736 static void
4737 bge_asf_driver_up(struct bge_softc *sc)
4738 {
4739 	if (sc->bge_asf_mode & ASF_STACKUP) {
4740 		/* Send ASF heartbeat aprox. every 2s */
4741 		if (sc->bge_asf_count)
4742 			sc->bge_asf_count --;
4743 		else {
4744 			sc->bge_asf_count = 2;
4745 			bge_writemem_ind(sc, BGE_SRAM_FW_CMD_MB,
4746 			    BGE_FW_CMD_DRV_ALIVE);
4747 			bge_writemem_ind(sc, BGE_SRAM_FW_CMD_LEN_MB, 4);
4748 			bge_writemem_ind(sc, BGE_SRAM_FW_CMD_DATA_MB,
4749 			    BGE_FW_HB_TIMEOUT_SEC);
4750 			CSR_WRITE_4(sc, BGE_RX_CPU_EVENT,
4751 			    CSR_READ_4(sc, BGE_RX_CPU_EVENT) |
4752 			    BGE_RX_CPU_DRV_EVENT);
4753 		}
4754 	}
4755 }
4756 
4757 static void
4758 bge_tick(void *xsc)
4759 {
4760 	struct bge_softc *sc = xsc;
4761 	struct mii_data *mii = NULL;
4762 
4763 	BGE_LOCK_ASSERT(sc);
4764 
4765 	/* Synchronize with possible callout reset/stop. */
4766 	if (callout_pending(&sc->bge_stat_ch) ||
4767 	    !callout_active(&sc->bge_stat_ch))
4768 		return;
4769 
4770 	if (BGE_IS_5705_PLUS(sc))
4771 		bge_stats_update_regs(sc);
4772 	else
4773 		bge_stats_update(sc);
4774 
4775 	/* XXX Add APE heartbeat check here? */
4776 
4777 	if ((sc->bge_flags & BGE_FLAG_TBI) == 0) {
4778 		mii = device_get_softc(sc->bge_miibus);
4779 		/*
4780 		 * Do not touch PHY if we have link up. This could break
4781 		 * IPMI/ASF mode or produce extra input errors
4782 		 * (extra errors was reported for bcm5701 & bcm5704).
4783 		 */
4784 		if (!sc->bge_link)
4785 			mii_tick(mii);
4786 	} else {
4787 		/*
4788 		 * Since in TBI mode auto-polling can't be used we should poll
4789 		 * link status manually. Here we register pending link event
4790 		 * and trigger interrupt.
4791 		 */
4792 #ifdef DEVICE_POLLING
4793 		/* In polling mode we poll link state in bge_poll(). */
4794 		if (!(if_getcapenable(sc->bge_ifp) & IFCAP_POLLING))
4795 #endif
4796 		{
4797 		sc->bge_link_evt++;
4798 		if (sc->bge_asicrev == BGE_ASICREV_BCM5700 ||
4799 		    sc->bge_flags & BGE_FLAG_5788)
4800 			BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
4801 		else
4802 			BGE_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
4803 		}
4804 	}
4805 
4806 	bge_asf_driver_up(sc);
4807 	bge_watchdog(sc);
4808 
4809 	callout_reset(&sc->bge_stat_ch, hz, bge_tick, sc);
4810 }
4811 
4812 static void
4813 bge_stats_update_regs(struct bge_softc *sc)
4814 {
4815 	if_t ifp;
4816 	struct bge_mac_stats *stats;
4817 	uint32_t val;
4818 
4819 	ifp = sc->bge_ifp;
4820 	stats = &sc->bge_mac_stats;
4821 
4822 	stats->ifHCOutOctets +=
4823 	    CSR_READ_4(sc, BGE_TX_MAC_STATS_OCTETS);
4824 	stats->etherStatsCollisions +=
4825 	    CSR_READ_4(sc, BGE_TX_MAC_STATS_COLLS);
4826 	stats->outXonSent +=
4827 	    CSR_READ_4(sc, BGE_TX_MAC_STATS_XON_SENT);
4828 	stats->outXoffSent +=
4829 	    CSR_READ_4(sc, BGE_TX_MAC_STATS_XOFF_SENT);
4830 	stats->dot3StatsInternalMacTransmitErrors +=
4831 	    CSR_READ_4(sc, BGE_TX_MAC_STATS_ERRORS);
4832 	stats->dot3StatsSingleCollisionFrames +=
4833 	    CSR_READ_4(sc, BGE_TX_MAC_STATS_SINGLE_COLL);
4834 	stats->dot3StatsMultipleCollisionFrames +=
4835 	    CSR_READ_4(sc, BGE_TX_MAC_STATS_MULTI_COLL);
4836 	stats->dot3StatsDeferredTransmissions +=
4837 	    CSR_READ_4(sc, BGE_TX_MAC_STATS_DEFERRED);
4838 	stats->dot3StatsExcessiveCollisions +=
4839 	    CSR_READ_4(sc, BGE_TX_MAC_STATS_EXCESS_COLL);
4840 	stats->dot3StatsLateCollisions +=
4841 	    CSR_READ_4(sc, BGE_TX_MAC_STATS_LATE_COLL);
4842 	stats->ifHCOutUcastPkts +=
4843 	    CSR_READ_4(sc, BGE_TX_MAC_STATS_UCAST);
4844 	stats->ifHCOutMulticastPkts +=
4845 	    CSR_READ_4(sc, BGE_TX_MAC_STATS_MCAST);
4846 	stats->ifHCOutBroadcastPkts +=
4847 	    CSR_READ_4(sc, BGE_TX_MAC_STATS_BCAST);
4848 
4849 	stats->ifHCInOctets +=
4850 	    CSR_READ_4(sc, BGE_RX_MAC_STATS_OCTESTS);
4851 	stats->etherStatsFragments +=
4852 	    CSR_READ_4(sc, BGE_RX_MAC_STATS_FRAGMENTS);
4853 	stats->ifHCInUcastPkts +=
4854 	    CSR_READ_4(sc, BGE_RX_MAC_STATS_UCAST);
4855 	stats->ifHCInMulticastPkts +=
4856 	    CSR_READ_4(sc, BGE_RX_MAC_STATS_MCAST);
4857 	stats->ifHCInBroadcastPkts +=
4858 	    CSR_READ_4(sc, BGE_RX_MAC_STATS_BCAST);
4859 	stats->dot3StatsFCSErrors +=
4860 	    CSR_READ_4(sc, BGE_RX_MAC_STATS_FCS_ERRORS);
4861 	stats->dot3StatsAlignmentErrors +=
4862 	    CSR_READ_4(sc, BGE_RX_MAC_STATS_ALGIN_ERRORS);
4863 	stats->xonPauseFramesReceived +=
4864 	    CSR_READ_4(sc, BGE_RX_MAC_STATS_XON_RCVD);
4865 	stats->xoffPauseFramesReceived +=
4866 	    CSR_READ_4(sc, BGE_RX_MAC_STATS_XOFF_RCVD);
4867 	stats->macControlFramesReceived +=
4868 	    CSR_READ_4(sc, BGE_RX_MAC_STATS_CTRL_RCVD);
4869 	stats->xoffStateEntered +=
4870 	    CSR_READ_4(sc, BGE_RX_MAC_STATS_XOFF_ENTERED);
4871 	stats->dot3StatsFramesTooLong +=
4872 	    CSR_READ_4(sc, BGE_RX_MAC_STATS_FRAME_TOO_LONG);
4873 	stats->etherStatsJabbers +=
4874 	    CSR_READ_4(sc, BGE_RX_MAC_STATS_JABBERS);
4875 	stats->etherStatsUndersizePkts +=
4876 	    CSR_READ_4(sc, BGE_RX_MAC_STATS_UNDERSIZE);
4877 
4878 	stats->FramesDroppedDueToFilters +=
4879 	    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_FILTDROP);
4880 	stats->DmaWriteQueueFull +=
4881 	    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_DMA_WRQ_FULL);
4882 	stats->DmaWriteHighPriQueueFull +=
4883 	    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_DMA_HPWRQ_FULL);
4884 	stats->NoMoreRxBDs +=
4885 	    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_OUT_OF_BDS);
4886 	/*
4887 	 * XXX
4888 	 * Unlike other controllers, BGE_RXLP_LOCSTAT_IFIN_DROPS
4889 	 * counter of BCM5717, BCM5718, BCM5719 A0 and BCM5720 A0
4890 	 * includes number of unwanted multicast frames.  This comes
4891 	 * from silicon bug and known workaround to get rough(not
4892 	 * exact) counter is to enable interrupt on MBUF low water
4893 	 * attention.  This can be accomplished by setting
4894 	 * BGE_HCCMODE_ATTN bit of BGE_HCC_MODE,
4895 	 * BGE_BMANMODE_LOMBUF_ATTN bit of BGE_BMAN_MODE and
4896 	 * BGE_MODECTL_FLOWCTL_ATTN_INTR bit of BGE_MODE_CTL.
4897 	 * However that change would generate more interrupts and
4898 	 * there are still possibilities of losing multiple frames
4899 	 * during BGE_MODECTL_FLOWCTL_ATTN_INTR interrupt handling.
4900 	 * Given that the workaround still would not get correct
4901 	 * counter I don't think it's worth to implement it.  So
4902 	 * ignore reading the counter on controllers that have the
4903 	 * silicon bug.
4904 	 */
4905 	if (sc->bge_asicrev != BGE_ASICREV_BCM5717 &&
4906 	    sc->bge_chipid != BGE_CHIPID_BCM5719_A0 &&
4907 	    sc->bge_chipid != BGE_CHIPID_BCM5720_A0)
4908 		stats->InputDiscards +=
4909 		    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_IFIN_DROPS);
4910 	stats->InputErrors +=
4911 	    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_IFIN_ERRORS);
4912 	stats->RecvThresholdHit +=
4913 	    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_RXTHRESH_HIT);
4914 
4915 	if (sc->bge_flags & BGE_FLAG_RDMA_BUG) {
4916 		/*
4917 		 * If controller transmitted more than BGE_NUM_RDMA_CHANNELS
4918 		 * frames, it's safe to disable workaround for DMA engine's
4919 		 * miscalculation of TXMBUF space.
