1/*-
2 * Copyright (c) 2001 Wind River Systems
3 * Copyright (c) 1997, 1998, 1999, 2001
4 *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 *    notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 *    notice, this list of conditions and the following disclaimer in the
13 *    documentation and/or other materials provided with the distribution.
14 * 3. All advertising materials mentioning features or use of this software
15 *    must display the following acknowledgement:
16 *	This product includes software developed by Bill Paul.
17 * 4. Neither the name of the author nor the names of any co-contributors
18 *    may be used to endorse or promote products derived from this software
19 *    without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
25 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
28 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
29 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
30 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
31 * THE POSSIBILITY OF SUCH DAMAGE.
32 */
33
34#include <sys/cdefs.h>
35__FBSDID("$FreeBSD$");
36
37/*
38 * Broadcom BCM57xx(x)/BCM590x NetXtreme and NetLink family Ethernet driver
39 *
40 * The Broadcom BCM5700 is based on technology originally developed by
41 * Alteon Networks as part of the Tigon I and Tigon II Gigabit Ethernet
42 * MAC chips. The BCM5700, sometimes referred to as the Tigon III, has
43 * two on-board MIPS R4000 CPUs and can have as much as 16MB of external
44 * SSRAM. The BCM5700 supports TCP, UDP and IP checksum offload, jumbo
45 * frames, highly configurable RX filtering, and 16 RX and TX queues
46 * (which, along with RX filter rules, can be used for QOS applications).
47 * Other features, such as TCP segmentation, may be available as part
48 * of value-added firmware updates. Unlike the Tigon I and Tigon II,
49 * firmware images can be stored in hardware and need not be compiled
50 * into the driver.
51 *
52 * The BCM5700 supports the PCI v2.2 and PCI-X v1.0 standards, and will
53 * function in a 32-bit/64-bit 33/66Mhz bus, or a 64-bit/133Mhz bus.
54 *
55 * The BCM5701 is a single-chip solution incorporating both the BCM5700
56 * MAC and a BCM5401 10/100/1000 PHY. Unlike the BCM5700, the BCM5701
57 * does not support external SSRAM.
58 *
59 * Broadcom also produces a variation of the BCM5700 under the "Altima"
60 * brand name, which is functionally similar but lacks PCI-X support.
61 *
62 * Without external SSRAM, you can only have at most 4 TX rings,
63 * and the use of the mini RX ring is disabled. This seems to imply
64 * that these features are simply not available on the BCM5701. As a
65 * result, this driver does not implement any support for the mini RX
66 * ring.
67 */
68
69#ifdef HAVE_KERNEL_OPTION_HEADERS
70#include "opt_device_polling.h"
71#endif
72
73#include <sys/param.h>
74#include <sys/endian.h>
75#include <sys/systm.h>
76#include <sys/sockio.h>
77#include <sys/mbuf.h>
78#include <sys/malloc.h>
79#include <sys/kernel.h>
80#include <sys/module.h>
81#include <sys/socket.h>
82#include <sys/sysctl.h>
83#include <sys/taskqueue.h>
84
85#include <net/if.h>
86#include <net/if_arp.h>
87#include <net/ethernet.h>
88#include <net/if_dl.h>
89#include <net/if_media.h>
90
91#include <net/bpf.h>
92
93#include <net/if_types.h>
94#include <net/if_vlan_var.h>
95
96#include <netinet/in_systm.h>
97#include <netinet/in.h>
98#include <netinet/ip.h>
99#include <netinet/tcp.h>
100
101#include <machine/bus.h>
102#include <machine/resource.h>
103#include <sys/bus.h>
104#include <sys/rman.h>
105
106#include <dev/mii/mii.h>
107#include <dev/mii/miivar.h>
108#include "miidevs.h"
109#include <dev/mii/brgphyreg.h>
110
111#ifdef __sparc64__
112#include <dev/ofw/ofw_bus.h>
113#include <dev/ofw/openfirm.h>
114#include <machine/ofw_machdep.h>
115#include <machine/ver.h>
116#endif
117
118#include <dev/pci/pcireg.h>
119#include <dev/pci/pcivar.h>
120
121#include <dev/bge/if_bgereg.h>
122
123#define	BGE_CSUM_FEATURES	(CSUM_IP | CSUM_TCP)
124#define	ETHER_MIN_NOPAD		(ETHER_MIN_LEN - ETHER_CRC_LEN) /* i.e., 60 */
125
126MODULE_DEPEND(bge, pci, 1, 1, 1);
127MODULE_DEPEND(bge, ether, 1, 1, 1);
128MODULE_DEPEND(bge, miibus, 1, 1, 1);
129
130/* "device miibus" required.  See GENERIC if you get errors here. */
131#include "miibus_if.h"
132
133/*
134 * Various supported device vendors/types and their names. Note: the
135 * spec seems to indicate that the hardware still has Alteon's vendor
136 * ID burned into it, though it will always be overriden by the vendor
137 * ID in the EEPROM. Just to be safe, we cover all possibilities.
138 */
139static const struct bge_type {
140	uint16_t	bge_vid;
141	uint16_t	bge_did;
142} bge_devs[] = {
143	{ ALTEON_VENDORID,	ALTEON_DEVICEID_BCM5700 },
144	{ ALTEON_VENDORID,	ALTEON_DEVICEID_BCM5701 },
145
146	{ ALTIMA_VENDORID,	ALTIMA_DEVICE_AC1000 },
147	{ ALTIMA_VENDORID,	ALTIMA_DEVICE_AC1002 },
148	{ ALTIMA_VENDORID,	ALTIMA_DEVICE_AC9100 },
149
150	{ APPLE_VENDORID,	APPLE_DEVICE_BCM5701 },
151
152	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5700 },
153	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5701 },
154	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5702 },
155	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5702_ALT },
156	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5702X },
157	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5703 },
158	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5703_ALT },
159	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5703X },
160	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5704C },
161	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5704S },
162	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5704S_ALT },
163	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5705 },
164	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5705F },
165	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5705K },
166	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5705M },
167	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5705M_ALT },
168	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5714C },
169	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5714S },
170	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5715 },
171	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5715S },
172	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5717 },
173	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5717C },
174	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5718 },
175	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5719 },
176	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5720 },
177	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5721 },
178	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5722 },
179	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5723 },
180	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5725 },
181	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5727 },
182	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5750 },
183	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5750M },
184	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5751 },
185	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5751F },
186	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5751M },
187	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5752 },
188	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5752M },
189	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5753 },
190	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5753F },
191	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5753M },
192	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5754 },
193	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5754M },
194	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5755 },
195	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5755M },
196	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5756 },
197	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5761 },
198	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5761E },
199	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5761S },
200	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5761SE },
201	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5762 },
202	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5764 },
203	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5780 },
204	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5780S },
205	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5781 },
206	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5782 },
207	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5784 },
208	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5785F },
209	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5785G },
210	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5786 },
211	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5787 },
212	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5787F },
213	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5787M },
214	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5788 },
215	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5789 },
216	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5901 },
217	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5901A2 },
218	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5903M },
219	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5906 },
220	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM5906M },
221	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57760 },
222	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57761 },
223	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57762 },
224	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57764 },
225	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57765 },
226	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57766 },
227	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57767 },
228	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57780 },
229	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57781 },
230	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57782 },
231	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57785 },
232	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57786 },
233	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57787 },
234	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57788 },
235	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57790 },
236	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57791 },
237	{ BCOM_VENDORID,	BCOM_DEVICEID_BCM57795 },
238
239	{ SK_VENDORID,		SK_DEVICEID_ALTIMA },
240
241	{ TC_VENDORID,		TC_DEVICEID_3C996 },
242
243	{ FJTSU_VENDORID,	FJTSU_DEVICEID_PW008GE4 },
244	{ FJTSU_VENDORID,	FJTSU_DEVICEID_PW008GE5 },
245	{ FJTSU_VENDORID,	FJTSU_DEVICEID_PP250450 },
246
247	{ 0, 0 }
248};
249
250static const struct bge_vendor {
251	uint16_t	v_id;
252	const char	*v_name;
253} bge_vendors[] = {
254	{ ALTEON_VENDORID,	"Alteon" },
255	{ ALTIMA_VENDORID,	"Altima" },
256	{ APPLE_VENDORID,	"Apple" },
257	{ BCOM_VENDORID,	"Broadcom" },
258	{ SK_VENDORID,		"SysKonnect" },
259	{ TC_VENDORID,		"3Com" },
260	{ FJTSU_VENDORID,	"Fujitsu" },
261
262	{ 0, NULL }
263};
264
265static const struct bge_revision {
266	uint32_t	br_chipid;
267	const char	*br_name;
268} bge_revisions[] = {
269	{ BGE_CHIPID_BCM5700_A0,	"BCM5700 A0" },
270	{ BGE_CHIPID_BCM5700_A1,	"BCM5700 A1" },
271	{ BGE_CHIPID_BCM5700_B0,	"BCM5700 B0" },
272	{ BGE_CHIPID_BCM5700_B1,	"BCM5700 B1" },
273	{ BGE_CHIPID_BCM5700_B2,	"BCM5700 B2" },
274	{ BGE_CHIPID_BCM5700_B3,	"BCM5700 B3" },
275	{ BGE_CHIPID_BCM5700_ALTIMA,	"BCM5700 Altima" },
276	{ BGE_CHIPID_BCM5700_C0,	"BCM5700 C0" },
277	{ BGE_CHIPID_BCM5701_A0,	"BCM5701 A0" },
278	{ BGE_CHIPID_BCM5701_B0,	"BCM5701 B0" },
279	{ BGE_CHIPID_BCM5701_B2,	"BCM5701 B2" },
280	{ BGE_CHIPID_BCM5701_B5,	"BCM5701 B5" },
281	{ BGE_CHIPID_BCM5703_A0,	"BCM5703 A0" },
282	{ BGE_CHIPID_BCM5703_A1,	"BCM5703 A1" },
283	{ BGE_CHIPID_BCM5703_A2,	"BCM5703 A2" },
284	{ BGE_CHIPID_BCM5703_A3,	"BCM5703 A3" },
285	{ BGE_CHIPID_BCM5703_B0,	"BCM5703 B0" },
286	{ BGE_CHIPID_BCM5704_A0,	"BCM5704 A0" },
287	{ BGE_CHIPID_BCM5704_A1,	"BCM5704 A1" },
288	{ BGE_CHIPID_BCM5704_A2,	"BCM5704 A2" },
289	{ BGE_CHIPID_BCM5704_A3,	"BCM5704 A3" },
290	{ BGE_CHIPID_BCM5704_B0,	"BCM5704 B0" },
291	{ BGE_CHIPID_BCM5705_A0,	"BCM5705 A0" },
292	{ BGE_CHIPID_BCM5705_A1,	"BCM5705 A1" },
293	{ BGE_CHIPID_BCM5705_A2,	"BCM5705 A2" },
294	{ BGE_CHIPID_BCM5705_A3,	"BCM5705 A3" },
295	{ BGE_CHIPID_BCM5750_A0,	"BCM5750 A0" },
296	{ BGE_CHIPID_BCM5750_A1,	"BCM5750 A1" },
297	{ BGE_CHIPID_BCM5750_A3,	"BCM5750 A3" },
298	{ BGE_CHIPID_BCM5750_B0,	"BCM5750 B0" },
299	{ BGE_CHIPID_BCM5750_B1,	"BCM5750 B1" },
300	{ BGE_CHIPID_BCM5750_C0,	"BCM5750 C0" },
301	{ BGE_CHIPID_BCM5750_C1,	"BCM5750 C1" },
302	{ BGE_CHIPID_BCM5750_C2,	"BCM5750 C2" },
303	{ BGE_CHIPID_BCM5714_A0,	"BCM5714 A0" },
304	{ BGE_CHIPID_BCM5752_A0,	"BCM5752 A0" },
305	{ BGE_CHIPID_BCM5752_A1,	"BCM5752 A1" },
306	{ BGE_CHIPID_BCM5752_A2,	"BCM5752 A2" },
307	{ BGE_CHIPID_BCM5714_B0,	"BCM5714 B0" },
308	{ BGE_CHIPID_BCM5714_B3,	"BCM5714 B3" },
309	{ BGE_CHIPID_BCM5715_A0,	"BCM5715 A0" },
310	{ BGE_CHIPID_BCM5715_A1,	"BCM5715 A1" },
311	{ BGE_CHIPID_BCM5715_A3,	"BCM5715 A3" },
312	{ BGE_CHIPID_BCM5717_A0,	"BCM5717 A0" },
313	{ BGE_CHIPID_BCM5717_B0,	"BCM5717 B0" },
314	{ BGE_CHIPID_BCM5717_C0,	"BCM5717 C0" },
315	{ BGE_CHIPID_BCM5719_A0,	"BCM5719 A0" },
316	{ BGE_CHIPID_BCM5720_A0,	"BCM5720 A0" },
317	{ BGE_CHIPID_BCM5755_A0,	"BCM5755 A0" },
318	{ BGE_CHIPID_BCM5755_A1,	"BCM5755 A1" },
319	{ BGE_CHIPID_BCM5755_A2,	"BCM5755 A2" },
320	{ BGE_CHIPID_BCM5722_A0,	"BCM5722 A0" },
321	{ BGE_CHIPID_BCM5761_A0,	"BCM5761 A0" },
322	{ BGE_CHIPID_BCM5761_A1,	"BCM5761 A1" },
323	{ BGE_CHIPID_BCM5762_A0,	"BCM5762 A0" },
324	{ BGE_CHIPID_BCM5784_A0,	"BCM5784 A0" },
325	{ BGE_CHIPID_BCM5784_A1,	"BCM5784 A1" },
326	/* 5754 and 5787 share the same ASIC ID */
327	{ BGE_CHIPID_BCM5787_A0,	"BCM5754/5787 A0" },
328	{ BGE_CHIPID_BCM5787_A1,	"BCM5754/5787 A1" },
329	{ BGE_CHIPID_BCM5787_A2,	"BCM5754/5787 A2" },
330	{ BGE_CHIPID_BCM5906_A1,	"BCM5906 A1" },
331	{ BGE_CHIPID_BCM5906_A2,	"BCM5906 A2" },
332	{ BGE_CHIPID_BCM57765_A0,	"BCM57765 A0" },
333	{ BGE_CHIPID_BCM57765_B0,	"BCM57765 B0" },
334	{ BGE_CHIPID_BCM57780_A0,	"BCM57780 A0" },
335	{ BGE_CHIPID_BCM57780_A1,	"BCM57780 A1" },
336
337	{ 0, NULL }
338};
339
340/*
341 * Some defaults for major revisions, so that newer steppings
342 * that we don't know about have a shot at working.
343 */
344static const struct bge_revision bge_majorrevs[] = {
345	{ BGE_ASICREV_BCM5700,		"unknown BCM5700" },
346	{ BGE_ASICREV_BCM5701,		"unknown BCM5701" },
347	{ BGE_ASICREV_BCM5703,		"unknown BCM5703" },
348	{ BGE_ASICREV_BCM5704,		"unknown BCM5704" },
349	{ BGE_ASICREV_BCM5705,		"unknown BCM5705" },
350	{ BGE_ASICREV_BCM5750,		"unknown BCM5750" },
351	{ BGE_ASICREV_BCM5714_A0,	"unknown BCM5714" },
352	{ BGE_ASICREV_BCM5752,		"unknown BCM5752" },
353	{ BGE_ASICREV_BCM5780,		"unknown BCM5780" },
354	{ BGE_ASICREV_BCM5714,		"unknown BCM5714" },
355	{ BGE_ASICREV_BCM5755,		"unknown BCM5755" },
356	{ BGE_ASICREV_BCM5761,		"unknown BCM5761" },
357	{ BGE_ASICREV_BCM5784,		"unknown BCM5784" },
358	{ BGE_ASICREV_BCM5785,		"unknown BCM5785" },
359	/* 5754 and 5787 share the same ASIC ID */
360	{ BGE_ASICREV_BCM5787,		"unknown BCM5754/5787" },
361	{ BGE_ASICREV_BCM5906,		"unknown BCM5906" },
362	{ BGE_ASICREV_BCM57765,		"unknown BCM57765" },
363	{ BGE_ASICREV_BCM57766,		"unknown BCM57766" },
364	{ BGE_ASICREV_BCM57780,		"unknown BCM57780" },
365	{ BGE_ASICREV_BCM5717,		"unknown BCM5717" },
366	{ BGE_ASICREV_BCM5719,		"unknown BCM5719" },
367	{ BGE_ASICREV_BCM5720,		"unknown BCM5720" },
368	{ BGE_ASICREV_BCM5762,		"unknown BCM5762" },
369
370	{ 0, NULL }
371};
372
373#define	BGE_IS_JUMBO_CAPABLE(sc)	((sc)->bge_flags & BGE_FLAG_JUMBO)
374#define	BGE_IS_5700_FAMILY(sc)		((sc)->bge_flags & BGE_FLAG_5700_FAMILY)
375#define	BGE_IS_5705_PLUS(sc)		((sc)->bge_flags & BGE_FLAG_5705_PLUS)
376#define	BGE_IS_5714_FAMILY(sc)		((sc)->bge_flags & BGE_FLAG_5714_FAMILY)
377#define	BGE_IS_575X_PLUS(sc)		((sc)->bge_flags & BGE_FLAG_575X_PLUS)
378#define	BGE_IS_5755_PLUS(sc)		((sc)->bge_flags & BGE_FLAG_5755_PLUS)
379#define	BGE_IS_5717_PLUS(sc)		((sc)->bge_flags & BGE_FLAG_5717_PLUS)
380#define	BGE_IS_57765_PLUS(sc)		((sc)->bge_flags & BGE_FLAG_57765_PLUS)
381
382static uint32_t bge_chipid(device_t);
383static const struct bge_vendor * bge_lookup_vendor(uint16_t);
384static const struct bge_revision * bge_lookup_rev(uint32_t);
385
386typedef int	(*bge_eaddr_fcn_t)(struct bge_softc *, uint8_t[]);
387
388static int bge_probe(device_t);
389static int bge_attach(device_t);
390static int bge_detach(device_t);
391static int bge_suspend(device_t);
392static int bge_resume(device_t);
393static void bge_release_resources(struct bge_softc *);
394static void bge_dma_map_addr(void *, bus_dma_segment_t *, int, int);
395static int bge_dma_alloc(struct bge_softc *);
396static void bge_dma_free(struct bge_softc *);
397static int bge_dma_ring_alloc(struct bge_softc *, bus_size_t, bus_size_t,
398    bus_dma_tag_t *, uint8_t **, bus_dmamap_t *, bus_addr_t *, const char *);
399
400static void bge_devinfo(struct bge_softc *);
401static int bge_mbox_reorder(struct bge_softc *);
402
403static int bge_get_eaddr_fw(struct bge_softc *sc, uint8_t ether_addr[]);
404static int bge_get_eaddr_mem(struct bge_softc *, uint8_t[]);
405static int bge_get_eaddr_nvram(struct bge_softc *, uint8_t[]);
406static int bge_get_eaddr_eeprom(struct bge_softc *, uint8_t[]);
407static int bge_get_eaddr(struct bge_softc *, uint8_t[]);
408
409static void bge_txeof(struct bge_softc *, uint16_t);
410static void bge_rxcsum(struct bge_softc *, struct bge_rx_bd *, struct mbuf *);
411static int bge_rxeof(struct bge_softc *, uint16_t, int);
412
413static void bge_asf_driver_up (struct bge_softc *);
414static void bge_tick(void *);
415static void bge_stats_clear_regs(struct bge_softc *);
416static void bge_stats_update(struct bge_softc *);
417static void bge_stats_update_regs(struct bge_softc *);
418static struct mbuf *bge_check_short_dma(struct mbuf *);
419static struct mbuf *bge_setup_tso(struct bge_softc *, struct mbuf *,
420    uint16_t *, uint16_t *);
421static int bge_encap(struct bge_softc *, struct mbuf **, uint32_t *);
422
423static void bge_intr(void *);
424static int bge_msi_intr(void *);
425static void bge_intr_task(void *, int);
426static void bge_start_locked(struct ifnet *);
427static void bge_start(struct ifnet *);
428static int bge_ioctl(struct ifnet *, u_long, caddr_t);
429static void bge_init_locked(struct bge_softc *);
430static void bge_init(void *);
431static void bge_stop_block(struct bge_softc *, bus_size_t, uint32_t);
432static void bge_stop(struct bge_softc *);
433static void bge_watchdog(struct bge_softc *);
434static int bge_shutdown(device_t);
435static int bge_ifmedia_upd_locked(struct ifnet *);
436static int bge_ifmedia_upd(struct ifnet *);
437static void bge_ifmedia_sts(struct ifnet *, struct ifmediareq *);
438
439static uint8_t bge_nvram_getbyte(struct bge_softc *, int, uint8_t *);
440static int bge_read_nvram(struct bge_softc *, caddr_t, int, int);
441
442static uint8_t bge_eeprom_getbyte(struct bge_softc *, int, uint8_t *);
443static int bge_read_eeprom(struct bge_softc *, caddr_t, int, int);
444
445static void bge_setpromisc(struct bge_softc *);
446static void bge_setmulti(struct bge_softc *);
447static void bge_setvlan(struct bge_softc *);
448
449static __inline void bge_rxreuse_std(struct bge_softc *, int);
450static __inline void bge_rxreuse_jumbo(struct bge_softc *, int);
451static int bge_newbuf_std(struct bge_softc *, int);
452static int bge_newbuf_jumbo(struct bge_softc *, int);
453static int bge_init_rx_ring_std(struct bge_softc *);
454static void bge_free_rx_ring_std(struct bge_softc *);
455static int bge_init_rx_ring_jumbo(struct bge_softc *);
456static void bge_free_rx_ring_jumbo(struct bge_softc *);
457static void bge_free_tx_ring(struct bge_softc *);
458static int bge_init_tx_ring(struct bge_softc *);
459
460static int bge_chipinit(struct bge_softc *);
461static int bge_blockinit(struct bge_softc *);
462static uint32_t bge_dma_swap_options(struct bge_softc *);
463
464static int bge_has_eaddr(struct bge_softc *);
465static uint32_t bge_readmem_ind(struct bge_softc *, int);
466static void bge_writemem_ind(struct bge_softc *, int, int);
467static void bge_writembx(struct bge_softc *, int, int);
468#ifdef notdef
469static uint32_t bge_readreg_ind(struct bge_softc *, int);
470#endif
471static void bge_writemem_direct(struct bge_softc *, int, int);
472static void bge_writereg_ind(struct bge_softc *, int, int);
473
474static int bge_miibus_readreg(device_t, int, int);
475static int bge_miibus_writereg(device_t, int, int, int);
476static void bge_miibus_statchg(device_t);
477#ifdef DEVICE_POLLING
478static int bge_poll(struct ifnet *ifp, enum poll_cmd cmd, int count);
479#endif
480
481#define	BGE_RESET_SHUTDOWN	0
482#define	BGE_RESET_START		1
483#define	BGE_RESET_SUSPEND	2
484static void bge_sig_post_reset(struct bge_softc *, int);
485static void bge_sig_legacy(struct bge_softc *, int);
486static void bge_sig_pre_reset(struct bge_softc *, int);
487static void bge_stop_fw(struct bge_softc *);
488static int bge_reset(struct bge_softc *);
489static void bge_link_upd(struct bge_softc *);
490
491static void bge_ape_lock_init(struct bge_softc *);
492static void bge_ape_read_fw_ver(struct bge_softc *);
493static int bge_ape_lock(struct bge_softc *, int);
494static void bge_ape_unlock(struct bge_softc *, int);
495static void bge_ape_send_event(struct bge_softc *, uint32_t);
496static void bge_ape_driver_state_change(struct bge_softc *, int);
497
498/*
499 * The BGE_REGISTER_DEBUG option is only for low-level debugging.  It may
500 * leak information to untrusted users.  It is also known to cause alignment
501 * traps on certain architectures.
502 */
503#ifdef BGE_REGISTER_DEBUG
504static int bge_sysctl_debug_info(SYSCTL_HANDLER_ARGS);
505static int bge_sysctl_reg_read(SYSCTL_HANDLER_ARGS);
506static int bge_sysctl_ape_read(SYSCTL_HANDLER_ARGS);
507static int bge_sysctl_mem_read(SYSCTL_HANDLER_ARGS);
508#endif
509static void bge_add_sysctls(struct bge_softc *);
510static void bge_add_sysctl_stats_regs(struct bge_softc *,
511    struct sysctl_ctx_list *, struct sysctl_oid_list *);
512static void bge_add_sysctl_stats(struct bge_softc *, struct sysctl_ctx_list *,
513    struct sysctl_oid_list *);
514static int bge_sysctl_stats(SYSCTL_HANDLER_ARGS);
515
516static device_method_t bge_methods[] = {
517	/* Device interface */
518	DEVMETHOD(device_probe,		bge_probe),
519	DEVMETHOD(device_attach,	bge_attach),
520	DEVMETHOD(device_detach,	bge_detach),
521	DEVMETHOD(device_shutdown,	bge_shutdown),
522	DEVMETHOD(device_suspend,	bge_suspend),
523	DEVMETHOD(device_resume,	bge_resume),
524
525	/* MII interface */
526	DEVMETHOD(miibus_readreg,	bge_miibus_readreg),
527	DEVMETHOD(miibus_writereg,	bge_miibus_writereg),
528	DEVMETHOD(miibus_statchg,	bge_miibus_statchg),
529
530	DEVMETHOD_END
531};
532
533static driver_t bge_driver = {
534	"bge",
535	bge_methods,
536	sizeof(struct bge_softc)
537};
538
539static devclass_t bge_devclass;
540
541DRIVER_MODULE(bge, pci, bge_driver, bge_devclass, 0, 0);
542DRIVER_MODULE(miibus, bge, miibus_driver, miibus_devclass, 0, 0);
543
544static int bge_allow_asf = 1;
545
546TUNABLE_INT("hw.bge.allow_asf", &bge_allow_asf);
547
548static SYSCTL_NODE(_hw, OID_AUTO, bge, CTLFLAG_RD, 0, "BGE driver parameters");
549SYSCTL_INT(_hw_bge, OID_AUTO, allow_asf, CTLFLAG_RD, &bge_allow_asf, 0,
550	"Allow ASF mode if available");
551
552#define	SPARC64_BLADE_1500_MODEL	"SUNW,Sun-Blade-1500"
553#define	SPARC64_BLADE_1500_PATH_BGE	"/pci@1f,700000/network@2"
554#define	SPARC64_BLADE_2500_MODEL	"SUNW,Sun-Blade-2500"
555#define	SPARC64_BLADE_2500_PATH_BGE	"/pci@1c,600000/network@3"
556#define	SPARC64_OFW_SUBVENDOR		"subsystem-vendor-id"
557
558static int
559bge_has_eaddr(struct bge_softc *sc)
560{
561#ifdef __sparc64__
562	char buf[sizeof(SPARC64_BLADE_1500_PATH_BGE)];
563	device_t dev;
564	uint32_t subvendor;
565
566	dev = sc->bge_dev;
567
568	/*
569	 * The on-board BGEs found in sun4u machines aren't fitted with
570	 * an EEPROM which means that we have to obtain the MAC address
571	 * via OFW and that some tests will always fail.  We distinguish
572	 * such BGEs by the subvendor ID, which also has to be obtained
573	 * from OFW instead of the PCI configuration space as the latter
574	 * indicates Broadcom as the subvendor of the netboot interface.
575	 * For early Blade 1500 and 2500 we even have to check the OFW
576	 * device path as the subvendor ID always defaults to Broadcom
577	 * there.
578	 */
579	if (OF_getprop(ofw_bus_get_node(dev), SPARC64_OFW_SUBVENDOR,
580	    &subvendor, sizeof(subvendor)) == sizeof(subvendor) &&
581	    (subvendor == FJTSU_VENDORID || subvendor == SUN_VENDORID))
582		return (0);
583	memset(buf, 0, sizeof(buf));
584	if (OF_package_to_path(ofw_bus_get_node(dev), buf, sizeof(buf)) > 0) {
585		if (strcmp(sparc64_model, SPARC64_BLADE_1500_MODEL) == 0 &&
586		    strcmp(buf, SPARC64_BLADE_1500_PATH_BGE) == 0)
587			return (0);
588		if (strcmp(sparc64_model, SPARC64_BLADE_2500_MODEL) == 0 &&
589		    strcmp(buf, SPARC64_BLADE_2500_PATH_BGE) == 0)
590			return (0);
591	}
592#endif
593	return (1);
594}
595
596static uint32_t
597bge_readmem_ind(struct bge_softc *sc, int off)
598{
599	device_t dev;
600	uint32_t val;
601
602	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
603	    off >= BGE_STATS_BLOCK && off < BGE_SEND_RING_1_TO_4)
604		return (0);
605
606	dev = sc->bge_dev;
607
608	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4);
609	val = pci_read_config(dev, BGE_PCI_MEMWIN_DATA, 4);
610	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4);
611	return (val);
612}
613
614static void
615bge_writemem_ind(struct bge_softc *sc, int off, int val)
616{
617	device_t dev;
618
619	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
620	    off >= BGE_STATS_BLOCK && off < BGE_SEND_RING_1_TO_4)
621		return;
622
623	dev = sc->bge_dev;
624
625	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4);
626	pci_write_config(dev, BGE_PCI_MEMWIN_DATA, val, 4);
627	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4);
628}
629
630#ifdef notdef
631static uint32_t
632bge_readreg_ind(struct bge_softc *sc, int off)
633{
634	device_t dev;
635
636	dev = sc->bge_dev;
637
638	pci_write_config(dev, BGE_PCI_REG_BASEADDR, off, 4);
639	return (pci_read_config(dev, BGE_PCI_REG_DATA, 4));
640}
641#endif
642
643static void
644bge_writereg_ind(struct bge_softc *sc, int off, int val)
645{
646	device_t dev;
647
648	dev = sc->bge_dev;
649
650	pci_write_config(dev, BGE_PCI_REG_BASEADDR, off, 4);
651	pci_write_config(dev, BGE_PCI_REG_DATA, val, 4);
652}
653
654static void
655bge_writemem_direct(struct bge_softc *sc, int off, int val)
656{
657	CSR_WRITE_4(sc, off, val);
658}
659
660static void
661bge_writembx(struct bge_softc *sc, int off, int val)
662{
663	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
664		off += BGE_LPMBX_IRQ0_HI - BGE_MBX_IRQ0_HI;
665
666	CSR_WRITE_4(sc, off, val);
667	if ((sc->bge_flags & BGE_FLAG_MBOX_REORDER) != 0)
668		CSR_READ_4(sc, off);
669}
670
671/*
672 * Clear all stale locks and select the lock for this driver instance.
673 */
674static void
675bge_ape_lock_init(struct bge_softc *sc)
676{
677	uint32_t bit, regbase;
678	int i;
679
680	if (sc->bge_asicrev == BGE_ASICREV_BCM5761)
681		regbase = BGE_APE_LOCK_GRANT;
682	else
683		regbase = BGE_APE_PER_LOCK_GRANT;
684
685	/* Clear any stale locks. */
686	for (i = BGE_APE_LOCK_PHY0; i <= BGE_APE_LOCK_GPIO; i++) {
687		switch (i) {
688		case BGE_APE_LOCK_PHY0:
689		case BGE_APE_LOCK_PHY1:
690		case BGE_APE_LOCK_PHY2:
691		case BGE_APE_LOCK_PHY3:
692			bit = BGE_APE_LOCK_GRANT_DRIVER0;
693			break;
694		default:
695			if (sc->bge_func_addr == 0)
696				bit = BGE_APE_LOCK_GRANT_DRIVER0;
697			else
698				bit = (1 << sc->bge_func_addr);
699		}
700		APE_WRITE_4(sc, regbase + 4 * i, bit);
701	}
702
703	/* Select the PHY lock based on the device's function number. */
704	switch (sc->bge_func_addr) {
705	case 0:
706		sc->bge_phy_ape_lock = BGE_APE_LOCK_PHY0;
707		break;
708	case 1:
709		sc->bge_phy_ape_lock = BGE_APE_LOCK_PHY1;
710		break;
711	case 2:
712		sc->bge_phy_ape_lock = BGE_APE_LOCK_PHY2;
713		break;
714	case 3:
715		sc->bge_phy_ape_lock = BGE_APE_LOCK_PHY3;
716		break;
717	default:
718		device_printf(sc->bge_dev,
719		    "PHY lock not supported on this function\n");
720	}
721}
722
723/*
724 * Check for APE firmware, set flags, and print version info.
725 */
726static void
727bge_ape_read_fw_ver(struct bge_softc *sc)
728{
729	const char *fwtype;
730	uint32_t apedata, features;
731
732	/* Check for a valid APE signature in shared memory. */
733	apedata = APE_READ_4(sc, BGE_APE_SEG_SIG);
734	if (apedata != BGE_APE_SEG_SIG_MAGIC) {
735		sc->bge_mfw_flags &= ~ BGE_MFW_ON_APE;
736		return;
737	}
738
739	/* Check if APE firmware is running. */
740	apedata = APE_READ_4(sc, BGE_APE_FW_STATUS);
741	if ((apedata & BGE_APE_FW_STATUS_READY) == 0) {
742		device_printf(sc->bge_dev, "APE signature found "
743		    "but FW status not ready! 0x%08x\n", apedata);
744		return;
745	}
746
747	sc->bge_mfw_flags |= BGE_MFW_ON_APE;
748
749	/* Fetch the APE firwmare type and version. */
750	apedata = APE_READ_4(sc, BGE_APE_FW_VERSION);
751	features = APE_READ_4(sc, BGE_APE_FW_FEATURES);
752	if ((features & BGE_APE_FW_FEATURE_NCSI) != 0) {
753		sc->bge_mfw_flags |= BGE_MFW_TYPE_NCSI;
754		fwtype = "NCSI";
755	} else if ((features & BGE_APE_FW_FEATURE_DASH) != 0) {
756		sc->bge_mfw_flags |= BGE_MFW_TYPE_DASH;
757		fwtype = "DASH";
758	} else
759		fwtype = "UNKN";
760
761	/* Print the APE firmware version. */
762	device_printf(sc->bge_dev, "APE FW version: %s v%d.%d.%d.%d\n",
763	    fwtype,
764	    (apedata & BGE_APE_FW_VERSION_MAJMSK) >> BGE_APE_FW_VERSION_MAJSFT,
765	    (apedata & BGE_APE_FW_VERSION_MINMSK) >> BGE_APE_FW_VERSION_MINSFT,
766	    (apedata & BGE_APE_FW_VERSION_REVMSK) >> BGE_APE_FW_VERSION_REVSFT,
767	    (apedata & BGE_APE_FW_VERSION_BLDMSK));
768}
769
770static int
771bge_ape_lock(struct bge_softc *sc, int locknum)
772{
773	uint32_t bit, gnt, req, status;
774	int i, off;
775
776	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) == 0)
777		return (0);
778
779	/* Lock request/grant registers have different bases. */
780	if (sc->bge_asicrev == BGE_ASICREV_BCM5761) {
781		req = BGE_APE_LOCK_REQ;
782		gnt = BGE_APE_LOCK_GRANT;
783	} else {
784		req = BGE_APE_PER_LOCK_REQ;
785		gnt = BGE_APE_PER_LOCK_GRANT;
786	}
787
788	off = 4 * locknum;
789
790	switch (locknum) {
791	case BGE_APE_LOCK_GPIO:
792		/* Lock required when using GPIO. */
793		if (sc->bge_asicrev == BGE_ASICREV_BCM5761)
794			return (0);
795		if (sc->bge_func_addr == 0)
796			bit = BGE_APE_LOCK_REQ_DRIVER0;
797		else
798			bit = (1 << sc->bge_func_addr);
799		break;
800	case BGE_APE_LOCK_GRC:
801		/* Lock required to reset the device. */
802		if (sc->bge_func_addr == 0)
803			bit = BGE_APE_LOCK_REQ_DRIVER0;
804		else
805			bit = (1 << sc->bge_func_addr);
806		break;
807	case BGE_APE_LOCK_MEM:
808		/* Lock required when accessing certain APE memory. */
809		if (sc->bge_func_addr == 0)
810			bit = BGE_APE_LOCK_REQ_DRIVER0;
811		else
812			bit = (1 << sc->bge_func_addr);
813		break;
814	case BGE_APE_LOCK_PHY0:
815	case BGE_APE_LOCK_PHY1:
816	case BGE_APE_LOCK_PHY2:
817	case BGE_APE_LOCK_PHY3:
818		/* Lock required when accessing PHYs. */
819		bit = BGE_APE_LOCK_REQ_DRIVER0;
820		break;
821	default:
822		return (EINVAL);
823	}
824
825	/* Request a lock. */
826	APE_WRITE_4(sc, req + off, bit);
827
828	/* Wait up to 1 second to acquire lock. */
829	for (i = 0; i < 20000; i++) {
830		status = APE_READ_4(sc, gnt + off);
831		if (status == bit)
832			break;
833		DELAY(50);
834	}
835
836	/* Handle any errors. */
837	if (status != bit) {
838		device_printf(sc->bge_dev, "APE lock %d request failed! "
839		    "request = 0x%04x[0x%04x], status = 0x%04x[0x%04x]\n",
840		    locknum, req + off, bit & 0xFFFF, gnt + off,
841		    status & 0xFFFF);
842		/* Revoke the lock request. */
843		APE_WRITE_4(sc, gnt + off, bit);
844		return (EBUSY);
845	}
846
847	return (0);
848}
849
850static void
851bge_ape_unlock(struct bge_softc *sc, int locknum)
852{
853	uint32_t bit, gnt;
854	int off;
855
856	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) == 0)
857		return;
858
859	if (sc->bge_asicrev == BGE_ASICREV_BCM5761)
860		gnt = BGE_APE_LOCK_GRANT;
861	else
862		gnt = BGE_APE_PER_LOCK_GRANT;
863
864	off = 4 * locknum;
865
866	switch (locknum) {
867	case BGE_APE_LOCK_GPIO:
868		if (sc->bge_asicrev == BGE_ASICREV_BCM5761)
869			return;
870		if (sc->bge_func_addr == 0)
871			bit = BGE_APE_LOCK_GRANT_DRIVER0;
872		else
873			bit = (1 << sc->bge_func_addr);
874		break;
875	case BGE_APE_LOCK_GRC:
876		if (sc->bge_func_addr == 0)
877			bit = BGE_APE_LOCK_GRANT_DRIVER0;
878		else
879			bit = (1 << sc->bge_func_addr);
880		break;
881	case BGE_APE_LOCK_MEM:
882		if (sc->bge_func_addr == 0)
883			bit = BGE_APE_LOCK_GRANT_DRIVER0;
884		else
885			bit = (1 << sc->bge_func_addr);
886		break;
887	case BGE_APE_LOCK_PHY0:
888	case BGE_APE_LOCK_PHY1:
889	case BGE_APE_LOCK_PHY2:
890	case BGE_APE_LOCK_PHY3:
891		bit = BGE_APE_LOCK_GRANT_DRIVER0;
892		break;
893	default:
894		return;
895	}
896
897	APE_WRITE_4(sc, gnt + off, bit);
898}
899
900/*
901 * Send an event to the APE firmware.
