xref: /dflybsd-src/sys/dev/netif/bge/if_bge.c (revision c666f28aa7c82e205ee3709528b79a41e8cc5308)
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_polling.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_types.h>
100 #include <net/ifq_var.h>
101 #include <net/vlan/if_vlan_var.h>
102 #include <net/vlan/if_vlan_ether.h>
103 
104 #include <dev/netif/mii_layer/mii.h>
105 #include <dev/netif/mii_layer/miivar.h>
106 #include <dev/netif/mii_layer/brgphyreg.h>
107 
108 #include <bus/pci/pcidevs.h>
109 #include <bus/pci/pcireg.h>
110 #include <bus/pci/pcivar.h>
111 
112 #include <dev/netif/bge/if_bgereg.h>
113 #include <dev/netif/bge/if_bgevar.h>
114 
115 /* "device miibus" required.  See GENERIC if you get errors here. */
116 #include "miibus_if.h"
117 
118 #define BGE_CSUM_FEATURES	(CSUM_IP | CSUM_TCP)
119 
120 static const struct bge_type {
121 	uint16_t		bge_vid;
122 	uint16_t		bge_did;
123 	char			*bge_name;
124 } bge_devs[] = {
125 	{ PCI_VENDOR_3COM, PCI_PRODUCT_3COM_3C996,
126 		"3COM 3C996 Gigabit Ethernet" },
127 
128 	{ PCI_VENDOR_ALTEON, PCI_PRODUCT_ALTEON_BCM5700,
129 		"Alteon BCM5700 Gigabit Ethernet" },
130 	{ PCI_VENDOR_ALTEON, PCI_PRODUCT_ALTEON_BCM5701,
131 		"Alteon BCM5701 Gigabit Ethernet" },
132 
133 	{ PCI_VENDOR_ALTIMA, PCI_PRODUCT_ALTIMA_AC1000,
134 		"Altima AC1000 Gigabit Ethernet" },
135 	{ PCI_VENDOR_ALTIMA, PCI_PRODUCT_ALTIMA_AC1001,
136 		"Altima AC1002 Gigabit Ethernet" },
137 	{ PCI_VENDOR_ALTIMA, PCI_PRODUCT_ALTIMA_AC9100,
138 		"Altima AC9100 Gigabit Ethernet" },
139 
140 	{ PCI_VENDOR_APPLE, PCI_PRODUCT_APPLE_BCM5701,
141 		"Apple BCM5701 Gigabit Ethernet" },
142 
143 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5700,
144 		"Broadcom BCM5700 Gigabit Ethernet" },
145 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5701,
146 		"Broadcom BCM5701 Gigabit Ethernet" },
147 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5702,
148 		"Broadcom BCM5702 Gigabit Ethernet" },
149 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5702X,
150 		"Broadcom BCM5702X Gigabit Ethernet" },
151 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5702_ALT,
152 		"Broadcom BCM5702 Gigabit Ethernet" },
153 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5703,
154 		"Broadcom BCM5703 Gigabit Ethernet" },
155 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5703X,
156 		"Broadcom BCM5703X Gigabit Ethernet" },
157 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5703A3,
158 		"Broadcom BCM5703 Gigabit Ethernet" },
159 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5704C,
160 		"Broadcom BCM5704C Dual Gigabit Ethernet" },
161 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5704S,
162 		"Broadcom BCM5704S Dual Gigabit Ethernet" },
163 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5704S_ALT,
164 		"Broadcom BCM5704S Dual Gigabit Ethernet" },
165 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5705,
166 		"Broadcom BCM5705 Gigabit Ethernet" },
167 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5705F,
168 		"Broadcom BCM5705F Gigabit Ethernet" },
169 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5705K,
170 		"Broadcom BCM5705K Gigabit Ethernet" },
171 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5705M,
172 		"Broadcom BCM5705M Gigabit Ethernet" },
173 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5705M_ALT,
174 		"Broadcom BCM5705M Gigabit Ethernet" },
175 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5714,
176 		"Broadcom BCM5714C Gigabit Ethernet" },
177 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5714S,
178 		"Broadcom BCM5714S Gigabit Ethernet" },
179 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5715,
180 		"Broadcom BCM5715 Gigabit Ethernet" },
181 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5715S,
182 		"Broadcom BCM5715S Gigabit Ethernet" },
183 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5720,
184 		"Broadcom BCM5720 Gigabit Ethernet" },
185 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5721,
186 		"Broadcom BCM5721 Gigabit Ethernet" },
187 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5722,
188 		"Broadcom BCM5722 Gigabit Ethernet" },
189 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5723,
190 		"Broadcom BCM5723 Gigabit Ethernet" },
191 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5750,
192 		"Broadcom BCM5750 Gigabit Ethernet" },
193 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5750M,
194 		"Broadcom BCM5750M Gigabit Ethernet" },
195 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5751,
196 		"Broadcom BCM5751 Gigabit Ethernet" },
197 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5751F,
198 		"Broadcom BCM5751F Gigabit Ethernet" },
199 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5751M,
200 		"Broadcom BCM5751M Gigabit Ethernet" },
201 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5752,
202 		"Broadcom BCM5752 Gigabit Ethernet" },
203 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5752M,
204 		"Broadcom BCM5752M Gigabit Ethernet" },
205 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5753,
206 		"Broadcom BCM5753 Gigabit Ethernet" },
207 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5753F,
208 		"Broadcom BCM5753F Gigabit Ethernet" },
209 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5753M,
210 		"Broadcom BCM5753M Gigabit Ethernet" },
211 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5754,
212 		"Broadcom BCM5754 Gigabit Ethernet" },
213 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5754M,
214 		"Broadcom BCM5754M Gigabit Ethernet" },
215 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5755,
216 		"Broadcom BCM5755 Gigabit Ethernet" },
217 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5755M,
218 		"Broadcom BCM5755M Gigabit Ethernet" },
219 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5756,
220 		"Broadcom BCM5756 Gigabit Ethernet" },
221 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5761,
222 		"Broadcom BCM5761 Gigabit Ethernet" },
223 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5761E,
224 		"Broadcom BCM5761E Gigabit Ethernet" },
225 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5761S,
226 		"Broadcom BCM5761S Gigabit Ethernet" },
227 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5761SE,
228 		"Broadcom BCM5761SE Gigabit Ethernet" },
229 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5764,
230 		"Broadcom BCM5764 Gigabit Ethernet" },
231 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5780,
232 		"Broadcom BCM5780 Gigabit Ethernet" },
233 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5780S,
234 		"Broadcom BCM5780S Gigabit Ethernet" },
235 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5781,
236 		"Broadcom BCM5781 Gigabit Ethernet" },
237 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5782,
238 		"Broadcom BCM5782 Gigabit Ethernet" },
239 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5784,
240 		"Broadcom BCM5784 Gigabit Ethernet" },
241 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5785F,
242 		"Broadcom BCM5785F Gigabit Ethernet" },
243 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5785G,
244 		"Broadcom BCM5785G Gigabit Ethernet" },
245 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5786,
246 		"Broadcom BCM5786 Gigabit Ethernet" },
247 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5787,
248 		"Broadcom BCM5787 Gigabit Ethernet" },
249 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5787F,
250 		"Broadcom BCM5787F Gigabit Ethernet" },
251 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5787M,
252 		"Broadcom BCM5787M Gigabit Ethernet" },
253 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5788,
254 		"Broadcom BCM5788 Gigabit Ethernet" },
255 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5789,
256 		"Broadcom BCM5789 Gigabit Ethernet" },
257 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5901,
258 		"Broadcom BCM5901 Fast Ethernet" },
259 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5901A2,
260 		"Broadcom BCM5901A2 Fast Ethernet" },
261 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5903M,
262 		"Broadcom BCM5903M Fast Ethernet" },
263 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5906,
264 		"Broadcom BCM5906 Fast Ethernet"},
265 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5906M,
266 		"Broadcom BCM5906M Fast Ethernet"},
267 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57760,
268 		"Broadcom BCM57760 Gigabit Ethernet"},
269 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57780,
270 		"Broadcom BCM57780 Gigabit Ethernet"},
271 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57788,
272 		"Broadcom BCM57788 Gigabit Ethernet"},
273 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57790,
274 		"Broadcom BCM57790 Gigabit Ethernet"},
275 	{ PCI_VENDOR_SCHNEIDERKOCH, PCI_PRODUCT_SCHNEIDERKOCH_SK_9DX1,
276 		"SysKonnect Gigabit Ethernet" },
277 
278 	{ 0, 0, NULL }
279 };
280 
281 #define BGE_IS_JUMBO_CAPABLE(sc)	((sc)->bge_flags & BGE_FLAG_JUMBO)
282 #define BGE_IS_5700_FAMILY(sc)		((sc)->bge_flags & BGE_FLAG_5700_FAMILY)
283 #define BGE_IS_5705_PLUS(sc)		((sc)->bge_flags & BGE_FLAG_5705_PLUS)
284 #define BGE_IS_5714_FAMILY(sc)		((sc)->bge_flags & BGE_FLAG_5714_FAMILY)
285 #define BGE_IS_575X_PLUS(sc)		((sc)->bge_flags & BGE_FLAG_575X_PLUS)
286 #define BGE_IS_5755_PLUS(sc)		((sc)->bge_flags & BGE_FLAG_5755_PLUS)
287 #define BGE_IS_5788(sc)			((sc)->bge_flags & BGE_FLAG_5788)
288 
289 #define BGE_IS_CRIPPLED(sc)		\
290 	(BGE_IS_5788((sc)) || (sc)->bge_asicrev == BGE_ASICREV_BCM5700)
291 
292 typedef int	(*bge_eaddr_fcn_t)(struct bge_softc *, uint8_t[]);
293 
294 static int	bge_probe(device_t);
295 static int	bge_attach(device_t);
296 static int	bge_detach(device_t);
297 static void	bge_txeof(struct bge_softc *, uint16_t);
298 static void	bge_rxeof(struct bge_softc *, uint16_t);
299 
300 static void	bge_tick(void *);
301 static void	bge_stats_update(struct bge_softc *);
302 static void	bge_stats_update_regs(struct bge_softc *);
303 static struct mbuf *
304 		bge_defrag_shortdma(struct mbuf *);
305 static int	bge_encap(struct bge_softc *, struct mbuf **, uint32_t *);
306 static int	bge_setup_tso(struct bge_softc *, struct mbuf **,
307 		    uint16_t *, uint16_t *);
308 
309 #ifdef DEVICE_POLLING
310 static void	bge_poll(struct ifnet *ifp, enum poll_cmd cmd, int count);
311 #endif
312 static void	bge_intr_crippled(void *);