4920 		 */
4921 		if (stats->ifHCOutUcastPkts + stats->ifHCOutMulticastPkts +
4922 		    stats->ifHCOutBroadcastPkts > BGE_NUM_RDMA_CHANNELS) {
4923 			val = CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL);
4924 			if (sc->bge_asicrev == BGE_ASICREV_BCM5719)
4925 				val &= ~BGE_RDMA_TX_LENGTH_WA_5719;
4926 			else
4927 				val &= ~BGE_RDMA_TX_LENGTH_WA_5720;
4928 			CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL, val);
4929 			sc->bge_flags &= ~BGE_FLAG_RDMA_BUG;
4930 		}
4931 	}
4932 }
4933 
4934 static void
4935 bge_stats_clear_regs(struct bge_softc *sc)
4936 {
4937 
4938 	CSR_READ_4(sc, BGE_TX_MAC_STATS_OCTETS);
4939 	CSR_READ_4(sc, BGE_TX_MAC_STATS_COLLS);
4940 	CSR_READ_4(sc, BGE_TX_MAC_STATS_XON_SENT);
4941 	CSR_READ_4(sc, BGE_TX_MAC_STATS_XOFF_SENT);
4942 	CSR_READ_4(sc, BGE_TX_MAC_STATS_ERRORS);
4943 	CSR_READ_4(sc, BGE_TX_MAC_STATS_SINGLE_COLL);
4944 	CSR_READ_4(sc, BGE_TX_MAC_STATS_MULTI_COLL);
4945 	CSR_READ_4(sc, BGE_TX_MAC_STATS_DEFERRED);
4946 	CSR_READ_4(sc, BGE_TX_MAC_STATS_EXCESS_COLL);
4947 	CSR_READ_4(sc, BGE_TX_MAC_STATS_LATE_COLL);
4948 	CSR_READ_4(sc, BGE_TX_MAC_STATS_UCAST);
4949 	CSR_READ_4(sc, BGE_TX_MAC_STATS_MCAST);
4950 	CSR_READ_4(sc, BGE_TX_MAC_STATS_BCAST);
4951 
4952 	CSR_READ_4(sc, BGE_RX_MAC_STATS_OCTESTS);
4953 	CSR_READ_4(sc, BGE_RX_MAC_STATS_FRAGMENTS);
4954 	CSR_READ_4(sc, BGE_RX_MAC_STATS_UCAST);
4955 	CSR_READ_4(sc, BGE_RX_MAC_STATS_MCAST);
4956 	CSR_READ_4(sc, BGE_RX_MAC_STATS_BCAST);
4957 	CSR_READ_4(sc, BGE_RX_MAC_STATS_FCS_ERRORS);
4958 	CSR_READ_4(sc, BGE_RX_MAC_STATS_ALGIN_ERRORS);
4959 	CSR_READ_4(sc, BGE_RX_MAC_STATS_XON_RCVD);
4960 	CSR_READ_4(sc, BGE_RX_MAC_STATS_XOFF_RCVD);
4961 	CSR_READ_4(sc, BGE_RX_MAC_STATS_CTRL_RCVD);
4962 	CSR_READ_4(sc, BGE_RX_MAC_STATS_XOFF_ENTERED);
4963 	CSR_READ_4(sc, BGE_RX_MAC_STATS_FRAME_TOO_LONG);
4964 	CSR_READ_4(sc, BGE_RX_MAC_STATS_JABBERS);
4965 	CSR_READ_4(sc, BGE_RX_MAC_STATS_UNDERSIZE);
4966 
4967 	CSR_READ_4(sc, BGE_RXLP_LOCSTAT_FILTDROP);
4968 	CSR_READ_4(sc, BGE_RXLP_LOCSTAT_DMA_WRQ_FULL);
4969 	CSR_READ_4(sc, BGE_RXLP_LOCSTAT_DMA_HPWRQ_FULL);
4970 	CSR_READ_4(sc, BGE_RXLP_LOCSTAT_OUT_OF_BDS);
4971 	CSR_READ_4(sc, BGE_RXLP_LOCSTAT_IFIN_DROPS);
4972 	CSR_READ_4(sc, BGE_RXLP_LOCSTAT_IFIN_ERRORS);
4973 	CSR_READ_4(sc, BGE_RXLP_LOCSTAT_RXTHRESH_HIT);
4974 }
4975 
4976 static void
4977 bge_stats_update(struct bge_softc *sc)
4978 {
4979 	if_t ifp;
4980 	bus_size_t stats;
4981 	uint32_t cnt;	/* current register value */
4982 
4983 	ifp = sc->bge_ifp;
4984 
4985 	stats = BGE_MEMWIN_START + BGE_STATS_BLOCK;
4986 
4987 #define	READ_STAT(sc, stats, stat) \
4988 	CSR_READ_4(sc, stats + offsetof(struct bge_stats, stat))
4989 
4990 	cnt = READ_STAT(sc, stats, txstats.etherStatsCollisions.bge_addr_lo);
4991 	if_inc_counter(ifp, IFCOUNTER_COLLISIONS, cnt - sc->bge_tx_collisions);
4992 	sc->bge_tx_collisions = cnt;
4993 
4994 	cnt = READ_STAT(sc, stats, nicNoMoreRxBDs.bge_addr_lo);
4995 	if_inc_counter(ifp, IFCOUNTER_IERRORS, cnt - sc->bge_rx_nobds);
4996 	sc->bge_rx_nobds = cnt;
4997 	cnt = READ_STAT(sc, stats, ifInErrors.bge_addr_lo);
4998 	if_inc_counter(ifp, IFCOUNTER_IERRORS, cnt - sc->bge_rx_inerrs);
4999 	sc->bge_rx_inerrs = cnt;
5000 	cnt = READ_STAT(sc, stats, ifInDiscards.bge_addr_lo);
5001 	if_inc_counter(ifp, IFCOUNTER_IERRORS, cnt - sc->bge_rx_discards);
5002 	sc->bge_rx_discards = cnt;
5003 
5004 	cnt = READ_STAT(sc, stats, txstats.ifOutDiscards.bge_addr_lo);
5005 	if_inc_counter(ifp, IFCOUNTER_OERRORS, cnt - sc->bge_tx_discards);
5006 	sc->bge_tx_discards = cnt;
5007 
5008 #undef	READ_STAT
5009 }
5010 
5011 /*
5012  * Pad outbound frame to ETHER_MIN_NOPAD for an unusual reason.
5013  * The bge hardware will pad out Tx runts to ETHER_MIN_NOPAD,
5014  * but when such padded frames employ the bge IP/TCP checksum offload,
5015  * the hardware checksum assist gives incorrect results (possibly
5016  * from incorporating its own padding into the UDP/TCP checksum; who knows).
5017  * If we pad such runts with zeros, the onboard checksum comes out correct.
5018  */
5019 static __inline int
5020 bge_cksum_pad(struct mbuf *m)
5021 {
5022 	int padlen = ETHER_MIN_NOPAD - m->m_pkthdr.len;
5023 	struct mbuf *last;
5024 
5025 	/* If there's only the packet-header and we can pad there, use it. */
5026 	if (m->m_pkthdr.len == m->m_len && M_WRITABLE(m) &&
5027 	    M_TRAILINGSPACE(m) >= padlen) {
5028 		last = m;
5029 	} else {
5030 		/*
5031 		 * Walk packet chain to find last mbuf. We will either
5032 		 * pad there, or append a new mbuf and pad it.
5033 		 */
5034 		for (last = m; last->m_next != NULL; last = last->m_next);
5035 		if (!(M_WRITABLE(last) && M_TRAILINGSPACE(last) >= padlen)) {
5036 			/* Allocate new empty mbuf, pad it. Compact later. */
5037 			struct mbuf *n;
5038 
5039 			MGET(n, M_NOWAIT, MT_DATA);
5040 			if (n == NULL)
5041 				return (ENOBUFS);
5042 			n->m_len = 0;
5043 			last->m_next = n;
5044 			last = n;
5045 		}
5046 	}
5047 
5048 	/* Now zero the pad area, to avoid the bge cksum-assist bug. */
5049 	memset(mtod(last, caddr_t) + last->m_len, 0, padlen);
5050 	last->m_len += padlen;
5051 	m->m_pkthdr.len += padlen;
5052 
5053 	return (0);
5054 }
5055 
5056 static struct mbuf *
5057 bge_check_short_dma(struct mbuf *m)
5058 {
5059 	struct mbuf *n;
5060 	int found;
5061 
5062 	/*
5063 	 * If device receive two back-to-back send BDs with less than
5064 	 * or equal to 8 total bytes then the device may hang.  The two
5065 	 * back-to-back send BDs must in the same frame for this failure
5066 	 * to occur.  Scan mbuf chains and see whether two back-to-back
5067 	 * send BDs are there. If this is the case, allocate new mbuf
5068 	 * and copy the frame to workaround the silicon bug.
5069 	 */
5070 	for (n = m, found = 0; n != NULL; n = n->m_next) {
5071 		if (n->m_len < 8) {
5072 			found++;
5073 			if (found > 1)
5074 				break;
5075 			continue;
5076 		}
5077 		found = 0;
5078 	}
5079 
5080 	if (found > 1) {
5081 		n = m_defrag(m, M_NOWAIT);
5082 		if (n == NULL)
5083 			m_freem(m);
5084 	} else
5085 		n = m;
5086 	return (n);
5087 }
5088 
5089 static struct mbuf *
5090 bge_setup_tso(struct bge_softc *sc, struct mbuf *m, uint16_t *mss,
5091     uint16_t *flags)
5092 {
5093 	struct ip *ip;
5094 	struct tcphdr *tcp;
5095 	struct mbuf *n;
5096 	uint16_t hlen;
5097 	uint32_t poff;
5098 
5099 	if (M_WRITABLE(m) == 0) {
5100 		/* Get a writable copy. */
5101 		n = m_dup(m, M_NOWAIT);
5102 		m_freem(m);
5103 		if (n == NULL)
5104 			return (NULL);
5105 		m = n;
5106 	}
5107 	m = m_pullup(m, sizeof(struct ether_header) + sizeof(struct ip));
5108 	if (m == NULL)
5109 		return (NULL);
5110 	ip = (struct ip *)(mtod(m, char *) + sizeof(struct ether_header));
5111 	poff = sizeof(struct ether_header) + (ip->ip_hl << 2);
5112 	m = m_pullup(m, poff + sizeof(struct tcphdr));
5113 	if (m == NULL)
5114 		return (NULL);
5115 	tcp = (struct tcphdr *)(mtod(m, char *) + poff);
5116 	m = m_pullup(m, poff + (tcp->th_off << 2));
5117 	if (m == NULL)
5118 		return (NULL);
5119 	/*
5120 	 * It seems controller doesn't modify IP length and TCP pseudo
5121 	 * checksum. These checksum computed by upper stack should be 0.
5122 	 */
5123 	*mss = m->m_pkthdr.tso_segsz;
5124 	ip = (struct ip *)(mtod(m, char *) + sizeof(struct ether_header));
5125 	ip->ip_sum = 0;
5126 	ip->ip_len = htons(*mss + (ip->ip_hl << 2) + (tcp->th_off << 2));
5127 	/* Clear pseudo checksum computed by TCP stack. */
5128 	tcp = (struct tcphdr *)(mtod(m, char *) + poff);
5129 	tcp->th_sum = 0;
5130 	/*
5131 	 * Broadcom controllers uses different descriptor format for
5132 	 * TSO depending on ASIC revision. Due to TSO-capable firmware
5133 	 * license issue and lower performance of firmware based TSO
5134 	 * we only support hardware based TSO.
5135 	 */
5136 	/* Calculate header length, incl. TCP/IP options, in 32 bit units. */
5137 	hlen = ((ip->ip_hl << 2) + (tcp->th_off << 2)) >> 2;
5138 	if (sc->bge_flags & BGE_FLAG_TSO3) {
5139 		/*
5140 		 * For BCM5717 and newer controllers, hardware based TSO
5141 		 * uses the 14 lower bits of the bge_mss field to store the
5142 		 * MSS and the upper 2 bits to store the lowest 2 bits of
5143 		 * the IP/TCP header length.  The upper 6 bits of the header
5144 		 * length are stored in the bge_flags[14:10,4] field.  Jumbo
5145 		 * frames are supported.
5146 		 */
5147 		*mss |= ((hlen & 0x3) << 14);
5148 		*flags |= ((hlen & 0xF8) << 7) | ((hlen & 0x4) << 2);
5149 	} else {
5150 		/*
5151 		 * For BCM5755 and newer controllers, hardware based TSO uses
5152 		 * the lower 11	bits to store the MSS and the upper 5 bits to
5153 		 * store the IP/TCP header length. Jumbo frames are not
5154 		 * supported.
5155 		 */
5156 		*mss |= (hlen << 11);
5157 	}
5158 	return (m);
5159 }
5160 
5161 /*
5162  * Encapsulate an mbuf chain in the tx ring  by coupling the mbuf data
5163  * pointers to descriptors.
5164  */
5165 static int
5166 bge_encap(struct bge_softc *sc, struct mbuf **m_head, uint32_t *txidx)
5167 {
5168 	bus_dma_segment_t	segs[BGE_NSEG_NEW];
5169 	bus_dmamap_t		map;
5170 	struct bge_tx_bd	*d;
5171 	struct mbuf		*m = *m_head;
5172 	uint32_t		idx = *txidx;
5173 	uint16_t		csum_flags, mss, vlan_tag;
5174 	int			nsegs, i, error;
5175 
5176 	csum_flags = 0;
5177 	mss = 0;
5178 	vlan_tag = 0;
5179 	if ((sc->bge_flags & BGE_FLAG_SHORT_DMA_BUG) != 0 &&
5180 	    m->m_next != NULL) {
5181 		*m_head = bge_check_short_dma(m);
5182 		if (*m_head == NULL)
5183 			return (ENOBUFS);
5184 		m = *m_head;
5185 	}
5186 	if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
5187 		*m_head = m = bge_setup_tso(sc, m, &mss, &csum_flags);
5188 		if (*m_head == NULL)
5189 			return (ENOBUFS);
5190 		csum_flags |= BGE_TXBDFLAG_CPU_PRE_DMA |
5191 		    BGE_TXBDFLAG_CPU_POST_DMA;
5192 	} else if ((m->m_pkthdr.csum_flags & sc->bge_csum_features) != 0) {
5193 		if (m->m_pkthdr.csum_flags & CSUM_IP)
5194 			csum_flags |= BGE_TXBDFLAG_IP_CSUM;
5195 		if (m->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP)) {
5196 			csum_flags |= BGE_TXBDFLAG_TCP_UDP_CSUM;
5197 			if (m->m_pkthdr.len < ETHER_MIN_NOPAD &&
5198 			    (error = bge_cksum_pad(m)) != 0) {
5199 				m_freem(m);
5200 				*m_head = NULL;
5201 				return (error);
5202 			}
5203 		}
5204 	}
5205 
5206 	if ((m->m_pkthdr.csum_flags & CSUM_TSO) == 0) {
5207 		if (sc->bge_flags & BGE_FLAG_JUMBO_FRAME &&
5208 		    m->m_pkthdr.len > ETHER_MAX_LEN)
5209 			csum_flags |= BGE_TXBDFLAG_JUMBO_FRAME;
5210 		if (sc->bge_forced_collapse > 0 &&
5211 		    (sc->bge_flags & BGE_FLAG_PCIE) != 0 && m->m_next != NULL) {
5212 			/*
5213 			 * Forcedly collapse mbuf chains to overcome hardware
5214 			 * limitation which only support a single outstanding
5215 			 * DMA read operation.