902 */
903static void
904bge_ape_send_event(struct bge_softc *sc, uint32_t event)
905{
906	uint32_t apedata;
907	int i;
908
909	/* NCSI does not support APE events. */
910	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) == 0)
911		return;
912
913	/* Wait up to 1ms for APE to service previous event. */
914	for (i = 10; i > 0; i--) {
915		if (bge_ape_lock(sc, BGE_APE_LOCK_MEM) != 0)
916			break;
917		apedata = APE_READ_4(sc, BGE_APE_EVENT_STATUS);
918		if ((apedata & BGE_APE_EVENT_STATUS_EVENT_PENDING) == 0) {
919			APE_WRITE_4(sc, BGE_APE_EVENT_STATUS, event |
920			    BGE_APE_EVENT_STATUS_EVENT_PENDING);
921			bge_ape_unlock(sc, BGE_APE_LOCK_MEM);
922			APE_WRITE_4(sc, BGE_APE_EVENT, BGE_APE_EVENT_1);
923			break;
924		}
925		bge_ape_unlock(sc, BGE_APE_LOCK_MEM);
926		DELAY(100);
927	}
928	if (i == 0)
929		device_printf(sc->bge_dev, "APE event 0x%08x send timed out\n",
930		    event);
931}
932
933static void
934bge_ape_driver_state_change(struct bge_softc *sc, int kind)
935{
936	uint32_t apedata, event;
937
938	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) == 0)
939		return;
940
941	switch (kind) {
942	case BGE_RESET_START:
943		/* If this is the first load, clear the load counter. */
944		apedata = APE_READ_4(sc, BGE_APE_HOST_SEG_SIG);
945		if (apedata != BGE_APE_HOST_SEG_SIG_MAGIC)
946			APE_WRITE_4(sc, BGE_APE_HOST_INIT_COUNT, 0);
947		else {
948			apedata = APE_READ_4(sc, BGE_APE_HOST_INIT_COUNT);
949			APE_WRITE_4(sc, BGE_APE_HOST_INIT_COUNT, ++apedata);
950		}
951		APE_WRITE_4(sc, BGE_APE_HOST_SEG_SIG,
952		    BGE_APE_HOST_SEG_SIG_MAGIC);
953		APE_WRITE_4(sc, BGE_APE_HOST_SEG_LEN,
954		    BGE_APE_HOST_SEG_LEN_MAGIC);
955
956		/* Add some version info if bge(4) supports it. */
957		APE_WRITE_4(sc, BGE_APE_HOST_DRIVER_ID,
958		    BGE_APE_HOST_DRIVER_ID_MAGIC(1, 0));
959		APE_WRITE_4(sc, BGE_APE_HOST_BEHAVIOR,
960		    BGE_APE_HOST_BEHAV_NO_PHYLOCK);
961		APE_WRITE_4(sc, BGE_APE_HOST_HEARTBEAT_INT_MS,
962		    BGE_APE_HOST_HEARTBEAT_INT_DISABLE);
963		APE_WRITE_4(sc, BGE_APE_HOST_DRVR_STATE,
964		    BGE_APE_HOST_DRVR_STATE_START);
965		event = BGE_APE_EVENT_STATUS_STATE_START;
966		break;
967	case BGE_RESET_SHUTDOWN:
968		APE_WRITE_4(sc, BGE_APE_HOST_DRVR_STATE,
969		    BGE_APE_HOST_DRVR_STATE_UNLOAD);
970		event = BGE_APE_EVENT_STATUS_STATE_UNLOAD;
971		break;
972	case BGE_RESET_SUSPEND:
973		event = BGE_APE_EVENT_STATUS_STATE_SUSPEND;
974		break;
975	default:
976		return;
977	}
978
979	bge_ape_send_event(sc, event | BGE_APE_EVENT_STATUS_DRIVER_EVNT |
980	    BGE_APE_EVENT_STATUS_STATE_CHNGE);
981}
982
983/*
984 * Map a single buffer address.
985 */
986
987static void
988bge_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
989{
990	struct bge_dmamap_arg *ctx;
991
992	if (error)
993		return;
994
995	KASSERT(nseg == 1, ("%s: %d segments returned!", __func__, nseg));
996
997	ctx = arg;
998	ctx->bge_busaddr = segs->ds_addr;
999}
1000
1001static uint8_t
1002bge_nvram_getbyte(struct bge_softc *sc, int addr, uint8_t *dest)
1003{
1004	uint32_t access, byte = 0;
1005	int i;
1006
1007	/* Lock. */
1008	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_SET1);
1009	for (i = 0; i < 8000; i++) {
1010		if (CSR_READ_4(sc, BGE_NVRAM_SWARB) & BGE_NVRAMSWARB_GNT1)
1011			break;
1012		DELAY(20);
1013	}
1014	if (i == 8000)
1015		return (1);
1016
1017	/* Enable access. */
1018	access = CSR_READ_4(sc, BGE_NVRAM_ACCESS);
1019	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access | BGE_NVRAMACC_ENABLE);
1020
1021	CSR_WRITE_4(sc, BGE_NVRAM_ADDR, addr & 0xfffffffc);
1022	CSR_WRITE_4(sc, BGE_NVRAM_CMD, BGE_NVRAM_READCMD);
1023	for (i = 0; i < BGE_TIMEOUT * 10; i++) {
1024		DELAY(10);
1025		if (CSR_READ_4(sc, BGE_NVRAM_CMD) & BGE_NVRAMCMD_DONE) {
1026			DELAY(10);
1027			break;
1028		}
1029	}
1030
1031	if (i == BGE_TIMEOUT * 10) {
1032		if_printf(sc->bge_ifp, "nvram read timed out\n");
1033		return (1);
1034	}
1035
1036	/* Get result. */
1037	byte = CSR_READ_4(sc, BGE_NVRAM_RDDATA);
1038
1039	*dest = (bswap32(byte) >> ((addr % 4) * 8)) & 0xFF;
1040
1041	/* Disable access. */
1042	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access);
1043
1044	/* Unlock. */
1045	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_CLR1);
1046	CSR_READ_4(sc, BGE_NVRAM_SWARB);
1047
1048	return (0);
1049}
1050
1051/*
1052 * Read a sequence of bytes from NVRAM.
1053 */
1054static int
1055bge_read_nvram(struct bge_softc *sc, caddr_t dest, int off, int cnt)
1056{
1057	int err = 0, i;
1058	uint8_t byte = 0;
1059
1060	if (sc->bge_asicrev != BGE_ASICREV_BCM5906)
1061		return (1);
1062
1063	for (i = 0; i < cnt; i++) {
1064		err = bge_nvram_getbyte(sc, off + i, &byte);
1065		if (err)
1066			break;
1067		*(dest + i) = byte;
1068	}
1069
1070	return (err ? 1 : 0);
1071}
1072
1073/*
1074 * Read a byte of data stored in the EEPROM at address 'addr.' The
1075 * BCM570x supports both the traditional bitbang interface and an
1076 * auto access interface for reading the EEPROM. We use the auto
1077 * access method.
1078 */
1079static uint8_t
1080bge_eeprom_getbyte(struct bge_softc *sc, int addr, uint8_t *dest)
1081{
1082	int i;
1083	uint32_t byte = 0;
1084
1085	/*
1086	 * Enable use of auto EEPROM access so we can avoid
1087	 * having to use the bitbang method.
1088	 */
1089	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_AUTO_EEPROM);
1090
1091	/* Reset the EEPROM, load the clock period. */
1092	CSR_WRITE_4(sc, BGE_EE_ADDR,
1093	    BGE_EEADDR_RESET | BGE_EEHALFCLK(BGE_HALFCLK_384SCL));
1094	DELAY(20);
1095
1096	/* Issue the read EEPROM command. */
1097	CSR_WRITE_4(sc, BGE_EE_ADDR, BGE_EE_READCMD | addr);
1098
1099	/* Wait for completion */
1100	for(i = 0; i < BGE_TIMEOUT * 10; i++) {
1101		DELAY(10);
1102		if (CSR_READ_4(sc, BGE_EE_ADDR) & BGE_EEADDR_DONE)
1103			break;
1104	}
1105
1106	if (i == BGE_TIMEOUT * 10) {
1107		device_printf(sc->bge_dev, "EEPROM read timed out\n");
1108		return (1);
1109	}
1110
1111	/* Get result. */
1112	byte = CSR_READ_4(sc, BGE_EE_DATA);
1113
1114	*dest = (byte >> ((addr % 4) * 8)) & 0xFF;
1115
1116	return (0);
1117}
1118
1119/*
1120 * Read a sequence of bytes from the EEPROM.
1121 */
1122static int
1123bge_read_eeprom(struct bge_softc *sc, caddr_t dest, int off, int cnt)
1124{
1125	int i, error = 0;
1126	uint8_t byte = 0;
1127
1128	for (i = 0; i < cnt; i++) {
1129		error = bge_eeprom_getbyte(sc, off + i, &byte);
1130		if (error)
1131			break;
1132		*(dest + i) = byte;
1133	}
1134
1135	return (error ? 1 : 0);
1136}
1137
1138static int
1139bge_miibus_readreg(device_t dev, int phy, int reg)
1140{
1141	struct bge_softc *sc;
1142	uint32_t val;
1143	int i;
1144
1145	sc = device_get_softc(dev);
1146
1147	if (bge_ape_lock(sc, sc->bge_phy_ape_lock) != 0)
1148		return (0);
1149
1150	/* Clear the autopoll bit if set, otherwise may trigger PCI errors. */
1151	if ((sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) != 0) {
1152		CSR_WRITE_4(sc, BGE_MI_MODE,
1153		    sc->bge_mi_mode & ~BGE_MIMODE_AUTOPOLL);
1154		DELAY(80);
1155	}
1156
1157	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_READ | BGE_MICOMM_BUSY |
1158	    BGE_MIPHY(phy) | BGE_MIREG(reg));
1159
1160	/* Poll for the PHY register access to complete. */
1161	for (i = 0; i < BGE_TIMEOUT; i++) {
1162		DELAY(10);
1163		val = CSR_READ_4(sc, BGE_MI_COMM);
1164		if ((val & BGE_MICOMM_BUSY) == 0) {
1165			DELAY(5);
1166			val = CSR_READ_4(sc, BGE_MI_COMM);
1167			break;
1168		}
1169	}
1170
1171	if (i == BGE_TIMEOUT) {
1172		device_printf(sc->bge_dev,
1173		    "PHY read timed out (phy %d, reg %d, val 0x%08x)\n",
1174		    phy, reg, val);
1175		val = 0;
1176	}
1177
1178	/* Restore the autopoll bit if necessary. */
1179	if ((sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) != 0) {
1180		CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode);
1181		DELAY(80);
1182	}
1183
1184	bge_ape_unlock(sc, sc->bge_phy_ape_lock);
1185
1186	if (val & BGE_MICOMM_READFAIL)
1187		return (0);
1188
1189	return (val & 0xFFFF);
1190}
1191
1192static int
1193bge_miibus_writereg(device_t dev, int phy, int reg, int val)
1194{
1195	struct bge_softc *sc;
1196	int i;
1197
1198	sc = device_get_softc(dev);
1199
1200	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
1201	    (reg == BRGPHY_MII_1000CTL || reg == BRGPHY_MII_AUXCTL))
1202		return (0);
1203
1204	if (bge_ape_lock(sc, sc->bge_phy_ape_lock) != 0)
1205		return (0);
1206
1207	/* Clear the autopoll bit if set, otherwise may trigger PCI errors. */
1208	if ((sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) != 0) {
1209		CSR_WRITE_4(sc, BGE_MI_MODE,
1210		    sc->bge_mi_mode & ~BGE_MIMODE_AUTOPOLL);
1211		DELAY(80);
1212	}
1213
1214	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_WRITE | BGE_MICOMM_BUSY |
1215	    BGE_MIPHY(phy) | BGE_MIREG(reg) | val);
1216
1217	for (i = 0; i < BGE_TIMEOUT; i++) {
1218		DELAY(10);
1219		if (!(CSR_READ_4(sc, BGE_MI_COMM) & BGE_MICOMM_BUSY)) {
1220			DELAY(5);
1221			CSR_READ_4(sc, BGE_MI_COMM); /* dummy read */
1222			break;
1223		}
1224	}
1225
1226	/* Restore the autopoll bit if necessary. */
1227	if ((sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) != 0) {
1228		CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode);
1229		DELAY(80);
1230	}
1231
1232	bge_ape_unlock(sc, sc->bge_phy_ape_lock);
1233
1234	if (i == BGE_TIMEOUT)
1235		device_printf(sc->bge_dev,
1236		    "PHY write timed out (phy %d, reg %d, val 0x%04x)\n",
1237		    phy, reg, val);
1238
1239	return (0);
1240}
1241
1242static void
1243bge_miibus_statchg(device_t dev)
1244{
1245	struct bge_softc *sc;
1246	struct mii_data *mii;
1247	uint32_t mac_mode, rx_mode, tx_mode;
1248
1249	sc = device_get_softc(dev);
1250	if ((sc->bge_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
1251		return;
1252	mii = device_get_softc(sc->bge_miibus);
1253
1254	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
1255	    (IFM_ACTIVE | IFM_AVALID)) {
1256		switch (IFM_SUBTYPE(mii->mii_media_active)) {
1257		case IFM_10_T:
1258		case IFM_100_TX:
1259			sc->bge_link = 1;
1260			break;
1261		case IFM_1000_T:
1262		case IFM_1000_SX:
1263		case IFM_2500_SX:
1264			if (sc->bge_asicrev != BGE_ASICREV_BCM5906)
1265				sc->bge_link = 1;
1266			else
1267				sc->bge_link = 0;
1268			break;
1269		default:
1270			sc->bge_link = 0;
1271			break;
1272		}
1273	} else
1274		sc->bge_link = 0;
1275	if (sc->bge_link == 0)
1276		return;
1277
1278	/*
1279	 * APE firmware touches these registers to keep the MAC
1280	 * connected to the outside world.  Try to keep the
1281	 * accesses atomic.
1282	 */
1283
1284	/* Set the port mode (MII/GMII) to match the link speed. */
1285	mac_mode = CSR_READ_4(sc, BGE_MAC_MODE) &
1286	    ~(BGE_MACMODE_PORTMODE | BGE_MACMODE_HALF_DUPLEX);
1287	tx_mode = CSR_READ_4(sc, BGE_TX_MODE);
1288	rx_mode = CSR_READ_4(sc, BGE_RX_MODE);
1289
1290	if (IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_T ||
1291	    IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_SX)
1292		mac_mode |= BGE_PORTMODE_GMII;
1293	else
1294		mac_mode |= BGE_PORTMODE_MII;
1295
1296	/* Set MAC flow control behavior to match link flow control settings. */
1297	tx_mode &= ~BGE_TXMODE_FLOWCTL_ENABLE;
1298	rx_mode &= ~BGE_RXMODE_FLOWCTL_ENABLE;
1299	if ((IFM_OPTIONS(mii->mii_media_active) & IFM_FDX) != 0) {
1300		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_TXPAUSE) != 0)
1301			tx_mode |= BGE_TXMODE_FLOWCTL_ENABLE;
1302		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_RXPAUSE) != 0)
1303			rx_mode |= BGE_RXMODE_FLOWCTL_ENABLE;
1304	} else
1305		mac_mode |= BGE_MACMODE_HALF_DUPLEX;
1306
1307	CSR_WRITE_4(sc, BGE_MAC_MODE, mac_mode);
1308	DELAY(40);
1309	CSR_WRITE_4(sc, BGE_TX_MODE, tx_mode);
1310	CSR_WRITE_4(sc, BGE_RX_MODE, rx_mode);
1311}
1312
1313/*
1314 * Intialize a standard receive ring descriptor.
1315 */
1316static int
1317bge_newbuf_std(struct bge_softc *sc, int i)
1318{
1319	struct mbuf *m;
1320	struct bge_rx_bd *r;
1321	bus_dma_segment_t segs[1];
1322	bus_dmamap_t map;
1323	int error, nsegs;
1324
1325	if (sc->bge_flags & BGE_FLAG_JUMBO_STD &&
1326	    (sc->bge_ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN +
1327	    ETHER_VLAN_ENCAP_LEN > (MCLBYTES - ETHER_ALIGN))) {
1328		m = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUM9BYTES);
1329		if (m == NULL)
1330			return (ENOBUFS);
1331		m->m_len = m->m_pkthdr.len = MJUM9BYTES;
1332	} else {
1333		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
1334		if (m == NULL)
1335			return (ENOBUFS);
1336		m->m_len = m->m_pkthdr.len = MCLBYTES;
1337	}
1338	if ((sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) == 0)
1339		m_adj(m, ETHER_ALIGN);
1340
1341	error = bus_dmamap_load_mbuf_sg(sc->bge_cdata.bge_rx_mtag,
1342	    sc->bge_cdata.bge_rx_std_sparemap, m, segs, &nsegs, 0);
1343	if (error != 0) {
1344		m_freem(m);
1345		return (error);
1346	}
1347	if (sc->bge_cdata.bge_rx_std_chain[i] != NULL) {
1348		bus_dmamap_sync(sc->bge_cdata.bge_rx_mtag,
1349		    sc->bge_cdata.bge_rx_std_dmamap[i], BUS_DMASYNC_POSTREAD);
1350		bus_dmamap_unload(sc->bge_cdata.bge_rx_mtag,
1351		    sc->bge_cdata.bge_rx_std_dmamap[i]);
1352	}
1353	map = sc->bge_cdata.bge_rx_std_dmamap[i];
1354	sc->bge_cdata.bge_rx_std_dmamap[i] = sc->bge_cdata.bge_rx_std_sparemap;
1355	sc->bge_cdata.bge_rx_std_sparemap = map;
1356	sc->bge_cdata.bge_rx_std_chain[i] = m;
1357	sc->bge_cdata.bge_rx_std_seglen[i] = segs[0].ds_len;
1358	r = &sc->bge_ldata.bge_rx_std_ring[sc->bge_std];
1359	r->bge_addr.bge_addr_lo = BGE_ADDR_LO(segs[0].ds_addr);
1360	r->bge_addr.bge_addr_hi = BGE_ADDR_HI(segs[0].ds_addr);
1361	r->bge_flags = BGE_RXBDFLAG_END;
1362	r->bge_len = segs[0].ds_len;
1363	r->bge_idx = i;
1364
1365	bus_dmamap_sync(sc->bge_cdata.bge_rx_mtag,
1366	    sc->bge_cdata.bge_rx_std_dmamap[i], BUS_DMASYNC_PREREAD);
1367
1368	return (0);
1369}
1370
1371/*
1372 * Initialize a jumbo receive ring descriptor. This allocates
1373 * a jumbo buffer from the pool managed internally by the driver.
1374 */
1375static int
1376bge_newbuf_jumbo(struct bge_softc *sc, int i)
1377{
1378	bus_dma_segment_t segs[BGE_NSEG_JUMBO];
1379	bus_dmamap_t map;
1380	struct bge_extrx_bd *r;
1381	struct mbuf *m;
1382	int error, nsegs;
1383
1384	MGETHDR(m, M_NOWAIT, MT_DATA);
1385	if (m == NULL)
1386		return (ENOBUFS);
1387
1388	m_cljget(m, M_NOWAIT, MJUM9BYTES);
1389	if (!(m->m_flags & M_EXT)) {
1390		m_freem(m);
1391		return (ENOBUFS);
1392	}
1393	m->m_len = m->m_pkthdr.len = MJUM9BYTES;
1394	if ((sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) == 0)
1395		m_adj(m, ETHER_ALIGN);
1396
1397	error = bus_dmamap_load_mbuf_sg(sc->bge_cdata.bge_mtag_jumbo,
1398	    sc->bge_cdata.bge_rx_jumbo_sparemap, m, segs, &nsegs, 0);
1399	if (error != 0) {
1400		m_freem(m);
1401		return (error);
1402	}
1403
1404	if (sc->bge_cdata.bge_rx_jumbo_chain[i] != NULL) {
1405		bus_dmamap_sync(sc->bge_cdata.bge_mtag_jumbo,
1406		    sc->bge_cdata.bge_rx_jumbo_dmamap[i], BUS_DMASYNC_POSTREAD);
1407		bus_dmamap_unload(sc->bge_cdata.bge_mtag_jumbo,
1408		    sc->bge_cdata.bge_rx_jumbo_dmamap[i]);
1409	}
1410	map = sc->bge_cdata.bge_rx_jumbo_dmamap[i];
1411	sc->bge_cdata.bge_rx_jumbo_dmamap[i] =
1412	    sc->bge_cdata.bge_rx_jumbo_sparemap;
1413	sc->bge_cdata.bge_rx_jumbo_sparemap = map;
1414	sc->bge_cdata.bge_rx_jumbo_chain[i] = m;
1415	sc->bge_cdata.bge_rx_jumbo_seglen[i][0] = 0;
1416	sc->bge_cdata.bge_rx_jumbo_seglen[i][1] = 0;
1417	sc->bge_cdata.bge_rx_jumbo_seglen[i][2] = 0;
1418	sc->bge_cdata.bge_rx_jumbo_seglen[i][3] = 0;
1419
1420	/*
1421	 * Fill in the extended RX buffer descriptor.
1422	 */
1423	r = &sc->bge_ldata.bge_rx_jumbo_ring[sc->bge_jumbo];
1424	r->bge_flags = BGE_RXBDFLAG_JUMBO_RING | BGE_RXBDFLAG_END;
1425	r->bge_idx = i;
1426	r->bge_len3 = r->bge_len2 = r->bge_len1 = 0;
1427	switch (nsegs) {
1428	case 4:
1429		r->bge_addr3.bge_addr_lo = BGE_ADDR_LO(segs[3].ds_addr);
1430		r->bge_addr3.bge_addr_hi = BGE_ADDR_HI(segs[3].ds_addr);
1431		r->bge_len3 = segs[3].ds_len;
1432		sc->bge_cdata.bge_rx_jumbo_seglen[i][3] = segs[3].ds_len;
1433	case 3:
1434		r->bge_addr2.bge_addr_lo = BGE_ADDR_LO(segs[2].ds_addr);
1435		r->bge_addr2.bge_addr_hi = BGE_ADDR_HI(segs[2].ds_addr);
1436		r->bge_len2 = segs[2].ds_len;
1437		sc->bge_cdata.bge_rx_jumbo_seglen[i][2] = segs[2].ds_len;
1438	case 2:
1439		r->bge_addr1.bge_addr_lo = BGE_ADDR_LO(segs[1].ds_addr);
1440		r->bge_addr1.bge_addr_hi = BGE_ADDR_HI(segs[1].ds_addr);
1441		r->bge_len1 = segs[1].ds_len;
1442		sc->bge_cdata.bge_rx_jumbo_seglen[i][1] = segs[1].ds_len;
1443	case 1:
1444		r->bge_addr0.bge_addr_lo = BGE_ADDR_LO(segs[0].ds_addr);
1445		r->bge_addr0.bge_addr_hi = BGE_ADDR_HI(segs[0].ds_addr);
1446		r->bge_len0 = segs[0].ds_len;
1447		sc->bge_cdata.bge_rx_jumbo_seglen[i][0] = segs[0].ds_len;
1448		break;
1449	default:
1450		panic("%s: %d segments\n", __func__, nsegs);
1451	}
1452
1453	bus_dmamap_sync(sc->bge_cdata.bge_mtag_jumbo,
1454	    sc->bge_cdata.bge_rx_jumbo_dmamap[i], BUS_DMASYNC_PREREAD);
1455
1456	return (0);
1457}
1458
1459static int
1460bge_init_rx_ring_std(struct bge_softc *sc)
1461{
1462	int error, i;
1463
1464	bzero(sc->bge_ldata.bge_rx_std_ring, BGE_STD_RX_RING_SZ);
1465	sc->bge_std = 0;
1466	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1467		if ((error = bge_newbuf_std(sc, i)) != 0)
1468			return (error);
1469		BGE_INC(sc->bge_std, BGE_STD_RX_RING_CNT);
1470	}
1471
1472	bus_dmamap_sync(sc->bge_cdata.bge_rx_std_ring_tag,
1473	    sc->bge_cdata.bge_rx_std_ring_map, BUS_DMASYNC_PREWRITE);
1474
1475	sc->bge_std = 0;
1476	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, BGE_STD_RX_RING_CNT - 1);
1477
1478	return (0);
1479}
1480
1481static void
1482bge_free_rx_ring_std(struct bge_softc *sc)
1483{
1484	int i;
1485
1486	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1487		if (sc->bge_cdata.bge_rx_std_chain[i] != NULL) {
1488			bus_dmamap_sync(sc->bge_cdata.bge_rx_mtag,
1489			    sc->bge_cdata.bge_rx_std_dmamap[i],
1490			    BUS_DMASYNC_POSTREAD);
1491			bus_dmamap_unload(sc->bge_cdata.bge_rx_mtag,
1492			    sc->bge_cdata.bge_rx_std_dmamap[i]);
1493			m_freem(sc->bge_cdata.bge_rx_std_chain[i]);
1494			sc->bge_cdata.bge_rx_std_chain[i] = NULL;
1495		}
1496		bzero((char *)&sc->bge_ldata.bge_rx_std_ring[i],
1497		    sizeof(struct bge_rx_bd));
1498	}
1499}
1500
1501static int
1502bge_init_rx_ring_jumbo(struct bge_softc *sc)
1503{
1504	struct bge_rcb *rcb;
1505	int error, i;
1506
1507	bzero(sc->bge_ldata.bge_rx_jumbo_ring, BGE_JUMBO_RX_RING_SZ);
1508	sc->bge_jumbo = 0;
1509	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1510		if ((error = bge_newbuf_jumbo(sc, i)) != 0)
1511			return (error);
1512		BGE_INC(sc->bge_jumbo, BGE_JUMBO_RX_RING_CNT);
1513	}
1514
1515	bus_dmamap_sync(sc->bge_cdata.bge_rx_jumbo_ring_tag,
1516	    sc->bge_cdata.bge_rx_jumbo_ring_map, BUS_DMASYNC_PREWRITE);
1517
1518	sc->bge_jumbo = 0;
1519
1520	/* Enable the jumbo receive producer ring. */
1521	rcb = &sc->bge_ldata.bge_info.bge_jumbo_rx_rcb;
1522	rcb->bge_maxlen_flags =
1523	    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_USE_EXT_RX_BD);
1524	CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1525
1526	bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, BGE_JUMBO_RX_RING_CNT - 1);
1527
1528	return (0);
1529}
1530
1531static void
1532bge_free_rx_ring_jumbo(struct bge_softc *sc)
1533{
1534	int i;
1535
1536	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1537		if (sc->bge_cdata.bge_rx_jumbo_chain[i] != NULL) {
1538			bus_dmamap_sync(sc->bge_cdata.bge_mtag_jumbo,
1539			    sc->bge_cdata.bge_rx_jumbo_dmamap[i],
1540			    BUS_DMASYNC_POSTREAD);
1541			bus_dmamap_unload(sc->bge_cdata.bge_mtag_jumbo,
1542			    sc->bge_cdata.bge_rx_jumbo_dmamap[i]);
1543			m_freem(sc->bge_cdata.bge_rx_jumbo_chain[i]);
1544			sc->bge_cdata.bge_rx_jumbo_chain[i] = NULL;
1545		}
1546		bzero((char *)&sc->bge_ldata.bge_rx_jumbo_ring[i],
1547		    sizeof(struct bge_extrx_bd));
1548	}
1549}
1550
1551static void
1552bge_free_tx_ring(struct bge_softc *sc)
1553{
1554	int i;
1555
1556	if (sc->bge_ldata.bge_tx_ring == NULL)
1557		return;
1558
1559	for (i = 0; i < BGE_TX_RING_CNT; i++) {
1560		if (sc->bge_cdata.bge_tx_chain[i] != NULL) {
1561			bus_dmamap_sync(sc->bge_cdata.bge_tx_mtag,
1562			    sc->bge_cdata.bge_tx_dmamap[i],
1563			    BUS_DMASYNC_POSTWRITE);
1564			bus_dmamap_unload(sc->bge_cdata.bge_tx_mtag,
1565			    sc->bge_cdata.bge_tx_dmamap[i]);
1566			m_freem(sc->bge_cdata.bge_tx_chain[i]);
1567			sc->bge_cdata.bge_tx_chain[i] = NULL;
1568		}
1569		bzero((char *)&sc->bge_ldata.bge_tx_ring[i],
1570		    sizeof(struct bge_tx_bd));
1571	}
1572}
1573
1574static int
1575bge_init_tx_ring(struct bge_softc *sc)
1576{
1577	sc->bge_txcnt = 0;
1578	sc->bge_tx_saved_considx = 0;
1579
1580	bzero(sc->bge_ldata.bge_tx_ring, BGE_TX_RING_SZ);
1581	bus_dmamap_sync(sc->bge_cdata.bge_tx_ring_tag,
1582	    sc->bge_cdata.bge_tx_ring_map, BUS_DMASYNC_PREWRITE);
1583
1584	/* Initialize transmit producer index for host-memory send ring. */
1585	sc->bge_tx_prodidx = 0;
1586	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1587
1588	/* 5700 b2 errata */
1589	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
1590		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1591
1592	/* NIC-memory send ring not used; initialize to zero. */
1593	bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1594	/* 5700 b2 errata */
1595	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
1596		bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1597
1598	return (0);
1599}
1600
1601static void
1602bge_setpromisc(struct bge_softc *sc)
1603{
1604	struct ifnet *ifp;
1605
1606	BGE_LOCK_ASSERT(sc);
1607
1608	ifp = sc->bge_ifp;
1609
1610	/* Enable or disable promiscuous mode as needed. */
1611	if (ifp->if_flags & IFF_PROMISC)
1612		BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
1613	else
1614		BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
1615}
1616
1617static void
1618bge_setmulti(struct bge_softc *sc)
1619{
1620	struct ifnet *ifp;
1621	struct ifmultiaddr *ifma;
1622	uint32_t hashes[4] = { 0, 0, 0, 0 };
1623	int h, i;
1624
1625	BGE_LOCK_ASSERT(sc);
1626
1627	ifp = sc->bge_ifp;
1628
1629	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
1630		for (i = 0; i < 4; i++)
1631			CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0xFFFFFFFF);
1632		return;
1633	}
1634
1635	/* First, zot all the existing filters. */
1636	for (i = 0; i < 4; i++)
1637		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0);
1638
1639	/* Now program new ones. */
1640	if_maddr_rlock(ifp);
1641	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1642		if (ifma->ifma_addr->sa_family != AF_LINK)
1643			continue;
1644		h = ether_crc32_le(LLADDR((struct sockaddr_dl *)
1645		    ifma->ifma_addr), ETHER_ADDR_LEN) & 0x7F;
1646		hashes[(h & 0x60) >> 5] |= 1 << (h & 0x1F);
1647	}
1648	if_maddr_runlock(ifp);
1649
1650	for (i = 0; i < 4; i++)
1651		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), hashes[i]);
1652}
1653
1654static void
1655bge_setvlan(struct bge_softc *sc)
1656{
1657	struct ifnet *ifp;
1658
1659	BGE_LOCK_ASSERT(sc);
1660
1661	ifp = sc->bge_ifp;
1662
1663	/* Enable or disable VLAN tag stripping as needed. */
1664	if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING)
1665		BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_KEEP_VLAN_DIAG);
1666	else
1667		BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_KEEP_VLAN_DIAG);
1668}
1669
1670static void
1671bge_sig_pre_reset(struct bge_softc *sc, int type)
1672{
1673
1674	/*
1675	 * Some chips don't like this so only do this if ASF is enabled
1676	 */
1677	if (sc->bge_asf_mode)
1678		bge_writemem_ind(sc, BGE_SRAM_FW_MB, BGE_SRAM_FW_MB_MAGIC);
1679
1680	if (sc->bge_asf_mode & ASF_NEW_HANDSHAKE) {
1681		switch (type) {
1682		case BGE_RESET_START:
1683			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1684			    BGE_FW_DRV_STATE_START);
1685			break;
1686		case BGE_RESET_SHUTDOWN:
1687			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1688			    BGE_FW_DRV_STATE_UNLOAD);
1689			break;
1690		case BGE_RESET_SUSPEND:
1691			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1692			    BGE_FW_DRV_STATE_SUSPEND);
1693			break;
1694		}
1695	}
1696
1697	if (type == BGE_RESET_START || type == BGE_RESET_SUSPEND)
1698		bge_ape_driver_state_change(sc, type);
1699}
1700
1701static void
1702bge_sig_post_reset(struct bge_softc *sc, int type)
1703{
1704
1705	if (sc->bge_asf_mode & ASF_NEW_HANDSHAKE) {
1706		switch (type) {
1707		case BGE_RESET_START:
1708			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1709			    BGE_FW_DRV_STATE_START_DONE);
1710			/* START DONE */
1711			break;
1712		case BGE_RESET_SHUTDOWN:
1713			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1714			    BGE_FW_DRV_STATE_UNLOAD_DONE);
1715			break;
1716		}
1717	}
1718	if (type == BGE_RESET_SHUTDOWN)
1719		bge_ape_driver_state_change(sc, type);
1720}
1721
1722static void
1723bge_sig_legacy(struct bge_softc *sc, int type)
1724{
1725
1726	if (sc->bge_asf_mode) {
1727		switch (type) {
1728		case BGE_RESET_START:
1729			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1730			    BGE_FW_DRV_STATE_START);
1731			break;
1732		case BGE_RESET_SHUTDOWN:
1733			bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB,
1734			    BGE_FW_DRV_STATE_UNLOAD);
1735			break;
1736		}
1737	}
1738}
1739
1740static void
1741bge_stop_fw(struct bge_softc *sc)
1742{
1743	int i;
1744
1745	if (sc->bge_asf_mode) {
1746		bge_writemem_ind(sc, BGE_SRAM_FW_CMD_MB, BGE_FW_CMD_PAUSE);
1747		CSR_WRITE_4(sc, BGE_RX_CPU_EVENT,
1748		    CSR_READ_4(sc, BGE_RX_CPU_EVENT) | BGE_RX_CPU_DRV_EVENT);
1749
1750		for (i = 0; i < 100; i++ ) {
1751			if (!(CSR_READ_4(sc, BGE_RX_CPU_EVENT) &
1752			    BGE_RX_CPU_DRV_EVENT))
1753				break;
1754			DELAY(10);
1755		}
1756	}
1757}
1758
1759static uint32_t
1760bge_dma_swap_options(struct bge_softc *sc)
1761{
1762	uint32_t dma_options;
1763
1764	dma_options = BGE_MODECTL_WORDSWAP_NONFRAME |
1765	    BGE_MODECTL_BYTESWAP_DATA | BGE_MODECTL_WORDSWAP_DATA;
1766#if BYTE_ORDER == BIG_ENDIAN
1767	dma_options |= BGE_MODECTL_BYTESWAP_NONFRAME;
1768#endif
1769	return (dma_options);
1770}
1771
1772/*
1773 * Do endian, PCI and DMA initialization.
1774 */
1775static int
1776bge_chipinit(struct bge_softc *sc)
1777{
1778	uint32_t dma_rw_ctl, misc_ctl, mode_ctl;
1779	uint16_t val;
1780	int i;
1781
1782	/* Set endianness before we access any non-PCI registers. */
1783	misc_ctl = BGE_INIT;
1784	if (sc->bge_flags & BGE_FLAG_TAGGED_STATUS)
1785		misc_ctl |= BGE_PCIMISCCTL_TAGGED_STATUS;
1786	pci_write_config(sc->bge_dev, BGE_PCI_MISC_CTL, misc_ctl, 4);
1787
1788	/*
1789	 * Clear the MAC statistics block in the NIC's
1790	 * internal memory.
1791	 */
1792	for (i = BGE_STATS_BLOCK;
1793	    i < BGE_STATS_BLOCK_END + 1; i += sizeof(uint32_t))
1794		BGE_MEMWIN_WRITE(sc, i, 0);
1795
1796	for (i = BGE_STATUS_BLOCK;
1797	    i < BGE_STATUS_BLOCK_END + 1; i += sizeof(uint32_t))
1798		BGE_MEMWIN_WRITE(sc, i, 0);
1799
1800	if (sc->bge_chiprev == BGE_CHIPREV_5704_BX) {
1801		/*
1802		 *  Fix data corruption caused by non-qword write with WB.
1803		 *  Fix master abort in PCI mode.
1804		 *  Fix PCI latency timer.
1805		 */
1806		val = pci_read_config(sc->bge_dev, BGE_PCI_MSI_DATA + 2, 2);
1807		val |= (1 << 10) | (1 << 12) | (1 << 13);
1808		pci_write_config(sc->bge_dev, BGE_PCI_MSI_DATA + 2, val, 2);
1809	}
1810
1811	if (sc->bge_asicrev == BGE_ASICREV_BCM57765 ||
1812	    sc->bge_asicrev == BGE_ASICREV_BCM57766) {
1813		/*
1814		 * For the 57766 and non Ax versions of 57765, bootcode
1815		 * needs to setup the PCIE Fast Training Sequence (FTS)
1816		 * value to prevent transmit hangs.
