xref: /dflybsd-src/sys/dev/netif/bge/if_bge.c (revision d88b9605be186eb2c93c4f2683c997a128758b5a)
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  * $FreeBSD: src/sys/dev/bge/if_bge.c,v 1.3.2.39 2005/07/03 03:41:18 silby Exp $
34  */
35 
36 /*
37  * Broadcom BCM570x family gigabit ethernet driver for FreeBSD.
38  *
39  * Written by Bill Paul <wpaul@windriver.com>
40  * Senior Engineer, Wind River Systems
41  */
42 
43 /*
44  * The Broadcom BCM5700 is based on technology originally developed by
45  * Alteon Networks as part of the Tigon I and Tigon II gigabit ethernet
46  * MAC chips. The BCM5700, sometimes refered to as the Tigon III, has
47  * two on-board MIPS R4000 CPUs and can have as much as 16MB of external
48  * SSRAM. The BCM5700 supports TCP, UDP and IP checksum offload, jumbo
49  * frames, highly configurable RX filtering, and 16 RX and TX queues
50  * (which, along with RX filter rules, can be used for QOS applications).
51  * Other features, such as TCP segmentation, may be available as part
52  * of value-added firmware updates. Unlike the Tigon I and Tigon II,
53  * firmware images can be stored in hardware and need not be compiled
54  * into the driver.
55  *
56  * The BCM5700 supports the PCI v2.2 and PCI-X v1.0 standards, and will
57  * function in a 32-bit/64-bit 33/66Mhz bus, or a 64-bit/133Mhz bus.
58  *
59  * The BCM5701 is a single-chip solution incorporating both the BCM5700
60  * MAC and a BCM5401 10/100/1000 PHY. Unlike the BCM5700, the BCM5701
61  * does not support external SSRAM.
62  *
63  * Broadcom also produces a variation of the BCM5700 under the "Altima"
64  * brand name, which is functionally similar but lacks PCI-X support.
65  *
66  * Without external SSRAM, you can only have at most 4 TX rings,
67  * and the use of the mini RX ring is disabled. This seems to imply
68  * that these features are simply not available on the BCM5701. As a
69  * result, this driver does not implement any support for the mini RX
70  * ring.
71  */
72 
73 #include "opt_ifpoll.h"
74 
75 #include <sys/param.h>
76 #include <sys/bus.h>
77 #include <sys/endian.h>
78 #include <sys/kernel.h>
79 #include <sys/ktr.h>
80 #include <sys/interrupt.h>
81 #include <sys/mbuf.h>
82 #include <sys/malloc.h>
83 #include <sys/queue.h>
84 #include <sys/rman.h>
85 #include <sys/serialize.h>
86 #include <sys/socket.h>
87 #include <sys/sockio.h>
88 #include <sys/sysctl.h>
89 
90 #include <netinet/ip.h>
91 #include <netinet/tcp.h>
92 
93 #include <net/bpf.h>
94 #include <net/ethernet.h>
95 #include <net/if.h>
96 #include <net/if_arp.h>
97 #include <net/if_dl.h>
98 #include <net/if_media.h>
99 #include <net/if_poll.h>
100 #include <net/if_types.h>
101 #include <net/ifq_var.h>
102 #include <net/vlan/if_vlan_var.h>
103 #include <net/vlan/if_vlan_ether.h>
104 
105 #include <dev/netif/mii_layer/mii.h>
106 #include <dev/netif/mii_layer/miivar.h>
107 #include <dev/netif/mii_layer/brgphyreg.h>
108 
109 #include <bus/pci/pcidevs.h>
110 #include <bus/pci/pcireg.h>
111 #include <bus/pci/pcivar.h>
112 
113 #include <dev/netif/bge/if_bgereg.h>
114 #include <dev/netif/bge/if_bgevar.h>
115 
116 /* "device miibus" required.  See GENERIC if you get errors here. */
117 #include "miibus_if.h"
118 
119 #define BGE_CSUM_FEATURES	(CSUM_IP | CSUM_TCP)
120 
121 static const struct bge_type {
122 	uint16_t		bge_vid;
123 	uint16_t		bge_did;
124 	char			*bge_name;
125 } bge_devs[] = {
126 	{ PCI_VENDOR_3COM, PCI_PRODUCT_3COM_3C996,
127 		"3COM 3C996 Gigabit Ethernet" },
128 
129 	{ PCI_VENDOR_ALTEON, PCI_PRODUCT_ALTEON_BCM5700,
130 		"Alteon BCM5700 Gigabit Ethernet" },
131 	{ PCI_VENDOR_ALTEON, PCI_PRODUCT_ALTEON_BCM5701,
132 		"Alteon BCM5701 Gigabit Ethernet" },
133 
134 	{ PCI_VENDOR_ALTIMA, PCI_PRODUCT_ALTIMA_AC1000,
135 		"Altima AC1000 Gigabit Ethernet" },
136 	{ PCI_VENDOR_ALTIMA, PCI_PRODUCT_ALTIMA_AC1001,
137 		"Altima AC1002 Gigabit Ethernet" },
138 	{ PCI_VENDOR_ALTIMA, PCI_PRODUCT_ALTIMA_AC9100,
139 		"Altima AC9100 Gigabit Ethernet" },
140 
141 	{ PCI_VENDOR_APPLE, PCI_PRODUCT_APPLE_BCM5701,
142 		"Apple BCM5701 Gigabit Ethernet" },
143 
144 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5700,
145 		"Broadcom BCM5700 Gigabit Ethernet" },
146 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5701,
147 		"Broadcom BCM5701 Gigabit Ethernet" },
148 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5702,
149 		"Broadcom BCM5702 Gigabit Ethernet" },
150 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5702X,
151 		"Broadcom BCM5702X Gigabit Ethernet" },
152 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5702_ALT,
153 		"Broadcom BCM5702 Gigabit Ethernet" },
154 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5703,
155 		"Broadcom BCM5703 Gigabit Ethernet" },
156 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5703X,
157 		"Broadcom BCM5703X Gigabit Ethernet" },
158 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5703A3,
159 		"Broadcom BCM5703 Gigabit Ethernet" },
160 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5704C,
161 		"Broadcom BCM5704C Dual Gigabit Ethernet" },
162 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5704S,
163 		"Broadcom BCM5704S Dual Gigabit Ethernet" },
164 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5704S_ALT,
165 		"Broadcom BCM5704S Dual Gigabit Ethernet" },
166 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5705,
167 		"Broadcom BCM5705 Gigabit Ethernet" },
168 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5705F,
169 		"Broadcom BCM5705F Gigabit Ethernet" },
170 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5705K,
171 		"Broadcom BCM5705K Gigabit Ethernet" },
172 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5705M,
173 		"Broadcom BCM5705M Gigabit Ethernet" },
174 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5705M_ALT,
175 		"Broadcom BCM5705M Gigabit Ethernet" },
176 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5714,
177 		"Broadcom BCM5714C Gigabit Ethernet" },
178 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5714S,
179 		"Broadcom BCM5714S Gigabit Ethernet" },
180 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5715,
181 		"Broadcom BCM5715 Gigabit Ethernet" },
182 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5715S,
183 		"Broadcom BCM5715S Gigabit Ethernet" },
184 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5720,
185 		"Broadcom BCM5720 Gigabit Ethernet" },
186 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5721,
187 		"Broadcom BCM5721 Gigabit Ethernet" },
188 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5722,
189 		"Broadcom BCM5722 Gigabit Ethernet" },
190 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5723,
191 		"Broadcom BCM5723 Gigabit Ethernet" },
192 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5750,
193 		"Broadcom BCM5750 Gigabit Ethernet" },
194 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5750M,
195 		"Broadcom BCM5750M Gigabit Ethernet" },
196 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5751,
197 		"Broadcom BCM5751 Gigabit Ethernet" },
198 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5751F,
199 		"Broadcom BCM5751F Gigabit Ethernet" },
200 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5751M,
201 		"Broadcom BCM5751M Gigabit Ethernet" },
202 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5752,
203 		"Broadcom BCM5752 Gigabit Ethernet" },
204 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5752M,
205 		"Broadcom BCM5752M Gigabit Ethernet" },
206 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5753,
207 		"Broadcom BCM5753 Gigabit Ethernet" },
208 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5753F,
209 		"Broadcom BCM5753F Gigabit Ethernet" },
210 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5753M,
211 		"Broadcom BCM5753M Gigabit Ethernet" },
212 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5754,
213 		"Broadcom BCM5754 Gigabit Ethernet" },
214 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5754M,
215 		"Broadcom BCM5754M Gigabit Ethernet" },
216 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5755,
217 		"Broadcom BCM5755 Gigabit Ethernet" },
218 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5755M,
219 		"Broadcom BCM5755M Gigabit Ethernet" },
220 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5756,
221 		"Broadcom BCM5756 Gigabit Ethernet" },
222 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5761,
223 		"Broadcom BCM5761 Gigabit Ethernet" },
224 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5761E,
225 		"Broadcom BCM5761E Gigabit Ethernet" },
226 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5761S,
227 		"Broadcom BCM5761S Gigabit Ethernet" },
228 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5761SE,
229 		"Broadcom BCM5761SE Gigabit Ethernet" },
230 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5764,
231 		"Broadcom BCM5764 Gigabit Ethernet" },
232 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5780,
233 		"Broadcom BCM5780 Gigabit Ethernet" },
234 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5780S,
235 		"Broadcom BCM5780S Gigabit Ethernet" },
236 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5781,
237 		"Broadcom BCM5781 Gigabit Ethernet" },
238 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5782,
239 		"Broadcom BCM5782 Gigabit Ethernet" },
240 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5784,
241 		"Broadcom BCM5784 Gigabit Ethernet" },
242 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5785F,
243 		"Broadcom BCM5785F Gigabit Ethernet" },
244 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5785G,
245 		"Broadcom BCM5785G Gigabit Ethernet" },
246 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5786,
247 		"Broadcom BCM5786 Gigabit Ethernet" },
248 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5787,
249 		"Broadcom BCM5787 Gigabit Ethernet" },
250 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5787F,
251 		"Broadcom BCM5787F Gigabit Ethernet" },
252 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5787M,
253 		"Broadcom BCM5787M Gigabit Ethernet" },
254 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5788,
255 		"Broadcom BCM5788 Gigabit Ethernet" },
256 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5789,
257 		"Broadcom BCM5789 Gigabit Ethernet" },
258 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5901,
259 		"Broadcom BCM5901 Fast Ethernet" },
260 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5901A2,
261 		"Broadcom BCM5901A2 Fast Ethernet" },
262 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5903M,
263 		"Broadcom BCM5903M Fast Ethernet" },
264 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5906,
265 		"Broadcom BCM5906 Fast Ethernet"},
266 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5906M,
267 		"Broadcom BCM5906M Fast Ethernet"},
268 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57760,
269 		"Broadcom BCM57760 Gigabit Ethernet"},
270 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57780,
271 		"Broadcom BCM57780 Gigabit Ethernet"},
272 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57788,
273 		"Broadcom BCM57788 Gigabit Ethernet"},
274 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57790,
275 		"Broadcom BCM57790 Gigabit Ethernet"},
276 	{ PCI_VENDOR_SCHNEIDERKOCH, PCI_PRODUCT_SCHNEIDERKOCH_SK_9DX1,
277 		"SysKonnect Gigabit Ethernet" },
278 
279 	{ 0, 0, NULL }
280 };
281 
282 #define BGE_IS_JUMBO_CAPABLE(sc)	((sc)->bge_flags & BGE_FLAG_JUMBO)
283 #define BGE_IS_5700_FAMILY(sc)		((sc)->bge_flags & BGE_FLAG_5700_FAMILY)
284 #define BGE_IS_5705_PLUS(sc)		((sc)->bge_flags & BGE_FLAG_5705_PLUS)
285 #define BGE_IS_5714_FAMILY(sc)		((sc)->bge_flags & BGE_FLAG_5714_FAMILY)
286 #define BGE_IS_575X_PLUS(sc)		((sc)->bge_flags & BGE_FLAG_575X_PLUS)
287 #define BGE_IS_5755_PLUS(sc)		((sc)->bge_flags & BGE_FLAG_5755_PLUS)
288 #define BGE_IS_5788(sc)			((sc)->bge_flags & BGE_FLAG_5788)
289 
290 #define BGE_IS_CRIPPLED(sc)		\
291 	(BGE_IS_5788((sc)) || (sc)->bge_asicrev == BGE_ASICREV_BCM5700)
292 
293 typedef int	(*bge_eaddr_fcn_t)(struct bge_softc *, uint8_t[]);
294 
295 static int	bge_probe(device_t);
296 static int	bge_attach(device_t);
297 static int	bge_detach(device_t);
298 static void	bge_txeof(struct bge_softc *, uint16_t);
299 static void	bge_rxeof(struct bge_softc *, uint16_t, int);
300 
301 static void	bge_tick(void *);
302 static void	bge_stats_update(struct bge_softc *);
303 static void	bge_stats_update_regs(struct bge_softc *);
304 static struct mbuf *
305 		bge_defrag_shortdma(struct mbuf *);
306 static int	bge_encap(struct bge_softc *, struct mbuf **,
307 		    uint32_t *, int *);
308 static void	bge_xmit(struct bge_softc *, uint32_t);
309 static int	bge_setup_tso(struct bge_softc *, struct mbuf **,
310 		    uint16_t *, uint16_t *);
311 
312 #ifdef IFPOLL_ENABLE
313 static void	bge_npoll(struct ifnet *, struct ifpoll_info *);
314 static void	bge_npoll_compat(struct ifnet *, void *, int );
315 #endif
316 static void	bge_intr_crippled(void *);
317 static void	bge_intr_legacy(void *);
318 static void	bge_msi(void *);
319 static void	bge_msi_oneshot(void *);
320 static void	bge_intr(struct bge_softc *);
321 static void	bge_enable_intr(struct bge_softc *);
322 static void	bge_disable_intr(struct bge_softc *);
323 static void	bge_start(struct ifnet *, struct ifaltq_subque *);
324 static int	bge_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
325 static void	bge_init(void *);
326 static void	bge_stop(struct bge_softc *);
327 static void	bge_watchdog(struct ifnet *);
328 static void	bge_shutdown(device_t);
329 static int	bge_suspend(device_t);
330 static int	bge_resume(device_t);
331 static int	bge_ifmedia_upd(struct ifnet *);
332 static void	bge_ifmedia_sts(struct ifnet *, struct ifmediareq *);
333 
334 static uint8_t	bge_nvram_getbyte(struct bge_softc *, int, uint8_t *);
335 static int	bge_read_nvram(struct bge_softc *, caddr_t, int, int);
336 
337 static uint8_t	bge_eeprom_getbyte(struct bge_softc *, uint32_t, uint8_t *);
338 static int	bge_read_eeprom(struct bge_softc *, caddr_t, uint32_t, size_t);
339 
340 static void	bge_setmulti(struct bge_softc *);
341 static void	bge_setpromisc(struct bge_softc *);
342 static void	bge_enable_msi(struct bge_softc *sc);
343 
344 static int	bge_alloc_jumbo_mem(struct bge_softc *);
345 static void	bge_free_jumbo_mem(struct bge_softc *);
346 static struct bge_jslot
347 		*bge_jalloc(struct bge_softc *);
348 static void	bge_jfree(void *);
349 static void	bge_jref(void *);
350 static int	bge_newbuf_std(struct bge_softc *, int, int);
351 static int	bge_newbuf_jumbo(struct bge_softc *, int, int);
352 static void	bge_setup_rxdesc_std(struct bge_softc *, int);
353 static void	bge_setup_rxdesc_jumbo(struct bge_softc *, int);
354 static int	bge_init_rx_ring_std(struct bge_softc *);
355 static void	bge_free_rx_ring_std(struct bge_softc *);
356 static int	bge_init_rx_ring_jumbo(struct bge_softc *);
357 static void	bge_free_rx_ring_jumbo(struct bge_softc *);
358 static void	bge_free_tx_ring(struct bge_softc *);
359 static int	bge_init_tx_ring(struct bge_softc *);
360 
361 static int	bge_chipinit(struct bge_softc *);
362 static int	bge_blockinit(struct bge_softc *);
363 static void	bge_stop_block(struct bge_softc *, bus_size_t, uint32_t);
364 
365 static uint32_t	bge_readmem_ind(struct bge_softc *, uint32_t);
366 static void	bge_writemem_ind(struct bge_softc *, uint32_t, uint32_t);
367 #ifdef notdef
368 static uint32_t	bge_readreg_ind(struct bge_softc *, uint32_t);
369 #endif
370 static void	bge_writereg_ind(struct bge_softc *, uint32_t, uint32_t);
371 static void	bge_writemem_direct(struct bge_softc *, uint32_t, uint32_t);
372 static void	bge_writembx(struct bge_softc *, int, int);
373 
374 static int	bge_miibus_readreg(device_t, int, int);
375 static int	bge_miibus_writereg(device_t, int, int, int);
376 static void	bge_miibus_statchg(device_t);
377 static void	bge_bcm5700_link_upd(struct bge_softc *, uint32_t);
378 static void	bge_tbi_link_upd(struct bge_softc *, uint32_t);
379 static void	bge_copper_link_upd(struct bge_softc *, uint32_t);
380 static void	bge_autopoll_link_upd(struct bge_softc *, uint32_t);
381 static void	bge_link_poll(struct bge_softc *);
382 
383 static void	bge_reset(struct bge_softc *);
384 
385 static int	bge_dma_alloc(struct bge_softc *);
386 static void	bge_dma_free(struct bge_softc *);
387 static int	bge_dma_block_alloc(struct bge_softc *, bus_size_t,
388 				    bus_dma_tag_t *, bus_dmamap_t *,
389 				    void **, bus_addr_t *);
390 static void	bge_dma_block_free(bus_dma_tag_t, bus_dmamap_t, void *);
391 
392 static int	bge_get_eaddr_mem(struct bge_softc *, uint8_t[]);
393 static int	bge_get_eaddr_nvram(struct bge_softc *, uint8_t[]);
394 static int	bge_get_eaddr_eeprom(struct bge_softc *, uint8_t[]);
395 static int	bge_get_eaddr(struct bge_softc *, uint8_t[]);
396 
397 static void	bge_coal_change(struct bge_softc *);
398 static int	bge_sysctl_rx_coal_ticks(SYSCTL_HANDLER_ARGS);
399 static int	bge_sysctl_tx_coal_ticks(SYSCTL_HANDLER_ARGS);
400 static int	bge_sysctl_rx_coal_bds(SYSCTL_HANDLER_ARGS);
401 static int	bge_sysctl_tx_coal_bds(SYSCTL_HANDLER_ARGS);
402 static int	bge_sysctl_rx_coal_ticks_int(SYSCTL_HANDLER_ARGS);
403 static int	bge_sysctl_tx_coal_ticks_int(SYSCTL_HANDLER_ARGS);
404 static int	bge_sysctl_rx_coal_bds_int(SYSCTL_HANDLER_ARGS);
405 static int	bge_sysctl_tx_coal_bds_int(SYSCTL_HANDLER_ARGS);
406 static int	bge_sysctl_coal_chg(SYSCTL_HANDLER_ARGS, uint32_t *,
407 		    int, int, uint32_t);
408 
409 /*
410  * Set following tunable to 1 for some IBM blade servers with the DNLK
411  * switch module. Auto negotiation is broken for those configurations.
412  */
413 static int	bge_fake_autoneg = 0;
414 TUNABLE_INT("hw.bge.fake_autoneg", &bge_fake_autoneg);
415 
416 static int	bge_msi_enable = 1;
417 TUNABLE_INT("hw.bge.msi.enable", &bge_msi_enable);
418 
419 #if !defined(KTR_IF_BGE)
420 #define KTR_IF_BGE	KTR_ALL
421 #endif
422 KTR_INFO_MASTER(if_bge);
423 KTR_INFO(KTR_IF_BGE, if_bge, intr, 0, "intr");
424 KTR_INFO(KTR_IF_BGE, if_bge, rx_pkt, 1, "rx_pkt");
425 KTR_INFO(KTR_IF_BGE, if_bge, tx_pkt, 2, "tx_pkt");
426 #define logif(name)	KTR_LOG(if_bge_ ## name)
427 
428 static device_method_t bge_methods[] = {
429 	/* Device interface */
430 	DEVMETHOD(device_probe,		bge_probe),
431 	DEVMETHOD(device_attach,	bge_attach),
432 	DEVMETHOD(device_detach,	bge_detach),
433 	DEVMETHOD(device_shutdown,	bge_shutdown),
434 	DEVMETHOD(device_suspend,	bge_suspend),
435 	DEVMETHOD(device_resume,	bge_resume),
436 
437 	/* bus interface */
438 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
439 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
440 
441 	/* MII interface */
442 	DEVMETHOD(miibus_readreg,	bge_miibus_readreg),
443 	DEVMETHOD(miibus_writereg,	bge_miibus_writereg),
444 	DEVMETHOD(miibus_statchg,	bge_miibus_statchg),
445 
446 	DEVMETHOD_END
447 };
448 
449 static DEFINE_CLASS_0(bge, bge_driver, bge_methods, sizeof(struct bge_softc));
450 static devclass_t bge_devclass;
451 
452 DECLARE_DUMMY_MODULE(if_bge);
453 DRIVER_MODULE(if_bge, pci, bge_driver, bge_devclass, NULL, NULL);
454 DRIVER_MODULE(miibus, bge, miibus_driver, miibus_devclass, NULL, NULL);
455 
456 static uint32_t
457 bge_readmem_ind(struct bge_softc *sc, uint32_t off)
458 {
459 	device_t dev = sc->bge_dev;
460 	uint32_t val;
461 
462 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
463 	    off >= BGE_STATS_BLOCK && off < BGE_SEND_RING_1_TO_4)
464 		return 0;
465 
466 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4);
467 	val = pci_read_config(dev, BGE_PCI_MEMWIN_DATA, 4);
468 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4);
469 	return (val);
470 }
471 
472 static void
473 bge_writemem_ind(struct bge_softc *sc, uint32_t off, uint32_t val)
474 {
475 	device_t dev = sc->bge_dev;
476 
477 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
478 	    off >= BGE_STATS_BLOCK && off < BGE_SEND_RING_1_TO_4)
479 		return;
480 
481 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4);
482 	pci_write_config(dev, BGE_PCI_MEMWIN_DATA, val, 4);
483 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4);
484 }
485 
486 #ifdef notdef
487 static uint32_t
488 bge_readreg_ind(struct bge_softc *sc, uin32_t off)
489 {
490 	device_t dev = sc->bge_dev;
491 
492 	pci_write_config(dev, BGE_PCI_REG_BASEADDR, off, 4);
493 	return(pci_read_config(dev, BGE_PCI_REG_DATA, 4));
494 }
495 #endif
496 
497 static void
498 bge_writereg_ind(struct bge_softc *sc, uint32_t off, uint32_t val)
499 {
500 	device_t dev = sc->bge_dev;
501 
502 	pci_write_config(dev, BGE_PCI_REG_BASEADDR, off, 4);
503 	pci_write_config(dev, BGE_PCI_REG_DATA, val, 4);
504 }
505 
506 static void
507 bge_writemem_direct(struct bge_softc *sc, uint32_t off, uint32_t val)
508 {
509 	CSR_WRITE_4(sc, off, val);
510 }
511 
512 static void
513 bge_writembx(struct bge_softc *sc, int off, int val)
514 {
515 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
516 		off += BGE_LPMBX_IRQ0_HI - BGE_MBX_IRQ0_HI;
517 
518 	CSR_WRITE_4(sc, off, val);
519 	if (sc->bge_mbox_reorder)
520 		CSR_READ_4(sc, off);
521 }
522 
523 static uint8_t
524 bge_nvram_getbyte(struct bge_softc *sc, int addr, uint8_t *dest)
525 {
526 	uint32_t access, byte = 0;
527 	int i;
528 
529 	/* Lock. */
530 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_SET1);
531 	for (i = 0; i < 8000; i++) {
532 		if (CSR_READ_4(sc, BGE_NVRAM_SWARB) & BGE_NVRAMSWARB_GNT1)
533 			break;
534 		DELAY(20);
535 	}
536 	if (i == 8000)
537 		return (1);
538 
539 	/* Enable access. */
540 	access = CSR_READ_4(sc, BGE_NVRAM_ACCESS);
541 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access | BGE_NVRAMACC_ENABLE);
542 
543 	CSR_WRITE_4(sc, BGE_NVRAM_ADDR, addr & 0xfffffffc);
544 	CSR_WRITE_4(sc, BGE_NVRAM_CMD, BGE_NVRAM_READCMD);
545 	for (i = 0; i < BGE_TIMEOUT * 10; i++) {
546 		DELAY(10);
547 		if (CSR_READ_4(sc, BGE_NVRAM_CMD) & BGE_NVRAMCMD_DONE) {
548 			DELAY(10);
549 			break;
550 		}
551 	}
552 
553 	if (i == BGE_TIMEOUT * 10) {
554 		if_printf(&sc->arpcom.ac_if, "nvram read timed out\n");
555 		return (1);
556 	}
557 
558 	/* Get result. */
559 	byte = CSR_READ_4(sc, BGE_NVRAM_RDDATA);
560 
561 	*dest = (bswap32(byte) >> ((addr % 4) * 8)) & 0xFF;
562 
563 	/* Disable access. */
564 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access);
565 
566 	/* Unlock. */
567 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_CLR1);
568 	CSR_READ_4(sc, BGE_NVRAM_SWARB);
569 
570 	return (0);
571 }
572 
573 /*
574  * Read a sequence of bytes from NVRAM.
575  */
576 static int
577 bge_read_nvram(struct bge_softc *sc, caddr_t dest, int off, int cnt)
578 {
579 	int err = 0, i;
580 	uint8_t byte = 0;
581 
582 	if (sc->bge_asicrev != BGE_ASICREV_BCM5906)
583 		return (1);
584 
585 	for (i = 0; i < cnt; i++) {
586 		err = bge_nvram_getbyte(sc, off + i, &byte);
587 		if (err)
588 			break;
589 		*(dest + i) = byte;
590 	}
591 
592 	return (err ? 1 : 0);
593 }
594 
595 /*
596  * Read a byte of data stored in the EEPROM at address 'addr.' The
597  * BCM570x supports both the traditional bitbang interface and an
598  * auto access interface for reading the EEPROM. We use the auto
599  * access method.
600  */
601 static uint8_t
602 bge_eeprom_getbyte(struct bge_softc *sc, uint32_t addr, uint8_t *dest)
603 {
604 	int i;
605 	uint32_t byte = 0;
606 
607 	/*
608 	 * Enable use of auto EEPROM access so we can avoid
609 	 * having to use the bitbang method.