5216 			 */
5217 			if (sc->bge_forced_collapse == 1)
5218 				m = m_defrag(m, M_NOWAIT);
5219 			else
5220 				m = m_collapse(m, M_NOWAIT,
5221 				    sc->bge_forced_collapse);
5222 			if (m == NULL)
5223 				m = *m_head;
5224 			*m_head = m;
5225 		}
5226 	}
5227 
5228 	map = sc->bge_cdata.bge_tx_dmamap[idx];
5229 	error = bus_dmamap_load_mbuf_sg(sc->bge_cdata.bge_tx_mtag, map, m, segs,
5230 	    &nsegs, BUS_DMA_NOWAIT);
5231 	if (error == EFBIG) {
5232 		m = m_collapse(m, M_NOWAIT, BGE_NSEG_NEW);
5233 		if (m == NULL) {
5234 			m_freem(*m_head);
5235 			*m_head = NULL;
5236 			return (ENOBUFS);
5237 		}
5238 		*m_head = m;
5239 		error = bus_dmamap_load_mbuf_sg(sc->bge_cdata.bge_tx_mtag, map,
5240 		    m, segs, &nsegs, BUS_DMA_NOWAIT);
5241 		if (error) {
5242 			m_freem(m);
5243 			*m_head = NULL;
5244 			return (error);
5245 		}
5246 	} else if (error != 0)
5247 		return (error);
5248 
5249 	/* Check if we have enough free send BDs. */
5250 	if (sc->bge_txcnt + nsegs >= BGE_TX_RING_CNT) {
5251 		bus_dmamap_unload(sc->bge_cdata.bge_tx_mtag, map);
5252 		return (ENOBUFS);
5253 	}
5254 
5255 	bus_dmamap_sync(sc->bge_cdata.bge_tx_mtag, map, BUS_DMASYNC_PREWRITE);
5256 
5257 	if (m->m_flags & M_VLANTAG) {
5258 		csum_flags |= BGE_TXBDFLAG_VLAN_TAG;
5259 		vlan_tag = m->m_pkthdr.ether_vtag;
5260 	}
5261 
5262 	if (sc->bge_asicrev == BGE_ASICREV_BCM5762 &&
5263 	    (m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
5264 		/*
5265 		 * 5725 family of devices corrupts TSO packets when TSO DMA
5266 		 * buffers cross into regions which are within MSS bytes of
5267 		 * a 4GB boundary.  If we encounter the condition, drop the
5268 		 * packet.
5269 		 */
5270 		for (i = 0; ; i++) {
5271 			d = &sc->bge_ldata.bge_tx_ring[idx];
5272 			d->bge_addr.bge_addr_lo = BGE_ADDR_LO(segs[i].ds_addr);
5273 			d->bge_addr.bge_addr_hi = BGE_ADDR_HI(segs[i].ds_addr);
5274 			d->bge_len = segs[i].ds_len;
5275 			if (d->bge_addr.bge_addr_lo + segs[i].ds_len + mss <
5276 			    d->bge_addr.bge_addr_lo)
5277 				break;
5278 			d->bge_flags = csum_flags;
5279 			d->bge_vlan_tag = vlan_tag;
5280 			d->bge_mss = mss;
5281 			if (i == nsegs - 1)
5282 				break;
5283 			BGE_INC(idx, BGE_TX_RING_CNT);
5284 		}
5285 		if (i != nsegs - 1) {
5286 			bus_dmamap_sync(sc->bge_cdata.bge_tx_mtag, map,
5287 			    BUS_DMASYNC_POSTWRITE);
5288 			bus_dmamap_unload(sc->bge_cdata.bge_tx_mtag, map);
5289 			m_freem(*m_head);
5290 			*m_head = NULL;
5291 			return (EIO);
5292 		}
5293 	} else {
5294 		for (i = 0; ; i++) {
5295 			d = &sc->bge_ldata.bge_tx_ring[idx];
5296 			d->bge_addr.bge_addr_lo = BGE_ADDR_LO(segs[i].ds_addr);
5297 			d->bge_addr.bge_addr_hi = BGE_ADDR_HI(segs[i].ds_addr);
5298 			d->bge_len = segs[i].ds_len;
5299 			d->bge_flags = csum_flags;
5300 			d->bge_vlan_tag = vlan_tag;
5301 			d->bge_mss = mss;
5302 			if (i == nsegs - 1)
5303 				break;
5304 			BGE_INC(idx, BGE_TX_RING_CNT);
5305 		}
5306 	}
5307 
5308 	/* Mark the last segment as end of packet... */
5309 	d->bge_flags |= BGE_TXBDFLAG_END;
5310 
5311 	/*
5312 	 * Insure that the map for this transmission
5313 	 * is placed at the array index of the last descriptor
5314 	 * in this chain.
5315 	 */
5316 	sc->bge_cdata.bge_tx_dmamap[*txidx] = sc->bge_cdata.bge_tx_dmamap[idx];
5317 	sc->bge_cdata.bge_tx_dmamap[idx] = map;
5318 	sc->bge_cdata.bge_tx_chain[idx] = m;
5319 	sc->bge_txcnt += nsegs;
5320 
5321 	BGE_INC(idx, BGE_TX_RING_CNT);
5322 	*txidx = idx;
5323 
5324 	return (0);
5325 }
5326 
5327 /*
5328  * Main transmit routine. To avoid having to do mbuf copies, we put pointers
5329  * to the mbuf data regions directly in the transmit descriptors.
5330  */
5331 static void
5332 bge_start_locked(if_t ifp)
5333 {
5334 	struct bge_softc *sc;
5335 	struct mbuf *m_head;
5336 	uint32_t prodidx;
5337 	int count;
5338 
5339 	sc = if_getsoftc(ifp);
5340 	BGE_LOCK_ASSERT(sc);
5341 
5342 	if (!sc->bge_link ||
5343 	    (if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
5344 	    IFF_DRV_RUNNING)
5345 		return;
5346 
5347 	prodidx = sc->bge_tx_prodidx;
5348 
5349 	for (count = 0; !if_sendq_empty(ifp);) {
5350 		if (sc->bge_txcnt > BGE_TX_RING_CNT - 16) {
5351 			if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, 0);
5352 			break;
5353 		}
5354 		m_head = if_dequeue(ifp);
5355 		if (m_head == NULL)
5356 			break;
5357 
5358 		/*
5359 		 * Pack the data into the transmit ring. If we
5360 		 * don't have room, set the OACTIVE flag and wait
5361 		 * for the NIC to drain the ring.
5362 		 */
5363 		if (bge_encap(sc, &m_head, &prodidx)) {
5364 			if (m_head == NULL)
5365 				break;
5366 			if_sendq_prepend(ifp, m_head);
5367 			if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, 0);
5368 			break;
5369 		}
5370 		++count;
5371 
5372 		/*
5373 		 * If there's a BPF listener, bounce a copy of this frame
5374 		 * to him.
5375 		 */
5376 		if_bpfmtap(ifp, m_head);
5377 	}
5378 
5379 	if (count > 0)
5380 		bge_start_tx(sc, prodidx);
5381 }
5382 
5383 static void
5384 bge_start_tx(struct bge_softc *sc, uint32_t prodidx)
5385 {
5386 
5387 	bus_dmamap_sync(sc->bge_cdata.bge_tx_ring_tag,
5388 	    sc->bge_cdata.bge_tx_ring_map, BUS_DMASYNC_PREWRITE);
5389 	/* Transmit. */
5390 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
5391 	/* 5700 b2 errata */
5392 	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
5393 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
5394 
5395 	sc->bge_tx_prodidx = prodidx;
5396 
5397 	/* Set a timeout in case the chip goes out to lunch. */
5398 	sc->bge_timer = BGE_TX_TIMEOUT;
5399 }
5400 
5401 /*
5402  * Main transmit routine. To avoid having to do mbuf copies, we put pointers
5403  * to the mbuf data regions directly in the transmit descriptors.
5404  */
5405 static void
5406 bge_start(if_t ifp)
5407 {
5408 	struct bge_softc *sc;
5409 
5410 	sc = if_getsoftc(ifp);
5411 	BGE_LOCK(sc);
5412 	bge_start_locked(ifp);
5413 	BGE_UNLOCK(sc);
5414 }
5415 
5416 static void
5417 bge_init_locked(struct bge_softc *sc)
5418 {
5419 	if_t ifp;
5420 	uint16_t *m;
5421 	uint32_t mode;
5422 
5423 	BGE_LOCK_ASSERT(sc);
5424 
5425 	ifp = sc->bge_ifp;
5426 
5427 	if (if_getdrvflags(ifp) & IFF_DRV_RUNNING)
5428 		return;
5429 
5430 	/* Cancel pending I/O and flush buffers. */
5431 	bge_stop(sc);
5432 
5433 	bge_stop_fw(sc);
5434 	bge_sig_pre_reset(sc, BGE_RESET_START);
5435 	bge_reset(sc);
5436 	bge_sig_legacy(sc, BGE_RESET_START);
5437 	bge_sig_post_reset(sc, BGE_RESET_START);
5438 
5439 	bge_chipinit(sc);
5440 
5441 	/*
5442 	 * Init the various state machines, ring
5443 	 * control blocks and firmware.
5444 	 */
5445 	if (bge_blockinit(sc)) {
5446 		device_printf(sc->bge_dev, "initialization failure\n");
5447 		return;
5448 	}
5449 
5450 	ifp = sc->bge_ifp;
5451 
5452 	/* Specify MTU. */
5453 	CSR_WRITE_4(sc, BGE_RX_MTU, if_getmtu(ifp) +
5454 	    ETHER_HDR_LEN + ETHER_CRC_LEN +
5455 	    (if_getcapenable(ifp) & IFCAP_VLAN_MTU ? ETHER_VLAN_ENCAP_LEN : 0));
5456 
5457 	/* Load our MAC address. */
5458 	m = (uint16_t *)IF_LLADDR(sc->bge_ifp);
5459 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_LO, htons(m[0]));
5460 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_HI, (htons(m[1]) << 16) | htons(m[2]));
5461 
5462 	/* Program promiscuous mode. */
5463 	bge_setpromisc(sc);
5464 
5465 	/* Program multicast filter. */
5466 	bge_setmulti(sc);
5467 
5468 	/* Program VLAN tag stripping. */
5469 	bge_setvlan(sc);
5470 
5471 	/* Override UDP checksum offloading. */
5472 	if (sc->bge_forced_udpcsum == 0)
5473 		sc->bge_csum_features &= ~CSUM_UDP;
5474 	else
5475 		sc->bge_csum_features |= CSUM_UDP;
5476 	if (if_getcapabilities(ifp) & IFCAP_TXCSUM &&
5477 	    if_getcapenable(ifp) & IFCAP_TXCSUM) {
5478 		if_sethwassistbits(ifp, 0, (BGE_CSUM_FEATURES | CSUM_UDP));
5479 		if_sethwassistbits(ifp, sc->bge_csum_features, 0);
5480 	}
5481 
5482 	/* Init RX ring. */
5483 	if (bge_init_rx_ring_std(sc) != 0) {
5484 		device_printf(sc->bge_dev, "no memory for std Rx buffers.\n");
5485 		bge_stop(sc);
5486 		return;
5487 	}
5488 
5489 	/*
5490 	 * Workaround for a bug in 5705 ASIC rev A0. Poll the NIC's
5491 	 * memory to insure that the chip has in fact read the first
5492 	 * entry of the ring.