1817		 */
1818		if (sc->bge_chiprev != BGE_CHIPREV_57765_AX) {
1819			CSR_WRITE_4(sc, BGE_CPMU_PADRNG_CTL,
1820			    CSR_READ_4(sc, BGE_CPMU_PADRNG_CTL) |
1821			    BGE_CPMU_PADRNG_CTL_RDIV2);
1822		}
1823	}
1824
1825	/*
1826	 * Set up the PCI DMA control register.
1827	 */
1828	dma_rw_ctl = BGE_PCIDMARWCTL_RD_CMD_SHIFT(6) |
1829	    BGE_PCIDMARWCTL_WR_CMD_SHIFT(7);
1830	if (sc->bge_flags & BGE_FLAG_PCIE) {
1831		if (sc->bge_mps >= 256)
1832			dma_rw_ctl |= BGE_PCIDMARWCTL_WR_WAT_SHIFT(7);
1833		else
1834			dma_rw_ctl |= BGE_PCIDMARWCTL_WR_WAT_SHIFT(3);
1835	} else if (sc->bge_flags & BGE_FLAG_PCIX) {
1836		if (BGE_IS_5714_FAMILY(sc)) {
1837			/* 256 bytes for read and write. */
1838			dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(2) |
1839			    BGE_PCIDMARWCTL_WR_WAT_SHIFT(2);
1840			dma_rw_ctl |= (sc->bge_asicrev == BGE_ASICREV_BCM5780) ?
1841			    BGE_PCIDMARWCTL_ONEDMA_ATONCE_GLOBAL :
1842			    BGE_PCIDMARWCTL_ONEDMA_ATONCE_LOCAL;
1843		} else if (sc->bge_asicrev == BGE_ASICREV_BCM5703) {
1844			/*
1845			 * In the BCM5703, the DMA read watermark should
1846			 * be set to less than or equal to the maximum
1847			 * memory read byte count of the PCI-X command
1848			 * register.
1849			 */
1850			dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(4) |
1851			    BGE_PCIDMARWCTL_WR_WAT_SHIFT(3);
1852		} else if (sc->bge_asicrev == BGE_ASICREV_BCM5704) {
1853			/* 1536 bytes for read, 384 bytes for write. */
1854			dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(7) |
1855			    BGE_PCIDMARWCTL_WR_WAT_SHIFT(3);
1856		} else {
1857			/* 384 bytes for read and write. */
1858			dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(3) |
1859			    BGE_PCIDMARWCTL_WR_WAT_SHIFT(3) |
1860			    0x0F;
1861		}
1862		if (sc->bge_asicrev == BGE_ASICREV_BCM5703 ||
1863		    sc->bge_asicrev == BGE_ASICREV_BCM5704) {
1864			uint32_t tmp;
1865
1866			/* Set ONE_DMA_AT_ONCE for hardware workaround. */
1867			tmp = CSR_READ_4(sc, BGE_PCI_CLKCTL) & 0x1F;
1868			if (tmp == 6 || tmp == 7)
1869				dma_rw_ctl |=
1870				    BGE_PCIDMARWCTL_ONEDMA_ATONCE_GLOBAL;
1871
1872			/* Set PCI-X DMA write workaround. */
1873			dma_rw_ctl |= BGE_PCIDMARWCTL_ASRT_ALL_BE;
1874		}
1875	} else {
1876		/* Conventional PCI bus: 256 bytes for read and write. */
1877		dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(7) |
1878		    BGE_PCIDMARWCTL_WR_WAT_SHIFT(7);
1879
1880		if (sc->bge_asicrev != BGE_ASICREV_BCM5705 &&
1881		    sc->bge_asicrev != BGE_ASICREV_BCM5750)
1882			dma_rw_ctl |= 0x0F;
1883	}
1884	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 ||
1885	    sc->bge_asicrev == BGE_ASICREV_BCM5701)
1886		dma_rw_ctl |= BGE_PCIDMARWCTL_USE_MRM |
1887		    BGE_PCIDMARWCTL_ASRT_ALL_BE;
1888	if (sc->bge_asicrev == BGE_ASICREV_BCM5703 ||
1889	    sc->bge_asicrev == BGE_ASICREV_BCM5704)
1890		dma_rw_ctl &= ~BGE_PCIDMARWCTL_MINDMA;
1891	if (BGE_IS_5717_PLUS(sc)) {
1892		dma_rw_ctl &= ~BGE_PCIDMARWCTL_DIS_CACHE_ALIGNMENT;
1893		if (sc->bge_chipid == BGE_CHIPID_BCM57765_A0)
1894			dma_rw_ctl &= ~BGE_PCIDMARWCTL_CRDRDR_RDMA_MRRS_MSK;
1895		/*
1896		 * Enable HW workaround for controllers that misinterpret
1897		 * a status tag update and leave interrupts permanently
1898		 * disabled.
1899		 */
1900		if (!BGE_IS_57765_PLUS(sc) &&
1901		    sc->bge_asicrev != BGE_ASICREV_BCM5717 &&
1902		    sc->bge_asicrev != BGE_ASICREV_BCM5762)
1903			dma_rw_ctl |= BGE_PCIDMARWCTL_TAGGED_STATUS_WA;
1904	}
1905	pci_write_config(sc->bge_dev, BGE_PCI_DMA_RW_CTL, dma_rw_ctl, 4);
1906
1907	/*
1908	 * Set up general mode register.
1909	 */
1910	mode_ctl = bge_dma_swap_options(sc);
1911	if (sc->bge_asicrev == BGE_ASICREV_BCM5720 ||
1912	    sc->bge_asicrev == BGE_ASICREV_BCM5762) {
1913		/* Retain Host-2-BMC settings written by APE firmware. */
1914		mode_ctl |= CSR_READ_4(sc, BGE_MODE_CTL) &
1915		    (BGE_MODECTL_BYTESWAP_B2HRX_DATA |
1916		    BGE_MODECTL_WORDSWAP_B2HRX_DATA |
1917		    BGE_MODECTL_B2HRX_ENABLE | BGE_MODECTL_HTX2B_ENABLE);
1918	}
1919	mode_ctl |= BGE_MODECTL_MAC_ATTN_INTR | BGE_MODECTL_HOST_SEND_BDS |
1920	    BGE_MODECTL_TX_NO_PHDR_CSUM;
1921
1922	/*
1923	 * BCM5701 B5 have a bug causing data corruption when using
1924	 * 64-bit DMA reads, which can be terminated early and then
1925	 * completed later as 32-bit accesses, in combination with
1926	 * certain bridges.
1927	 */
1928	if (sc->bge_asicrev == BGE_ASICREV_BCM5701 &&
1929	    sc->bge_chipid == BGE_CHIPID_BCM5701_B5)
1930		mode_ctl |= BGE_MODECTL_FORCE_PCI32;
1931
1932	/*
1933	 * Tell the firmware the driver is running
1934	 */
1935	if (sc->bge_asf_mode & ASF_STACKUP)
1936		mode_ctl |= BGE_MODECTL_STACKUP;
1937
1938	CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
1939
1940	/*
1941	 * Disable memory write invalidate.  Apparently it is not supported
1942	 * properly by these devices.
1943	 */
1944	PCI_CLRBIT(sc->bge_dev, BGE_PCI_CMD, PCIM_CMD_MWIEN, 4);
1945
1946	/* Set the timer prescaler (always 66 MHz). */
1947	CSR_WRITE_4(sc, BGE_MISC_CFG, BGE_32BITTIME_66MHZ);
1948
1949	/* XXX: The Linux tg3 driver does this at the start of brgphy_reset. */
1950	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
1951		DELAY(40);	/* XXX */
1952
1953		/* Put PHY into ready state */
1954		BGE_CLRBIT(sc, BGE_MISC_CFG, BGE_MISCCFG_EPHY_IDDQ);
1955		CSR_READ_4(sc, BGE_MISC_CFG); /* Flush */
1956		DELAY(40);
1957	}
1958
1959	return (0);
1960}
1961
1962static int
1963bge_blockinit(struct bge_softc *sc)
1964{
1965	struct bge_rcb *rcb;
1966	bus_size_t vrcb;
1967	bge_hostaddr taddr;
1968	uint32_t dmactl, rdmareg, val;
1969	int i, limit;
1970
1971	/*
1972	 * Initialize the memory window pointer register so that
1973	 * we can access the first 32K of internal NIC RAM. This will
1974	 * allow us to set up the TX send ring RCBs and the RX return
1975	 * ring RCBs, plus other things which live in NIC memory.
1976	 */
1977	CSR_WRITE_4(sc, BGE_PCI_MEMWIN_BASEADDR, 0);
1978
1979	/* Note: the BCM5704 has a smaller mbuf space than other chips. */
1980
1981	if (!(BGE_IS_5705_PLUS(sc))) {
1982		/* Configure mbuf memory pool */
1983		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_BASEADDR, BGE_BUFFPOOL_1);
1984		if (sc->bge_asicrev == BGE_ASICREV_BCM5704)
1985			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x10000);
1986		else
1987			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x18000);
1988
1989		/* Configure DMA resource pool */
1990		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_BASEADDR,
1991		    BGE_DMA_DESCRIPTORS);
1992		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LEN, 0x2000);
1993	}
1994
1995	/* Configure mbuf pool watermarks */
1996	if (BGE_IS_5717_PLUS(sc)) {
1997		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1998		if (sc->bge_ifp->if_mtu > ETHERMTU) {
1999			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x7e);
2000			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0xea);
2001		} else {
2002			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x2a);
2003			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0xa0);
2004		}
2005	} else if (!BGE_IS_5705_PLUS(sc)) {
2006		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x50);
2007		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x20);
2008		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
2009	} else if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
2010		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
2011		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x04);
2012		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x10);
2013	} else {
2014		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
2015		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x10);
2016		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
2017	}
2018
2019	/* Configure DMA resource watermarks */
2020	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LOWAT, 5);
2021	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_HIWAT, 10);
2022
2023	/* Enable buffer manager */
2024	val = BGE_BMANMODE_ENABLE | BGE_BMANMODE_LOMBUF_ATTN;
2025	/*
2026	 * Change the arbitration algorithm of TXMBUF read request to
2027	 * round-robin instead of priority based for BCM5719.  When
2028	 * TXFIFO is almost empty, RDMA will hold its request until
2029	 * TXFIFO is not almost empty.
2030	 */
2031	if (sc->bge_asicrev == BGE_ASICREV_BCM5719)
2032		val |= BGE_BMANMODE_NO_TX_UNDERRUN;
2033	CSR_WRITE_4(sc, BGE_BMAN_MODE, val);
2034
2035	/* Poll for buffer manager start indication */
2036	for (i = 0; i < BGE_TIMEOUT; i++) {
2037		DELAY(10);
2038		if (CSR_READ_4(sc, BGE_BMAN_MODE) & BGE_BMANMODE_ENABLE)
2039			break;
2040	}
2041
2042	if (i == BGE_TIMEOUT) {
2043		device_printf(sc->bge_dev, "buffer manager failed to start\n");
2044		return (ENXIO);
2045	}
2046
2047	/* Enable flow-through queues */
2048	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
2049	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
2050
2051	/* Wait until queue initialization is complete */
2052	for (i = 0; i < BGE_TIMEOUT; i++) {
2053		DELAY(10);
2054		if (CSR_READ_4(sc, BGE_FTQ_RESET) == 0)
2055			break;
2056	}
2057
2058	if (i == BGE_TIMEOUT) {
2059		device_printf(sc->bge_dev, "flow-through queue init failed\n");
2060		return (ENXIO);
2061	}
2062
2063	/*
2064	 * Summary of rings supported by the controller:
2065	 *
2066	 * Standard Receive Producer Ring
2067	 * - This ring is used to feed receive buffers for "standard"
2068	 *   sized frames (typically 1536 bytes) to the controller.
2069	 *
2070	 * Jumbo Receive Producer Ring
2071	 * - This ring is used to feed receive buffers for jumbo sized
2072	 *   frames (i.e. anything bigger than the "standard" frames)
2073	 *   to the controller.
2074	 *
2075	 * Mini Receive Producer Ring
2076	 * - This ring is used to feed receive buffers for "mini"
2077	 *   sized frames to the controller.
2078	 * - This feature required external memory for the controller
2079	 *   but was never used in a production system.  Should always
2080	 *   be disabled.
2081	 *
2082	 * Receive Return Ring
2083	 * - After the controller has placed an incoming frame into a
2084	 *   receive buffer that buffer is moved into a receive return
2085	 *   ring.  The driver is then responsible to passing the
2086	 *   buffer up to the stack.  Many versions of the controller
2087	 *   support multiple RR rings.
2088	 *
2089	 * Send Ring
2090	 * - This ring is used for outgoing frames.  Many versions of
2091	 *   the controller support multiple send rings.
2092	 */
2093
2094	/* Initialize the standard receive producer ring control block. */
2095	rcb = &sc->bge_ldata.bge_info.bge_std_rx_rcb;
2096	rcb->bge_hostaddr.bge_addr_lo =
2097	    BGE_ADDR_LO(sc->bge_ldata.bge_rx_std_ring_paddr);
2098	rcb->bge_hostaddr.bge_addr_hi =
2099	    BGE_ADDR_HI(sc->bge_ldata.bge_rx_std_ring_paddr);
2100	bus_dmamap_sync(sc->bge_cdata.bge_rx_std_ring_tag,
2101	    sc->bge_cdata.bge_rx_std_ring_map, BUS_DMASYNC_PREREAD);
2102	if (BGE_IS_5717_PLUS(sc)) {
2103		/*
2104		 * Bits 31-16: Programmable ring size (2048, 1024, 512, .., 32)
2105		 * Bits 15-2 : Maximum RX frame size
2106		 * Bit 1     : 1 = Ring Disabled, 0 = Ring ENabled
2107		 * Bit 0     : Reserved
2108		 */
2109		rcb->bge_maxlen_flags =
2110		    BGE_RCB_MAXLEN_FLAGS(512, BGE_MAX_FRAMELEN << 2);
2111	} else if (BGE_IS_5705_PLUS(sc)) {
2112		/*
2113		 * Bits 31-16: Programmable ring size (512, 256, 128, 64, 32)
2114		 * Bits 15-2 : Reserved (should be 0)
2115		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
2116		 * Bit 0     : Reserved
2117		 */
2118		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(512, 0);
2119	} else {
2120		/*
2121		 * Ring size is always XXX entries
2122		 * Bits 31-16: Maximum RX frame size
2123		 * Bits 15-2 : Reserved (should be 0)
2124		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
2125		 * Bit 0     : Reserved
2126		 */
2127		rcb->bge_maxlen_flags =
2128		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN, 0);
2129	}
2130	if (sc->bge_asicrev == BGE_ASICREV_BCM5717 ||
2131	    sc->bge_asicrev == BGE_ASICREV_BCM5719 ||
2132	    sc->bge_asicrev == BGE_ASICREV_BCM5720)
2133		rcb->bge_nicaddr = BGE_STD_RX_RINGS_5717;
2134	else
2135		rcb->bge_nicaddr = BGE_STD_RX_RINGS;
2136	/* Write the standard receive producer ring control block. */
2137	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_HI, rcb->bge_hostaddr.bge_addr_hi);
2138	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_LO, rcb->bge_hostaddr.bge_addr_lo);
2139	CSR_WRITE_4(sc, BGE_RX_STD_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
2140	CSR_WRITE_4(sc, BGE_RX_STD_RCB_NICADDR, rcb->bge_nicaddr);
2141
2142	/* Reset the standard receive producer ring producer index. */
2143	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, 0);
2144
2145	/*
2146	 * Initialize the jumbo RX producer ring control
2147	 * block.  We set the 'ring disabled' bit in the
2148	 * flags field until we're actually ready to start
2149	 * using this ring (i.e. once we set the MTU
2150	 * high enough to require it).
2151	 */
2152	if (BGE_IS_JUMBO_CAPABLE(sc)) {
2153		rcb = &sc->bge_ldata.bge_info.bge_jumbo_rx_rcb;
2154		/* Get the jumbo receive producer ring RCB parameters. */
2155		rcb->bge_hostaddr.bge_addr_lo =
2156		    BGE_ADDR_LO(sc->bge_ldata.bge_rx_jumbo_ring_paddr);
2157		rcb->bge_hostaddr.bge_addr_hi =
2158		    BGE_ADDR_HI(sc->bge_ldata.bge_rx_jumbo_ring_paddr);
2159		bus_dmamap_sync(sc->bge_cdata.bge_rx_jumbo_ring_tag,
2160		    sc->bge_cdata.bge_rx_jumbo_ring_map,
2161		    BUS_DMASYNC_PREREAD);
2162		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(0,
2163		    BGE_RCB_FLAG_USE_EXT_RX_BD | BGE_RCB_FLAG_RING_DISABLED);
2164		if (sc->bge_asicrev == BGE_ASICREV_BCM5717 ||
2165		    sc->bge_asicrev == BGE_ASICREV_BCM5719 ||
2166		    sc->bge_asicrev == BGE_ASICREV_BCM5720)
2167			rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS_5717;
2168		else
2169			rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS;
2170		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_HI,
2171		    rcb->bge_hostaddr.bge_addr_hi);
2172		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_LO,
2173		    rcb->bge_hostaddr.bge_addr_lo);
2174		/* Program the jumbo receive producer ring RCB parameters. */
2175		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS,
2176		    rcb->bge_maxlen_flags);
2177		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_NICADDR, rcb->bge_nicaddr);
2178		/* Reset the jumbo receive producer ring producer index. */
2179		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, 0);
2180	}
2181
2182	/* Disable the mini receive producer ring RCB. */
2183	if (BGE_IS_5700_FAMILY(sc)) {
2184		rcb = &sc->bge_ldata.bge_info.bge_mini_rx_rcb;
2185		rcb->bge_maxlen_flags =
2186		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED);
2187		CSR_WRITE_4(sc, BGE_RX_MINI_RCB_MAXLEN_FLAGS,
2188		    rcb->bge_maxlen_flags);
2189		/* Reset the mini receive producer ring producer index. */
2190		bge_writembx(sc, BGE_MBX_RX_MINI_PROD_LO, 0);
2191	}
2192
2193	/* Choose de-pipeline mode for BCM5906 A0, A1 and A2. */
2194	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
2195		if (sc->bge_chipid == BGE_CHIPID_BCM5906_A0 ||
2196		    sc->bge_chipid == BGE_CHIPID_BCM5906_A1 ||
2197		    sc->bge_chipid == BGE_CHIPID_BCM5906_A2)
2198			CSR_WRITE_4(sc, BGE_ISO_PKT_TX,
2199			    (CSR_READ_4(sc, BGE_ISO_PKT_TX) & ~3) | 2);
2200	}
2201	/*
2202	 * The BD ring replenish thresholds control how often the
2203	 * hardware fetches new BD's from the producer rings in host
2204	 * memory.  Setting the value too low on a busy system can
2205	 * starve the hardware and recue the throughpout.
2206	 *
2207	 * Set the BD ring replentish thresholds. The recommended
2208	 * values are 1/8th the number of descriptors allocated to
2209	 * each ring.
2210	 * XXX The 5754 requires a lower threshold, so it might be a
2211	 * requirement of all 575x family chips.  The Linux driver sets
2212	 * the lower threshold for all 5705 family chips as well, but there
2213	 * are reports that it might not need to be so strict.
2214	 *
2215	 * XXX Linux does some extra fiddling here for the 5906 parts as
2216	 * well.
2217	 */
2218	if (BGE_IS_5705_PLUS(sc))
2219		val = 8;
2220	else
2221		val = BGE_STD_RX_RING_CNT / 8;
2222	CSR_WRITE_4(sc, BGE_RBDI_STD_REPL_THRESH, val);
2223	if (BGE_IS_JUMBO_CAPABLE(sc))
2224		CSR_WRITE_4(sc, BGE_RBDI_JUMBO_REPL_THRESH,
2225		    BGE_JUMBO_RX_RING_CNT/8);
2226	if (BGE_IS_5717_PLUS(sc)) {
2227		CSR_WRITE_4(sc, BGE_STD_REPLENISH_LWM, 32);
2228		CSR_WRITE_4(sc, BGE_JMB_REPLENISH_LWM, 16);
2229	}
2230
2231	/*
2232	 * Disable all send rings by setting the 'ring disabled' bit
2233	 * in the flags field of all the TX send ring control blocks,
2234	 * located in NIC memory.
2235	 */
2236	if (!BGE_IS_5705_PLUS(sc))
2237		/* 5700 to 5704 had 16 send rings. */
2238		limit = BGE_TX_RINGS_EXTSSRAM_MAX;
2239	else if (BGE_IS_57765_PLUS(sc) ||
2240	    sc->bge_asicrev == BGE_ASICREV_BCM5762)
2241		limit = 2;
2242	else if (BGE_IS_5717_PLUS(sc))
2243		limit = 4;
2244	else
2245		limit = 1;
2246	vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
2247	for (i = 0; i < limit; i++) {
2248		RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
2249		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED));
2250		RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
2251		vrcb += sizeof(struct bge_rcb);
2252	}
2253
2254	/* Configure send ring RCB 0 (we use only the first ring) */
2255	vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
2256	BGE_HOSTADDR(taddr, sc->bge_ldata.bge_tx_ring_paddr);
2257	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
2258	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
2259	if (sc->bge_asicrev == BGE_ASICREV_BCM5717 ||
2260	    sc->bge_asicrev == BGE_ASICREV_BCM5719 ||
2261	    sc->bge_asicrev == BGE_ASICREV_BCM5720)
2262		RCB_WRITE_4(sc, vrcb, bge_nicaddr, BGE_SEND_RING_5717);
2263	else
2264		RCB_WRITE_4(sc, vrcb, bge_nicaddr,
2265		    BGE_NIC_TXRING_ADDR(0, BGE_TX_RING_CNT));
2266	RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
2267	    BGE_RCB_MAXLEN_FLAGS(BGE_TX_RING_CNT, 0));
2268
2269	/*
2270	 * Disable all receive return rings by setting the
2271	 * 'ring diabled' bit in the flags field of all the receive
2272	 * return ring control blocks, located in NIC memory.
2273	 */
2274	if (sc->bge_asicrev == BGE_ASICREV_BCM5717 ||
2275	    sc->bge_asicrev == BGE_ASICREV_BCM5719 ||
2276	    sc->bge_asicrev == BGE_ASICREV_BCM5720) {
2277		/* Should be 17, use 16 until we get an SRAM map. */
2278		limit = 16;
2279	} else if (!BGE_IS_5705_PLUS(sc))
2280		limit = BGE_RX_RINGS_MAX;
2281	else if (sc->bge_asicrev == BGE_ASICREV_BCM5755 ||
2282	    sc->bge_asicrev == BGE_ASICREV_BCM5762 ||
2283	    BGE_IS_57765_PLUS(sc))
2284		limit = 4;
2285	else
2286		limit = 1;
2287	/* Disable all receive return rings. */
2288	vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
2289	for (i = 0; i < limit; i++) {
2290		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, 0);
2291		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, 0);
2292		RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
2293		    BGE_RCB_FLAG_RING_DISABLED);
2294		RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
2295		bge_writembx(sc, BGE_MBX_RX_CONS0_LO +
2296		    (i * (sizeof(uint64_t))), 0);
2297		vrcb += sizeof(struct bge_rcb);
2298	}
2299
2300	/*
2301	 * Set up receive return ring 0.  Note that the NIC address
2302	 * for RX return rings is 0x0.  The return rings live entirely
2303	 * within the host, so the nicaddr field in the RCB isn't used.
2304	 */
2305	vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
2306	BGE_HOSTADDR(taddr, sc->bge_ldata.bge_rx_return_ring_paddr);
2307	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
2308	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
2309	RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
2310	RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
2311	    BGE_RCB_MAXLEN_FLAGS(sc->bge_return_ring_cnt, 0));
2312
2313	/* Set random backoff seed for TX */
2314	CSR_WRITE_4(sc, BGE_TX_RANDOM_BACKOFF,
2315	    (IF_LLADDR(sc->bge_ifp)[0] + IF_LLADDR(sc->bge_ifp)[1] +
2316	    IF_LLADDR(sc->bge_ifp)[2] + IF_LLADDR(sc->bge_ifp)[3] +
2317	    IF_LLADDR(sc->bge_ifp)[4] + IF_LLADDR(sc->bge_ifp)[5]) &
2318	    BGE_TX_BACKOFF_SEED_MASK);
2319
2320	/* Set inter-packet gap */
2321	val = 0x2620;
2322	if (sc->bge_asicrev == BGE_ASICREV_BCM5720 ||
2323	    sc->bge_asicrev == BGE_ASICREV_BCM5762)
2324		val |= CSR_READ_4(sc, BGE_TX_LENGTHS) &
2325		    (BGE_TXLEN_JMB_FRM_LEN_MSK | BGE_TXLEN_CNT_DN_VAL_MSK);
2326	CSR_WRITE_4(sc, BGE_TX_LENGTHS, val);
2327
2328	/*
2329	 * Specify which ring to use for packets that don't match
2330	 * any RX rules.
2331	 */
2332	CSR_WRITE_4(sc, BGE_RX_RULES_CFG, 0x08);
2333
2334	/*
2335	 * Configure number of RX lists. One interrupt distribution
2336	 * list, sixteen active lists, one bad frames class.
2337	 */
2338	CSR_WRITE_4(sc, BGE_RXLP_CFG, 0x181);
2339
2340	/* Inialize RX list placement stats mask. */
2341	CSR_WRITE_4(sc, BGE_RXLP_STATS_ENABLE_MASK, 0x007FFFFF);
2342	CSR_WRITE_4(sc, BGE_RXLP_STATS_CTL, 0x1);
2343
2344	/* Disable host coalescing until we get it set up */
2345	CSR_WRITE_4(sc, BGE_HCC_MODE, 0x00000000);
2346
2347	/* Poll to make sure it's shut down. */
2348	for (i = 0; i < BGE_TIMEOUT; i++) {
2349		DELAY(10);
2350		if (!(CSR_READ_4(sc, BGE_HCC_MODE) & BGE_HCCMODE_ENABLE))
2351			break;
2352	}
2353
2354	if (i == BGE_TIMEOUT) {
2355		device_printf(sc->bge_dev,
2356		    "host coalescing engine failed to idle\n");
2357		return (ENXIO);
2358	}
2359
2360	/* Set up host coalescing defaults */
2361	CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, sc->bge_rx_coal_ticks);
2362	CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS, sc->bge_tx_coal_ticks);
2363	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, sc->bge_rx_max_coal_bds);
2364	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS, sc->bge_tx_max_coal_bds);
2365	if (!(BGE_IS_5705_PLUS(sc))) {
2366		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS_INT, 0);
2367		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS_INT, 0);
2368	}
2369	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT, 1);
2370	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT, 1);
2371
2372	/* Set up address of statistics block */
2373	if (!(BGE_IS_5705_PLUS(sc))) {
2374		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_HI,
2375		    BGE_ADDR_HI(sc->bge_ldata.bge_stats_paddr));
2376		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_LO,
2377		    BGE_ADDR_LO(sc->bge_ldata.bge_stats_paddr));
2378		CSR_WRITE_4(sc, BGE_HCC_STATS_BASEADDR, BGE_STATS_BLOCK);
2379		CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_BASEADDR, BGE_STATUS_BLOCK);
2380		CSR_WRITE_4(sc, BGE_HCC_STATS_TICKS, sc->bge_stat_ticks);
2381	}
2382
2383	/* Set up address of status block */
2384	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_HI,
2385	    BGE_ADDR_HI(sc->bge_ldata.bge_status_block_paddr));
2386	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_LO,
2387	    BGE_ADDR_LO(sc->bge_ldata.bge_status_block_paddr));
2388
2389	/* Set up status block size. */
2390	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
2391	    sc->bge_chipid != BGE_CHIPID_BCM5700_C0) {
2392		val = BGE_STATBLKSZ_FULL;
2393		bzero(sc->bge_ldata.bge_status_block, BGE_STATUS_BLK_SZ);
2394	} else {
2395		val = BGE_STATBLKSZ_32BYTE;
2396		bzero(sc->bge_ldata.bge_status_block, 32);
2397	}
2398	bus_dmamap_sync(sc->bge_cdata.bge_status_tag,
2399	    sc->bge_cdata.bge_status_map,
2400	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2401
2402	/* Turn on host coalescing state machine */
2403	CSR_WRITE_4(sc, BGE_HCC_MODE, val | BGE_HCCMODE_ENABLE);
2404
2405	/* Turn on RX BD completion state machine and enable attentions */
2406	CSR_WRITE_4(sc, BGE_RBDC_MODE,
2407	    BGE_RBDCMODE_ENABLE | BGE_RBDCMODE_ATTN);
2408
2409	/* Turn on RX list placement state machine */
2410	CSR_WRITE_4(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
2411
2412	/* Turn on RX list selector state machine. */
2413	if (!(BGE_IS_5705_PLUS(sc)))
2414		CSR_WRITE_4(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
2415
2416	/* Turn on DMA, clear stats. */
2417	val = BGE_MACMODE_TXDMA_ENB | BGE_MACMODE_RXDMA_ENB |
2418	    BGE_MACMODE_RX_STATS_CLEAR | BGE_MACMODE_TX_STATS_CLEAR |
2419	    BGE_MACMODE_RX_STATS_ENB | BGE_MACMODE_TX_STATS_ENB |
2420	    BGE_MACMODE_FRMHDR_DMA_ENB;
2421
2422	if (sc->bge_flags & BGE_FLAG_TBI)
2423		val |= BGE_PORTMODE_TBI;
2424	else if (sc->bge_flags & BGE_FLAG_MII_SERDES)
2425		val |= BGE_PORTMODE_GMII;
2426	else
2427		val |= BGE_PORTMODE_MII;
2428
2429	/* Allow APE to send/receive frames. */
2430	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) != 0)
2431		val |= BGE_MACMODE_APE_RX_EN | BGE_MACMODE_APE_TX_EN;
2432
2433	CSR_WRITE_4(sc, BGE_MAC_MODE, val);
2434	DELAY(40);
2435
2436	/* Set misc. local control, enable interrupts on attentions */
2437	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_ONATTN);
2438
2439#ifdef notdef
2440	/* Assert GPIO pins for PHY reset */
2441	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUT0 |
2442	    BGE_MLC_MISCIO_OUT1 | BGE_MLC_MISCIO_OUT2);
2443	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUTEN0 |
2444	    BGE_MLC_MISCIO_OUTEN1 | BGE_MLC_MISCIO_OUTEN2);
2445#endif
2446
2447	/* Turn on DMA completion state machine */
2448	if (!(BGE_IS_5705_PLUS(sc)))
2449		CSR_WRITE_4(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
2450
2451	val = BGE_WDMAMODE_ENABLE | BGE_WDMAMODE_ALL_ATTNS;
2452
2453	/* Enable host coalescing bug fix. */
2454	if (BGE_IS_5755_PLUS(sc))
2455		val |= BGE_WDMAMODE_STATUS_TAG_FIX;
2456
2457	/* Request larger DMA burst size to get better performance. */
2458	if (sc->bge_asicrev == BGE_ASICREV_BCM5785)
2459		val |= BGE_WDMAMODE_BURST_ALL_DATA;
2460
2461	/* Turn on write DMA state machine */
2462	CSR_WRITE_4(sc, BGE_WDMA_MODE, val);
2463	DELAY(40);
2464
2465	/* Turn on read DMA state machine */
2466	val = BGE_RDMAMODE_ENABLE | BGE_RDMAMODE_ALL_ATTNS;
2467
2468	if (sc->bge_asicrev == BGE_ASICREV_BCM5717)
2469		val |= BGE_RDMAMODE_MULT_DMA_RD_DIS;
2470
2471	if (sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
2472	    sc->bge_asicrev == BGE_ASICREV_BCM5785 ||
2473	    sc->bge_asicrev == BGE_ASICREV_BCM57780)
2474		val |= BGE_RDMAMODE_BD_SBD_CRPT_ATTN |
2475		    BGE_RDMAMODE_MBUF_RBD_CRPT_ATTN |
2476		    BGE_RDMAMODE_MBUF_SBD_CRPT_ATTN;
2477	if (sc->bge_flags & BGE_FLAG_PCIE)
2478		val |= BGE_RDMAMODE_FIFO_LONG_BURST;
2479	if (sc->bge_flags & (BGE_FLAG_TSO | BGE_FLAG_TSO3)) {
2480		val |= BGE_RDMAMODE_TSO4_ENABLE;
2481		if (sc->bge_flags & BGE_FLAG_TSO3 ||
2482		    sc->bge_asicrev == BGE_ASICREV_BCM5785 ||
2483		    sc->bge_asicrev == BGE_ASICREV_BCM57780)
2484			val |= BGE_RDMAMODE_TSO6_ENABLE;
2485	}
2486
2487	if (sc->bge_asicrev == BGE_ASICREV_BCM5720 ||
2488	    sc->bge_asicrev == BGE_ASICREV_BCM5762) {
2489		val |= CSR_READ_4(sc, BGE_RDMA_MODE) &
2490			BGE_RDMAMODE_H2BNC_VLAN_DET;
2491		/*
2492		 * Allow multiple outstanding read requests from
2493		 * non-LSO read DMA engine.
2494		 */
2495		val &= ~BGE_RDMAMODE_MULT_DMA_RD_DIS;
2496	}
2497
2498	if (sc->bge_asicrev == BGE_ASICREV_BCM5761 ||
2499	    sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
2500	    sc->bge_asicrev == BGE_ASICREV_BCM5785 ||
2501	    sc->bge_asicrev == BGE_ASICREV_BCM57780 ||
2502	    BGE_IS_5717_PLUS(sc) || BGE_IS_57765_PLUS(sc)) {
2503		if (sc->bge_asicrev == BGE_ASICREV_BCM5762)
2504			rdmareg = BGE_RDMA_RSRVCTRL_REG2;
2505		else
2506			rdmareg = BGE_RDMA_RSRVCTRL;
2507		dmactl = CSR_READ_4(sc, rdmareg);
2508		/*
2509		 * Adjust tx margin to prevent TX data corruption and
2510		 * fix internal FIFO overflow.
2511		 */
2512		if (sc->bge_chipid == BGE_CHIPID_BCM5719_A0 ||
2513		    sc->bge_asicrev == BGE_ASICREV_BCM5762) {
2514			dmactl &= ~(BGE_RDMA_RSRVCTRL_FIFO_LWM_MASK |
2515			    BGE_RDMA_RSRVCTRL_FIFO_HWM_MASK |
2516			    BGE_RDMA_RSRVCTRL_TXMRGN_MASK);
2517			dmactl |= BGE_RDMA_RSRVCTRL_FIFO_LWM_1_5K |
2518			    BGE_RDMA_RSRVCTRL_FIFO_HWM_1_5K |
2519			    BGE_RDMA_RSRVCTRL_TXMRGN_320B;
2520		}
2521		/*
2522		 * Enable fix for read DMA FIFO overruns.
2523		 * The fix is to limit the number of RX BDs
2524		 * the hardware would fetch at a fime.
2525		 */
2526		CSR_WRITE_4(sc, rdmareg, dmactl |
2527		    BGE_RDMA_RSRVCTRL_FIFO_OFLW_FIX);
2528	}
2529
2530	if (sc->bge_asicrev == BGE_ASICREV_BCM5719) {
2531		CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL,
2532		    CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL) |
2533		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_4K |
2534		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
2535	} else if (sc->bge_asicrev == BGE_ASICREV_BCM5720) {
2536		/*
2537		 * Allow 4KB burst length reads for non-LSO frames.
2538		 * Enable 512B burst length reads for buffer descriptors.
2539		 */
2540		CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL,
2541		    CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL) |
2542		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_512 |
2543		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
2544	} else if (sc->bge_asicrev == BGE_ASICREV_BCM5762) {
2545		CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL_REG2,
2546		    CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL_REG2) |
2547		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_4K |
2548		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
2549	}
2550
2551	CSR_WRITE_4(sc, BGE_RDMA_MODE, val);
2552	DELAY(40);
2553
2554	if (sc->bge_flags & BGE_FLAG_RDMA_BUG) {
2555		for (i = 0; i < BGE_NUM_RDMA_CHANNELS / 2; i++) {
2556			val = CSR_READ_4(sc, BGE_RDMA_LENGTH + i * 4);
2557			if ((val & 0xFFFF) > BGE_FRAMELEN)
2558				break;
2559			if (((val >> 16) & 0xFFFF) > BGE_FRAMELEN)
2560				break;
2561		}
2562		if (i != BGE_NUM_RDMA_CHANNELS / 2) {
2563			val = CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL);
2564			if (sc->bge_asicrev == BGE_ASICREV_BCM5719)
2565				val |= BGE_RDMA_TX_LENGTH_WA_5719;
2566			else
2567				val |= BGE_RDMA_TX_LENGTH_WA_5720;
2568			CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL, val);
2569		}
2570	}
2571
2572	/* Turn on RX data completion state machine */
2573	CSR_WRITE_4(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
2574
2575	/* Turn on RX BD initiator state machine */
2576	CSR_WRITE_4(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
2577
2578	/* Turn on RX data and RX BD initiator state machine */
2579	CSR_WRITE_4(sc, BGE_RDBDI_MODE, BGE_RDBDIMODE_ENABLE);
2580
2581	/* Turn on Mbuf cluster free state machine */
2582	if (!(BGE_IS_5705_PLUS(sc)))
2583		CSR_WRITE_4(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
2584
2585	/* Turn on send BD completion state machine */
2586	CSR_WRITE_4(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
2587
2588	/* Turn on send data completion state machine */
2589	val = BGE_SDCMODE_ENABLE;
2590	if (sc->bge_asicrev == BGE_ASICREV_BCM5761)
2591		val |= BGE_SDCMODE_CDELAY;
2592	CSR_WRITE_4(sc, BGE_SDC_MODE, val);
2593
2594	/* Turn on send data initiator state machine */
2595	if (sc->bge_flags & (BGE_FLAG_TSO | BGE_FLAG_TSO3))
2596		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE |
2597		    BGE_SDIMODE_HW_LSO_PRE_DMA);
2598	else
2599		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
2600
2601	/* Turn on send BD initiator state machine */
2602	CSR_WRITE_4(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
2603
2604	/* Turn on send BD selector state machine */
2605	CSR_WRITE_4(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
2606
2607	CSR_WRITE_4(sc, BGE_SDI_STATS_ENABLE_MASK, 0x007FFFFF);
2608	CSR_WRITE_4(sc, BGE_SDI_STATS_CTL,
2609	    BGE_SDISTATSCTL_ENABLE | BGE_SDISTATSCTL_FASTER);
2610
2611	/* ack/clear link change events */
2612	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
2613	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
2614	    BGE_MACSTAT_LINK_CHANGED);
2615	CSR_WRITE_4(sc, BGE_MI_STS, 0);
2616
2617	/*
2618	 * Enable attention when the link has changed state for
2619	 * devices that use auto polling.