610 	 */
611 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_AUTO_EEPROM);
612 
613 	/* Reset the EEPROM, load the clock period. */
614 	CSR_WRITE_4(sc, BGE_EE_ADDR,
615 	    BGE_EEADDR_RESET|BGE_EEHALFCLK(BGE_HALFCLK_384SCL));
616 	DELAY(20);
617 
618 	/* Issue the read EEPROM command. */
619 	CSR_WRITE_4(sc, BGE_EE_ADDR, BGE_EE_READCMD | addr);
620 
621 	/* Wait for completion */
622 	for(i = 0; i < BGE_TIMEOUT * 10; i++) {
623 		DELAY(10);
624 		if (CSR_READ_4(sc, BGE_EE_ADDR) & BGE_EEADDR_DONE)
625 			break;
626 	}
627 
628 	if (i == BGE_TIMEOUT) {
629 		if_printf(&sc->arpcom.ac_if, "eeprom read timed out\n");
630 		return(1);
631 	}
632 
633 	/* Get result. */
634 	byte = CSR_READ_4(sc, BGE_EE_DATA);
635 
636         *dest = (byte >> ((addr % 4) * 8)) & 0xFF;
637 
638 	return(0);
639 }
640 
641 /*
642  * Read a sequence of bytes from the EEPROM.
643  */
644 static int
645 bge_read_eeprom(struct bge_softc *sc, caddr_t dest, uint32_t off, size_t len)
646 {
647 	size_t i;
648 	int err;
649 	uint8_t byte;
650 
651 	for (byte = 0, err = 0, i = 0; i < len; i++) {
652 		err = bge_eeprom_getbyte(sc, off + i, &byte);
653 		if (err)
654 			break;
655 		*(dest + i) = byte;
656 	}
657 
658 	return(err ? 1 : 0);
659 }
660 
661 static int
662 bge_miibus_readreg(device_t dev, int phy, int reg)
663 {
664 	struct bge_softc *sc = device_get_softc(dev);
665 	uint32_t val;
666 	int i;
667 
668 	KASSERT(phy == sc->bge_phyno,
669 	    ("invalid phyno %d, should be %d", phy, sc->bge_phyno));
670 
671 	/* Clear the autopoll bit if set, otherwise may trigger PCI errors. */
672 	if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
673 		CSR_WRITE_4(sc, BGE_MI_MODE,
674 		    sc->bge_mi_mode & ~BGE_MIMODE_AUTOPOLL);
675 		DELAY(80);
676 	}
677 
678 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_READ | BGE_MICOMM_BUSY |
679 	    BGE_MIPHY(phy) | BGE_MIREG(reg));
680 
681 	/* Poll for the PHY register access to complete. */
682 	for (i = 0; i < BGE_TIMEOUT; i++) {
683 		DELAY(10);
684 		val = CSR_READ_4(sc, BGE_MI_COMM);
685 		if ((val & BGE_MICOMM_BUSY) == 0) {
686 			DELAY(5);
687 			val = CSR_READ_4(sc, BGE_MI_COMM);
688 			break;
689 		}
690 	}
691 	if (i == BGE_TIMEOUT) {
692 		if_printf(&sc->arpcom.ac_if, "PHY read timed out "
693 		    "(phy %d, reg %d, val 0x%08x)\n", phy, reg, val);
694 		val = 0;
695 	}
696 
697 	/* Restore the autopoll bit if necessary. */
698 	if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
699 		CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode);
700 		DELAY(80);
701 	}
702 
703 	if (val & BGE_MICOMM_READFAIL)
704 		return 0;
705 
706 	return (val & 0xFFFF);
707 }
708 
709 static int
710 bge_miibus_writereg(device_t dev, int phy, int reg, int val)
711 {
712 	struct bge_softc *sc = device_get_softc(dev);
713 	int i;
714 
715 	KASSERT(phy == sc->bge_phyno,
716 	    ("invalid phyno %d, should be %d", phy, sc->bge_phyno));
717 
718 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
719 	    (reg == BRGPHY_MII_1000CTL || reg == BRGPHY_MII_AUXCTL))
720 	       return 0;
721 
722 	/* Clear the autopoll bit if set, otherwise may trigger PCI errors. */
723 	if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
724 		CSR_WRITE_4(sc, BGE_MI_MODE,
725 		    sc->bge_mi_mode & ~BGE_MIMODE_AUTOPOLL);
726 		DELAY(80);
727 	}
728 
729 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_WRITE | BGE_MICOMM_BUSY |
730 	    BGE_MIPHY(phy) | BGE_MIREG(reg) | val);
731 
732 	for (i = 0; i < BGE_TIMEOUT; i++) {
733 		DELAY(10);
734 		if (!(CSR_READ_4(sc, BGE_MI_COMM) & BGE_MICOMM_BUSY)) {
735 			DELAY(5);
736 			CSR_READ_4(sc, BGE_MI_COMM); /* dummy read */
737 			break;
738 		}
739 	}
740 	if (i == BGE_TIMEOUT) {
741 		if_printf(&sc->arpcom.ac_if, "PHY write timed out "
742 		    "(phy %d, reg %d, val %d)\n", phy, reg, val);
743 	}
744 
745 	/* Restore the autopoll bit if necessary. */
746 	if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
747 		CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode);
748 		DELAY(80);
749 	}
750 
751 	return 0;
752 }
753 
754 static void
755 bge_miibus_statchg(device_t dev)
756 {
757 	struct bge_softc *sc;
758 	struct mii_data *mii;
759 
760 	sc = device_get_softc(dev);
761 	mii = device_get_softc(sc->bge_miibus);
762 
763 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
764 	    (IFM_ACTIVE | IFM_AVALID)) {
765 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
766 		case IFM_10_T:
767 		case IFM_100_TX:
768 			sc->bge_link = 1;
769 			break;
770 		case IFM_1000_T:
771 		case IFM_1000_SX:
772 		case IFM_2500_SX:
773 			if (sc->bge_asicrev != BGE_ASICREV_BCM5906)
774 				sc->bge_link = 1;
775 			else
776 				sc->bge_link = 0;
777 			break;
778 		default:
779 			sc->bge_link = 0;
780 			break;
781 		}
782 	} else {
783 		sc->bge_link = 0;
784 	}
785 	if (sc->bge_link == 0)
786 		return;
787 
788 	BGE_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_PORTMODE);
789 	if (IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_T ||
790 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_SX) {
791 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_GMII);
792 	} else {
793 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_MII);
794 	}
795 
796 	if ((mii->mii_media_active & IFM_GMASK) == IFM_FDX) {
797 		BGE_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
798 	} else {
799 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
800 	}
801 }
802 
803 /*
804  * Memory management for jumbo frames.
805  */
806 static int
807 bge_alloc_jumbo_mem(struct bge_softc *sc)
808 {
809 	struct ifnet *ifp = &sc->arpcom.ac_if;
810 	struct bge_jslot *entry;
811 	uint8_t *ptr;
812 	bus_addr_t paddr;
813 	int i, error;
814 
815 	/*
816 	 * Create tag for jumbo mbufs.
817 	 * This is really a bit of a kludge. We allocate a special
818 	 * jumbo buffer pool which (thanks to the way our DMA
819 	 * memory allocation works) will consist of contiguous
820 	 * pages. This means that even though a jumbo buffer might
821 	 * be larger than a page size, we don't really need to
822 	 * map it into more than one DMA segment. However, the
823 	 * default mbuf tag will result in multi-segment mappings,
824 	 * so we have to create a special jumbo mbuf tag that
825 	 * lets us get away with mapping the jumbo buffers as
826 	 * a single segment. I think eventually the driver should
827 	 * be changed so that it uses ordinary mbufs and cluster
828 	 * buffers, i.e. jumbo frames can span multiple DMA
829 	 * descriptors. But that's a project for another day.
830 	 */
831 
832 	/*
833 	 * Create DMA stuffs for jumbo RX ring.
834 	 */
835 	error = bge_dma_block_alloc(sc, BGE_JUMBO_RX_RING_SZ,
836 				    &sc->bge_cdata.bge_rx_jumbo_ring_tag,
837 				    &sc->bge_cdata.bge_rx_jumbo_ring_map,
838 				    (void *)&sc->bge_ldata.bge_rx_jumbo_ring,
839 				    &sc->bge_ldata.bge_rx_jumbo_ring_paddr);
840 	if (error) {
841 		if_printf(ifp, "could not create jumbo RX ring\n");
842 		return error;
843 	}
844 
845 	/*
846 	 * Create DMA stuffs for jumbo buffer block.
847 	 */
848 	error = bge_dma_block_alloc(sc, BGE_JMEM,
849 				    &sc->bge_cdata.bge_jumbo_tag,
850 				    &sc->bge_cdata.bge_jumbo_map,
851 				    (void **)&sc->bge_ldata.bge_jumbo_buf,
852 				    &paddr);
853 	if (error) {
854 		if_printf(ifp, "could not create jumbo buffer\n");
855 		return error;
856 	}
857 
858 	SLIST_INIT(&sc->bge_jfree_listhead);
859 
860 	/*
861 	 * Now divide it up into 9K pieces and save the addresses
862 	 * in an array. Note that we play an evil trick here by using
863 	 * the first few bytes in the buffer to hold the the address
864 	 * of the softc structure for this interface. This is because
865 	 * bge_jfree() needs it, but it is called by the mbuf management
866 	 * code which will not pass it to us explicitly.
867 	 */
868 	for (i = 0, ptr = sc->bge_ldata.bge_jumbo_buf; i < BGE_JSLOTS; i++) {
869 		entry = &sc->bge_cdata.bge_jslots[i];
870 		entry->bge_sc = sc;
871 		entry->bge_buf = ptr;
872 		entry->bge_paddr = paddr;
873 		entry->bge_inuse = 0;
874 		entry->bge_slot = i;
875 		SLIST_INSERT_HEAD(&sc->bge_jfree_listhead, entry, jslot_link);
876 
877 		ptr += BGE_JLEN;
878 		paddr += BGE_JLEN;
879 	}
880 	return 0;
881 }
882 
883 static void
884 bge_free_jumbo_mem(struct bge_softc *sc)
885 {
886 	/* Destroy jumbo RX ring. */
887 	bge_dma_block_free(sc->bge_cdata.bge_rx_jumbo_ring_tag,
888 			   sc->bge_cdata.bge_rx_jumbo_ring_map,
889 			   sc->bge_ldata.bge_rx_jumbo_ring);
890 
891 	/* Destroy jumbo buffer block. */
892 	bge_dma_block_free(sc->bge_cdata.bge_jumbo_tag,
893 			   sc->bge_cdata.bge_jumbo_map,
894 			   sc->bge_ldata.bge_jumbo_buf);
895 }
896 
897 /*
898  * Allocate a jumbo buffer.
899  */
900 static struct bge_jslot *
901 bge_jalloc(struct bge_softc *sc)
902 {
903 	struct bge_jslot *entry;
904 
905 	lwkt_serialize_enter(&sc->bge_jslot_serializer);
906 	entry = SLIST_FIRST(&sc->bge_jfree_listhead);
907 	if (entry) {
908 		SLIST_REMOVE_HEAD(&sc->bge_jfree_listhead, jslot_link);
909 		entry->bge_inuse = 1;
910 	} else {
911 		if_printf(&sc->arpcom.ac_if, "no free jumbo buffers\n");
912 	}
913 	lwkt_serialize_exit(&sc->bge_jslot_serializer);
914 	return(entry);
915 }
916 
917 /*
918  * Adjust usage count on a jumbo buffer.
919  */
920 static void
921 bge_jref(void *arg)
922 {
923 	struct bge_jslot *entry = (struct bge_jslot *)arg;
924 	struct bge_softc *sc = entry->bge_sc;
925 
926 	if (sc == NULL)
927 		panic("bge_jref: can't find softc pointer!");
928 
929 	if (&sc->bge_cdata.bge_jslots[entry->bge_slot] != entry) {
930 		panic("bge_jref: asked to reference buffer "
931 		    "that we don't manage!");
932 	} else if (entry->bge_inuse == 0) {
933 		panic("bge_jref: buffer already free!");
934 	} else {
935 		atomic_add_int(&entry->bge_inuse, 1);
936 	}
937 }
938 
939 /*
940  * Release a jumbo buffer.
941  */
942 static void
943 bge_jfree(void *arg)
944 {
945 	struct bge_jslot *entry = (struct bge_jslot *)arg;
946 	struct bge_softc *sc = entry->bge_sc;
947 
948 	if (sc == NULL)
949 		panic("bge_jfree: can't find softc pointer!");
950 
951 	if (&sc->bge_cdata.bge_jslots[entry->bge_slot] != entry) {
952 		panic("bge_jfree: asked to free buffer that we don't manage!");
953 	} else if (entry->bge_inuse == 0) {
954 		panic("bge_jfree: buffer already free!");
955 	} else {
956 		/*
957 		 * Possible MP race to 0, use the serializer.  The atomic insn
958 		 * is still needed for races against bge_jref().
959 		 */
960 		lwkt_serialize_enter(&sc->bge_jslot_serializer);
961 		atomic_subtract_int(&entry->bge_inuse, 1);
962 		if (entry->bge_inuse == 0) {
963 			SLIST_INSERT_HEAD(&sc->bge_jfree_listhead,
964 					  entry, jslot_link);
965 		}
966 		lwkt_serialize_exit(&sc->bge_jslot_serializer);
967 	}
968 }
969 
970 
971 /*
972  * Intialize a standard receive ring descriptor.
973  */
974 static int
975 bge_newbuf_std(struct bge_softc *sc, int i, int init)
976 {
977 	struct mbuf *m_new = NULL;
978 	bus_dma_segment_t seg;
979 	bus_dmamap_t map;
980 	int error, nsegs;
981 
982 	m_new = m_getcl(init ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
983 	if (m_new == NULL)
984 		return ENOBUFS;
985 	m_new->m_len = m_new->m_pkthdr.len = MCLBYTES;
986 
987 	if ((sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) == 0)
988 		m_adj(m_new, ETHER_ALIGN);
989 
990 	error = bus_dmamap_load_mbuf_segment(sc->bge_cdata.bge_rx_mtag,
991 			sc->bge_cdata.bge_rx_tmpmap, m_new,
992 			&seg, 1, &nsegs, BUS_DMA_NOWAIT);
993 	if (error) {
994 		m_freem(m_new);
995 		return error;
996 	}
997 
998 	if (!init) {
999 		bus_dmamap_sync(sc->bge_cdata.bge_rx_mtag,
1000 				sc->bge_cdata.bge_rx_std_dmamap[i],
1001 				BUS_DMASYNC_POSTREAD);
1002 		bus_dmamap_unload(sc->bge_cdata.bge_rx_mtag,
1003 			sc->bge_cdata.bge_rx_std_dmamap[i]);
1004 	}
1005 
1006 	map = sc->bge_cdata.bge_rx_tmpmap;
1007 	sc->bge_cdata.bge_rx_tmpmap = sc->bge_cdata.bge_rx_std_dmamap[i];
1008 	sc->bge_cdata.bge_rx_std_dmamap[i] = map;
1009 
1010 	sc->bge_cdata.bge_rx_std_chain[i].bge_mbuf = m_new;
1011 	sc->bge_cdata.bge_rx_std_chain[i].bge_paddr = seg.ds_addr;
1012 
1013 	bge_setup_rxdesc_std(sc, i);
1014 	return 0;
1015 }
1016 
1017 static void
1018 bge_setup_rxdesc_std(struct bge_softc *sc, int i)
1019 {
1020 	struct bge_rxchain *rc;
1021 	struct bge_rx_bd *r;
1022 
1023 	rc = &sc->bge_cdata.bge_rx_std_chain[i];
1024 	r = &sc->bge_ldata.bge_rx_std_ring[i];
1025 
1026 	r->bge_addr.bge_addr_lo = BGE_ADDR_LO(rc->bge_paddr);
1027 	r->bge_addr.bge_addr_hi = BGE_ADDR_HI(rc->bge_paddr);
1028 	r->bge_len = rc->bge_mbuf->m_len;
1029 	r->bge_idx = i;
1030 	r->bge_flags = BGE_RXBDFLAG_END;
1031 }
1032 
1033 /*
1034  * Initialize a jumbo receive ring descriptor. This allocates
1035  * a jumbo buffer from the pool managed internally by the driver.
1036  */
1037 static int
1038 bge_newbuf_jumbo(struct bge_softc *sc, int i, int init)
1039 {
1040 	struct mbuf *m_new = NULL;
1041 	struct bge_jslot *buf;
1042 	bus_addr_t paddr;
1043 
1044 	/* Allocate the mbuf. */
1045 	MGETHDR(m_new, init ? MB_WAIT : MB_DONTWAIT, MT_DATA);
1046 	if (m_new == NULL)
1047 		return ENOBUFS;
1048 
1049 	/* Allocate the jumbo buffer */
1050 	buf = bge_jalloc(sc);
1051 	if (buf == NULL) {
1052 		m_freem(m_new);
1053 		return ENOBUFS;
1054 	}
1055 
1056 	/* Attach the buffer to the mbuf. */
1057 	m_new->m_ext.ext_arg = buf;
1058 	m_new->m_ext.ext_buf = buf->bge_buf;
1059 	m_new->m_ext.ext_free = bge_jfree;
1060 	m_new->m_ext.ext_ref = bge_jref;
1061 	m_new->m_ext.ext_size = BGE_JUMBO_FRAMELEN;
1062 
1063 	m_new->m_flags |= M_EXT;
1064 
1065 	m_new->m_data = m_new->m_ext.ext_buf;
1066 	m_new->m_len = m_new->m_pkthdr.len = m_new->m_ext.ext_size;
1067 
1068 	paddr = buf->bge_paddr;
1069 	if ((sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) == 0) {
1070 		m_adj(m_new, ETHER_ALIGN);
1071 		paddr += ETHER_ALIGN;
1072 	}
1073 
1074 	/* Save necessary information */
1075 	sc->bge_cdata.bge_rx_jumbo_chain[i].bge_mbuf = m_new;
1076 	sc->bge_cdata.bge_rx_jumbo_chain[i].bge_paddr = paddr;
1077 
1078 	/* Set up the descriptor. */
1079 	bge_setup_rxdesc_jumbo(sc, i);
1080 	return 0;
1081 }
1082 
1083 static void
1084 bge_setup_rxdesc_jumbo(struct bge_softc *sc, int i)
1085 {
1086 	struct bge_rx_bd *r;
1087 	struct bge_rxchain *rc;
1088 
1089 	r = &sc->bge_ldata.bge_rx_jumbo_ring[i];
1090 	rc = &sc->bge_cdata.bge_rx_jumbo_chain[i];
1091 
1092 	r->bge_addr.bge_addr_lo = BGE_ADDR_LO(rc->bge_paddr);
1093 	r->bge_addr.bge_addr_hi = BGE_ADDR_HI(rc->bge_paddr);
1094 	r->bge_len = rc->bge_mbuf->m_len;
1095 	r->bge_idx = i;
1096 	r->bge_flags = BGE_RXBDFLAG_END|BGE_RXBDFLAG_JUMBO_RING;
1097 }
1098 
1099 static int
1100 bge_init_rx_ring_std(struct bge_softc *sc)
1101 {
1102 	int i, error;
1103 
1104 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1105 		error = bge_newbuf_std(sc, i, 1);
1106 		if (error)
1107 			return error;
1108 	}
1109 
1110 	sc->bge_std = BGE_STD_RX_RING_CNT - 1;
1111 	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bge_std);
1112 
1113 	return(0);
1114 }
1115 
1116 static void
1117 bge_free_rx_ring_std(struct bge_softc *sc)
1118 {
1119 	int i;
1120 
1121 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1122 		struct bge_rxchain *rc = &sc->bge_cdata.bge_rx_std_chain[i];
1123 
1124 		if (rc->bge_mbuf != NULL) {
1125 			bus_dmamap_unload(sc->bge_cdata.bge_rx_mtag,
1126 					  sc->bge_cdata.bge_rx_std_dmamap[i]);
1127 			m_freem(rc->bge_mbuf);
1128 			rc->bge_mbuf = NULL;
1129 		}
1130 		bzero(&sc->bge_ldata.bge_rx_std_ring[i],
1131 		    sizeof(struct bge_rx_bd));
1132 	}
1133 }
1134 
1135 static int
1136 bge_init_rx_ring_jumbo(struct bge_softc *sc)
1137 {
1138 	struct bge_rcb *rcb;
1139 	int i, error;
1140 
1141 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1142 		error = bge_newbuf_jumbo(sc, i, 1);
1143 		if (error)
1144 			return error;
1145 	}
1146 
1147 	sc->bge_jumbo = BGE_JUMBO_RX_RING_CNT - 1;
1148 
1149 	rcb = &sc->bge_ldata.bge_info.bge_jumbo_rx_rcb;
1150 	rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(0, 0);
1151 	CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1152 
1153 	bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
1154 
1155 	return(0);
1156 }
1157 
1158 static void
1159 bge_free_rx_ring_jumbo(struct bge_softc *sc)
1160 {
1161 	int i;
1162 
1163 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1164 		struct bge_rxchain *rc = &sc->bge_cdata.bge_rx_jumbo_chain[i];
1165 
1166 		if (rc->bge_mbuf != NULL) {
1167 			m_freem(rc->bge_mbuf);
1168 			rc->bge_mbuf = NULL;
1169 		}
1170 		bzero(&sc->bge_ldata.bge_rx_jumbo_ring[i],
1171 		    sizeof(struct bge_rx_bd));
1172 	}
1173 }
1174 
1175 static void
1176 bge_free_tx_ring(struct bge_softc *sc)
1177 {
1178 	int i;
1179 
1180 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
1181 		if (sc->bge_cdata.bge_tx_chain[i] != NULL) {
1182 			bus_dmamap_unload(sc->bge_cdata.bge_tx_mtag,
1183 					  sc->bge_cdata.bge_tx_dmamap[i]);
1184 			m_freem(sc->bge_cdata.bge_tx_chain[i]);
1185 			sc->bge_cdata.bge_tx_chain[i] = NULL;
1186 		}
1187 		bzero(&sc->bge_ldata.bge_tx_ring[i],
1188 		    sizeof(struct bge_tx_bd));
1189 	}
1190 }
1191 
1192 static int
1193 bge_init_tx_ring(struct bge_softc *sc)
1194 {
1195 	sc->bge_txcnt = 0;
1196 	sc->bge_tx_saved_considx = 0;
1197 	sc->bge_tx_prodidx = 0;
1198 
1199 	/* Initialize transmit producer index for host-memory send ring. */
1200 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1201 
1202 	/* 5700 b2 errata */
1203 	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
1204 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1205 
1206 	bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1207 	/* 5700 b2 errata */
1208 	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
1209 		bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1210 
1211 	return(0);
1212 }
1213 
1214 static void
1215 bge_setmulti(struct bge_softc *sc)
1216 {
1217 	struct ifnet *ifp;
1218 	struct ifmultiaddr *ifma;
1219 	uint32_t hashes[4] = { 0, 0, 0, 0 };
1220 	int h, i;
1221 
1222 	ifp = &sc->arpcom.ac_if;
1223 
1224 	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
1225 		for (i = 0; i < 4; i++)
1226 			CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0xFFFFFFFF);
1227 		return;
1228 	}
1229 
1230 	/* First, zot all the existing filters. */
1231 	for (i = 0; i < 4; i++)
1232 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0);
1233 
1234 	/* Now program new ones. */
1235 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1236 		if (ifma->ifma_addr->sa_family != AF_LINK)
1237 			continue;
1238 		h = ether_crc32_le(
1239 		    LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1240 		    ETHER_ADDR_LEN) & 0x7f;
1241 		hashes[(h & 0x60) >> 5] |= 1 << (h & 0x1F);
1242 	}
1243 
1244 	for (i = 0; i < 4; i++)
1245 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), hashes[i]);
1246 }
1247 
1248 /*
1249  * Do endian, PCI and DMA initialization. Also check the on-board ROM
1250  * self-test results.
1251  */
1252 static int
1253 bge_chipinit(struct bge_softc *sc)
1254 {
1255 	int i;
1256 	uint32_t dma_rw_ctl;
1257 	uint16_t val;
1258 
1259 	/* Set endian type before we access any non-PCI registers. */
1260 	pci_write_config(sc->bge_dev, BGE_PCI_MISC_CTL,
1261 	    BGE_INIT | sc->bge_pci_miscctl, 4);
1262 
1263 	/* Clear the MAC control register */
1264 	CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
1265 
1266 	/*
1267 	 * Clear the MAC statistics block in the NIC's
1268 	 * internal memory.
1269 	 */
1270 	for (i = BGE_STATS_BLOCK;
1271 	    i < BGE_STATS_BLOCK_END + 1; i += sizeof(uint32_t))
1272 		BGE_MEMWIN_WRITE(sc, i, 0);
1273 
1274 	for (i = BGE_STATUS_BLOCK;
1275 	    i < BGE_STATUS_BLOCK_END + 1; i += sizeof(uint32_t))
1276 		BGE_MEMWIN_WRITE(sc, i, 0);
1277 
1278 	if (sc->bge_chiprev == BGE_CHIPREV_5704_BX) {
1279 		/*
1280 		 * Fix data corruption caused by non-qword write with WB.
1281 		 * Fix master abort in PCI mode.
1282 		 * Fix PCI latency timer.