5493 	 */
5494 	if (sc->bge_chipid == BGE_CHIPID_BCM5705_A0) {
5495 		uint32_t		v, i;
5496 		for (i = 0; i < 10; i++) {
5497 			DELAY(20);
5498 			v = bge_readmem_ind(sc, BGE_STD_RX_RINGS + 8);
5499 			if (v == (MCLBYTES - ETHER_ALIGN))
5500 				break;
5501 		}
5502 		if (i == 10)
5503 			device_printf (sc->bge_dev,
5504 			    "5705 A0 chip failed to load RX ring\n");
5505 	}
5506 
5507 	/* Init jumbo RX ring. */
5508 	if (BGE_IS_JUMBO_CAPABLE(sc) &&
5509 	    if_getmtu(ifp) + ETHER_HDR_LEN + ETHER_CRC_LEN +
5510      	    ETHER_VLAN_ENCAP_LEN > (MCLBYTES - ETHER_ALIGN)) {
5511 		if (bge_init_rx_ring_jumbo(sc) != 0) {
5512 			device_printf(sc->bge_dev,
5513 			    "no memory for jumbo Rx buffers.\n");
5514 			bge_stop(sc);
5515 			return;
5516 		}
5517 	}
5518 
5519 	/* Init our RX return ring index. */
5520 	sc->bge_rx_saved_considx = 0;
5521 
5522 	/* Init our RX/TX stat counters. */
5523 	sc->bge_rx_discards = sc->bge_tx_discards = sc->bge_tx_collisions = 0;
5524 
5525 	/* Init TX ring. */
5526 	bge_init_tx_ring(sc);
5527 
5528 	/* Enable TX MAC state machine lockup fix. */
5529 	mode = CSR_READ_4(sc, BGE_TX_MODE);
5530 	if (BGE_IS_5755_PLUS(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5906)
5531 		mode |= BGE_TXMODE_MBUF_LOCKUP_FIX;
5532 	if (sc->bge_asicrev == BGE_ASICREV_BCM5720 ||
5533 	    sc->bge_asicrev == BGE_ASICREV_BCM5762) {
5534 		mode &= ~(BGE_TXMODE_JMB_FRM_LEN | BGE_TXMODE_CNT_DN_MODE);
5535 		mode |= CSR_READ_4(sc, BGE_TX_MODE) &
5536 		    (BGE_TXMODE_JMB_FRM_LEN | BGE_TXMODE_CNT_DN_MODE);
5537 	}
5538 	/* Turn on transmitter. */
5539 	CSR_WRITE_4(sc, BGE_TX_MODE, mode | BGE_TXMODE_ENABLE);
5540 	DELAY(100);
5541 
5542 	/* Turn on receiver. */
5543 	mode = CSR_READ_4(sc, BGE_RX_MODE);
5544 	if (BGE_IS_5755_PLUS(sc))
5545 		mode |= BGE_RXMODE_IPV6_ENABLE;
5546 	if (sc->bge_asicrev == BGE_ASICREV_BCM5762)
5547 		mode |= BGE_RXMODE_IPV4_FRAG_FIX;
5548 	CSR_WRITE_4(sc,BGE_RX_MODE, mode | BGE_RXMODE_ENABLE);
5549 	DELAY(10);
5550 
5551 	/*
5552 	 * Set the number of good frames to receive after RX MBUF
5553 	 * Low Watermark has been reached. After the RX MAC receives
5554 	 * this number of frames, it will drop subsequent incoming
5555 	 * frames until the MBUF High Watermark is reached.
5556 	 */
5557 	if (BGE_IS_57765_PLUS(sc))
5558 		CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 1);
5559 	else
5560 		CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 2);
5561 
5562 	/* Clear MAC statistics. */
5563 	if (BGE_IS_5705_PLUS(sc))
5564 		bge_stats_clear_regs(sc);
5565 
5566 	/* Tell firmware we're alive. */
5567 	BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
5568 
5569 #ifdef DEVICE_POLLING
5570 	/* Disable interrupts if we are polling. */
5571 	if (if_getcapenable(ifp) & IFCAP_POLLING) {
5572 		BGE_SETBIT(sc, BGE_PCI_MISC_CTL,
5573 		    BGE_PCIMISCCTL_MASK_PCI_INTR);
5574 		bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
5575 	} else
5576 #endif
5577 
5578 	/* Enable host interrupts. */
5579 	{
5580 	BGE_SETBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_CLEAR_INTA);
5581 	BGE_CLRBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR);
5582 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 0);
5583 	}
5584 
5585 	if_setdrvflagbits(ifp, IFF_DRV_RUNNING, 0);
5586 	if_setdrvflagbits(ifp, 0, IFF_DRV_OACTIVE);
5587 
5588 	bge_ifmedia_upd_locked(ifp);
5589 
5590 	callout_reset(&sc->bge_stat_ch, hz, bge_tick, sc);
5591 }
5592 
5593 static void
5594 bge_init(void *xsc)
5595 {
5596 	struct bge_softc *sc = xsc;
5597 
5598 	BGE_LOCK(sc);
5599 	bge_init_locked(sc);
5600 	BGE_UNLOCK(sc);
5601 }
5602 
5603 /*
5604  * Set media options.
5605  */
5606 static int
5607 bge_ifmedia_upd(if_t ifp)
5608 {
5609 	struct bge_softc *sc = if_getsoftc(ifp);
5610 	int res;
5611 
5612 	BGE_LOCK(sc);
5613 	res = bge_ifmedia_upd_locked(ifp);
5614 	BGE_UNLOCK(sc);
5615 
5616 	return (res);
5617 }
5618 
5619 static int
5620 bge_ifmedia_upd_locked(if_t ifp)
5621 {
5622 	struct bge_softc *sc = if_getsoftc(ifp);
5623 	struct mii_data *mii;
5624 	struct mii_softc *miisc;
5625 	struct ifmedia *ifm;
5626 
5627 	BGE_LOCK_ASSERT(sc);
5628 
5629 	ifm = &sc->bge_ifmedia;
5630 
5631 	/* If this is a 1000baseX NIC, enable the TBI port. */
5632 	if (sc->bge_flags & BGE_FLAG_TBI) {
5633 		if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
5634 			return (EINVAL);
5635 		switch(IFM_SUBTYPE(ifm->ifm_media)) {
5636 		case IFM_AUTO:
5637 			/*
5638 			 * The BCM5704 ASIC appears to have a special
5639 			 * mechanism for programming the autoneg
5640 			 * advertisement registers in TBI mode.
5641 			 */
5642 			if (sc->bge_asicrev == BGE_ASICREV_BCM5704) {
5643 				uint32_t sgdig;
5644 				sgdig = CSR_READ_4(sc, BGE_SGDIG_STS);
5645 				if (sgdig & BGE_SGDIGSTS_DONE) {
5646 					CSR_WRITE_4(sc, BGE_TX_TBI_AUTONEG, 0);
5647 					sgdig = CSR_READ_4(sc, BGE_SGDIG_CFG);
5648 					sgdig |= BGE_SGDIGCFG_AUTO |
5649 					    BGE_SGDIGCFG_PAUSE_CAP |
5650 					    BGE_SGDIGCFG_ASYM_PAUSE;
5651 					CSR_WRITE_4(sc, BGE_SGDIG_CFG,
5652 					    sgdig | BGE_SGDIGCFG_SEND);
5653 					DELAY(5);
5654 					CSR_WRITE_4(sc, BGE_SGDIG_CFG, sgdig);
5655 				}
5656 			}
5657 			break;
5658 		case IFM_1000_SX:
5659 			if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX) {
5660 				BGE_CLRBIT(sc, BGE_MAC_MODE,
5661 				    BGE_MACMODE_HALF_DUPLEX);
5662 			} else {
5663 				BGE_SETBIT(sc, BGE_MAC_MODE,
5664 				    BGE_MACMODE_HALF_DUPLEX);
5665 			}
5666 			DELAY(40);
5667 			break;
5668 		default:
5669 			return (EINVAL);
5670 		}
5671 		return (0);
5672 	}
5673 
5674 	sc->bge_link_evt++;
5675 	mii = device_get_softc(sc->bge_miibus);
5676 	LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
5677 		PHY_RESET(miisc);
5678 	mii_mediachg(mii);
5679 
5680 	/*
5681 	 * Force an interrupt so that we will call bge_link_upd
5682 	 * if needed and clear any pending link state attention.
5683 	 * Without this we are not getting any further interrupts
5684 	 * for link state changes and thus will not UP the link and
5685 	 * not be able to send in bge_start_locked. The only
5686 	 * way to get things working was to receive a packet and
5687 	 * get an RX intr.
5688 	 * bge_tick should help for fiber cards and we might not
5689 	 * need to do this here if BGE_FLAG_TBI is set but as
5690 	 * we poll for fiber anyway it should not harm.
5691 	 */
5692 	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 ||
5693 	    sc->bge_flags & BGE_FLAG_5788)
5694 		BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
5695 	else
5696 		BGE_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
5697 
5698 	return (0);
5699 }
5700 
5701 /*
5702  * Report current media status.
5703  */
5704 static void
5705 bge_ifmedia_sts(if_t ifp, struct ifmediareq *ifmr)
5706 {
5707 	struct bge_softc *sc = if_getsoftc(ifp);
5708 	struct mii_data *mii;
5709 
5710 	BGE_LOCK(sc);
5711 
5712 	if ((if_getflags(ifp) & IFF_UP) == 0) {
5713 		BGE_UNLOCK(sc);
5714 		return;
5715 	}
5716 	if (sc->bge_flags & BGE_FLAG_TBI) {
5717 		ifmr->ifm_status = IFM_AVALID;
5718 		ifmr->ifm_active = IFM_ETHER;
5719 		if (CSR_READ_4(sc, BGE_MAC_STS) &
5720 		    BGE_MACSTAT_TBI_PCS_SYNCHED)
5721 			ifmr->ifm_status |= IFM_ACTIVE;
5722 		else {
5723 			ifmr->ifm_active |= IFM_NONE;
5724 			BGE_UNLOCK(sc);
5725 			return;
5726 		}
5727 		ifmr->ifm_active |= IFM_1000_SX;
5728 		if (CSR_READ_4(sc, BGE_MAC_MODE) & BGE_MACMODE_HALF_DUPLEX)
5729 			ifmr->ifm_active |= IFM_HDX;
5730 		else
5731 			ifmr->ifm_active |= IFM_FDX;
5732 		BGE_UNLOCK(sc);
5733 		return;
5734 	}
5735 
5736 	mii = device_get_softc(sc->bge_miibus);
5737 	mii_pollstat(mii);
5738 	ifmr->ifm_active = mii->mii_media_active;
5739 	ifmr->ifm_status = mii->mii_media_status;
5740 
5741 	BGE_UNLOCK(sc);
5742 }
5743 
5744 static int
5745 bge_ioctl(if_t ifp, u_long command, caddr_t data)
5746 {
5747 	struct bge_softc *sc = if_getsoftc(ifp);
5748 	struct ifreq *ifr = (struct ifreq *) data;
5749 	struct mii_data *mii;
5750 	int flags, mask, error = 0;
5751 
5752 	switch (command) {
5753 	case SIOCSIFMTU:
5754 		if (BGE_IS_JUMBO_CAPABLE(sc) ||
5755 		    (sc->bge_flags & BGE_FLAG_JUMBO_STD)) {
5756 			if (ifr->ifr_mtu < ETHERMIN ||
5757 			    ifr->ifr_mtu > BGE_JUMBO_MTU) {
5758 				error = EINVAL;
5759 				break;
5760 			}
5761 		} else if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ETHERMTU) {
5762 			error = EINVAL;
5763 			break;
5764 		}
5765 		BGE_LOCK(sc);
5766 		if (if_getmtu(ifp) != ifr->ifr_mtu) {
5767 			if_setmtu(ifp, ifr->ifr_mtu);
5768 			if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
5769 				if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
5770 				bge_init_locked(sc);
5771 			}
5772 		}
5773 		BGE_UNLOCK(sc);
5774 		break;
5775 	case SIOCSIFFLAGS:
5776 		BGE_LOCK(sc);
5777 		if (if_getflags(ifp) & IFF_UP) {
5778 			/*
5779 			 * If only the state of the PROMISC flag changed,
5780 			 * then just use the 'set promisc mode' command
5781 			 * instead of reinitializing the entire NIC. Doing
5782 			 * a full re-init means reloading the firmware and
5783 			 * waiting for it to start up, which may take a
5784 			 * second or two.  Similarly for ALLMULTI.