2620	 */
2621	if (sc->bge_flags & BGE_FLAG_TBI) {
2622		CSR_WRITE_4(sc, BGE_MI_STS, BGE_MISTS_LINK);
2623	} else {
2624		if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
2625			CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode);
2626			DELAY(80);
2627		}
2628		if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
2629		    sc->bge_chipid != BGE_CHIPID_BCM5700_B2)
2630			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
2631			    BGE_EVTENB_MI_INTERRUPT);
2632	}
2633
2634	/*
2635	 * Clear any pending link state attention.
2636	 * Otherwise some link state change events may be lost until attention
2637	 * is cleared by bge_intr() -> bge_link_upd() sequence.
2638	 * It's not necessary on newer BCM chips - perhaps enabling link
2639	 * state change attentions implies clearing pending attention.
2640	 */
2641	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
2642	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
2643	    BGE_MACSTAT_LINK_CHANGED);
2644
2645	/* Enable link state change attentions. */
2646	BGE_SETBIT(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_LINK_CHANGED);
2647
2648	return (0);
2649}
2650
2651static const struct bge_revision *
2652bge_lookup_rev(uint32_t chipid)
2653{
2654	const struct bge_revision *br;
2655
2656	for (br = bge_revisions; br->br_name != NULL; br++) {
2657		if (br->br_chipid == chipid)
2658			return (br);
2659	}
2660
2661	for (br = bge_majorrevs; br->br_name != NULL; br++) {
2662		if (br->br_chipid == BGE_ASICREV(chipid))
2663			return (br);
2664	}
2665
2666	return (NULL);
2667}
2668
2669static const struct bge_vendor *
2670bge_lookup_vendor(uint16_t vid)
2671{
2672	const struct bge_vendor *v;
2673
2674	for (v = bge_vendors; v->v_name != NULL; v++)
2675		if (v->v_id == vid)
2676			return (v);
2677
2678	return (NULL);
2679}
2680
2681static uint32_t
2682bge_chipid(device_t dev)
2683{
2684	uint32_t id;
2685
2686	id = pci_read_config(dev, BGE_PCI_MISC_CTL, 4) >>
2687	    BGE_PCIMISCCTL_ASICREV_SHIFT;
2688	if (BGE_ASICREV(id) == BGE_ASICREV_USE_PRODID_REG) {
2689		/*
2690		 * Find the ASCI revision.  Different chips use different
2691		 * registers.
2692		 */
2693		switch (pci_get_device(dev)) {
2694		case BCOM_DEVICEID_BCM5717C:
2695			/* 5717 C0 seems to belong to 5720 line. */
2696			id = BGE_CHIPID_BCM5720_A0;
2697			break;
2698		case BCOM_DEVICEID_BCM5717:
2699		case BCOM_DEVICEID_BCM5718:
2700		case BCOM_DEVICEID_BCM5719:
2701		case BCOM_DEVICEID_BCM5720:
2702		case BCOM_DEVICEID_BCM5725:
2703		case BCOM_DEVICEID_BCM5727:
2704		case BCOM_DEVICEID_BCM5762:
2705		case BCOM_DEVICEID_BCM57764:
2706		case BCOM_DEVICEID_BCM57767:
2707		case BCOM_DEVICEID_BCM57787:
2708			id = pci_read_config(dev,
2709			    BGE_PCI_GEN2_PRODID_ASICREV, 4);
2710			break;
2711		case BCOM_DEVICEID_BCM57761:
2712		case BCOM_DEVICEID_BCM57762:
2713		case BCOM_DEVICEID_BCM57765:
2714		case BCOM_DEVICEID_BCM57766:
2715		case BCOM_DEVICEID_BCM57781:
2716		case BCOM_DEVICEID_BCM57782:
2717		case BCOM_DEVICEID_BCM57785:
2718		case BCOM_DEVICEID_BCM57786:
2719		case BCOM_DEVICEID_BCM57791:
2720		case BCOM_DEVICEID_BCM57795:
2721			id = pci_read_config(dev,
2722			    BGE_PCI_GEN15_PRODID_ASICREV, 4);
2723			break;
2724		default:
2725			id = pci_read_config(dev, BGE_PCI_PRODID_ASICREV, 4);
2726		}
2727	}
2728	return (id);
2729}
2730
2731/*
2732 * Probe for a Broadcom chip. Check the PCI vendor and device IDs
2733 * against our list and return its name if we find a match.
2734 *
2735 * Note that since the Broadcom controller contains VPD support, we
2736 * try to get the device name string from the controller itself instead
2737 * of the compiled-in string. It guarantees we'll always announce the
2738 * right product name. We fall back to the compiled-in string when
2739 * VPD is unavailable or corrupt.
2740 */
2741static int
2742bge_probe(device_t dev)
2743{
2744	char buf[96];
2745	char model[64];
2746	const struct bge_revision *br;
2747	const char *pname;
2748	struct bge_softc *sc;
2749	const struct bge_type *t = bge_devs;
2750	const struct bge_vendor *v;
2751	uint32_t id;
2752	uint16_t did, vid;
2753
2754	sc = device_get_softc(dev);
2755	sc->bge_dev = dev;
2756	vid = pci_get_vendor(dev);
2757	did = pci_get_device(dev);
2758	while(t->bge_vid != 0) {
2759		if ((vid == t->bge_vid) && (did == t->bge_did)) {
2760			id = bge_chipid(dev);
2761			br = bge_lookup_rev(id);
2762			if (bge_has_eaddr(sc) &&
2763			    pci_get_vpd_ident(dev, &pname) == 0)
2764				snprintf(model, sizeof(model), "%s", pname);
2765			else {
2766				v = bge_lookup_vendor(vid);
2767				snprintf(model, sizeof(model), "%s %s",
2768				    v != NULL ? v->v_name : "Unknown",
2769				    br != NULL ? br->br_name :
2770				    "NetXtreme/NetLink Ethernet Controller");
2771			}
2772			snprintf(buf, sizeof(buf), "%s, %sASIC rev. %#08x",
2773			    model, br != NULL ? "" : "unknown ", id);
2774			device_set_desc_copy(dev, buf);
2775			return (BUS_PROBE_DEFAULT);
2776		}
2777		t++;
2778	}
2779
2780	return (ENXIO);
2781}
2782
2783static void
2784bge_dma_free(struct bge_softc *sc)
2785{
2786	int i;
2787
2788	/* Destroy DMA maps for RX buffers. */
2789	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
2790		if (sc->bge_cdata.bge_rx_std_dmamap[i])
2791			bus_dmamap_destroy(sc->bge_cdata.bge_rx_mtag,
2792			    sc->bge_cdata.bge_rx_std_dmamap[i]);
2793	}
2794	if (sc->bge_cdata.bge_rx_std_sparemap)
2795		bus_dmamap_destroy(sc->bge_cdata.bge_rx_mtag,
2796		    sc->bge_cdata.bge_rx_std_sparemap);
2797
2798	/* Destroy DMA maps for jumbo RX buffers. */
2799	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
2800		if (sc->bge_cdata.bge_rx_jumbo_dmamap[i])
2801			bus_dmamap_destroy(sc->bge_cdata.bge_mtag_jumbo,
2802			    sc->bge_cdata.bge_rx_jumbo_dmamap[i]);
2803	}
2804	if (sc->bge_cdata.bge_rx_jumbo_sparemap)
2805		bus_dmamap_destroy(sc->bge_cdata.bge_mtag_jumbo,
2806		    sc->bge_cdata.bge_rx_jumbo_sparemap);
2807
2808	/* Destroy DMA maps for TX buffers. */
2809	for (i = 0; i < BGE_TX_RING_CNT; i++) {
2810		if (sc->bge_cdata.bge_tx_dmamap[i])
2811			bus_dmamap_destroy(sc->bge_cdata.bge_tx_mtag,
2812			    sc->bge_cdata.bge_tx_dmamap[i]);
2813	}
2814
2815	if (sc->bge_cdata.bge_rx_mtag)
2816		bus_dma_tag_destroy(sc->bge_cdata.bge_rx_mtag);
2817	if (sc->bge_cdata.bge_mtag_jumbo)
2818		bus_dma_tag_destroy(sc->bge_cdata.bge_mtag_jumbo);
2819	if (sc->bge_cdata.bge_tx_mtag)
2820		bus_dma_tag_destroy(sc->bge_cdata.bge_tx_mtag);
2821
2822	/* Destroy standard RX ring. */
2823	if (sc->bge_cdata.bge_rx_std_ring_map)
2824		bus_dmamap_unload(sc->bge_cdata.bge_rx_std_ring_tag,
2825		    sc->bge_cdata.bge_rx_std_ring_map);
2826	if (sc->bge_cdata.bge_rx_std_ring_map && sc->bge_ldata.bge_rx_std_ring)
2827		bus_dmamem_free(sc->bge_cdata.bge_rx_std_ring_tag,
2828		    sc->bge_ldata.bge_rx_std_ring,
2829		    sc->bge_cdata.bge_rx_std_ring_map);
2830
2831	if (sc->bge_cdata.bge_rx_std_ring_tag)
2832		bus_dma_tag_destroy(sc->bge_cdata.bge_rx_std_ring_tag);
2833
2834	/* Destroy jumbo RX ring. */
2835	if (sc->bge_cdata.bge_rx_jumbo_ring_map)
2836		bus_dmamap_unload(sc->bge_cdata.bge_rx_jumbo_ring_tag,
2837		    sc->bge_cdata.bge_rx_jumbo_ring_map);
2838
2839	if (sc->bge_cdata.bge_rx_jumbo_ring_map &&
2840	    sc->bge_ldata.bge_rx_jumbo_ring)
2841		bus_dmamem_free(sc->bge_cdata.bge_rx_jumbo_ring_tag,
2842		    sc->bge_ldata.bge_rx_jumbo_ring,
2843		    sc->bge_cdata.bge_rx_jumbo_ring_map);
2844
2845	if (sc->bge_cdata.bge_rx_jumbo_ring_tag)
2846		bus_dma_tag_destroy(sc->bge_cdata.bge_rx_jumbo_ring_tag);
2847
2848	/* Destroy RX return ring. */
2849	if (sc->bge_cdata.bge_rx_return_ring_map)
2850		bus_dmamap_unload(sc->bge_cdata.bge_rx_return_ring_tag,
2851		    sc->bge_cdata.bge_rx_return_ring_map);
2852
2853	if (sc->bge_cdata.bge_rx_return_ring_map &&
2854	    sc->bge_ldata.bge_rx_return_ring)
2855		bus_dmamem_free(sc->bge_cdata.bge_rx_return_ring_tag,
2856		    sc->bge_ldata.bge_rx_return_ring,
2857		    sc->bge_cdata.bge_rx_return_ring_map);
2858
2859	if (sc->bge_cdata.bge_rx_return_ring_tag)
2860		bus_dma_tag_destroy(sc->bge_cdata.bge_rx_return_ring_tag);
2861
2862	/* Destroy TX ring. */
2863	if (sc->bge_cdata.bge_tx_ring_map)
2864		bus_dmamap_unload(sc->bge_cdata.bge_tx_ring_tag,
2865		    sc->bge_cdata.bge_tx_ring_map);
2866
2867	if (sc->bge_cdata.bge_tx_ring_map && sc->bge_ldata.bge_tx_ring)
2868		bus_dmamem_free(sc->bge_cdata.bge_tx_ring_tag,
2869		    sc->bge_ldata.bge_tx_ring,
2870		    sc->bge_cdata.bge_tx_ring_map);
2871
2872	if (sc->bge_cdata.bge_tx_ring_tag)
2873		bus_dma_tag_destroy(sc->bge_cdata.bge_tx_ring_tag);
2874
2875	/* Destroy status block. */
2876	if (sc->bge_cdata.bge_status_map)
2877		bus_dmamap_unload(sc->bge_cdata.bge_status_tag,
2878		    sc->bge_cdata.bge_status_map);
2879
2880	if (sc->bge_cdata.bge_status_map && sc->bge_ldata.bge_status_block)
2881		bus_dmamem_free(sc->bge_cdata.bge_status_tag,
2882		    sc->bge_ldata.bge_status_block,
2883		    sc->bge_cdata.bge_status_map);
2884
2885	if (sc->bge_cdata.bge_status_tag)
2886		bus_dma_tag_destroy(sc->bge_cdata.bge_status_tag);
2887
2888	/* Destroy statistics block. */
2889	if (sc->bge_cdata.bge_stats_map)
2890		bus_dmamap_unload(sc->bge_cdata.bge_stats_tag,
2891		    sc->bge_cdata.bge_stats_map);
2892
2893	if (sc->bge_cdata.bge_stats_map && sc->bge_ldata.bge_stats)
2894		bus_dmamem_free(sc->bge_cdata.bge_stats_tag,
2895		    sc->bge_ldata.bge_stats,
2896		    sc->bge_cdata.bge_stats_map);
2897
2898	if (sc->bge_cdata.bge_stats_tag)
2899		bus_dma_tag_destroy(sc->bge_cdata.bge_stats_tag);
2900
2901	if (sc->bge_cdata.bge_buffer_tag)
2902		bus_dma_tag_destroy(sc->bge_cdata.bge_buffer_tag);
2903
2904	/* Destroy the parent tag. */
2905	if (sc->bge_cdata.bge_parent_tag)
2906		bus_dma_tag_destroy(sc->bge_cdata.bge_parent_tag);
2907}
2908
2909static int
2910bge_dma_ring_alloc(struct bge_softc *sc, bus_size_t alignment,
2911    bus_size_t maxsize, bus_dma_tag_t *tag, uint8_t **ring, bus_dmamap_t *map,
2912    bus_addr_t *paddr, const char *msg)
2913{
2914	struct bge_dmamap_arg ctx;
2915	int error;
2916
2917	error = bus_dma_tag_create(sc->bge_cdata.bge_parent_tag,
2918	    alignment, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
2919	    NULL, maxsize, 1, maxsize, 0, NULL, NULL, tag);
2920	if (error != 0) {
2921		device_printf(sc->bge_dev,
2922		    "could not create %s dma tag\n", msg);
2923		return (ENOMEM);
2924	}
2925	/* Allocate DMA'able memory for ring. */
2926	error = bus_dmamem_alloc(*tag, (void **)ring,
2927	    BUS_DMA_NOWAIT | BUS_DMA_ZERO | BUS_DMA_COHERENT, map);
2928	if (error != 0) {
2929		device_printf(sc->bge_dev,
2930		    "could not allocate DMA'able memory for %s\n", msg);
2931		return (ENOMEM);
2932	}
2933	/* Load the address of the ring. */
2934	ctx.bge_busaddr = 0;
2935	error = bus_dmamap_load(*tag, *map, *ring, maxsize, bge_dma_map_addr,
2936	    &ctx, BUS_DMA_NOWAIT);
2937	if (error != 0) {
2938		device_printf(sc->bge_dev,
2939		    "could not load DMA'able memory for %s\n", msg);
2940		return (ENOMEM);
2941	}
2942	*paddr = ctx.bge_busaddr;
2943	return (0);
2944}
2945
2946static int
2947bge_dma_alloc(struct bge_softc *sc)
2948{
2949	bus_addr_t lowaddr;
2950	bus_size_t rxmaxsegsz, sbsz, txsegsz, txmaxsegsz;
2951	int i, error;
2952
2953	lowaddr = BUS_SPACE_MAXADDR;
2954	if ((sc->bge_flags & BGE_FLAG_40BIT_BUG) != 0)
2955		lowaddr = BGE_DMA_MAXADDR;
2956	/*
2957	 * Allocate the parent bus DMA tag appropriate for PCI.
2958	 */
2959	error = bus_dma_tag_create(bus_get_dma_tag(sc->bge_dev),
2960	    1, 0, lowaddr, BUS_SPACE_MAXADDR, NULL,
2961	    NULL, BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT,
2962	    0, NULL, NULL, &sc->bge_cdata.bge_parent_tag);
2963	if (error != 0) {
2964		device_printf(sc->bge_dev,
2965		    "could not allocate parent dma tag\n");
2966		return (ENOMEM);
2967	}
2968
2969	/* Create tag for standard RX ring. */
2970	error = bge_dma_ring_alloc(sc, PAGE_SIZE, BGE_STD_RX_RING_SZ,
2971	    &sc->bge_cdata.bge_rx_std_ring_tag,
2972	    (uint8_t **)&sc->bge_ldata.bge_rx_std_ring,
2973	    &sc->bge_cdata.bge_rx_std_ring_map,
2974	    &sc->bge_ldata.bge_rx_std_ring_paddr, "RX ring");
2975	if (error)
2976		return (error);
2977
2978	/* Create tag for RX return ring. */
2979	error = bge_dma_ring_alloc(sc, PAGE_SIZE, BGE_RX_RTN_RING_SZ(sc),
2980	    &sc->bge_cdata.bge_rx_return_ring_tag,
2981	    (uint8_t **)&sc->bge_ldata.bge_rx_return_ring,
2982	    &sc->bge_cdata.bge_rx_return_ring_map,
2983	    &sc->bge_ldata.bge_rx_return_ring_paddr, "RX return ring");
2984	if (error)
2985		return (error);
2986
2987	/* Create tag for TX ring. */
2988	error = bge_dma_ring_alloc(sc, PAGE_SIZE, BGE_TX_RING_SZ,
2989	    &sc->bge_cdata.bge_tx_ring_tag,
2990	    (uint8_t **)&sc->bge_ldata.bge_tx_ring,
2991	    &sc->bge_cdata.bge_tx_ring_map,
2992	    &sc->bge_ldata.bge_tx_ring_paddr, "TX ring");
2993	if (error)
2994		return (error);
2995
2996	/*
2997	 * Create tag for status block.
2998	 * Because we only use single Tx/Rx/Rx return ring, use
2999	 * minimum status block size except BCM5700 AX/BX which
3000	 * seems to want to see full status block size regardless
3001	 * of configured number of ring.
3002	 */
3003	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
3004	    sc->bge_chipid != BGE_CHIPID_BCM5700_C0)
3005		sbsz = BGE_STATUS_BLK_SZ;
3006	else
3007		sbsz = 32;
3008	error = bge_dma_ring_alloc(sc, PAGE_SIZE, sbsz,
3009	    &sc->bge_cdata.bge_status_tag,
3010	    (uint8_t **)&sc->bge_ldata.bge_status_block,
3011	    &sc->bge_cdata.bge_status_map,
3012	    &sc->bge_ldata.bge_status_block_paddr, "status block");
3013	if (error)
3014		return (error);
3015
3016	/* Create tag for statistics block. */
3017	error = bge_dma_ring_alloc(sc, PAGE_SIZE, BGE_STATS_SZ,
3018	    &sc->bge_cdata.bge_stats_tag,
3019	    (uint8_t **)&sc->bge_ldata.bge_stats,
3020	    &sc->bge_cdata.bge_stats_map,
3021	    &sc->bge_ldata.bge_stats_paddr, "statistics block");
3022	if (error)
3023		return (error);
3024
3025	/* Create tag for jumbo RX ring. */
3026	if (BGE_IS_JUMBO_CAPABLE(sc)) {
3027		error = bge_dma_ring_alloc(sc, PAGE_SIZE, BGE_JUMBO_RX_RING_SZ,
3028		    &sc->bge_cdata.bge_rx_jumbo_ring_tag,
3029		    (uint8_t **)&sc->bge_ldata.bge_rx_jumbo_ring,
3030		    &sc->bge_cdata.bge_rx_jumbo_ring_map,
3031		    &sc->bge_ldata.bge_rx_jumbo_ring_paddr, "jumbo RX ring");
3032		if (error)
3033			return (error);
3034	}
3035
3036	/* Create parent tag for buffers. */
3037	if ((sc->bge_flags & BGE_FLAG_4G_BNDRY_BUG) != 0) {
3038		/*
3039		 * XXX
3040		 * watchdog timeout issue was observed on BCM5704 which
3041		 * lives behind PCI-X bridge(e.g AMD 8131 PCI-X bridge).
3042		 * Both limiting DMA address space to 32bits and flushing
3043		 * mailbox write seem to address the issue.
3044		 */
3045		if (sc->bge_pcixcap != 0)
3046			lowaddr = BUS_SPACE_MAXADDR_32BIT;
3047	}
3048	error = bus_dma_tag_create(bus_get_dma_tag(sc->bge_dev), 1, 0, lowaddr,
3049	    BUS_SPACE_MAXADDR, NULL, NULL, BUS_SPACE_MAXSIZE_32BIT, 0,
3050	    BUS_SPACE_MAXSIZE_32BIT, 0, NULL, NULL,
3051	    &sc->bge_cdata.bge_buffer_tag);
3052	if (error != 0) {
3053		device_printf(sc->bge_dev,
3054		    "could not allocate buffer dma tag\n");
3055		return (ENOMEM);
3056	}
3057	/* Create tag for Tx mbufs. */
3058	if (sc->bge_flags & (BGE_FLAG_TSO | BGE_FLAG_TSO3)) {
3059		txsegsz = BGE_TSOSEG_SZ;
3060		txmaxsegsz = 65535 + sizeof(struct ether_vlan_header);
3061	} else {
3062		txsegsz = MCLBYTES;
3063		txmaxsegsz = MCLBYTES * BGE_NSEG_NEW;
3064	}
3065	error = bus_dma_tag_create(sc->bge_cdata.bge_buffer_tag, 1,
3066	    0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
3067	    txmaxsegsz, BGE_NSEG_NEW, txsegsz, 0, NULL, NULL,
3068	    &sc->bge_cdata.bge_tx_mtag);
3069
3070	if (error) {
3071		device_printf(sc->bge_dev, "could not allocate TX dma tag\n");
3072		return (ENOMEM);
3073	}
3074
3075	/* Create tag for Rx mbufs. */
3076	if (sc->bge_flags & BGE_FLAG_JUMBO_STD)
3077		rxmaxsegsz = MJUM9BYTES;
3078	else
3079		rxmaxsegsz = MCLBYTES;
3080	error = bus_dma_tag_create(sc->bge_cdata.bge_buffer_tag, 1, 0,
3081	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, rxmaxsegsz, 1,
3082	    rxmaxsegsz, 0, NULL, NULL, &sc->bge_cdata.bge_rx_mtag);
3083
3084	if (error) {
3085		device_printf(sc->bge_dev, "could not allocate RX dma tag\n");
3086		return (ENOMEM);
3087	}
3088
3089	/* Create DMA maps for RX buffers. */
3090	error = bus_dmamap_create(sc->bge_cdata.bge_rx_mtag, 0,
3091	    &sc->bge_cdata.bge_rx_std_sparemap);
3092	if (error) {
3093		device_printf(sc->bge_dev,
3094		    "can't create spare DMA map for RX\n");
3095		return (ENOMEM);
3096	}
3097	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
3098		error = bus_dmamap_create(sc->bge_cdata.bge_rx_mtag, 0,
3099			    &sc->bge_cdata.bge_rx_std_dmamap[i]);
3100		if (error) {
3101			device_printf(sc->bge_dev,
3102			    "can't create DMA map for RX\n");
3103			return (ENOMEM);
3104		}
3105	}
3106
3107	/* Create DMA maps for TX buffers. */
3108	for (i = 0; i < BGE_TX_RING_CNT; i++) {
3109		error = bus_dmamap_create(sc->bge_cdata.bge_tx_mtag, 0,
3110			    &sc->bge_cdata.bge_tx_dmamap[i]);
3111		if (error) {
3112			device_printf(sc->bge_dev,
3113			    "can't create DMA map for TX\n");
3114			return (ENOMEM);
3115		}
3116	}
3117
3118	/* Create tags for jumbo RX buffers. */
3119	if (BGE_IS_JUMBO_CAPABLE(sc)) {
3120		error = bus_dma_tag_create(sc->bge_cdata.bge_buffer_tag,
3121		    1, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
3122		    NULL, MJUM9BYTES, BGE_NSEG_JUMBO, PAGE_SIZE,
3123		    0, NULL, NULL, &sc->bge_cdata.bge_mtag_jumbo);
3124		if (error) {
3125			device_printf(sc->bge_dev,
3126			    "could not allocate jumbo dma tag\n");
3127			return (ENOMEM);
3128		}
3129		/* Create DMA maps for jumbo RX buffers. */
3130		error = bus_dmamap_create(sc->bge_cdata.bge_mtag_jumbo,
3131		    0, &sc->bge_cdata.bge_rx_jumbo_sparemap);
3132		if (error) {
3133			device_printf(sc->bge_dev,
3134			    "can't create spare DMA map for jumbo RX\n");
3135			return (ENOMEM);
3136		}
3137		for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
3138			error = bus_dmamap_create(sc->bge_cdata.bge_mtag_jumbo,
3139				    0, &sc->bge_cdata.bge_rx_jumbo_dmamap[i]);
3140			if (error) {
3141				device_printf(sc->bge_dev,
3142				    "can't create DMA map for jumbo RX\n");
3143				return (ENOMEM);
3144			}
3145		}
3146	}
3147
3148	return (0);
3149}
3150
3151/*
3152 * Return true if this device has more than one port.
3153 */
3154static int
3155bge_has_multiple_ports(struct bge_softc *sc)
3156{
3157	device_t dev = sc->bge_dev;
3158	u_int b, d, f, fscan, s;
3159
3160	d = pci_get_domain(dev);
3161	b = pci_get_bus(dev);
3162	s = pci_get_slot(dev);
3163	f = pci_get_function(dev);
3164	for (fscan = 0; fscan <= PCI_FUNCMAX; fscan++)
3165		if (fscan != f && pci_find_dbsf(d, b, s, fscan) != NULL)
3166			return (1);
3167	return (0);
3168}
3169
3170/*
3171 * Return true if MSI can be used with this device.
3172 */
3173static int
3174bge_can_use_msi(struct bge_softc *sc)
3175{
3176	int can_use_msi = 0;
3177
3178	if (sc->bge_msi == 0)
3179		return (0);
3180
3181	/* Disable MSI for polling(4). */
3182#ifdef DEVICE_POLLING
3183	return (0);
3184#endif
3185	switch (sc->bge_asicrev) {
3186	case BGE_ASICREV_BCM5714_A0:
3187	case BGE_ASICREV_BCM5714:
3188		/*
3189		 * Apparently, MSI doesn't work when these chips are
3190		 * configured in single-port mode.
3191		 */
3192		if (bge_has_multiple_ports(sc))
3193			can_use_msi = 1;
3194		break;
3195	case BGE_ASICREV_BCM5750:
3196		if (sc->bge_chiprev != BGE_CHIPREV_5750_AX &&
3197		    sc->bge_chiprev != BGE_CHIPREV_5750_BX)
3198			can_use_msi = 1;
3199		break;
3200	default:
3201		if (BGE_IS_575X_PLUS(sc))
3202			can_use_msi = 1;
3203	}
3204	return (can_use_msi);
3205}
3206
3207static int
3208bge_mbox_reorder(struct bge_softc *sc)
3209{
3210	/* Lists of PCI bridges that are known to reorder mailbox writes. */
3211	static const struct mbox_reorder {
3212		const uint16_t vendor;
3213		const uint16_t device;
3214		const char *desc;
3215	} mbox_reorder_lists[] = {
3216		{ 0x1022, 0x7450, "AMD-8131 PCI-X Bridge" },
3217	};
3218	devclass_t pci, pcib;
3219	device_t bus, dev;
3220	int i;
3221
3222	pci = devclass_find("pci");
3223	pcib = devclass_find("pcib");
3224	dev = sc->bge_dev;
3225	bus = device_get_parent(dev);
3226	for (;;) {
3227		dev = device_get_parent(bus);
3228		bus = device_get_parent(dev);
3229		if (device_get_devclass(dev) != pcib)
3230			break;
3231		for (i = 0; i < nitems(mbox_reorder_lists); i++) {
3232			if (pci_get_vendor(dev) ==
3233			    mbox_reorder_lists[i].vendor &&
3234			    pci_get_device(dev) ==
3235			    mbox_reorder_lists[i].device) {
3236				device_printf(sc->bge_dev,
3237				    "enabling MBOX workaround for %s\n",
3238				    mbox_reorder_lists[i].desc);
3239				return (1);
3240			}
3241		}
3242		if (device_get_devclass(bus) != pci)
3243			break;
3244	}
3245	return (0);
3246}
3247
3248static void
3249bge_devinfo(struct bge_softc *sc)
3250{
3251	uint32_t cfg, clk;
3252
3253	device_printf(sc->bge_dev,
3254	    "CHIP ID 0x%08x; ASIC REV 0x%02x; CHIP REV 0x%02x; ",
3255	    sc->bge_chipid, sc->bge_asicrev, sc->bge_chiprev);
3256	if (sc->bge_flags & BGE_FLAG_PCIE)
3257		printf("PCI-E\n");
3258	else if (sc->bge_flags & BGE_FLAG_PCIX) {
3259		printf("PCI-X ");
3260		cfg = CSR_READ_4(sc, BGE_MISC_CFG) & BGE_MISCCFG_BOARD_ID_MASK;
3261		if (cfg == BGE_MISCCFG_BOARD_ID_5704CIOBE)
3262			clk = 133;
3263		else {
3264			clk = CSR_READ_4(sc, BGE_PCI_CLKCTL) & 0x1F;
3265			switch (clk) {
3266			case 0:
3267				clk = 33;
3268				break;
3269			case 2:
3270				clk = 50;
3271				break;
3272			case 4:
3273				clk = 66;
3274				break;
3275			case 6:
3276				clk = 100;
3277				break;
3278			case 7:
3279				clk = 133;
3280				break;
3281			}
3282		}
3283		printf("%u MHz\n", clk);
3284	} else {
3285		if (sc->bge_pcixcap != 0)
3286			printf("PCI on PCI-X ");
3287		else
3288			printf("PCI ");
3289		cfg = pci_read_config(sc->bge_dev, BGE_PCI_PCISTATE, 4);
3290		if (cfg & BGE_PCISTATE_PCI_BUSSPEED)
3291			clk = 66;
3292		else
3293			clk = 33;
3294		if (cfg & BGE_PCISTATE_32BIT_BUS)
3295			printf("%u MHz; 32bit\n", clk);
3296		else
3297			printf("%u MHz; 64bit\n", clk);
3298	}
3299}
3300
3301static int
3302bge_attach(device_t dev)
3303{
3304	struct ifnet *ifp;
3305	struct bge_softc *sc;
3306	uint32_t hwcfg = 0, misccfg, pcistate;
3307	u_char eaddr[ETHER_ADDR_LEN];
3308	int capmask, error, reg, rid, trys;
3309
3310	sc = device_get_softc(dev);
3311	sc->bge_dev = dev;
3312
3313	BGE_LOCK_INIT(sc, device_get_nameunit(dev));
3314	TASK_INIT(&sc->bge_intr_task, 0, bge_intr_task, sc);
3315	callout_init_mtx(&sc->bge_stat_ch, &sc->bge_mtx, 0);
3316
3317	pci_enable_busmaster(dev);
3318
3319	/*
3320	 * Allocate control/status registers.
3321	 */
3322	rid = PCIR_BAR(0);
3323	sc->bge_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
3324	    RF_ACTIVE);
3325
3326	if (sc->bge_res == NULL) {
3327		device_printf (sc->bge_dev, "couldn't map BAR0 memory\n");
3328		error = ENXIO;
3329		goto fail;
3330	}
3331
3332	/* Save various chip information. */
3333	sc->bge_func_addr = pci_get_function(dev);
3334	sc->bge_chipid = bge_chipid(dev);
3335	sc->bge_asicrev = BGE_ASICREV(sc->bge_chipid);
3336	sc->bge_chiprev = BGE_CHIPREV(sc->bge_chipid);
3337
3338	/* Set default PHY address. */
3339	sc->bge_phy_addr = 1;
3340	 /*
3341	  * PHY address mapping for various devices.
3342	  *
3343	  *          | F0 Cu | F0 Sr | F1 Cu | F1 Sr |
3344	  * ---------+-------+-------+-------+-------+
3345	  * BCM57XX  |   1   |   X   |   X   |   X   |
3346	  * BCM5704  |   1   |   X   |   1   |   X   |
3347	  * BCM5717  |   1   |   8   |   2   |   9   |
3348	  * BCM5719  |   1   |   8   |   2   |   9   |
3349	  * BCM5720  |   1   |   8   |   2   |   9   |
3350	  *
3351	  *          | F2 Cu | F2 Sr | F3 Cu | F3 Sr |
3352	  * ---------+-------+-------+-------+-------+
3353	  * BCM57XX  |   X   |   X   |   X   |   X   |
3354	  * BCM5704  |   X   |   X   |   X   |   X   |
3355	  * BCM5717  |   X   |   X   |   X   |   X   |
3356	  * BCM5719  |   3   |   10  |   4   |   11  |
3357	  * BCM5720  |   X   |   X   |   X   |   X   |
3358	  *
3359	  * Other addresses may respond but they are not
3360	  * IEEE compliant PHYs and should be ignored.
3361	  */
3362	if (sc->bge_asicrev == BGE_ASICREV_BCM5717 ||
3363	    sc->bge_asicrev == BGE_ASICREV_BCM5719 ||
3364	    sc->bge_asicrev == BGE_ASICREV_BCM5720) {
3365		if (sc->bge_chipid != BGE_CHIPID_BCM5717_A0) {
3366			if (CSR_READ_4(sc, BGE_SGDIG_STS) &
3367			    BGE_SGDIGSTS_IS_SERDES)
3368				sc->bge_phy_addr = sc->bge_func_addr + 8;
3369			else
3370				sc->bge_phy_addr = sc->bge_func_addr + 1;
3371		} else {
3372			if (CSR_READ_4(sc, BGE_CPMU_PHY_STRAP) &
3373			    BGE_CPMU_PHY_STRAP_IS_SERDES)
3374				sc->bge_phy_addr = sc->bge_func_addr + 8;
3375			else
3376				sc->bge_phy_addr = sc->bge_func_addr + 1;
3377		}
3378	}
3379
3380	if (bge_has_eaddr(sc))
3381		sc->bge_flags |= BGE_FLAG_EADDR;
3382
3383	/* Save chipset family. */
3384	switch (sc->bge_asicrev) {
3385	case BGE_ASICREV_BCM5762:
3386	case BGE_ASICREV_BCM57765:
3387	case BGE_ASICREV_BCM57766:
3388		sc->bge_flags |= BGE_FLAG_57765_PLUS;
3389		/* FALLTHROUGH */
3390	case BGE_ASICREV_BCM5717:
3391	case BGE_ASICREV_BCM5719:
3392	case BGE_ASICREV_BCM5720:
3393		sc->bge_flags |= BGE_FLAG_5717_PLUS | BGE_FLAG_5755_PLUS |
3394		    BGE_FLAG_575X_PLUS | BGE_FLAG_5705_PLUS | BGE_FLAG_JUMBO |
3395		    BGE_FLAG_JUMBO_FRAME;
3396		if (sc->bge_asicrev == BGE_ASICREV_BCM5719 ||
3397		    sc->bge_asicrev == BGE_ASICREV_BCM5720) {
3398			/*
3399			 * Enable work around for DMA engine miscalculation
3400			 * of TXMBUF available space.