1283 		 */
1284 		val = pci_read_config(sc->bge_dev, BGE_PCI_MSI_DATA + 2, 2);
1285 		val |= (1 << 10) | (1 << 12) | (1 << 13);
1286 		pci_write_config(sc->bge_dev, BGE_PCI_MSI_DATA + 2, val, 2);
1287 	}
1288 
1289 	/* Set up the PCI DMA control register. */
1290 	dma_rw_ctl = BGE_PCI_READ_CMD | BGE_PCI_WRITE_CMD;
1291 	if (sc->bge_flags & BGE_FLAG_PCIE) {
1292 		/* PCI-E bus */
1293 		/* DMA read watermark not used on PCI-E */
1294 		dma_rw_ctl |= (0x3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1295 	} else if (sc->bge_flags & BGE_FLAG_PCIX) {
1296 		/* PCI-X bus */
1297 		if (sc->bge_asicrev == BGE_ASICREV_BCM5780) {
1298 			dma_rw_ctl |= (0x4 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1299 			    (0x2 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1300 			dma_rw_ctl |= BGE_PCIDMARWCTL_ONEDMA_ATONCE_GLOBAL;
1301 		} else if (sc->bge_asicrev == BGE_ASICREV_BCM5714) {
1302 			dma_rw_ctl |= (0x4 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1303 			    (0x2 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1304 			dma_rw_ctl |= BGE_PCIDMARWCTL_ONEDMA_ATONCE_LOCAL;
1305 		} else if (sc->bge_asicrev == BGE_ASICREV_BCM5703 ||
1306 		    sc->bge_asicrev == BGE_ASICREV_BCM5704) {
1307 			uint32_t rd_wat = 0x7;
1308 			uint32_t clkctl;
1309 
1310 			clkctl = CSR_READ_4(sc, BGE_PCI_CLKCTL) & 0x1f;
1311 			if ((sc->bge_flags & BGE_FLAG_MAXADDR_40BIT) &&
1312 			    sc->bge_asicrev == BGE_ASICREV_BCM5704) {
1313 				dma_rw_ctl |=
1314 				    BGE_PCIDMARWCTL_ONEDMA_ATONCE_LOCAL;
1315 			} else if (clkctl == 0x6 || clkctl == 0x7) {
1316 				dma_rw_ctl |=
1317 				    BGE_PCIDMARWCTL_ONEDMA_ATONCE_GLOBAL;
1318 			}
1319 			if (sc->bge_asicrev == BGE_ASICREV_BCM5703)
1320 				rd_wat = 0x4;
1321 
1322 			dma_rw_ctl |= (rd_wat << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1323 			    (3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1324 			dma_rw_ctl |= BGE_PCIDMARWCTL_ASRT_ALL_BE;
1325 		} else {
1326 			dma_rw_ctl |= (0x3 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1327 			    (0x3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1328 			dma_rw_ctl |= 0xf;
1329 		}
1330 	} else {
1331 		/* Conventional PCI bus */
1332 		dma_rw_ctl |= (0x7 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1333 		    (0x7 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1334 		if (sc->bge_asicrev != BGE_ASICREV_BCM5705 &&
1335 		    sc->bge_asicrev != BGE_ASICREV_BCM5750)
1336 			dma_rw_ctl |= 0xf;
1337 	}
1338 
1339 	if (sc->bge_asicrev == BGE_ASICREV_BCM5703 ||
1340 	    sc->bge_asicrev == BGE_ASICREV_BCM5704) {
1341 		dma_rw_ctl &= ~BGE_PCIDMARWCTL_MINDMA;
1342 	} else if (sc->bge_asicrev == BGE_ASICREV_BCM5700 ||
1343 	    sc->bge_asicrev == BGE_ASICREV_BCM5701) {
1344 		dma_rw_ctl |= BGE_PCIDMARWCTL_USE_MRM |
1345 		    BGE_PCIDMARWCTL_ASRT_ALL_BE;
1346 	}
1347 	pci_write_config(sc->bge_dev, BGE_PCI_DMA_RW_CTL, dma_rw_ctl, 4);
1348 
1349 	/*
1350 	 * Set up general mode register.
1351 	 */
1352 	CSR_WRITE_4(sc, BGE_MODE_CTL, BGE_DMA_SWAP_OPTIONS|
1353 	    BGE_MODECTL_MAC_ATTN_INTR|BGE_MODECTL_HOST_SEND_BDS|
1354 	    BGE_MODECTL_TX_NO_PHDR_CSUM);
1355 
1356 	/*
1357 	 * BCM5701 B5 have a bug causing data corruption when using
1358 	 * 64-bit DMA reads, which can be terminated early and then
1359 	 * completed later as 32-bit accesses, in combination with
1360 	 * certain bridges.
1361 	 */
1362 	if (sc->bge_asicrev == BGE_ASICREV_BCM5701 &&
1363 	    sc->bge_chipid == BGE_CHIPID_BCM5701_B5)
1364 		BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_FORCE_PCI32);
1365 
1366 	/*
1367 	 * Disable memory write invalidate.  Apparently it is not supported
1368 	 * properly by these devices.  Also ensure that INTx isn't disabled,
1369 	 * as these chips need it even when using MSI.
1370 	 */
1371 	PCI_CLRBIT(sc->bge_dev, BGE_PCI_CMD,
1372 	    (PCIM_CMD_MWRICEN | PCIM_CMD_INTxDIS), 4);
1373 
1374 	/* Set the timer prescaler (always 66Mhz) */
1375 	CSR_WRITE_4(sc, BGE_MISC_CFG, 65 << 1/*BGE_32BITTIME_66MHZ*/);
1376 
1377 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
1378 		DELAY(40);	/* XXX */
1379 
1380 		/* Put PHY into ready state */
1381 		BGE_CLRBIT(sc, BGE_MISC_CFG, BGE_MISCCFG_EPHY_IDDQ);
1382 		CSR_READ_4(sc, BGE_MISC_CFG); /* Flush */
1383 		DELAY(40);
1384 	}
1385 
1386 	return(0);
1387 }
1388 
1389 static int
1390 bge_blockinit(struct bge_softc *sc)
1391 {
1392 	struct bge_rcb *rcb;
1393 	bus_size_t vrcb;
1394 	bge_hostaddr taddr;
1395 	uint32_t val;
1396 	int i, limit;
1397 
1398 	/*
1399 	 * Initialize the memory window pointer register so that
1400 	 * we can access the first 32K of internal NIC RAM. This will
1401 	 * allow us to set up the TX send ring RCBs and the RX return
1402 	 * ring RCBs, plus other things which live in NIC memory.
1403 	 */
1404 	CSR_WRITE_4(sc, BGE_PCI_MEMWIN_BASEADDR, 0);
1405 
1406 	/* Note: the BCM5704 has a smaller mbuf space than other chips. */
1407 
1408 	if (!BGE_IS_5705_PLUS(sc)) {
1409 		/* Configure mbuf memory pool */
1410 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_BASEADDR, BGE_BUFFPOOL_1);
1411 		if (sc->bge_asicrev == BGE_ASICREV_BCM5704)
1412 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x10000);
1413 		else
1414 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x18000);
1415 
1416 		/* Configure DMA resource pool */
1417 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_BASEADDR,
1418 		    BGE_DMA_DESCRIPTORS);
1419 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LEN, 0x2000);
1420 	}
1421 
1422 	/* Configure mbuf pool watermarks */
1423 	if (!BGE_IS_5705_PLUS(sc)) {
1424 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x50);
1425 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x20);
1426 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1427 	} else if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
1428 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1429 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x04);
1430 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x10);
1431 	} else {
1432 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1433 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x10);
1434 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1435 	}
1436 
1437 	/* Configure DMA resource watermarks */
1438 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LOWAT, 5);
1439 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_HIWAT, 10);
1440 
1441 	/* Enable buffer manager */
1442 	CSR_WRITE_4(sc, BGE_BMAN_MODE,
1443 	    BGE_BMANMODE_ENABLE|BGE_BMANMODE_LOMBUF_ATTN);
1444 
1445 	/* Poll for buffer manager start indication */
1446 	for (i = 0; i < BGE_TIMEOUT; i++) {
1447 		if (CSR_READ_4(sc, BGE_BMAN_MODE) & BGE_BMANMODE_ENABLE)
1448 			break;
1449 		DELAY(10);
1450 	}
1451 
1452 	if (i == BGE_TIMEOUT) {
1453 		if_printf(&sc->arpcom.ac_if,
1454 			  "buffer manager failed to start\n");
1455 		return(ENXIO);
1456 	}
1457 
1458 	/* Enable flow-through queues */
1459 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
1460 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
1461 
1462 	/* Wait until queue initialization is complete */
1463 	for (i = 0; i < BGE_TIMEOUT; i++) {
1464 		if (CSR_READ_4(sc, BGE_FTQ_RESET) == 0)
1465 			break;
1466 		DELAY(10);
1467 	}
1468 
1469 	if (i == BGE_TIMEOUT) {
1470 		if_printf(&sc->arpcom.ac_if,
1471 			  "flow-through queue init failed\n");
1472 		return(ENXIO);
1473 	}
1474 
1475 	/*
1476 	 * Summary of rings supported by the controller:
1477 	 *
1478 	 * Standard Receive Producer Ring
1479 	 * - This ring is used to feed receive buffers for "standard"
1480 	 *   sized frames (typically 1536 bytes) to the controller.
1481 	 *
1482 	 * Jumbo Receive Producer Ring
1483 	 * - This ring is used to feed receive buffers for jumbo sized
1484 	 *   frames (i.e. anything bigger than the "standard" frames)
1485 	 *   to the controller.
1486 	 *
1487 	 * Mini Receive Producer Ring
1488 	 * - This ring is used to feed receive buffers for "mini"
1489 	 *   sized frames to the controller.
1490 	 * - This feature required external memory for the controller
1491 	 *   but was never used in a production system.  Should always
1492 	 *   be disabled.
1493 	 *
1494 	 * Receive Return Ring
1495 	 * - After the controller has placed an incoming frame into a
1496 	 *   receive buffer that buffer is moved into a receive return
1497 	 *   ring.  The driver is then responsible to passing the
1498 	 *   buffer up to the stack.  Many versions of the controller
1499 	 *   support multiple RR rings.
1500 	 *
1501 	 * Send Ring
1502 	 * - This ring is used for outgoing frames.  Many versions of
1503 	 *   the controller support multiple send rings.
1504 	 */
1505 
1506 	/* Initialize the standard receive producer ring control block. */
1507 	rcb = &sc->bge_ldata.bge_info.bge_std_rx_rcb;
1508 	rcb->bge_hostaddr.bge_addr_lo =
1509 	    BGE_ADDR_LO(sc->bge_ldata.bge_rx_std_ring_paddr);
1510 	rcb->bge_hostaddr.bge_addr_hi =
1511 	    BGE_ADDR_HI(sc->bge_ldata.bge_rx_std_ring_paddr);
1512 	if (BGE_IS_5705_PLUS(sc)) {
1513 		/*
1514 		 * Bits 31-16: Programmable ring size (512, 256, 128, 64, 32)
1515 		 * Bits 15-2 : Reserved (should be 0)
1516 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
1517 		 * Bit 0     : Reserved
1518 		 */
1519 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(512, 0);
1520 	} else {
1521 		/*
1522 		 * Ring size is always XXX entries
1523 		 * Bits 31-16: Maximum RX frame size
1524 		 * Bits 15-2 : Reserved (should be 0)
1525 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
1526 		 * Bit 0     : Reserved
1527 		 */
1528 		rcb->bge_maxlen_flags =
1529 		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN, 0);
1530 	}
1531 	rcb->bge_nicaddr = BGE_STD_RX_RINGS;
1532 	/* Write the standard receive producer ring control block. */
1533 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_HI, rcb->bge_hostaddr.bge_addr_hi);
1534 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_LO, rcb->bge_hostaddr.bge_addr_lo);
1535 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1536 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_NICADDR, rcb->bge_nicaddr);
1537 	/* Reset the standard receive producer ring producer index. */
1538 	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, 0);
1539 
1540 	/*
1541 	 * Initialize the jumbo RX producer ring control
1542 	 * block.  We set the 'ring disabled' bit in the
1543 	 * flags field until we're actually ready to start
1544 	 * using this ring (i.e. once we set the MTU
1545 	 * high enough to require it).
1546 	 */
1547 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
1548 		rcb = &sc->bge_ldata.bge_info.bge_jumbo_rx_rcb;
1549 		/* Get the jumbo receive producer ring RCB parameters. */
1550 		rcb->bge_hostaddr.bge_addr_lo =
1551 		    BGE_ADDR_LO(sc->bge_ldata.bge_rx_jumbo_ring_paddr);
1552 		rcb->bge_hostaddr.bge_addr_hi =
1553 		    BGE_ADDR_HI(sc->bge_ldata.bge_rx_jumbo_ring_paddr);
1554 		rcb->bge_maxlen_flags =
1555 		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN,
1556 		    BGE_RCB_FLAG_RING_DISABLED);
1557 		rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS;
1558 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_HI,
1559 		    rcb->bge_hostaddr.bge_addr_hi);
1560 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_LO,
1561 		    rcb->bge_hostaddr.bge_addr_lo);
1562 		/* Program the jumbo receive producer ring RCB parameters. */
1563 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS,
1564 		    rcb->bge_maxlen_flags);
1565 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_NICADDR, rcb->bge_nicaddr);
1566 		/* Reset the jumbo receive producer ring producer index. */
1567 		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, 0);
1568 	}
1569 
1570 	/* Disable the mini receive producer ring RCB. */
1571 	if (BGE_IS_5700_FAMILY(sc)) {
1572 		rcb = &sc->bge_ldata.bge_info.bge_mini_rx_rcb;
1573 		rcb->bge_maxlen_flags =
1574 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED);
1575 		CSR_WRITE_4(sc, BGE_RX_MINI_RCB_MAXLEN_FLAGS,
1576 		    rcb->bge_maxlen_flags);
1577 		/* Reset the mini receive producer ring producer index. */
1578 		bge_writembx(sc, BGE_MBX_RX_MINI_PROD_LO, 0);
1579 	}
1580 
1581 	/* Choose de-pipeline mode for BCM5906 A0, A1 and A2. */
1582 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
1583 	    (sc->bge_chipid == BGE_CHIPID_BCM5906_A0 ||
1584 	     sc->bge_chipid == BGE_CHIPID_BCM5906_A1 ||
1585 	     sc->bge_chipid == BGE_CHIPID_BCM5906_A2)) {
1586 		CSR_WRITE_4(sc, BGE_ISO_PKT_TX,
1587 		    (CSR_READ_4(sc, BGE_ISO_PKT_TX) & ~3) | 2);
1588 	}
1589 
1590 	/*
1591 	 * The BD ring replenish thresholds control how often the
1592 	 * hardware fetches new BD's from the producer rings in host
1593 	 * memory.  Setting the value too low on a busy system can
1594 	 * starve the hardware and recue the throughpout.
1595 	 *
1596 	 * Set the BD ring replentish thresholds. The recommended
1597 	 * values are 1/8th the number of descriptors allocated to
1598 	 * each ring.
1599 	 */
1600 	if (BGE_IS_5705_PLUS(sc))
1601 		val = 8;
1602 	else
1603 		val = BGE_STD_RX_RING_CNT / 8;
1604 	CSR_WRITE_4(sc, BGE_RBDI_STD_REPL_THRESH, val);
1605 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
1606 		CSR_WRITE_4(sc, BGE_RBDI_JUMBO_REPL_THRESH,
1607 		    BGE_JUMBO_RX_RING_CNT/8);
1608 	}
1609 
1610 	/*
1611 	 * Disable all send rings by setting the 'ring disabled' bit
1612 	 * in the flags field of all the TX send ring control blocks,
1613 	 * located in NIC memory.
1614 	 */
1615 	if (!BGE_IS_5705_PLUS(sc)) {
1616 		/* 5700 to 5704 had 16 send rings. */
1617 		limit = BGE_TX_RINGS_EXTSSRAM_MAX;
1618 	} else {
1619 		limit = 1;
1620 	}
1621 	vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1622 	for (i = 0; i < limit; i++) {
1623 		RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1624 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED));
1625 		RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1626 		vrcb += sizeof(struct bge_rcb);
1627 	}
1628 
1629 	/* Configure send ring RCB 0 (we use only the first ring) */
1630 	vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1631 	BGE_HOSTADDR(taddr, sc->bge_ldata.bge_tx_ring_paddr);
1632 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1633 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1634 	RCB_WRITE_4(sc, vrcb, bge_nicaddr,
1635 	    BGE_NIC_TXRING_ADDR(0, BGE_TX_RING_CNT));
1636 	RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1637 	    BGE_RCB_MAXLEN_FLAGS(BGE_TX_RING_CNT, 0));
1638 
1639 	/*
1640 	 * Disable all receive return rings by setting the
1641 	 * 'ring diabled' bit in the flags field of all the receive
1642 	 * return ring control blocks, located in NIC memory.
1643 	 */
1644 	if (!BGE_IS_5705_PLUS(sc))
1645 		limit = BGE_RX_RINGS_MAX;
1646 	else if (sc->bge_asicrev == BGE_ASICREV_BCM5755)
1647 		limit = 4;
1648 	else
1649 		limit = 1;
1650 	/* Disable all receive return rings. */
1651 	vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1652 	for (i = 0; i < limit; i++) {
1653 		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, 0);
1654 		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, 0);
1655 		RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1656 		    BGE_RCB_FLAG_RING_DISABLED);
1657 		RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1658 		bge_writembx(sc, BGE_MBX_RX_CONS0_LO +
1659 		    (i * (sizeof(uint64_t))), 0);
1660 		vrcb += sizeof(struct bge_rcb);
1661 	}
1662 
1663 	/*
1664 	 * Set up receive return ring 0.  Note that the NIC address
1665 	 * for RX return rings is 0x0.  The return rings live entirely
1666 	 * within the host, so the nicaddr field in the RCB isn't used.
1667 	 */
1668 	vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1669 	BGE_HOSTADDR(taddr, sc->bge_ldata.bge_rx_return_ring_paddr);
1670 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1671 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1672 	RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1673 	RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1674 	    BGE_RCB_MAXLEN_FLAGS(sc->bge_return_ring_cnt, 0));
1675 
1676 	/* Set random backoff seed for TX */
1677 	CSR_WRITE_4(sc, BGE_TX_RANDOM_BACKOFF,
1678 	    sc->arpcom.ac_enaddr[0] + sc->arpcom.ac_enaddr[1] +
1679 	    sc->arpcom.ac_enaddr[2] + sc->arpcom.ac_enaddr[3] +
1680 	    sc->arpcom.ac_enaddr[4] + sc->arpcom.ac_enaddr[5] +
1681 	    BGE_TX_BACKOFF_SEED_MASK);
1682 
1683 	/* Set inter-packet gap */
1684 	CSR_WRITE_4(sc, BGE_TX_LENGTHS, 0x2620);
1685 
1686 	/*
1687 	 * Specify which ring to use for packets that don't match
1688 	 * any RX rules.
1689 	 */
1690 	CSR_WRITE_4(sc, BGE_RX_RULES_CFG, 0x08);
1691 
1692 	/*
1693 	 * Configure number of RX lists. One interrupt distribution
1694 	 * list, sixteen active lists, one bad frames class.
1695 	 */
1696 	CSR_WRITE_4(sc, BGE_RXLP_CFG, 0x181);
1697 
1698 	/* Inialize RX list placement stats mask. */
1699 	CSR_WRITE_4(sc, BGE_RXLP_STATS_ENABLE_MASK, 0x007FFFFF);
1700 	CSR_WRITE_4(sc, BGE_RXLP_STATS_CTL, 0x1);
1701 
1702 	/* Disable host coalescing until we get it set up */
1703 	CSR_WRITE_4(sc, BGE_HCC_MODE, 0x00000000);
1704 
1705 	/* Poll to make sure it's shut down. */
1706 	for (i = 0; i < BGE_TIMEOUT; i++) {
1707 		if (!(CSR_READ_4(sc, BGE_HCC_MODE) & BGE_HCCMODE_ENABLE))
1708 			break;
1709 		DELAY(10);
1710 	}
1711 
1712 	if (i == BGE_TIMEOUT) {
1713 		if_printf(&sc->arpcom.ac_if,
1714 			  "host coalescing engine failed to idle\n");
1715 		return(ENXIO);
1716 	}
1717 
1718 	/* Set up host coalescing defaults */
1719 	CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, sc->bge_rx_coal_ticks);
1720 	CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS, sc->bge_tx_coal_ticks);
1721 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, sc->bge_rx_coal_bds);
1722 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS, sc->bge_tx_coal_bds);
1723 	if (!BGE_IS_5705_PLUS(sc)) {
1724 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS_INT,
1725 		    sc->bge_rx_coal_ticks_int);
1726 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS_INT,
1727 		    sc->bge_tx_coal_ticks_int);
1728 	}
1729 	/*
1730 	 * NOTE:
1731 	 * The datasheet (57XX-PG105-R) says BCM5705+ do not
1732 	 * have following two registers; obviously it is wrong.
1733 	 */
1734 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT, sc->bge_rx_coal_bds_int);
1735 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT, sc->bge_tx_coal_bds_int);
1736 
1737 	/* Set up address of statistics block */
1738 	if (!BGE_IS_5705_PLUS(sc)) {
1739 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_HI,
1740 		    BGE_ADDR_HI(sc->bge_ldata.bge_stats_paddr));
1741 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_LO,
1742 		    BGE_ADDR_LO(sc->bge_ldata.bge_stats_paddr));
1743 
1744 		CSR_WRITE_4(sc, BGE_HCC_STATS_BASEADDR, BGE_STATS_BLOCK);
1745 		CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_BASEADDR, BGE_STATUS_BLOCK);
1746 		CSR_WRITE_4(sc, BGE_HCC_STATS_TICKS, sc->bge_stat_ticks);
1747 	}
1748 
1749 	/* Set up address of status block */
1750 	bzero(sc->bge_ldata.bge_status_block, BGE_STATUS_BLK_SZ);
1751 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_HI,
1752 	    BGE_ADDR_HI(sc->bge_ldata.bge_status_block_paddr));
1753 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_LO,
1754 	    BGE_ADDR_LO(sc->bge_ldata.bge_status_block_paddr));
1755 
1756 	/*
1757 	 * Set up status block partail update size.
1758 	 *
1759 	 * Because only single TX ring, RX produce ring and Rx return ring
1760 	 * are used, ask device to update only minimum part of status block
1761 	 * except for BCM5700 AX/BX, whose status block partial update size
1762 	 * can't be configured.
1763 	 */
1764 	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
1765 	    sc->bge_chipid != BGE_CHIPID_BCM5700_C0) {
1766 		/* XXX Actually reserved on BCM5700 AX/BX */
1767 		val = BGE_STATBLKSZ_FULL;
1768 	} else {
1769 		val = BGE_STATBLKSZ_32BYTE;
1770 	}
1771 #if 0
1772 	/*
1773 	 * Does not seem to have visible effect in both
1774 	 * bulk data (1472B UDP datagram) and tiny data
1775 	 * (18B UDP datagram) TX tests.
1776 	 */
1777 	if (!BGE_IS_CRIPPLED(sc))
1778 		val |= BGE_HCCMODE_CLRTICK_TX;
1779 #endif
1780 
1781 	/* Turn on host coalescing state machine */
1782 	CSR_WRITE_4(sc, BGE_HCC_MODE, val | BGE_HCCMODE_ENABLE);
1783 
1784 	/* Turn on RX BD completion state machine and enable attentions */
1785 	CSR_WRITE_4(sc, BGE_RBDC_MODE,
1786 	    BGE_RBDCMODE_ENABLE|BGE_RBDCMODE_ATTN);
1787 
1788 	/* Turn on RX list placement state machine */
1789 	CSR_WRITE_4(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
1790 
1791 	/* Turn on RX list selector state machine. */
1792 	if (!BGE_IS_5705_PLUS(sc))
1793 		CSR_WRITE_4(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
1794 
1795 	val = BGE_MACMODE_TXDMA_ENB | BGE_MACMODE_RXDMA_ENB |
1796 	    BGE_MACMODE_RX_STATS_CLEAR | BGE_MACMODE_TX_STATS_CLEAR |
1797 	    BGE_MACMODE_RX_STATS_ENB | BGE_MACMODE_TX_STATS_ENB |
1798 	    BGE_MACMODE_FRMHDR_DMA_ENB;
1799 
1800 	if (sc->bge_flags & BGE_FLAG_TBI)
1801 		val |= BGE_PORTMODE_TBI;
1802 	else if (sc->bge_flags & BGE_FLAG_MII_SERDES)
1803 		val |= BGE_PORTMODE_GMII;
1804 	else
1805 		val |= BGE_PORTMODE_MII;
1806 
1807 	/* Turn on DMA, clear stats */
1808 	CSR_WRITE_4(sc, BGE_MAC_MODE, val);
1809 
1810 	/* Set misc. local control, enable interrupts on attentions */
1811 	CSR_WRITE_4(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_ONATTN);
1812 
1813 #ifdef notdef
1814 	/* Assert GPIO pins for PHY reset */
1815 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUT0|
1816 	    BGE_MLC_MISCIO_OUT1|BGE_MLC_MISCIO_OUT2);
1817 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUTEN0|
1818 	    BGE_MLC_MISCIO_OUTEN1|BGE_MLC_MISCIO_OUTEN2);
1819 #endif
1820 
1821 	/* Turn on DMA completion state machine */
1822 	if (!BGE_IS_5705_PLUS(sc))
1823 		CSR_WRITE_4(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
1824 
1825 	/* Turn on write DMA state machine */
1826 	val = BGE_WDMAMODE_ENABLE|BGE_WDMAMODE_ALL_ATTNS;
1827 	if (BGE_IS_5755_PLUS(sc)) {
1828 		/* Enable host coalescing bug fix. */
1829 		val |= BGE_WDMAMODE_STATUS_TAG_FIX;
1830 	}
1831 	if (sc->bge_asicrev == BGE_ASICREV_BCM5785) {
1832 		/* Request larger DMA burst size to get better performance. */
1833 		val |= BGE_WDMAMODE_BURST_ALL_DATA;
1834 	}
1835 	CSR_WRITE_4(sc, BGE_WDMA_MODE, val);
1836 	DELAY(40);
1837 
1838 	if (sc->bge_asicrev == BGE_ASICREV_BCM5761 ||
1839 	    sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
1840 	    sc->bge_asicrev == BGE_ASICREV_BCM5785 ||
1841 	    sc->bge_asicrev == BGE_ASICREV_BCM57780) {
1842 		/*
1843 		 * Enable fix for read DMA FIFO overruns.
1844 		 * The fix is to limit the number of RX BDs
1845 		 * the hardware would fetch at a fime.
1846 		 */
1847 		val = CSR_READ_4(sc, BGE_RDMA_RSRVCTRL);
1848 		CSR_WRITE_4(sc, BGE_RDMA_RSRVCTRL,
1849 		    val| BGE_RDMA_RSRVCTRL_FIFO_OFLW_FIX);
1850 	}
1851 
1852 	/* Turn on read DMA state machine */
1853 	val = BGE_RDMAMODE_ENABLE | BGE_RDMAMODE_ALL_ATTNS;
1854         if (sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
1855             sc->bge_asicrev == BGE_ASICREV_BCM5785 ||
1856             sc->bge_asicrev == BGE_ASICREV_BCM57780)
1857 		val |= BGE_RDMAMODE_BD_SBD_CRPT_ATTN |
1858                   BGE_RDMAMODE_MBUF_RBD_CRPT_ATTN |
1859                   BGE_RDMAMODE_MBUF_SBD_CRPT_ATTN;
1860 	if (sc->bge_flags & BGE_FLAG_PCIE)
1861 		val |= BGE_RDMAMODE_FIFO_LONG_BURST;
1862 	if (sc->bge_flags & BGE_FLAG_TSO)
1863 		val |= BGE_RDMAMODE_TSO4_ENABLE;
1864 	CSR_WRITE_4(sc, BGE_RDMA_MODE, val);
1865 	DELAY(40);
1866 
1867 	/* Turn on RX data completion state machine */
1868 	CSR_WRITE_4(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
1869 
1870 	/* Turn on RX BD initiator state machine */
1871 	CSR_WRITE_4(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
1872 
1873 	/* Turn on RX data and RX BD initiator state machine */
1874 	CSR_WRITE_4(sc, BGE_RDBDI_MODE, BGE_RDBDIMODE_ENABLE);
1875 
1876 	/* Turn on Mbuf cluster free state machine */
1877 	if (!BGE_IS_5705_PLUS(sc))
1878 		CSR_WRITE_4(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
1879 
1880 	/* Turn on send BD completion state machine */
1881 	CSR_WRITE_4(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
1882 
1883 	/* Turn on send data completion state machine */
1884 	val = BGE_SDCMODE_ENABLE;
1885 	if (sc->bge_asicrev == BGE_ASICREV_BCM5761)
1886 		val |= BGE_SDCMODE_CDELAY;
1887 	CSR_WRITE_4(sc, BGE_SDC_MODE, val);
1888 
1889 	/* Turn on send data initiator state machine */
1890 	if (sc->bge_flags & BGE_FLAG_TSO)
1891 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE |
1892 		    BGE_SDIMODE_HW_LSO_PRE_DMA);
1893 	else
1894 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
1895 
1896 	/* Turn on send BD initiator state machine */
1897 	CSR_WRITE_4(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
1898 
1899 	/* Turn on send BD selector state machine */
1900 	CSR_WRITE_4(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
1901 
1902 	CSR_WRITE_4(sc, BGE_SDI_STATS_ENABLE_MASK, 0x007FFFFF);
1903 	CSR_WRITE_4(sc, BGE_SDI_STATS_CTL,
1904 	    BGE_SDISTATSCTL_ENABLE|BGE_SDISTATSCTL_FASTER);
1905 
1906 	/* ack/clear link change events */
1907 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
1908 	    BGE_MACSTAT_CFG_CHANGED|BGE_MACSTAT_MI_COMPLETE|
1909 	    BGE_MACSTAT_LINK_CHANGED);
1910 	CSR_WRITE_4(sc, BGE_MI_STS, 0);
1911 
1912 	/*
1913 	 * Enable attention when the link has changed state for
1914 	 * devices that use auto polling.