5785 			 */
5786 			if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
5787 				flags = if_getflags(ifp) ^ sc->bge_if_flags;
5788 				if (flags & IFF_PROMISC)
5789 					bge_setpromisc(sc);
5790 				if (flags & IFF_ALLMULTI)
5791 					bge_setmulti(sc);
5792 			} else
5793 				bge_init_locked(sc);
5794 		} else {
5795 			if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
5796 				bge_stop(sc);
5797 			}
5798 		}
5799 		sc->bge_if_flags = if_getflags(ifp);
5800 		BGE_UNLOCK(sc);
5801 		error = 0;
5802 		break;
5803 	case SIOCADDMULTI:
5804 	case SIOCDELMULTI:
5805 		if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
5806 			BGE_LOCK(sc);
5807 			bge_setmulti(sc);
5808 			BGE_UNLOCK(sc);
5809 			error = 0;
5810 		}
5811 		break;
5812 	case SIOCSIFMEDIA:
5813 	case SIOCGIFMEDIA:
5814 		if (sc->bge_flags & BGE_FLAG_TBI) {
5815 			error = ifmedia_ioctl(ifp, ifr,
5816 			    &sc->bge_ifmedia, command);
5817 		} else {
5818 			mii = device_get_softc(sc->bge_miibus);
5819 			error = ifmedia_ioctl(ifp, ifr,
5820 			    &mii->mii_media, command);
5821 		}
5822 		break;
5823 	case SIOCSIFCAP:
5824 		mask = ifr->ifr_reqcap ^ if_getcapenable(ifp);
5825 #ifdef DEVICE_POLLING
5826 		if (mask & IFCAP_POLLING) {
5827 			if (ifr->ifr_reqcap & IFCAP_POLLING) {
5828 				error = ether_poll_register(bge_poll, ifp);
5829 				if (error)
5830 					return (error);
5831 				BGE_LOCK(sc);
5832 				BGE_SETBIT(sc, BGE_PCI_MISC_CTL,
5833 				    BGE_PCIMISCCTL_MASK_PCI_INTR);
5834 				bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
5835 				if_setcapenablebit(ifp, IFCAP_POLLING, 0);
5836 				BGE_UNLOCK(sc);
5837 			} else {
5838 				error = ether_poll_deregister(ifp);
5839 				/* Enable interrupt even in error case */
5840 				BGE_LOCK(sc);
5841 				BGE_CLRBIT(sc, BGE_PCI_MISC_CTL,
5842 				    BGE_PCIMISCCTL_MASK_PCI_INTR);
5843 				bge_writembx(sc, BGE_MBX_IRQ0_LO, 0);
5844 				if_setcapenablebit(ifp, 0, IFCAP_POLLING);
5845 				BGE_UNLOCK(sc);
5846 			}
5847 		}
5848 #endif
5849 		if ((mask & IFCAP_TXCSUM) != 0 &&
5850 		    (if_getcapabilities(ifp) & IFCAP_TXCSUM) != 0) {
5851 			if_togglecapenable(ifp, IFCAP_TXCSUM);
5852 			if ((if_getcapenable(ifp) & IFCAP_TXCSUM) != 0)
5853 				if_sethwassistbits(ifp,
5854 				    sc->bge_csum_features, 0);
5855 			else
5856 				if_sethwassistbits(ifp, 0,
5857 				    sc->bge_csum_features);
5858 		}
5859 
5860 		if ((mask & IFCAP_RXCSUM) != 0 &&
5861 		    (if_getcapabilities(ifp) & IFCAP_RXCSUM) != 0)
5862 			if_togglecapenable(ifp, IFCAP_RXCSUM);
5863 
5864 		if ((mask & IFCAP_TSO4) != 0 &&
5865 		    (if_getcapabilities(ifp) & IFCAP_TSO4) != 0) {
5866 			if_togglecapenable(ifp, IFCAP_TSO4);
5867 			if ((if_getcapenable(ifp) & IFCAP_TSO4) != 0)
5868 				if_sethwassistbits(ifp, CSUM_TSO, 0);
5869 			else
5870 				if_sethwassistbits(ifp, 0, CSUM_TSO);
5871 		}
5872 
5873 		if (mask & IFCAP_VLAN_MTU) {
5874 			if_togglecapenable(ifp, IFCAP_VLAN_MTU);
5875 			if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
5876 			bge_init(sc);
5877 		}
5878 
5879 		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
5880 		    (if_getcapabilities(ifp) & IFCAP_VLAN_HWTSO) != 0)
5881 			if_togglecapenable(ifp, IFCAP_VLAN_HWTSO);
5882 		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
5883 		    (if_getcapabilities(ifp) & IFCAP_VLAN_HWTAGGING) != 0) {
5884 			if_togglecapenable(ifp, IFCAP_VLAN_HWTAGGING);
5885 			if ((if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) == 0)
5886 				if_setcapenablebit(ifp, 0, IFCAP_VLAN_HWTSO);
5887 			BGE_LOCK(sc);
5888 			bge_setvlan(sc);
5889 			BGE_UNLOCK(sc);
5890 		}
5891 #ifdef VLAN_CAPABILITIES
5892 		if_vlancap(ifp);
5893 #endif
5894 		break;
5895 	default:
5896 		error = ether_ioctl(ifp, command, data);
5897 		break;
5898 	}
5899 
5900 	return (error);
5901 }
5902 
5903 static void
5904 bge_watchdog(struct bge_softc *sc)
5905 {
5906 	if_t ifp;
5907 	uint32_t status;
5908 
5909 	BGE_LOCK_ASSERT(sc);
5910 
5911 	if (sc->bge_timer == 0 || --sc->bge_timer)
5912 		return;
5913 
5914 	/* If pause frames are active then don't reset the hardware. */
5915 	if ((CSR_READ_4(sc, BGE_RX_MODE) & BGE_RXMODE_FLOWCTL_ENABLE) != 0) {
5916 		status = CSR_READ_4(sc, BGE_RX_STS);
5917 		if ((status & BGE_RXSTAT_REMOTE_XOFFED) != 0) {
5918 			/*
5919 			 * If link partner has us in XOFF state then wait for
5920 			 * the condition to clear.
5921 			 */
5922 			CSR_WRITE_4(sc, BGE_RX_STS, status);
5923 			sc->bge_timer = BGE_TX_TIMEOUT;
5924 			return;
5925 		} else if ((status & BGE_RXSTAT_RCVD_XOFF) != 0 &&
5926 		    (status & BGE_RXSTAT_RCVD_XON) != 0) {
5927 			/*
5928 			 * If link partner has us in XOFF state then wait for
5929 			 * the condition to clear.
5930 			 */
5931 			CSR_WRITE_4(sc, BGE_RX_STS, status);
5932 			sc->bge_timer = BGE_TX_TIMEOUT;
5933 			return;
5934 		}
5935 		/*
5936 		 * Any other condition is unexpected and the controller
5937 		 * should be reset.
5938 		 */
5939 	}
5940 
5941 	ifp = sc->bge_ifp;
5942 
5943 	if_printf(ifp, "watchdog timeout -- resetting\n");
5944 
5945 	if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
5946 	bge_init_locked(sc);
5947 
5948 	if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
5949 }
5950 
5951 static void
5952 bge_stop_block(struct bge_softc *sc, bus_size_t reg, uint32_t bit)
5953 {
5954 	int i;
5955 
5956 	BGE_CLRBIT(sc, reg, bit);
5957 
5958 	for (i = 0; i < BGE_TIMEOUT; i++) {
5959 		if ((CSR_READ_4(sc, reg) & bit) == 0)
5960 			return;
5961 		DELAY(100);
5962         }
5963 }
5964 
5965 /*
5966  * Stop the adapter and free any mbufs allocated to the
5967  * RX and TX lists.
5968  */
5969 static void
5970 bge_stop(struct bge_softc *sc)
5971 {
5972 	if_t ifp;
5973 
5974 	BGE_LOCK_ASSERT(sc);
5975 
5976 	ifp = sc->bge_ifp;
5977 
5978 	callout_stop(&sc->bge_stat_ch);
5979 
5980 	/* Disable host interrupts. */
5981 	BGE_SETBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR);
5982 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
5983 
5984 	/*
5985 	 * Tell firmware we're shutting down.
5986 	 */
5987 	bge_stop_fw(sc);
5988 	bge_sig_pre_reset(sc, BGE_RESET_SHUTDOWN);
5989 
5990 	/*
5991 	 * Disable all of the receiver blocks.
5992 	 */
5993 	bge_stop_block(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
5994 	bge_stop_block(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
5995 	bge_stop_block(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
5996 	if (BGE_IS_5700_FAMILY(sc))
5997 		bge_stop_block(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
5998 	bge_stop_block(sc, BGE_RDBDI_MODE, BGE_RBDIMODE_ENABLE);
5999 	bge_stop_block(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
6000 	bge_stop_block(sc, BGE_RBDC_MODE, BGE_RBDCMODE_ENABLE);
6001 
6002 	/*
6003 	 * Disable all of the transmit blocks.
6004 	 */
6005 	bge_stop_block(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
6006 	bge_stop_block(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
6007 	bge_stop_block(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
6008 	bge_stop_block(sc, BGE_RDMA_MODE, BGE_RDMAMODE_ENABLE);
6009 	bge_stop_block(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE);
6010 	if (BGE_IS_5700_FAMILY(sc))
6011 		bge_stop_block(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
6012 	bge_stop_block(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
6013 
6014 	/*
6015 	 * Shut down all of the memory managers and related
6016 	 * state machines.
6017 	 */
6018 	bge_stop_block(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE);
6019 	bge_stop_block(sc, BGE_WDMA_MODE, BGE_WDMAMODE_ENABLE);
6020 	if (BGE_IS_5700_FAMILY(sc))
6021 		bge_stop_block(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
6022 
6023 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
6024 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
6025 	if (!(BGE_IS_5705_PLUS(sc))) {
6026 		BGE_CLRBIT(sc, BGE_BMAN_MODE, BGE_BMANMODE_ENABLE);
6027 		BGE_CLRBIT(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
6028 	}
6029 	/* Update MAC statistics. */
6030 	if (BGE_IS_5705_PLUS(sc))
6031 		bge_stats_update_regs(sc);
6032 
6033 	bge_reset(sc);
6034 	bge_sig_legacy(sc, BGE_RESET_SHUTDOWN);
6035 	bge_sig_post_reset(sc, BGE_RESET_SHUTDOWN);
6036 
6037 	/*
6038 	 * Keep the ASF firmware running if up.
6039 	 */
6040 	if (sc->bge_asf_mode & ASF_STACKUP)
6041 		BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
6042 	else
6043 		BGE_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
6044 
6045 	/* Free the RX lists. */
6046 	bge_free_rx_ring_std(sc);
6047 
6048 	/* Free jumbo RX list. */
6049 	if (BGE_IS_JUMBO_CAPABLE(sc))
6050 		bge_free_rx_ring_jumbo(sc);
6051 
6052 	/* Free TX buffers. */
6053 	bge_free_tx_ring(sc);
6054 
6055 	sc->bge_tx_saved_considx = BGE_TXCONS_UNSET;
6056 
6057 	/* Clear MAC's link state (PHY may still have link UP). */
6058 	if (bootverbose && sc->bge_link)
6059 		if_printf(sc->bge_ifp, "link DOWN\n");
6060 	sc->bge_link = 0;
6061 
6062 	if_setdrvflagbits(ifp, 0, (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
6063 }
6064 
6065 /*
6066  * Stop all chip I/O so that the kernel's probe routines don't
6067  * get confused by errant DMAs when rebooting.
6068  */
6069 static int
6070 bge_shutdown(device_t dev)
6071 {
6072 	struct bge_softc *sc;
6073 
6074 	sc = device_get_softc(dev);
6075 	BGE_LOCK(sc);
6076 	bge_stop(sc);
6077 	BGE_UNLOCK(sc);
6078 
6079 	return (0);
6080 }
6081 
6082 static int
6083 bge_suspend(device_t dev)
6084 {
6085 	struct bge_softc *sc;
6086 
6087 	sc = device_get_softc(dev);
6088 	BGE_LOCK(sc);
6089 	bge_stop(sc);
6090 	BGE_UNLOCK(sc);
6091 
6092 	return (0);
6093 }
6094 
6095 static int
6096 bge_resume(device_t dev)
6097 {
6098 	struct bge_softc *sc;
6099 	if_t ifp;
6100 
6101 	sc = device_get_softc(dev);
6102 	BGE_LOCK(sc);
6103 	ifp = sc->bge_ifp;
6104 	if (if_getflags(ifp) & IFF_UP) {
6105 		bge_init_locked(sc);
6106 		if (if_getdrvflags(ifp) & IFF_DRV_RUNNING)
6107 			bge_start_locked(ifp);
6108 	}
6109 	BGE_UNLOCK(sc);
6110 
6111 	return (0);
6112 }
6113 
6114 static void
6115 bge_link_upd(struct bge_softc *sc)
6116 {
6117 	struct mii_data *mii;
6118 	uint32_t link, status;
6119 
6120 	BGE_LOCK_ASSERT(sc);
6121 
6122 	/* Clear 'pending link event' flag. */
6123 	sc->bge_link_evt = 0;
6124 
6125 	/*
6126 	 * Process link state changes.