3401			 */
3402			sc->bge_flags |= BGE_FLAG_RDMA_BUG;
3403			if (sc->bge_asicrev == BGE_ASICREV_BCM5719 &&
3404			    sc->bge_chipid == BGE_CHIPID_BCM5719_A0) {
3405				/* Jumbo frame on BCM5719 A0 does not work. */
3406				sc->bge_flags &= ~BGE_FLAG_JUMBO;
3407			}
3408		}
3409		break;
3410	case BGE_ASICREV_BCM5755:
3411	case BGE_ASICREV_BCM5761:
3412	case BGE_ASICREV_BCM5784:
3413	case BGE_ASICREV_BCM5785:
3414	case BGE_ASICREV_BCM5787:
3415	case BGE_ASICREV_BCM57780:
3416		sc->bge_flags |= BGE_FLAG_5755_PLUS | BGE_FLAG_575X_PLUS |
3417		    BGE_FLAG_5705_PLUS;
3418		break;
3419	case BGE_ASICREV_BCM5700:
3420	case BGE_ASICREV_BCM5701:
3421	case BGE_ASICREV_BCM5703:
3422	case BGE_ASICREV_BCM5704:
3423		sc->bge_flags |= BGE_FLAG_5700_FAMILY | BGE_FLAG_JUMBO;
3424		break;
3425	case BGE_ASICREV_BCM5714_A0:
3426	case BGE_ASICREV_BCM5780:
3427	case BGE_ASICREV_BCM5714:
3428		sc->bge_flags |= BGE_FLAG_5714_FAMILY | BGE_FLAG_JUMBO_STD;
3429		/* FALLTHROUGH */
3430	case BGE_ASICREV_BCM5750:
3431	case BGE_ASICREV_BCM5752:
3432	case BGE_ASICREV_BCM5906:
3433		sc->bge_flags |= BGE_FLAG_575X_PLUS;
3434		/* FALLTHROUGH */
3435	case BGE_ASICREV_BCM5705:
3436		sc->bge_flags |= BGE_FLAG_5705_PLUS;
3437		break;
3438	}
3439
3440	/* Identify chips with APE processor. */
3441	switch (sc->bge_asicrev) {
3442	case BGE_ASICREV_BCM5717:
3443	case BGE_ASICREV_BCM5719:
3444	case BGE_ASICREV_BCM5720:
3445	case BGE_ASICREV_BCM5761:
3446	case BGE_ASICREV_BCM5762:
3447		sc->bge_flags |= BGE_FLAG_APE;
3448		break;
3449	}
3450
3451	/* Chips with APE need BAR2 access for APE registers/memory. */
3452	if ((sc->bge_flags & BGE_FLAG_APE) != 0) {
3453		rid = PCIR_BAR(2);
3454		sc->bge_res2 = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
3455		    RF_ACTIVE);
3456		if (sc->bge_res2 == NULL) {
3457			device_printf (sc->bge_dev,
3458			    "couldn't map BAR2 memory\n");
3459			error = ENXIO;
3460			goto fail;
3461		}
3462
3463		/* Enable APE register/memory access by host driver. */
3464		pcistate = pci_read_config(dev, BGE_PCI_PCISTATE, 4);
3465		pcistate |= BGE_PCISTATE_ALLOW_APE_CTLSPC_WR |
3466		    BGE_PCISTATE_ALLOW_APE_SHMEM_WR |
3467		    BGE_PCISTATE_ALLOW_APE_PSPACE_WR;
3468		pci_write_config(dev, BGE_PCI_PCISTATE, pcistate, 4);
3469
3470		bge_ape_lock_init(sc);
3471		bge_ape_read_fw_ver(sc);
3472	}
3473
3474	/* Add SYSCTLs, requires the chipset family to be set. */
3475	bge_add_sysctls(sc);
3476
3477	/* Identify the chips that use an CPMU. */
3478	if (BGE_IS_5717_PLUS(sc) ||
3479	    sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
3480	    sc->bge_asicrev == BGE_ASICREV_BCM5761 ||
3481	    sc->bge_asicrev == BGE_ASICREV_BCM5785 ||
3482	    sc->bge_asicrev == BGE_ASICREV_BCM57780)
3483		sc->bge_flags |= BGE_FLAG_CPMU_PRESENT;
3484	if ((sc->bge_flags & BGE_FLAG_CPMU_PRESENT) != 0)
3485		sc->bge_mi_mode = BGE_MIMODE_500KHZ_CONST;
3486	else
3487		sc->bge_mi_mode = BGE_MIMODE_BASE;
3488	/* Enable auto polling for BCM570[0-5]. */
3489	if (BGE_IS_5700_FAMILY(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5705)
3490		sc->bge_mi_mode |= BGE_MIMODE_AUTOPOLL;
3491
3492	/*
3493	 * All Broadcom controllers have 4GB boundary DMA bug.
3494	 * Whenever an address crosses a multiple of the 4GB boundary
3495	 * (including 4GB, 8Gb, 12Gb, etc.) and makes the transition
3496	 * from 0xX_FFFF_FFFF to 0x(X+1)_0000_0000 an internal DMA
3497	 * state machine will lockup and cause the device to hang.
3498	 */
3499	sc->bge_flags |= BGE_FLAG_4G_BNDRY_BUG;
3500
3501	/* BCM5755 or higher and BCM5906 have short DMA bug. */
3502	if (BGE_IS_5755_PLUS(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5906)
3503		sc->bge_flags |= BGE_FLAG_SHORT_DMA_BUG;
3504
3505	/*
3506	 * BCM5719 cannot handle DMA requests for DMA segments that
3507	 * have larger than 4KB in size.  However the maximum DMA
3508	 * segment size created in DMA tag is 4KB for TSO, so we
3509	 * wouldn't encounter the issue here.
3510	 */
3511	if (sc->bge_asicrev == BGE_ASICREV_BCM5719)
3512		sc->bge_flags |= BGE_FLAG_4K_RDMA_BUG;
3513
3514	misccfg = CSR_READ_4(sc, BGE_MISC_CFG) & BGE_MISCCFG_BOARD_ID_MASK;
3515	if (sc->bge_asicrev == BGE_ASICREV_BCM5705) {
3516		if (misccfg == BGE_MISCCFG_BOARD_ID_5788 ||
3517		    misccfg == BGE_MISCCFG_BOARD_ID_5788M)
3518			sc->bge_flags |= BGE_FLAG_5788;
3519	}
3520
3521	capmask = BMSR_DEFCAPMASK;
3522	if ((sc->bge_asicrev == BGE_ASICREV_BCM5703 &&
3523	    (misccfg == 0x4000 || misccfg == 0x8000)) ||
3524	    (sc->bge_asicrev == BGE_ASICREV_BCM5705 &&
3525	    pci_get_vendor(dev) == BCOM_VENDORID &&
3526	    (pci_get_device(dev) == BCOM_DEVICEID_BCM5901 ||
3527	    pci_get_device(dev) == BCOM_DEVICEID_BCM5901A2 ||
3528	    pci_get_device(dev) == BCOM_DEVICEID_BCM5705F)) ||
3529	    (pci_get_vendor(dev) == BCOM_VENDORID &&
3530	    (pci_get_device(dev) == BCOM_DEVICEID_BCM5751F ||
3531	    pci_get_device(dev) == BCOM_DEVICEID_BCM5753F ||
3532	    pci_get_device(dev) == BCOM_DEVICEID_BCM5787F)) ||
3533	    pci_get_device(dev) == BCOM_DEVICEID_BCM57790 ||
3534	    pci_get_device(dev) == BCOM_DEVICEID_BCM57791 ||
3535	    pci_get_device(dev) == BCOM_DEVICEID_BCM57795 ||
3536	    sc->bge_asicrev == BGE_ASICREV_BCM5906) {
3537		/* These chips are 10/100 only. */
3538		capmask &= ~BMSR_EXTSTAT;
3539		sc->bge_phy_flags |= BGE_PHY_NO_WIRESPEED;
3540	}
3541
3542	/*
3543	 * Some controllers seem to require a special firmware to use
3544	 * TSO. But the firmware is not available to FreeBSD and Linux
3545	 * claims that the TSO performed by the firmware is slower than
3546	 * hardware based TSO. Moreover the firmware based TSO has one
3547	 * known bug which can't handle TSO if Ethernet header + IP/TCP
3548	 * header is greater than 80 bytes. A workaround for the TSO
3549	 * bug exist but it seems it's too expensive than not using
3550	 * TSO at all. Some hardwares also have the TSO bug so limit
3551	 * the TSO to the controllers that are not affected TSO issues
3552	 * (e.g. 5755 or higher).
3553	 */
3554	if (BGE_IS_5717_PLUS(sc)) {
3555		/* BCM5717 requires different TSO configuration. */
3556		sc->bge_flags |= BGE_FLAG_TSO3;
3557		if (sc->bge_asicrev == BGE_ASICREV_BCM5719 &&
3558		    sc->bge_chipid == BGE_CHIPID_BCM5719_A0) {
3559			/* TSO on BCM5719 A0 does not work. */
3560			sc->bge_flags &= ~BGE_FLAG_TSO3;
3561		}
3562	} else if (BGE_IS_5755_PLUS(sc)) {
3563		/*
3564		 * BCM5754 and BCM5787 shares the same ASIC id so
3565		 * explicit device id check is required.
3566		 * Due to unknown reason TSO does not work on BCM5755M.
3567		 */
3568		if (pci_get_device(dev) != BCOM_DEVICEID_BCM5754 &&
3569		    pci_get_device(dev) != BCOM_DEVICEID_BCM5754M &&
3570		    pci_get_device(dev) != BCOM_DEVICEID_BCM5755M)
3571			sc->bge_flags |= BGE_FLAG_TSO;
3572	}
3573
3574	/*
3575	 * Check if this is a PCI-X or PCI Express device.
3576	 */
3577	if (pci_find_cap(dev, PCIY_EXPRESS, &reg) == 0) {
3578		/*
3579		 * Found a PCI Express capabilities register, this
3580		 * must be a PCI Express device.
3581		 */
3582		sc->bge_flags |= BGE_FLAG_PCIE;
3583		sc->bge_expcap = reg;
3584		/* Extract supported maximum payload size. */
3585		sc->bge_mps = pci_read_config(dev, sc->bge_expcap +
3586		    PCIER_DEVICE_CAP, 2);
3587		sc->bge_mps = 128 << (sc->bge_mps & PCIEM_CAP_MAX_PAYLOAD);
3588		if (sc->bge_asicrev == BGE_ASICREV_BCM5719 ||
3589		    sc->bge_asicrev == BGE_ASICREV_BCM5720)
3590			sc->bge_expmrq = 2048;
3591		else
3592			sc->bge_expmrq = 4096;
3593		pci_set_max_read_req(dev, sc->bge_expmrq);
3594	} else {
3595		/*
3596		 * Check if the device is in PCI-X Mode.
3597		 * (This bit is not valid on PCI Express controllers.)
3598		 */
3599		if (pci_find_cap(dev, PCIY_PCIX, &reg) == 0)
3600			sc->bge_pcixcap = reg;
3601		if ((pci_read_config(dev, BGE_PCI_PCISTATE, 4) &
3602		    BGE_PCISTATE_PCI_BUSMODE) == 0)
3603			sc->bge_flags |= BGE_FLAG_PCIX;
3604	}
3605
3606	/*
3607	 * The 40bit DMA bug applies to the 5714/5715 controllers and is
3608	 * not actually a MAC controller bug but an issue with the embedded
3609	 * PCIe to PCI-X bridge in the device. Use 40bit DMA workaround.
3610	 */
3611	if (BGE_IS_5714_FAMILY(sc) && (sc->bge_flags & BGE_FLAG_PCIX))
3612		sc->bge_flags |= BGE_FLAG_40BIT_BUG;
3613	/*
3614	 * Some PCI-X bridges are known to trigger write reordering to
3615	 * the mailbox registers. Typical phenomena is watchdog timeouts
3616	 * caused by out-of-order TX completions.  Enable workaround for
3617	 * PCI-X devices that live behind these bridges.
3618	 * Note, PCI-X controllers can run in PCI mode so we can't use
3619	 * BGE_FLAG_PCIX flag to detect PCI-X controllers.
3620	 */
3621	if (sc->bge_pcixcap != 0 && bge_mbox_reorder(sc) != 0)
3622		sc->bge_flags |= BGE_FLAG_MBOX_REORDER;
3623	/*
3624	 * Allocate the interrupt, using MSI if possible.  These devices
3625	 * support 8 MSI messages, but only the first one is used in
3626	 * normal operation.
3627	 */
3628	rid = 0;
3629	if (pci_find_cap(sc->bge_dev, PCIY_MSI, &reg) == 0) {
3630		sc->bge_msicap = reg;
3631		reg = 1;
3632		if (bge_can_use_msi(sc) && pci_alloc_msi(dev, &reg) == 0) {
3633			rid = 1;
3634			sc->bge_flags |= BGE_FLAG_MSI;
3635		}
3636	}
3637
3638	/*
3639	 * All controllers except BCM5700 supports tagged status but
3640	 * we use tagged status only for MSI case on BCM5717. Otherwise
3641	 * MSI on BCM5717 does not work.
3642	 */
3643#ifndef DEVICE_POLLING
3644	if (sc->bge_flags & BGE_FLAG_MSI && BGE_IS_5717_PLUS(sc))
3645		sc->bge_flags |= BGE_FLAG_TAGGED_STATUS;
3646#endif
3647
3648	sc->bge_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
3649	    RF_ACTIVE | (rid != 0 ? 0 : RF_SHAREABLE));
3650
3651	if (sc->bge_irq == NULL) {
3652		device_printf(sc->bge_dev, "couldn't map interrupt\n");
3653		error = ENXIO;
3654		goto fail;
3655	}
3656
3657	bge_devinfo(sc);
3658
3659	sc->bge_asf_mode = 0;
3660	/* No ASF if APE present. */
3661	if ((sc->bge_flags & BGE_FLAG_APE) == 0) {
3662		if (bge_allow_asf && (bge_readmem_ind(sc, BGE_SRAM_DATA_SIG) ==
3663		    BGE_SRAM_DATA_SIG_MAGIC)) {
3664			if (bge_readmem_ind(sc, BGE_SRAM_DATA_CFG) &
3665			    BGE_HWCFG_ASF) {
3666				sc->bge_asf_mode |= ASF_ENABLE;
3667				sc->bge_asf_mode |= ASF_STACKUP;
3668				if (BGE_IS_575X_PLUS(sc))
3669					sc->bge_asf_mode |= ASF_NEW_HANDSHAKE;
3670			}
3671		}
3672	}
3673
3674	bge_stop_fw(sc);
3675	bge_sig_pre_reset(sc, BGE_RESET_SHUTDOWN);
3676	if (bge_reset(sc)) {
3677		device_printf(sc->bge_dev, "chip reset failed\n");
3678		error = ENXIO;
3679		goto fail;
3680	}
3681
3682	bge_sig_legacy(sc, BGE_RESET_SHUTDOWN);
3683	bge_sig_post_reset(sc, BGE_RESET_SHUTDOWN);
3684
3685	if (bge_chipinit(sc)) {
3686		device_printf(sc->bge_dev, "chip initialization failed\n");
3687		error = ENXIO;
3688		goto fail;
3689	}
3690
3691	error = bge_get_eaddr(sc, eaddr);
3692	if (error) {
3693		device_printf(sc->bge_dev,
3694		    "failed to read station address\n");
3695		error = ENXIO;
3696		goto fail;
3697	}
3698
3699	/* 5705 limits RX return ring to 512 entries. */
3700	if (BGE_IS_5717_PLUS(sc))
3701		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT;
3702	else if (BGE_IS_5705_PLUS(sc))
3703		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT_5705;
3704	else
3705		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT;
3706
3707	if (bge_dma_alloc(sc)) {
3708		device_printf(sc->bge_dev,
3709		    "failed to allocate DMA resources\n");
3710		error = ENXIO;
3711		goto fail;
3712	}
3713
3714	/* Set default tuneable values. */
3715	sc->bge_stat_ticks = BGE_TICKS_PER_SEC;
3716	sc->bge_rx_coal_ticks = 150;
3717	sc->bge_tx_coal_ticks = 150;
3718	sc->bge_rx_max_coal_bds = 10;
3719	sc->bge_tx_max_coal_bds = 10;
3720
3721	/* Initialize checksum features to use. */
3722	sc->bge_csum_features = BGE_CSUM_FEATURES;
3723	if (sc->bge_forced_udpcsum != 0)
3724		sc->bge_csum_features |= CSUM_UDP;
3725
3726	/* Set up ifnet structure */
3727	ifp = sc->bge_ifp = if_alloc(IFT_ETHER);
3728	if (ifp == NULL) {
3729		device_printf(sc->bge_dev, "failed to if_alloc()\n");
3730		error = ENXIO;
3731		goto fail;
3732	}
3733	ifp->if_softc = sc;
3734	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
3735	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
3736	ifp->if_ioctl = bge_ioctl;
3737	ifp->if_start = bge_start;
3738	ifp->if_init = bge_init;
3739	ifp->if_snd.ifq_drv_maxlen = BGE_TX_RING_CNT - 1;
3740	IFQ_SET_MAXLEN(&ifp->if_snd, ifp->if_snd.ifq_drv_maxlen);
3741	IFQ_SET_READY(&ifp->if_snd);
3742	ifp->if_hwassist = sc->bge_csum_features;
3743	ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_VLAN_HWTAGGING |
3744	    IFCAP_VLAN_MTU;
3745	if ((sc->bge_flags & (BGE_FLAG_TSO | BGE_FLAG_TSO3)) != 0) {
3746		ifp->if_hwassist |= CSUM_TSO;
3747		ifp->if_capabilities |= IFCAP_TSO4 | IFCAP_VLAN_HWTSO;
3748	}
3749#ifdef IFCAP_VLAN_HWCSUM
3750	ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
3751#endif
3752	ifp->if_capenable = ifp->if_capabilities;
3753#ifdef DEVICE_POLLING
3754	ifp->if_capabilities |= IFCAP_POLLING;
3755#endif
3756
3757	/*
3758	 * 5700 B0 chips do not support checksumming correctly due
3759	 * to hardware bugs.
3760	 */
3761	if (sc->bge_chipid == BGE_CHIPID_BCM5700_B0) {
3762		ifp->if_capabilities &= ~IFCAP_HWCSUM;
3763		ifp->if_capenable &= ~IFCAP_HWCSUM;
3764		ifp->if_hwassist = 0;
3765	}
3766
3767	/*
3768	 * Figure out what sort of media we have by checking the
3769	 * hardware config word in the first 32k of NIC internal memory,
3770	 * or fall back to examining the EEPROM if necessary.
3771	 * Note: on some BCM5700 cards, this value appears to be unset.
3772	 * If that's the case, we have to rely on identifying the NIC
3773	 * by its PCI subsystem ID, as we do below for the SysKonnect
3774	 * SK-9D41.
3775	 */
3776	if (bge_readmem_ind(sc, BGE_SRAM_DATA_SIG) == BGE_SRAM_DATA_SIG_MAGIC)
3777		hwcfg = bge_readmem_ind(sc, BGE_SRAM_DATA_CFG);
3778	else if ((sc->bge_flags & BGE_FLAG_EADDR) &&
3779	    (sc->bge_asicrev != BGE_ASICREV_BCM5906)) {
3780		if (bge_read_eeprom(sc, (caddr_t)&hwcfg, BGE_EE_HWCFG_OFFSET,
3781		    sizeof(hwcfg))) {
3782			device_printf(sc->bge_dev, "failed to read EEPROM\n");
3783			error = ENXIO;
3784			goto fail;
3785		}
3786		hwcfg = ntohl(hwcfg);
3787	}
3788
3789	/* The SysKonnect SK-9D41 is a 1000baseSX card. */
3790	if ((pci_read_config(dev, BGE_PCI_SUBSYS, 4) >> 16) ==
3791	    SK_SUBSYSID_9D41 || (hwcfg & BGE_HWCFG_MEDIA) == BGE_MEDIA_FIBER) {
3792		if (BGE_IS_5705_PLUS(sc)) {
3793			sc->bge_flags |= BGE_FLAG_MII_SERDES;
3794			sc->bge_phy_flags |= BGE_PHY_NO_WIRESPEED;
3795		} else
3796			sc->bge_flags |= BGE_FLAG_TBI;
3797	}
3798
3799	/* Set various PHY bug flags. */
3800	if (sc->bge_chipid == BGE_CHIPID_BCM5701_A0 ||
3801	    sc->bge_chipid == BGE_CHIPID_BCM5701_B0)
3802		sc->bge_phy_flags |= BGE_PHY_CRC_BUG;
3803	if (sc->bge_chiprev == BGE_CHIPREV_5703_AX ||
3804	    sc->bge_chiprev == BGE_CHIPREV_5704_AX)
3805		sc->bge_phy_flags |= BGE_PHY_ADC_BUG;
3806	if (sc->bge_chipid == BGE_CHIPID_BCM5704_A0)
3807		sc->bge_phy_flags |= BGE_PHY_5704_A0_BUG;
3808	if (pci_get_subvendor(dev) == DELL_VENDORID)
3809		sc->bge_phy_flags |= BGE_PHY_NO_3LED;
3810	if ((BGE_IS_5705_PLUS(sc)) &&
3811	    sc->bge_asicrev != BGE_ASICREV_BCM5906 &&
3812	    sc->bge_asicrev != BGE_ASICREV_BCM5785 &&
3813	    sc->bge_asicrev != BGE_ASICREV_BCM57780 &&
3814	    !BGE_IS_5717_PLUS(sc)) {
3815		if (sc->bge_asicrev == BGE_ASICREV_BCM5755 ||
3816		    sc->bge_asicrev == BGE_ASICREV_BCM5761 ||
3817		    sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
3818		    sc->bge_asicrev == BGE_ASICREV_BCM5787) {
3819			if (pci_get_device(dev) != BCOM_DEVICEID_BCM5722 &&
3820			    pci_get_device(dev) != BCOM_DEVICEID_BCM5756)
3821				sc->bge_phy_flags |= BGE_PHY_JITTER_BUG;
3822			if (pci_get_device(dev) == BCOM_DEVICEID_BCM5755M)
3823				sc->bge_phy_flags |= BGE_PHY_ADJUST_TRIM;
3824		} else
3825			sc->bge_phy_flags |= BGE_PHY_BER_BUG;
3826	}
3827
3828	/*
3829	 * Don't enable Ethernet@WireSpeed for the 5700 or the
3830	 * 5705 A0 and A1 chips.
3831	 */
3832	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 ||
3833	    (sc->bge_asicrev == BGE_ASICREV_BCM5705 &&
3834	    (sc->bge_chipid != BGE_CHIPID_BCM5705_A0 &&
3835	    sc->bge_chipid != BGE_CHIPID_BCM5705_A1)))
3836		sc->bge_phy_flags |= BGE_PHY_NO_WIRESPEED;
3837
3838	if (sc->bge_flags & BGE_FLAG_TBI) {
3839		ifmedia_init(&sc->bge_ifmedia, IFM_IMASK, bge_ifmedia_upd,
3840		    bge_ifmedia_sts);
3841		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER | IFM_1000_SX, 0, NULL);
3842		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER | IFM_1000_SX | IFM_FDX,
3843		    0, NULL);
3844		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER | IFM_AUTO, 0, NULL);
3845		ifmedia_set(&sc->bge_ifmedia, IFM_ETHER | IFM_AUTO);
3846		sc->bge_ifmedia.ifm_media = sc->bge_ifmedia.ifm_cur->ifm_media;
3847	} else {
3848		/*
3849		 * Do transceiver setup and tell the firmware the
3850		 * driver is down so we can try to get access the
3851		 * probe if ASF is running.  Retry a couple of times
3852		 * if we get a conflict with the ASF firmware accessing
3853		 * the PHY.
3854		 */
3855		trys = 0;
3856		BGE_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3857again:
3858		bge_asf_driver_up(sc);
3859
3860		error = mii_attach(dev, &sc->bge_miibus, ifp, bge_ifmedia_upd,
3861		    bge_ifmedia_sts, capmask, sc->bge_phy_addr, MII_OFFSET_ANY,
3862		    MIIF_DOPAUSE);
3863		if (error != 0) {
3864			if (trys++ < 4) {
3865				device_printf(sc->bge_dev, "Try again\n");
3866				bge_miibus_writereg(sc->bge_dev,
3867				    sc->bge_phy_addr, MII_BMCR, BMCR_RESET);
3868				goto again;
3869			}
3870			device_printf(sc->bge_dev, "attaching PHYs failed\n");
3871			goto fail;
3872		}
3873
3874		/*
3875		 * Now tell the firmware we are going up after probing the PHY
3876		 */
3877		if (sc->bge_asf_mode & ASF_STACKUP)
3878			BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3879	}
3880
3881	/*
3882	 * When using the BCM5701 in PCI-X mode, data corruption has
3883	 * been observed in the first few bytes of some received packets.
3884	 * Aligning the packet buffer in memory eliminates the corruption.
3885	 * Unfortunately, this misaligns the packet payloads.  On platforms
3886	 * which do not support unaligned accesses, we will realign the
3887	 * payloads by copying the received packets.
3888	 */
3889	if (sc->bge_asicrev == BGE_ASICREV_BCM5701 &&
3890	    sc->bge_flags & BGE_FLAG_PCIX)
3891                sc->bge_flags |= BGE_FLAG_RX_ALIGNBUG;
3892
3893	/*
3894	 * Call MI attach routine.
3895	 */
3896	ether_ifattach(ifp, eaddr);
3897
3898	/* Tell upper layer we support long frames. */
3899	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
3900
3901	/*
3902	 * Hookup IRQ last.
3903	 */
3904	if (BGE_IS_5755_PLUS(sc) && sc->bge_flags & BGE_FLAG_MSI) {
3905		/* Take advantage of single-shot MSI. */
3906		CSR_WRITE_4(sc, BGE_MSI_MODE, CSR_READ_4(sc, BGE_MSI_MODE) &
3907		    ~BGE_MSIMODE_ONE_SHOT_DISABLE);
3908		sc->bge_tq = taskqueue_create_fast("bge_taskq", M_WAITOK,
3909		    taskqueue_thread_enqueue, &sc->bge_tq);
3910		if (sc->bge_tq == NULL) {
3911			device_printf(dev, "could not create taskqueue.\n");
3912			ether_ifdetach(ifp);
3913			error = ENOMEM;
3914			goto fail;
3915		}
3916		error = taskqueue_start_threads(&sc->bge_tq, 1, PI_NET,
3917		    "%s taskq", device_get_nameunit(sc->bge_dev));
3918		if (error != 0) {
3919			device_printf(dev, "could not start threads.\n");
3920			ether_ifdetach(ifp);
3921			goto fail;
3922		}
3923		error = bus_setup_intr(dev, sc->bge_irq,
3924		    INTR_TYPE_NET | INTR_MPSAFE, bge_msi_intr, NULL, sc,
3925		    &sc->bge_intrhand);
3926	} else
3927		error = bus_setup_intr(dev, sc->bge_irq,
3928		    INTR_TYPE_NET | INTR_MPSAFE, NULL, bge_intr, sc,
3929		    &sc->bge_intrhand);
3930
3931	if (error) {
3932		ether_ifdetach(ifp);
3933		device_printf(sc->bge_dev, "couldn't set up irq\n");
3934	}
3935
3936fail:
3937	if (error)
3938		bge_detach(dev);
3939	return (error);
3940}
3941
3942static int
3943bge_detach(device_t dev)
3944{
3945	struct bge_softc *sc;
3946	struct ifnet *ifp;
3947
3948	sc = device_get_softc(dev);
3949	ifp = sc->bge_ifp;
3950
3951#ifdef DEVICE_POLLING
3952	if (ifp->if_capenable & IFCAP_POLLING)
3953		ether_poll_deregister(ifp);
3954#endif
3955
3956	if (device_is_attached(dev)) {
3957		ether_ifdetach(ifp);
3958		BGE_LOCK(sc);
3959		bge_stop(sc);
3960		BGE_UNLOCK(sc);
3961		callout_drain(&sc->bge_stat_ch);
3962	}
3963
3964	if (sc->bge_tq)
3965		taskqueue_drain(sc->bge_tq, &sc->bge_intr_task);
3966
3967	if (sc->bge_flags & BGE_FLAG_TBI)
3968		ifmedia_removeall(&sc->bge_ifmedia);
3969	else if (sc->bge_miibus != NULL) {
3970		bus_generic_detach(dev);
3971		device_delete_child(dev, sc->bge_miibus);
3972	}
3973
3974	bge_release_resources(sc);
3975
3976	return (0);
3977}
3978
3979static void
3980bge_release_resources(struct bge_softc *sc)
3981{
3982	device_t dev;
3983
3984	dev = sc->bge_dev;
3985
3986	if (sc->bge_tq != NULL)
3987		taskqueue_free(sc->bge_tq);
3988
3989	if (sc->bge_intrhand != NULL)
3990		bus_teardown_intr(dev, sc->bge_irq, sc->bge_intrhand);
3991
3992	if (sc->bge_irq != NULL) {
3993		bus_release_resource(dev, SYS_RES_IRQ,
3994		    rman_get_rid(sc->bge_irq), sc->bge_irq);
3995		pci_release_msi(dev);
3996	}
3997
3998	if (sc->bge_res != NULL)
3999		bus_release_resource(dev, SYS_RES_MEMORY,
4000		    rman_get_rid(sc->bge_res), sc->bge_res);
4001
4002	if (sc->bge_res2 != NULL)
4003		bus_release_resource(dev, SYS_RES_MEMORY,
4004		    rman_get_rid(sc->bge_res2), sc->bge_res2);
4005
4006	if (sc->bge_ifp != NULL)
4007		if_free(sc->bge_ifp);
4008
4009	bge_dma_free(sc);
4010
4011	if (mtx_initialized(&sc->bge_mtx))	/* XXX */
4012		BGE_LOCK_DESTROY(sc);
4013}
4014
4015static int
4016bge_reset(struct bge_softc *sc)
4017{
4018	device_t dev;
4019	uint32_t cachesize, command, mac_mode, mac_mode_mask, reset, val;
4020	void (*write_op)(struct bge_softc *, int, int);
4021	uint16_t devctl;
4022	int i;
4023
4024	dev = sc->bge_dev;
4025
4026	mac_mode_mask = BGE_MACMODE_HALF_DUPLEX | BGE_MACMODE_PORTMODE;
4027	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) != 0)
4028		mac_mode_mask |= BGE_MACMODE_APE_RX_EN | BGE_MACMODE_APE_TX_EN;
4029	mac_mode = CSR_READ_4(sc, BGE_MAC_MODE) & mac_mode_mask;
4030
4031	if (BGE_IS_575X_PLUS(sc) && !BGE_IS_5714_FAMILY(sc) &&
4032	    (sc->bge_asicrev != BGE_ASICREV_BCM5906)) {
4033		if (sc->bge_flags & BGE_FLAG_PCIE)
4034			write_op = bge_writemem_direct;
4035		else
4036			write_op = bge_writemem_ind;
4037	} else
4038		write_op = bge_writereg_ind;
4039
4040	if (sc->bge_asicrev != BGE_ASICREV_BCM5700 &&
4041	    sc->bge_asicrev != BGE_ASICREV_BCM5701) {
4042		CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_SET1);
4043		for (i = 0; i < 8000; i++) {
4044			if (CSR_READ_4(sc, BGE_NVRAM_SWARB) &
4045			    BGE_NVRAMSWARB_GNT1)
4046				break;
4047			DELAY(20);
4048		}
4049		if (i == 8000) {
4050			if (bootverbose)
4051				device_printf(dev, "NVRAM lock timedout!\n");
4052		}
4053	}
4054	/* Take APE lock when performing reset. */
4055	bge_ape_lock(sc, BGE_APE_LOCK_GRC);
4056
4057	/* Save some important PCI state. */
4058	cachesize = pci_read_config(dev, BGE_PCI_CACHESZ, 4);
4059	command = pci_read_config(dev, BGE_PCI_CMD, 4);
4060
4061	pci_write_config(dev, BGE_PCI_MISC_CTL,
4062	    BGE_PCIMISCCTL_INDIRECT_ACCESS | BGE_PCIMISCCTL_MASK_PCI_INTR |
4063	    BGE_HIF_SWAP_OPTIONS | BGE_PCIMISCCTL_PCISTATE_RW, 4);
4064
4065	/* Disable fastboot on controllers that support it. */
4066	if (sc->bge_asicrev == BGE_ASICREV_BCM5752 ||
4067	    BGE_IS_5755_PLUS(sc)) {
4068		if (bootverbose)
4069			device_printf(dev, "Disabling fastboot\n");
4070		CSR_WRITE_4(sc, BGE_FASTBOOT_PC, 0x0);
4071	}
4072
4073	/*
4074	 * Write the magic number to SRAM at offset 0xB50.
4075	 * When firmware finishes its initialization it will
4076	 * write ~BGE_SRAM_FW_MB_MAGIC to the same location.
4077	 */
4078	bge_writemem_ind(sc, BGE_SRAM_FW_MB, BGE_SRAM_FW_MB_MAGIC);
4079
4080	reset = BGE_MISCCFG_RESET_CORE_CLOCKS | BGE_32BITTIME_66MHZ;
4081
4082	/* XXX: Broadcom Linux driver. */
4083	if (sc->bge_flags & BGE_FLAG_PCIE) {
4084		if (sc->bge_asicrev != BGE_ASICREV_BCM5785 &&
4085		    (sc->bge_flags & BGE_FLAG_5717_PLUS) == 0) {
4086			if (CSR_READ_4(sc, 0x7E2C) == 0x60)	/* PCIE 1.0 */
4087				CSR_WRITE_4(sc, 0x7E2C, 0x20);
4088		}
4089		if (sc->bge_chipid != BGE_CHIPID_BCM5750_A0) {
4090			/* Prevent PCIE link training during global reset */
4091			CSR_WRITE_4(sc, BGE_MISC_CFG, 1 << 29);
4092			reset |= 1 << 29;
4093		}
4094	}
4095
4096	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
4097		val = CSR_READ_4(sc, BGE_VCPU_STATUS);
4098		CSR_WRITE_4(sc, BGE_VCPU_STATUS,
4099		    val | BGE_VCPU_STATUS_DRV_RESET);
4100		val = CSR_READ_4(sc, BGE_VCPU_EXT_CTRL);
4101		CSR_WRITE_4(sc, BGE_VCPU_EXT_CTRL,
4102		    val & ~BGE_VCPU_EXT_CTRL_HALT_CPU);
4103	}
4104
4105	/*
4106	 * Set GPHY Power Down Override to leave GPHY
4107	 * powered up in D0 uninitialized.
4108	 */
4109	if (BGE_IS_5705_PLUS(sc) &&
4110	    (sc->bge_flags & BGE_FLAG_CPMU_PRESENT) == 0)
4111		reset |= BGE_MISCCFG_GPHY_PD_OVERRIDE;
4112
4113	/* Issue global reset */
4114	write_op(sc, BGE_MISC_CFG, reset);
4115
4116	if (sc->bge_flags & BGE_FLAG_PCIE)
4117		DELAY(100 * 1000);
4118	else
4119		DELAY(1000);
4120
4121	/* XXX: Broadcom Linux driver. */
4122	if (sc->bge_flags & BGE_FLAG_PCIE) {
4123		if (sc->bge_chipid == BGE_CHIPID_BCM5750_A0) {
4124			DELAY(500000); /* wait for link training to complete */
4125			val = pci_read_config(dev, 0xC4, 4);
4126			pci_write_config(dev, 0xC4, val | (1 << 15), 4);
4127		}
4128		devctl = pci_read_config(dev,
4129		    sc->bge_expcap + PCIER_DEVICE_CTL, 2);
4130		/* Clear enable no snoop and disable relaxed ordering. */
4131		devctl &= ~(PCIEM_CTL_RELAXED_ORD_ENABLE |
4132		    PCIEM_CTL_NOSNOOP_ENABLE);
4133		pci_write_config(dev, sc->bge_expcap + PCIER_DEVICE_CTL,
4134		    devctl, 2);
4135		pci_set_max_read_req(dev, sc->bge_expmrq);
4136		/* Clear error status. */
4137		pci_write_config(dev, sc->bge_expcap + PCIER_DEVICE_STA,
4138		    PCIEM_STA_CORRECTABLE_ERROR |
4139		    PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR |
4140		    PCIEM_STA_UNSUPPORTED_REQ, 2);
4141	}
4142
4143	/* Reset some of the PCI state that got zapped by reset. */
4144	pci_write_config(dev, BGE_PCI_MISC_CTL,
4145	    BGE_PCIMISCCTL_INDIRECT_ACCESS | BGE_PCIMISCCTL_MASK_PCI_INTR |
4146	    BGE_HIF_SWAP_OPTIONS | BGE_PCIMISCCTL_PCISTATE_RW, 4);
4147	val = BGE_PCISTATE_ROM_ENABLE | BGE_PCISTATE_ROM_RETRY_ENABLE;
4148	if (sc->bge_chipid == BGE_CHIPID_BCM5704_A0 &&
4149	    (sc->bge_flags & BGE_FLAG_PCIX) != 0)
4150		val |= BGE_PCISTATE_RETRY_SAME_DMA;
4151	if ((sc->bge_mfw_flags & BGE_MFW_ON_APE) != 0)
4152		val |= BGE_PCISTATE_ALLOW_APE_CTLSPC_WR |
4153		    BGE_PCISTATE_ALLOW_APE_SHMEM_WR |
4154		    BGE_PCISTATE_ALLOW_APE_PSPACE_WR;
4155	pci_write_config(dev, BGE_PCI_PCISTATE, val, 4);
4156	pci_write_config(dev, BGE_PCI_CACHESZ, cachesize, 4);
4157	pci_write_config(dev, BGE_PCI_CMD, command, 4);
4158	/*
4159	 * Disable PCI-X relaxed ordering to ensure status block update
4160	 * comes first then packet buffer DMA. Otherwise driver may
4161	 * read stale status block.
4162	 */
4163	if (sc->bge_flags & BGE_FLAG_PCIX) {
4164		devctl = pci_read_config(dev,
4165		    sc->bge_pcixcap + PCIXR_COMMAND, 2);
4166		devctl &= ~PCIXM_COMMAND_ERO;
4167		if (sc->bge_asicrev == BGE_ASICREV_BCM5703) {
4168			devctl &= ~PCIXM_COMMAND_MAX_READ;
4169			devctl |= PCIXM_COMMAND_MAX_READ_2048;
4170		} else if (sc->bge_asicrev == BGE_ASICREV_BCM5704) {
4171			devctl &= ~(PCIXM_COMMAND_MAX_SPLITS |
4172			    PCIXM_COMMAND_MAX_READ);
4173			devctl |= PCIXM_COMMAND_MAX_READ_2048;
4174		}
4175		pci_write_config(dev, sc->bge_pcixcap + PCIXR_COMMAND,
4176		    devctl, 2);
4177	}
4178	/* Re-enable MSI, if necessary, and enable the memory arbiter. */
4179	if (BGE_IS_5714_FAMILY(sc)) {
4180		/* This chip disables MSI on reset. */
4181		if (sc->bge_flags & BGE_FLAG_MSI) {
4182			val = pci_read_config(dev,
4183			    sc->bge_msicap + PCIR_MSI_CTRL, 2);
4184			pci_write_config(dev,
4185			    sc->bge_msicap + PCIR_MSI_CTRL,
4186			    val | PCIM_MSICTRL_MSI_ENABLE, 2);
4187			val = CSR_READ_4(sc, BGE_MSI_MODE);
4188			CSR_WRITE_4(sc, BGE_MSI_MODE,
4189			    val | BGE_MSIMODE_ENABLE);
4190		}
4191		val = CSR_READ_4(sc, BGE_MARB_MODE);
4192		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE | val);
4193	} else
4194		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
4195
4196	/* Fix up byte swapping. */
4197	CSR_WRITE_4(sc, BGE_MODE_CTL, bge_dma_swap_options(sc));
4198
4199	val = CSR_READ_4(sc, BGE_MAC_MODE);
4200	val = (val & ~mac_mode_mask) | mac_mode;
4201	CSR_WRITE_4(sc, BGE_MAC_MODE, val);
4202	DELAY(40);
4203
4204	bge_ape_unlock(sc, BGE_APE_LOCK_GRC);
4205
4206	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
4207		for (i = 0; i < BGE_TIMEOUT; i++) {
4208			val = CSR_READ_4(sc, BGE_VCPU_STATUS);
4209			if (val & BGE_VCPU_STATUS_INIT_DONE)
4210				break;
4211			DELAY(100);
4212		}
4213		if (i == BGE_TIMEOUT) {
4214			device_printf(dev, "reset timed out\n");
4215			return (1);
4216		}
4217	} else {
4218		/*
4219		 * Poll until we see the 1's complement of the magic number.