1915 	 */
1916 	if (sc->bge_flags & BGE_FLAG_TBI) {
1917 		CSR_WRITE_4(sc, BGE_MI_STS, BGE_MISTS_LINK);
1918  	} else {
1919 		if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
1920 			CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode);
1921 			DELAY(80);
1922 		}
1923 		if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
1924 		    sc->bge_chipid != BGE_CHIPID_BCM5700_B2) {
1925 			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
1926 			    BGE_EVTENB_MI_INTERRUPT);
1927 		}
1928 	}
1929 
1930 	/*
1931 	 * Clear any pending link state attention.
1932 	 * Otherwise some link state change events may be lost until attention
1933 	 * is cleared by bge_intr() -> bge_softc.bge_link_upd() sequence.
1934 	 * It's not necessary on newer BCM chips - perhaps enabling link
1935 	 * state change attentions implies clearing pending attention.
1936 	 */
1937 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
1938 	    BGE_MACSTAT_CFG_CHANGED|BGE_MACSTAT_MI_COMPLETE|
1939 	    BGE_MACSTAT_LINK_CHANGED);
1940 
1941 	/* Enable link state change attentions. */
1942 	BGE_SETBIT(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_LINK_CHANGED);
1943 
1944 	return(0);
1945 }
1946 
1947 /*
1948  * Probe for a Broadcom chip. Check the PCI vendor and device IDs
1949  * against our list and return its name if we find a match. Note
1950  * that since the Broadcom controller contains VPD support, we
1951  * can get the device name string from the controller itself instead
1952  * of the compiled-in string. This is a little slow, but it guarantees
1953  * we'll always announce the right product name.
1954  */
1955 static int
1956 bge_probe(device_t dev)
1957 {
1958 	const struct bge_type *t;
1959 	uint16_t product, vendor;
1960 
1961 	product = pci_get_device(dev);
1962 	vendor = pci_get_vendor(dev);
1963 
1964 	for (t = bge_devs; t->bge_name != NULL; t++) {
1965 		if (vendor == t->bge_vid && product == t->bge_did)
1966 			break;
1967 	}
1968 	if (t->bge_name == NULL)
1969 		return(ENXIO);
1970 
1971 	device_set_desc(dev, t->bge_name);
1972 	return(0);
1973 }
1974 
1975 static int
1976 bge_attach(device_t dev)
1977 {
1978 	struct ifnet *ifp;
1979 	struct bge_softc *sc;
1980 	uint32_t hwcfg = 0, misccfg;
1981 	int error = 0, rid, capmask;
1982 	uint8_t ether_addr[ETHER_ADDR_LEN];
1983 	uint16_t product, vendor;
1984 	driver_intr_t *intr_func;
1985 	uintptr_t mii_priv = 0;
1986 	u_int intr_flags;
1987 	int msi_enable;
1988 
1989 	sc = device_get_softc(dev);
1990 	sc->bge_dev = dev;
1991 	callout_init_mp(&sc->bge_stat_timer);
1992 	lwkt_serialize_init(&sc->bge_jslot_serializer);
1993 
1994 	product = pci_get_device(dev);
1995 	vendor = pci_get_vendor(dev);
1996 
1997 #ifndef BURN_BRIDGES
1998 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1999 		uint32_t irq, mem;
2000 
2001 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
2002 		mem = pci_read_config(dev, BGE_PCI_BAR0, 4);
2003 
2004 		device_printf(dev, "chip is in D%d power mode "
2005 		    "-- setting to D0\n", pci_get_powerstate(dev));
2006 
2007 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
2008 
2009 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
2010 		pci_write_config(dev, BGE_PCI_BAR0, mem, 4);
2011 	}
2012 #endif	/* !BURN_BRIDGE */
2013 
2014 	/*
2015 	 * Map control/status registers.
2016 	 */
2017 	pci_enable_busmaster(dev);
2018 
2019 	rid = BGE_PCI_BAR0;
2020 	sc->bge_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
2021 	    RF_ACTIVE);
2022 
2023 	if (sc->bge_res == NULL) {
2024 		device_printf(dev, "couldn't map memory\n");
2025 		return ENXIO;
2026 	}
2027 
2028 	sc->bge_btag = rman_get_bustag(sc->bge_res);
2029 	sc->bge_bhandle = rman_get_bushandle(sc->bge_res);
2030 
2031 	/* Save various chip information */
2032 	sc->bge_chipid =
2033 	    pci_read_config(dev, BGE_PCI_MISC_CTL, 4) >>
2034 	    BGE_PCIMISCCTL_ASICREV_SHIFT;
2035 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_USE_PRODID_REG) {
2036 		/* All chips, which use BGE_PCI_PRODID_ASICREV, have CPMU */
2037 		sc->bge_flags |= BGE_FLAG_CPMU;
2038 		sc->bge_chipid = pci_read_config(dev, BGE_PCI_PRODID_ASICREV, 4);
2039 	}
2040 	sc->bge_asicrev = BGE_ASICREV(sc->bge_chipid);
2041 	sc->bge_chiprev = BGE_CHIPREV(sc->bge_chipid);
2042 
2043 	/* Save chipset family. */
2044 	switch (sc->bge_asicrev) {
2045 	case BGE_ASICREV_BCM5755:
2046 	case BGE_ASICREV_BCM5761:
2047 	case BGE_ASICREV_BCM5784:
2048 	case BGE_ASICREV_BCM5785:
2049 	case BGE_ASICREV_BCM5787:
2050 	case BGE_ASICREV_BCM57780:
2051 	    sc->bge_flags |= BGE_FLAG_5755_PLUS | BGE_FLAG_575X_PLUS |
2052 		BGE_FLAG_5705_PLUS;
2053 	    break;
2054 
2055 	case BGE_ASICREV_BCM5700:
2056 	case BGE_ASICREV_BCM5701:
2057 	case BGE_ASICREV_BCM5703:
2058 	case BGE_ASICREV_BCM5704:
2059 		sc->bge_flags |= BGE_FLAG_5700_FAMILY | BGE_FLAG_JUMBO;
2060 		break;
2061 
2062 	case BGE_ASICREV_BCM5714_A0:
2063 	case BGE_ASICREV_BCM5780:
2064 	case BGE_ASICREV_BCM5714:
2065 		sc->bge_flags |= BGE_FLAG_5714_FAMILY;
2066 		/* Fall through */
2067 
2068 	case BGE_ASICREV_BCM5750:
2069 	case BGE_ASICREV_BCM5752:
2070 	case BGE_ASICREV_BCM5906:
2071 		sc->bge_flags |= BGE_FLAG_575X_PLUS;
2072 		/* Fall through */
2073 
2074 	case BGE_ASICREV_BCM5705:
2075 		sc->bge_flags |= BGE_FLAG_5705_PLUS;
2076 		break;
2077 	}
2078 
2079 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
2080 		sc->bge_flags |= BGE_FLAG_NO_EEPROM;
2081 
2082 	misccfg = CSR_READ_4(sc, BGE_MISC_CFG) & BGE_MISCCFG_BOARD_ID_MASK;
2083 	if (sc->bge_asicrev == BGE_ASICREV_BCM5705 &&
2084 	    (misccfg == BGE_MISCCFG_BOARD_ID_5788 ||
2085 	     misccfg == BGE_MISCCFG_BOARD_ID_5788M))
2086 		sc->bge_flags |= BGE_FLAG_5788;
2087 
2088 	/* BCM5755 or higher and BCM5906 have short DMA bug. */
2089 	if (BGE_IS_5755_PLUS(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5906)
2090 		sc->bge_flags |= BGE_FLAG_SHORTDMA;
2091 
2092   	/*
2093 	 * Check if this is a PCI-X or PCI Express device.
2094   	 */
2095 	if (BGE_IS_5705_PLUS(sc)) {
2096 		if (pci_is_pcie(dev)) {
2097 			sc->bge_flags |= BGE_FLAG_PCIE;
2098 			sc->bge_pciecap = pci_get_pciecap_ptr(sc->bge_dev);
2099 			pcie_set_max_readrq(dev, PCIEM_DEVCTL_MAX_READRQ_4096);
2100 		}
2101 	} else {
2102 		/*
2103 		 * Check if the device is in PCI-X Mode.
2104 		 * (This bit is not valid on PCI Express controllers.)
2105 		 */
2106 		if ((pci_read_config(sc->bge_dev, BGE_PCI_PCISTATE, 4) &
2107 		    BGE_PCISTATE_PCI_BUSMODE) == 0) {
2108 			sc->bge_flags |= BGE_FLAG_PCIX;
2109 			sc->bge_pcixcap = pci_get_pcixcap_ptr(sc->bge_dev);
2110 			sc->bge_mbox_reorder = device_getenv_int(sc->bge_dev,
2111 			    "mbox_reorder", 0);
2112 		}
2113  	}
2114 	device_printf(dev, "CHIP ID 0x%08x; "
2115 		      "ASIC REV 0x%02x; CHIP REV 0x%02x; %s\n",
2116 		      sc->bge_chipid, sc->bge_asicrev, sc->bge_chiprev,
2117 		      (sc->bge_flags & BGE_FLAG_PCIX) ? "PCI-X"
2118 		      : ((sc->bge_flags & BGE_FLAG_PCIE) ?
2119 			"PCI-E" : "PCI"));
2120 
2121 	/*
2122 	 * The 40bit DMA bug applies to the 5714/5715 controllers and is
2123 	 * not actually a MAC controller bug but an issue with the embedded
2124 	 * PCIe to PCI-X bridge in the device. Use 40bit DMA workaround.
2125 	 */
2126 	if ((sc->bge_flags & BGE_FLAG_PCIX) &&
2127 	    (BGE_IS_5714_FAMILY(sc) || device_getenv_int(dev, "dma40b", 0)))
2128 		sc->bge_flags |= BGE_FLAG_MAXADDR_40BIT;
2129 
2130 	/*
2131 	 * When using the BCM5701 in PCI-X mode, data corruption has
2132 	 * been observed in the first few bytes of some received packets.
2133 	 * Aligning the packet buffer in memory eliminates the corruption.
2134 	 * Unfortunately, this misaligns the packet payloads.  On platforms
2135 	 * which do not support unaligned accesses, we will realign the
2136 	 * payloads by copying the received packets.
2137 	 */
2138 	if (sc->bge_asicrev == BGE_ASICREV_BCM5701 &&
2139 	    (sc->bge_flags & BGE_FLAG_PCIX))
2140 		sc->bge_flags |= BGE_FLAG_RX_ALIGNBUG;
2141 
2142 	if (!BGE_IS_CRIPPLED(sc)) {
2143 		if (device_getenv_int(dev, "status_tag", 1)) {
2144 			sc->bge_flags |= BGE_FLAG_STATUS_TAG;
2145 			sc->bge_pci_miscctl = BGE_PCIMISCCTL_TAGGED_STATUS;
2146 			if (bootverbose)
2147 				device_printf(dev, "enable status tag\n");
2148 		}
2149 	}
2150 
2151 	if (BGE_IS_5755_PLUS(sc)) {
2152 		/*
2153 		 * BCM5754 and BCM5787 shares the same ASIC id so
2154 		 * explicit device id check is required.
2155 		 * Due to unknown reason TSO does not work on BCM5755M.
2156 		 */
2157 		if (product != PCI_PRODUCT_BROADCOM_BCM5754 &&
2158 		    product != PCI_PRODUCT_BROADCOM_BCM5754M &&
2159 		    product != PCI_PRODUCT_BROADCOM_BCM5755M)
2160 			sc->bge_flags |= BGE_FLAG_TSO;
2161 	}
2162 
2163 	/*
2164 	 * Set various PHY quirk flags.
2165 	 */
2166 
2167 	if ((sc->bge_asicrev == BGE_ASICREV_BCM5700 ||
2168 	     sc->bge_asicrev == BGE_ASICREV_BCM5701) &&
2169 	    pci_get_subvendor(dev) == PCI_VENDOR_DELL)
2170 		mii_priv |= BRGPHY_FLAG_NO_3LED;
2171 
2172 	capmask = MII_CAPMASK_DEFAULT;
2173 	if ((sc->bge_asicrev == BGE_ASICREV_BCM5703 &&
2174 	     (misccfg == 0x4000 || misccfg == 0x8000)) ||
2175 	    (sc->bge_asicrev == BGE_ASICREV_BCM5705 &&
2176 	     vendor == PCI_VENDOR_BROADCOM &&
2177 	     (product == PCI_PRODUCT_BROADCOM_BCM5901 ||
2178 	      product == PCI_PRODUCT_BROADCOM_BCM5901A2 ||
2179 	      product == PCI_PRODUCT_BROADCOM_BCM5705F)) ||
2180 	    (vendor == PCI_VENDOR_BROADCOM &&
2181 	     (product == PCI_PRODUCT_BROADCOM_BCM5751F ||
2182 	      product == PCI_PRODUCT_BROADCOM_BCM5753F ||
2183 	      product == PCI_PRODUCT_BROADCOM_BCM5787F)) ||
2184 	    product == PCI_PRODUCT_BROADCOM_BCM57790 ||
2185 	    sc->bge_asicrev == BGE_ASICREV_BCM5906) {
2186 		/* 10/100 only */
2187 		capmask &= ~BMSR_EXTSTAT;
2188 	}
2189 
2190 	mii_priv |= BRGPHY_FLAG_WIRESPEED;
2191 	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 ||
2192 	    (sc->bge_asicrev == BGE_ASICREV_BCM5705 &&
2193 	     (sc->bge_chipid != BGE_CHIPID_BCM5705_A0 &&
2194 	      sc->bge_chipid != BGE_CHIPID_BCM5705_A1)) ||
2195 	    sc->bge_asicrev == BGE_ASICREV_BCM5906)
2196 		mii_priv &= ~BRGPHY_FLAG_WIRESPEED;
2197 
2198 	if (sc->bge_chipid == BGE_CHIPID_BCM5701_A0 ||
2199 	    sc->bge_chipid == BGE_CHIPID_BCM5701_B0)
2200 		mii_priv |= BRGPHY_FLAG_CRC_BUG;
2201 
2202 	if (sc->bge_chiprev == BGE_CHIPREV_5703_AX ||
2203 	    sc->bge_chiprev == BGE_CHIPREV_5704_AX)
2204 		mii_priv |= BRGPHY_FLAG_ADC_BUG;
2205 
2206 	if (sc->bge_chipid == BGE_CHIPID_BCM5704_A0)
2207 		mii_priv |= BRGPHY_FLAG_5704_A0;
2208 
2209 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
2210 		mii_priv |= BRGPHY_FLAG_5906;
2211 
2212 	if (BGE_IS_5705_PLUS(sc) &&
2213 	    sc->bge_asicrev != BGE_ASICREV_BCM5906 &&
2214 	    /* sc->bge_asicrev != BGE_ASICREV_BCM5717 && */
2215 	    sc->bge_asicrev != BGE_ASICREV_BCM5785 &&
2216 	    /* sc->bge_asicrev != BGE_ASICREV_BCM57765 && */
2217 	    sc->bge_asicrev != BGE_ASICREV_BCM57780) {
2218 		if (sc->bge_asicrev == BGE_ASICREV_BCM5755 ||
2219 		    sc->bge_asicrev == BGE_ASICREV_BCM5761 ||
2220 		    sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
2221 		    sc->bge_asicrev == BGE_ASICREV_BCM5787) {
2222 			if (product != PCI_PRODUCT_BROADCOM_BCM5722 &&
2223 			    product != PCI_PRODUCT_BROADCOM_BCM5756)
2224 				mii_priv |= BRGPHY_FLAG_JITTER_BUG;
2225 			if (product == PCI_PRODUCT_BROADCOM_BCM5755M)
2226 				mii_priv |= BRGPHY_FLAG_ADJUST_TRIM;
2227 		} else {
2228 			mii_priv |= BRGPHY_FLAG_BER_BUG;
2229 		}
2230 	}
2231 
2232 	/*
2233 	 * Allocate interrupt
2234 	 */
2235 	msi_enable = bge_msi_enable;
2236 	if ((sc->bge_flags & BGE_FLAG_STATUS_TAG) == 0) {
2237 		/* If "tagged status" is disabled, don't enable MSI */
2238 		msi_enable = 0;
2239 	} else if (msi_enable) {
2240 		msi_enable = 0; /* Disable by default */
2241 		if (BGE_IS_575X_PLUS(sc)) {
2242 			msi_enable = 1;
2243 			/* XXX we filter all 5714 chips */
2244 			if (sc->bge_asicrev == BGE_ASICREV_BCM5714 ||
2245 			    (sc->bge_asicrev == BGE_ASICREV_BCM5750 &&
2246 			     (sc->bge_chiprev == BGE_CHIPREV_5750_AX ||
2247 			      sc->bge_chiprev == BGE_CHIPREV_5750_BX)))
2248 				msi_enable = 0;
2249 			else if (BGE_IS_5755_PLUS(sc) ||
2250 			    sc->bge_asicrev == BGE_ASICREV_BCM5906)
2251 				sc->bge_flags |= BGE_FLAG_ONESHOT_MSI;
2252 		}
2253 	}
2254 	if (msi_enable) {
2255 		if (pci_find_extcap(dev, PCIY_MSI, &sc->bge_msicap)) {
2256 			device_printf(dev, "no MSI capability\n");
2257 			msi_enable = 0;
2258 		}
2259 	}
2260 
2261 	sc->bge_irq_type = pci_alloc_1intr(dev, msi_enable, &sc->bge_irq_rid,
2262 	    &intr_flags);
2263 
2264 	sc->bge_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &sc->bge_irq_rid,
2265 	    intr_flags);
2266 	if (sc->bge_irq == NULL) {
2267 		device_printf(dev, "couldn't map interrupt\n");
2268 		error = ENXIO;
2269 		goto fail;
2270 	}
2271 
2272 	if (sc->bge_irq_type == PCI_INTR_TYPE_MSI)
2273 		bge_enable_msi(sc);
2274 	else
2275 		sc->bge_flags &= ~BGE_FLAG_ONESHOT_MSI;
2276 
2277 	/* Initialize if_name earlier, so if_printf could be used */
2278 	ifp = &sc->arpcom.ac_if;
2279 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
2280 
2281 	/* Try to reset the chip. */
2282 	bge_reset(sc);
2283 
2284 	if (bge_chipinit(sc)) {
2285 		device_printf(dev, "chip initialization failed\n");
2286 		error = ENXIO;
2287 		goto fail;
2288 	}
2289 
2290 	/*
2291 	 * Get station address
2292 	 */
2293 	error = bge_get_eaddr(sc, ether_addr);
2294 	if (error) {
2295 		device_printf(dev, "failed to read station address\n");
2296 		goto fail;
2297 	}
2298 
2299 	/* 5705/5750 limits RX return ring to 512 entries. */
2300 	if (BGE_IS_5705_PLUS(sc))
2301 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT_5705;
2302 	else
2303 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT;
2304 
2305 	error = bge_dma_alloc(sc);
2306 	if (error)
2307 		goto fail;
2308 
2309 	/* Set default tuneable values. */
2310 	sc->bge_stat_ticks = BGE_TICKS_PER_SEC;
2311 	sc->bge_rx_coal_ticks = BGE_RX_COAL_TICKS_DEF;
2312 	sc->bge_tx_coal_ticks = BGE_TX_COAL_TICKS_DEF;
2313 	sc->bge_rx_coal_bds = BGE_RX_COAL_BDS_DEF;
2314 	sc->bge_tx_coal_bds = BGE_TX_COAL_BDS_DEF;
2315 	if (sc->bge_flags & BGE_FLAG_STATUS_TAG) {
2316 		sc->bge_rx_coal_ticks_int = BGE_RX_COAL_TICKS_DEF;
2317 		sc->bge_tx_coal_ticks_int = BGE_TX_COAL_TICKS_DEF;
2318 		sc->bge_rx_coal_bds_int = BGE_RX_COAL_BDS_DEF;
2319 		sc->bge_tx_coal_bds_int = BGE_TX_COAL_BDS_DEF;
2320 	} else {
2321 		sc->bge_rx_coal_ticks_int = BGE_RX_COAL_TICKS_MIN;
2322 		sc->bge_tx_coal_ticks_int = BGE_TX_COAL_TICKS_MIN;
2323 		sc->bge_rx_coal_bds_int = BGE_RX_COAL_BDS_MIN;
2324 		sc->bge_tx_coal_bds_int = BGE_TX_COAL_BDS_MIN;
2325 	}
2326 	sc->bge_tx_wreg = BGE_TX_WREG_NSEGS;
2327 
2328 	/* Set up TX spare and reserved descriptor count */
2329 	if (sc->bge_flags & BGE_FLAG_TSO) {
2330 		sc->bge_txspare = BGE_NSEG_SPARE_TSO;
2331 		sc->bge_txrsvd = BGE_NSEG_RSVD_TSO;
2332 	} else {
2333 		sc->bge_txspare = BGE_NSEG_SPARE;
2334 		sc->bge_txrsvd = BGE_NSEG_RSVD;
2335 	}
2336 
2337 	/* Set up ifnet structure */
2338 	ifp->if_softc = sc;
2339 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2340 	ifp->if_ioctl = bge_ioctl;
2341 	ifp->if_start = bge_start;
2342 #ifdef IFPOLL_ENABLE
2343 	ifp->if_npoll = bge_npoll;
2344 #endif
2345 	ifp->if_watchdog = bge_watchdog;
2346 	ifp->if_init = bge_init;
2347 	ifp->if_mtu = ETHERMTU;
2348 	ifp->if_capabilities = IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU;
2349 	ifq_set_maxlen(&ifp->if_snd, BGE_TX_RING_CNT - 1);
2350 	ifq_set_ready(&ifp->if_snd);
2351 
2352 	/*
2353 	 * 5700 B0 chips do not support checksumming correctly due
2354 	 * to hardware bugs.
2355 	 */
2356 	if (sc->bge_chipid != BGE_CHIPID_BCM5700_B0) {
2357 		ifp->if_capabilities |= IFCAP_HWCSUM;
2358 		ifp->if_hwassist |= BGE_CSUM_FEATURES;
2359 	}
2360 	if (sc->bge_flags & BGE_FLAG_TSO) {
2361 		ifp->if_capabilities |= IFCAP_TSO;
2362 		ifp->if_hwassist |= CSUM_TSO;
2363 	}
2364 	ifp->if_capenable = ifp->if_capabilities;
2365 
2366 	/*
2367 	 * Figure out what sort of media we have by checking the
2368 	 * hardware config word in the first 32k of NIC internal memory,
2369 	 * or fall back to examining the EEPROM if necessary.
2370 	 * Note: on some BCM5700 cards, this value appears to be unset.
2371 	 * If that's the case, we have to rely on identifying the NIC
2372 	 * by its PCI subsystem ID, as we do below for the SysKonnect
2373 	 * SK-9D41.
2374 	 */
2375 	if (bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_SIG) == BGE_MAGIC_NUMBER) {
2376 		hwcfg = bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_NICCFG);
2377 	} else {
2378 		if (bge_read_eeprom(sc, (caddr_t)&hwcfg, BGE_EE_HWCFG_OFFSET,
2379 				    sizeof(hwcfg))) {
2380 			device_printf(dev, "failed to read EEPROM\n");
2381 			error = ENXIO;
2382 			goto fail;
2383 		}
2384 		hwcfg = ntohl(hwcfg);
2385 	}
2386 
2387 	/* The SysKonnect SK-9D41 is a 1000baseSX card. */
2388 	if (pci_get_subvendor(dev) == PCI_PRODUCT_SCHNEIDERKOCH_SK_9D41 ||
2389 	    (hwcfg & BGE_HWCFG_MEDIA) == BGE_MEDIA_FIBER) {
2390 		if (BGE_IS_5714_FAMILY(sc))
2391 			sc->bge_flags |= BGE_FLAG_MII_SERDES;
2392 		else
2393 			sc->bge_flags |= BGE_FLAG_TBI;
2394 	}
2395 
2396 	/* Setup MI MODE */
2397 	if (sc->bge_flags & BGE_FLAG_CPMU)
2398 		sc->bge_mi_mode = BGE_MIMODE_500KHZ_CONST;
2399 	else
2400 		sc->bge_mi_mode = BGE_MIMODE_BASE;
2401 	if (BGE_IS_5700_FAMILY(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5705) {
2402 		/* Enable auto polling for BCM570[0-5]. */
2403 		sc->bge_mi_mode |= BGE_MIMODE_AUTOPOLL;
2404 	}
2405 
2406 	/* Setup link status update stuffs */
2407 	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
2408 	    sc->bge_chipid != BGE_CHIPID_BCM5700_B2) {
2409 		sc->bge_link_upd = bge_bcm5700_link_upd;
2410 		sc->bge_link_chg = BGE_MACSTAT_MI_INTERRUPT;
2411 	} else if (sc->bge_flags & BGE_FLAG_TBI) {
2412 		sc->bge_link_upd = bge_tbi_link_upd;
2413 		sc->bge_link_chg = BGE_MACSTAT_LINK_CHANGED;
2414 	} else if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
2415 		sc->bge_link_upd = bge_autopoll_link_upd;
2416 		sc->bge_link_chg = BGE_MACSTAT_LINK_CHANGED;
2417 	} else {
2418 		sc->bge_link_upd = bge_copper_link_upd;
2419 		sc->bge_link_chg = BGE_MACSTAT_LINK_CHANGED;
2420 	}
2421 
2422 	/*
2423 	 * Broadcom's own driver always assumes the internal
2424 	 * PHY is at GMII address 1.  On some chips, the PHY responds
2425 	 * to accesses at all addresses, which could cause us to
2426 	 * bogusly attach the PHY 32 times at probe type.  Always
2427 	 * restricting the lookup to address 1 is simpler than
2428 	 * trying to figure out which chips revisions should be
2429 	 * special-cased.