6127 	 * Grrr. The link status word in the status block does
6128 	 * not work correctly on the BCM5700 rev AX and BX chips,
6129 	 * according to all available information. Hence, we have
6130 	 * to enable MII interrupts in order to properly obtain
6131 	 * async link changes. Unfortunately, this also means that
6132 	 * we have to read the MAC status register to detect link
6133 	 * changes, thereby adding an additional register access to
6134 	 * the interrupt handler.
6135 	 *
6136 	 * XXX: perhaps link state detection procedure used for
6137 	 * BGE_CHIPID_BCM5700_B2 can be used for others BCM5700 revisions.
6138 	 */
6139 
6140 	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
6141 	    sc->bge_chipid != BGE_CHIPID_BCM5700_B2) {
6142 		status = CSR_READ_4(sc, BGE_MAC_STS);
6143 		if (status & BGE_MACSTAT_MI_INTERRUPT) {
6144 			mii = device_get_softc(sc->bge_miibus);
6145 			mii_pollstat(mii);
6146 			if (!sc->bge_link &&
6147 			    mii->mii_media_status & IFM_ACTIVE &&
6148 			    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
6149 				sc->bge_link++;
6150 				if (bootverbose)
6151 					if_printf(sc->bge_ifp, "link UP\n");
6152 			} else if (sc->bge_link &&
6153 			    (!(mii->mii_media_status & IFM_ACTIVE) ||
6154 			    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE)) {
6155 				sc->bge_link = 0;
6156 				if (bootverbose)
6157 					if_printf(sc->bge_ifp, "link DOWN\n");
6158 			}
6159 
6160 			/* Clear the interrupt. */
6161 			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
6162 			    BGE_EVTENB_MI_INTERRUPT);
6163 			bge_miibus_readreg(sc->bge_dev, sc->bge_phy_addr,
6164 			    BRGPHY_MII_ISR);
6165 			bge_miibus_writereg(sc->bge_dev, sc->bge_phy_addr,
6166 			    BRGPHY_MII_IMR, BRGPHY_INTRS);
6167 		}
6168 		return;
6169 	}
6170 
6171 	if (sc->bge_flags & BGE_FLAG_TBI) {
6172 		status = CSR_READ_4(sc, BGE_MAC_STS);
6173 		if (status & BGE_MACSTAT_TBI_PCS_SYNCHED) {
6174 			if (!sc->bge_link) {
6175 				sc->bge_link++;
6176 				if (sc->bge_asicrev == BGE_ASICREV_BCM5704) {
6177 					BGE_CLRBIT(sc, BGE_MAC_MODE,
6178 					    BGE_MACMODE_TBI_SEND_CFGS);
6179 					DELAY(40);
6180 				}
6181 				CSR_WRITE_4(sc, BGE_MAC_STS, 0xFFFFFFFF);
6182 				if (bootverbose)
6183 					if_printf(sc->bge_ifp, "link UP\n");
6184 				if_link_state_change(sc->bge_ifp,
6185 				    LINK_STATE_UP);
6186 			}
6187 		} else if (sc->bge_link) {
6188 			sc->bge_link = 0;
6189 			if (bootverbose)
6190 				if_printf(sc->bge_ifp, "link DOWN\n");
6191 			if_link_state_change(sc->bge_ifp, LINK_STATE_DOWN);
6192 		}
6193 	} else if ((sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) != 0) {
6194 		/*
6195 		 * Some broken BCM chips have BGE_STATFLAG_LINKSTATE_CHANGED bit
6196 		 * in status word always set. Workaround this bug by reading
6197 		 * PHY link status directly.
6198 		 */
6199 		link = (CSR_READ_4(sc, BGE_MI_STS) & BGE_MISTS_LINK) ? 1 : 0;
6200 
6201 		if (link != sc->bge_link ||
6202 		    sc->bge_asicrev == BGE_ASICREV_BCM5700) {
6203 			mii = device_get_softc(sc->bge_miibus);
6204 			mii_pollstat(mii);
6205 			if (!sc->bge_link &&
6206 			    mii->mii_media_status & IFM_ACTIVE &&
6207 			    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
6208 				sc->bge_link++;
6209 				if (bootverbose)
6210 					if_printf(sc->bge_ifp, "link UP\n");
6211 			} else if (sc->bge_link &&
6212 			    (!(mii->mii_media_status & IFM_ACTIVE) ||
6213 			    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE)) {
6214 				sc->bge_link = 0;
6215 				if (bootverbose)
6216 					if_printf(sc->bge_ifp, "link DOWN\n");
6217 			}
6218 		}
6219 	} else {
6220 		/*
6221 		 * For controllers that call mii_tick, we have to poll
6222 		 * link status.
6223 		 */
6224 		mii = device_get_softc(sc->bge_miibus);
6225 		mii_pollstat(mii);
6226 		bge_miibus_statchg(sc->bge_dev);
6227 	}
6228 
6229 	/* Disable MAC attention when link is up. */
6230 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
6231 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
6232 	    BGE_MACSTAT_LINK_CHANGED);
6233 }
6234 
6235 static void
6236 bge_add_sysctls(struct bge_softc *sc)
6237 {
6238 	struct sysctl_ctx_list *ctx;
6239 	struct sysctl_oid_list *children;
6240 	int unit;
6241 
6242 	ctx = device_get_sysctl_ctx(sc->bge_dev);
6243 	children = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->bge_dev));
6244 
6245 #ifdef BGE_REGISTER_DEBUG
6246 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "debug_info",
6247 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, sc, 0,
6248 	    bge_sysctl_debug_info, "I", "Debug Information");
6249 
6250 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "reg_read",
6251 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, sc, 0,
6252 	    bge_sysctl_reg_read, "I", "MAC Register Read");
6253 
6254 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ape_read",
6255 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, sc, 0,
6256 	    bge_sysctl_ape_read, "I", "APE Register Read");
6257 
6258 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mem_read",
6259 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, sc, 0,
6260 	    bge_sysctl_mem_read, "I", "Memory Read");
6261 
6262 #endif
6263 
6264 	unit = device_get_unit(sc->bge_dev);
6265 	/*
6266 	 * A common design characteristic for many Broadcom client controllers
6267 	 * is that they only support a single outstanding DMA read operation
6268 	 * on the PCIe bus. This means that it will take twice as long to fetch
6269 	 * a TX frame that is split into header and payload buffers as it does
6270 	 * to fetch a single, contiguous TX frame (2 reads vs. 1 read). For
6271 	 * these controllers, coalescing buffers to reduce the number of memory
6272 	 * reads is effective way to get maximum performance(about 940Mbps).
6273 	 * Without collapsing TX buffers the maximum TCP bulk transfer
6274 	 * performance is about 850Mbps. However forcing coalescing mbufs
6275 	 * consumes a lot of CPU cycles, so leave it off by default.
6276 	 */
6277 	sc->bge_forced_collapse = 0;
6278 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "forced_collapse",
6279 	    CTLFLAG_RWTUN, &sc->bge_forced_collapse, 0,
6280 	    "Number of fragmented TX buffers of a frame allowed before "
6281 	    "forced collapsing");
6282 
6283 	sc->bge_msi = 1;
6284 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "msi",
6285 	    CTLFLAG_RDTUN, &sc->bge_msi, 0, "Enable MSI");
6286 
6287 	/*
6288 	 * It seems all Broadcom controllers have a bug that can generate UDP
6289 	 * datagrams with checksum value 0 when TX UDP checksum offloading is
6290 	 * enabled.  Generating UDP checksum value 0 is RFC 768 violation.
6291 	 * Even though the probability of generating such UDP datagrams is
6292 	 * low, I don't want to see FreeBSD boxes to inject such datagrams
6293 	 * into network so disable UDP checksum offloading by default.  Users
6294 	 * still override this behavior by setting a sysctl variable,
6295 	 * dev.bge.0.forced_udpcsum.
6296 	 */
6297 	sc->bge_forced_udpcsum = 0;
6298 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "forced_udpcsum",
6299 	    CTLFLAG_RWTUN, &sc->bge_forced_udpcsum, 0,
6300 	    "Enable UDP checksum offloading even if controller can "
6301 	    "generate UDP checksum value 0");
6302 
6303 	if (BGE_IS_5705_PLUS(sc))
6304 		bge_add_sysctl_stats_regs(sc, ctx, children);
6305 	else
6306 		bge_add_sysctl_stats(sc, ctx, children);
6307 }
6308 
6309 #define BGE_SYSCTL_STAT(sc, ctx, desc, parent, node, oid) \
6310     SYSCTL_ADD_PROC(ctx, parent, OID_AUTO, oid, \
6311         CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, \
6312 	offsetof(struct bge_stats, node), bge_sysctl_stats, "IU", desc)
6313 
6314 static void
6315 bge_add_sysctl_stats(struct bge_softc *sc, struct sysctl_ctx_list *ctx,
6316     struct sysctl_oid_list *parent)
6317 {
6318 	struct sysctl_oid *tree;
6319 	struct sysctl_oid_list *children, *schildren;
6320 
6321 	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "stats",
6322 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "BGE Statistics");
6323 	schildren = children = SYSCTL_CHILDREN(tree);
6324 	BGE_SYSCTL_STAT(sc, ctx, "Frames Dropped Due To Filters",
6325 	    children, COSFramesDroppedDueToFilters,
6326 	    "FramesDroppedDueToFilters");
6327 	BGE_SYSCTL_STAT(sc, ctx, "NIC DMA Write Queue Full",
6328 	    children, nicDmaWriteQueueFull, "DmaWriteQueueFull");
6329 	BGE_SYSCTL_STAT(sc, ctx, "NIC DMA Write High Priority Queue Full",
6330 	    children, nicDmaWriteHighPriQueueFull, "DmaWriteHighPriQueueFull");
6331 	BGE_SYSCTL_STAT(sc, ctx, "NIC No More RX Buffer Descriptors",
6332 	    children, nicNoMoreRxBDs, "NoMoreRxBDs");
6333 	BGE_SYSCTL_STAT(sc, ctx, "Discarded Input Frames",
6334 	    children, ifInDiscards, "InputDiscards");
6335 	BGE_SYSCTL_STAT(sc, ctx, "Input Errors",
6336 	    children, ifInErrors, "InputErrors");
6337 	BGE_SYSCTL_STAT(sc, ctx, "NIC Recv Threshold Hit",
6338 	    children, nicRecvThresholdHit, "RecvThresholdHit");
6339 	BGE_SYSCTL_STAT(sc, ctx, "NIC DMA Read Queue Full",
6340 	    children, nicDmaReadQueueFull, "DmaReadQueueFull");
6341 	BGE_SYSCTL_STAT(sc, ctx, "NIC DMA Read High Priority Queue Full",
6342 	    children, nicDmaReadHighPriQueueFull, "DmaReadHighPriQueueFull");
6343 	BGE_SYSCTL_STAT(sc, ctx, "NIC Send Data Complete Queue Full",
6344 	    children, nicSendDataCompQueueFull, "SendDataCompQueueFull");
6345 	BGE_SYSCTL_STAT(sc, ctx, "NIC Ring Set Send Producer Index",