4220		 * This indicates that the firmware initialization is complete.
4221		 * We expect this to fail if no chip containing the Ethernet
4222		 * address is fitted though.
4223		 */
4224		for (i = 0; i < BGE_TIMEOUT; i++) {
4225			DELAY(10);
4226			val = bge_readmem_ind(sc, BGE_SRAM_FW_MB);
4227			if (val == ~BGE_SRAM_FW_MB_MAGIC)
4228				break;
4229		}
4230
4231		if ((sc->bge_flags & BGE_FLAG_EADDR) && i == BGE_TIMEOUT)
4232			device_printf(dev,
4233			    "firmware handshake timed out, found 0x%08x\n",
4234			    val);
4235		/* BCM57765 A0 needs additional time before accessing. */
4236		if (sc->bge_chipid == BGE_CHIPID_BCM57765_A0)
4237			DELAY(10 * 1000);	/* XXX */
4238	}
4239
4240	/*
4241	 * The 5704 in TBI mode apparently needs some special
4242	 * adjustment to insure the SERDES drive level is set
4243	 * to 1.2V.
4244	 */
4245	if (sc->bge_asicrev == BGE_ASICREV_BCM5704 &&
4246	    sc->bge_flags & BGE_FLAG_TBI) {
4247		val = CSR_READ_4(sc, BGE_SERDES_CFG);
4248		val = (val & ~0xFFF) | 0x880;
4249		CSR_WRITE_4(sc, BGE_SERDES_CFG, val);
4250	}
4251
4252	/* XXX: Broadcom Linux driver. */
4253	if (sc->bge_flags & BGE_FLAG_PCIE &&
4254	    !BGE_IS_5717_PLUS(sc) &&
4255	    sc->bge_chipid != BGE_CHIPID_BCM5750_A0 &&
4256	    sc->bge_asicrev != BGE_ASICREV_BCM5785) {
4257		/* Enable Data FIFO protection. */
4258		val = CSR_READ_4(sc, 0x7C00);
4259		CSR_WRITE_4(sc, 0x7C00, val | (1 << 25));
4260	}
4261
4262	if (sc->bge_asicrev == BGE_ASICREV_BCM5720)
4263		BGE_CLRBIT(sc, BGE_CPMU_CLCK_ORIDE,
4264		    CPMU_CLCK_ORIDE_MAC_ORIDE_EN);
4265
4266	return (0);
4267}
4268
4269static __inline void
4270bge_rxreuse_std(struct bge_softc *sc, int i)
4271{
4272	struct bge_rx_bd *r;
4273
4274	r = &sc->bge_ldata.bge_rx_std_ring[sc->bge_std];
4275	r->bge_flags = BGE_RXBDFLAG_END;
4276	r->bge_len = sc->bge_cdata.bge_rx_std_seglen[i];
4277	r->bge_idx = i;
4278	BGE_INC(sc->bge_std, BGE_STD_RX_RING_CNT);
4279}
4280
4281static __inline void
4282bge_rxreuse_jumbo(struct bge_softc *sc, int i)
4283{
4284	struct bge_extrx_bd *r;
4285
4286	r = &sc->bge_ldata.bge_rx_jumbo_ring[sc->bge_jumbo];
4287	r->bge_flags = BGE_RXBDFLAG_JUMBO_RING | BGE_RXBDFLAG_END;
4288	r->bge_len0 = sc->bge_cdata.bge_rx_jumbo_seglen[i][0];
4289	r->bge_len1 = sc->bge_cdata.bge_rx_jumbo_seglen[i][1];
4290	r->bge_len2 = sc->bge_cdata.bge_rx_jumbo_seglen[i][2];
4291	r->bge_len3 = sc->bge_cdata.bge_rx_jumbo_seglen[i][3];
4292	r->bge_idx = i;
4293	BGE_INC(sc->bge_jumbo, BGE_JUMBO_RX_RING_CNT);
4294}
4295
4296/*
4297 * Frame reception handling. This is called if there's a frame
4298 * on the receive return list.
4299 *
4300 * Note: we have to be able to handle two possibilities here:
4301 * 1) the frame is from the jumbo receive ring
4302 * 2) the frame is from the standard receive ring
4303 */
4304
4305static int
4306bge_rxeof(struct bge_softc *sc, uint16_t rx_prod, int holdlck)
4307{
4308	struct ifnet *ifp;
4309	int rx_npkts = 0, stdcnt = 0, jumbocnt = 0;
4310	uint16_t rx_cons;
4311
4312	rx_cons = sc->bge_rx_saved_considx;
4313
4314	/* Nothing to do. */
4315	if (rx_cons == rx_prod)
4316		return (rx_npkts);
4317
4318	ifp = sc->bge_ifp;
4319
4320	bus_dmamap_sync(sc->bge_cdata.bge_rx_return_ring_tag,
4321	    sc->bge_cdata.bge_rx_return_ring_map, BUS_DMASYNC_POSTREAD);
4322	bus_dmamap_sync(sc->bge_cdata.bge_rx_std_ring_tag,
4323	    sc->bge_cdata.bge_rx_std_ring_map, BUS_DMASYNC_POSTWRITE);
4324	if (BGE_IS_JUMBO_CAPABLE(sc) &&
4325	    ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN >
4326	    (MCLBYTES - ETHER_ALIGN))
4327		bus_dmamap_sync(sc->bge_cdata.bge_rx_jumbo_ring_tag,
4328		    sc->bge_cdata.bge_rx_jumbo_ring_map, BUS_DMASYNC_POSTWRITE);
4329
4330	while (rx_cons != rx_prod) {
4331		struct bge_rx_bd	*cur_rx;
4332		uint32_t		rxidx;
4333		struct mbuf		*m = NULL;
4334		uint16_t		vlan_tag = 0;
4335		int			have_tag = 0;
4336
4337#ifdef DEVICE_POLLING
4338		if (ifp->if_capenable & IFCAP_POLLING) {
4339			if (sc->rxcycles <= 0)
4340				break;
4341			sc->rxcycles--;
4342		}
4343#endif
4344
4345		cur_rx = &sc->bge_ldata.bge_rx_return_ring[rx_cons];
4346
4347		rxidx = cur_rx->bge_idx;
4348		BGE_INC(rx_cons, sc->bge_return_ring_cnt);
4349
4350		if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING &&
4351		    cur_rx->bge_flags & BGE_RXBDFLAG_VLAN_TAG) {
4352			have_tag = 1;
4353			vlan_tag = cur_rx->bge_vlan_tag;
4354		}
4355
4356		if (cur_rx->bge_flags & BGE_RXBDFLAG_JUMBO_RING) {
4357			jumbocnt++;
4358			m = sc->bge_cdata.bge_rx_jumbo_chain[rxidx];
4359			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
4360				bge_rxreuse_jumbo(sc, rxidx);
4361				continue;
4362			}
4363			if (bge_newbuf_jumbo(sc, rxidx) != 0) {
4364				bge_rxreuse_jumbo(sc, rxidx);
4365				ifp->if_iqdrops++;
4366				continue;
4367			}
4368			BGE_INC(sc->bge_jumbo, BGE_JUMBO_RX_RING_CNT);
4369		} else {
4370			stdcnt++;
4371			m = sc->bge_cdata.bge_rx_std_chain[rxidx];
4372			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
4373				bge_rxreuse_std(sc, rxidx);
4374				continue;
4375			}
4376			if (bge_newbuf_std(sc, rxidx) != 0) {
4377				bge_rxreuse_std(sc, rxidx);
4378				ifp->if_iqdrops++;
4379				continue;
4380			}
4381			BGE_INC(sc->bge_std, BGE_STD_RX_RING_CNT);
4382		}
4383
4384		ifp->if_ipackets++;
4385#ifndef __NO_STRICT_ALIGNMENT
4386		/*
4387		 * For architectures with strict alignment we must make sure
4388		 * the payload is aligned.
4389		 */
4390		if (sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) {
4391			bcopy(m->m_data, m->m_data + ETHER_ALIGN,
4392			    cur_rx->bge_len);
4393			m->m_data += ETHER_ALIGN;
4394		}
4395#endif
4396		m->m_pkthdr.len = m->m_len = cur_rx->bge_len - ETHER_CRC_LEN;
4397		m->m_pkthdr.rcvif = ifp;
4398
4399		if (ifp->if_capenable & IFCAP_RXCSUM)
4400			bge_rxcsum(sc, cur_rx, m);
4401
4402		/*
4403		 * If we received a packet with a vlan tag,
4404		 * attach that information to the packet.
4405		 */
4406		if (have_tag) {
4407			m->m_pkthdr.ether_vtag = vlan_tag;
4408			m->m_flags |= M_VLANTAG;
4409		}
4410
4411		if (holdlck != 0) {
4412			BGE_UNLOCK(sc);
4413			(*ifp->if_input)(ifp, m);
4414			BGE_LOCK(sc);
4415		} else
4416			(*ifp->if_input)(ifp, m);
4417		rx_npkts++;
4418
4419		if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
4420			return (rx_npkts);
4421	}
4422
4423	bus_dmamap_sync(sc->bge_cdata.bge_rx_return_ring_tag,
4424	    sc->bge_cdata.bge_rx_return_ring_map, BUS_DMASYNC_PREREAD);
4425	if (stdcnt > 0)
4426		bus_dmamap_sync(sc->bge_cdata.bge_rx_std_ring_tag,
4427		    sc->bge_cdata.bge_rx_std_ring_map, BUS_DMASYNC_PREWRITE);
4428
4429	if (jumbocnt > 0)
4430		bus_dmamap_sync(sc->bge_cdata.bge_rx_jumbo_ring_tag,
4431		    sc->bge_cdata.bge_rx_jumbo_ring_map, BUS_DMASYNC_PREWRITE);
4432
4433	sc->bge_rx_saved_considx = rx_cons;
4434	bge_writembx(sc, BGE_MBX_RX_CONS0_LO, sc->bge_rx_saved_considx);
4435	if (stdcnt)
4436		bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, (sc->bge_std +
4437		    BGE_STD_RX_RING_CNT - 1) % BGE_STD_RX_RING_CNT);
4438	if (jumbocnt)
4439		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, (sc->bge_jumbo +
4440		    BGE_JUMBO_RX_RING_CNT - 1) % BGE_JUMBO_RX_RING_CNT);
4441#ifdef notyet
4442	/*
4443	 * This register wraps very quickly under heavy packet drops.
4444	 * If you need correct statistics, you can enable this check.
4445	 */
4446	if (BGE_IS_5705_PLUS(sc))
4447		ifp->if_ierrors += CSR_READ_4(sc, BGE_RXLP_LOCSTAT_IFIN_DROPS);
4448#endif
4449	return (rx_npkts);
4450}
4451
4452static void
4453bge_rxcsum(struct bge_softc *sc, struct bge_rx_bd *cur_rx, struct mbuf *m)
4454{
4455
4456	if (BGE_IS_5717_PLUS(sc)) {
4457		if ((cur_rx->bge_flags & BGE_RXBDFLAG_IPV6) == 0) {
4458			if (cur_rx->bge_flags & BGE_RXBDFLAG_IP_CSUM) {
4459				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
4460				if ((cur_rx->bge_error_flag &
4461				    BGE_RXERRFLAG_IP_CSUM_NOK) == 0)
4462					m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
4463			}
4464			if (cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM) {
4465				m->m_pkthdr.csum_data =
4466				    cur_rx->bge_tcp_udp_csum;
4467				m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
4468				    CSUM_PSEUDO_HDR;
4469			}
4470		}
4471	} else {
4472		if (cur_rx->bge_flags & BGE_RXBDFLAG_IP_CSUM) {
4473			m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
4474			if ((cur_rx->bge_ip_csum ^ 0xFFFF) == 0)
4475				m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
4476		}
4477		if (cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM &&
4478		    m->m_pkthdr.len >= ETHER_MIN_NOPAD) {
4479			m->m_pkthdr.csum_data =
4480			    cur_rx->bge_tcp_udp_csum;
4481			m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
4482			    CSUM_PSEUDO_HDR;
4483		}
4484	}
4485}
4486
4487static void
4488bge_txeof(struct bge_softc *sc, uint16_t tx_cons)
4489{
4490	struct bge_tx_bd *cur_tx;
4491	struct ifnet *ifp;
4492
4493	BGE_LOCK_ASSERT(sc);
4494
4495	/* Nothing to do. */
4496	if (sc->bge_tx_saved_considx == tx_cons)
4497		return;
4498
4499	ifp = sc->bge_ifp;
4500
4501	bus_dmamap_sync(sc->bge_cdata.bge_tx_ring_tag,
4502	    sc->bge_cdata.bge_tx_ring_map, BUS_DMASYNC_POSTWRITE);
4503	/*
4504	 * Go through our tx ring and free mbufs for those
4505	 * frames that have been sent.
4506	 */
4507	while (sc->bge_tx_saved_considx != tx_cons) {
4508		uint32_t		idx;
4509
4510		idx = sc->bge_tx_saved_considx;
4511		cur_tx = &sc->bge_ldata.bge_tx_ring[idx];
4512		if (cur_tx->bge_flags & BGE_TXBDFLAG_END)
4513			ifp->if_opackets++;
4514		if (sc->bge_cdata.bge_tx_chain[idx] != NULL) {
4515			bus_dmamap_sync(sc->bge_cdata.bge_tx_mtag,
4516			    sc->bge_cdata.bge_tx_dmamap[idx],
4517			    BUS_DMASYNC_POSTWRITE);
4518			bus_dmamap_unload(sc->bge_cdata.bge_tx_mtag,
4519			    sc->bge_cdata.bge_tx_dmamap[idx]);
4520			m_freem(sc->bge_cdata.bge_tx_chain[idx]);
4521			sc->bge_cdata.bge_tx_chain[idx] = NULL;
4522		}
4523		sc->bge_txcnt--;
4524		BGE_INC(sc->bge_tx_saved_considx, BGE_TX_RING_CNT);
4525	}
4526
4527	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
4528	if (sc->bge_txcnt == 0)
4529		sc->bge_timer = 0;
4530}
4531
4532#ifdef DEVICE_POLLING
4533static int
4534bge_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
4535{
4536	struct bge_softc *sc = ifp->if_softc;
4537	uint16_t rx_prod, tx_cons;
4538	uint32_t statusword;
4539	int rx_npkts = 0;
4540
4541	BGE_LOCK(sc);
4542	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4543		BGE_UNLOCK(sc);
4544		return (rx_npkts);
4545	}
4546
4547	bus_dmamap_sync(sc->bge_cdata.bge_status_tag,
4548	    sc->bge_cdata.bge_status_map,
4549	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
4550	/* Fetch updates from the status block. */
4551	rx_prod = sc->bge_ldata.bge_status_block->bge_idx[0].bge_rx_prod_idx;
4552	tx_cons = sc->bge_ldata.bge_status_block->bge_idx[0].bge_tx_cons_idx;
4553
4554	statusword = sc->bge_ldata.bge_status_block->bge_status;
4555	/* Clear the status so the next pass only sees the changes. */
4556	sc->bge_ldata.bge_status_block->bge_status = 0;
4557
4558	bus_dmamap_sync(sc->bge_cdata.bge_status_tag,
4559	    sc->bge_cdata.bge_status_map,
4560	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
4561
4562	/* Note link event. It will be processed by POLL_AND_CHECK_STATUS. */
4563	if (statusword & BGE_STATFLAG_LINKSTATE_CHANGED)
4564		sc->bge_link_evt++;
4565
4566	if (cmd == POLL_AND_CHECK_STATUS)
4567		if ((sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
4568		    sc->bge_chipid != BGE_CHIPID_BCM5700_B2) ||
4569		    sc->bge_link_evt || (sc->bge_flags & BGE_FLAG_TBI))
4570			bge_link_upd(sc);
4571
4572	sc->rxcycles = count;
4573	rx_npkts = bge_rxeof(sc, rx_prod, 1);
4574	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4575		BGE_UNLOCK(sc);
4576		return (rx_npkts);
4577	}
4578	bge_txeof(sc, tx_cons);
4579	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
4580		bge_start_locked(ifp);
4581
4582	BGE_UNLOCK(sc);
4583	return (rx_npkts);
4584}
4585#endif /* DEVICE_POLLING */
4586
4587static int
4588bge_msi_intr(void *arg)
4589{
4590	struct bge_softc *sc;
4591
4592	sc = (struct bge_softc *)arg;
4593	/*
4594	 * This interrupt is not shared and controller already
4595	 * disabled further interrupt.
4596	 */
4597	taskqueue_enqueue(sc->bge_tq, &sc->bge_intr_task);
4598	return (FILTER_HANDLED);
4599}
4600
4601static void
4602bge_intr_task(void *arg, int pending)
4603{
4604	struct bge_softc *sc;
4605	struct ifnet *ifp;
4606	uint32_t status, status_tag;
4607	uint16_t rx_prod, tx_cons;
4608
4609	sc = (struct bge_softc *)arg;
4610	ifp = sc->bge_ifp;
4611
4612	BGE_LOCK(sc);
4613	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
4614		BGE_UNLOCK(sc);
4615		return;
4616	}
4617
4618	/* Get updated status block. */
4619	bus_dmamap_sync(sc->bge_cdata.bge_status_tag,
4620	    sc->bge_cdata.bge_status_map,
4621	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
4622
4623	/* Save producer/consumer indices. */
4624	rx_prod = sc->bge_ldata.bge_status_block->bge_idx[0].bge_rx_prod_idx;
4625	tx_cons = sc->bge_ldata.bge_status_block->bge_idx[0].bge_tx_cons_idx;
4626	status = sc->bge_ldata.bge_status_block->bge_status;
4627	status_tag = sc->bge_ldata.bge_status_block->bge_status_tag << 24;
4628	/* Dirty the status flag. */
4629	sc->bge_ldata.bge_status_block->bge_status = 0;
4630	bus_dmamap_sync(sc->bge_cdata.bge_status_tag,
4631	    sc->bge_cdata.bge_status_map,
4632	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
4633	if ((sc->bge_flags & BGE_FLAG_TAGGED_STATUS) == 0)
4634		status_tag = 0;
4635
4636	if ((status & BGE_STATFLAG_LINKSTATE_CHANGED) != 0)
4637		bge_link_upd(sc);
4638
4639	/* Let controller work. */
4640	bge_writembx(sc, BGE_MBX_IRQ0_LO, status_tag);
4641
4642	if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
4643	    sc->bge_rx_saved_considx != rx_prod) {
4644		/* Check RX return ring producer/consumer. */
4645		BGE_UNLOCK(sc);
4646		bge_rxeof(sc, rx_prod, 0);
4647		BGE_LOCK(sc);
4648	}
4649	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
4650		/* Check TX ring producer/consumer. */
4651		bge_txeof(sc, tx_cons);
4652		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
4653			bge_start_locked(ifp);
4654	}
4655	BGE_UNLOCK(sc);
4656}
4657
4658static void
4659bge_intr(void *xsc)
4660{
4661	struct bge_softc *sc;
4662	struct ifnet *ifp;
4663	uint32_t statusword;
4664	uint16_t rx_prod, tx_cons;
4665
4666	sc = xsc;
4667
4668	BGE_LOCK(sc);
4669
4670	ifp = sc->bge_ifp;
4671
4672#ifdef DEVICE_POLLING
4673	if (ifp->if_capenable & IFCAP_POLLING) {
4674		BGE_UNLOCK(sc);
4675		return;
4676	}
4677#endif
4678
4679	/*
4680	 * Ack the interrupt by writing something to BGE_MBX_IRQ0_LO.  Don't
4681	 * disable interrupts by writing nonzero like we used to, since with
4682	 * our current organization this just gives complications and
4683	 * pessimizations for re-enabling interrupts.  We used to have races
4684	 * instead of the necessary complications.  Disabling interrupts
4685	 * would just reduce the chance of a status update while we are
4686	 * running (by switching to the interrupt-mode coalescence
4687	 * parameters), but this chance is already very low so it is more
4688	 * efficient to get another interrupt than prevent it.
4689	 *
4690	 * We do the ack first to ensure another interrupt if there is a
4691	 * status update after the ack.  We don't check for the status
4692	 * changing later because it is more efficient to get another
4693	 * interrupt than prevent it, not quite as above (not checking is
4694	 * a smaller optimization than not toggling the interrupt enable,
4695	 * since checking doesn't involve PCI accesses and toggling require
4696	 * the status check).  So toggling would probably be a pessimization
4697	 * even with MSI.  It would only be needed for using a task queue.
4698	 */
4699	bge_writembx(sc, BGE_MBX_IRQ0_LO, 0);
4700
4701	/*
4702	 * Do the mandatory PCI flush as well as get the link status.
4703	 */
4704	statusword = CSR_READ_4(sc, BGE_MAC_STS) & BGE_MACSTAT_LINK_CHANGED;
4705
4706	/* Make sure the descriptor ring indexes are coherent. */
4707	bus_dmamap_sync(sc->bge_cdata.bge_status_tag,
4708	    sc->bge_cdata.bge_status_map,
4709	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
4710	rx_prod = sc->bge_ldata.bge_status_block->bge_idx[0].bge_rx_prod_idx;
4711	tx_cons = sc->bge_ldata.bge_status_block->bge_idx[0].bge_tx_cons_idx;
4712	sc->bge_ldata.bge_status_block->bge_status = 0;
4713	bus_dmamap_sync(sc->bge_cdata.bge_status_tag,
4714	    sc->bge_cdata.bge_status_map,
4715	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
4716
4717	if ((sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
4718	    sc->bge_chipid != BGE_CHIPID_BCM5700_B2) ||
4719	    statusword || sc->bge_link_evt)
4720		bge_link_upd(sc);
4721
4722	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
4723		/* Check RX return ring producer/consumer. */
4724		bge_rxeof(sc, rx_prod, 1);
4725	}
4726
4727	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
4728		/* Check TX ring producer/consumer. */
4729		bge_txeof(sc, tx_cons);
4730	}
4731
4732	if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
4733	    !IFQ_DRV_IS_EMPTY(&ifp->if_snd))
4734		bge_start_locked(ifp);
4735
4736	BGE_UNLOCK(sc);
4737}
4738
4739static void
4740bge_asf_driver_up(struct bge_softc *sc)
4741{
4742	if (sc->bge_asf_mode & ASF_STACKUP) {
4743		/* Send ASF heartbeat aprox. every 2s */
4744		if (sc->bge_asf_count)
4745			sc->bge_asf_count --;
4746		else {
4747			sc->bge_asf_count = 2;
4748			bge_writemem_ind(sc, BGE_SRAM_FW_CMD_MB,
4749			    BGE_FW_CMD_DRV_ALIVE);
4750			bge_writemem_ind(sc, BGE_SRAM_FW_CMD_LEN_MB, 4);
4751			bge_writemem_ind(sc, BGE_SRAM_FW_CMD_DATA_MB,
4752			    BGE_FW_HB_TIMEOUT_SEC);
4753			CSR_WRITE_4(sc, BGE_RX_CPU_EVENT,
4754			    CSR_READ_4(sc, BGE_RX_CPU_EVENT) |
4755			    BGE_RX_CPU_DRV_EVENT);
4756		}
4757	}
4758}
4759
4760static void
4761bge_tick(void *xsc)
4762{
4763	struct bge_softc *sc = xsc;
4764	struct mii_data *mii = NULL;
4765
4766	BGE_LOCK_ASSERT(sc);
4767
4768	/* Synchronize with possible callout reset/stop. */
4769	if (callout_pending(&sc->bge_stat_ch) ||
4770	    !callout_active(&sc->bge_stat_ch))
4771		return;
4772
4773	if (BGE_IS_5705_PLUS(sc))
4774		bge_stats_update_regs(sc);
4775	else
4776		bge_stats_update(sc);
4777
4778	/* XXX Add APE heartbeat check here? */
4779
4780	if ((sc->bge_flags & BGE_FLAG_TBI) == 0) {
4781		mii = device_get_softc(sc->bge_miibus);
4782		/*
4783		 * Do not touch PHY if we have link up. This could break
4784		 * IPMI/ASF mode or produce extra input errors
4785		 * (extra errors was reported for bcm5701 & bcm5704).
4786		 */
4787		if (!sc->bge_link)
4788			mii_tick(mii);
4789	} else {
4790		/*
4791		 * Since in TBI mode auto-polling can't be used we should poll
4792		 * link status manually. Here we register pending link event
4793		 * and trigger interrupt.
4794		 */
4795#ifdef DEVICE_POLLING
4796		/* In polling mode we poll link state in bge_poll(). */
4797		if (!(sc->bge_ifp->if_capenable & IFCAP_POLLING))
4798#endif
4799		{
4800		sc->bge_link_evt++;
4801		if (sc->bge_asicrev == BGE_ASICREV_BCM5700 ||
4802		    sc->bge_flags & BGE_FLAG_5788)
4803			BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
4804		else
4805			BGE_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
4806		}
4807	}
4808
4809	bge_asf_driver_up(sc);
4810	bge_watchdog(sc);
4811
4812	callout_reset(&sc->bge_stat_ch, hz, bge_tick, sc);
4813}
4814
4815static void
4816bge_stats_update_regs(struct bge_softc *sc)
4817{
4818	struct ifnet *ifp;
4819	struct bge_mac_stats *stats;
4820	uint32_t val;
4821
4822	ifp = sc->bge_ifp;
4823	stats = &sc->bge_mac_stats;
4824
4825	stats->ifHCOutOctets +=
4826	    CSR_READ_4(sc, BGE_TX_MAC_STATS_OCTETS);
4827	stats->etherStatsCollisions +=
4828	    CSR_READ_4(sc, BGE_TX_MAC_STATS_COLLS);
4829	stats->outXonSent +=
4830	    CSR_READ_4(sc, BGE_TX_MAC_STATS_XON_SENT);
4831	stats->outXoffSent +=
4832	    CSR_READ_4(sc, BGE_TX_MAC_STATS_XOFF_SENT);
4833	stats->dot3StatsInternalMacTransmitErrors +=
4834	    CSR_READ_4(sc, BGE_TX_MAC_STATS_ERRORS);
4835	stats->dot3StatsSingleCollisionFrames +=
4836	    CSR_READ_4(sc, BGE_TX_MAC_STATS_SINGLE_COLL);
4837	stats->dot3StatsMultipleCollisionFrames +=
4838	    CSR_READ_4(sc, BGE_TX_MAC_STATS_MULTI_COLL);
4839	stats->dot3StatsDeferredTransmissions +=
4840	    CSR_READ_4(sc, BGE_TX_MAC_STATS_DEFERRED);
4841	stats->dot3StatsExcessiveCollisions +=
4842	    CSR_READ_4(sc, BGE_TX_MAC_STATS_EXCESS_COLL);
4843	stats->dot3StatsLateCollisions +=
4844	    CSR_READ_4(sc, BGE_TX_MAC_STATS_LATE_COLL);
4845	stats->ifHCOutUcastPkts +=
4846	    CSR_READ_4(sc, BGE_TX_MAC_STATS_UCAST);
4847	stats->ifHCOutMulticastPkts +=
4848	    CSR_READ_4(sc, BGE_TX_MAC_STATS_MCAST);
4849	stats->ifHCOutBroadcastPkts +=
4850	    CSR_READ_4(sc, BGE_TX_MAC_STATS_BCAST);
4851
4852	stats->ifHCInOctets +=
4853	    CSR_READ_4(sc, BGE_RX_MAC_STATS_OCTESTS);
4854	stats->etherStatsFragments +=
4855	    CSR_READ_4(sc, BGE_RX_MAC_STATS_FRAGMENTS);
4856	stats->ifHCInUcastPkts +=
4857	    CSR_READ_4(sc, BGE_RX_MAC_STATS_UCAST);
4858	stats->ifHCInMulticastPkts +=
4859	    CSR_READ_4(sc, BGE_RX_MAC_STATS_MCAST);
4860	stats->ifHCInBroadcastPkts +=
4861	    CSR_READ_4(sc, BGE_RX_MAC_STATS_BCAST);
4862	stats->dot3StatsFCSErrors +=
4863	    CSR_READ_4(sc, BGE_RX_MAC_STATS_FCS_ERRORS);
4864	stats->dot3StatsAlignmentErrors +=
4865	    CSR_READ_4(sc, BGE_RX_MAC_STATS_ALGIN_ERRORS);
4866	stats->xonPauseFramesReceived +=
4867	    CSR_READ_4(sc, BGE_RX_MAC_STATS_XON_RCVD);
4868	stats->xoffPauseFramesReceived +=
4869	    CSR_READ_4(sc, BGE_RX_MAC_STATS_XOFF_RCVD);
4870	stats->macControlFramesReceived +=
4871	    CSR_READ_4(sc, BGE_RX_MAC_STATS_CTRL_RCVD);
4872	stats->xoffStateEntered +=
4873	    CSR_READ_4(sc, BGE_RX_MAC_STATS_XOFF_ENTERED);
4874	stats->dot3StatsFramesTooLong +=
4875	    CSR_READ_4(sc, BGE_RX_MAC_STATS_FRAME_TOO_LONG);
4876	stats->etherStatsJabbers +=
4877	    CSR_READ_4(sc, BGE_RX_MAC_STATS_JABBERS);
4878	stats->etherStatsUndersizePkts +=
4879	    CSR_READ_4(sc, BGE_RX_MAC_STATS_UNDERSIZE);
4880
4881	stats->FramesDroppedDueToFilters +=
4882	    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_FILTDROP);
4883	stats->DmaWriteQueueFull +=
4884	    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_DMA_WRQ_FULL);
4885	stats->DmaWriteHighPriQueueFull +=
4886	    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_DMA_HPWRQ_FULL);
4887	stats->NoMoreRxBDs +=
4888	    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_OUT_OF_BDS);
4889	/*
4890	 * XXX
4891	 * Unlike other controllers, BGE_RXLP_LOCSTAT_IFIN_DROPS
4892	 * counter of BCM5717, BCM5718, BCM5719 A0 and BCM5720 A0
4893	 * includes number of unwanted multicast frames.  This comes
4894	 * from silicon bug and known workaround to get rough(not
4895	 * exact) counter is to enable interrupt on MBUF low water
4896	 * attention.  This can be accomplished by setting
4897	 * BGE_HCCMODE_ATTN bit of BGE_HCC_MODE,
4898	 * BGE_BMANMODE_LOMBUF_ATTN bit of BGE_BMAN_MODE and
4899	 * BGE_MODECTL_FLOWCTL_ATTN_INTR bit of BGE_MODE_CTL.
4900	 * However that change would generate more interrupts and
4901	 * there are still possibilities of losing multiple frames
4902	 * during BGE_MODECTL_FLOWCTL_ATTN_INTR interrupt handling.
4903	 * Given that the workaround still would not get correct
4904	 * counter I don't think it's worth to implement it.  So
4905	 * ignore reading the counter on controllers that have the
4906	 * silicon bug.
4907	 */
4908	if (sc->bge_asicrev != BGE_ASICREV_BCM5717 &&
4909	    sc->bge_chipid != BGE_CHIPID_BCM5719_A0 &&
4910	    sc->bge_chipid != BGE_CHIPID_BCM5720_A0)
4911		stats->InputDiscards +=
4912		    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_IFIN_DROPS);
4913	stats->InputErrors +=
4914	    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_IFIN_ERRORS);
4915	stats->RecvThresholdHit +=
4916	    CSR_READ_4(sc, BGE_RXLP_LOCSTAT_RXTHRESH_HIT);
4917
4918	ifp->if_collisions = (u_long)stats->etherStatsCollisions;
4919	ifp->if_ierrors = (u_long)(stats->NoMoreRxBDs + stats->InputDiscards +
4920	    stats->InputErrors);
4921
4922	if (sc->bge_flags & BGE_FLAG_RDMA_BUG) {
4923		/*
4924		 * If controller transmitted more than BGE_NUM_RDMA_CHANNELS
4925		 * frames, it's safe to disable workaround for DMA engine's
4926		 * miscalculation of TXMBUF space.
4927		 */
4928		if (stats->ifHCOutUcastPkts + stats->ifHCOutMulticastPkts +
4929		    stats->ifHCOutBroadcastPkts > BGE_NUM_RDMA_CHANNELS) {
4930			val = CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL);
4931			if (sc->bge_asicrev == BGE_ASICREV_BCM5719)
4932				val &= ~BGE_RDMA_TX_LENGTH_WA_5719;
4933			else
4934				val &= ~BGE_RDMA_TX_LENGTH_WA_5720;
4935			CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL, val);
4936			sc->bge_flags &= ~BGE_FLAG_RDMA_BUG;
4937		}
4938	}
4939}
4940
4941static void
4942bge_stats_clear_regs(struct bge_softc *sc)
4943{
4944
4945	CSR_READ_4(sc, BGE_TX_MAC_STATS_OCTETS);
4946	CSR_READ_4(sc, BGE_TX_MAC_STATS_COLLS);
4947	CSR_READ_4(sc, BGE_TX_MAC_STATS_XON_SENT);
4948	CSR_READ_4(sc, BGE_TX_MAC_STATS_XOFF_SENT);
4949	CSR_READ_4(sc, BGE_TX_MAC_STATS_ERRORS);
4950	CSR_READ_4(sc, BGE_TX_MAC_STATS_SINGLE_COLL);
4951	CSR_READ_4(sc, BGE_TX_MAC_STATS_MULTI_COLL);
4952	CSR_READ_4(sc, BGE_TX_MAC_STATS_DEFERRED);
4953	CSR_READ_4(sc, BGE_TX_MAC_STATS_EXCESS_COLL);
4954	CSR_READ_4(sc, BGE_TX_MAC_STATS_LATE_COLL);
4955	CSR_READ_4(sc, BGE_TX_MAC_STATS_UCAST);
4956	CSR_READ_4(sc, BGE_TX_MAC_STATS_MCAST);
4957	CSR_READ_4(sc, BGE_TX_MAC_STATS_BCAST);
4958
4959	CSR_READ_4(sc, BGE_RX_MAC_STATS_OCTESTS);
4960	CSR_READ_4(sc, BGE_RX_MAC_STATS_FRAGMENTS);
4961	CSR_READ_4(sc, BGE_RX_MAC_STATS_UCAST);
4962	CSR_READ_4(sc, BGE_RX_MAC_STATS_MCAST);
4963	CSR_READ_4(sc, BGE_RX_MAC_STATS_BCAST);
4964	CSR_READ_4(sc, BGE_RX_MAC_STATS_FCS_ERRORS);
4965	CSR_READ_4(sc, BGE_RX_MAC_STATS_ALGIN_ERRORS);
4966	CSR_READ_4(sc, BGE_RX_MAC_STATS_XON_RCVD);
4967	CSR_READ_4(sc, BGE_RX_MAC_STATS_XOFF_RCVD);
4968	CSR_READ_4(sc, BGE_RX_MAC_STATS_CTRL_RCVD);
4969	CSR_READ_4(sc, BGE_RX_MAC_STATS_XOFF_ENTERED);
4970	CSR_READ_4(sc, BGE_RX_MAC_STATS_FRAME_TOO_LONG);
4971	CSR_READ_4(sc, BGE_RX_MAC_STATS_JABBERS);
4972	CSR_READ_4(sc, BGE_RX_MAC_STATS_UNDERSIZE);
4973
4974	CSR_READ_4(sc, BGE_RXLP_LOCSTAT_FILTDROP);
4975	CSR_READ_4(sc, BGE_RXLP_LOCSTAT_DMA_WRQ_FULL);
4976	CSR_READ_4(sc, BGE_RXLP_LOCSTAT_DMA_HPWRQ_FULL);
4977	CSR_READ_4(sc, BGE_RXLP_LOCSTAT_OUT_OF_BDS);
4978	CSR_READ_4(sc, BGE_RXLP_LOCSTAT_IFIN_DROPS);
4979	CSR_READ_4(sc, BGE_RXLP_LOCSTAT_IFIN_ERRORS);
4980	CSR_READ_4(sc, BGE_RXLP_LOCSTAT_RXTHRESH_HIT);
4981}
4982
4983static void
4984bge_stats_update(struct bge_softc *sc)
4985{
4986	struct ifnet *ifp;
4987	bus_size_t stats;
4988	uint32_t cnt;	/* current register value */
4989
4990	ifp = sc->bge_ifp;
4991
4992	stats = BGE_MEMWIN_START + BGE_STATS_BLOCK;
4993
4994#define	READ_STAT(sc, stats, stat) \
4995	CSR_READ_4(sc, stats + offsetof(struct bge_stats, stat))
4996
4997	cnt = READ_STAT(sc, stats, txstats.etherStatsCollisions.bge_addr_lo);
4998	ifp->if_collisions += (uint32_t)(cnt - sc->bge_tx_collisions);
4999	sc->bge_tx_collisions = cnt;
5000
5001	cnt = READ_STAT(sc, stats, nicNoMoreRxBDs.bge_addr_lo);
5002	ifp->if_ierrors += (uint32_t)(cnt - sc->bge_rx_nobds);
5003	sc->bge_rx_nobds = cnt;
5004	cnt = READ_STAT(sc, stats, ifInErrors.bge_addr_lo);
5005	ifp->if_ierrors += (uint32_t)(cnt - sc->bge_rx_inerrs);
5006	sc->bge_rx_inerrs = cnt;
5007	cnt = READ_STAT(sc, stats, ifInDiscards.bge_addr_lo);
5008	ifp->if_ierrors += (uint32_t)(cnt - sc->bge_rx_discards);
5009	sc->bge_rx_discards = cnt;
5010
5011	cnt = READ_STAT(sc, stats, txstats.ifOutDiscards.bge_addr_lo);
5012	ifp->if_oerrors += (uint32_t)(cnt - sc->bge_tx_discards);
5013	sc->bge_tx_discards = cnt;
5014
5015#undef	READ_STAT
5016}
5017
5018/*
5019 * Pad outbound frame to ETHER_MIN_NOPAD for an unusual reason.