2430 	 */
2431 	sc->bge_phyno = 1;
2432 
2433 	if (sc->bge_flags & BGE_FLAG_TBI) {
2434 		ifmedia_init(&sc->bge_ifmedia, IFM_IMASK,
2435 		    bge_ifmedia_upd, bge_ifmedia_sts);
2436 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER|IFM_1000_SX, 0, NULL);
2437 		ifmedia_add(&sc->bge_ifmedia,
2438 		    IFM_ETHER|IFM_1000_SX|IFM_FDX, 0, NULL);
2439 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER|IFM_AUTO, 0, NULL);
2440 		ifmedia_set(&sc->bge_ifmedia, IFM_ETHER|IFM_AUTO);
2441 		sc->bge_ifmedia.ifm_media = sc->bge_ifmedia.ifm_cur->ifm_media;
2442 	} else {
2443 		struct mii_probe_args mii_args;
2444 
2445 		mii_probe_args_init(&mii_args, bge_ifmedia_upd, bge_ifmedia_sts);
2446 		mii_args.mii_probemask = 1 << sc->bge_phyno;
2447 		mii_args.mii_capmask = capmask;
2448 		mii_args.mii_privtag = MII_PRIVTAG_BRGPHY;
2449 		mii_args.mii_priv = mii_priv;
2450 
2451 		error = mii_probe(dev, &sc->bge_miibus, &mii_args);
2452 		if (error) {
2453 			device_printf(dev, "MII without any PHY!\n");
2454 			goto fail;
2455 		}
2456 	}
2457 
2458 	/*
2459 	 * Create sysctl nodes.
2460 	 */
2461 	sysctl_ctx_init(&sc->bge_sysctl_ctx);
2462 	sc->bge_sysctl_tree = SYSCTL_ADD_NODE(&sc->bge_sysctl_ctx,
2463 					      SYSCTL_STATIC_CHILDREN(_hw),
2464 					      OID_AUTO,
2465 					      device_get_nameunit(dev),
2466 					      CTLFLAG_RD, 0, "");
2467 	if (sc->bge_sysctl_tree == NULL) {
2468 		device_printf(dev, "can't add sysctl node\n");
2469 		error = ENXIO;
2470 		goto fail;
2471 	}
2472 
2473 	SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2474 			SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2475 			OID_AUTO, "rx_coal_ticks",
2476 			CTLTYPE_INT | CTLFLAG_RW,
2477 			sc, 0, bge_sysctl_rx_coal_ticks, "I",
2478 			"Receive coalescing ticks (usec).");
2479 	SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2480 			SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2481 			OID_AUTO, "tx_coal_ticks",
2482 			CTLTYPE_INT | CTLFLAG_RW,
2483 			sc, 0, bge_sysctl_tx_coal_ticks, "I",
2484 			"Transmit coalescing ticks (usec).");
2485 	SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2486 			SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2487 			OID_AUTO, "rx_coal_bds",
2488 			CTLTYPE_INT | CTLFLAG_RW,
2489 			sc, 0, bge_sysctl_rx_coal_bds, "I",
2490 			"Receive max coalesced BD count.");
2491 	SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2492 			SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2493 			OID_AUTO, "tx_coal_bds",
2494 			CTLTYPE_INT | CTLFLAG_RW,
2495 			sc, 0, bge_sysctl_tx_coal_bds, "I",
2496 			"Transmit max coalesced BD count.");
2497 
2498 	SYSCTL_ADD_INT(&sc->bge_sysctl_ctx,
2499 		       SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2500 		       OID_AUTO, "tx_wreg", CTLFLAG_RW,
2501 		       &sc->bge_tx_wreg, 0,
2502 		       "# of segments before writing to hardware register");
2503 
2504 	if (sc->bge_flags & BGE_FLAG_PCIE) {
2505 		/*
2506 		 * A common design characteristic for many Broadcom
2507 		 * client controllers is that they only support a
2508 		 * single outstanding DMA read operation on the PCIe
2509 		 * bus. This means that it will take twice as long to
2510 		 * fetch a TX frame that is split into header and
2511 		 * payload buffers as it does to fetch a single,
2512 		 * contiguous TX frame (2 reads vs. 1 read). For these
2513 		 * controllers, coalescing buffers to reduce the number
2514 		 * of memory reads is effective way to get maximum
2515 		 * performance(about 940Mbps).  Without collapsing TX
2516 		 * buffers the maximum TCP bulk transfer performance
2517 		 * is about 850Mbps. However forcing coalescing mbufs
2518 		 * consumes a lot of CPU cycles, so leave it off by
2519 		 * default.
2520 		 */
2521 		SYSCTL_ADD_INT(&sc->bge_sysctl_ctx,
2522 			       SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2523 			       OID_AUTO, "force_defrag", CTLFLAG_RW,
2524 			       &sc->bge_force_defrag, 0,
2525 			       "Force defragment on TX path");
2526 	}
2527 	if (sc->bge_flags & BGE_FLAG_STATUS_TAG) {
2528 		if (!BGE_IS_5705_PLUS(sc)) {
2529 			SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2530 			    SYSCTL_CHILDREN(sc->bge_sysctl_tree), OID_AUTO,
2531 			    "rx_coal_ticks_int", CTLTYPE_INT | CTLFLAG_RW,
2532 			    sc, 0, bge_sysctl_rx_coal_ticks_int, "I",
2533 			    "Receive coalescing ticks "
2534 			    "during interrupt (usec).");
2535 			SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2536 			    SYSCTL_CHILDREN(sc->bge_sysctl_tree), OID_AUTO,
2537 			    "tx_coal_ticks_int", CTLTYPE_INT | CTLFLAG_RW,
2538 			    sc, 0, bge_sysctl_tx_coal_ticks_int, "I",
2539 			    "Transmit coalescing ticks "
2540 			    "during interrupt (usec).");
2541 		}
2542 		SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2543 		    SYSCTL_CHILDREN(sc->bge_sysctl_tree), OID_AUTO,
2544 		    "rx_coal_bds_int", CTLTYPE_INT | CTLFLAG_RW,
2545 		    sc, 0, bge_sysctl_rx_coal_bds_int, "I",
2546 		    "Receive max coalesced BD count during interrupt.");
2547 		SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2548 		    SYSCTL_CHILDREN(sc->bge_sysctl_tree), OID_AUTO,
2549 		    "tx_coal_bds_int", CTLTYPE_INT | CTLFLAG_RW,
2550 		    sc, 0, bge_sysctl_tx_coal_bds_int, "I",
2551 		    "Transmit max coalesced BD count during interrupt.");
2552 	}
2553 
2554 	/*
2555 	 * Call MI attach routine.
2556 	 */
2557 	ether_ifattach(ifp, ether_addr, NULL);
2558 
2559 	ifq_set_cpuid(&ifp->if_snd, rman_get_cpuid(sc->bge_irq));
2560 
2561 #ifdef IFPOLL_ENABLE
2562 	/* Polling setup */
2563 	ifpoll_compat_setup(&sc->bge_npoll,
2564 	    &sc->bge_sysctl_ctx, sc->bge_sysctl_tree, device_get_unit(dev),
2565 	    ifp->if_serializer);
2566 #endif
2567 
2568 	if (sc->bge_irq_type == PCI_INTR_TYPE_MSI) {
2569 		if (sc->bge_flags & BGE_FLAG_ONESHOT_MSI) {
2570 			intr_func = bge_msi_oneshot;
2571 			if (bootverbose)
2572 				device_printf(dev, "oneshot MSI\n");
2573 		} else {
2574 			intr_func = bge_msi;
2575 		}
2576 	} else if (sc->bge_flags & BGE_FLAG_STATUS_TAG) {
2577 		intr_func = bge_intr_legacy;
2578 	} else {
2579 		intr_func = bge_intr_crippled;
2580 	}
2581 	error = bus_setup_intr(dev, sc->bge_irq, INTR_MPSAFE, intr_func, sc,
2582 	    &sc->bge_intrhand, ifp->if_serializer);
2583 	if (error) {
2584 		ether_ifdetach(ifp);
2585 		device_printf(dev, "couldn't set up irq\n");
2586 		goto fail;
2587 	}
2588 
2589 	return(0);
2590 fail:
2591 	bge_detach(dev);
2592 	return(error);
2593 }
2594 
2595 static int
2596 bge_detach(device_t dev)
2597 {
2598 	struct bge_softc *sc = device_get_softc(dev);
2599 
2600 	if (device_is_attached(dev)) {
2601 		struct ifnet *ifp = &sc->arpcom.ac_if;
2602 
2603 		lwkt_serialize_enter(ifp->if_serializer);
2604 		bge_stop(sc);
2605 		bge_reset(sc);
2606 		bus_teardown_intr(dev, sc->bge_irq, sc->bge_intrhand);
2607 		lwkt_serialize_exit(ifp->if_serializer);
2608 
2609 		ether_ifdetach(ifp);
2610 	}
2611 
2612 	if (sc->bge_flags & BGE_FLAG_TBI)
2613 		ifmedia_removeall(&sc->bge_ifmedia);
2614 	if (sc->bge_miibus)
2615 		device_delete_child(dev, sc->bge_miibus);
2616 	bus_generic_detach(dev);
2617 
2618 	if (sc->bge_irq != NULL) {
2619 		bus_release_resource(dev, SYS_RES_IRQ, sc->bge_irq_rid,
2620 		    sc->bge_irq);
2621 	}
2622 	if (sc->bge_irq_type == PCI_INTR_TYPE_MSI)
2623 		pci_release_msi(dev);
2624 
2625 	if (sc->bge_res != NULL) {
2626 		bus_release_resource(dev, SYS_RES_MEMORY,
2627 		    BGE_PCI_BAR0, sc->bge_res);
2628 	}
2629 
2630 	if (sc->bge_sysctl_tree != NULL)
2631 		sysctl_ctx_free(&sc->bge_sysctl_ctx);
2632 
2633 	bge_dma_free(sc);
2634 
2635 	return 0;
2636 }
2637 
2638 static void
2639 bge_reset(struct bge_softc *sc)
2640 {
2641 	device_t dev;
2642 	uint32_t cachesize, command, pcistate, reset;
2643 	void (*write_op)(struct bge_softc *, uint32_t, uint32_t);
2644 	int i, val = 0;
2645 
2646 	dev = sc->bge_dev;
2647 
2648 	if (BGE_IS_575X_PLUS(sc) && !BGE_IS_5714_FAMILY(sc) &&
2649 	    sc->bge_asicrev != BGE_ASICREV_BCM5906) {
2650 		if (sc->bge_flags & BGE_FLAG_PCIE)
2651 			write_op = bge_writemem_direct;
2652 		else
2653 			write_op = bge_writemem_ind;
2654 	} else {
2655 		write_op = bge_writereg_ind;
2656 	}
2657 
2658 	/* Save some important PCI state. */
2659 	cachesize = pci_read_config(dev, BGE_PCI_CACHESZ, 4);
2660 	command = pci_read_config(dev, BGE_PCI_CMD, 4);
2661 	pcistate = pci_read_config(dev, BGE_PCI_PCISTATE, 4);
2662 
2663 	pci_write_config(dev, BGE_PCI_MISC_CTL,
2664 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2665 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW|
2666 	    sc->bge_pci_miscctl, 4);
2667 
2668 	/* Disable fastboot on controllers that support it. */
2669 	if (sc->bge_asicrev == BGE_ASICREV_BCM5752 ||
2670 	    BGE_IS_5755_PLUS(sc)) {
2671 		if (bootverbose)
2672 			if_printf(&sc->arpcom.ac_if, "Disabling fastboot\n");
2673 		CSR_WRITE_4(sc, BGE_FASTBOOT_PC, 0x0);
2674 	}
2675 
2676 	/*
2677 	 * Write the magic number to SRAM at offset 0xB50.
2678 	 * When firmware finishes its initialization it will
2679 	 * write ~BGE_MAGIC_NUMBER to the same location.
2680 	 */
2681 	bge_writemem_ind(sc, BGE_SOFTWARE_GENCOMM, BGE_MAGIC_NUMBER);
2682 
2683 	reset = BGE_MISCCFG_RESET_CORE_CLOCKS|(65<<1);
2684 
2685 	/* XXX: Broadcom Linux driver. */
2686 	if (sc->bge_flags & BGE_FLAG_PCIE) {
2687 		/* Force PCI-E 1.0a mode */
2688 		if (sc->bge_asicrev != BGE_ASICREV_BCM5785 &&
2689 		    CSR_READ_4(sc, BGE_PCIE_PHY_TSTCTL) ==
2690 		    (BGE_PCIE_PHY_TSTCTL_PSCRAM |
2691 		     BGE_PCIE_PHY_TSTCTL_PCIE10)) {
2692 			CSR_WRITE_4(sc, BGE_PCIE_PHY_TSTCTL,
2693 			    BGE_PCIE_PHY_TSTCTL_PSCRAM);
2694 		}
2695 		if (sc->bge_chipid != BGE_CHIPID_BCM5750_A0) {
2696 			/* Prevent PCIE link training during global reset */
2697 			CSR_WRITE_4(sc, BGE_MISC_CFG, (1<<29));
2698 			reset |= (1<<29);
2699 		}
2700 	}
2701 
2702 	/*
2703 	 * Set GPHY Power Down Override to leave GPHY
2704 	 * powered up in D0 uninitialized.
2705 	 */
2706 	if (BGE_IS_5705_PLUS(sc) && (sc->bge_flags & BGE_FLAG_CPMU) == 0)
2707 		reset |= BGE_MISCCFG_GPHY_PD_OVERRIDE;
2708 
2709 	/* Issue global reset */
2710 	write_op(sc, BGE_MISC_CFG, reset);
2711 
2712 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
2713 		uint32_t status, ctrl;
2714 
2715 		status = CSR_READ_4(sc, BGE_VCPU_STATUS);
2716 		CSR_WRITE_4(sc, BGE_VCPU_STATUS,
2717 		    status | BGE_VCPU_STATUS_DRV_RESET);
2718 		ctrl = CSR_READ_4(sc, BGE_VCPU_EXT_CTRL);
2719 		CSR_WRITE_4(sc, BGE_VCPU_EXT_CTRL,
2720 		    ctrl & ~BGE_VCPU_EXT_CTRL_HALT_CPU);
2721 	}
2722 
2723 	DELAY(1000);
2724 
2725 	/* XXX: Broadcom Linux driver. */
2726 	if (sc->bge_flags & BGE_FLAG_PCIE) {
2727 		uint16_t devctl;
2728 
2729 		if (sc->bge_chipid == BGE_CHIPID_BCM5750_A0) {
2730 			uint32_t v;
2731 
2732 			DELAY(500000); /* wait for link training to complete */
2733 			v = pci_read_config(dev, 0xc4, 4);
2734 			pci_write_config(dev, 0xc4, v | (1<<15), 4);
2735 		}
2736 
2737 		devctl = pci_read_config(dev,
2738 		    sc->bge_pciecap + PCIER_DEVCTRL, 2);
2739 
2740 		/* Disable no snoop and disable relaxed ordering. */
2741 		devctl &= ~(PCIEM_DEVCTL_RELAX_ORDER | PCIEM_DEVCTL_NOSNOOP);
2742 
2743 		/* Old PCI-E chips only support 128 bytes Max PayLoad Size. */
2744 		if ((sc->bge_flags & BGE_FLAG_CPMU) == 0) {
2745 			devctl &= ~PCIEM_DEVCTL_MAX_PAYLOAD_MASK;
2746 			devctl |= PCIEM_DEVCTL_MAX_PAYLOAD_128;
2747 		}
2748 
2749 		pci_write_config(dev, sc->bge_pciecap + PCIER_DEVCTRL,
2750 		    devctl, 2);
2751 
2752 		/* Clear error status. */
2753 		pci_write_config(dev, sc->bge_pciecap + PCIER_DEVSTS,
2754 		    PCIEM_DEVSTS_CORR_ERR |
2755 		    PCIEM_DEVSTS_NFATAL_ERR |
2756 		    PCIEM_DEVSTS_FATAL_ERR |
2757 		    PCIEM_DEVSTS_UNSUPP_REQ, 2);
2758 	}
2759 
2760 	/* Reset some of the PCI state that got zapped by reset */
2761 	pci_write_config(dev, BGE_PCI_MISC_CTL,
2762 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2763 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW|
2764 	    sc->bge_pci_miscctl, 4);
2765 	pci_write_config(dev, BGE_PCI_CACHESZ, cachesize, 4);
2766 	pci_write_config(dev, BGE_PCI_CMD, command, 4);
2767 	write_op(sc, BGE_MISC_CFG, (65 << 1));
2768 
2769 	/*
2770 	 * Disable PCI-X relaxed ordering to ensure status block update
2771 	 * comes first then packet buffer DMA. Otherwise driver may
2772 	 * read stale status block.
2773 	 */
2774 	if (sc->bge_flags & BGE_FLAG_PCIX) {
2775 		uint16_t devctl;
2776 
2777 		devctl = pci_read_config(dev,
2778 		    sc->bge_pcixcap + PCIXR_COMMAND, 2);
2779 		devctl &= ~PCIXM_COMMAND_ERO;
2780 		if (sc->bge_asicrev == BGE_ASICREV_BCM5703) {
2781 			devctl &= ~PCIXM_COMMAND_MAX_READ;
2782 			devctl |= PCIXM_COMMAND_MAX_READ_2048;
2783 		} else if (sc->bge_asicrev == BGE_ASICREV_BCM5704) {
2784 			devctl &= ~(PCIXM_COMMAND_MAX_SPLITS |
2785 			    PCIXM_COMMAND_MAX_READ);
2786 			devctl |= PCIXM_COMMAND_MAX_READ_2048;
2787 		}
2788 		pci_write_config(dev, sc->bge_pcixcap + PCIXR_COMMAND,
2789 		    devctl, 2);
2790 	}
2791 
2792 	/*
2793 	 * Enable memory arbiter and re-enable MSI if necessary.
2794 	 */
2795 	if (BGE_IS_5714_FAMILY(sc)) {
2796 		uint32_t val;
2797 
2798 		if (sc->bge_irq_type == PCI_INTR_TYPE_MSI) {
2799 			/*
2800 			 * Resetting BCM5714 family will clear MSI
2801 			 * enable bit; restore it after resetting.
2802 			 */
2803 			PCI_SETBIT(sc->bge_dev, sc->bge_msicap + PCIR_MSI_CTRL,
2804 			    PCIM_MSICTRL_MSI_ENABLE, 2);
2805 			BGE_SETBIT(sc, BGE_MSI_MODE, BGE_MSIMODE_ENABLE);
2806 		}
2807 		val = CSR_READ_4(sc, BGE_MARB_MODE);
2808 		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE | val);
2809 	} else {
2810 		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
2811 	}
2812 
2813 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
2814 		for (i = 0; i < BGE_TIMEOUT; i++) {
2815 			val = CSR_READ_4(sc, BGE_VCPU_STATUS);
2816 			if (val & BGE_VCPU_STATUS_INIT_DONE)
2817 				break;
2818 			DELAY(100);
2819 		}
2820 		if (i == BGE_TIMEOUT) {
2821 			if_printf(&sc->arpcom.ac_if, "reset timed out\n");
2822 			return;
2823 		}
2824 	} else {
2825 		/*
2826 		 * Poll until we see the 1's complement of the magic number.
2827 		 * This indicates that the firmware initialization
2828 		 * is complete.
2829 		 */
2830 		for (i = 0; i < BGE_FIRMWARE_TIMEOUT; i++) {
2831 			val = bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM);
2832 			if (val == ~BGE_MAGIC_NUMBER)
2833 				break;
2834 			DELAY(10);
2835 		}
2836 		if (i == BGE_FIRMWARE_TIMEOUT) {
2837 			if_printf(&sc->arpcom.ac_if, "firmware handshake "
2838 				  "timed out, found 0x%08x\n", val);
2839 		}
2840 	}
2841 
2842 	/*
2843 	 * XXX Wait for the value of the PCISTATE register to
2844 	 * return to its original pre-reset state. This is a
2845 	 * fairly good indicator of reset completion. If we don't
2846 	 * wait for the reset to fully complete, trying to read
2847 	 * from the device's non-PCI registers may yield garbage
2848 	 * results.
2849 	 */
2850 	for (i = 0; i < BGE_TIMEOUT; i++) {
2851 		if (pci_read_config(dev, BGE_PCI_PCISTATE, 4) == pcistate)
2852 			break;
2853 		DELAY(10);
2854 	}
2855 
2856 	/* Fix up byte swapping */
2857 	CSR_WRITE_4(sc, BGE_MODE_CTL, BGE_DMA_SWAP_OPTIONS |
2858 	    BGE_MODECTL_BYTESWAP_DATA);
2859 
2860 	CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
2861 
2862 	/*
2863 	 * The 5704 in TBI mode apparently needs some special
2864 	 * adjustment to insure the SERDES drive level is set
2865 	 * to 1.2V.
2866 	 */
2867 	if (sc->bge_asicrev == BGE_ASICREV_BCM5704 &&
2868 	    (sc->bge_flags & BGE_FLAG_TBI)) {
2869 		uint32_t serdescfg;
2870 
2871 		serdescfg = CSR_READ_4(sc, BGE_SERDES_CFG);
2872 		serdescfg = (serdescfg & ~0xFFF) | 0x880;
2873 		CSR_WRITE_4(sc, BGE_SERDES_CFG, serdescfg);
2874 	}
2875 
2876 	/* XXX: Broadcom Linux driver. */
2877 	if ((sc->bge_flags & BGE_FLAG_PCIE) &&
2878 	    sc->bge_chipid != BGE_CHIPID_BCM5750_A0 &&
2879 	    sc->bge_asicrev != BGE_ASICREV_BCM5785) {
2880 		uint32_t v;
2881 
2882 		/* Enable Data FIFO protection. */
2883 		v = CSR_READ_4(sc, BGE_PCIE_TLDLPL_PORT);
2884 		CSR_WRITE_4(sc, BGE_PCIE_TLDLPL_PORT, v | (1 << 25));
2885 	}
2886 
2887 	DELAY(10000);
2888 }
2889 
2890 /*
2891  * Frame reception handling. This is called if there's a frame
2892  * on the receive return list.
2893  *
2894  * Note: we have to be able to handle two possibilities here:
2895  * 1) the frame is from the jumbo recieve ring
2896  * 2) the frame is from the standard receive ring
2897  */
2898 
2899 static void
2900 bge_rxeof(struct bge_softc *sc, uint16_t rx_prod, int count)
2901 {
2902 	struct ifnet *ifp;
2903 	int stdcnt = 0, jumbocnt = 0;
2904 
2905 	ifp = &sc->arpcom.ac_if;
2906 
2907 	while (sc->bge_rx_saved_considx != rx_prod && count != 0) {
2908 		struct bge_rx_bd	*cur_rx;
2909 		uint32_t		rxidx;
2910 		struct mbuf		*m = NULL;
2911 		uint16_t		vlan_tag = 0;
2912 		int			have_tag = 0;
2913 
2914 		--count;
2915 
2916 		cur_rx =
2917 	    &sc->bge_ldata.bge_rx_return_ring[sc->bge_rx_saved_considx];
2918 
2919 		rxidx = cur_rx->bge_idx;
2920 		BGE_INC(sc->bge_rx_saved_considx, sc->bge_return_ring_cnt);
2921 		logif(rx_pkt);
2922 
2923 		if (cur_rx->bge_flags & BGE_RXBDFLAG_VLAN_TAG) {
2924 			have_tag = 1;
2925 			vlan_tag = cur_rx->bge_vlan_tag;
2926 		}
2927 
2928 		if (cur_rx->bge_flags & BGE_RXBDFLAG_JUMBO_RING) {
2929 			BGE_INC(sc->bge_jumbo, BGE_JUMBO_RX_RING_CNT);
2930 			jumbocnt++;
2931 
2932 			if (rxidx != sc->bge_jumbo) {
2933 				IFNET_STAT_INC(ifp, ierrors, 1);
2934 				if_printf(ifp, "sw jumbo index(%d) "
2935 				    "and hw jumbo index(%d) mismatch, drop!\n",
2936 				    sc->bge_jumbo, rxidx);
2937 				bge_setup_rxdesc_jumbo(sc, rxidx);
2938 				continue;
2939 			}
2940 
2941 			m = sc->bge_cdata.bge_rx_jumbo_chain[rxidx].bge_mbuf;
2942 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
2943 				IFNET_STAT_INC(ifp, ierrors, 1);
2944 				bge_setup_rxdesc_jumbo(sc, sc->bge_jumbo);
2945 				continue;
2946 			}
2947 			if (bge_newbuf_jumbo(sc, sc->bge_jumbo, 0)) {
2948 				IFNET_STAT_INC(ifp, ierrors, 1);
2949 				bge_setup_rxdesc_jumbo(sc, sc->bge_jumbo);
2950 				continue;
2951 			}
2952 		} else {
2953 			BGE_INC(sc->bge_std, BGE_STD_RX_RING_CNT);
2954 			stdcnt++;
2955 
2956 			if (rxidx != sc->bge_std) {
2957 				IFNET_STAT_INC(ifp, ierrors, 1);
2958 				if_printf(ifp, "sw std index(%d) "
2959 				    "and hw std index(%d) mismatch, drop!\n",
2960 				    sc->bge_std, rxidx);
2961 				bge_setup_rxdesc_std(sc, rxidx);
2962 				continue;
2963 			}
2964 
2965 			m = sc->bge_cdata.bge_rx_std_chain[rxidx].bge_mbuf;
2966 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
2967 				IFNET_STAT_INC(ifp, ierrors, 1);
2968 				bge_setup_rxdesc_std(sc, sc->bge_std);
2969 				continue;
2970 			}
2971 			if (bge_newbuf_std(sc, sc->bge_std, 0)) {
2972 				IFNET_STAT_INC(ifp, ierrors, 1);
2973 				bge_setup_rxdesc_std(sc, sc->bge_std);
2974 				continue;
2975 			}
2976 		}
2977 
2978 		IFNET_STAT_INC(ifp, ipackets, 1);
2979 #if !defined(__i386__) && !defined(__x86_64__)
2980 		/*
2981 		 * The x86 allows unaligned accesses, but for other
2982 		 * platforms we must make sure the payload is aligned.