313 static void	bge_intr_legacy(void *);
314 static void	bge_msi(void *);
315 static void	bge_msi_oneshot(void *);
316 static void	bge_intr(struct bge_softc *);
317 static void	bge_enable_intr(struct bge_softc *);
318 static void	bge_disable_intr(struct bge_softc *);
319 static void	bge_start(struct ifnet *);
320 static int	bge_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
321 static void	bge_init(void *);
322 static void	bge_stop(struct bge_softc *);
323 static void	bge_watchdog(struct ifnet *);
324 static void	bge_shutdown(device_t);
325 static int	bge_suspend(device_t);
326 static int	bge_resume(device_t);
327 static int	bge_ifmedia_upd(struct ifnet *);
328 static void	bge_ifmedia_sts(struct ifnet *, struct ifmediareq *);
329 
330 static uint8_t	bge_nvram_getbyte(struct bge_softc *, int, uint8_t *);
331 static int	bge_read_nvram(struct bge_softc *, caddr_t, int, int);
332 
333 static uint8_t	bge_eeprom_getbyte(struct bge_softc *, uint32_t, uint8_t *);
334 static int	bge_read_eeprom(struct bge_softc *, caddr_t, uint32_t, size_t);
335 
336 static void	bge_setmulti(struct bge_softc *);
337 static void	bge_setpromisc(struct bge_softc *);
338 static void	bge_enable_msi(struct bge_softc *sc);
339 
340 static int	bge_alloc_jumbo_mem(struct bge_softc *);
341 static void	bge_free_jumbo_mem(struct bge_softc *);
342 static struct bge_jslot
343 		*bge_jalloc(struct bge_softc *);
344 static void	bge_jfree(void *);
345 static void	bge_jref(void *);
346 static int	bge_newbuf_std(struct bge_softc *, int, int);
347 static int	bge_newbuf_jumbo(struct bge_softc *, int, int);
348 static void	bge_setup_rxdesc_std(struct bge_softc *, int);
349 static void	bge_setup_rxdesc_jumbo(struct bge_softc *, int);
350 static int	bge_init_rx_ring_std(struct bge_softc *);
351 static void	bge_free_rx_ring_std(struct bge_softc *);
352 static int	bge_init_rx_ring_jumbo(struct bge_softc *);
353 static void	bge_free_rx_ring_jumbo(struct bge_softc *);
354 static void	bge_free_tx_ring(struct bge_softc *);
355 static int	bge_init_tx_ring(struct bge_softc *);
356 
357 static int	bge_chipinit(struct bge_softc *);
358 static int	bge_blockinit(struct bge_softc *);
359 static void	bge_stop_block(struct bge_softc *, bus_size_t, uint32_t);
360 
361 static uint32_t	bge_readmem_ind(struct bge_softc *, uint32_t);
362 static void	bge_writemem_ind(struct bge_softc *, uint32_t, uint32_t);
363 #ifdef notdef
364 static uint32_t	bge_readreg_ind(struct bge_softc *, uint32_t);
365 #endif
366 static void	bge_writereg_ind(struct bge_softc *, uint32_t, uint32_t);
367 static void	bge_writemem_direct(struct bge_softc *, uint32_t, uint32_t);
368 static void	bge_writembx(struct bge_softc *, int, int);
369 
370 static int	bge_miibus_readreg(device_t, int, int);
371 static int	bge_miibus_writereg(device_t, int, int, int);
372 static void	bge_miibus_statchg(device_t);
373 static void	bge_bcm5700_link_upd(struct bge_softc *, uint32_t);
374 static void	bge_tbi_link_upd(struct bge_softc *, uint32_t);
375 static void	bge_copper_link_upd(struct bge_softc *, uint32_t);
376 static void	bge_autopoll_link_upd(struct bge_softc *, uint32_t);
377 static void	bge_link_poll(struct bge_softc *);
378 
379 static void	bge_reset(struct bge_softc *);
380 
381 static int	bge_dma_alloc(struct bge_softc *);
382 static void	bge_dma_free(struct bge_softc *);
383 static int	bge_dma_block_alloc(struct bge_softc *, bus_size_t,
384 				    bus_dma_tag_t *, bus_dmamap_t *,
385 				    void **, bus_addr_t *);
386 static void	bge_dma_block_free(bus_dma_tag_t, bus_dmamap_t, void *);
387 
388 static int	bge_get_eaddr_mem(struct bge_softc *, uint8_t[]);
389 static int	bge_get_eaddr_nvram(struct bge_softc *, uint8_t[]);
390 static int	bge_get_eaddr_eeprom(struct bge_softc *, uint8_t[]);
391 static int	bge_get_eaddr(struct bge_softc *, uint8_t[]);
392 
393 static void	bge_coal_change(struct bge_softc *);
394 static int	bge_sysctl_rx_coal_ticks(SYSCTL_HANDLER_ARGS);
395 static int	bge_sysctl_tx_coal_ticks(SYSCTL_HANDLER_ARGS);
396 static int	bge_sysctl_rx_coal_bds(SYSCTL_HANDLER_ARGS);
397 static int	bge_sysctl_tx_coal_bds(SYSCTL_HANDLER_ARGS);
398 static int	bge_sysctl_rx_coal_ticks_int(SYSCTL_HANDLER_ARGS);
399 static int	bge_sysctl_tx_coal_ticks_int(SYSCTL_HANDLER_ARGS);
400 static int	bge_sysctl_rx_coal_bds_int(SYSCTL_HANDLER_ARGS);
401 static int	bge_sysctl_tx_coal_bds_int(SYSCTL_HANDLER_ARGS);
402 static int	bge_sysctl_coal_chg(SYSCTL_HANDLER_ARGS, uint32_t *,
403 		    int, int, uint32_t);
404 
405 /*
406  * Set following tunable to 1 for some IBM blade servers with the DNLK
407  * switch module. Auto negotiation is broken for those configurations.
408  */
409 static int	bge_fake_autoneg = 0;
410 TUNABLE_INT("hw.bge.fake_autoneg", &bge_fake_autoneg);
411 
412 static int	bge_msi_enable = 1;
413 TUNABLE_INT("hw.bge.msi.enable", &bge_msi_enable);
414 
415 #if !defined(KTR_IF_BGE)
416 #define KTR_IF_BGE	KTR_ALL
417 #endif
418 KTR_INFO_MASTER(if_bge);
419 KTR_INFO(KTR_IF_BGE, if_bge, intr, 0, "intr");
420 KTR_INFO(KTR_IF_BGE, if_bge, rx_pkt, 1, "rx_pkt");
421 KTR_INFO(KTR_IF_BGE, if_bge, tx_pkt, 2, "tx_pkt");
422 #define logif(name)	KTR_LOG(if_bge_ ## name)
423 
424 static device_method_t bge_methods[] = {
425 	/* Device interface */
426 	DEVMETHOD(device_probe,		bge_probe),
427 	DEVMETHOD(device_attach,	bge_attach),
428 	DEVMETHOD(device_detach,	bge_detach),
429 	DEVMETHOD(device_shutdown,	bge_shutdown),
430 	DEVMETHOD(device_suspend,	bge_suspend),
431 	DEVMETHOD(device_resume,	bge_resume),
432 
433 	/* bus interface */
434 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
435 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
436 
437 	/* MII interface */
438 	DEVMETHOD(miibus_readreg,	bge_miibus_readreg),
439 	DEVMETHOD(miibus_writereg,	bge_miibus_writereg),
440 	DEVMETHOD(miibus_statchg,	bge_miibus_statchg),
441 
442 	{ 0, 0 }
443 };
444 
445 static DEFINE_CLASS_0(bge, bge_driver, bge_methods, sizeof(struct bge_softc));
446 static devclass_t bge_devclass;
447 
448 DECLARE_DUMMY_MODULE(if_bge);
449 DRIVER_MODULE(if_bge, pci, bge_driver, bge_devclass, NULL, NULL);
450 DRIVER_MODULE(miibus, bge, miibus_driver, miibus_devclass, NULL, NULL);
451 
452 static uint32_t
453 bge_readmem_ind(struct bge_softc *sc, uint32_t off)
454 {
455 	device_t dev = sc->bge_dev;
456 	uint32_t val;
457 
458 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
459 	    off >= BGE_STATS_BLOCK && off < BGE_SEND_RING_1_TO_4)
460 		return 0;
461 
462 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4);
463 	val = pci_read_config(dev, BGE_PCI_MEMWIN_DATA, 4);
464 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4);
465 	return (val);
466 }
467 
468 static void
469 bge_writemem_ind(struct bge_softc *sc, uint32_t off, uint32_t val)
470 {
471 	device_t dev = sc->bge_dev;
472 
473 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
474 	    off >= BGE_STATS_BLOCK && off < BGE_SEND_RING_1_TO_4)
475 		return;
476 
477 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4);
478 	pci_write_config(dev, BGE_PCI_MEMWIN_DATA, val, 4);
479 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4);
480 }
481 
482 #ifdef notdef
483 static uint32_t
484 bge_readreg_ind(struct bge_softc *sc, uin32_t off)
485 {
486 	device_t dev = sc->bge_dev;
487 
488 	pci_write_config(dev, BGE_PCI_REG_BASEADDR, off, 4);
489 	return(pci_read_config(dev, BGE_PCI_REG_DATA, 4));
490 }
491 #endif
492 
493 static void
494 bge_writereg_ind(struct bge_softc *sc, uint32_t off, uint32_t val)
495 {
496 	device_t dev = sc->bge_dev;
497 
498 	pci_write_config(dev, BGE_PCI_REG_BASEADDR, off, 4);
499 	pci_write_config(dev, BGE_PCI_REG_DATA, val, 4);
500 }
501 
502 static void
503 bge_writemem_direct(struct bge_softc *sc, uint32_t off, uint32_t val)
504 {
505 	CSR_WRITE_4(sc, off, val);
506 }
507 
508 static void
509 bge_writembx(struct bge_softc *sc, int off, int val)
510 {
511 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
512 		off += BGE_LPMBX_IRQ0_HI - BGE_MBX_IRQ0_HI;
513 
514 	CSR_WRITE_4(sc, off, val);
515 	if (sc->bge_mbox_reorder)
516 		CSR_READ_4(sc, off);
517 }
518 
519 static uint8_t
520 bge_nvram_getbyte(struct bge_softc *sc, int addr, uint8_t *dest)
521 {
522 	uint32_t access, byte = 0;
523 	int i;
524 
525 	/* Lock. */
526 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_SET1);
527 	for (i = 0; i < 8000; i++) {
528 		if (CSR_READ_4(sc, BGE_NVRAM_SWARB) & BGE_NVRAMSWARB_GNT1)
529 			break;
530 		DELAY(20);
531 	}
532 	if (i == 8000)
533 		return (1);
534 
535 	/* Enable access. */
536 	access = CSR_READ_4(sc, BGE_NVRAM_ACCESS);
537 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access | BGE_NVRAMACC_ENABLE);
538 
539 	CSR_WRITE_4(sc, BGE_NVRAM_ADDR, addr & 0xfffffffc);
540 	CSR_WRITE_4(sc, BGE_NVRAM_CMD, BGE_NVRAM_READCMD);
541 	for (i = 0; i < BGE_TIMEOUT * 10; i++) {
542 		DELAY(10);
543 		if (CSR_READ_4(sc, BGE_NVRAM_CMD) & BGE_NVRAMCMD_DONE) {
544 			DELAY(10);
545 			break;
546 		}
547 	}
548 
549 	if (i == BGE_TIMEOUT * 10) {
550 		if_printf(&sc->arpcom.ac_if, "nvram read timed out\n");
551 		return (1);
552 	}
553 
554 	/* Get result. */
555 	byte = CSR_READ_4(sc, BGE_NVRAM_RDDATA);
556 
557 	*dest = (bswap32(byte) >> ((addr % 4) * 8)) & 0xFF;
558 
559 	/* Disable access. */
560 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access);
561 
562 	/* Unlock. */
563 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_CLR1);
564 	CSR_READ_4(sc, BGE_NVRAM_SWARB);
565 
566 	return (0);
567 }
568 
569 /*
570  * Read a sequence of bytes from NVRAM.
571  */
572 static int
573 bge_read_nvram(struct bge_softc *sc, caddr_t dest, int off, int cnt)
574 {
575 	int err = 0, i;
576 	uint8_t byte = 0;
577 
578 	if (sc->bge_asicrev != BGE_ASICREV_BCM5906)
579 		return (1);
580 
581 	for (i = 0; i < cnt; i++) {
582 		err = bge_nvram_getbyte(sc, off + i, &byte);
583 		if (err)
584 			break;
585 		*(dest + i) = byte;
586 	}
587 
588 	return (err ? 1 : 0);
589 }
590 
591 /*
592  * Read a byte of data stored in the EEPROM at address 'addr.' The
593  * BCM570x supports both the traditional bitbang interface and an
594  * auto access interface for reading the EEPROM. We use the auto
595  * access method.
596  */
597 static uint8_t
598 bge_eeprom_getbyte(struct bge_softc *sc, uint32_t addr, uint8_t *dest)
599 {
600 	int i;
601 	uint32_t byte = 0;
602 
603 	/*
604 	 * Enable use of auto EEPROM access so we can avoid
605 	 * having to use the bitbang method.