6346 	    children, nicRingSetSendProdIndex, "RingSetSendProdIndex");
6347 	BGE_SYSCTL_STAT(sc, ctx, "NIC Ring Status Update",
6348 	    children, nicRingStatusUpdate, "RingStatusUpdate");
6349 	BGE_SYSCTL_STAT(sc, ctx, "NIC Interrupts",
6350 	    children, nicInterrupts, "Interrupts");
6351 	BGE_SYSCTL_STAT(sc, ctx, "NIC Avoided Interrupts",
6352 	    children, nicAvoidedInterrupts, "AvoidedInterrupts");
6353 	BGE_SYSCTL_STAT(sc, ctx, "NIC Send Threshold Hit",
6354 	    children, nicSendThresholdHit, "SendThresholdHit");
6355 
6356 	tree = SYSCTL_ADD_NODE(ctx, schildren, OID_AUTO, "rx",
6357 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "BGE RX Statistics");
6358 	children = SYSCTL_CHILDREN(tree);
6359 	BGE_SYSCTL_STAT(sc, ctx, "Inbound Octets",
6360 	    children, rxstats.ifHCInOctets, "ifHCInOctets");
6361 	BGE_SYSCTL_STAT(sc, ctx, "Fragments",
6362 	    children, rxstats.etherStatsFragments, "Fragments");
6363 	BGE_SYSCTL_STAT(sc, ctx, "Inbound Unicast Packets",
6364 	    children, rxstats.ifHCInUcastPkts, "UnicastPkts");
6365 	BGE_SYSCTL_STAT(sc, ctx, "Inbound Multicast Packets",
6366 	    children, rxstats.ifHCInMulticastPkts, "MulticastPkts");
6367 	BGE_SYSCTL_STAT(sc, ctx, "FCS Errors",
6368 	    children, rxstats.dot3StatsFCSErrors, "FCSErrors");
6369 	BGE_SYSCTL_STAT(sc, ctx, "Alignment Errors",
6370 	    children, rxstats.dot3StatsAlignmentErrors, "AlignmentErrors");
6371 	BGE_SYSCTL_STAT(sc, ctx, "XON Pause Frames Received",
6372 	    children, rxstats.xonPauseFramesReceived, "xonPauseFramesReceived");
6373 	BGE_SYSCTL_STAT(sc, ctx, "XOFF Pause Frames Received",
6374 	    children, rxstats.xoffPauseFramesReceived,
6375 	    "xoffPauseFramesReceived");
6376 	BGE_SYSCTL_STAT(sc, ctx, "MAC Control Frames Received",
6377 	    children, rxstats.macControlFramesReceived,
6378 	    "ControlFramesReceived");
6379 	BGE_SYSCTL_STAT(sc, ctx, "XOFF State Entered",
6380 	    children, rxstats.xoffStateEntered, "xoffStateEntered");
6381 	BGE_SYSCTL_STAT(sc, ctx, "Frames Too Long",
6382 	    children, rxstats.dot3StatsFramesTooLong, "FramesTooLong");
6383 	BGE_SYSCTL_STAT(sc, ctx, "Jabbers",
6384 	    children, rxstats.etherStatsJabbers, "Jabbers");
6385 	BGE_SYSCTL_STAT(sc, ctx, "Undersized Packets",
6386 	    children, rxstats.etherStatsUndersizePkts, "UndersizePkts");
6387 	BGE_SYSCTL_STAT(sc, ctx, "Inbound Range Length Errors",
6388 	    children, rxstats.inRangeLengthError, "inRangeLengthError");
6389 	BGE_SYSCTL_STAT(sc, ctx, "Outbound Range Length Errors",
6390 	    children, rxstats.outRangeLengthError, "outRangeLengthError");
6391 
6392 	tree = SYSCTL_ADD_NODE(ctx, schildren, OID_AUTO, "tx",
6393 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "BGE TX Statistics");
6394 	children = SYSCTL_CHILDREN(tree);
6395 	BGE_SYSCTL_STAT(sc, ctx, "Outbound Octets",
6396 	    children, txstats.ifHCOutOctets, "ifHCOutOctets");
6397 	BGE_SYSCTL_STAT(sc, ctx, "TX Collisions",
6398 	    children, txstats.etherStatsCollisions, "Collisions");
6399 	BGE_SYSCTL_STAT(sc, ctx, "XON Sent",
6400 	    children, txstats.outXonSent, "XonSent");
6401 	BGE_SYSCTL_STAT(sc, ctx, "XOFF Sent",
6402 	    children, txstats.outXoffSent, "XoffSent");
6403 	BGE_SYSCTL_STAT(sc, ctx, "Flow Control Done",
6404 	    children, txstats.flowControlDone, "flowControlDone");
6405 	BGE_SYSCTL_STAT(sc, ctx, "Internal MAC TX errors",
6406 	    children, txstats.dot3StatsInternalMacTransmitErrors,
6407 	    "InternalMacTransmitErrors");
6408 	BGE_SYSCTL_STAT(sc, ctx, "Single Collision Frames",
6409 	    children, txstats.dot3StatsSingleCollisionFrames,
6410 	    "SingleCollisionFrames");
6411 	BGE_SYSCTL_STAT(sc, ctx, "Multiple Collision Frames",
6412 	    children, txstats.dot3StatsMultipleCollisionFrames,
6413 	    "MultipleCollisionFrames");
6414 	BGE_SYSCTL_STAT(sc, ctx, "Deferred Transmissions",
6415 	    children, txstats.dot3StatsDeferredTransmissions,
6416 	    "DeferredTransmissions");
6417 	BGE_SYSCTL_STAT(sc, ctx, "Excessive Collisions",
6418 	    children, txstats.dot3StatsExcessiveCollisions,
6419 	    "ExcessiveCollisions");
6420 	BGE_SYSCTL_STAT(sc, ctx, "Late Collisions",
6421 	    children, txstats.dot3StatsLateCollisions,
6422 	    "LateCollisions");
6423 	BGE_SYSCTL_STAT(sc, ctx, "Outbound Unicast Packets",
6424 	    children, txstats.ifHCOutUcastPkts, "UnicastPkts");
6425 	BGE_SYSCTL_STAT(sc, ctx, "Outbound Multicast Packets",
6426 	    children, txstats.ifHCOutMulticastPkts, "MulticastPkts");
6427 	BGE_SYSCTL_STAT(sc, ctx, "Outbound Broadcast Packets",
6428 	    children, txstats.ifHCOutBroadcastPkts, "BroadcastPkts");
6429 	BGE_SYSCTL_STAT(sc, ctx, "Carrier Sense Errors",
6430 	    children, txstats.dot3StatsCarrierSenseErrors,
6431 	    "CarrierSenseErrors");
6432 	BGE_SYSCTL_STAT(sc, ctx, "Outbound Discards",
6433 	    children, txstats.ifOutDiscards, "Discards");
6434 	BGE_SYSCTL_STAT(sc, ctx, "Outbound Errors",
6435 	    children, txstats.ifOutErrors, "Errors");
6436 }
6437 
6438 #undef BGE_SYSCTL_STAT
6439 
6440 #define	BGE_SYSCTL_STAT_ADD64(c, h, n, p, d)	\
6441 	    SYSCTL_ADD_UQUAD(c, h, OID_AUTO, n, CTLFLAG_RD, p, d)
6442 
6443 static void
6444 bge_add_sysctl_stats_regs(struct bge_softc *sc, struct sysctl_ctx_list *ctx,
6445     struct sysctl_oid_list *parent)
6446 {
6447 	struct sysctl_oid *tree;
6448 	struct sysctl_oid_list *child, *schild;
6449 	struct bge_mac_stats *stats;
6450 
6451 	stats = &sc->bge_mac_stats;
6452 	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "stats",
6453 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "BGE Statistics");
6454 	schild = child = SYSCTL_CHILDREN(tree);
6455 	BGE_SYSCTL_STAT_ADD64(ctx, child, "FramesDroppedDueToFilters",
6456 	    &stats->FramesDroppedDueToFilters, "Frames Dropped Due to Filters");
6457 	BGE_SYSCTL_STAT_ADD64(ctx, child, "DmaWriteQueueFull",
6458 	    &stats->DmaWriteQueueFull, "NIC DMA Write Queue Full");
6459 	BGE_SYSCTL_STAT_ADD64(ctx, child, "DmaWriteHighPriQueueFull",
6460 	    &stats->DmaWriteHighPriQueueFull,
6461 	    "NIC DMA Write High Priority Queue Full");
6462 	BGE_SYSCTL_STAT_ADD64(ctx, child, "NoMoreRxBDs",
6463 	    &stats->NoMoreRxBDs, "NIC No More RX Buffer Descriptors");
6464 	BGE_SYSCTL_STAT_ADD64(ctx, child, "InputDiscards",
6465 	    &stats->InputDiscards, "Discarded Input Frames");
6466 	BGE_SYSCTL_STAT_ADD64(ctx, child, "InputErrors",
6467 	    &stats->InputErrors, "Input Errors");
6468 	BGE_SYSCTL_STAT_ADD64(ctx, child, "RecvThresholdHit",
6469 	    &stats->RecvThresholdHit, "NIC Recv Threshold Hit");
6470 
6471 	tree = SYSCTL_ADD_NODE(ctx, schild, OID_AUTO, "rx",
6472 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "BGE RX Statistics");
6473 	child = SYSCTL_CHILDREN(tree);
6474 	BGE_SYSCTL_STAT_ADD64(ctx, child, "ifHCInOctets",
6475 	    &stats->ifHCInOctets, "Inbound Octets");
6476 	BGE_SYSCTL_STAT_ADD64(ctx, child, "Fragments",
6477 	    &stats->etherStatsFragments, "Fragments");
6478 	BGE_SYSCTL_STAT_ADD64(ctx, child, "UnicastPkts",
6479 	    &stats->ifHCInUcastPkts, "Inbound Unicast Packets");
6480 	BGE_SYSCTL_STAT_ADD64(ctx, child, "MulticastPkts",
6481 	    &stats->ifHCInMulticastPkts, "Inbound Multicast Packets");
6482 	BGE_SYSCTL_STAT_ADD64(ctx, child, "BroadcastPkts",
6483 	    &stats->ifHCInBroadcastPkts, "Inbound Broadcast Packets");
6484 	BGE_SYSCTL_STAT_ADD64(ctx, child, "FCSErrors",
6485 	    &stats->dot3StatsFCSErrors, "FCS Errors");
6486 	BGE_SYSCTL_STAT_ADD64(ctx, child, "AlignmentErrors",
6487 	    &stats->dot3StatsAlignmentErrors, "Alignment Errors");
6488 	BGE_SYSCTL_STAT_ADD64(ctx, child, "xonPauseFramesReceived",
6489 	    &stats->xonPauseFramesReceived, "XON Pause Frames Received");
6490 	BGE_SYSCTL_STAT_ADD64(ctx, child, "xoffPauseFramesReceived",
6491 	    &stats->xoffPauseFramesReceived, "XOFF Pause Frames Received");
6492 	BGE_SYSCTL_STAT_ADD64(ctx, child, "ControlFramesReceived",
6493 	    &stats->macControlFramesReceived, "MAC Control Frames Received");
6494 	BGE_SYSCTL_STAT_ADD64(ctx, child, "xoffStateEntered",
6495 	    &stats->xoffStateEntered, "XOFF State Entered");
6496 	BGE_SYSCTL_STAT_ADD64(ctx, child, "FramesTooLong",
6497 	    &stats->dot3StatsFramesTooLong, "Frames Too Long");
6498 	BGE_SYSCTL_STAT_ADD64(ctx, child, "Jabbers",
6499 	    &stats->etherStatsJabbers, "Jabbers");
6500 	BGE_SYSCTL_STAT_ADD64(ctx, child, "UndersizePkts",
6501 	    &stats->etherStatsUndersizePkts, "Undersized Packets");
6502 
6503 	tree = SYSCTL_ADD_NODE(ctx, schild, OID_AUTO, "tx",
6504 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "BGE TX Statistics");
6505 	child = SYSCTL_CHILDREN(tree);
6506 	BGE_SYSCTL_STAT_ADD64(ctx, child, "ifHCOutOctets",
6507 	    &stats->ifHCOutOctets, "Outbound Octets");
6508 	BGE_SYSCTL_STAT_ADD64(ctx, child, "Collisions",
6509 	    &stats->etherStatsCollisions, "TX Collisions");
6510 	BGE_SYSCTL_STAT_ADD64(ctx, child, "XonSent",
6511 	    &stats->outXonSent, "XON Sent");
6512 	BGE_SYSCTL_STAT_ADD64(ctx, child, "XoffSent",
6513 	    &stats->outXoffSent, "XOFF Sent");
6514 	BGE_SYSCTL_STAT_ADD64(ctx, child, "InternalMacTransmitErrors",
6515 	    &stats->dot3StatsInternalMacTransmitErrors,
6516 	    "Internal MAC TX Errors");
6517 	BGE_SYSCTL_STAT_ADD64(ctx, child, "SingleCollisionFrames",
6518 	    &stats->dot3StatsSingleCollisionFrames, "Single Collision Frames");
6519 	BGE_SYSCTL_STAT_ADD64(ctx, child, "MultipleCollisionFrames",
6520 	    &stats->dot3StatsMultipleCollisionFrames,
6521 	    "Multiple Collision Frames");
6522 	BGE_SYSCTL_STAT_ADD64(ctx, child, "DeferredTransmissions",