2983 		 */
2984 		if (sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) {
2985 			bcopy(m->m_data, m->m_data + ETHER_ALIGN,
2986 			    cur_rx->bge_len);
2987 			m->m_data += ETHER_ALIGN;
2988 		}
2989 #endif
2990 		m->m_pkthdr.len = m->m_len = cur_rx->bge_len - ETHER_CRC_LEN;
2991 		m->m_pkthdr.rcvif = ifp;
2992 
2993 		if (ifp->if_capenable & IFCAP_RXCSUM) {
2994 			if (cur_rx->bge_flags & BGE_RXBDFLAG_IP_CSUM) {
2995 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2996 				if ((cur_rx->bge_ip_csum ^ 0xffff) == 0)
2997 					m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2998 			}
2999 			if ((cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM) &&
3000 			    m->m_pkthdr.len >= BGE_MIN_FRAMELEN) {
3001 				m->m_pkthdr.csum_data =
3002 					cur_rx->bge_tcp_udp_csum;
3003 				m->m_pkthdr.csum_flags |=
3004 					CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
3005 			}
3006 		}
3007 
3008 		/*
3009 		 * If we received a packet with a vlan tag, pass it
3010 		 * to vlan_input() instead of ether_input().
3011 		 */
3012 		if (have_tag) {
3013 			m->m_flags |= M_VLANTAG;
3014 			m->m_pkthdr.ether_vlantag = vlan_tag;
3015 		}
3016 		ifp->if_input(ifp, m);
3017 	}
3018 
3019 	bge_writembx(sc, BGE_MBX_RX_CONS0_LO, sc->bge_rx_saved_considx);
3020 	if (stdcnt)
3021 		bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bge_std);
3022 	if (jumbocnt)
3023 		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
3024 }
3025 
3026 static void
3027 bge_txeof(struct bge_softc *sc, uint16_t tx_cons)
3028 {
3029 	struct ifnet *ifp;
3030 
3031 	ifp = &sc->arpcom.ac_if;
3032 
3033 	/*
3034 	 * Go through our tx ring and free mbufs for those
3035 	 * frames that have been sent.
3036 	 */
3037 	while (sc->bge_tx_saved_considx != tx_cons) {
3038 		uint32_t idx = 0;
3039 
3040 		idx = sc->bge_tx_saved_considx;
3041 		if (sc->bge_cdata.bge_tx_chain[idx] != NULL) {
3042 			IFNET_STAT_INC(ifp, opackets, 1);
3043 			bus_dmamap_unload(sc->bge_cdata.bge_tx_mtag,
3044 			    sc->bge_cdata.bge_tx_dmamap[idx]);
3045 			m_freem(sc->bge_cdata.bge_tx_chain[idx]);
3046 			sc->bge_cdata.bge_tx_chain[idx] = NULL;
3047 		}
3048 		sc->bge_txcnt--;
3049 		BGE_INC(sc->bge_tx_saved_considx, BGE_TX_RING_CNT);
3050 		logif(tx_pkt);
3051 	}
3052 
3053 	if ((BGE_TX_RING_CNT - sc->bge_txcnt) >=
3054 	    (sc->bge_txrsvd + sc->bge_txspare))
3055 		ifq_clr_oactive(&ifp->if_snd);
3056 
3057 	if (sc->bge_txcnt == 0)
3058 		ifp->if_timer = 0;
3059 
3060 	if (!ifq_is_empty(&ifp->if_snd))
3061 		if_devstart(ifp);
3062 }
3063 
3064 #ifdef IFPOLL_ENABLE
3065 
3066 static void
3067 bge_npoll_compat(struct ifnet *ifp, void *arg __unused, int cycles)
3068 {
3069 	struct bge_softc *sc = ifp->if_softc;
3070 	struct bge_status_block *sblk = sc->bge_ldata.bge_status_block;
3071 	uint16_t rx_prod, tx_cons;
3072 
3073 	ASSERT_SERIALIZED(ifp->if_serializer);
3074 
3075 	if (sc->bge_npoll.ifpc_stcount-- == 0) {
3076 		sc->bge_npoll.ifpc_stcount = sc->bge_npoll.ifpc_stfrac;
3077 		/*
3078 		 * Process link state changes.
3079 		 */
3080 		bge_link_poll(sc);
3081 	}
3082 
3083 	if (sc->bge_flags & BGE_FLAG_STATUS_TAG) {
3084 		sc->bge_status_tag = sblk->bge_status_tag;
3085 		/*
3086 		 * Use a load fence to ensure that status_tag
3087 		 * is saved  before rx_prod and tx_cons.
3088 		 */
3089 		cpu_lfence();
3090 	}
3091 
3092 	rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
3093 	if (sc->bge_rx_saved_considx != rx_prod)
3094 		bge_rxeof(sc, rx_prod, cycles);
3095 
3096 	tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
3097 	if (sc->bge_tx_saved_considx != tx_cons)
3098 		bge_txeof(sc, tx_cons);
3099 
3100 	if (sc->bge_flags & BGE_FLAG_STATUS_TAG)
3101 		bge_writembx(sc, BGE_MBX_IRQ0_LO, sc->bge_status_tag << 24);
3102 
3103 	if (sc->bge_coal_chg)
3104 		bge_coal_change(sc);
3105 }
3106 
3107 static void
3108 bge_npoll(struct ifnet *ifp, struct ifpoll_info *info)
3109 {
3110 	struct bge_softc *sc = ifp->if_softc;
3111 
3112 	ASSERT_SERIALIZED(ifp->if_serializer);
3113 
3114 	if (info != NULL) {
3115 		int cpuid = sc->bge_npoll.ifpc_cpuid;
3116 
3117 		info->ifpi_rx[cpuid].poll_func = bge_npoll_compat;
3118 		info->ifpi_rx[cpuid].arg = NULL;
3119 		info->ifpi_rx[cpuid].serializer = ifp->if_serializer;
3120 
3121 		if (ifp->if_flags & IFF_RUNNING)
3122 			bge_disable_intr(sc);
3123 		ifq_set_cpuid(&ifp->if_snd, cpuid);
3124 	} else {
3125 		if (ifp->if_flags & IFF_RUNNING)
3126 			bge_enable_intr(sc);
3127 		ifq_set_cpuid(&ifp->if_snd, rman_get_cpuid(sc->bge_irq));
3128 	}
3129 }
3130 
3131 #endif	/* IFPOLL_ENABLE */
3132 
3133 static void
3134 bge_intr_crippled(void *xsc)
3135 {
3136 	struct bge_softc *sc = xsc;
3137 	struct ifnet *ifp = &sc->arpcom.ac_if;
3138 
3139 	logif(intr);
3140 
3141  	/*
3142 	 * Ack the interrupt by writing something to BGE_MBX_IRQ0_LO.  Don't
3143 	 * disable interrupts by writing nonzero like we used to, since with
3144 	 * our current organization this just gives complications and
3145 	 * pessimizations for re-enabling interrupts.  We used to have races
3146 	 * instead of the necessary complications.  Disabling interrupts
3147 	 * would just reduce the chance of a status update while we are
3148 	 * running (by switching to the interrupt-mode coalescence
3149 	 * parameters), but this chance is already very low so it is more
3150 	 * efficient to get another interrupt than prevent it.
3151 	 *
3152 	 * We do the ack first to ensure another interrupt if there is a
3153 	 * status update after the ack.  We don't check for the status
3154 	 * changing later because it is more efficient to get another
3155 	 * interrupt than prevent it, not quite as above (not checking is
3156 	 * a smaller optimization than not toggling the interrupt enable,
3157 	 * since checking doesn't involve PCI accesses and toggling require
3158 	 * the status check).  So toggling would probably be a pessimization
3159 	 * even with MSI.  It would only be needed for using a task queue.
3160 	 */
3161 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 0);
3162 
3163 	/*
3164 	 * Process link state changes.
3165 	 */
3166 	bge_link_poll(sc);
3167 
3168 	if (ifp->if_flags & IFF_RUNNING) {
3169 		struct bge_status_block *sblk = sc->bge_ldata.bge_status_block;
3170 		uint16_t rx_prod, tx_cons;
3171 
3172 		rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
3173 		if (sc->bge_rx_saved_considx != rx_prod)
3174 			bge_rxeof(sc, rx_prod, -1);
3175 
3176 		tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
3177 		if (sc->bge_tx_saved_considx != tx_cons)
3178 			bge_txeof(sc, tx_cons);
3179 	}
3180 
3181 	if (sc->bge_coal_chg)
3182 		bge_coal_change(sc);
3183 }
3184 
3185 static void
3186 bge_intr_legacy(void *xsc)
3187 {
3188 	struct bge_softc *sc = xsc;
3189 	struct bge_status_block *sblk = sc->bge_ldata.bge_status_block;
3190 
3191 	if (sc->bge_status_tag == sblk->bge_status_tag) {
3192 		uint32_t val;
3193 
3194 		val = pci_read_config(sc->bge_dev, BGE_PCI_PCISTATE, 4);
3195 		if (val & BGE_PCISTAT_INTR_NOTACT)
3196 			return;
3197 	}
3198 
3199 	/*
3200 	 * NOTE:
3201 	 * Interrupt will have to be disabled if tagged status
3202 	 * is used, else interrupt will always be asserted on
3203 	 * certain chips (at least on BCM5750 AX/BX).
3204 	 */
3205 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
3206 
3207 	bge_intr(sc);
3208 }
3209 
3210 static void
3211 bge_msi(void *xsc)
3212 {
3213 	struct bge_softc *sc = xsc;
3214 
3215 	/* Disable interrupt first */
3216 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
3217 	bge_intr(sc);
3218 }
3219 
3220 static void
3221 bge_msi_oneshot(void *xsc)
3222 {
3223 	bge_intr(xsc);
3224 }
3225 
3226 static void
3227 bge_intr(struct bge_softc *sc)
3228 {
3229 	struct ifnet *ifp = &sc->arpcom.ac_if;
3230 	struct bge_status_block *sblk = sc->bge_ldata.bge_status_block;
3231 	uint16_t rx_prod, tx_cons;
3232 	uint32_t status;
3233 
3234 	sc->bge_status_tag = sblk->bge_status_tag;
3235 	/*
3236 	 * Use a load fence to ensure that status_tag is saved
3237 	 * before rx_prod, tx_cons and status.
3238 	 */
3239 	cpu_lfence();
3240 
3241 	rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
3242 	tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
3243 	status = sblk->bge_status;
3244 
3245 	if ((status & BGE_STATFLAG_LINKSTATE_CHANGED) || sc->bge_link_evt)
3246 		bge_link_poll(sc);
3247 
3248 	if (ifp->if_flags & IFF_RUNNING) {
3249 		if (sc->bge_rx_saved_considx != rx_prod)
3250 			bge_rxeof(sc, rx_prod, -1);
3251 
3252 		if (sc->bge_tx_saved_considx != tx_cons)
3253 			bge_txeof(sc, tx_cons);
3254 	}
3255 
3256 	bge_writembx(sc, BGE_MBX_IRQ0_LO, sc->bge_status_tag << 24);
3257 
3258 	if (sc->bge_coal_chg)
3259 		bge_coal_change(sc);
3260 }
3261 
3262 static void
3263 bge_tick(void *xsc)
3264 {
3265 	struct bge_softc *sc = xsc;
3266 	struct ifnet *ifp = &sc->arpcom.ac_if;
3267 
3268 	lwkt_serialize_enter(ifp->if_serializer);
3269 
3270 	if (BGE_IS_5705_PLUS(sc))
3271 		bge_stats_update_regs(sc);
3272 	else
3273 		bge_stats_update(sc);
3274 
3275 	if (sc->bge_flags & BGE_FLAG_TBI) {
3276 		/*
3277 		 * Since in TBI mode auto-polling can't be used we should poll
3278 		 * link status manually. Here we register pending link event
3279 		 * and trigger interrupt.
3280 		 */
3281 		sc->bge_link_evt++;
3282 		if (BGE_IS_CRIPPLED(sc))
3283 			BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
3284 		else
3285 			BGE_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
3286 	} else if (!sc->bge_link) {
3287 		mii_tick(device_get_softc(sc->bge_miibus));
3288 	}
3289 
3290 	callout_reset(&sc->bge_stat_timer, hz, bge_tick, sc);
3291 
3292 	lwkt_serialize_exit(ifp->if_serializer);
3293 }
3294 
3295 static void
3296 bge_stats_update_regs(struct bge_softc *sc)
3297 {
3298 	struct ifnet *ifp = &sc->arpcom.ac_if;
3299 	struct bge_mac_stats_regs stats;
3300 	uint32_t *s;
3301 	int i;
3302 
3303 	s = (uint32_t *)&stats;
3304 	for (i = 0; i < sizeof(struct bge_mac_stats_regs); i += 4) {
3305 		*s = CSR_READ_4(sc, BGE_RX_STATS + i);
3306 		s++;
3307 	}
3308 
3309 	IFNET_STAT_SET(ifp, collisions,
3310 	   (stats.dot3StatsSingleCollisionFrames +
3311 	   stats.dot3StatsMultipleCollisionFrames +
3312 	   stats.dot3StatsExcessiveCollisions +
3313 	   stats.dot3StatsLateCollisions));
3314 }
3315 
3316 static void
3317 bge_stats_update(struct bge_softc *sc)
3318 {
3319 	struct ifnet *ifp = &sc->arpcom.ac_if;
3320 	bus_size_t stats;
3321 
3322 	stats = BGE_MEMWIN_START + BGE_STATS_BLOCK;
3323 
3324 #define READ_STAT(sc, stats, stat)	\
3325 	CSR_READ_4(sc, stats + offsetof(struct bge_stats, stat))
3326 
3327 	IFNET_STAT_SET(ifp, collisions,
3328 	   (READ_STAT(sc, stats,
3329 		txstats.dot3StatsSingleCollisionFrames.bge_addr_lo) +
3330 	    READ_STAT(sc, stats,
3331 		txstats.dot3StatsMultipleCollisionFrames.bge_addr_lo) +
3332 	    READ_STAT(sc, stats,
3333 		txstats.dot3StatsExcessiveCollisions.bge_addr_lo) +
3334 	    READ_STAT(sc, stats,
3335 		txstats.dot3StatsLateCollisions.bge_addr_lo)));
3336 
3337 #undef READ_STAT
3338 
3339 #ifdef notdef
3340 	IFNET_STAT_SET(ifp, collisions,
3341 	   (sc->bge_rdata->bge_info.bge_stats.dot3StatsSingleCollisionFrames +
3342 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsMultipleCollisionFrames +
3343 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsExcessiveCollisions +
3344 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsLateCollisions));
3345 #endif
3346 }
3347 
3348 /*
3349  * Encapsulate an mbuf chain in the tx ring  by coupling the mbuf data
3350  * pointers to descriptors.
3351  */
3352 static int
3353 bge_encap(struct bge_softc *sc, struct mbuf **m_head0, uint32_t *txidx,
3354     int *segs_used)
3355 {
3356 	struct bge_tx_bd *d = NULL, *last_d;
3357 	uint16_t csum_flags = 0, mss = 0;
3358 	bus_dma_segment_t segs[BGE_NSEG_NEW];
3359 	bus_dmamap_t map;
3360 	int error, maxsegs, nsegs, idx, i;
3361 	struct mbuf *m_head = *m_head0, *m_new;
3362 
3363 	if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
3364 		error = bge_setup_tso(sc, m_head0, &mss, &csum_flags);
3365 		if (error)
3366 			return ENOBUFS;
3367 		m_head = *m_head0;
3368 	} else if (m_head->m_pkthdr.csum_flags & BGE_CSUM_FEATURES) {
3369 		if (m_head->m_pkthdr.csum_flags & CSUM_IP)
3370 			csum_flags |= BGE_TXBDFLAG_IP_CSUM;
3371 		if (m_head->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP))
3372 			csum_flags |= BGE_TXBDFLAG_TCP_UDP_CSUM;
3373 		if (m_head->m_flags & M_LASTFRAG)
3374 			csum_flags |= BGE_TXBDFLAG_IP_FRAG_END;
3375 		else if (m_head->m_flags & M_FRAG)
3376 			csum_flags |= BGE_TXBDFLAG_IP_FRAG;
3377 	}
3378 
3379 	idx = *txidx;
3380 	map = sc->bge_cdata.bge_tx_dmamap[idx];
3381 
3382 	maxsegs = (BGE_TX_RING_CNT - sc->bge_txcnt) - sc->bge_txrsvd;
3383 	KASSERT(maxsegs >= sc->bge_txspare,
3384 		("not enough segments %d", maxsegs));
3385 
3386 	if (maxsegs > BGE_NSEG_NEW)
3387 		maxsegs = BGE_NSEG_NEW;
3388 
3389 	/*
3390 	 * Pad outbound frame to BGE_MIN_FRAMELEN for an unusual reason.
3391 	 * The bge hardware will pad out Tx runts to BGE_MIN_FRAMELEN,
3392 	 * but when such padded frames employ the bge IP/TCP checksum
3393 	 * offload, the hardware checksum assist gives incorrect results
3394 	 * (possibly from incorporating its own padding into the UDP/TCP
3395 	 * checksum; who knows).  If we pad such runts with zeros, the
3396 	 * onboard checksum comes out correct.
3397 	 */
3398 	if ((csum_flags & BGE_TXBDFLAG_TCP_UDP_CSUM) &&
3399 	    m_head->m_pkthdr.len < BGE_MIN_FRAMELEN) {
3400 		error = m_devpad(m_head, BGE_MIN_FRAMELEN);
3401 		if (error)
3402 			goto back;
3403 	}
3404 
3405 	if ((sc->bge_flags & BGE_FLAG_SHORTDMA) && m_head->m_next != NULL) {
3406 		m_new = bge_defrag_shortdma(m_head);
3407 		if (m_new == NULL) {
3408 			error = ENOBUFS;
3409 			goto back;
3410 		}
3411 		*m_head0 = m_head = m_new;
3412 	}
3413 	if ((m_head->m_pkthdr.csum_flags & CSUM_TSO) == 0 &&
3414 	    sc->bge_force_defrag && (sc->bge_flags & BGE_FLAG_PCIE) &&
3415 	    m_head->m_next != NULL) {
3416 		/*
3417 		 * Forcefully defragment mbuf chain to overcome hardware
3418 		 * limitation which only support a single outstanding
3419 		 * DMA read operation.  If it fails, keep moving on using
3420 		 * the original mbuf chain.
3421 		 */
3422 		m_new = m_defrag(m_head, MB_DONTWAIT);
3423 		if (m_new != NULL)
3424 			*m_head0 = m_head = m_new;
3425 	}
3426 
3427 	error = bus_dmamap_load_mbuf_defrag(sc->bge_cdata.bge_tx_mtag, map,
3428 			m_head0, segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
3429 	if (error)
3430 		goto back;
3431 	*segs_used += nsegs;
3432 
3433 	m_head = *m_head0;
3434 	bus_dmamap_sync(sc->bge_cdata.bge_tx_mtag, map, BUS_DMASYNC_PREWRITE);
3435 
3436 	for (i = 0; ; i++) {
3437 		d = &sc->bge_ldata.bge_tx_ring[idx];
3438 
3439 		d->bge_addr.bge_addr_lo = BGE_ADDR_LO(segs[i].ds_addr);
3440 		d->bge_addr.bge_addr_hi = BGE_ADDR_HI(segs[i].ds_addr);
3441 		d->bge_len = segs[i].ds_len;
3442 		d->bge_flags = csum_flags;
3443 		d->bge_mss = mss;
3444 
3445 		if (i == nsegs - 1)
3446 			break;
3447 		BGE_INC(idx, BGE_TX_RING_CNT);
3448 	}
3449 	last_d = d;
3450 
3451 	/* Set vlan tag to the first segment of the packet. */
3452 	d = &sc->bge_ldata.bge_tx_ring[*txidx];
3453 	if (m_head->m_flags & M_VLANTAG) {
3454 		d->bge_flags |= BGE_TXBDFLAG_VLAN_TAG;
3455 		d->bge_vlan_tag = m_head->m_pkthdr.ether_vlantag;
3456 	} else {
3457 		d->bge_vlan_tag = 0;
3458 	}
3459 
3460 	/* Mark the last segment as end of packet... */
3461 	last_d->bge_flags |= BGE_TXBDFLAG_END;
3462 
3463 	/*
3464 	 * Insure that the map for this transmission is placed at
3465 	 * the array index of the last descriptor in this chain.
3466 	 */
3467 	sc->bge_cdata.bge_tx_dmamap[*txidx] = sc->bge_cdata.bge_tx_dmamap[idx];
3468 	sc->bge_cdata.bge_tx_dmamap[idx] = map;
3469 	sc->bge_cdata.bge_tx_chain[idx] = m_head;
3470 	sc->bge_txcnt += nsegs;
3471 
3472 	BGE_INC(idx, BGE_TX_RING_CNT);
3473 	*txidx = idx;
3474 back:
3475 	if (error) {
3476 		m_freem(*m_head0);
3477 		*m_head0 = NULL;
3478 	}
3479 	return error;
3480 }
3481 
3482 static void
3483 bge_xmit(struct bge_softc *sc, uint32_t prodidx)
3484 {
3485 	/* Transmit */
3486 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
3487 	/* 5700 b2 errata */
3488 	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
3489 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
3490 }
3491 
3492 /*
3493  * Main transmit routine. To avoid having to do mbuf copies, we put pointers
3494  * to the mbuf data regions directly in the transmit descriptors.
3495  */
3496 static void
3497 bge_start(struct ifnet *ifp, struct ifaltq_subque *ifsq)
3498 {
3499 	struct bge_softc *sc = ifp->if_softc;
3500 	struct mbuf *m_head = NULL;
3501 	uint32_t prodidx;
3502 	int nsegs = 0;
3503 
3504 	ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq);
3505 
3506 	if ((ifp->if_flags & IFF_RUNNING) == 0 || ifq_is_oactive(&ifp->if_snd))
3507 		return;
3508 
3509 	prodidx = sc->bge_tx_prodidx;
3510 
3511 	while (sc->bge_cdata.bge_tx_chain[prodidx] == NULL) {
3512 		m_head = ifq_dequeue(&ifp->if_snd, NULL);
3513 		if (m_head == NULL)
3514 			break;
3515 
3516 		/*
3517 		 * XXX
3518 		 * The code inside the if() block is never reached since we
3519 		 * must mark CSUM_IP_FRAGS in our if_hwassist to start getting
3520 		 * requests to checksum TCP/UDP in a fragmented packet.
3521 		 *
3522 		 * XXX
3523 		 * safety overkill.  If this is a fragmented packet chain
3524 		 * with delayed TCP/UDP checksums, then only encapsulate
3525 		 * it if we have enough descriptors to handle the entire
3526 		 * chain at once.
3527 		 * (paranoia -- may not actually be needed)
3528 		 */
3529 		if ((m_head->m_flags & M_FIRSTFRAG) &&
3530 		    (m_head->m_pkthdr.csum_flags & CSUM_DELAY_DATA)) {
3531 			if ((BGE_TX_RING_CNT - sc->bge_txcnt) <
3532 			    m_head->m_pkthdr.csum_data + sc->bge_txrsvd) {
3533 				ifq_set_oactive(&ifp->if_snd);
3534 				ifq_prepend(&ifp->if_snd, m_head);
3535 				break;
3536 			}
3537 		}
3538 
3539 		/*
3540 		 * Sanity check: avoid coming within bge_txrsvd
3541 		 * descriptors of the end of the ring.  Also make
3542 		 * sure there are bge_txspare descriptors for
3543 		 * jumbo buffers' defragmentation.
3544 		 */
3545 		if ((BGE_TX_RING_CNT - sc->bge_txcnt) <
3546 		    (sc->bge_txrsvd + sc->bge_txspare)) {
3547 			ifq_set_oactive(&ifp->if_snd);
3548 			ifq_prepend(&ifp->if_snd, m_head);
3549 			break;
3550 		}
3551 
3552 		/*
3553 		 * Pack the data into the transmit ring. If we
3554 		 * don't have room, set the OACTIVE flag and wait
3555 		 * for the NIC to drain the ring.
3556 		 */
3557 		if (bge_encap(sc, &m_head, &prodidx, &nsegs)) {
3558 			ifq_set_oactive(&ifp->if_snd);
3559 			IFNET_STAT_INC(ifp, oerrors, 1);
3560 			break;
3561 		}
3562 
3563 		if (nsegs >= sc->bge_tx_wreg) {
3564 			bge_xmit(sc, prodidx);
3565 			nsegs = 0;
3566 		}
3567 
3568 		ETHER_BPF_MTAP(ifp, m_head);
3569 
3570 		/*
3571 		 * Set a timeout in case the chip goes out to lunch.
3572 		 */
3573 		ifp->if_timer = 5;
3574 	}
3575 
3576 	if (nsegs > 0)
3577 		bge_xmit(sc, prodidx);
3578 	sc->bge_tx_prodidx = prodidx;
3579 }
3580 
3581 static void
3582 bge_init(void *xsc)
3583 {
3584 	struct bge_softc *sc = xsc;
3585 	struct ifnet *ifp = &sc->arpcom.ac_if;
3586 	uint16_t *m;
3587 	uint32_t mode;
3588 
3589 	ASSERT_SERIALIZED(ifp->if_serializer);
3590 
3591 	/* Cancel pending I/O and flush buffers. */
3592 	bge_stop(sc);
3593 	bge_reset(sc);
3594 	bge_chipinit(sc);
3595 
3596 	/*
3597 	 * Init the various state machines, ring
3598 	 * control blocks and firmware.
3599 	 */
3600 	if (bge_blockinit(sc)) {
3601 		if_printf(ifp, "initialization failure\n");
3602 		bge_stop(sc);
3603 		return;
3604 	}
3605 
3606 	/* Specify MTU. */
3607 	CSR_WRITE_4(sc, BGE_RX_MTU, ifp->if_mtu +
3608 	    ETHER_HDR_LEN + ETHER_CRC_LEN + EVL_ENCAPLEN);
3609 
3610 	/* Load our MAC address. */
3611 	m = (uint16_t *)&sc->arpcom.ac_enaddr[0];
3612 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_LO, htons(m[0]));
3613 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_HI, (htons(m[1]) << 16) | htons(m[2]));
3614 
3615 	/* Enable or disable promiscuous mode as needed. */
3616 	bge_setpromisc(sc);
3617 
3618 	/* Program multicast filter. */
3619 	bge_setmulti(sc);
3620 
3621 	/* Init RX ring. */
3622 	if (bge_init_rx_ring_std(sc)) {
3623 		if_printf(ifp, "RX ring initialization failed\n");
3624 		bge_stop(sc);
3625 		return;
3626 	}
3627 
3628 	/*
3629 	 * Workaround for a bug in 5705 ASIC rev A0. Poll the NIC's
3630 	 * memory to insure that the chip has in fact read the first
3631 	 * entry of the ring.