606 	 */
607 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_AUTO_EEPROM);
608 
609 	/* Reset the EEPROM, load the clock period. */
610 	CSR_WRITE_4(sc, BGE_EE_ADDR,
611 	    BGE_EEADDR_RESET|BGE_EEHALFCLK(BGE_HALFCLK_384SCL));
612 	DELAY(20);
613 
614 	/* Issue the read EEPROM command. */
615 	CSR_WRITE_4(sc, BGE_EE_ADDR, BGE_EE_READCMD | addr);
616 
617 	/* Wait for completion */
618 	for(i = 0; i < BGE_TIMEOUT * 10; i++) {
619 		DELAY(10);
620 		if (CSR_READ_4(sc, BGE_EE_ADDR) & BGE_EEADDR_DONE)
621 			break;
622 	}
623 
624 	if (i == BGE_TIMEOUT) {
625 		if_printf(&sc->arpcom.ac_if, "eeprom read timed out\n");
626 		return(1);
627 	}
628 
629 	/* Get result. */
630 	byte = CSR_READ_4(sc, BGE_EE_DATA);
631 
632         *dest = (byte >> ((addr % 4) * 8)) & 0xFF;
633 
634 	return(0);
635 }
636 
637 /*
638  * Read a sequence of bytes from the EEPROM.
639  */
640 static int
641 bge_read_eeprom(struct bge_softc *sc, caddr_t dest, uint32_t off, size_t len)
642 {
643 	size_t i;
644 	int err;
645 	uint8_t byte;
646 
647 	for (byte = 0, err = 0, i = 0; i < len; i++) {
648 		err = bge_eeprom_getbyte(sc, off + i, &byte);
649 		if (err)
650 			break;
651 		*(dest + i) = byte;
652 	}
653 
654 	return(err ? 1 : 0);
655 }
656 
657 static int
658 bge_miibus_readreg(device_t dev, int phy, int reg)
659 {
660 	struct bge_softc *sc = device_get_softc(dev);
661 	uint32_t val;
662 	int i;
663 
664 	KASSERT(phy == sc->bge_phyno,
665 	    ("invalid phyno %d, should be %d", phy, sc->bge_phyno));
666 
667 	/* Clear the autopoll bit if set, otherwise may trigger PCI errors. */
668 	if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
669 		CSR_WRITE_4(sc, BGE_MI_MODE,
670 		    sc->bge_mi_mode & ~BGE_MIMODE_AUTOPOLL);
671 		DELAY(80);
672 	}
673 
674 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_READ | BGE_MICOMM_BUSY |
675 	    BGE_MIPHY(phy) | BGE_MIREG(reg));
676 
677 	/* Poll for the PHY register access to complete. */
678 	for (i = 0; i < BGE_TIMEOUT; i++) {
679 		DELAY(10);
680 		val = CSR_READ_4(sc, BGE_MI_COMM);
681 		if ((val & BGE_MICOMM_BUSY) == 0) {
682 			DELAY(5);
683 			val = CSR_READ_4(sc, BGE_MI_COMM);
684 			break;
685 		}
686 	}
687 	if (i == BGE_TIMEOUT) {
688 		if_printf(&sc->arpcom.ac_if, "PHY read timed out "
689 		    "(phy %d, reg %d, val 0x%08x)\n", phy, reg, val);
690 		val = 0;
691 	}
692 
693 	/* Restore the autopoll bit if necessary. */
694 	if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
695 		CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode);
696 		DELAY(80);
697 	}
698 
699 	if (val & BGE_MICOMM_READFAIL)
700 		return 0;
701 
702 	return (val & 0xFFFF);
703 }
704 
705 static int
706 bge_miibus_writereg(device_t dev, int phy, int reg, int val)
707 {
708 	struct bge_softc *sc = device_get_softc(dev);
709 	int i;
710 
711 	KASSERT(phy == sc->bge_phyno,
712 	    ("invalid phyno %d, should be %d", phy, sc->bge_phyno));
713 
714 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
715 	    (reg == BRGPHY_MII_1000CTL || reg == BRGPHY_MII_AUXCTL))
716 	       return 0;
717 
718 	/* Clear the autopoll bit if set, otherwise may trigger PCI errors. */
719 	if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
720 		CSR_WRITE_4(sc, BGE_MI_MODE,
721 		    sc->bge_mi_mode & ~BGE_MIMODE_AUTOPOLL);
722 		DELAY(80);
723 	}
724 
725 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_WRITE | BGE_MICOMM_BUSY |
726 	    BGE_MIPHY(phy) | BGE_MIREG(reg) | val);
727 
728 	for (i = 0; i < BGE_TIMEOUT; i++) {
729 		DELAY(10);
730 		if (!(CSR_READ_4(sc, BGE_MI_COMM) & BGE_MICOMM_BUSY)) {
731 			DELAY(5);
732 			CSR_READ_4(sc, BGE_MI_COMM); /* dummy read */
733 			break;
734 		}
735 	}
736 	if (i == BGE_TIMEOUT) {
737 		if_printf(&sc->arpcom.ac_if, "PHY write timed out "
738 		    "(phy %d, reg %d, val %d)\n", phy, reg, val);
739 	}
740 
741 	/* Restore the autopoll bit if necessary. */
742 	if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
743 		CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode);
744 		DELAY(80);
745 	}
746 
747 	return 0;
748 }
749 
750 static void
751 bge_miibus_statchg(device_t dev)
752 {
753 	struct bge_softc *sc;
754 	struct mii_data *mii;
755 
756 	sc = device_get_softc(dev);
757 	mii = device_get_softc(sc->bge_miibus);
758 
759 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
760 	    (IFM_ACTIVE | IFM_AVALID)) {
761 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
762 		case IFM_10_T:
763 		case IFM_100_TX:
764 			sc->bge_link = 1;
765 			break;
766 		case IFM_1000_T:
767 		case IFM_1000_SX:
768 		case IFM_2500_SX:
769 			if (sc->bge_asicrev != BGE_ASICREV_BCM5906)
770 				sc->bge_link = 1;
771 			else
772 				sc->bge_link = 0;
773 			break;
774 		default:
775 			sc->bge_link = 0;
776 			break;
777 		}
778 	} else {
779 		sc->bge_link = 0;
780 	}
781 	if (sc->bge_link == 0)
782 		return;
783 
784 	BGE_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_PORTMODE);
785 	if (IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_T ||
786 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_SX) {
787 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_GMII);
788 	} else {
789 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_MII);
790 	}
791 
792 	if ((mii->mii_media_active & IFM_GMASK) == IFM_FDX) {
793 		BGE_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
794 	} else {
795 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
796 	}
797 }
798 
799 /*
800  * Memory management for jumbo frames.
801  */
802 static int
803 bge_alloc_jumbo_mem(struct bge_softc *sc)
804 {
805 	struct ifnet *ifp = &sc->arpcom.ac_if;
806 	struct bge_jslot *entry;
807 	uint8_t *ptr;
808 	bus_addr_t paddr;
809 	int i, error;
810 
811 	/*
812 	 * Create tag for jumbo mbufs.
813 	 * This is really a bit of a kludge. We allocate a special
814 	 * jumbo buffer pool which (thanks to the way our DMA
815 	 * memory allocation works) will consist of contiguous
816 	 * pages. This means that even though a jumbo buffer might
817 	 * be larger than a page size, we don't really need to
818 	 * map it into more than one DMA segment. However, the
819 	 * default mbuf tag will result in multi-segment mappings,
820 	 * so we have to create a special jumbo mbuf tag that
821 	 * lets us get away with mapping the jumbo buffers as
822 	 * a single segment. I think eventually the driver should
823 	 * be changed so that it uses ordinary mbufs and cluster
824 	 * buffers, i.e. jumbo frames can span multiple DMA
825 	 * descriptors. But that's a project for another day.
826 	 */
827 
828 	/*
829 	 * Create DMA stuffs for jumbo RX ring.
830 	 */
831 	error = bge_dma_block_alloc(sc, BGE_JUMBO_RX_RING_SZ,
832 				    &sc->bge_cdata.bge_rx_jumbo_ring_tag,
833 				    &sc->bge_cdata.bge_rx_jumbo_ring_map,
834 				    (void *)&sc->bge_ldata.bge_rx_jumbo_ring,
835 				    &sc->bge_ldata.bge_rx_jumbo_ring_paddr);
836 	if (error) {
837 		if_printf(ifp, "could not create jumbo RX ring\n");
838 		return error;
839 	}
840 
841 	/*
842 	 * Create DMA stuffs for jumbo buffer block.
843 	 */
844 	error = bge_dma_block_alloc(sc, BGE_JMEM,
845 				    &sc->bge_cdata.bge_jumbo_tag,
846 				    &sc->bge_cdata.bge_jumbo_map,
847 				    (void **)&sc->bge_ldata.bge_jumbo_buf,
848 				    &paddr);
849 	if (error) {
850 		if_printf(ifp, "could not create jumbo buffer\n");
851 		return error;
852 	}
853 
854 	SLIST_INIT(&sc->bge_jfree_listhead);
855 
856 	/*
857 	 * Now divide it up into 9K pieces and save the addresses
858 	 * in an array. Note that we play an evil trick here by using
859 	 * the first few bytes in the buffer to hold the the address
860 	 * of the softc structure for this interface. This is because
861 	 * bge_jfree() needs it, but it is called by the mbuf management
862 	 * code which will not pass it to us explicitly.
863 	 */
864 	for (i = 0, ptr = sc->bge_ldata.bge_jumbo_buf; i < BGE_JSLOTS; i++) {
865 		entry = &sc->bge_cdata.bge_jslots[i];
866 		entry->bge_sc = sc;
867 		entry->bge_buf = ptr;
868 		entry->bge_paddr = paddr;
869 		entry->bge_inuse = 0;
870 		entry->bge_slot = i;
871 		SLIST_INSERT_HEAD(&sc->bge_jfree_listhead, entry, jslot_link);
872 
873 		ptr += BGE_JLEN;
874 		paddr += BGE_JLEN;
875 	}
876 	return 0;
877 }
878 
879 static void
880 bge_free_jumbo_mem(struct bge_softc *sc)
881 {
882 	/* Destroy jumbo RX ring. */
883 	bge_dma_block_free(sc->bge_cdata.bge_rx_jumbo_ring_tag,
884 			   sc->bge_cdata.bge_rx_jumbo_ring_map,
885 			   sc->bge_ldata.bge_rx_jumbo_ring);
886 
887 	/* Destroy jumbo buffer block. */
888 	bge_dma_block_free(sc->bge_cdata.bge_jumbo_tag,
889 			   sc->bge_cdata.bge_jumbo_map,
890 			   sc->bge_ldata.bge_jumbo_buf);
891 }
892 
893 /*
894  * Allocate a jumbo buffer.
895  */
896 static struct bge_jslot *
897 bge_jalloc(struct bge_softc *sc)
898 {
899 	struct bge_jslot *entry;
900 
901 	lwkt_serialize_enter(&sc->bge_jslot_serializer);
902 	entry = SLIST_FIRST(&sc->bge_jfree_listhead);
903 	if (entry) {
904 		SLIST_REMOVE_HEAD(&sc->bge_jfree_listhead, jslot_link);
905 		entry->bge_inuse = 1;
906 	} else {
907 		if_printf(&sc->arpcom.ac_if, "no free jumbo buffers\n");
908 	}
909 	lwkt_serialize_exit(&sc->bge_jslot_serializer);
910 	return(entry);
911 }
912 
913 /*
914  * Adjust usage count on a jumbo buffer.
915  */
916 static void
917 bge_jref(void *arg)
918 {
919 	struct bge_jslot *entry = (struct bge_jslot *)arg;
920 	struct bge_softc *sc = entry->bge_sc;
921 
922 	if (sc == NULL)
923 		panic("bge_jref: can't find softc pointer!");
924 
925 	if (&sc->bge_cdata.bge_jslots[entry->bge_slot] != entry) {
926 		panic("bge_jref: asked to reference buffer "
927 		    "that we don't manage!");
928 	} else if (entry->bge_inuse == 0) {
929 		panic("bge_jref: buffer already free!");
930 	} else {
931 		atomic_add_int(&entry->bge_inuse, 1);
932 	}
933 }
934 
935 /*
936  * Release a jumbo buffer.