6523 	    &stats->dot3StatsDeferredTransmissions, "Deferred Transmissions");
6524 	BGE_SYSCTL_STAT_ADD64(ctx, child, "ExcessiveCollisions",
6525 	    &stats->dot3StatsExcessiveCollisions, "Excessive Collisions");
6526 	BGE_SYSCTL_STAT_ADD64(ctx, child, "LateCollisions",
6527 	    &stats->dot3StatsLateCollisions, "Late Collisions");
6528 	BGE_SYSCTL_STAT_ADD64(ctx, child, "UnicastPkts",
6529 	    &stats->ifHCOutUcastPkts, "Outbound Unicast Packets");
6530 	BGE_SYSCTL_STAT_ADD64(ctx, child, "MulticastPkts",
6531 	    &stats->ifHCOutMulticastPkts, "Outbound Multicast Packets");
6532 	BGE_SYSCTL_STAT_ADD64(ctx, child, "BroadcastPkts",
6533 	    &stats->ifHCOutBroadcastPkts, "Outbound Broadcast Packets");
6534 }
6535 
6536 #undef	BGE_SYSCTL_STAT_ADD64
6537 
6538 static int
6539 bge_sysctl_stats(SYSCTL_HANDLER_ARGS)
6540 {
6541 	struct bge_softc *sc;
6542 	uint32_t result;
6543 	int offset;
6544 
6545 	sc = (struct bge_softc *)arg1;
6546 	offset = arg2;
6547 	result = CSR_READ_4(sc, BGE_MEMWIN_START + BGE_STATS_BLOCK + offset +
6548 	    offsetof(bge_hostaddr, bge_addr_lo));
6549 	return (sysctl_handle_int(oidp, &result, 0, req));
6550 }
6551 
6552 #ifdef BGE_REGISTER_DEBUG
6553 static int
6554 bge_sysctl_debug_info(SYSCTL_HANDLER_ARGS)
6555 {
6556 	struct bge_softc *sc;
6557 	uint16_t *sbdata;
6558 	int error, result, sbsz;
6559 	int i, j;
6560 
6561 	result = -1;
6562 	error = sysctl_handle_int(oidp, &result, 0, req);
6563 	if (error || (req->newptr == NULL))
6564 		return (error);
6565 
6566 	if (result == 1) {
6567 		sc = (struct bge_softc *)arg1;
6568 
6569 		if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
6570 		    sc->bge_chipid != BGE_CHIPID_BCM5700_C0)
6571 			sbsz = BGE_STATUS_BLK_SZ;
6572 		else
6573 			sbsz = 32;
6574 		sbdata = (uint16_t *)sc->bge_ldata.bge_status_block;
6575 		printf("Status Block:\n");
6576 		BGE_LOCK(sc);
6577 		bus_dmamap_sync(sc->bge_cdata.bge_status_tag,
6578 		    sc->bge_cdata.bge_status_map,
6579 		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
6580 		for (i = 0x0; i < sbsz / sizeof(uint16_t); ) {
6581 			printf("%06x:", i);
6582 			for (j = 0; j < 8; j++)
6583 				printf(" %04x", sbdata[i++]);
6584 			printf("\n");
6585 		}
6586 
6587 		printf("Registers:\n");
6588 		for (i = 0x800; i < 0xA00; ) {
6589 			printf("%06x:", i);
6590 			for (j = 0; j < 8; j++) {
6591 				printf(" %08x", CSR_READ_4(sc, i));
6592 				i += 4;
6593 			}
6594 			printf("\n");
6595 		}
6596 		BGE_UNLOCK(sc);
6597 
6598 		printf("Hardware Flags:\n");
6599 		if (BGE_IS_5717_PLUS(sc))
6600 			printf(" - 5717 Plus\n");
6601 		if (BGE_IS_5755_PLUS(sc))
6602 			printf(" - 5755 Plus\n");
6603 		if (BGE_IS_575X_PLUS(sc))
6604 			printf(" - 575X Plus\n");
6605 		if (BGE_IS_5705_PLUS(sc))
6606 			printf(" - 5705 Plus\n");
6607 		if (BGE_IS_5714_FAMILY(sc))
6608 			printf(" - 5714 Family\n");
6609 		if (BGE_IS_5700_FAMILY(sc))
6610 			printf(" - 5700 Family\n");
6611 		if (sc->bge_flags & BGE_FLAG_JUMBO)
6612 			printf(" - Supports Jumbo Frames\n");
6613 		if (sc->bge_flags & BGE_FLAG_PCIX)
6614 			printf(" - PCI-X Bus\n");
6615 		if (sc->bge_flags & BGE_FLAG_PCIE)
6616 			printf(" - PCI Express Bus\n");
6617 		if (sc->bge_phy_flags & BGE_PHY_NO_3LED)
6618 			printf(" - No 3 LEDs\n");
6619 		if (sc->bge_flags & BGE_FLAG_RX_ALIGNBUG)
6620 			printf(" - RX Alignment Bug\n");
6621 	}
6622 
6623 	return (error);
6624 }
6625 
6626 static int
6627 bge_sysctl_reg_read(SYSCTL_HANDLER_ARGS)
6628 {
6629 	struct bge_softc *sc;
6630 	int error;
6631 	uint16_t result;
6632 	uint32_t val;
6633 
6634 	result = -1;
6635 	error = sysctl_handle_int(oidp, &result, 0, req);
6636 	if (error || (req->newptr == NULL))
6637 		return (error);
6638 
6639 	if (result < 0x8000) {
6640 		sc = (struct bge_softc *)arg1;
6641 		val = CSR_READ_4(sc, result);
6642 		printf("reg 0x%06X = 0x%08X\n", result, val);
6643 	}
6644 
6645 	return (error);
6646 }
6647 
6648 static int
6649 bge_sysctl_ape_read(SYSCTL_HANDLER_ARGS)
6650 {
6651 	struct bge_softc *sc;
6652 	int error;
6653 	uint16_t result;
6654 	uint32_t val;
6655 
6656 	result = -1;
6657 	error = sysctl_handle_int(oidp, &result, 0, req);
6658 	if (error || (req->newptr == NULL))
6659 		return (error);
6660 
6661 	if (result < 0x8000) {
6662 		sc = (struct bge_softc *)arg1;
6663 		val = APE_READ_4(sc, result);
6664 		printf("reg 0x%06X = 0x%08X\n", result, val);
6665 	}
6666 
6667 	return (error);
6668 }
6669 
6670 static int
6671 bge_sysctl_mem_read(SYSCTL_HANDLER_ARGS)
6672 {
6673 	struct bge_softc *sc;
6674 	int error;
6675 	uint16_t result;
6676 	uint32_t val;
6677 
6678 	result = -1;
6679 	error = sysctl_handle_int(oidp, &result, 0, req);
6680 	if (error || (req->newptr == NULL))
6681 		return (error);
6682 
6683 	if (result < 0x8000) {
6684 		sc = (struct bge_softc *)arg1;
6685 		val = bge_readmem_ind(sc, result);
6686 		printf("mem 0x%06X = 0x%08X\n", result, val);
6687 	}
6688 
6689 	return (error);
6690 }
6691 #endif
6692 
6693 static int
6694 bge_get_eaddr_fw(struct bge_softc *sc, uint8_t ether_addr[])
6695 {
6696 	return (1);
6697 }
6698 
6699 static int
6700 bge_get_eaddr_mem(struct bge_softc *sc, uint8_t ether_addr[])
6701 {
6702 	uint32_t mac_addr;
6703 
6704 	mac_addr = bge_readmem_ind(sc, BGE_SRAM_MAC_ADDR_HIGH_MB);
6705 	if ((mac_addr >> 16) == 0x484b) {
6706 		ether_addr[0] = (uint8_t)(mac_addr >> 8);
6707 		ether_addr[1] = (uint8_t)mac_addr;
6708 		mac_addr = bge_readmem_ind(sc, BGE_SRAM_MAC_ADDR_LOW_MB);
6709 		ether_addr[2] = (uint8_t)(mac_addr >> 24);
6710 		ether_addr[3] = (uint8_t)(mac_addr >> 16);
6711 		ether_addr[4] = (uint8_t)(mac_addr >> 8);
6712 		ether_addr[5] = (uint8_t)mac_addr;
6713 		return (0);
6714 	}
6715 	return (1);
6716 }
6717 
6718 static int
6719 bge_get_eaddr_nvram(struct bge_softc *sc, uint8_t ether_addr[])
6720 {
6721 	int mac_offset = BGE_EE_MAC_OFFSET;
6722 
6723 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
6724 		mac_offset = BGE_EE_MAC_OFFSET_5906;
6725 
6726 	return (bge_read_nvram(sc, ether_addr, mac_offset + 2,
6727 	    ETHER_ADDR_LEN));
6728 }
6729 
6730 static int
6731 bge_get_eaddr_eeprom(struct bge_softc *sc, uint8_t ether_addr[])
6732 {
6733 
6734 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
6735 		return (1);
6736 
6737 	return (bge_read_eeprom(sc, ether_addr, BGE_EE_MAC_OFFSET + 2,
6738 	   ETHER_ADDR_LEN));
6739 }
6740 
6741 static int
6742 bge_get_eaddr(struct bge_softc *sc, uint8_t eaddr[])
6743 {
6744 	static const bge_eaddr_fcn_t bge_eaddr_funcs[] = {
6745 		/* NOTE: Order is critical */
6746 		bge_get_eaddr_fw,
6747 		bge_get_eaddr_mem,
6748 		bge_get_eaddr_nvram,
6749 		bge_get_eaddr_eeprom,
6750 		NULL
6751 	};
6752 	const bge_eaddr_fcn_t *func;
6753 
6754 	for (func = bge_eaddr_funcs; *func != NULL; ++func) {
6755 		if ((*func)(sc, eaddr) == 0)
6756 			break;
6757 	}
6758 	return (*func == NULL ? ENXIO : 0);
6759 }
6760 
6761 static uint64_t
6762 bge_get_counter(if_t ifp, ift_counter cnt)
6763 {
6764 	struct bge_softc *sc;
6765 	struct bge_mac_stats *stats;
6766 
6767 	sc = if_getsoftc(ifp);
6768 	if (!BGE_IS_5705_PLUS(sc))
6769 		return (if_get_counter_default(ifp, cnt));
6770 	stats = &sc->bge_mac_stats;
6771 
6772 	switch (cnt) {
6773 	case IFCOUNTER_IERRORS:
6774 		return (stats->NoMoreRxBDs + stats->InputDiscards +
6775 		    stats->InputErrors);
6776 	case IFCOUNTER_COLLISIONS:
6777 		return (stats->etherStatsCollisions);
6778 	default:
6779 		return (if_get_counter_default(ifp, cnt));
6780 	}
6781 }
6782 
6783 #ifdef DEBUGNET
6784 static void
6785 bge_debugnet_init(if_t ifp, int *nrxr, int *ncl, int *clsize)
6786 {
6787 	struct bge_softc *sc;
6788 
6789 	sc = if_getsoftc(ifp);
6790 	BGE_LOCK(sc);
6791 	*nrxr = sc->bge_return_ring_cnt;
6792 	*ncl = DEBUGNET_MAX_IN_FLIGHT;
6793 	if ((sc->bge_flags & BGE_FLAG_JUMBO_STD) != 0 &&
6794 	    (if_getmtu(sc->bge_ifp) + ETHER_HDR_LEN + ETHER_CRC_LEN +
6795 	    ETHER_VLAN_ENCAP_LEN > (MCLBYTES - ETHER_ALIGN)))
6796 		*clsize = MJUM9BYTES;
6797 	else
6798 		*clsize = MCLBYTES;
6799 	BGE_UNLOCK(sc);
6800 }
6801 
6802 static void
6803 bge_debugnet_event(if_t ifp __unused, enum debugnet_ev event __unused)
6804 {
6805 }
6806 
6807 static int
6808 bge_debugnet_transmit(if_t ifp, struct mbuf *m)
6809 {
6810 	struct bge_softc *sc;
6811 	uint32_t prodidx;
6812 	int error;
6813 
6814 	sc = if_getsoftc(ifp);
6815 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
6816 	    IFF_DRV_RUNNING)
6817 		return (1);
6818 
6819 	prodidx = sc->bge_tx_prodidx;
6820 	error = bge_encap(sc, &m, &prodidx);
6821 	if (error == 0)
6822 		bge_start_tx(sc, prodidx);
6823 	return (error);
6824 }
6825 
6826 static int
6827 bge_debugnet_poll(if_t ifp, int count)
6828 {
6829 	struct bge_softc *sc;
6830 	uint32_t rx_prod, tx_cons;
6831 
6832 	sc = if_getsoftc(ifp);
6833 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
6834 	    IFF_DRV_RUNNING)
6835 		return (1);
6836 
6837 	bus_dmamap_sync(sc->bge_cdata.bge_status_tag,
6838 	    sc->bge_cdata.bge_status_map,
6839 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
6840 
6841 	rx_prod = sc->bge_ldata.bge_status_block->bge_idx[0].bge_rx_prod_idx;
6842 	tx_cons = sc->bge_ldata.bge_status_block->bge_idx[0].bge_tx_cons_idx;
6843 
6844 	bus_dmamap_sync(sc->bge_cdata.bge_status_tag,
6845 	    sc->bge_cdata.bge_status_map,
6846 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
6847 
6848 	(void)bge_rxeof(sc, rx_prod, 0);
6849 	bge_txeof(sc, tx_cons);
6850 	return (0);
6851 }
6852 #endif /* DEBUGNET */
6853