3632 	 */
3633 	if (sc->bge_chipid == BGE_CHIPID_BCM5705_A0) {
3634 		uint32_t		v, i;
3635 		for (i = 0; i < 10; i++) {
3636 			DELAY(20);
3637 			v = bge_readmem_ind(sc, BGE_STD_RX_RINGS + 8);
3638 			if (v == (MCLBYTES - ETHER_ALIGN))
3639 				break;
3640 		}
3641 		if (i == 10)
3642 			if_printf(ifp, "5705 A0 chip failed to load RX ring\n");
3643 	}
3644 
3645 	/* Init jumbo RX ring. */
3646 	if (ifp->if_mtu > (ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN)) {
3647 		if (bge_init_rx_ring_jumbo(sc)) {
3648 			if_printf(ifp, "Jumbo RX ring initialization failed\n");
3649 			bge_stop(sc);
3650 			return;
3651 		}
3652 	}
3653 
3654 	/* Init our RX return ring index */
3655 	sc->bge_rx_saved_considx = 0;
3656 
3657 	/* Init TX ring. */
3658 	bge_init_tx_ring(sc);
3659 
3660 	/* Enable TX MAC state machine lockup fix. */
3661 	mode = CSR_READ_4(sc, BGE_TX_MODE);
3662 	if (BGE_IS_5755_PLUS(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5906)
3663 		mode |= BGE_TXMODE_MBUF_LOCKUP_FIX;
3664 	/* Turn on transmitter */
3665 	CSR_WRITE_4(sc, BGE_TX_MODE, mode | BGE_TXMODE_ENABLE);
3666 
3667 	/* Turn on receiver */
3668 	BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
3669 
3670 	/*
3671 	 * Set the number of good frames to receive after RX MBUF
3672 	 * Low Watermark has been reached.  After the RX MAC receives
3673 	 * this number of frames, it will drop subsequent incoming
3674 	 * frames until the MBUF High Watermark is reached.
3675 	 */
3676 	CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 2);
3677 
3678 	if (sc->bge_irq_type == PCI_INTR_TYPE_MSI) {
3679 		if (bootverbose) {
3680 			if_printf(ifp, "MSI_MODE: %#x\n",
3681 			    CSR_READ_4(sc, BGE_MSI_MODE));
3682 		}
3683 
3684 		/*
3685 		 * XXX
3686 		 * Linux driver turns it on for all chips supporting MSI?!
3687 		 */
3688 		if (sc->bge_flags & BGE_FLAG_ONESHOT_MSI) {
3689 			/*
3690 			 * XXX
3691 			 * According to 5722-PG101-R,
3692 			 * BGE_PCIE_TRANSACT_ONESHOT_MSI applies only to
3693 			 * BCM5906.
3694 			 */
3695 			BGE_SETBIT(sc, BGE_PCIE_TRANSACT,
3696 			    BGE_PCIE_TRANSACT_ONESHOT_MSI);
3697 		}
3698 	}
3699 
3700 	/* Tell firmware we're alive. */
3701 	BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3702 
3703 	/* Enable host interrupts if polling(4) is not enabled. */
3704 	PCI_SETBIT(sc->bge_dev, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_CLEAR_INTA, 4);
3705 #ifdef IFPOLL_ENABLE
3706 	if (ifp->if_flags & IFF_NPOLLING)
3707 		bge_disable_intr(sc);
3708 	else
3709 #endif
3710 	bge_enable_intr(sc);
3711 
3712 	bge_ifmedia_upd(ifp);
3713 
3714 	ifp->if_flags |= IFF_RUNNING;
3715 	ifq_clr_oactive(&ifp->if_snd);
3716 
3717 	callout_reset(&sc->bge_stat_timer, hz, bge_tick, sc);
3718 }
3719 
3720 /*
3721  * Set media options.
3722  */
3723 static int
3724 bge_ifmedia_upd(struct ifnet *ifp)
3725 {
3726 	struct bge_softc *sc = ifp->if_softc;
3727 
3728 	/* If this is a 1000baseX NIC, enable the TBI port. */
3729 	if (sc->bge_flags & BGE_FLAG_TBI) {
3730 		struct ifmedia *ifm = &sc->bge_ifmedia;
3731 
3732 		if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
3733 			return(EINVAL);
3734 
3735 		switch(IFM_SUBTYPE(ifm->ifm_media)) {
3736 		case IFM_AUTO:
3737 			/*
3738 			 * The BCM5704 ASIC appears to have a special
3739 			 * mechanism for programming the autoneg
3740 			 * advertisement registers in TBI mode.
3741 			 */
3742 			if (!bge_fake_autoneg &&
3743 			    sc->bge_asicrev == BGE_ASICREV_BCM5704) {
3744 				uint32_t sgdig;
3745 
3746 				CSR_WRITE_4(sc, BGE_TX_TBI_AUTONEG, 0);
3747 				sgdig = CSR_READ_4(sc, BGE_SGDIG_CFG);
3748 				sgdig |= BGE_SGDIGCFG_AUTO |
3749 					 BGE_SGDIGCFG_PAUSE_CAP |
3750 					 BGE_SGDIGCFG_ASYM_PAUSE;
3751 				CSR_WRITE_4(sc, BGE_SGDIG_CFG,
3752 					    sgdig | BGE_SGDIGCFG_SEND);
3753 				DELAY(5);
3754 				CSR_WRITE_4(sc, BGE_SGDIG_CFG, sgdig);
3755 			}
3756 			break;
3757 		case IFM_1000_SX:
3758 			if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX) {
3759 				BGE_CLRBIT(sc, BGE_MAC_MODE,
3760 				    BGE_MACMODE_HALF_DUPLEX);
3761 			} else {
3762 				BGE_SETBIT(sc, BGE_MAC_MODE,
3763 				    BGE_MACMODE_HALF_DUPLEX);
3764 			}
3765 			break;
3766 		default:
3767 			return(EINVAL);
3768 		}
3769 	} else {
3770 		struct mii_data *mii = device_get_softc(sc->bge_miibus);
3771 
3772 		sc->bge_link_evt++;
3773 		sc->bge_link = 0;
3774 		if (mii->mii_instance) {
3775 			struct mii_softc *miisc;
3776 
3777 			LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
3778 				mii_phy_reset(miisc);
3779 		}
3780 		mii_mediachg(mii);
3781 
3782 		/*
3783 		 * Force an interrupt so that we will call bge_link_upd
3784 		 * if needed and clear any pending link state attention.
3785 		 * Without this we are not getting any further interrupts
3786 		 * for link state changes and thus will not UP the link and
3787 		 * not be able to send in bge_start.  The only way to get
3788 		 * things working was to receive a packet and get an RX
3789 		 * intr.
3790 		 *
3791 		 * bge_tick should help for fiber cards and we might not
3792 		 * need to do this here if BGE_FLAG_TBI is set but as
3793 		 * we poll for fiber anyway it should not harm.
3794 		 */
3795 		if (BGE_IS_CRIPPLED(sc))
3796 			BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
3797 		else
3798 			BGE_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
3799 	}
3800 	return(0);
3801 }
3802 
3803 /*
3804  * Report current media status.
3805  */
3806 static void
3807 bge_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3808 {
3809 	struct bge_softc *sc = ifp->if_softc;
3810 
3811 	if (sc->bge_flags & BGE_FLAG_TBI) {
3812 		ifmr->ifm_status = IFM_AVALID;
3813 		ifmr->ifm_active = IFM_ETHER;
3814 		if (CSR_READ_4(sc, BGE_MAC_STS) &
3815 		    BGE_MACSTAT_TBI_PCS_SYNCHED) {
3816 			ifmr->ifm_status |= IFM_ACTIVE;
3817 		} else {
3818 			ifmr->ifm_active |= IFM_NONE;
3819 			return;
3820 		}
3821 
3822 		ifmr->ifm_active |= IFM_1000_SX;
3823 		if (CSR_READ_4(sc, BGE_MAC_MODE) & BGE_MACMODE_HALF_DUPLEX)
3824 			ifmr->ifm_active |= IFM_HDX;
3825 		else
3826 			ifmr->ifm_active |= IFM_FDX;
3827 	} else {
3828 		struct mii_data *mii = device_get_softc(sc->bge_miibus);
3829 
3830 		mii_pollstat(mii);
3831 		ifmr->ifm_active = mii->mii_media_active;
3832 		ifmr->ifm_status = mii->mii_media_status;
3833 	}
3834 }
3835 
3836 static int
3837 bge_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
3838 {
3839 	struct bge_softc *sc = ifp->if_softc;
3840 	struct ifreq *ifr = (struct ifreq *)data;
3841 	int mask, error = 0;
3842 
3843 	ASSERT_SERIALIZED(ifp->if_serializer);
3844 
3845 	switch (command) {
3846 	case SIOCSIFMTU:
3847 		if ((!BGE_IS_JUMBO_CAPABLE(sc) && ifr->ifr_mtu > ETHERMTU) ||
3848 		    (BGE_IS_JUMBO_CAPABLE(sc) &&
3849 		     ifr->ifr_mtu > BGE_JUMBO_MTU)) {
3850 			error = EINVAL;
3851 		} else if (ifp->if_mtu != ifr->ifr_mtu) {
3852 			ifp->if_mtu = ifr->ifr_mtu;
3853 			if (ifp->if_flags & IFF_RUNNING)
3854 				bge_init(sc);
3855 		}
3856 		break;
3857 	case SIOCSIFFLAGS:
3858 		if (ifp->if_flags & IFF_UP) {
3859 			if (ifp->if_flags & IFF_RUNNING) {
3860 				mask = ifp->if_flags ^ sc->bge_if_flags;
3861 
3862 				/*
3863 				 * If only the state of the PROMISC flag
3864 				 * changed, then just use the 'set promisc
3865 				 * mode' command instead of reinitializing
3866 				 * the entire NIC. Doing a full re-init
3867 				 * means reloading the firmware and waiting
3868 				 * for it to start up, which may take a
3869 				 * second or two.  Similarly for ALLMULTI.
3870 				 */
3871 				if (mask & IFF_PROMISC)
3872 					bge_setpromisc(sc);
3873 				if (mask & IFF_ALLMULTI)
3874 					bge_setmulti(sc);
3875 			} else {
3876 				bge_init(sc);
3877 			}
3878 		} else if (ifp->if_flags & IFF_RUNNING) {
3879 			bge_stop(sc);
3880 		}
3881 		sc->bge_if_flags = ifp->if_flags;
3882 		break;
3883 	case SIOCADDMULTI:
3884 	case SIOCDELMULTI:
3885 		if (ifp->if_flags & IFF_RUNNING)
3886 			bge_setmulti(sc);
3887 		break;
3888 	case SIOCSIFMEDIA:
3889 	case SIOCGIFMEDIA:
3890 		if (sc->bge_flags & BGE_FLAG_TBI) {
3891 			error = ifmedia_ioctl(ifp, ifr,
3892 			    &sc->bge_ifmedia, command);
3893 		} else {
3894 			struct mii_data *mii;
3895 
3896 			mii = device_get_softc(sc->bge_miibus);
3897 			error = ifmedia_ioctl(ifp, ifr,
3898 					      &mii->mii_media, command);
3899 		}
3900 		break;
3901         case SIOCSIFCAP:
3902 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3903 		if (mask & IFCAP_HWCSUM) {
3904 			ifp->if_capenable ^= (mask & IFCAP_HWCSUM);
3905 			if (ifp->if_capenable & IFCAP_TXCSUM)
3906 				ifp->if_hwassist |= BGE_CSUM_FEATURES;
3907 			else
3908 				ifp->if_hwassist &= ~BGE_CSUM_FEATURES;
3909 		}
3910 		if (mask & IFCAP_TSO) {
3911 			ifp->if_capenable ^= IFCAP_TSO;
3912 			if (ifp->if_capenable & IFCAP_TSO)
3913 				ifp->if_hwassist |= CSUM_TSO;
3914 			else
3915 				ifp->if_hwassist &= ~CSUM_TSO;
3916 		}
3917 		break;
3918 	default:
3919 		error = ether_ioctl(ifp, command, data);
3920 		break;
3921 	}
3922 	return error;
3923 }
3924 
3925 static void
3926 bge_watchdog(struct ifnet *ifp)
3927 {
3928 	struct bge_softc *sc = ifp->if_softc;
3929 
3930 	if_printf(ifp, "watchdog timeout -- resetting\n");
3931 
3932 	bge_init(sc);
3933 
3934 	IFNET_STAT_INC(ifp, oerrors, 1);
3935 
3936 	if (!ifq_is_empty(&ifp->if_snd))
3937 		if_devstart(ifp);
3938 }
3939 
3940 /*
3941  * Stop the adapter and free any mbufs allocated to the
3942  * RX and TX lists.
3943  */
3944 static void
3945 bge_stop(struct bge_softc *sc)
3946 {
3947 	struct ifnet *ifp = &sc->arpcom.ac_if;
3948 
3949 	ASSERT_SERIALIZED(ifp->if_serializer);
3950 
3951 	callout_stop(&sc->bge_stat_timer);
3952 
3953 	/*
3954 	 * Disable all of the receiver blocks
3955 	 */
3956 	bge_stop_block(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
3957 	bge_stop_block(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
3958 	bge_stop_block(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
3959 	if (BGE_IS_5700_FAMILY(sc))
3960 		bge_stop_block(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
3961 	bge_stop_block(sc, BGE_RDBDI_MODE, BGE_RBDIMODE_ENABLE);
3962 	bge_stop_block(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
3963 	bge_stop_block(sc, BGE_RBDC_MODE, BGE_RBDCMODE_ENABLE);
3964 
3965 	/*
3966 	 * Disable all of the transmit blocks
3967 	 */
3968 	bge_stop_block(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
3969 	bge_stop_block(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
3970 	bge_stop_block(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
3971 	bge_stop_block(sc, BGE_RDMA_MODE, BGE_RDMAMODE_ENABLE);
3972 	bge_stop_block(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE);
3973 	if (BGE_IS_5700_FAMILY(sc))
3974 		bge_stop_block(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
3975 	bge_stop_block(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
3976 
3977 	/*
3978 	 * Shut down all of the memory managers and related
3979 	 * state machines.
3980 	 */
3981 	bge_stop_block(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE);
3982 	bge_stop_block(sc, BGE_WDMA_MODE, BGE_WDMAMODE_ENABLE);
3983 	if (BGE_IS_5700_FAMILY(sc))
3984 		bge_stop_block(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
3985 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
3986 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
3987 	if (!BGE_IS_5705_PLUS(sc)) {
3988 		BGE_CLRBIT(sc, BGE_BMAN_MODE, BGE_BMANMODE_ENABLE);
3989 		BGE_CLRBIT(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
3990 	}
3991 
3992 	/* Disable host interrupts. */
3993 	bge_disable_intr(sc);
3994 
3995 	/*
3996 	 * Tell firmware we're shutting down.
3997 	 */
3998 	BGE_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3999 
4000 	/* Free the RX lists. */
4001 	bge_free_rx_ring_std(sc);
4002 
4003 	/* Free jumbo RX list. */
4004 	if (BGE_IS_JUMBO_CAPABLE(sc))
4005 		bge_free_rx_ring_jumbo(sc);
4006 
4007 	/* Free TX buffers. */
4008 	bge_free_tx_ring(sc);
4009 
4010 	sc->bge_status_tag = 0;
4011 	sc->bge_link = 0;
4012 	sc->bge_coal_chg = 0;
4013 
4014 	sc->bge_tx_saved_considx = BGE_TXCONS_UNSET;
4015 
4016 	ifp->if_flags &= ~IFF_RUNNING;
4017 	ifq_clr_oactive(&ifp->if_snd);
4018 	ifp->if_timer = 0;
4019 }
4020 
4021 /*
4022  * Stop all chip I/O so that the kernel's probe routines don't
4023  * get confused by errant DMAs when rebooting.
4024  */
4025 static void
4026 bge_shutdown(device_t dev)
4027 {
4028 	struct bge_softc *sc = device_get_softc(dev);
4029 	struct ifnet *ifp = &sc->arpcom.ac_if;
4030 
4031 	lwkt_serialize_enter(ifp->if_serializer);
4032 	bge_stop(sc);
4033 	bge_reset(sc);
4034 	lwkt_serialize_exit(ifp->if_serializer);
4035 }
4036 
4037 static int
4038 bge_suspend(device_t dev)
4039 {
4040 	struct bge_softc *sc = device_get_softc(dev);
4041 	struct ifnet *ifp = &sc->arpcom.ac_if;
4042 
4043 	lwkt_serialize_enter(ifp->if_serializer);
4044 	bge_stop(sc);
4045 	lwkt_serialize_exit(ifp->if_serializer);
4046 
4047 	return 0;
4048 }
4049 
4050 static int
4051 bge_resume(device_t dev)
4052 {
4053 	struct bge_softc *sc = device_get_softc(dev);
4054 	struct ifnet *ifp = &sc->arpcom.ac_if;
4055 
4056 	lwkt_serialize_enter(ifp->if_serializer);
4057 
4058 	if (ifp->if_flags & IFF_UP) {
4059 		bge_init(sc);
4060 
4061 		if (!ifq_is_empty(&ifp->if_snd))
4062 			if_devstart(ifp);
4063 	}
4064 
4065 	lwkt_serialize_exit(ifp->if_serializer);
4066 
4067 	return 0;
4068 }
4069 
4070 static void
4071 bge_setpromisc(struct bge_softc *sc)
4072 {
4073 	struct ifnet *ifp = &sc->arpcom.ac_if;
4074 
4075 	if (ifp->if_flags & IFF_PROMISC)
4076 		BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
4077 	else
4078 		BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
4079 }
4080 
4081 static void
4082 bge_dma_free(struct bge_softc *sc)
4083 {
4084 	int i;
4085 
4086 	/* Destroy RX mbuf DMA stuffs. */
4087 	if (sc->bge_cdata.bge_rx_mtag != NULL) {
4088 		for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
4089 			bus_dmamap_destroy(sc->bge_cdata.bge_rx_mtag,
4090 			    sc->bge_cdata.bge_rx_std_dmamap[i]);
4091 		}
4092 		bus_dmamap_destroy(sc->bge_cdata.bge_rx_mtag,
4093 				   sc->bge_cdata.bge_rx_tmpmap);
4094 		bus_dma_tag_destroy(sc->bge_cdata.bge_rx_mtag);
4095 	}
4096 
4097 	/* Destroy TX mbuf DMA stuffs. */
4098 	if (sc->bge_cdata.bge_tx_mtag != NULL) {
4099 		for (i = 0; i < BGE_TX_RING_CNT; i++) {
4100 			bus_dmamap_destroy(sc->bge_cdata.bge_tx_mtag,
4101 			    sc->bge_cdata.bge_tx_dmamap[i]);
4102 		}
4103 		bus_dma_tag_destroy(sc->bge_cdata.bge_tx_mtag);
4104 	}
4105 
4106 	/* Destroy standard RX ring */
4107 	bge_dma_block_free(sc->bge_cdata.bge_rx_std_ring_tag,
4108 			   sc->bge_cdata.bge_rx_std_ring_map,
4109 			   sc->bge_ldata.bge_rx_std_ring);
4110 
4111 	if (BGE_IS_JUMBO_CAPABLE(sc))
4112 		bge_free_jumbo_mem(sc);
4113 
4114 	/* Destroy RX return ring */
4115 	bge_dma_block_free(sc->bge_cdata.bge_rx_return_ring_tag,
4116 			   sc->bge_cdata.bge_rx_return_ring_map,
4117 			   sc->bge_ldata.bge_rx_return_ring);
4118 
4119 	/* Destroy TX ring */
4120 	bge_dma_block_free(sc->bge_cdata.bge_tx_ring_tag,
4121 			   sc->bge_cdata.bge_tx_ring_map,
4122 			   sc->bge_ldata.bge_tx_ring);
4123 
4124 	/* Destroy status block */
4125 	bge_dma_block_free(sc->bge_cdata.bge_status_tag,
4126 			   sc->bge_cdata.bge_status_map,
4127 			   sc->bge_ldata.bge_status_block);
4128 
4129 	/* Destroy statistics block */
4130 	bge_dma_block_free(sc->bge_cdata.bge_stats_tag,
4131 			   sc->bge_cdata.bge_stats_map,
4132 			   sc->bge_ldata.bge_stats);
4133 
4134 	/* Destroy the parent tag */
4135 	if (sc->bge_cdata.bge_parent_tag != NULL)
4136 		bus_dma_tag_destroy(sc->bge_cdata.bge_parent_tag);
4137 }
4138 
4139 static int
4140 bge_dma_alloc(struct bge_softc *sc)
4141 {
4142 	struct ifnet *ifp = &sc->arpcom.ac_if;
4143 	int i, error;
4144 	bus_addr_t lowaddr;
4145 	bus_size_t txmaxsz;
4146 
4147 	lowaddr = BUS_SPACE_MAXADDR;
4148 	if (sc->bge_flags & BGE_FLAG_MAXADDR_40BIT)
4149 		lowaddr = BGE_DMA_MAXADDR_40BIT;
4150 
4151 	/*
4152 	 * Allocate the parent bus DMA tag appropriate for PCI.
4153 	 *
4154 	 * All of the NetExtreme/NetLink controllers have 4GB boundary
4155 	 * DMA bug.
4156 	 * Whenever an address crosses a multiple of the 4GB boundary
4157 	 * (including 4GB, 8Gb, 12Gb, etc.) and makes the transition
4158 	 * from 0xX_FFFF_FFFF to 0x(X+1)_0000_0000 an internal DMA
4159 	 * state machine will lockup and cause the device to hang.
4160 	 */
4161 	error = bus_dma_tag_create(NULL, 1, BGE_DMA_BOUNDARY_4G,
4162 				   lowaddr, BUS_SPACE_MAXADDR,
4163 				   NULL, NULL,
4164 				   BUS_SPACE_MAXSIZE_32BIT, 0,
4165 				   BUS_SPACE_MAXSIZE_32BIT,
4166 				   0, &sc->bge_cdata.bge_parent_tag);
4167 	if (error) {
4168 		if_printf(ifp, "could not allocate parent dma tag\n");
4169 		return error;
4170 	}
4171 
4172 	/*
4173 	 * Create DMA tag and maps for RX mbufs.
4174 	 */
4175 	error = bus_dma_tag_create(sc->bge_cdata.bge_parent_tag, 1, 0,
4176 				   BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
4177 				   NULL, NULL, MCLBYTES, 1, MCLBYTES,
4178 				   BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK,
4179 				   &sc->bge_cdata.bge_rx_mtag);
4180 	if (error) {
4181 		if_printf(ifp, "could not allocate RX mbuf dma tag\n");
4182 		return error;
4183 	}
4184 
4185 	error = bus_dmamap_create(sc->bge_cdata.bge_rx_mtag,
4186 				  BUS_DMA_WAITOK, &sc->bge_cdata.bge_rx_tmpmap);
4187 	if (error) {
4188 		bus_dma_tag_destroy(sc->bge_cdata.bge_rx_mtag);
4189 		sc->bge_cdata.bge_rx_mtag = NULL;
4190 		return error;
4191 	}
4192 
4193 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
4194 		error = bus_dmamap_create(sc->bge_cdata.bge_rx_mtag,
4195 					  BUS_DMA_WAITOK,
4196 					  &sc->bge_cdata.bge_rx_std_dmamap[i]);
4197 		if (error) {
4198 			int j;
4199 
4200 			for (j = 0; j < i; ++j) {
4201 				bus_dmamap_destroy(sc->bge_cdata.bge_rx_mtag,
4202 					sc->bge_cdata.bge_rx_std_dmamap[j]);
4203 			}
4204 			bus_dma_tag_destroy(sc->bge_cdata.bge_rx_mtag);
4205 			sc->bge_cdata.bge_rx_mtag = NULL;
4206 
4207 			if_printf(ifp, "could not create DMA map for RX\n");
4208 			return error;
4209 		}
4210 	}
4211 
4212 	/*
4213 	 * Create DMA tag and maps for TX mbufs.
4214 	 */
4215 	if (sc->bge_flags & BGE_FLAG_TSO)
4216 		txmaxsz = IP_MAXPACKET + sizeof(struct ether_vlan_header);
4217 	else
4218 		txmaxsz = BGE_JUMBO_FRAMELEN;
4219 	error = bus_dma_tag_create(sc->bge_cdata.bge_parent_tag, 1, 0,
4220 				   BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
4221 				   NULL, NULL,
4222 				   txmaxsz, BGE_NSEG_NEW, PAGE_SIZE,
4223 				   BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK |
4224 				   BUS_DMA_ONEBPAGE,
4225 				   &sc->bge_cdata.bge_tx_mtag);
4226 	if (error) {
4227 		if_printf(ifp, "could not allocate TX mbuf dma tag\n");
4228 		return error;
4229 	}
4230 
4231 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
4232 		error = bus_dmamap_create(sc->bge_cdata.bge_tx_mtag,
4233 					  BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
4234 					  &sc->bge_cdata.bge_tx_dmamap[i]);
4235 		if (error) {
4236 			int j;
4237 
4238 			for (j = 0; j < i; ++j) {
4239 				bus_dmamap_destroy(sc->bge_cdata.bge_tx_mtag,
4240 					sc->bge_cdata.bge_tx_dmamap[j]);
4241 			}
4242 			bus_dma_tag_destroy(sc->bge_cdata.bge_tx_mtag);
4243 			sc->bge_cdata.bge_tx_mtag = NULL;
4244 
4245 			if_printf(ifp, "could not create DMA map for TX\n");
4246 			return error;
4247 		}
4248 	}
4249 
4250 	/*
4251 	 * Create DMA stuffs for standard RX ring.
4252 	 */
4253 	error = bge_dma_block_alloc(sc, BGE_STD_RX_RING_SZ,
4254 				    &sc->bge_cdata.bge_rx_std_ring_tag,
4255 				    &sc->bge_cdata.bge_rx_std_ring_map,
4256 				    (void *)&sc->bge_ldata.bge_rx_std_ring,
4257 				    &sc->bge_ldata.bge_rx_std_ring_paddr);
4258 	if (error) {
4259 		if_printf(ifp, "could not create std RX ring\n");
4260 		return error;
4261 	}
4262 
4263 	/*
4264 	 * Create jumbo buffer pool.
4265 	 */
4266 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
4267 		error = bge_alloc_jumbo_mem(sc);
4268 		if (error) {
4269 			if_printf(ifp, "could not create jumbo buffer pool\n");
4270 			return error;
4271 		}
4272 	}
4273 
4274 	/*
4275 	 * Create DMA stuffs for RX return ring.
4276 	 */
4277 	error = bge_dma_block_alloc(sc,
4278 	    BGE_RX_RTN_RING_SZ(sc->bge_return_ring_cnt),
4279 	    &sc->bge_cdata.bge_rx_return_ring_tag,
4280 	    &sc->bge_cdata.bge_rx_return_ring_map,
4281 	    (void *)&sc->bge_ldata.bge_rx_return_ring,
4282 	    &sc->bge_ldata.bge_rx_return_ring_paddr);
4283 	if (error) {
4284 		if_printf(ifp, "could not create RX ret ring\n");
4285 		return error;
4286 	}
4287 
4288 	/*
4289 	 * Create DMA stuffs for TX ring.