937  */
938 static void
939 bge_jfree(void *arg)
940 {
941 	struct bge_jslot *entry = (struct bge_jslot *)arg;
942 	struct bge_softc *sc = entry->bge_sc;
943 
944 	if (sc == NULL)
945 		panic("bge_jfree: can't find softc pointer!");
946 
947 	if (&sc->bge_cdata.bge_jslots[entry->bge_slot] != entry) {
948 		panic("bge_jfree: asked to free buffer that we don't manage!");
949 	} else if (entry->bge_inuse == 0) {
950 		panic("bge_jfree: buffer already free!");
951 	} else {
952 		/*
953 		 * Possible MP race to 0, use the serializer.  The atomic insn
954 		 * is still needed for races against bge_jref().
955 		 */
956 		lwkt_serialize_enter(&sc->bge_jslot_serializer);
957 		atomic_subtract_int(&entry->bge_inuse, 1);
958 		if (entry->bge_inuse == 0) {
959 			SLIST_INSERT_HEAD(&sc->bge_jfree_listhead,
960 					  entry, jslot_link);
961 		}
962 		lwkt_serialize_exit(&sc->bge_jslot_serializer);
963 	}
964 }
965 
966 
967 /*
968  * Intialize a standard receive ring descriptor.
969  */
970 static int
971 bge_newbuf_std(struct bge_softc *sc, int i, int init)
972 {
973 	struct mbuf *m_new = NULL;
974 	bus_dma_segment_t seg;
975 	bus_dmamap_t map;
976 	int error, nsegs;
977 
978 	m_new = m_getcl(init ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
979 	if (m_new == NULL)
980 		return ENOBUFS;
981 	m_new->m_len = m_new->m_pkthdr.len = MCLBYTES;
982 
983 	if ((sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) == 0)
984 		m_adj(m_new, ETHER_ALIGN);
985 
986 	error = bus_dmamap_load_mbuf_segment(sc->bge_cdata.bge_rx_mtag,
987 			sc->bge_cdata.bge_rx_tmpmap, m_new,
988 			&seg, 1, &nsegs, BUS_DMA_NOWAIT);
989 	if (error) {
990 		m_freem(m_new);
991 		return error;
992 	}
993 
994 	if (!init) {
995 		bus_dmamap_sync(sc->bge_cdata.bge_rx_mtag,
996 				sc->bge_cdata.bge_rx_std_dmamap[i],
997 				BUS_DMASYNC_POSTREAD);
998 		bus_dmamap_unload(sc->bge_cdata.bge_rx_mtag,
999 			sc->bge_cdata.bge_rx_std_dmamap[i]);
1000 	}
1001 
1002 	map = sc->bge_cdata.bge_rx_tmpmap;
1003 	sc->bge_cdata.bge_rx_tmpmap = sc->bge_cdata.bge_rx_std_dmamap[i];
1004 	sc->bge_cdata.bge_rx_std_dmamap[i] = map;
1005 
1006 	sc->bge_cdata.bge_rx_std_chain[i].bge_mbuf = m_new;
1007 	sc->bge_cdata.bge_rx_std_chain[i].bge_paddr = seg.ds_addr;
1008 
1009 	bge_setup_rxdesc_std(sc, i);
1010 	return 0;
1011 }
1012 
1013 static void
1014 bge_setup_rxdesc_std(struct bge_softc *sc, int i)
1015 {
1016 	struct bge_rxchain *rc;
1017 	struct bge_rx_bd *r;
1018 
1019 	rc = &sc->bge_cdata.bge_rx_std_chain[i];
1020 	r = &sc->bge_ldata.bge_rx_std_ring[i];
1021 
1022 	r->bge_addr.bge_addr_lo = BGE_ADDR_LO(rc->bge_paddr);
1023 	r->bge_addr.bge_addr_hi = BGE_ADDR_HI(rc->bge_paddr);
1024 	r->bge_len = rc->bge_mbuf->m_len;
1025 	r->bge_idx = i;
1026 	r->bge_flags = BGE_RXBDFLAG_END;
1027 }
1028 
1029 /*
1030  * Initialize a jumbo receive ring descriptor. This allocates
1031  * a jumbo buffer from the pool managed internally by the driver.
1032  */
1033 static int
1034 bge_newbuf_jumbo(struct bge_softc *sc, int i, int init)
1035 {
1036 	struct mbuf *m_new = NULL;
1037 	struct bge_jslot *buf;
1038 	bus_addr_t paddr;
1039 
1040 	/* Allocate the mbuf. */
1041 	MGETHDR(m_new, init ? MB_WAIT : MB_DONTWAIT, MT_DATA);
1042 	if (m_new == NULL)
1043 		return ENOBUFS;
1044 
1045 	/* Allocate the jumbo buffer */
1046 	buf = bge_jalloc(sc);
1047 	if (buf == NULL) {
1048 		m_freem(m_new);
1049 		return ENOBUFS;
1050 	}
1051 
1052 	/* Attach the buffer to the mbuf. */
1053 	m_new->m_ext.ext_arg = buf;
1054 	m_new->m_ext.ext_buf = buf->bge_buf;
1055 	m_new->m_ext.ext_free = bge_jfree;
1056 	m_new->m_ext.ext_ref = bge_jref;
1057 	m_new->m_ext.ext_size = BGE_JUMBO_FRAMELEN;
1058 
1059 	m_new->m_flags |= M_EXT;
1060 
1061 	m_new->m_data = m_new->m_ext.ext_buf;
1062 	m_new->m_len = m_new->m_pkthdr.len = m_new->m_ext.ext_size;
1063 
1064 	paddr = buf->bge_paddr;
1065 	if ((sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) == 0) {
1066 		m_adj(m_new, ETHER_ALIGN);
1067 		paddr += ETHER_ALIGN;
1068 	}
1069 
1070 	/* Save necessary information */
1071 	sc->bge_cdata.bge_rx_jumbo_chain[i].bge_mbuf = m_new;
1072 	sc->bge_cdata.bge_rx_jumbo_chain[i].bge_paddr = paddr;
1073 
1074 	/* Set up the descriptor. */
1075 	bge_setup_rxdesc_jumbo(sc, i);
1076 	return 0;
1077 }
1078 
1079 static void
1080 bge_setup_rxdesc_jumbo(struct bge_softc *sc, int i)
1081 {
1082 	struct bge_rx_bd *r;
1083 	struct bge_rxchain *rc;
1084 
1085 	r = &sc->bge_ldata.bge_rx_jumbo_ring[i];
1086 	rc = &sc->bge_cdata.bge_rx_jumbo_chain[i];
1087 
1088 	r->bge_addr.bge_addr_lo = BGE_ADDR_LO(rc->bge_paddr);
1089 	r->bge_addr.bge_addr_hi = BGE_ADDR_HI(rc->bge_paddr);
1090 	r->bge_len = rc->bge_mbuf->m_len;
1091 	r->bge_idx = i;
1092 	r->bge_flags = BGE_RXBDFLAG_END|BGE_RXBDFLAG_JUMBO_RING;
1093 }
1094 
1095 static int
1096 bge_init_rx_ring_std(struct bge_softc *sc)
1097 {
1098 	int i, error;
1099 
1100 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1101 		error = bge_newbuf_std(sc, i, 1);
1102 		if (error)
1103 			return error;
1104 	};
1105 
1106 	sc->bge_std = BGE_STD_RX_RING_CNT - 1;
1107 	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bge_std);
1108 
1109 	return(0);
1110 }
1111 
1112 static void
1113 bge_free_rx_ring_std(struct bge_softc *sc)
1114 {
1115 	int i;
1116 
1117 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1118 		struct bge_rxchain *rc = &sc->bge_cdata.bge_rx_std_chain[i];
1119 
1120 		if (rc->bge_mbuf != NULL) {
1121 			bus_dmamap_unload(sc->bge_cdata.bge_rx_mtag,
1122 					  sc->bge_cdata.bge_rx_std_dmamap[i]);
1123 			m_freem(rc->bge_mbuf);
1124 			rc->bge_mbuf = NULL;
1125 		}
1126 		bzero(&sc->bge_ldata.bge_rx_std_ring[i],
1127 		    sizeof(struct bge_rx_bd));
1128 	}
1129 }
1130 
1131 static int
1132 bge_init_rx_ring_jumbo(struct bge_softc *sc)
1133 {
1134 	struct bge_rcb *rcb;
1135 	int i, error;
1136 
1137 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1138 		error = bge_newbuf_jumbo(sc, i, 1);
1139 		if (error)
1140 			return error;
1141 	};
1142 
1143 	sc->bge_jumbo = BGE_JUMBO_RX_RING_CNT - 1;
1144 
1145 	rcb = &sc->bge_ldata.bge_info.bge_jumbo_rx_rcb;
1146 	rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(0, 0);
1147 	CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1148 
1149 	bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
1150 
1151 	return(0);
1152 }
1153 
1154 static void
1155 bge_free_rx_ring_jumbo(struct bge_softc *sc)
1156 {
1157 	int i;
1158 
1159 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1160 		struct bge_rxchain *rc = &sc->bge_cdata.bge_rx_jumbo_chain[i];
1161 
1162 		if (rc->bge_mbuf != NULL) {
1163 			m_freem(rc->bge_mbuf);
1164 			rc->bge_mbuf = NULL;
1165 		}
1166 		bzero(&sc->bge_ldata.bge_rx_jumbo_ring[i],
1167 		    sizeof(struct bge_rx_bd));
1168 	}
1169 }
1170 
1171 static void
1172 bge_free_tx_ring(struct bge_softc *sc)
1173 {
1174 	int i;
1175 
1176 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
1177 		if (sc->bge_cdata.bge_tx_chain[i] != NULL) {
1178 			bus_dmamap_unload(sc->bge_cdata.bge_tx_mtag,
1179 					  sc->bge_cdata.bge_tx_dmamap[i]);
1180 			m_freem(sc->bge_cdata.bge_tx_chain[i]);
1181 			sc->bge_cdata.bge_tx_chain[i] = NULL;
1182 		}
1183 		bzero(&sc->bge_ldata.bge_tx_ring[i],
1184 		    sizeof(struct bge_tx_bd));
1185 	}
1186 }
1187 
1188 static int
1189 bge_init_tx_ring(struct bge_softc *sc)
1190 {
1191 	sc->bge_txcnt = 0;
1192 	sc->bge_tx_saved_considx = 0;
1193 	sc->bge_tx_prodidx = 0;
1194 
1195 	/* Initialize transmit producer index for host-memory send ring. */
1196 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1197 
1198 	/* 5700 b2 errata */
1199 	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
1200 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1201 
1202 	bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1203 	/* 5700 b2 errata */
1204 	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
1205 		bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1206 
1207 	return(0);
1208 }
1209 
1210 static void
1211 bge_setmulti(struct bge_softc *sc)
1212 {
1213 	struct ifnet *ifp;
1214 	struct ifmultiaddr *ifma;
1215 	uint32_t hashes[4] = { 0, 0, 0, 0 };
1216 	int h, i;
1217 
1218 	ifp = &sc->arpcom.ac_if;
1219 
1220 	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
1221 		for (i = 0; i < 4; i++)
1222 			CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0xFFFFFFFF);
1223 		return;
1224 	}
1225 
1226 	/* First, zot all the existing filters. */
1227 	for (i = 0; i < 4; i++)
1228 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0);
1229 
1230 	/* Now program new ones. */
1231 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1232 		if (ifma->ifma_addr->sa_family != AF_LINK)
1233 			continue;
1234 		h = ether_crc32_le(
1235 		    LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1236 		    ETHER_ADDR_LEN) & 0x7f;
1237 		hashes[(h & 0x60) >> 5] |= 1 << (h & 0x1F);
1238 	}
1239 
1240 	for (i = 0; i < 4; i++)
1241 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), hashes[i]);
1242 }
1243 
1244 /*
1245  * Do endian, PCI and DMA initialization. Also check the on-board ROM
1246  * self-test results.