5020 * The bge hardware will pad out Tx runts to ETHER_MIN_NOPAD,
5021 * but when such padded frames employ the bge IP/TCP checksum offload,
5022 * the hardware checksum assist gives incorrect results (possibly
5023 * from incorporating its own padding into the UDP/TCP checksum; who knows).
5024 * If we pad such runts with zeros, the onboard checksum comes out correct.
5025 */
5026static __inline int
5027bge_cksum_pad(struct mbuf *m)
5028{
5029	int padlen = ETHER_MIN_NOPAD - m->m_pkthdr.len;
5030	struct mbuf *last;
5031
5032	/* If there's only the packet-header and we can pad there, use it. */
5033	if (m->m_pkthdr.len == m->m_len && M_WRITABLE(m) &&
5034	    M_TRAILINGSPACE(m) >= padlen) {
5035		last = m;
5036	} else {
5037		/*
5038		 * Walk packet chain to find last mbuf. We will either
5039		 * pad there, or append a new mbuf and pad it.
5040		 */
5041		for (last = m; last->m_next != NULL; last = last->m_next);
5042		if (!(M_WRITABLE(last) && M_TRAILINGSPACE(last) >= padlen)) {
5043			/* Allocate new empty mbuf, pad it. Compact later. */
5044			struct mbuf *n;
5045
5046			MGET(n, M_NOWAIT, MT_DATA);
5047			if (n == NULL)
5048				return (ENOBUFS);
5049			n->m_len = 0;
5050			last->m_next = n;
5051			last = n;
5052		}
5053	}
5054
5055	/* Now zero the pad area, to avoid the bge cksum-assist bug. */
5056	memset(mtod(last, caddr_t) + last->m_len, 0, padlen);
5057	last->m_len += padlen;
5058	m->m_pkthdr.len += padlen;
5059
5060	return (0);
5061}
5062
5063static struct mbuf *
5064bge_check_short_dma(struct mbuf *m)
5065{
5066	struct mbuf *n;
5067	int found;
5068
5069	/*
5070	 * If device receive two back-to-back send BDs with less than
5071	 * or equal to 8 total bytes then the device may hang.  The two
5072	 * back-to-back send BDs must in the same frame for this failure
5073	 * to occur.  Scan mbuf chains and see whether two back-to-back
5074	 * send BDs are there. If this is the case, allocate new mbuf
5075	 * and copy the frame to workaround the silicon bug.
5076	 */
5077	for (n = m, found = 0; n != NULL; n = n->m_next) {
5078		if (n->m_len < 8) {
5079			found++;
5080			if (found > 1)
5081				break;
5082			continue;
5083		}
5084		found = 0;
5085	}
5086
5087	if (found > 1) {
5088		n = m_defrag(m, M_NOWAIT);
5089		if (n == NULL)
5090			m_freem(m);
5091	} else
5092		n = m;
5093	return (n);
5094}
5095
5096static struct mbuf *
5097bge_setup_tso(struct bge_softc *sc, struct mbuf *m, uint16_t *mss,
5098    uint16_t *flags)
5099{
5100	struct ip *ip;
5101	struct tcphdr *tcp;
5102	struct mbuf *n;
5103	uint16_t hlen;
5104	uint32_t poff;
5105
5106	if (M_WRITABLE(m) == 0) {
5107		/* Get a writable copy. */
5108		n = m_dup(m, M_NOWAIT);
5109		m_freem(m);
5110		if (n == NULL)
5111			return (NULL);
5112		m = n;
5113	}
5114	m = m_pullup(m, sizeof(struct ether_header) + sizeof(struct ip));
5115	if (m == NULL)
5116		return (NULL);
5117	ip = (struct ip *)(mtod(m, char *) + sizeof(struct ether_header));
5118	poff = sizeof(struct ether_header) + (ip->ip_hl << 2);
5119	m = m_pullup(m, poff + sizeof(struct tcphdr));
5120	if (m == NULL)
5121		return (NULL);
5122	tcp = (struct tcphdr *)(mtod(m, char *) + poff);
5123	m = m_pullup(m, poff + (tcp->th_off << 2));
5124	if (m == NULL)
5125		return (NULL);
5126	/*
5127	 * It seems controller doesn't modify IP length and TCP pseudo
5128	 * checksum. These checksum computed by upper stack should be 0.
5129	 */
5130	*mss = m->m_pkthdr.tso_segsz;
5131	ip = (struct ip *)(mtod(m, char *) + sizeof(struct ether_header));
5132	ip->ip_sum = 0;
5133	ip->ip_len = htons(*mss + (ip->ip_hl << 2) + (tcp->th_off << 2));
5134	/* Clear pseudo checksum computed by TCP stack. */
5135	tcp = (struct tcphdr *)(mtod(m, char *) + poff);
5136	tcp->th_sum = 0;
5137	/*
5138	 * Broadcom controllers uses different descriptor format for
5139	 * TSO depending on ASIC revision. Due to TSO-capable firmware
5140	 * license issue and lower performance of firmware based TSO
5141	 * we only support hardware based TSO.
5142	 */
5143	/* Calculate header length, incl. TCP/IP options, in 32 bit units. */
5144	hlen = ((ip->ip_hl << 2) + (tcp->th_off << 2)) >> 2;
5145	if (sc->bge_flags & BGE_FLAG_TSO3) {
5146		/*
5147		 * For BCM5717 and newer controllers, hardware based TSO
5148		 * uses the 14 lower bits of the bge_mss field to store the
5149		 * MSS and the upper 2 bits to store the lowest 2 bits of
5150		 * the IP/TCP header length.  The upper 6 bits of the header
5151		 * length are stored in the bge_flags[14:10,4] field.  Jumbo
5152		 * frames are supported.
5153		 */
5154		*mss |= ((hlen & 0x3) << 14);
5155		*flags |= ((hlen & 0xF8) << 7) | ((hlen & 0x4) << 2);
5156	} else {
5157		/*
5158		 * For BCM5755 and newer controllers, hardware based TSO uses
5159		 * the lower 11	bits to store the MSS and the upper 5 bits to
5160		 * store the IP/TCP header length. Jumbo frames are not
5161		 * supported.
5162		 */
5163		*mss |= (hlen << 11);
5164	}
5165	return (m);
5166}
5167
5168/*
5169 * Encapsulate an mbuf chain in the tx ring  by coupling the mbuf data
5170 * pointers to descriptors.
5171 */
5172static int
5173bge_encap(struct bge_softc *sc, struct mbuf **m_head, uint32_t *txidx)
5174{
5175	bus_dma_segment_t	segs[BGE_NSEG_NEW];
5176	bus_dmamap_t		map;
5177	struct bge_tx_bd	*d;
5178	struct mbuf		*m = *m_head;
5179	uint32_t		idx = *txidx;
5180	uint16_t		csum_flags, mss, vlan_tag;
5181	int			nsegs, i, error;
5182
5183	csum_flags = 0;
5184	mss = 0;
5185	vlan_tag = 0;
5186	if ((sc->bge_flags & BGE_FLAG_SHORT_DMA_BUG) != 0 &&
5187	    m->m_next != NULL) {
5188		*m_head = bge_check_short_dma(m);
5189		if (*m_head == NULL)
5190			return (ENOBUFS);
5191		m = *m_head;
5192	}
5193	if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
5194		*m_head = m = bge_setup_tso(sc, m, &mss, &csum_flags);
5195		if (*m_head == NULL)
5196			return (ENOBUFS);
5197		csum_flags |= BGE_TXBDFLAG_CPU_PRE_DMA |
5198		    BGE_TXBDFLAG_CPU_POST_DMA;
5199	} else if ((m->m_pkthdr.csum_flags & sc->bge_csum_features) != 0) {
5200		if (m->m_pkthdr.csum_flags & CSUM_IP)
5201			csum_flags |= BGE_TXBDFLAG_IP_CSUM;
5202		if (m->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP)) {
5203			csum_flags |= BGE_TXBDFLAG_TCP_UDP_CSUM;
5204			if (m->m_pkthdr.len < ETHER_MIN_NOPAD &&
5205			    (error = bge_cksum_pad(m)) != 0) {
5206				m_freem(m);
5207				*m_head = NULL;
5208				return (error);
5209			}
5210		}
5211	}
5212
5213	if ((m->m_pkthdr.csum_flags & CSUM_TSO) == 0) {
5214		if (sc->bge_flags & BGE_FLAG_JUMBO_FRAME &&
5215		    m->m_pkthdr.len > ETHER_MAX_LEN)
5216			csum_flags |= BGE_TXBDFLAG_JUMBO_FRAME;
5217		if (sc->bge_forced_collapse > 0 &&
5218		    (sc->bge_flags & BGE_FLAG_PCIE) != 0 && m->m_next != NULL) {
5219			/*
5220			 * Forcedly collapse mbuf chains to overcome hardware
5221			 * limitation which only support a single outstanding
5222			 * DMA read operation.
5223			 */
5224			if (sc->bge_forced_collapse == 1)
5225				m = m_defrag(m, M_NOWAIT);
5226			else
5227				m = m_collapse(m, M_NOWAIT,
5228				    sc->bge_forced_collapse);
5229			if (m == NULL)
5230				m = *m_head;
5231			*m_head = m;
5232		}
5233	}
5234
5235	map = sc->bge_cdata.bge_tx_dmamap[idx];
5236	error = bus_dmamap_load_mbuf_sg(sc->bge_cdata.bge_tx_mtag, map, m, segs,
5237	    &nsegs, BUS_DMA_NOWAIT);
5238	if (error == EFBIG) {
5239		m = m_collapse(m, M_NOWAIT, BGE_NSEG_NEW);
5240		if (m == NULL) {
5241			m_freem(*m_head);
5242			*m_head = NULL;
5243			return (ENOBUFS);
5244		}
5245		*m_head = m;
5246		error = bus_dmamap_load_mbuf_sg(sc->bge_cdata.bge_tx_mtag, map,
5247		    m, segs, &nsegs, BUS_DMA_NOWAIT);
5248		if (error) {
5249			m_freem(m);
5250			*m_head = NULL;
5251			return (error);
5252		}
5253	} else if (error != 0)
5254		return (error);
5255
5256	/* Check if we have enough free send BDs. */
5257	if (sc->bge_txcnt + nsegs >= BGE_TX_RING_CNT) {
5258		bus_dmamap_unload(sc->bge_cdata.bge_tx_mtag, map);
5259		return (ENOBUFS);
5260	}
5261
5262	bus_dmamap_sync(sc->bge_cdata.bge_tx_mtag, map, BUS_DMASYNC_PREWRITE);
5263
5264	if (m->m_flags & M_VLANTAG) {
5265		csum_flags |= BGE_TXBDFLAG_VLAN_TAG;
5266		vlan_tag = m->m_pkthdr.ether_vtag;
5267	}
5268
5269	if (sc->bge_asicrev == BGE_ASICREV_BCM5762 &&
5270	    (m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
5271		/*
5272		 * 5725 family of devices corrupts TSO packets when TSO DMA
5273		 * buffers cross into regions which are within MSS bytes of
5274		 * a 4GB boundary.  If we encounter the condition, drop the
5275		 * packet.
5276		 */
5277		for (i = 0; ; i++) {
5278			d = &sc->bge_ldata.bge_tx_ring[idx];
5279			d->bge_addr.bge_addr_lo = BGE_ADDR_LO(segs[i].ds_addr);
5280			d->bge_addr.bge_addr_hi = BGE_ADDR_HI(segs[i].ds_addr);
5281			d->bge_len = segs[i].ds_len;
5282			if (d->bge_addr.bge_addr_lo + segs[i].ds_len + mss <
5283			    d->bge_addr.bge_addr_lo)
5284				break;
5285			d->bge_flags = csum_flags;
5286			d->bge_vlan_tag = vlan_tag;
5287			d->bge_mss = mss;
5288			if (i == nsegs - 1)
5289				break;
5290			BGE_INC(idx, BGE_TX_RING_CNT);
5291		}
5292		if (i != nsegs - 1) {
5293			bus_dmamap_sync(sc->bge_cdata.bge_tx_mtag, map,
5294			    BUS_DMASYNC_POSTWRITE);
5295			bus_dmamap_unload(sc->bge_cdata.bge_tx_mtag, map);
5296			m_freem(*m_head);
5297			*m_head = NULL;
5298			return (EIO);
5299		}
5300	} else {
5301		for (i = 0; ; i++) {
5302			d = &sc->bge_ldata.bge_tx_ring[idx];
5303			d->bge_addr.bge_addr_lo = BGE_ADDR_LO(segs[i].ds_addr);
5304			d->bge_addr.bge_addr_hi = BGE_ADDR_HI(segs[i].ds_addr);
5305			d->bge_len = segs[i].ds_len;
5306			d->bge_flags = csum_flags;
5307			d->bge_vlan_tag = vlan_tag;
5308			d->bge_mss = mss;
5309			if (i == nsegs - 1)
5310				break;
5311			BGE_INC(idx, BGE_TX_RING_CNT);
5312		}
5313	}
5314
5315	/* Mark the last segment as end of packet... */
5316	d->bge_flags |= BGE_TXBDFLAG_END;
5317
5318	/*
5319	 * Insure that the map for this transmission
5320	 * is placed at the array index of the last descriptor
5321	 * in this chain.
5322	 */
5323	sc->bge_cdata.bge_tx_dmamap[*txidx] = sc->bge_cdata.bge_tx_dmamap[idx];
5324	sc->bge_cdata.bge_tx_dmamap[idx] = map;
5325	sc->bge_cdata.bge_tx_chain[idx] = m;
5326	sc->bge_txcnt += nsegs;
5327
5328	BGE_INC(idx, BGE_TX_RING_CNT);
5329	*txidx = idx;
5330
5331	return (0);
5332}
5333
5334/*
5335 * Main transmit routine. To avoid having to do mbuf copies, we put pointers
5336 * to the mbuf data regions directly in the transmit descriptors.
5337 */
5338static void
5339bge_start_locked(struct ifnet *ifp)
5340{
5341	struct bge_softc *sc;
5342	struct mbuf *m_head;
5343	uint32_t prodidx;
5344	int count;
5345
5346	sc = ifp->if_softc;
5347	BGE_LOCK_ASSERT(sc);
5348
5349	if (!sc->bge_link ||
5350	    (ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
5351	    IFF_DRV_RUNNING)
5352		return;
5353
5354	prodidx = sc->bge_tx_prodidx;
5355
5356	for (count = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd);) {
5357		if (sc->bge_txcnt > BGE_TX_RING_CNT - 16) {
5358			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
5359			break;
5360		}
5361		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
5362		if (m_head == NULL)
5363			break;
5364
5365		/*
5366		 * Pack the data into the transmit ring. If we
5367		 * don't have room, set the OACTIVE flag and wait
5368		 * for the NIC to drain the ring.
5369		 */
5370		if (bge_encap(sc, &m_head, &prodidx)) {
5371			if (m_head == NULL)
5372				break;
5373			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
5374			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
5375			break;
5376		}
5377		++count;
5378
5379		/*
5380		 * If there's a BPF listener, bounce a copy of this frame
5381		 * to him.
5382		 */
5383#ifdef ETHER_BPF_MTAP
5384		ETHER_BPF_MTAP(ifp, m_head);
5385#else
5386		BPF_MTAP(ifp, m_head);
5387#endif
5388	}
5389
5390	if (count > 0) {
5391		bus_dmamap_sync(sc->bge_cdata.bge_tx_ring_tag,
5392		    sc->bge_cdata.bge_tx_ring_map, BUS_DMASYNC_PREWRITE);
5393		/* Transmit. */
5394		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
5395		/* 5700 b2 errata */
5396		if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
5397			bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
5398
5399		sc->bge_tx_prodidx = prodidx;
5400
5401		/*
5402		 * Set a timeout in case the chip goes out to lunch.
5403		 */
5404		sc->bge_timer = BGE_TX_TIMEOUT;
5405	}
5406}
5407
5408/*
5409 * Main transmit routine. To avoid having to do mbuf copies, we put pointers
5410 * to the mbuf data regions directly in the transmit descriptors.
5411 */
5412static void
5413bge_start(struct ifnet *ifp)
5414{
5415	struct bge_softc *sc;
5416
5417	sc = ifp->if_softc;
5418	BGE_LOCK(sc);
5419	bge_start_locked(ifp);
5420	BGE_UNLOCK(sc);
5421}
5422
5423static void
5424bge_init_locked(struct bge_softc *sc)
5425{
5426	struct ifnet *ifp;
5427	uint16_t *m;
5428	uint32_t mode;
5429
5430	BGE_LOCK_ASSERT(sc);
5431
5432	ifp = sc->bge_ifp;
5433
5434	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
5435		return;
5436
5437	/* Cancel pending I/O and flush buffers. */
5438	bge_stop(sc);
5439
5440	bge_stop_fw(sc);
5441	bge_sig_pre_reset(sc, BGE_RESET_START);
5442	bge_reset(sc);
5443	bge_sig_legacy(sc, BGE_RESET_START);
5444	bge_sig_post_reset(sc, BGE_RESET_START);
5445
5446	bge_chipinit(sc);
5447
5448	/*
5449	 * Init the various state machines, ring
5450	 * control blocks and firmware.
5451	 */
5452	if (bge_blockinit(sc)) {
5453		device_printf(sc->bge_dev, "initialization failure\n");
5454		return;
5455	}
5456
5457	ifp = sc->bge_ifp;
5458
5459	/* Specify MTU. */
5460	CSR_WRITE_4(sc, BGE_RX_MTU, ifp->if_mtu +
5461	    ETHER_HDR_LEN + ETHER_CRC_LEN +
5462	    (ifp->if_capenable & IFCAP_VLAN_MTU ? ETHER_VLAN_ENCAP_LEN : 0));
5463
5464	/* Load our MAC address. */
5465	m = (uint16_t *)IF_LLADDR(sc->bge_ifp);
5466	CSR_WRITE_4(sc, BGE_MAC_ADDR1_LO, htons(m[0]));
5467	CSR_WRITE_4(sc, BGE_MAC_ADDR1_HI, (htons(m[1]) << 16) | htons(m[2]));
5468
5469	/* Program promiscuous mode. */
5470	bge_setpromisc(sc);
5471
5472	/* Program multicast filter. */
5473	bge_setmulti(sc);
5474
5475	/* Program VLAN tag stripping. */
5476	bge_setvlan(sc);
5477
5478	/* Override UDP checksum offloading. */
5479	if (sc->bge_forced_udpcsum == 0)
5480		sc->bge_csum_features &= ~CSUM_UDP;
5481	else
5482		sc->bge_csum_features |= CSUM_UDP;
5483	if (ifp->if_capabilities & IFCAP_TXCSUM &&
5484	    ifp->if_capenable & IFCAP_TXCSUM) {
5485		ifp->if_hwassist &= ~(BGE_CSUM_FEATURES | CSUM_UDP);
5486		ifp->if_hwassist |= sc->bge_csum_features;
5487	}
5488
5489	/* Init RX ring. */
5490	if (bge_init_rx_ring_std(sc) != 0) {
5491		device_printf(sc->bge_dev, "no memory for std Rx buffers.\n");
5492		bge_stop(sc);
5493		return;
5494	}
5495
5496	/*
5497	 * Workaround for a bug in 5705 ASIC rev A0. Poll the NIC's
5498	 * memory to insure that the chip has in fact read the first
5499	 * entry of the ring.
5500	 */
5501	if (sc->bge_chipid == BGE_CHIPID_BCM5705_A0) {
5502		uint32_t		v, i;
5503		for (i = 0; i < 10; i++) {
5504			DELAY(20);
5505			v = bge_readmem_ind(sc, BGE_STD_RX_RINGS + 8);
5506			if (v == (MCLBYTES - ETHER_ALIGN))
5507				break;
5508		}
5509		if (i == 10)
5510			device_printf (sc->bge_dev,
5511			    "5705 A0 chip failed to load RX ring\n");
5512	}
5513
5514	/* Init jumbo RX ring. */
5515	if (BGE_IS_JUMBO_CAPABLE(sc) &&
5516	    ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN >
5517	    (MCLBYTES - ETHER_ALIGN)) {
5518		if (bge_init_rx_ring_jumbo(sc) != 0) {
5519			device_printf(sc->bge_dev,
5520			    "no memory for jumbo Rx buffers.\n");
5521			bge_stop(sc);
5522			return;
5523		}
5524	}
5525
5526	/* Init our RX return ring index. */
5527	sc->bge_rx_saved_considx = 0;
5528
5529	/* Init our RX/TX stat counters. */
5530	sc->bge_rx_discards = sc->bge_tx_discards = sc->bge_tx_collisions = 0;
5531
5532	/* Init TX ring. */
5533	bge_init_tx_ring(sc);
5534
5535	/* Enable TX MAC state machine lockup fix. */
5536	mode = CSR_READ_4(sc, BGE_TX_MODE);
5537	if (BGE_IS_5755_PLUS(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5906)
5538		mode |= BGE_TXMODE_MBUF_LOCKUP_FIX;
5539	if (sc->bge_asicrev == BGE_ASICREV_BCM5720 ||
5540	    sc->bge_asicrev == BGE_ASICREV_BCM5762) {
5541		mode &= ~(BGE_TXMODE_JMB_FRM_LEN | BGE_TXMODE_CNT_DN_MODE);
5542		mode |= CSR_READ_4(sc, BGE_TX_MODE) &
5543		    (BGE_TXMODE_JMB_FRM_LEN | BGE_TXMODE_CNT_DN_MODE);
5544	}
5545	/* Turn on transmitter. */
5546	CSR_WRITE_4(sc, BGE_TX_MODE, mode | BGE_TXMODE_ENABLE);
5547	DELAY(100);
5548
5549	/* Turn on receiver. */
5550	mode = CSR_READ_4(sc, BGE_RX_MODE);
5551	if (BGE_IS_5755_PLUS(sc))
5552		mode |= BGE_RXMODE_IPV6_ENABLE;
5553	if (sc->bge_asicrev == BGE_ASICREV_BCM5762)
5554		mode |= BGE_RXMODE_IPV4_FRAG_FIX;
5555	CSR_WRITE_4(sc,BGE_RX_MODE, mode | BGE_RXMODE_ENABLE);
5556	DELAY(10);
5557
5558	/*
5559	 * Set the number of good frames to receive after RX MBUF
5560	 * Low Watermark has been reached. After the RX MAC receives
5561	 * this number of frames, it will drop subsequent incoming
5562	 * frames until the MBUF High Watermark is reached.
5563	 */
5564	if (BGE_IS_57765_PLUS(sc))
5565		CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 1);
5566	else
5567		CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 2);
5568
5569	/* Clear MAC statistics. */
5570	if (BGE_IS_5705_PLUS(sc))
5571		bge_stats_clear_regs(sc);
5572
5573	/* Tell firmware we're alive. */
5574	BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
5575
5576#ifdef DEVICE_POLLING
5577	/* Disable interrupts if we are polling. */
5578	if (ifp->if_capenable & IFCAP_POLLING) {
5579		BGE_SETBIT(sc, BGE_PCI_MISC_CTL,
5580		    BGE_PCIMISCCTL_MASK_PCI_INTR);
5581		bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
5582	} else
5583#endif
5584
5585	/* Enable host interrupts. */
5586	{
5587	BGE_SETBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_CLEAR_INTA);
5588	BGE_CLRBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR);
5589	bge_writembx(sc, BGE_MBX_IRQ0_LO, 0);
5590	}
5591
5592	ifp->if_drv_flags |= IFF_DRV_RUNNING;
5593	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
5594
5595	bge_ifmedia_upd_locked(ifp);
5596
5597	callout_reset(&sc->bge_stat_ch, hz, bge_tick, sc);
5598}
5599
5600static void
5601bge_init(void *xsc)
5602{
5603	struct bge_softc *sc = xsc;
5604
5605	BGE_LOCK(sc);
5606	bge_init_locked(sc);
5607	BGE_UNLOCK(sc);
5608}
5609
5610/*
5611 * Set media options.
5612 */
5613static int
5614bge_ifmedia_upd(struct ifnet *ifp)
5615{
5616	struct bge_softc *sc = ifp->if_softc;
5617	int res;
5618
5619	BGE_LOCK(sc);
5620	res = bge_ifmedia_upd_locked(ifp);
5621	BGE_UNLOCK(sc);
5622
5623	return (res);
5624}
5625
5626static int
5627bge_ifmedia_upd_locked(struct ifnet *ifp)
5628{
5629	struct bge_softc *sc = ifp->if_softc;
5630	struct mii_data *mii;
5631	struct mii_softc *miisc;
5632	struct ifmedia *ifm;
5633
5634	BGE_LOCK_ASSERT(sc);
5635
5636	ifm = &sc->bge_ifmedia;
5637
5638	/* If this is a 1000baseX NIC, enable the TBI port. */
5639	if (sc->bge_flags & BGE_FLAG_TBI) {
5640		if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
5641			return (EINVAL);
5642		switch(IFM_SUBTYPE(ifm->ifm_media)) {
5643		case IFM_AUTO:
5644			/*
5645			 * The BCM5704 ASIC appears to have a special
5646			 * mechanism for programming the autoneg
5647			 * advertisement registers in TBI mode.
5648			 */
5649			if (sc->bge_asicrev == BGE_ASICREV_BCM5704) {
5650				uint32_t sgdig;
5651				sgdig = CSR_READ_4(sc, BGE_SGDIG_STS);
5652				if (sgdig & BGE_SGDIGSTS_DONE) {
5653					CSR_WRITE_4(sc, BGE_TX_TBI_AUTONEG, 0);
5654					sgdig = CSR_READ_4(sc, BGE_SGDIG_CFG);
5655					sgdig |= BGE_SGDIGCFG_AUTO |
5656					    BGE_SGDIGCFG_PAUSE_CAP |
5657					    BGE_SGDIGCFG_ASYM_PAUSE;
5658					CSR_WRITE_4(sc, BGE_SGDIG_CFG,
5659					    sgdig | BGE_SGDIGCFG_SEND);
5660					DELAY(5);
5661					CSR_WRITE_4(sc, BGE_SGDIG_CFG, sgdig);
5662				}
5663			}
5664			break;
5665		case IFM_1000_SX:
5666			if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX) {
5667				BGE_CLRBIT(sc, BGE_MAC_MODE,
5668				    BGE_MACMODE_HALF_DUPLEX);
5669			} else {
5670				BGE_SETBIT(sc, BGE_MAC_MODE,
5671				    BGE_MACMODE_HALF_DUPLEX);
5672			}
5673			DELAY(40);
5674			break;
5675		default:
5676			return (EINVAL);
5677		}
5678		return (0);
5679	}
5680
5681	sc->bge_link_evt++;
5682	mii = device_get_softc(sc->bge_miibus);
5683	LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
5684		PHY_RESET(miisc);
5685	mii_mediachg(mii);
5686
5687	/*
5688	 * Force an interrupt so that we will call bge_link_upd
5689	 * if needed and clear any pending link state attention.
5690	 * Without this we are not getting any further interrupts
5691	 * for link state changes and thus will not UP the link and
5692	 * not be able to send in bge_start_locked. The only
5693	 * way to get things working was to receive a packet and
5694	 * get an RX intr.
5695	 * bge_tick should help for fiber cards and we might not
5696	 * need to do this here if BGE_FLAG_TBI is set but as
5697	 * we poll for fiber anyway it should not harm.
5698	 */
5699	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 ||
5700	    sc->bge_flags & BGE_FLAG_5788)
5701		BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
5702	else
5703		BGE_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
5704
5705	return (0);
5706}
5707
5708/*
5709 * Report current media status.
5710 */
5711static void
5712bge_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
5713{
5714	struct bge_softc *sc = ifp->if_softc;
5715	struct mii_data *mii;
5716
5717	BGE_LOCK(sc);
5718
5719	if ((ifp->if_flags & IFF_UP) == 0) {
5720		BGE_UNLOCK(sc);
5721		return;
5722	}
5723	if (sc->bge_flags & BGE_FLAG_TBI) {
5724		ifmr->ifm_status = IFM_AVALID;
5725		ifmr->ifm_active = IFM_ETHER;
5726		if (CSR_READ_4(sc, BGE_MAC_STS) &
5727		    BGE_MACSTAT_TBI_PCS_SYNCHED)
5728			ifmr->ifm_status |= IFM_ACTIVE;
5729		else {
5730			ifmr->ifm_active |= IFM_NONE;
5731			BGE_UNLOCK(sc);
5732			return;
5733		}
5734		ifmr->ifm_active |= IFM_1000_SX;
5735		if (CSR_READ_4(sc, BGE_MAC_MODE) & BGE_MACMODE_HALF_DUPLEX)
5736			ifmr->ifm_active |= IFM_HDX;
5737		else
5738			ifmr->ifm_active |= IFM_FDX;
5739		BGE_UNLOCK(sc);
5740		return;
5741	}
5742
5743	mii = device_get_softc(sc->bge_miibus);
5744	mii_pollstat(mii);
5745	ifmr->ifm_active = mii->mii_media_active;
5746	ifmr->ifm_status = mii->mii_media_status;
5747
5748	BGE_UNLOCK(sc);
5749}
5750
5751static int
5752bge_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
5753{
5754	struct bge_softc *sc = ifp->if_softc;
5755	struct ifreq *ifr = (struct ifreq *) data;
5756	struct mii_data *mii;
5757	int flags, mask, error = 0;
5758
5759	switch (command) {
5760	case SIOCSIFMTU:
5761		if (BGE_IS_JUMBO_CAPABLE(sc) ||
5762		    (sc->bge_flags & BGE_FLAG_JUMBO_STD)) {
5763			if (ifr->ifr_mtu < ETHERMIN ||
5764			    ifr->ifr_mtu > BGE_JUMBO_MTU) {
5765				error = EINVAL;
5766				break;
5767			}
5768		} else if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ETHERMTU) {
5769			error = EINVAL;
5770			break;
5771		}
5772		BGE_LOCK(sc);
5773		if (ifp->if_mtu != ifr->ifr_mtu) {
5774			ifp->if_mtu = ifr->ifr_mtu;
5775			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
5776				ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
5777				bge_init_locked(sc);
5778			}
5779		}
5780		BGE_UNLOCK(sc);
5781		break;
5782	case SIOCSIFFLAGS:
5783		BGE_LOCK(sc);
5784		if (ifp->if_flags & IFF_UP) {
5785			/*
5786			 * If only the state of the PROMISC flag changed,
5787			 * then just use the 'set promisc mode' command
5788			 * instead of reinitializing the entire NIC. Doing
5789			 * a full re-init means reloading the firmware and
5790			 * waiting for it to start up, which may take a
5791			 * second or two.  Similarly for ALLMULTI.
5792			 */
5793			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
5794				flags = ifp->if_flags ^ sc->bge_if_flags;
5795				if (flags & IFF_PROMISC)
5796					bge_setpromisc(sc);
5797				if (flags & IFF_ALLMULTI)
5798					bge_setmulti(sc);
5799			} else
5800				bge_init_locked(sc);
5801		} else {
5802			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
5803				bge_stop(sc);
5804			}
5805		}
5806		sc->bge_if_flags = ifp->if_flags;
5807		BGE_UNLOCK(sc);
5808		error = 0;
5809		break;
5810	case SIOCADDMULTI:
5811	case SIOCDELMULTI:
5812		if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
5813			BGE_LOCK(sc);
5814			bge_setmulti(sc);
5815			BGE_UNLOCK(sc);
5816			error = 0;
5817		}
5818		break;
5819	case SIOCSIFMEDIA:
5820	case SIOCGIFMEDIA:
5821		if (sc->bge_flags & BGE_FLAG_TBI) {
5822			error = ifmedia_ioctl(ifp, ifr,
5823			    &sc->bge_ifmedia, command);
5824		} else {
5825			mii = device_get_softc(sc->bge_miibus);
5826			error = ifmedia_ioctl(ifp, ifr,
5827			    &mii->mii_media, command);
5828		}
5829		break;
5830	case SIOCSIFCAP:
5831		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
5832#ifdef DEVICE_POLLING
5833		if (mask & IFCAP_POLLING) {
5834			if (ifr->ifr_reqcap & IFCAP_POLLING) {
5835				error = ether_poll_register(bge_poll, ifp);
5836				if (error)
5837					return (error);
5838				BGE_LOCK(sc);
5839				BGE_SETBIT(sc, BGE_PCI_MISC_CTL,
5840				    BGE_PCIMISCCTL_MASK_PCI_INTR);
5841				bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
5842				ifp->if_capenable |= IFCAP_POLLING;
5843				BGE_UNLOCK(sc);
5844			} else {
5845				error = ether_poll_deregister(ifp);
5846				/* Enable interrupt even in error case */
5847				BGE_LOCK(sc);
5848				BGE_CLRBIT(sc, BGE_PCI_MISC_CTL,
5849				    BGE_PCIMISCCTL_MASK_PCI_INTR);
5850				bge_writembx(sc, BGE_MBX_IRQ0_LO, 0);
5851				ifp->if_capenable &= ~IFCAP_POLLING;
5852				BGE_UNLOCK(sc);
5853			}
5854		}
5855#endif
5856		if ((mask & IFCAP_TXCSUM) != 0 &&
5857		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
5858			ifp->if_capenable ^= IFCAP_TXCSUM;
5859			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
5860				ifp->if_hwassist |= sc->bge_csum_features;
5861			else
5862				ifp->if_hwassist &= ~sc->bge_csum_features;
5863		}
5864
5865		if ((mask & IFCAP_RXCSUM) != 0 &&
5866		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0)
5867			ifp->if_capenable ^= IFCAP_RXCSUM;
5868
5869		if ((mask & IFCAP_TSO4) != 0 &&
5870		    (ifp->if_capabilities & IFCAP_TSO4) != 0) {
5871			ifp->if_capenable ^= IFCAP_TSO4;
5872			if ((ifp->if_capenable & IFCAP_TSO4) != 0)
5873				ifp->if_hwassist |= CSUM_TSO;
5874			else
5875				ifp->if_hwassist &= ~CSUM_TSO;
5876		}
5877
5878		if (mask & IFCAP_VLAN_MTU) {
5879			ifp->if_capenable ^= IFCAP_VLAN_MTU;
5880			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
5881			bge_init(sc);
5882		}
5883
5884		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
5885		    (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0)
5886			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
5887		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
5888		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
5889			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
5890			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
5891				ifp->if_capenable &= ~IFCAP_VLAN_HWTSO;
5892			BGE_LOCK(sc);
5893			bge_setvlan(sc);
5894			BGE_UNLOCK(sc);
5895		}
5896#ifdef VLAN_CAPABILITIES
5897		VLAN_CAPABILITIES(ifp);
5898#endif
5899		break;
5900	default:
5901		error = ether_ioctl(ifp, command, data);
5902		break;
5903	}
5904
5905	return (error);
5906}
5907
5908static void
5909bge_watchdog(struct bge_softc *sc)
5910{
5911	struct ifnet *ifp;
5912	uint32_t status;
5913
5914	BGE_LOCK_ASSERT(sc);
5915
5916	if (sc->bge_timer == 0 || --sc->bge_timer)
5917		return;
5918
5919	/* If pause frames are active then don't reset the hardware. */
5920	if ((CSR_READ_4(sc, BGE_RX_MODE) & BGE_RXMODE_FLOWCTL_ENABLE) != 0) {
5921		status = CSR_READ_4(sc, BGE_RX_STS);
5922		if ((status & BGE_RXSTAT_REMOTE_XOFFED) != 0) {
5923			/*
5924			 * If link partner has us in XOFF state then wait for
5925			 * the condition to clear.
5926			 */
5927			CSR_WRITE_4(sc, BGE_RX_STS, status);
5928			sc->bge_timer = BGE_TX_TIMEOUT;
5929			return;
5930		} else if ((status & BGE_RXSTAT_RCVD_XOFF) != 0 &&
5931		    (status & BGE_RXSTAT_RCVD_XON) != 0) {
5932			/*
5933			 * If link partner has us in XOFF state then wait for
5934			 * the condition to clear.
5935			 */
5936			CSR_WRITE_4(sc, BGE_RX_STS, status);
5937			sc->bge_timer = BGE_TX_TIMEOUT;
5938			return;
5939		}
5940		/*
5941		 * Any other condition is unexpected and the controller
5942		 * should be reset.
5943		 */
5944	}
5945
5946	ifp = sc->bge_ifp;
5947
5948	if_printf(ifp, "watchdog timeout -- resetting\n");
5949
5950	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
5951	bge_init_locked(sc);
5952
5953	ifp->if_oerrors++;
5954}
5955
5956static void
5957bge_stop_block(struct bge_softc *sc, bus_size_t reg, uint32_t bit)
5958{
5959	int i;
5960
5961	BGE_CLRBIT(sc, reg, bit);
5962
5963	for (i = 0; i < BGE_TIMEOUT; i++) {
5964		if ((CSR_READ_4(sc, reg) & bit) == 0)
5965			return;
5966		DELAY(100);
5967        }
5968}
5969
5970/*
5971 * Stop the adapter and free any mbufs allocated to the
5972 * RX and TX lists.
5973 */
5974static void
5975bge_stop(struct bge_softc *sc)
5976{
5977	struct ifnet *ifp;
5978
5979	BGE_LOCK_ASSERT(sc);
5980
5981	ifp = sc->bge_ifp;
5982
5983	callout_stop(&sc->bge_stat_ch);
5984
5985	/* Disable host interrupts. */
5986	BGE_SETBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR);
5987	bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
5988
5989	/*
5990	 * Tell firmware we're shutting down.