4290 	 */
4291 	error = bge_dma_block_alloc(sc, BGE_TX_RING_SZ,
4292 				    &sc->bge_cdata.bge_tx_ring_tag,
4293 				    &sc->bge_cdata.bge_tx_ring_map,
4294 				    (void *)&sc->bge_ldata.bge_tx_ring,
4295 				    &sc->bge_ldata.bge_tx_ring_paddr);
4296 	if (error) {
4297 		if_printf(ifp, "could not create TX ring\n");
4298 		return error;
4299 	}
4300 
4301 	/*
4302 	 * Create DMA stuffs for status block.
4303 	 */
4304 	error = bge_dma_block_alloc(sc, BGE_STATUS_BLK_SZ,
4305 				    &sc->bge_cdata.bge_status_tag,
4306 				    &sc->bge_cdata.bge_status_map,
4307 				    (void *)&sc->bge_ldata.bge_status_block,
4308 				    &sc->bge_ldata.bge_status_block_paddr);
4309 	if (error) {
4310 		if_printf(ifp, "could not create status block\n");
4311 		return error;
4312 	}
4313 
4314 	/*
4315 	 * Create DMA stuffs for statistics block.
4316 	 */
4317 	error = bge_dma_block_alloc(sc, BGE_STATS_SZ,
4318 				    &sc->bge_cdata.bge_stats_tag,
4319 				    &sc->bge_cdata.bge_stats_map,
4320 				    (void *)&sc->bge_ldata.bge_stats,
4321 				    &sc->bge_ldata.bge_stats_paddr);
4322 	if (error) {
4323 		if_printf(ifp, "could not create stats block\n");
4324 		return error;
4325 	}
4326 	return 0;
4327 }
4328 
4329 static int
4330 bge_dma_block_alloc(struct bge_softc *sc, bus_size_t size, bus_dma_tag_t *tag,
4331 		    bus_dmamap_t *map, void **addr, bus_addr_t *paddr)
4332 {
4333 	bus_dmamem_t dmem;
4334 	int error;
4335 
4336 	error = bus_dmamem_coherent(sc->bge_cdata.bge_parent_tag, PAGE_SIZE, 0,
4337 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
4338 				    size, BUS_DMA_WAITOK | BUS_DMA_ZERO, &dmem);
4339 	if (error)
4340 		return error;
4341 
4342 	*tag = dmem.dmem_tag;
4343 	*map = dmem.dmem_map;
4344 	*addr = dmem.dmem_addr;
4345 	*paddr = dmem.dmem_busaddr;
4346 
4347 	return 0;
4348 }
4349 
4350 static void
4351 bge_dma_block_free(bus_dma_tag_t tag, bus_dmamap_t map, void *addr)
4352 {
4353 	if (tag != NULL) {
4354 		bus_dmamap_unload(tag, map);
4355 		bus_dmamem_free(tag, addr, map);
4356 		bus_dma_tag_destroy(tag);
4357 	}
4358 }
4359 
4360 /*
4361  * Grrr. The link status word in the status block does
4362  * not work correctly on the BCM5700 rev AX and BX chips,
4363  * according to all available information. Hence, we have
4364  * to enable MII interrupts in order to properly obtain
4365  * async link changes. Unfortunately, this also means that
4366  * we have to read the MAC status register to detect link
4367  * changes, thereby adding an additional register access to
4368  * the interrupt handler.
4369  *
4370  * XXX: perhaps link state detection procedure used for
4371  * BGE_CHIPID_BCM5700_B2 can be used for others BCM5700 revisions.
4372  */
4373 static void
4374 bge_bcm5700_link_upd(struct bge_softc *sc, uint32_t status __unused)
4375 {
4376 	struct ifnet *ifp = &sc->arpcom.ac_if;
4377 	struct mii_data *mii = device_get_softc(sc->bge_miibus);
4378 
4379 	mii_pollstat(mii);
4380 
4381 	if (!sc->bge_link &&
4382 	    (mii->mii_media_status & IFM_ACTIVE) &&
4383 	    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
4384 		sc->bge_link++;
4385 		if (bootverbose)
4386 			if_printf(ifp, "link UP\n");
4387 	} else if (sc->bge_link &&
4388 	    (!(mii->mii_media_status & IFM_ACTIVE) ||
4389 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE)) {
4390 		sc->bge_link = 0;
4391 		if (bootverbose)
4392 			if_printf(ifp, "link DOWN\n");
4393 	}
4394 
4395 	/* Clear the interrupt. */
4396 	CSR_WRITE_4(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_MI_INTERRUPT);
4397 	bge_miibus_readreg(sc->bge_dev, 1, BRGPHY_MII_ISR);
4398 	bge_miibus_writereg(sc->bge_dev, 1, BRGPHY_MII_IMR, BRGPHY_INTRS);
4399 }
4400 
4401 static void
4402 bge_tbi_link_upd(struct bge_softc *sc, uint32_t status)
4403 {
4404 	struct ifnet *ifp = &sc->arpcom.ac_if;
4405 
4406 #define PCS_ENCODE_ERR	(BGE_MACSTAT_PORT_DECODE_ERROR|BGE_MACSTAT_MI_COMPLETE)
4407 
4408 	/*
4409 	 * Sometimes PCS encoding errors are detected in
4410 	 * TBI mode (on fiber NICs), and for some reason
4411 	 * the chip will signal them as link changes.
4412 	 * If we get a link change event, but the 'PCS
4413 	 * encoding error' bit in the MAC status register
4414 	 * is set, don't bother doing a link check.
4415 	 * This avoids spurious "gigabit link up" messages
4416 	 * that sometimes appear on fiber NICs during
4417 	 * periods of heavy traffic.
4418 	 */
4419 	if (status & BGE_MACSTAT_TBI_PCS_SYNCHED) {
4420 		if (!sc->bge_link) {
4421 			sc->bge_link++;
4422 			if (sc->bge_asicrev == BGE_ASICREV_BCM5704) {
4423 				BGE_CLRBIT(sc, BGE_MAC_MODE,
4424 				    BGE_MACMODE_TBI_SEND_CFGS);
4425 			}
4426 			CSR_WRITE_4(sc, BGE_MAC_STS, 0xFFFFFFFF);
4427 
4428 			if (bootverbose)
4429 				if_printf(ifp, "link UP\n");
4430 
4431 			ifp->if_link_state = LINK_STATE_UP;
4432 			if_link_state_change(ifp);
4433 		}
4434 	} else if ((status & PCS_ENCODE_ERR) != PCS_ENCODE_ERR) {
4435 		if (sc->bge_link) {
4436 			sc->bge_link = 0;
4437 
4438 			if (bootverbose)
4439 				if_printf(ifp, "link DOWN\n");
4440 
4441 			ifp->if_link_state = LINK_STATE_DOWN;
4442 			if_link_state_change(ifp);
4443 		}
4444 	}
4445 
4446 #undef PCS_ENCODE_ERR
4447 
4448 	/* Clear the attention. */
4449 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
4450 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
4451 	    BGE_MACSTAT_LINK_CHANGED);
4452 }
4453 
4454 static void
4455 bge_copper_link_upd(struct bge_softc *sc, uint32_t status __unused)
4456 {
4457 	struct ifnet *ifp = &sc->arpcom.ac_if;
4458 	struct mii_data *mii = device_get_softc(sc->bge_miibus);
4459 
4460 	mii_pollstat(mii);
4461 	bge_miibus_statchg(sc->bge_dev);
4462 
4463 	if (bootverbose) {
4464 		if (sc->bge_link)
4465 			if_printf(ifp, "link UP\n");
4466 		else
4467 			if_printf(ifp, "link DOWN\n");
4468 	}
4469 
4470 	/* Clear the attention. */
4471 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
4472 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
4473 	    BGE_MACSTAT_LINK_CHANGED);
4474 }
4475 
4476 static void
4477 bge_autopoll_link_upd(struct bge_softc *sc, uint32_t status __unused)
4478 {
4479 	struct ifnet *ifp = &sc->arpcom.ac_if;
4480 	struct mii_data *mii = device_get_softc(sc->bge_miibus);
4481 
4482 	mii_pollstat(mii);
4483 
4484 	if (!sc->bge_link &&
4485 	    (mii->mii_media_status & IFM_ACTIVE) &&
4486 	    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
4487 		sc->bge_link++;
4488 		if (bootverbose)
4489 			if_printf(ifp, "link UP\n");
4490 	} else if (sc->bge_link &&
4491 	    (!(mii->mii_media_status & IFM_ACTIVE) ||
4492 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE)) {
4493 		sc->bge_link = 0;
4494 		if (bootverbose)
4495 			if_printf(ifp, "link DOWN\n");
4496 	}
4497 
4498 	/* Clear the attention. */
4499 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
4500 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
4501 	    BGE_MACSTAT_LINK_CHANGED);
4502 }
4503 
4504 static int
4505 bge_sysctl_rx_coal_ticks(SYSCTL_HANDLER_ARGS)
4506 {
4507 	struct bge_softc *sc = arg1;
4508 
4509 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4510 	    &sc->bge_rx_coal_ticks,
4511 	    BGE_RX_COAL_TICKS_MIN, BGE_RX_COAL_TICKS_MAX,
4512 	    BGE_RX_COAL_TICKS_CHG);
4513 }
4514 
4515 static int
4516 bge_sysctl_tx_coal_ticks(SYSCTL_HANDLER_ARGS)
4517 {
4518 	struct bge_softc *sc = arg1;
4519 
4520 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4521 	    &sc->bge_tx_coal_ticks,
4522 	    BGE_TX_COAL_TICKS_MIN, BGE_TX_COAL_TICKS_MAX,
4523 	    BGE_TX_COAL_TICKS_CHG);
4524 }
4525 
4526 static int
4527 bge_sysctl_rx_coal_bds(SYSCTL_HANDLER_ARGS)
4528 {
4529 	struct bge_softc *sc = arg1;
4530 
4531 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4532 	    &sc->bge_rx_coal_bds,
4533 	    BGE_RX_COAL_BDS_MIN, BGE_RX_COAL_BDS_MAX,
4534 	    BGE_RX_COAL_BDS_CHG);
4535 }
4536 
4537 static int
4538 bge_sysctl_tx_coal_bds(SYSCTL_HANDLER_ARGS)
4539 {
4540 	struct bge_softc *sc = arg1;
4541 
4542 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4543 	    &sc->bge_tx_coal_bds,
4544 	    BGE_TX_COAL_BDS_MIN, BGE_TX_COAL_BDS_MAX,
4545 	    BGE_TX_COAL_BDS_CHG);
4546 }
4547 
4548 static int
4549 bge_sysctl_rx_coal_ticks_int(SYSCTL_HANDLER_ARGS)
4550 {
4551 	struct bge_softc *sc = arg1;
4552 
4553 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4554 	    &sc->bge_rx_coal_ticks_int,
4555 	    BGE_RX_COAL_TICKS_MIN, BGE_RX_COAL_TICKS_MAX,
4556 	    BGE_RX_COAL_TICKS_INT_CHG);
4557 }
4558 
4559 static int
4560 bge_sysctl_tx_coal_ticks_int(SYSCTL_HANDLER_ARGS)
4561 {
4562 	struct bge_softc *sc = arg1;
4563 
4564 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4565 	    &sc->bge_tx_coal_ticks_int,
4566 	    BGE_TX_COAL_TICKS_MIN, BGE_TX_COAL_TICKS_MAX,
4567 	    BGE_TX_COAL_TICKS_INT_CHG);
4568 }
4569 
4570 static int
4571 bge_sysctl_rx_coal_bds_int(SYSCTL_HANDLER_ARGS)
4572 {
4573 	struct bge_softc *sc = arg1;
4574 
4575 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4576 	    &sc->bge_rx_coal_bds_int,
4577 	    BGE_RX_COAL_BDS_MIN, BGE_RX_COAL_BDS_MAX,
4578 	    BGE_RX_COAL_BDS_INT_CHG);
4579 }
4580 
4581 static int
4582 bge_sysctl_tx_coal_bds_int(SYSCTL_HANDLER_ARGS)
4583 {
4584 	struct bge_softc *sc = arg1;
4585 
4586 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4587 	    &sc->bge_tx_coal_bds_int,
4588 	    BGE_TX_COAL_BDS_MIN, BGE_TX_COAL_BDS_MAX,
4589 	    BGE_TX_COAL_BDS_INT_CHG);
4590 }
4591 
4592 static int
4593 bge_sysctl_coal_chg(SYSCTL_HANDLER_ARGS, uint32_t *coal,
4594     int coal_min, int coal_max, uint32_t coal_chg_mask)
4595 {
4596 	struct bge_softc *sc = arg1;
4597 	struct ifnet *ifp = &sc->arpcom.ac_if;
4598 	int error = 0, v;
4599 
4600 	lwkt_serialize_enter(ifp->if_serializer);
4601 
4602 	v = *coal;
4603 	error = sysctl_handle_int(oidp, &v, 0, req);
4604 	if (!error && req->newptr != NULL) {
4605 		if (v < coal_min || v > coal_max) {
4606 			error = EINVAL;
4607 		} else {
4608 			*coal = v;
4609 			sc->bge_coal_chg |= coal_chg_mask;
4610 		}
4611 	}
4612 
4613 	lwkt_serialize_exit(ifp->if_serializer);
4614 	return error;
4615 }
4616 
4617 static void
4618 bge_coal_change(struct bge_softc *sc)
4619 {
4620 	struct ifnet *ifp = &sc->arpcom.ac_if;
4621 
4622 	ASSERT_SERIALIZED(ifp->if_serializer);
4623 
4624 	if (sc->bge_coal_chg & BGE_RX_COAL_TICKS_CHG) {
4625 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS,
4626 			    sc->bge_rx_coal_ticks);
4627 		DELAY(10);
4628 		CSR_READ_4(sc, BGE_HCC_RX_COAL_TICKS);
4629 
4630 		if (bootverbose) {
4631 			if_printf(ifp, "rx_coal_ticks -> %u\n",
4632 				  sc->bge_rx_coal_ticks);
4633 		}
4634 	}
4635 
4636 	if (sc->bge_coal_chg & BGE_TX_COAL_TICKS_CHG) {
4637 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS,
4638 			    sc->bge_tx_coal_ticks);
4639 		DELAY(10);
4640 		CSR_READ_4(sc, BGE_HCC_TX_COAL_TICKS);
4641 
4642 		if (bootverbose) {
4643 			if_printf(ifp, "tx_coal_ticks -> %u\n",
4644 				  sc->bge_tx_coal_ticks);
4645 		}
4646 	}
4647 
4648 	if (sc->bge_coal_chg & BGE_RX_COAL_BDS_CHG) {
4649 		CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS,
4650 			    sc->bge_rx_coal_bds);
4651 		DELAY(10);
4652 		CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS);
4653 
4654 		if (bootverbose) {
4655 			if_printf(ifp, "rx_coal_bds -> %u\n",
4656 				  sc->bge_rx_coal_bds);
4657 		}
4658 	}
4659 
4660 	if (sc->bge_coal_chg & BGE_TX_COAL_BDS_CHG) {
4661 		CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS,
4662 			    sc->bge_tx_coal_bds);
4663 		DELAY(10);
4664 		CSR_READ_4(sc, BGE_HCC_TX_MAX_COAL_BDS);
4665 
4666 		if (bootverbose) {
4667 			if_printf(ifp, "tx_max_coal_bds -> %u\n",
4668 				  sc->bge_tx_coal_bds);
4669 		}
4670 	}
4671 
4672 	if (sc->bge_coal_chg & BGE_RX_COAL_TICKS_INT_CHG) {
4673 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS_INT,
4674 		    sc->bge_rx_coal_ticks_int);
4675 		DELAY(10);
4676 		CSR_READ_4(sc, BGE_HCC_RX_COAL_TICKS_INT);
4677 
4678 		if (bootverbose) {
4679 			if_printf(ifp, "rx_coal_ticks_int -> %u\n",
4680 			    sc->bge_rx_coal_ticks_int);
4681 		}
4682 	}
4683 
4684 	if (sc->bge_coal_chg & BGE_TX_COAL_TICKS_INT_CHG) {
4685 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS_INT,
4686 		    sc->bge_tx_coal_ticks_int);
4687 		DELAY(10);
4688 		CSR_READ_4(sc, BGE_HCC_TX_COAL_TICKS_INT);
4689 
4690 		if (bootverbose) {
4691 			if_printf(ifp, "tx_coal_ticks_int -> %u\n",
4692 			    sc->bge_tx_coal_ticks_int);
4693 		}
4694 	}
4695 
4696 	if (sc->bge_coal_chg & BGE_RX_COAL_BDS_INT_CHG) {
4697 		CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT,
4698 		    sc->bge_rx_coal_bds_int);
4699 		DELAY(10);
4700 		CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT);
4701 
4702 		if (bootverbose) {
4703 			if_printf(ifp, "rx_coal_bds_int -> %u\n",
4704 			    sc->bge_rx_coal_bds_int);
4705 		}
4706 	}
4707 
4708 	if (sc->bge_coal_chg & BGE_TX_COAL_BDS_INT_CHG) {
4709 		CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT,
4710 		    sc->bge_tx_coal_bds_int);
4711 		DELAY(10);
4712 		CSR_READ_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT);
4713 
4714 		if (bootverbose) {
4715 			if_printf(ifp, "tx_coal_bds_int -> %u\n",
4716 			    sc->bge_tx_coal_bds_int);
4717 		}
4718 	}
4719 
4720 	sc->bge_coal_chg = 0;
4721 }
4722 
4723 static void
4724 bge_enable_intr(struct bge_softc *sc)
4725 {
4726 	struct ifnet *ifp = &sc->arpcom.ac_if;
4727 
4728 	lwkt_serialize_handler_enable(ifp->if_serializer);
4729 
4730 	/*
4731 	 * Enable interrupt.
4732 	 */
4733 	bge_writembx(sc, BGE_MBX_IRQ0_LO, sc->bge_status_tag << 24);
4734 	if (sc->bge_flags & BGE_FLAG_ONESHOT_MSI) {
4735 		/* XXX Linux driver */
4736 		bge_writembx(sc, BGE_MBX_IRQ0_LO, sc->bge_status_tag << 24);
4737 	}
4738 
4739 	/*
4740 	 * Unmask the interrupt when we stop polling.
4741 	 */
4742 	PCI_CLRBIT(sc->bge_dev, BGE_PCI_MISC_CTL,
4743 	    BGE_PCIMISCCTL_MASK_PCI_INTR, 4);
4744 
4745 	/*
4746 	 * Trigger another interrupt, since above writing
4747 	 * to interrupt mailbox0 may acknowledge pending
4748 	 * interrupt.
4749 	 */
4750 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
4751 }
4752 
4753 static void
4754 bge_disable_intr(struct bge_softc *sc)
4755 {
4756 	struct ifnet *ifp = &sc->arpcom.ac_if;
4757 
4758 	/*
4759 	 * Mask the interrupt when we start polling.
4760 	 */
4761 	PCI_SETBIT(sc->bge_dev, BGE_PCI_MISC_CTL,
4762 	    BGE_PCIMISCCTL_MASK_PCI_INTR, 4);
4763 
4764 	/*
4765 	 * Acknowledge possible asserted interrupt.
4766 	 */
4767 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
4768 
4769 	sc->bge_npoll.ifpc_stcount = 0;
4770 
4771 	lwkt_serialize_handler_disable(ifp->if_serializer);
4772 }
4773 
4774 static int
4775 bge_get_eaddr_mem(struct bge_softc *sc, uint8_t ether_addr[])
4776 {
4777 	uint32_t mac_addr;
4778 	int ret = 1;
4779 
4780 	mac_addr = bge_readmem_ind(sc, 0x0c14);
4781 	if ((mac_addr >> 16) == 0x484b) {
4782 		ether_addr[0] = (uint8_t)(mac_addr >> 8);
4783 		ether_addr[1] = (uint8_t)mac_addr;
4784 		mac_addr = bge_readmem_ind(sc, 0x0c18);
4785 		ether_addr[2] = (uint8_t)(mac_addr >> 24);
4786 		ether_addr[3] = (uint8_t)(mac_addr >> 16);
4787 		ether_addr[4] = (uint8_t)(mac_addr >> 8);
4788 		ether_addr[5] = (uint8_t)mac_addr;
4789 		ret = 0;
4790 	}
4791 	return ret;
4792 }
4793 
4794 static int
4795 bge_get_eaddr_nvram(struct bge_softc *sc, uint8_t ether_addr[])
4796 {
4797 	int mac_offset = BGE_EE_MAC_OFFSET;
4798 
4799 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
4800 		mac_offset = BGE_EE_MAC_OFFSET_5906;
4801 
4802 	return bge_read_nvram(sc, ether_addr, mac_offset + 2, ETHER_ADDR_LEN);
4803 }
4804 
4805 static int
4806 bge_get_eaddr_eeprom(struct bge_softc *sc, uint8_t ether_addr[])
4807 {
4808 	if (sc->bge_flags & BGE_FLAG_NO_EEPROM)
4809 		return 1;
4810 
4811 	return bge_read_eeprom(sc, ether_addr, BGE_EE_MAC_OFFSET + 2,
4812 			       ETHER_ADDR_LEN);
4813 }
4814 
4815 static int
4816 bge_get_eaddr(struct bge_softc *sc, uint8_t eaddr[])
4817 {
4818 	static const bge_eaddr_fcn_t bge_eaddr_funcs[] = {
4819 		/* NOTE: Order is critical */
4820 		bge_get_eaddr_mem,
4821 		bge_get_eaddr_nvram,
4822 		bge_get_eaddr_eeprom,
4823 		NULL
4824 	};
4825 	const bge_eaddr_fcn_t *func;
4826 
4827 	for (func = bge_eaddr_funcs; *func != NULL; ++func) {
4828 		if ((*func)(sc, eaddr) == 0)
4829 			break;
4830 	}
4831 	return (*func == NULL ? ENXIO : 0);
4832 }
4833 
4834 /*
4835  * NOTE: 'm' is not freed upon failure
4836  */
4837 struct mbuf *
4838 bge_defrag_shortdma(struct mbuf *m)
4839 {
4840 	struct mbuf *n;
4841 	int found;
4842 
4843 	/*
4844 	 * If device receive two back-to-back send BDs with less than
4845 	 * or equal to 8 total bytes then the device may hang.  The two
4846 	 * back-to-back send BDs must in the same frame for this failure
4847 	 * to occur.  Scan mbuf chains and see whether two back-to-back
4848 	 * send BDs are there.  If this is the case, allocate new mbuf
4849 	 * and copy the frame to workaround the silicon bug.
4850 	 */
4851 	for (n = m, found = 0; n != NULL; n = n->m_next) {
4852 		if (n->m_len < 8) {
4853 			found++;
4854 			if (found > 1)
4855 				break;
4856 			continue;
4857 		}
4858 		found = 0;
4859 	}
4860 
4861 	if (found > 1)
4862 		n = m_defrag(m, MB_DONTWAIT);
4863 	else
4864 		n = m;
4865 	return n;
4866 }
4867 
4868 static void
4869 bge_stop_block(struct bge_softc *sc, bus_size_t reg, uint32_t bit)
4870 {
4871 	int i;
4872 
4873 	BGE_CLRBIT(sc, reg, bit);
4874 	for (i = 0; i < BGE_TIMEOUT; i++) {
4875 		if ((CSR_READ_4(sc, reg) & bit) == 0)
4876 			return;
4877 		DELAY(100);
4878 	}
4879 }
4880 
4881 static void
4882 bge_link_poll(struct bge_softc *sc)
4883 {
4884 	uint32_t status;
4885 
4886 	status = CSR_READ_4(sc, BGE_MAC_STS);
4887 	if ((status & sc->bge_link_chg) || sc->bge_link_evt) {
4888 		sc->bge_link_evt = 0;
4889 		sc->bge_link_upd(sc, status);
4890 	}
4891 }
4892 
4893 static void
4894 bge_enable_msi(struct bge_softc *sc)
4895 {
4896 	uint32_t msi_mode;
4897 
4898 	msi_mode = CSR_READ_4(sc, BGE_MSI_MODE);
4899 	msi_mode |= BGE_MSIMODE_ENABLE;
4900 	if (sc->bge_flags & BGE_FLAG_ONESHOT_MSI) {
4901 		/*
4902 		 * According to all of the datasheets that are publicly
4903 		 * available, bit 5 of the MSI_MODE is defined to be
4904 		 * "MSI FIFO Underrun Attn" for BCM5755+ and BCM5906, on
4905 		 * which "oneshot MSI" is enabled.  However, it is always
4906 		 * safe to clear it here.
4907 		 */
4908 		msi_mode &= ~BGE_MSIMODE_ONESHOT_DISABLE;
4909 	}
4910 	CSR_WRITE_4(sc, BGE_MSI_MODE, msi_mode);
4911 }
4912 
4913 static int
4914 bge_setup_tso(struct bge_softc *sc, struct mbuf **mp,
4915     uint16_t *mss0, uint16_t *flags0)
4916 {
4917 	struct mbuf *m;
4918 	struct ip *ip;
4919 	struct tcphdr *th;
4920 	int thoff, iphlen, hoff, hlen;
4921 	uint16_t flags, mss;
4922 
4923 	m = *mp;
4924 	KASSERT(M_WRITABLE(m), ("TSO mbuf not writable"));
4925 
4926 	hoff = m->m_pkthdr.csum_lhlen;
4927 	iphlen = m->m_pkthdr.csum_iphlen;
4928 	thoff = m->m_pkthdr.csum_thlen;
4929 
4930 	KASSERT(hoff > 0, ("invalid ether header len"));
4931 	KASSERT(iphlen > 0, ("invalid ip header len"));
4932 	KASSERT(thoff > 0, ("invalid tcp header len"));
4933 
4934 	if (__predict_false(m->m_len < hoff + iphlen + thoff)) {
4935 		m = m_pullup(m, hoff + iphlen + thoff);
4936 		if (m == NULL) {
4937 			*mp = NULL;
4938 			return ENOBUFS;
4939 		}
4940 		*mp = m;
4941 	}
4942 	ip = mtodoff(m, struct ip *, hoff);
4943 	th = mtodoff(m, struct tcphdr *, hoff + iphlen);
4944 
4945 	mss = m->m_pkthdr.tso_segsz;
4946 	flags = BGE_TXBDFLAG_CPU_PRE_DMA | BGE_TXBDFLAG_CPU_POST_DMA;
4947 
4948 	ip->ip_len = htons(mss + iphlen + thoff);
4949 	th->th_sum = 0;
4950 
4951 	hlen = (iphlen + thoff) >> 2;
4952 	mss |= (hlen << 11);
4953 
4954 	*mss0 = mss;
4955 	*flags0 = flags;
4956 
4957 	return 0;
4958 }
4959