1247  */
1248 static int
1249 bge_chipinit(struct bge_softc *sc)
1250 {
1251 	int i;
1252 	uint32_t dma_rw_ctl;
1253 	uint16_t val;
1254 
1255 	/* Set endian type before we access any non-PCI registers. */
1256 	pci_write_config(sc->bge_dev, BGE_PCI_MISC_CTL,
1257 	    BGE_INIT | sc->bge_pci_miscctl, 4);
1258 
1259 	/* Clear the MAC control register */
1260 	CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
1261 
1262 	/*
1263 	 * Clear the MAC statistics block in the NIC's
1264 	 * internal memory.
1265 	 */
1266 	for (i = BGE_STATS_BLOCK;
1267 	    i < BGE_STATS_BLOCK_END + 1; i += sizeof(uint32_t))
1268 		BGE_MEMWIN_WRITE(sc, i, 0);
1269 
1270 	for (i = BGE_STATUS_BLOCK;
1271 	    i < BGE_STATUS_BLOCK_END + 1; i += sizeof(uint32_t))
1272 		BGE_MEMWIN_WRITE(sc, i, 0);
1273 
1274 	if (sc->bge_chiprev == BGE_CHIPREV_5704_BX) {
1275 		/*
1276 		 * Fix data corruption caused by non-qword write with WB.
1277 		 * Fix master abort in PCI mode.
1278 		 * Fix PCI latency timer.
1279 		 */
1280 		val = pci_read_config(sc->bge_dev, BGE_PCI_MSI_DATA + 2, 2);
1281 		val |= (1 << 10) | (1 << 12) | (1 << 13);
1282 		pci_write_config(sc->bge_dev, BGE_PCI_MSI_DATA + 2, val, 2);
1283 	}
1284 
1285 	/* Set up the PCI DMA control register. */
1286 	dma_rw_ctl = BGE_PCI_READ_CMD | BGE_PCI_WRITE_CMD;
1287 	if (sc->bge_flags & BGE_FLAG_PCIE) {
1288 		/* PCI-E bus */
1289 		/* DMA read watermark not used on PCI-E */
1290 		dma_rw_ctl |= (0x3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1291 	} else if (sc->bge_flags & BGE_FLAG_PCIX) {
1292 		/* PCI-X bus */
1293 		if (sc->bge_asicrev == BGE_ASICREV_BCM5780) {
1294 			dma_rw_ctl |= (0x4 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1295 			    (0x2 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1296 			dma_rw_ctl |= BGE_PCIDMARWCTL_ONEDMA_ATONCE_GLOBAL;
1297 		} else if (sc->bge_asicrev == BGE_ASICREV_BCM5714) {
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_LOCAL;
1301 		} else if (sc->bge_asicrev == BGE_ASICREV_BCM5703 ||
1302 		    sc->bge_asicrev == BGE_ASICREV_BCM5704) {
1303 			uint32_t rd_wat = 0x7;
1304 			uint32_t clkctl;
1305 
1306 			clkctl = CSR_READ_4(sc, BGE_PCI_CLKCTL) & 0x1f;
1307 			if ((sc->bge_flags & BGE_FLAG_MAXADDR_40BIT) &&
1308 			    sc->bge_asicrev == BGE_ASICREV_BCM5704) {
1309 				dma_rw_ctl |=
1310 				    BGE_PCIDMARWCTL_ONEDMA_ATONCE_LOCAL;
1311 			} else if (clkctl == 0x6 || clkctl == 0x7) {
1312 				dma_rw_ctl |=
1313 				    BGE_PCIDMARWCTL_ONEDMA_ATONCE_GLOBAL;
1314 			}
1315 			if (sc->bge_asicrev == BGE_ASICREV_BCM5703)
1316 				rd_wat = 0x4;
1317 
1318 			dma_rw_ctl |= (rd_wat << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1319 			    (3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1320 			dma_rw_ctl |= BGE_PCIDMARWCTL_ASRT_ALL_BE;
1321 		} else {
1322 			dma_rw_ctl |= (0x3 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1323 			    (0x3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1324 			dma_rw_ctl |= 0xf;
1325 		}
1326 	} else {
1327 		/* Conventional PCI bus */
1328 		dma_rw_ctl |= (0x7 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1329 		    (0x7 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1330 		if (sc->bge_asicrev != BGE_ASICREV_BCM5705 &&
1331 		    sc->bge_asicrev != BGE_ASICREV_BCM5750)
1332 			dma_rw_ctl |= 0xf;
1333 	}
1334 
1335 	if (sc->bge_asicrev == BGE_ASICREV_BCM5703 ||
1336 	    sc->bge_asicrev == BGE_ASICREV_BCM5704) {
1337 		dma_rw_ctl &= ~BGE_PCIDMARWCTL_MINDMA;
1338 	} else if (sc->bge_asicrev == BGE_ASICREV_BCM5700 ||
1339 	    sc->bge_asicrev == BGE_ASICREV_BCM5701) {
1340 		dma_rw_ctl |= BGE_PCIDMARWCTL_USE_MRM |
1341 		    BGE_PCIDMARWCTL_ASRT_ALL_BE;
1342 	}
1343 	pci_write_config(sc->bge_dev, BGE_PCI_DMA_RW_CTL, dma_rw_ctl, 4);
1344 
1345 	/*
1346 	 * Set up general mode register.
1347 	 */
1348 	CSR_WRITE_4(sc, BGE_MODE_CTL, BGE_DMA_SWAP_OPTIONS|
1349 	    BGE_MODECTL_MAC_ATTN_INTR|BGE_MODECTL_HOST_SEND_BDS|
1350 	    BGE_MODECTL_TX_NO_PHDR_CSUM);
1351 
1352 	/*
1353 	 * BCM5701 B5 have a bug causing data corruption when using
1354 	 * 64-bit DMA reads, which can be terminated early and then
1355 	 * completed later as 32-bit accesses, in combination with
1356 	 * certain bridges.
1357 	 */
1358 	if (sc->bge_asicrev == BGE_ASICREV_BCM5701 &&
1359 	    sc->bge_chipid == BGE_CHIPID_BCM5701_B5)
1360 		BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_FORCE_PCI32);
1361 
1362 	/*
1363 	 * Disable memory write invalidate.  Apparently it is not supported
1364 	 * properly by these devices.  Also ensure that INTx isn't disabled,
1365 	 * as these chips need it even when using MSI.
1366 	 */
1367 	PCI_CLRBIT(sc->bge_dev, BGE_PCI_CMD,
1368 	    (PCIM_CMD_MWRICEN | PCIM_CMD_INTxDIS), 4);
1369 
1370 	/* Set the timer prescaler (always 66Mhz) */
1371 	CSR_WRITE_4(sc, BGE_MISC_CFG, 65 << 1/*BGE_32BITTIME_66MHZ*/);
1372 
1373 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
1374 		DELAY(40);	/* XXX */
1375 
1376 		/* Put PHY into ready state */
1377 		BGE_CLRBIT(sc, BGE_MISC_CFG, BGE_MISCCFG_EPHY_IDDQ);
1378 		CSR_READ_4(sc, BGE_MISC_CFG); /* Flush */
1379 		DELAY(40);
1380 	}
1381 
1382 	return(0);
1383 }
1384 
1385 static int
1386 bge_blockinit(struct bge_softc *sc)
1387 {
1388 	struct bge_rcb *rcb;
1389 	bus_size_t vrcb;
1390 	bge_hostaddr taddr;
1391 	uint32_t val;
1392 	int i, limit;
1393 
1394 	/*
1395 	 * Initialize the memory window pointer register so that
1396 	 * we can access the first 32K of internal NIC RAM. This will
1397 	 * allow us to set up the TX send ring RCBs and the RX return
1398 	 * ring RCBs, plus other things which live in NIC memory.
1399 	 */
1400 	CSR_WRITE_4(sc, BGE_PCI_MEMWIN_BASEADDR, 0);
1401 
1402 	/* Note: the BCM5704 has a smaller mbuf space than other chips. */
1403 
1404 	if (!BGE_IS_5705_PLUS(sc)) {
1405 		/* Configure mbuf memory pool */
1406 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_BASEADDR, BGE_BUFFPOOL_1);
1407 		if (sc->bge_asicrev == BGE_ASICREV_BCM5704)
1408 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x10000);
1409 		else
1410 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x18000);
1411 
1412 		/* Configure DMA resource pool */
1413 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_BASEADDR,
1414 		    BGE_DMA_DESCRIPTORS);
1415 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LEN, 0x2000);
1416 	}
1417 
1418 	/* Configure mbuf pool watermarks */
1419 	if (!BGE_IS_5705_PLUS(sc)) {
1420 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x50);
1421 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x20);
1422 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1423 	} else if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
1424 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1425 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x04);
1426 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x10);
1427 	} else {
1428 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1429 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x10);
1430 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1431 	}
1432 
1433 	/* Configure DMA resource watermarks */
1434 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LOWAT, 5);
1435 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_HIWAT, 10);
1436 
1437 	/* Enable buffer manager */
1438 	CSR_WRITE_4(sc, BGE_BMAN_MODE,
1439 	    BGE_BMANMODE_ENABLE|BGE_BMANMODE_LOMBUF_ATTN);
1440 
1441 	/* Poll for buffer manager start indication */
1442 	for (i = 0; i < BGE_TIMEOUT; i++) {
1443 		if (CSR_READ_4(sc, BGE_BMAN_MODE) & BGE_BMANMODE_ENABLE)
1444 			break;
1445 		DELAY(10);
1446 	}
1447 
1448 	if (i == BGE_TIMEOUT) {
1449 		if_printf(&sc->arpcom.ac_if,
1450 			  "buffer manager failed to start\n");
1451 		return(ENXIO);
1452 	}
1453 
1454 	/* Enable flow-through queues */
1455 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
1456 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
1457 
1458 	/* Wait until queue initialization is complete */
1459 	for (i = 0; i < BGE_TIMEOUT; i++) {
1460 		if (CSR_READ_4(sc, BGE_FTQ_RESET) == 0)
1461 			break;
1462 		DELAY(10);
1463 	}
1464 
1465 	if (i == BGE_TIMEOUT) {
1466 		if_printf(&sc->arpcom.ac_if,
1467 			  "flow-through queue init failed\n");
1468 		return(ENXIO);
1469 	}
1470 
1471 	/*
1472 	 * Summary of rings supported by the controller:
1473 	 *
1474 	 * Standard Receive Producer Ring
1475 	 * - This ring is used to feed receive buffers for "standard"
1476 	 *   sized frames (typically 1536 bytes) to the controller.
1477 	 *
1478 	 * Jumbo Receive Producer Ring
1479 	 * - This ring is used to feed receive buffers for jumbo sized
1480 	 *   frames (i.e. anything bigger than the "standard" frames)
1481 	 *   to the controller.
1482 	 *
1483 	 * Mini Receive Producer Ring
1484 	 * - This ring is used to feed receive buffers for "mini"
1485 	 *   sized frames to the controller.
1486 	 * - This feature required external memory for the controller
1487 	 *   but was never used in a production system.  Should always
1488 	 *   be disabled.
1489 	 *
1490 	 * Receive Return Ring
1491 	 * - After the controller has placed an incoming frame into a
1492 	 *   receive buffer that buffer is moved into a receive return
1493 	 *   ring.  The driver is then responsible to passing the
1494 	 *   buffer up to the stack.  Many versions of the controller
1495 	 *   support multiple RR rings.
1496 	 *
1497 	 * Send Ring
1498 	 * - This ring is used for outgoing frames.  Many versions of
1499 	 *   the controller support multiple send rings.