5991	 */
5992	bge_stop_fw(sc);
5993	bge_sig_pre_reset(sc, BGE_RESET_SHUTDOWN);
5994
5995	/*
5996	 * Disable all of the receiver blocks.
5997	 */
5998	bge_stop_block(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
5999	bge_stop_block(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
6000	bge_stop_block(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
6001	if (BGE_IS_5700_FAMILY(sc))
6002		bge_stop_block(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
6003	bge_stop_block(sc, BGE_RDBDI_MODE, BGE_RBDIMODE_ENABLE);
6004	bge_stop_block(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
6005	bge_stop_block(sc, BGE_RBDC_MODE, BGE_RBDCMODE_ENABLE);
6006
6007	/*
6008	 * Disable all of the transmit blocks.
6009	 */
6010	bge_stop_block(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
6011	bge_stop_block(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
6012	bge_stop_block(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
6013	bge_stop_block(sc, BGE_RDMA_MODE, BGE_RDMAMODE_ENABLE);
6014	bge_stop_block(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE);
6015	if (BGE_IS_5700_FAMILY(sc))
6016		bge_stop_block(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
6017	bge_stop_block(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
6018
6019	/*
6020	 * Shut down all of the memory managers and related
6021	 * state machines.
6022	 */
6023	bge_stop_block(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE);
6024	bge_stop_block(sc, BGE_WDMA_MODE, BGE_WDMAMODE_ENABLE);
6025	if (BGE_IS_5700_FAMILY(sc))
6026		bge_stop_block(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
6027
6028	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
6029	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
6030	if (!(BGE_IS_5705_PLUS(sc))) {
6031		BGE_CLRBIT(sc, BGE_BMAN_MODE, BGE_BMANMODE_ENABLE);
6032		BGE_CLRBIT(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
6033	}
6034	/* Update MAC statistics. */
6035	if (BGE_IS_5705_PLUS(sc))
6036		bge_stats_update_regs(sc);
6037
6038	bge_reset(sc);
6039	bge_sig_legacy(sc, BGE_RESET_SHUTDOWN);
6040	bge_sig_post_reset(sc, BGE_RESET_SHUTDOWN);
6041
6042	/*
6043	 * Keep the ASF firmware running if up.
6044	 */
6045	if (sc->bge_asf_mode & ASF_STACKUP)
6046		BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
6047	else
6048		BGE_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
6049
6050	/* Free the RX lists. */
6051	bge_free_rx_ring_std(sc);
6052
6053	/* Free jumbo RX list. */
6054	if (BGE_IS_JUMBO_CAPABLE(sc))
6055		bge_free_rx_ring_jumbo(sc);
6056
6057	/* Free TX buffers. */
6058	bge_free_tx_ring(sc);
6059
6060	sc->bge_tx_saved_considx = BGE_TXCONS_UNSET;
6061
6062	/* Clear MAC's link state (PHY may still have link UP). */
6063	if (bootverbose && sc->bge_link)
6064		if_printf(sc->bge_ifp, "link DOWN\n");
6065	sc->bge_link = 0;
6066
6067	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
6068}
6069
6070/*
6071 * Stop all chip I/O so that the kernel's probe routines don't
6072 * get confused by errant DMAs when rebooting.
6073 */
6074static int
6075bge_shutdown(device_t dev)
6076{
6077	struct bge_softc *sc;
6078
6079	sc = device_get_softc(dev);
6080	BGE_LOCK(sc);
6081	bge_stop(sc);
6082	BGE_UNLOCK(sc);
6083
6084	return (0);
6085}
6086
6087static int
6088bge_suspend(device_t dev)
6089{
6090	struct bge_softc *sc;
6091
6092	sc = device_get_softc(dev);
6093	BGE_LOCK(sc);
6094	bge_stop(sc);
6095	BGE_UNLOCK(sc);
6096
6097	return (0);
6098}
6099
6100static int
6101bge_resume(device_t dev)
6102{
6103	struct bge_softc *sc;
6104	struct ifnet *ifp;
6105
6106	sc = device_get_softc(dev);
6107	BGE_LOCK(sc);
6108	ifp = sc->bge_ifp;
6109	if (ifp->if_flags & IFF_UP) {
6110		bge_init_locked(sc);
6111		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
6112			bge_start_locked(ifp);
6113	}
6114	BGE_UNLOCK(sc);
6115
6116	return (0);
6117}
6118
6119static void
6120bge_link_upd(struct bge_softc *sc)
6121{
6122	struct mii_data *mii;
6123	uint32_t link, status;
6124
6125	BGE_LOCK_ASSERT(sc);
6126
6127	/* Clear 'pending link event' flag. */
6128	sc->bge_link_evt = 0;
6129
6130	/*
6131	 * Process link state changes.
6132	 * Grrr. The link status word in the status block does
6133	 * not work correctly on the BCM5700 rev AX and BX chips,
6134	 * according to all available information. Hence, we have
6135	 * to enable MII interrupts in order to properly obtain
6136	 * async link changes. Unfortunately, this also means that
6137	 * we have to read the MAC status register to detect link
6138	 * changes, thereby adding an additional register access to
6139	 * the interrupt handler.
6140	 *
6141	 * XXX: perhaps link state detection procedure used for
6142	 * BGE_CHIPID_BCM5700_B2 can be used for others BCM5700 revisions.
6143	 */
6144
6145	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
6146	    sc->bge_chipid != BGE_CHIPID_BCM5700_B2) {
6147		status = CSR_READ_4(sc, BGE_MAC_STS);
6148		if (status & BGE_MACSTAT_MI_INTERRUPT) {
6149			mii = device_get_softc(sc->bge_miibus);
6150			mii_pollstat(mii);
6151			if (!sc->bge_link &&
6152			    mii->mii_media_status & IFM_ACTIVE &&
6153			    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
6154				sc->bge_link++;
6155				if (bootverbose)
6156					if_printf(sc->bge_ifp, "link UP\n");
6157			} else if (sc->bge_link &&
6158			    (!(mii->mii_media_status & IFM_ACTIVE) ||
6159			    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE)) {
6160				sc->bge_link = 0;
6161				if (bootverbose)
6162					if_printf(sc->bge_ifp, "link DOWN\n");
6163			}
6164
6165			/* Clear the interrupt. */
6166			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
6167			    BGE_EVTENB_MI_INTERRUPT);
6168			bge_miibus_readreg(sc->bge_dev, sc->bge_phy_addr,
6169			    BRGPHY_MII_ISR);
6170			bge_miibus_writereg(sc->bge_dev, sc->bge_phy_addr,
6171			    BRGPHY_MII_IMR, BRGPHY_INTRS);
6172		}
6173		return;
6174	}
6175
6176	if (sc->bge_flags & BGE_FLAG_TBI) {
6177		status = CSR_READ_4(sc, BGE_MAC_STS);
6178		if (status & BGE_MACSTAT_TBI_PCS_SYNCHED) {
6179			if (!sc->bge_link) {
6180				sc->bge_link++;
6181				if (sc->bge_asicrev == BGE_ASICREV_BCM5704) {
6182					BGE_CLRBIT(sc, BGE_MAC_MODE,
6183					    BGE_MACMODE_TBI_SEND_CFGS);
6184					DELAY(40);
6185				}
6186				CSR_WRITE_4(sc, BGE_MAC_STS, 0xFFFFFFFF);
6187				if (bootverbose)
6188					if_printf(sc->bge_ifp, "link UP\n");
6189				if_link_state_change(sc->bge_ifp,
6190				    LINK_STATE_UP);
6191			}
6192		} else if (sc->bge_link) {
6193			sc->bge_link = 0;
6194			if (bootverbose)
6195				if_printf(sc->bge_ifp, "link DOWN\n");
6196			if_link_state_change(sc->bge_ifp, LINK_STATE_DOWN);
6197		}
6198	} else if ((sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) != 0) {
6199		/*
6200		 * Some broken BCM chips have BGE_STATFLAG_LINKSTATE_CHANGED bit
6201		 * in status word always set. Workaround this bug by reading
6202		 * PHY link status directly.
6203		 */
6204		link = (CSR_READ_4(sc, BGE_MI_STS) & BGE_MISTS_LINK) ? 1 : 0;
6205
6206		if (link != sc->bge_link ||
6207		    sc->bge_asicrev == BGE_ASICREV_BCM5700) {
6208			mii = device_get_softc(sc->bge_miibus);
6209			mii_pollstat(mii);
6210			if (!sc->bge_link &&
6211			    mii->mii_media_status & IFM_ACTIVE &&
6212			    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
6213				sc->bge_link++;
6214				if (bootverbose)
6215					if_printf(sc->bge_ifp, "link UP\n");
6216			} else if (sc->bge_link &&
6217			    (!(mii->mii_media_status & IFM_ACTIVE) ||
6218			    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE)) {
6219				sc->bge_link = 0;
6220				if (bootverbose)
6221					if_printf(sc->bge_ifp, "link DOWN\n");
6222			}
6223		}
6224	} else {
6225		/*
6226		 * For controllers that call mii_tick, we have to poll
6227		 * link status.
6228		 */
6229		mii = device_get_softc(sc->bge_miibus);
6230		mii_pollstat(mii);
6231		bge_miibus_statchg(sc->bge_dev);
6232	}
6233
6234	/* Disable MAC attention when link is up. */
6235	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
6236	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
6237	    BGE_MACSTAT_LINK_CHANGED);
6238}
6239
6240static void
6241bge_add_sysctls(struct bge_softc *sc)
6242{
6243	struct sysctl_ctx_list *ctx;
6244	struct sysctl_oid_list *children;
6245	char tn[32];
6246	int unit;
6247
6248	ctx = device_get_sysctl_ctx(sc->bge_dev);
6249	children = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->bge_dev));
6250
6251#ifdef BGE_REGISTER_DEBUG
6252	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "debug_info",
6253	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, bge_sysctl_debug_info, "I",
6254	    "Debug Information");
6255
6256	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "reg_read",
6257	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, bge_sysctl_reg_read, "I",
6258	    "MAC Register Read");
6259
6260	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ape_read",
6261	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, bge_sysctl_ape_read, "I",
6262	    "APE Register Read");
6263
6264	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mem_read",
6265	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, bge_sysctl_mem_read, "I",
6266	    "Memory Read");
6267
6268#endif
6269
6270	unit = device_get_unit(sc->bge_dev);
6271	/*
6272	 * A common design characteristic for many Broadcom client controllers
6273	 * is that they only support a single outstanding DMA read operation
6274	 * on the PCIe bus. This means that it will take twice as long to fetch
6275	 * a TX frame that is split into header and payload buffers as it does
6276	 * to fetch a single, contiguous TX frame (2 reads vs. 1 read). For
6277	 * these controllers, coalescing buffers to reduce the number of memory
6278	 * reads is effective way to get maximum performance(about 940Mbps).
6279	 * Without collapsing TX buffers the maximum TCP bulk transfer
6280	 * performance is about 850Mbps. However forcing coalescing mbufs
6281	 * consumes a lot of CPU cycles, so leave it off by default.
6282	 */
6283	sc->bge_forced_collapse = 0;
6284	snprintf(tn, sizeof(tn), "dev.bge.%d.forced_collapse", unit);
6285	TUNABLE_INT_FETCH(tn, &sc->bge_forced_collapse);
6286	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "forced_collapse",
6287	    CTLFLAG_RW, &sc->bge_forced_collapse, 0,
6288	    "Number of fragmented TX buffers of a frame allowed before "
6289	    "forced collapsing");
6290
6291	sc->bge_msi = 1;
6292	snprintf(tn, sizeof(tn), "dev.bge.%d.msi", unit);
6293	TUNABLE_INT_FETCH(tn, &sc->bge_msi);
6294	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "msi",
6295	    CTLFLAG_RD, &sc->bge_msi, 0, "Enable MSI");
6296
6297	/*
6298	 * It seems all Broadcom controllers have a bug that can generate UDP
6299	 * datagrams with checksum value 0 when TX UDP checksum offloading is
6300	 * enabled.  Generating UDP checksum value 0 is RFC 768 violation.
6301	 * Even though the probability of generating such UDP datagrams is
6302	 * low, I don't want to see FreeBSD boxes to inject such datagrams
6303	 * into network so disable UDP checksum offloading by default.  Users
6304	 * still override this behavior by setting a sysctl variable,
6305	 * dev.bge.0.forced_udpcsum.
6306	 */
6307	sc->bge_forced_udpcsum = 0;
6308	snprintf(tn, sizeof(tn), "dev.bge.%d.bge_forced_udpcsum", unit);
6309	TUNABLE_INT_FETCH(tn, &sc->bge_forced_udpcsum);
6310	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "forced_udpcsum",
6311	    CTLFLAG_RW, &sc->bge_forced_udpcsum, 0,
6312	    "Enable UDP checksum offloading even if controller can "
6313	    "generate UDP checksum value 0");
6314
6315	if (BGE_IS_5705_PLUS(sc))
6316		bge_add_sysctl_stats_regs(sc, ctx, children);
6317	else
6318		bge_add_sysctl_stats(sc, ctx, children);
6319}
6320
6321#define BGE_SYSCTL_STAT(sc, ctx, desc, parent, node, oid) \
6322	SYSCTL_ADD_PROC(ctx, parent, OID_AUTO, oid, CTLTYPE_UINT|CTLFLAG_RD, \
6323	    sc, offsetof(struct bge_stats, node), bge_sysctl_stats, "IU", \
6324	    desc)
6325
6326static void
6327bge_add_sysctl_stats(struct bge_softc *sc, struct sysctl_ctx_list *ctx,
6328    struct sysctl_oid_list *parent)
6329{
6330	struct sysctl_oid *tree;
6331	struct sysctl_oid_list *children, *schildren;
6332
6333	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "stats", CTLFLAG_RD,
6334	    NULL, "BGE Statistics");
6335	schildren = children = SYSCTL_CHILDREN(tree);
6336	BGE_SYSCTL_STAT(sc, ctx, "Frames Dropped Due To Filters",
6337	    children, COSFramesDroppedDueToFilters,
6338	    "FramesDroppedDueToFilters");
6339	BGE_SYSCTL_STAT(sc, ctx, "NIC DMA Write Queue Full",
6340	    children, nicDmaWriteQueueFull, "DmaWriteQueueFull");
6341	BGE_SYSCTL_STAT(sc, ctx, "NIC DMA Write High Priority Queue Full",
6342	    children, nicDmaWriteHighPriQueueFull, "DmaWriteHighPriQueueFull");
6343	BGE_SYSCTL_STAT(sc, ctx, "NIC No More RX Buffer Descriptors",
6344	    children, nicNoMoreRxBDs, "NoMoreRxBDs");
6345	BGE_SYSCTL_STAT(sc, ctx, "Discarded Input Frames",
6346	    children, ifInDiscards, "InputDiscards");
6347	BGE_SYSCTL_STAT(sc, ctx, "Input Errors",
6348	    children, ifInErrors, "InputErrors");
6349	BGE_SYSCTL_STAT(sc, ctx, "NIC Recv Threshold Hit",
6350	    children, nicRecvThresholdHit, "RecvThresholdHit");
6351	BGE_SYSCTL_STAT(sc, ctx, "NIC DMA Read Queue Full",
6352	    children, nicDmaReadQueueFull, "DmaReadQueueFull");
6353	BGE_SYSCTL_STAT(sc, ctx, "NIC DMA Read High Priority Queue Full",
6354	    children, nicDmaReadHighPriQueueFull, "DmaReadHighPriQueueFull");
6355	BGE_SYSCTL_STAT(sc, ctx, "NIC Send Data Complete Queue Full",
6356	    children, nicSendDataCompQueueFull, "SendDataCompQueueFull");
6357	BGE_SYSCTL_STAT(sc, ctx, "NIC Ring Set Send Producer Index",
6358	    children, nicRingSetSendProdIndex, "RingSetSendProdIndex");
6359	BGE_SYSCTL_STAT(sc, ctx, "NIC Ring Status Update",
6360	    children, nicRingStatusUpdate, "RingStatusUpdate");
6361	BGE_SYSCTL_STAT(sc, ctx, "NIC Interrupts",
6362	    children, nicInterrupts, "Interrupts");
6363	BGE_SYSCTL_STAT(sc, ctx, "NIC Avoided Interrupts",
6364	    children, nicAvoidedInterrupts, "AvoidedInterrupts");
6365	BGE_SYSCTL_STAT(sc, ctx, "NIC Send Threshold Hit",
6366	    children, nicSendThresholdHit, "SendThresholdHit");
6367
6368	tree = SYSCTL_ADD_NODE(ctx, schildren, OID_AUTO, "rx", CTLFLAG_RD,
6369	    NULL, "BGE RX Statistics");
6370	children = SYSCTL_CHILDREN(tree);
6371	BGE_SYSCTL_STAT(sc, ctx, "Inbound Octets",
6372	    children, rxstats.ifHCInOctets, "ifHCInOctets");
6373	BGE_SYSCTL_STAT(sc, ctx, "Fragments",
6374	    children, rxstats.etherStatsFragments, "Fragments");
6375	BGE_SYSCTL_STAT(sc, ctx, "Inbound Unicast Packets",
6376	    children, rxstats.ifHCInUcastPkts, "UnicastPkts");
6377	BGE_SYSCTL_STAT(sc, ctx, "Inbound Multicast Packets",
6378	    children, rxstats.ifHCInMulticastPkts, "MulticastPkts");
6379	BGE_SYSCTL_STAT(sc, ctx, "FCS Errors",
6380	    children, rxstats.dot3StatsFCSErrors, "FCSErrors");
6381	BGE_SYSCTL_STAT(sc, ctx, "Alignment Errors",
6382	    children, rxstats.dot3StatsAlignmentErrors, "AlignmentErrors");
6383	BGE_SYSCTL_STAT(sc, ctx, "XON Pause Frames Received",
6384	    children, rxstats.xonPauseFramesReceived, "xonPauseFramesReceived");
6385	BGE_SYSCTL_STAT(sc, ctx, "XOFF Pause Frames Received",
6386	    children, rxstats.xoffPauseFramesReceived,
6387	    "xoffPauseFramesReceived");
6388	BGE_SYSCTL_STAT(sc, ctx, "MAC Control Frames Received",
6389	    children, rxstats.macControlFramesReceived,
6390	    "ControlFramesReceived");
6391	BGE_SYSCTL_STAT(sc, ctx, "XOFF State Entered",
6392	    children, rxstats.xoffStateEntered, "xoffStateEntered");
6393	BGE_SYSCTL_STAT(sc, ctx, "Frames Too Long",
6394	    children, rxstats.dot3StatsFramesTooLong, "FramesTooLong");
6395	BGE_SYSCTL_STAT(sc, ctx, "Jabbers",
6396	    children, rxstats.etherStatsJabbers, "Jabbers");
6397	BGE_SYSCTL_STAT(sc, ctx, "Undersized Packets",
6398	    children, rxstats.etherStatsUndersizePkts, "UndersizePkts");
6399	BGE_SYSCTL_STAT(sc, ctx, "Inbound Range Length Errors",
6400	    children, rxstats.inRangeLengthError, "inRangeLengthError");
6401	BGE_SYSCTL_STAT(sc, ctx, "Outbound Range Length Errors",
6402	    children, rxstats.outRangeLengthError, "outRangeLengthError");
6403
6404	tree = SYSCTL_ADD_NODE(ctx, schildren, OID_AUTO, "tx", CTLFLAG_RD,
6405	    NULL, "BGE TX Statistics");
6406	children = SYSCTL_CHILDREN(tree);
6407	BGE_SYSCTL_STAT(sc, ctx, "Outbound Octets",
6408	    children, txstats.ifHCOutOctets, "ifHCOutOctets");
6409	BGE_SYSCTL_STAT(sc, ctx, "TX Collisions",
6410	    children, txstats.etherStatsCollisions, "Collisions");
6411	BGE_SYSCTL_STAT(sc, ctx, "XON Sent",
6412	    children, txstats.outXonSent, "XonSent");
6413	BGE_SYSCTL_STAT(sc, ctx, "XOFF Sent",
6414	    children, txstats.outXoffSent, "XoffSent");
6415	BGE_SYSCTL_STAT(sc, ctx, "Flow Control Done",
6416	    children, txstats.flowControlDone, "flowControlDone");
6417	BGE_SYSCTL_STAT(sc, ctx, "Internal MAC TX errors",
6418	    children, txstats.dot3StatsInternalMacTransmitErrors,
6419	    "InternalMacTransmitErrors");
6420	BGE_SYSCTL_STAT(sc, ctx, "Single Collision Frames",
6421	    children, txstats.dot3StatsSingleCollisionFrames,
6422	    "SingleCollisionFrames");
6423	BGE_SYSCTL_STAT(sc, ctx, "Multiple Collision Frames",
6424	    children, txstats.dot3StatsMultipleCollisionFrames,
6425	    "MultipleCollisionFrames");
6426	BGE_SYSCTL_STAT(sc, ctx, "Deferred Transmissions",
6427	    children, txstats.dot3StatsDeferredTransmissions,
6428	    "DeferredTransmissions");
6429	BGE_SYSCTL_STAT(sc, ctx, "Excessive Collisions",
6430	    children, txstats.dot3StatsExcessiveCollisions,
6431	    "ExcessiveCollisions");
6432	BGE_SYSCTL_STAT(sc, ctx, "Late Collisions",
6433	    children, txstats.dot3StatsLateCollisions,
6434	    "LateCollisions");
6435	BGE_SYSCTL_STAT(sc, ctx, "Outbound Unicast Packets",
6436	    children, txstats.ifHCOutUcastPkts, "UnicastPkts");
6437	BGE_SYSCTL_STAT(sc, ctx, "Outbound Multicast Packets",
6438	    children, txstats.ifHCOutMulticastPkts, "MulticastPkts");
6439	BGE_SYSCTL_STAT(sc, ctx, "Outbound Broadcast Packets",
6440	    children, txstats.ifHCOutBroadcastPkts, "BroadcastPkts");
6441	BGE_SYSCTL_STAT(sc, ctx, "Carrier Sense Errors",
6442	    children, txstats.dot3StatsCarrierSenseErrors,
6443	    "CarrierSenseErrors");
6444	BGE_SYSCTL_STAT(sc, ctx, "Outbound Discards",
6445	    children, txstats.ifOutDiscards, "Discards");
6446	BGE_SYSCTL_STAT(sc, ctx, "Outbound Errors",
6447	    children, txstats.ifOutErrors, "Errors");
6448}
6449
6450#undef BGE_SYSCTL_STAT
6451
6452#define	BGE_SYSCTL_STAT_ADD64(c, h, n, p, d)	\
6453	    SYSCTL_ADD_UQUAD(c, h, OID_AUTO, n, CTLFLAG_RD, p, d)
6454
6455static void
6456bge_add_sysctl_stats_regs(struct bge_softc *sc, struct sysctl_ctx_list *ctx,
6457    struct sysctl_oid_list *parent)
6458{
6459	struct sysctl_oid *tree;
6460	struct sysctl_oid_list *child, *schild;
6461	struct bge_mac_stats *stats;
6462
6463	stats = &sc->bge_mac_stats;
6464	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "stats", CTLFLAG_RD,
6465	    NULL, "BGE Statistics");
6466	schild = child = SYSCTL_CHILDREN(tree);
6467	BGE_SYSCTL_STAT_ADD64(ctx, child, "FramesDroppedDueToFilters",
6468	    &stats->FramesDroppedDueToFilters, "Frames Dropped Due to Filters");
6469	BGE_SYSCTL_STAT_ADD64(ctx, child, "DmaWriteQueueFull",
6470	    &stats->DmaWriteQueueFull, "NIC DMA Write Queue Full");
6471	BGE_SYSCTL_STAT_ADD64(ctx, child, "DmaWriteHighPriQueueFull",
6472	    &stats->DmaWriteHighPriQueueFull,
6473	    "NIC DMA Write High Priority Queue Full");
6474	BGE_SYSCTL_STAT_ADD64(ctx, child, "NoMoreRxBDs",
6475	    &stats->NoMoreRxBDs, "NIC No More RX Buffer Descriptors");
6476	BGE_SYSCTL_STAT_ADD64(ctx, child, "InputDiscards",
6477	    &stats->InputDiscards, "Discarded Input Frames");
6478	BGE_SYSCTL_STAT_ADD64(ctx, child, "InputErrors",
6479	    &stats->InputErrors, "Input Errors");
6480	BGE_SYSCTL_STAT_ADD64(ctx, child, "RecvThresholdHit",
6481	    &stats->RecvThresholdHit, "NIC Recv Threshold Hit");
6482
6483	tree = SYSCTL_ADD_NODE(ctx, schild, OID_AUTO, "rx", CTLFLAG_RD,
6484	    NULL, "BGE RX Statistics");
6485	child = SYSCTL_CHILDREN(tree);
6486	BGE_SYSCTL_STAT_ADD64(ctx, child, "ifHCInOctets",
6487	    &stats->ifHCInOctets, "Inbound Octets");
6488	BGE_SYSCTL_STAT_ADD64(ctx, child, "Fragments",
6489	    &stats->etherStatsFragments, "Fragments");
6490	BGE_SYSCTL_STAT_ADD64(ctx, child, "UnicastPkts",
6491	    &stats->ifHCInUcastPkts, "Inbound Unicast Packets");
6492	BGE_SYSCTL_STAT_ADD64(ctx, child, "MulticastPkts",
6493	    &stats->ifHCInMulticastPkts, "Inbound Multicast Packets");
6494	BGE_SYSCTL_STAT_ADD64(ctx, child, "BroadcastPkts",
6495	    &stats->ifHCInBroadcastPkts, "Inbound Broadcast Packets");
6496	BGE_SYSCTL_STAT_ADD64(ctx, child, "FCSErrors",
6497	    &stats->dot3StatsFCSErrors, "FCS Errors");
6498	BGE_SYSCTL_STAT_ADD64(ctx, child, "AlignmentErrors",
6499	    &stats->dot3StatsAlignmentErrors, "Alignment Errors");
6500	BGE_SYSCTL_STAT_ADD64(ctx, child, "xonPauseFramesReceived",
6501	    &stats->xonPauseFramesReceived, "XON Pause Frames Received");
6502	BGE_SYSCTL_STAT_ADD64(ctx, child, "xoffPauseFramesReceived",
6503	    &stats->xoffPauseFramesReceived, "XOFF Pause Frames Received");
6504	BGE_SYSCTL_STAT_ADD64(ctx, child, "ControlFramesReceived",
6505	    &stats->macControlFramesReceived, "MAC Control Frames Received");
6506	BGE_SYSCTL_STAT_ADD64(ctx, child, "xoffStateEntered",
6507	    &stats->xoffStateEntered, "XOFF State Entered");
6508	BGE_SYSCTL_STAT_ADD64(ctx, child, "FramesTooLong",
6509	    &stats->dot3StatsFramesTooLong, "Frames Too Long");
6510	BGE_SYSCTL_STAT_ADD64(ctx, child, "Jabbers",
6511	    &stats->etherStatsJabbers, "Jabbers");
6512	BGE_SYSCTL_STAT_ADD64(ctx, child, "UndersizePkts",
6513	    &stats->etherStatsUndersizePkts, "Undersized Packets");
6514
6515	tree = SYSCTL_ADD_NODE(ctx, schild, OID_AUTO, "tx", CTLFLAG_RD,
6516	    NULL, "BGE TX Statistics");
6517	child = SYSCTL_CHILDREN(tree);
6518	BGE_SYSCTL_STAT_ADD64(ctx, child, "ifHCOutOctets",
6519	    &stats->ifHCOutOctets, "Outbound Octets");
6520	BGE_SYSCTL_STAT_ADD64(ctx, child, "Collisions",
6521	    &stats->etherStatsCollisions, "TX Collisions");
6522	BGE_SYSCTL_STAT_ADD64(ctx, child, "XonSent",
6523	    &stats->outXonSent, "XON Sent");
6524	BGE_SYSCTL_STAT_ADD64(ctx, child, "XoffSent",
6525	    &stats->outXoffSent, "XOFF Sent");
6526	BGE_SYSCTL_STAT_ADD64(ctx, child, "InternalMacTransmitErrors",
6527	    &stats->dot3StatsInternalMacTransmitErrors,
6528	    "Internal MAC TX Errors");
6529	BGE_SYSCTL_STAT_ADD64(ctx, child, "SingleCollisionFrames",
6530	    &stats->dot3StatsSingleCollisionFrames, "Single Collision Frames");
6531	BGE_SYSCTL_STAT_ADD64(ctx, child, "MultipleCollisionFrames",
6532	    &stats->dot3StatsMultipleCollisionFrames,
6533	    "Multiple Collision Frames");
6534	BGE_SYSCTL_STAT_ADD64(ctx, child, "DeferredTransmissions",
6535	    &stats->dot3StatsDeferredTransmissions, "Deferred Transmissions");
6536	BGE_SYSCTL_STAT_ADD64(ctx, child, "ExcessiveCollisions",
6537	    &stats->dot3StatsExcessiveCollisions, "Excessive Collisions");
6538	BGE_SYSCTL_STAT_ADD64(ctx, child, "LateCollisions",
6539	    &stats->dot3StatsLateCollisions, "Late Collisions");
6540	BGE_SYSCTL_STAT_ADD64(ctx, child, "UnicastPkts",
6541	    &stats->ifHCOutUcastPkts, "Outbound Unicast Packets");
6542	BGE_SYSCTL_STAT_ADD64(ctx, child, "MulticastPkts",
6543	    &stats->ifHCOutMulticastPkts, "Outbound Multicast Packets");
6544	BGE_SYSCTL_STAT_ADD64(ctx, child, "BroadcastPkts",
6545	    &stats->ifHCOutBroadcastPkts, "Outbound Broadcast Packets");
6546}
6547
6548#undef	BGE_SYSCTL_STAT_ADD64
6549
6550static int
6551bge_sysctl_stats(SYSCTL_HANDLER_ARGS)
6552{
6553	struct bge_softc *sc;
6554	uint32_t result;
6555	int offset;
6556
6557	sc = (struct bge_softc *)arg1;
6558	offset = arg2;
6559	result = CSR_READ_4(sc, BGE_MEMWIN_START + BGE_STATS_BLOCK + offset +
6560	    offsetof(bge_hostaddr, bge_addr_lo));
6561	return (sysctl_handle_int(oidp, &result, 0, req));
6562}
6563
6564#ifdef BGE_REGISTER_DEBUG
6565static int
6566bge_sysctl_debug_info(SYSCTL_HANDLER_ARGS)
6567{
6568	struct bge_softc *sc;
6569	uint16_t *sbdata;
6570	int error, result, sbsz;
6571	int i, j;
6572
6573	result = -1;
6574	error = sysctl_handle_int(oidp, &result, 0, req);
6575	if (error || (req->newptr == NULL))
6576		return (error);
6577
6578	if (result == 1) {
6579		sc = (struct bge_softc *)arg1;
6580
6581		if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
6582		    sc->bge_chipid != BGE_CHIPID_BCM5700_C0)
6583			sbsz = BGE_STATUS_BLK_SZ;
6584		else
6585			sbsz = 32;
6586		sbdata = (uint16_t *)sc->bge_ldata.bge_status_block;
6587		printf("Status Block:\n");
6588		BGE_LOCK(sc);
6589		bus_dmamap_sync(sc->bge_cdata.bge_status_tag,
6590		    sc->bge_cdata.bge_status_map,
6591		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
6592		for (i = 0x0; i < sbsz / sizeof(uint16_t); ) {
6593			printf("%06x:", i);
6594			for (j = 0; j < 8; j++)
6595				printf(" %04x", sbdata[i++]);
6596			printf("\n");
6597		}
6598
6599		printf("Registers:\n");
6600		for (i = 0x800; i < 0xA00; ) {
6601			printf("%06x:", i);
6602			for (j = 0; j < 8; j++) {
6603				printf(" %08x", CSR_READ_4(sc, i));
6604				i += 4;
6605			}
6606			printf("\n");
6607		}
6608		BGE_UNLOCK(sc);
6609
6610		printf("Hardware Flags:\n");
6611		if (BGE_IS_5717_PLUS(sc))
6612			printf(" - 5717 Plus\n");
6613		if (BGE_IS_5755_PLUS(sc))
6614			printf(" - 5755 Plus\n");
6615		if (BGE_IS_575X_PLUS(sc))
6616			printf(" - 575X Plus\n");
6617		if (BGE_IS_5705_PLUS(sc))
6618			printf(" - 5705 Plus\n");
6619		if (BGE_IS_5714_FAMILY(sc))
6620			printf(" - 5714 Family\n");
6621		if (BGE_IS_5700_FAMILY(sc))
6622			printf(" - 5700 Family\n");
6623		if (sc->bge_flags & BGE_FLAG_JUMBO)
6624			printf(" - Supports Jumbo Frames\n");
6625		if (sc->bge_flags & BGE_FLAG_PCIX)
6626			printf(" - PCI-X Bus\n");
6627		if (sc->bge_flags & BGE_FLAG_PCIE)
6628			printf(" - PCI Express Bus\n");
6629		if (sc->bge_phy_flags & BGE_PHY_NO_3LED)
6630			printf(" - No 3 LEDs\n");
6631		if (sc->bge_flags & BGE_FLAG_RX_ALIGNBUG)
6632			printf(" - RX Alignment Bug\n");
6633	}
6634
6635	return (error);
6636}
6637
6638static int
6639bge_sysctl_reg_read(SYSCTL_HANDLER_ARGS)
6640{
6641	struct bge_softc *sc;
6642	int error;
6643	uint16_t result;
6644	uint32_t val;
6645
6646	result = -1;
6647	error = sysctl_handle_int(oidp, &result, 0, req);
6648	if (error || (req->newptr == NULL))
6649		return (error);
6650
6651	if (result < 0x8000) {
6652		sc = (struct bge_softc *)arg1;
6653		val = CSR_READ_4(sc, result);
6654		printf("reg 0x%06X = 0x%08X\n", result, val);
6655	}
6656
6657	return (error);
6658}
6659
6660static int
6661bge_sysctl_ape_read(SYSCTL_HANDLER_ARGS)
6662{
6663	struct bge_softc *sc;
6664	int error;
6665	uint16_t result;
6666	uint32_t val;
6667
6668	result = -1;
6669	error = sysctl_handle_int(oidp, &result, 0, req);
6670	if (error || (req->newptr == NULL))
6671		return (error);
6672
6673	if (result < 0x8000) {
6674		sc = (struct bge_softc *)arg1;
6675		val = APE_READ_4(sc, result);
6676		printf("reg 0x%06X = 0x%08X\n", result, val);
6677	}
6678
6679	return (error);
6680}
6681
6682static int
6683bge_sysctl_mem_read(SYSCTL_HANDLER_ARGS)
6684{
6685	struct bge_softc *sc;
6686	int error;
6687	uint16_t result;
6688	uint32_t val;
6689
6690	result = -1;
6691	error = sysctl_handle_int(oidp, &result, 0, req);
6692	if (error || (req->newptr == NULL))
6693		return (error);
6694
6695	if (result < 0x8000) {
6696		sc = (struct bge_softc *)arg1;
6697		val = bge_readmem_ind(sc, result);
6698		printf("mem 0x%06X = 0x%08X\n", result, val);
6699	}
6700
6701	return (error);
6702}
6703#endif
6704
6705static int
6706bge_get_eaddr_fw(struct bge_softc *sc, uint8_t ether_addr[])
6707{
6708
6709	if (sc->bge_flags & BGE_FLAG_EADDR)
6710		return (1);
6711
6712#ifdef __sparc64__
6713	OF_getetheraddr(sc->bge_dev, ether_addr);
6714	return (0);
6715#endif
6716	return (1);
6717}
6718
6719static int
6720bge_get_eaddr_mem(struct bge_softc *sc, uint8_t ether_addr[])
6721{
6722	uint32_t mac_addr;
6723
6724	mac_addr = bge_readmem_ind(sc, BGE_SRAM_MAC_ADDR_HIGH_MB);
6725	if ((mac_addr >> 16) == 0x484b) {
6726		ether_addr[0] = (uint8_t)(mac_addr >> 8);
6727		ether_addr[1] = (uint8_t)mac_addr;
6728		mac_addr = bge_readmem_ind(sc, BGE_SRAM_MAC_ADDR_LOW_MB);
6729		ether_addr[2] = (uint8_t)(mac_addr >> 24);
6730		ether_addr[3] = (uint8_t)(mac_addr >> 16);
6731		ether_addr[4] = (uint8_t)(mac_addr >> 8);
6732		ether_addr[5] = (uint8_t)mac_addr;
6733		return (0);
6734	}
6735	return (1);
6736}
6737
6738static int
6739bge_get_eaddr_nvram(struct bge_softc *sc, uint8_t ether_addr[])
6740{
6741	int mac_offset = BGE_EE_MAC_OFFSET;
6742
6743	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
6744		mac_offset = BGE_EE_MAC_OFFSET_5906;
6745
6746	return (bge_read_nvram(sc, ether_addr, mac_offset + 2,
6747	    ETHER_ADDR_LEN));
6748}
6749
6750static int
6751bge_get_eaddr_eeprom(struct bge_softc *sc, uint8_t ether_addr[])
6752{
6753
6754	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
6755		return (1);
6756
6757	return (bge_read_eeprom(sc, ether_addr, BGE_EE_MAC_OFFSET + 2,
6758	   ETHER_ADDR_LEN));
6759}
6760
6761static int
6762bge_get_eaddr(struct bge_softc *sc, uint8_t eaddr[])
6763{
6764	static const bge_eaddr_fcn_t bge_eaddr_funcs[] = {
6765		/* NOTE: Order is critical */
6766		bge_get_eaddr_fw,
6767		bge_get_eaddr_mem,
6768		bge_get_eaddr_nvram,
6769		bge_get_eaddr_eeprom,
6770		NULL
6771	};
6772	const bge_eaddr_fcn_t *func;
6773
6774	for (func = bge_eaddr_funcs; *func != NULL; ++func) {
6775		if ((*func)(sc, eaddr) == 0)
6776			break;
6777	}
6778	return (*func == NULL ? ENXIO : 0);
6779}
6780