1500 	 */
1501 
1502 	/* Initialize the standard receive producer ring control block. */
1503 	rcb = &sc->bge_ldata.bge_info.bge_std_rx_rcb;
1504 	rcb->bge_hostaddr.bge_addr_lo =
1505 	    BGE_ADDR_LO(sc->bge_ldata.bge_rx_std_ring_paddr);
1506 	rcb->bge_hostaddr.bge_addr_hi =
1507 	    BGE_ADDR_HI(sc->bge_ldata.bge_rx_std_ring_paddr);
1508 	if (BGE_IS_5705_PLUS(sc)) {
1509 		/*
1510 		 * Bits 31-16: Programmable ring size (512, 256, 128, 64, 32)
1511 		 * Bits 15-2 : Reserved (should be 0)
1512 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
1513 		 * Bit 0     : Reserved
1514 		 */
1515 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(512, 0);
1516 	} else {
1517 		/*
1518 		 * Ring size is always XXX entries
1519 		 * Bits 31-16: Maximum RX frame size
1520 		 * Bits 15-2 : Reserved (should be 0)
1521 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
1522 		 * Bit 0     : Reserved
1523 		 */
1524 		rcb->bge_maxlen_flags =
1525 		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN, 0);
1526 	}
1527 	rcb->bge_nicaddr = BGE_STD_RX_RINGS;
1528 	/* Write the standard receive producer ring control block. */
1529 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_HI, rcb->bge_hostaddr.bge_addr_hi);
1530 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_LO, rcb->bge_hostaddr.bge_addr_lo);
1531 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1532 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_NICADDR, rcb->bge_nicaddr);
1533 	/* Reset the standard receive producer ring producer index. */
1534 	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, 0);
1535 
1536 	/*
1537 	 * Initialize the jumbo RX producer ring control
1538 	 * block.  We set the 'ring disabled' bit in the
1539 	 * flags field until we're actually ready to start
1540 	 * using this ring (i.e. once we set the MTU
1541 	 * high enough to require it).
1542 	 */
1543 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
1544 		rcb = &sc->bge_ldata.bge_info.bge_jumbo_rx_rcb;
1545 		/* Get the jumbo receive producer ring RCB parameters. */
1546 		rcb->bge_hostaddr.bge_addr_lo =
1547 		    BGE_ADDR_LO(sc->bge_ldata.bge_rx_jumbo_ring_paddr);
1548 		rcb->bge_hostaddr.bge_addr_hi =
1549 		    BGE_ADDR_HI(sc->bge_ldata.bge_rx_jumbo_ring_paddr);
1550 		rcb->bge_maxlen_flags =
1551 		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN,
1552 		    BGE_RCB_FLAG_RING_DISABLED);
1553 		rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS;
1554 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_HI,
1555 		    rcb->bge_hostaddr.bge_addr_hi);
1556 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_LO,
1557 		    rcb->bge_hostaddr.bge_addr_lo);
1558 		/* Program the jumbo receive producer ring RCB parameters. */
1559 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS,
1560 		    rcb->bge_maxlen_flags);
1561 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_NICADDR, rcb->bge_nicaddr);
1562 		/* Reset the jumbo receive producer ring producer index. */
1563 		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, 0);
1564 	}
1565 
1566 	/* Disable the mini receive producer ring RCB. */
1567 	if (BGE_IS_5700_FAMILY(sc)) {
1568 		rcb = &sc->bge_ldata.bge_info.bge_mini_rx_rcb;
1569 		rcb->bge_maxlen_flags =
1570 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED);
1571 		CSR_WRITE_4(sc, BGE_RX_MINI_RCB_MAXLEN_FLAGS,
1572 		    rcb->bge_maxlen_flags);
1573 		/* Reset the mini receive producer ring producer index. */
1574 		bge_writembx(sc, BGE_MBX_RX_MINI_PROD_LO, 0);
1575 	}
1576 
1577 	/* Choose de-pipeline mode for BCM5906 A0, A1 and A2. */
1578 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
1579 	    (sc->bge_chipid == BGE_CHIPID_BCM5906_A0 ||
1580 	     sc->bge_chipid == BGE_CHIPID_BCM5906_A1 ||
1581 	     sc->bge_chipid == BGE_CHIPID_BCM5906_A2)) {
1582 		CSR_WRITE_4(sc, BGE_ISO_PKT_TX,
1583 		    (CSR_READ_4(sc, BGE_ISO_PKT_TX) & ~3) | 2);
1584 	}
1585 
1586 	/*
1587 	 * The BD ring replenish thresholds control how often the
1588 	 * hardware fetches new BD's from the producer rings in host
1589 	 * memory.  Setting the value too low on a busy system can
1590 	 * starve the hardware and recue the throughpout.
1591 	 *
1592 	 * Set the BD ring replentish thresholds. The recommended
1593 	 * values are 1/8th the number of descriptors allocated to
1594 	 * each ring.
1595 	 */
1596 	if (BGE_IS_5705_PLUS(sc))
1597 		val = 8;
1598 	else
1599 		val = BGE_STD_RX_RING_CNT / 8;
1600 	CSR_WRITE_4(sc, BGE_RBDI_STD_REPL_THRESH, val);
1601 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
1602 		CSR_WRITE_4(sc, BGE_RBDI_JUMBO_REPL_THRESH,
1603 		    BGE_JUMBO_RX_RING_CNT/8);
1604 	}
1605 
1606 	/*
1607 	 * Disable all send rings by setting the 'ring disabled' bit
1608 	 * in the flags field of all the TX send ring control blocks,
1609 	 * located in NIC memory.
1610 	 */
1611 	if (!BGE_IS_5705_PLUS(sc)) {
1612 		/* 5700 to 5704 had 16 send rings. */
1613 		limit = BGE_TX_RINGS_EXTSSRAM_MAX;
1614 	} else {
1615 		limit = 1;
1616 	}
1617 	vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1618 	for (i = 0; i < limit; i++) {
1619 		RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1620 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED));
1621 		RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1622 		vrcb += sizeof(struct bge_rcb);
1623 	}
1624 
1625 	/* Configure send ring RCB 0 (we use only the first ring) */
1626 	vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1627 	BGE_HOSTADDR(taddr, sc->bge_ldata.bge_tx_ring_paddr);
1628 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1629 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1630 	RCB_WRITE_4(sc, vrcb, bge_nicaddr,
1631 	    BGE_NIC_TXRING_ADDR(0, BGE_TX_RING_CNT));
1632 	RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1633 	    BGE_RCB_MAXLEN_FLAGS(BGE_TX_RING_CNT, 0));
1634 
1635 	/*
1636 	 * Disable all receive return rings by setting the
1637 	 * 'ring diabled' bit in the flags field of all the receive
1638 	 * return ring control blocks, located in NIC memory.
1639 	 */
1640 	if (!BGE_IS_5705_PLUS(sc))
1641 		limit = BGE_RX_RINGS_MAX;
1642 	else if (sc->bge_asicrev == BGE_ASICREV_BCM5755)
1643 		limit = 4;
1644 	else
1645 		limit = 1;
1646 	/* Disable all receive return rings. */
1647 	vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1648 	for (i = 0; i < limit; i++) {
1649 		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, 0);
1650 		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, 0);
1651 		RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1652 		    BGE_RCB_FLAG_RING_DISABLED);
1653 		RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1654 		bge_writembx(sc, BGE_MBX_RX_CONS0_LO +
1655 		    (i * (sizeof(uint64_t))), 0);
1656 		vrcb += sizeof(struct bge_rcb);
1657 	}
1658 
1659 	/*
1660 	 * Set up receive return ring 0.  Note that the NIC address
1661 	 * for RX return rings is 0x0.  The return rings live entirely
1662 	 * within the host, so the nicaddr field in the RCB isn't used.
1663 	 */
1664 	vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1665 	BGE_HOSTADDR(taddr, sc->bge_ldata.bge_rx_return_ring_paddr);
1666 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1667 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1668 	RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1669 	RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1670 	    BGE_RCB_MAXLEN_FLAGS(sc->bge_return_ring_cnt, 0));
1671 
1672 	/* Set random backoff seed for TX */
1673 	CSR_WRITE_4(sc, BGE_TX_RANDOM_BACKOFF,
1674 	    sc->arpcom.ac_enaddr[0] + sc->arpcom.ac_enaddr[1] +
1675 	    sc->arpcom.ac_enaddr[2] + sc->arpcom.ac_enaddr[3] +
1676 	    sc->arpcom.ac_enaddr[4] + sc->arpcom.ac_enaddr[5] +
1677 	    BGE_TX_BACKOFF_SEED_MASK);
1678 
1679 	/* Set inter-packet gap */
1680 	CSR_WRITE_4(sc, BGE_TX_LENGTHS, 0x2620);
1681 
1682 	/*
1683 	 * Specify which ring to use for packets that don't match
1684 	 * any RX rules.
1685 	 */
1686 	CSR_WRITE_4(sc, BGE_RX_RULES_CFG, 0x08);
1687 
1688 	/*
1689 	 * Configure number of RX lists. One interrupt distribution
1690 	 * list, sixteen active lists, one bad frames class.
1691 	 */
1692 	CSR_WRITE_4(sc, BGE_RXLP_CFG, 0x181);
1693 
1694 	/* Inialize RX list placement stats mask. */
1695 	CSR_WRITE_4(sc, BGE_RXLP_STATS_ENABLE_MASK, 0x007FFFFF);
1696 	CSR_WRITE_4(sc, BGE_RXLP_STATS_CTL, 0x1);
1697 
1698 	/* Disable host coalescing until we get it set up */
1699 	CSR_WRITE_4(sc, BGE_HCC_MODE, 0x00000000);
1700 
1701 	/* Poll to make sure it's shut down. */
1702 	for (i = 0; i < BGE_TIMEOUT; i++) {
1703 		if (!(CSR_READ_4(sc, BGE_HCC_MODE) & BGE_HCCMODE_ENABLE))
1704 			break;
1705 		DELAY(10);
1706 	}
1707 
1708 	if (i == BGE_TIMEOUT) {
1709 		if_printf(&sc->arpcom.ac_if,
1710 			  "host coalescing engine failed to idle\n");
1711 		return(ENXIO);
1712 	}
1713 
1714 	/* Set up host coalescing defaults */
1715 	CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, sc->bge_rx_coal_ticks);
1716 	CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS, sc->bge_tx_coal_ticks);
1717 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, sc->bge_rx_coal_bds);
1718 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS, sc->bge_tx_coal_bds);
1719 	if (!BGE_IS_5705_PLUS(sc)) {
1720 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS_INT,
1721 		    sc->bge_rx_coal_ticks_int);
1722 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS_INT,
1723 		    sc->bge_tx_coal_ticks_int);
1724 	}
1725 	/*
1726 	 * NOTE:
1727 	 * The datasheet (57XX-PG105-R) says BCM5705+ do not
1728 	 * have following two registers; obviously it is wrong.
1729 	 */
1730 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT, sc->bge_rx_coal_bds_int);
1731 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT, sc->bge_tx_coal_bds_int);
1732 
1733 	/* Set up address of statistics block */
1734 	if (!BGE_IS_5705_PLUS(sc)) {
1735 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_HI,
1736 		    BGE_ADDR_HI(sc->bge_ldata.bge_stats_paddr));
1737 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_LO,
1738 		    BGE_ADDR_LO(sc->bge_ldata.bge_stats_paddr));
1739 
1740 		CSR_WRITE_4(sc, BGE_HCC_STATS_BASEADDR, BGE_STATS_BLOCK);
1741 		CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_BASEADDR, BGE_STATUS_BLOCK);
1742 		CSR_WRITE_4(sc, BGE_HCC_STATS_TICKS, sc->bge_stat_ticks);
1743 	}
1744 
1745 	/* Set up address of status block */
1746 	bzero(sc->bge_ldata.bge_status_block, BGE_STATUS_BLK_SZ);
1747 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_HI,
1748 	    BGE_ADDR_HI(sc->bge_ldata.bge_status_block_paddr));
1749 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_LO,
1750 	    BGE_ADDR_LO(sc->bge_ldata.bge_status_block_paddr));
1751 
1752 	/*
1753 	 * Set up status block partail update size.
1754 	 *
1755 	 * Because only single TX ring, RX produce ring and Rx return ring
1756 	 * are used, ask device to update only minimum part of status block
1757 	 * except for BCM5700 AX/BX, whose status block partial update size
1758 	 * can't be configured.
1759 	 */
1760 	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
1761 	    sc->bge_chipid != BGE_CHIPID_BCM5700_C0) {
1762 		/* XXX Actually reserved on BCM5700 AX/BX */
1763 		val = BGE_STATBLKSZ_FULL;
1764 	} else {
1765 		val = BGE_STATBLKSZ_32BYTE;
1766 	}
1767 #if 0
1768 	/*
1769 	 * Does not seem to have visible effect in both
1770 	 * bulk data (1472B UDP datagram) and tiny data
1771 	 * (18B UDP datagram) TX tests.
1772 	 */
1773 	if (!BGE_IS_CRIPPLED(sc))
1774 		val |= BGE_HCCMODE_CLRTICK_TX;
1775 #endif
1776 
1777 	/* Turn on host coalescing state machine */
1778 	CSR_WRITE_4(sc, BGE_HCC_MODE, val | BGE_HCCMODE_ENABLE);
1779 
1780 	/* Turn on RX BD completion state machine and enable attentions */
1781 	CSR_WRITE_4(sc, BGE_RBDC_MODE,
1782 	    BGE_RBDCMODE_ENABLE|BGE_RBDCMODE_ATTN);
1783 
1784 	/* Turn on RX list placement state machine */
1785 	CSR_WRITE_4(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
1786 
1787 	/* Turn on RX list selector state machine. */
1788 	if (!BGE_IS_5705_PLUS(sc))
1789 		CSR_WRITE_4(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
1790 
1791 	val = BGE_MACMODE_TXDMA_ENB | BGE_MACMODE_RXDMA_ENB |
1792 	    BGE_MACMODE_RX_STATS_CLEAR | BGE_MACMODE_TX_STATS_CLEAR |
1793 	    BGE_MACMODE_RX_STATS_ENB | BGE_MACMODE_TX_STATS_ENB |
1794 	    BGE_MACMODE_FRMHDR_DMA_ENB;
1795 
1796 	if (sc->bge_flags & BGE_FLAG_TBI)
1797 		val |= BGE_PORTMODE_TBI;
1798 	else if (sc->bge_flags & BGE_FLAG_MII_SERDES)
1799 		val |= BGE_PORTMODE_GMII;
1800 	else
1801 		val |= BGE_PORTMODE_MII;
1802 
1803 	/* Turn on DMA, clear stats */
1804 	CSR_WRITE_4(sc, BGE_MAC_MODE, val);
1805 
1806 	/* Set misc. local control, enable interrupts on attentions */
1807 	CSR_WRITE_4(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_ONATTN);
1808 
1809 #ifdef notdef
1810 	/* Assert GPIO pins for PHY reset */
1811 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUT0|
1812 	    BGE_MLC_MISCIO_OUT1|BGE_MLC_MISCIO_OUT2);
1813 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUTEN0|
1814 	    BGE_MLC_MISCIO_OUTEN1|BGE_MLC_MISCIO_OUTEN2);
1815 #endif
1816 
1817 	/* Turn on DMA completion state machine */
1818 	if (!BGE_IS_5705_PLUS(sc))
1819 		CSR_WRITE_4(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
1820 
1821 	/* Turn on write DMA state machine */
1822 	val = BGE_WDMAMODE_ENABLE|BGE_WDMAMODE_ALL_ATTNS;
1823 	if (BGE_IS_5755_PLUS(sc)) {
1824 		/* Enable host coalescing bug fix. */
1825 		val |= BGE_WDMAMODE_STATUS_TAG_FIX;
1826 	}
1827 	if (sc->bge_asicrev == BGE_ASICREV_BCM5785) {
1828 		/* Request larger DMA burst size to get better performance. */
1829 		val |= BGE_WDMAMODE_BURST_ALL_DATA;
1830 	}
1831 	CSR_WRITE_4(sc, BGE_WDMA_MODE, val);
1832 	DELAY(40);
1833 
1834 	if (sc->bge_asicrev == BGE_ASICREV_BCM5761 ||
1835 	    sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
1836 	    sc->bge_asicrev == BGE_ASICREV_BCM5785 ||
1837 	    sc->bge_asicrev == BGE_ASICREV_BCM57780) {
1838 		/*
1839 		 * Enable fix for read DMA FIFO overruns.
1840 		 * The fix is to limit the number of RX BDs
1841 		 * the hardware would fetch at a fime.
1842 		 */
1843 		val = CSR_READ_4(sc, BGE_RDMA_RSRVCTRL);
1844 		CSR_WRITE_4(sc, BGE_RDMA_RSRVCTRL,
1845 		    val| BGE_RDMA_RSRVCTRL_FIFO_OFLW_FIX);
1846 	}
1847 
1848 	/* Turn on read DMA state machine */
1849 	val = BGE_RDMAMODE_ENABLE | BGE_RDMAMODE_ALL_ATTNS;
1850         if (sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
1851             sc->bge_asicrev == BGE_ASICREV_BCM5785 ||
1852             sc->bge_asicrev == BGE_ASICREV_BCM57780)
1853 		val |= BGE_RDMAMODE_BD_SBD_CRPT_ATTN |
1854                   BGE_RDMAMODE_MBUF_RBD_CRPT_ATTN |
1855                   BGE_RDMAMODE_MBUF_SBD_CRPT_ATTN;
1856 	if (sc->bge_flags & BGE_FLAG_PCIE)
1857 		val |= BGE_RDMAMODE_FIFO_LONG_BURST;
1858 	if (sc->bge_flags & BGE_FLAG_TSO)
1859 		val |= BGE_RDMAMODE_TSO4_ENABLE;
1860 	CSR_WRITE_4(sc, BGE_RDMA_MODE, val);
1861 	DELAY(40);
1862 
1863 	/* Turn on RX data completion state machine */
1864 	CSR_WRITE_4(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
1865 
1866 	/* Turn on RX BD initiator state machine */
1867 	CSR_WRITE_4(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
1868 
1869 	/* Turn on RX data and RX BD initiator state machine */
1870 	CSR_WRITE_4(sc, BGE_RDBDI_MODE, BGE_RDBDIMODE_ENABLE);
1871 
1872 	/* Turn on Mbuf cluster free state machine */
1873 	if (!BGE_IS_5705_PLUS(sc))
1874 		CSR_WRITE_4(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
1875 
1876 	/* Turn on send BD completion state machine */
1877 	CSR_WRITE_4(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
1878 
1879 	/* Turn on send data completion state machine */
1880 	val = BGE_SDCMODE_ENABLE;
1881 	if (sc->bge_asicrev == BGE_ASICREV_BCM5761)
1882 		val |= BGE_SDCMODE_CDELAY;
1883 	CSR_WRITE_4(sc, BGE_SDC_MODE, val);
1884 
1885 	/* Turn on send data initiator state machine */
1886 	if (sc->bge_flags & BGE_FLAG_TSO)
1887 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE |
1888 		    BGE_SDIMODE_HW_LSO_PRE_DMA);
1889 	else
1890 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
1891 
1892 	/* Turn on send BD initiator state machine */
1893 	CSR_WRITE_4(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
1894 
1895 	/* Turn on send BD selector state machine */
1896 	CSR_WRITE_4(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
1897 
1898 	CSR_WRITE_4(sc, BGE_SDI_STATS_ENABLE_MASK, 0x007FFFFF);
1899 	CSR_WRITE_4(sc, BGE_SDI_STATS_CTL,
1900 	    BGE_SDISTATSCTL_ENABLE|BGE_SDISTATSCTL_FASTER);
1901 
1902 	/* ack/clear link change events */
1903 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
1904 	    BGE_MACSTAT_CFG_CHANGED|BGE_MACSTAT_MI_COMPLETE|
1905 	    BGE_MACSTAT_LINK_CHANGED);
1906 	CSR_WRITE_4(sc, BGE_MI_STS, 0);
1907 
1908 	/*
1909 	 * Enable attention when the link has changed state for
1910 	 * devices that use auto polling.
1911 	 */
1912 	if (sc->bge_flags & BGE_FLAG_TBI) {
1913 		CSR_WRITE_4(sc, BGE_MI_STS, BGE_MISTS_LINK);
1914  	} else {
1915 		if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
1916 			CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode);
1917 			DELAY(80);
1918 		}
1919 		if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
1920 		    sc->bge_chipid != BGE_CHIPID_BCM5700_B2) {
1921 			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
1922 			    BGE_EVTENB_MI_INTERRUPT);
1923 		}
1924 	}
1925 
1926 	/*
1927 	 * Clear any pending link state attention.
1928 	 * Otherwise some link state change events may be lost until attention
1929 	 * is cleared by bge_intr() -> bge_softc.bge_link_upd() sequence.
1930 	 * It's not necessary on newer BCM chips - perhaps enabling link
1931 	 * state change attentions implies clearing pending attention.
1932 	 */
1933 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
1934 	    BGE_MACSTAT_CFG_CHANGED|BGE_MACSTAT_MI_COMPLETE|
1935 	    BGE_MACSTAT_LINK_CHANGED);
1936 
1937 	/* Enable link state change attentions. */
1938 	BGE_SETBIT(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_LINK_CHANGED);
1939 
1940 	return(0);
1941 }
1942 
1943 /*
1944  * Probe for a Broadcom chip. Check the PCI vendor and device IDs
1945  * against our list and return its name if we find a match. Note
1946  * that since the Broadcom controller contains VPD support, we
1947  * can get the device name string from the controller itself instead
1948  * of the compiled-in string. This is a little slow, but it guarantees
1949  * we'll always announce the right product name.
1950  */
1951 static int
1952 bge_probe(device_t dev)
1953 {
1954 	const struct bge_type *t;
1955 	uint16_t product, vendor;
1956 
1957 	product = pci_get_device(dev);
1958 	vendor = pci_get_vendor(dev);
1959 
1960 	for (t = bge_devs; t->bge_name != NULL; t++) {
1961 		if (vendor == t->bge_vid && product == t->bge_did)
1962 			break;
1963 	}
1964 	if (t->bge_name == NULL)
1965 		return(ENXIO);
1966 
1967 	device_set_desc(dev, t->bge_name);
1968 	return(0);
1969 }
1970 
1971 static int
1972 bge_attach(device_t dev)
1973 {
1974 	struct ifnet *ifp;
1975 	struct bge_softc *sc;
1976 	uint32_t hwcfg = 0, misccfg;
1977 	int error = 0, rid, capmask;
1978 	uint8_t ether_addr[ETHER_ADDR_LEN];
1979 	uint16_t product, vendor;
1980 	driver_intr_t *intr_func;
1981 	uintptr_t mii_priv = 0;
1982 	u_int intr_flags;
1983 	int msi_enable;
1984 
1985 	sc = device_get_softc(dev);
1986 	sc->bge_dev = dev;
1987 	callout_init_mp(&sc->bge_stat_timer);
1988 	lwkt_serialize_init(&sc->bge_jslot_serializer);
1989 
1990 	product = pci_get_device(dev);
1991 	vendor = pci_get_vendor(dev);
1992 
1993 #ifndef BURN_BRIDGES
1994 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1995 		uint32_t irq, mem;
1996 
1997 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
1998 		mem = pci_read_config(dev, BGE_PCI_BAR0, 4);
1999 
2000 		device_printf(dev, "chip is in D%d power mode "
2001 		    "-- setting to D0\n", pci_get_powerstate(dev));
2002 
2003 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
2004 
2005 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
2006 		pci_write_config(dev, BGE_PCI_BAR0, mem, 4);
2007 	}
2008 #endif	/* !BURN_BRIDGE */
2009 
2010 	/*
2011 	 * Map control/status registers.
2012 	 */
2013 	pci_enable_busmaster(dev);
2014 
2015 	rid = BGE_PCI_BAR0;
2016 	sc->bge_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
2017 	    RF_ACTIVE);
2018 
2019 	if (sc->bge_res == NULL) {
2020 		device_printf(dev, "couldn't map memory\n");
2021 		return ENXIO;
2022 	}
2023 
2024 	sc->bge_btag = rman_get_bustag(sc->bge_res);
2025 	sc->bge_bhandle = rman_get_bushandle(sc->bge_res);
2026 
2027 	/* Save various chip information */
2028 	sc->bge_chipid =
2029 	    pci_read_config(dev, BGE_PCI_MISC_CTL, 4) >>
2030 	    BGE_PCIMISCCTL_ASICREV_SHIFT;
2031 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_USE_PRODID_REG) {
2032 		/* All chips, which use BGE_PCI_PRODID_ASICREV, have CPMU */
2033 		sc->bge_flags |= BGE_FLAG_CPMU;
2034 		sc->bge_chipid = pci_read_config(dev, BGE_PCI_PRODID_ASICREV, 4);
2035 	}
2036 	sc->bge_asicrev = BGE_ASICREV(sc->bge_chipid);
2037 	sc->bge_chiprev = BGE_CHIPREV(sc->bge_chipid);
2038 
2039 	/* Save chipset family. */
2040 	switch (sc->bge_asicrev) {
2041 	case BGE_ASICREV_BCM5755:
2042 	case BGE_ASICREV_BCM5761:
2043 	case BGE_ASICREV_BCM5784:
2044 	case BGE_ASICREV_BCM5785:
2045 	case BGE_ASICREV_BCM5787:
2046 	case BGE_ASICREV_BCM57780:
2047 	    sc->bge_flags |= BGE_FLAG_5755_PLUS | BGE_FLAG_575X_PLUS |
2048 		BGE_FLAG_5705_PLUS;
2049 	    break;
2050 
2051 	case BGE_ASICREV_BCM5700:
2052 	case BGE_ASICREV_BCM5701:
2053 	case BGE_ASICREV_BCM5703:
2054 	case BGE_ASICREV_BCM5704:
2055 		sc->bge_flags |= BGE_FLAG_5700_FAMILY | BGE_FLAG_JUMBO;
2056 		break;
2057 
2058 	case BGE_ASICREV_BCM5714_A0:
2059 	case BGE_ASICREV_BCM5780:
2060 	case BGE_ASICREV_BCM5714:
2061 		sc->bge_flags |= BGE_FLAG_5714_FAMILY;
2062 		/* Fall through */
2063 
2064 	case BGE_ASICREV_BCM5750:
2065 	case BGE_ASICREV_BCM5752:
2066 	case BGE_ASICREV_BCM5906:
2067 		sc->bge_flags |= BGE_FLAG_575X_PLUS;
2068 		/* Fall through */
2069 
2070 	case BGE_ASICREV_BCM5705:
2071 		sc->bge_flags |= BGE_FLAG_5705_PLUS;
2072 		break;
2073 	}
2074 
2075 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
2076 		sc->bge_flags |= BGE_FLAG_NO_EEPROM;
2077 
2078 	misccfg = CSR_READ_4(sc, BGE_MISC_CFG) & BGE_MISCCFG_BOARD_ID_MASK;
2079 	if (sc->bge_asicrev == BGE_ASICREV_BCM5705 &&
2080 	    (misccfg == BGE_MISCCFG_BOARD_ID_5788 ||
2081 	     misccfg == BGE_MISCCFG_BOARD_ID_5788M))
2082 		sc->bge_flags |= BGE_FLAG_5788;
2083 
2084 	/* BCM5755 or higher and BCM5906 have short DMA bug. */
2085 	if (BGE_IS_5755_PLUS(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5906)
2086 		sc->bge_flags |= BGE_FLAG_SHORTDMA;
2087 
2088   	/*
2089 	 * Check if this is a PCI-X or PCI Express device.
2090   	 */
2091 	if (BGE_IS_5705_PLUS(sc)) {
2092 		if (pci_is_pcie(dev)) {
2093 			sc->bge_flags |= BGE_FLAG_PCIE;
2094 			sc->bge_pciecap = pci_get_pciecap_ptr(sc->bge_dev);
2095 			pcie_set_max_readrq(dev, PCIEM_DEVCTL_MAX_READRQ_4096);
2096 		}
2097 	} else {
2098 		/*
2099 		 * Check if the device is in PCI-X Mode.
2100 		 * (This bit is not valid on PCI Express controllers.)
2101 		 */
2102 		if ((pci_read_config(sc->bge_dev, BGE_PCI_PCISTATE, 4) &
2103 		    BGE_PCISTATE_PCI_BUSMODE) == 0) {
2104 			sc->bge_flags |= BGE_FLAG_PCIX;
2105 			sc->bge_pcixcap = pci_get_pcixcap_ptr(sc->bge_dev);
2106 			sc->bge_mbox_reorder = device_getenv_int(sc->bge_dev,
2107 			    "mbox_reorder", 0);
2108 		}
2109  	}
2110 	device_printf(dev, "CHIP ID 0x%08x; "
2111 		      "ASIC REV 0x%02x; CHIP REV 0x%02x; %s\n",
2112 		      sc->bge_chipid, sc->bge_asicrev, sc->bge_chiprev,
2113 		      (sc->bge_flags & BGE_FLAG_PCIX) ? "PCI-X"
2114 		      : ((sc->bge_flags & BGE_FLAG_PCIE) ?
2115 			"PCI-E" : "PCI"));
2116 
2117 	/*
2118 	 * The 40bit DMA bug applies to the 5714/5715 controllers and is
2119 	 * not actually a MAC controller bug but an issue with the embedded
2120 	 * PCIe to PCI-X bridge in the device. Use 40bit DMA workaround.
2121 	 */
2122 	if ((sc->bge_flags & BGE_FLAG_PCIX) &&
2123 	    (BGE_IS_5714_FAMILY(sc) || device_getenv_int(dev, "dma40b", 0)))
2124 		sc->bge_flags |= BGE_FLAG_MAXADDR_40BIT;
2125 
2126 	/*
2127 	 * When using the BCM5701 in PCI-X mode, data corruption has
2128 	 * been observed in the first few bytes of some received packets.
2129 	 * Aligning the packet buffer in memory eliminates the corruption.
2130 	 * Unfortunately, this misaligns the packet payloads.  On platforms
2131 	 * which do not support unaligned accesses, we will realign the
2132 	 * payloads by copying the received packets.
2133 	 */
2134 	if (sc->bge_asicrev == BGE_ASICREV_BCM5701 &&
2135 	    (sc->bge_flags & BGE_FLAG_PCIX))
2136 		sc->bge_flags |= BGE_FLAG_RX_ALIGNBUG;
2137 
2138 	if (!BGE_IS_CRIPPLED(sc)) {
2139 		if (device_getenv_int(dev, "status_tag", 1)) {
2140 			sc->bge_flags |= BGE_FLAG_STATUS_TAG;
2141 			sc->bge_pci_miscctl = BGE_PCIMISCCTL_TAGGED_STATUS;
2142 			if (bootverbose)
2143 				device_printf(dev, "enable status tag\n");
2144 		}
2145 	}
2146 
2147 	if (BGE_IS_5755_PLUS(sc)) {
2148 		/*
2149 		 * BCM5754 and BCM5787 shares the same ASIC id so
2150 		 * explicit device id check is required.
2151 		 * Due to unknown reason TSO does not work on BCM5755M.
2152 		 */
2153 		if (product != PCI_PRODUCT_BROADCOM_BCM5754 &&
2154 		    product != PCI_PRODUCT_BROADCOM_BCM5754M &&
2155 		    product != PCI_PRODUCT_BROADCOM_BCM5755M)
2156 			sc->bge_flags |= BGE_FLAG_TSO;
2157 	}
2158 
2159 	/*
2160 	 * Set various PHY quirk flags.
2161 	 */
2162 
2163 	if ((sc->bge_asicrev == BGE_ASICREV_BCM5700 ||
2164 	     sc->bge_asicrev == BGE_ASICREV_BCM5701) &&
2165 	    pci_get_subvendor(dev) == PCI_VENDOR_DELL)
2166 		mii_priv |= BRGPHY_FLAG_NO_3LED;
2167 
2168 	capmask = MII_CAPMASK_DEFAULT;
2169 	if ((sc->bge_asicrev == BGE_ASICREV_BCM5703 &&
2170 	     (misccfg == 0x4000 || misccfg == 0x8000)) ||
2171 	    (sc->bge_asicrev == BGE_ASICREV_BCM5705 &&
2172 	     vendor == PCI_VENDOR_BROADCOM &&
2173 	     (product == PCI_PRODUCT_BROADCOM_BCM5901 ||
2174 	      product == PCI_PRODUCT_BROADCOM_BCM5901A2 ||
2175 	      product == PCI_PRODUCT_BROADCOM_BCM5705F)) ||
2176 	    (vendor == PCI_VENDOR_BROADCOM &&
2177 	     (product == PCI_PRODUCT_BROADCOM_BCM5751F ||
2178 	      product == PCI_PRODUCT_BROADCOM_BCM5753F ||
2179 	      product == PCI_PRODUCT_BROADCOM_BCM5787F)) ||
2180 	    product == PCI_PRODUCT_BROADCOM_BCM57790 ||
2181 	    sc->bge_asicrev == BGE_ASICREV_BCM5906) {
2182 		/* 10/100 only */
2183 		capmask &= ~BMSR_EXTSTAT;
2184 	}
2185 
2186 	mii_priv |= BRGPHY_FLAG_WIRESPEED;
2187 	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 ||
2188 	    (sc->bge_asicrev == BGE_ASICREV_BCM5705 &&
2189 	     (sc->bge_chipid != BGE_CHIPID_BCM5705_A0 &&
2190 	      sc->bge_chipid != BGE_CHIPID_BCM5705_A1)) ||
2191 	    sc->bge_asicrev == BGE_ASICREV_BCM5906)
2192 		mii_priv &= ~BRGPHY_FLAG_WIRESPEED;
2193 
2194 	if (sc->bge_chipid == BGE_CHIPID_BCM5701_A0 ||
2195 	    sc->bge_chipid == BGE_CHIPID_BCM5701_B0)
2196 		mii_priv |= BRGPHY_FLAG_CRC_BUG;
2197 
2198 	if (sc->bge_chiprev == BGE_CHIPREV_5703_AX ||
2199 	    sc->bge_chiprev == BGE_CHIPREV_5704_AX)
2200 		mii_priv |= BRGPHY_FLAG_ADC_BUG;
2201 
2202 	if (sc->bge_chipid == BGE_CHIPID_BCM5704_A0)
2203 		mii_priv |= BRGPHY_FLAG_5704_A0;
2204 
2205 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
2206 		mii_priv |= BRGPHY_FLAG_5906;
2207 
2208 	if (BGE_IS_5705_PLUS(sc) &&
2209 	    sc->bge_asicrev != BGE_ASICREV_BCM5906 &&
2210 	    /* sc->bge_asicrev != BGE_ASICREV_BCM5717 && */
2211 	    sc->bge_asicrev != BGE_ASICREV_BCM5785 &&
2212 	    /* sc->bge_asicrev != BGE_ASICREV_BCM57765 && */
2213 	    sc->bge_asicrev != BGE_ASICREV_BCM57780) {
2214 		if (sc->bge_asicrev == BGE_ASICREV_BCM5755 ||
2215 		    sc->bge_asicrev == BGE_ASICREV_BCM5761 ||
2216 		    sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
2217 		    sc->bge_asicrev == BGE_ASICREV_BCM5787) {
2218 			if (product != PCI_PRODUCT_BROADCOM_BCM5722 &&
2219 			    product != PCI_PRODUCT_BROADCOM_BCM5756)
2220 				mii_priv |= BRGPHY_FLAG_JITTER_BUG;
2221 			if (product == PCI_PRODUCT_BROADCOM_BCM5755M)
2222 				mii_priv |= BRGPHY_FLAG_ADJUST_TRIM;
2223 		} else {
2224 			mii_priv |= BRGPHY_FLAG_BER_BUG;
2225 		}
2226 	}
2227 
2228 	/*
2229 	 * Allocate interrupt
2230 	 */
2231 	msi_enable = bge_msi_enable;
2232 	if ((sc->bge_flags & BGE_FLAG_STATUS_TAG) == 0) {
2233 		/* If "tagged status" is disabled, don't enable MSI */
2234 		msi_enable = 0;
2235 	} else if (msi_enable) {
2236 		msi_enable = 0; /* Disable by default */
2237 		if (BGE_IS_575X_PLUS(sc)) {
2238 			msi_enable = 1;
2239 			/* XXX we filter all 5714 chips */
2240 			if (sc->bge_asicrev == BGE_ASICREV_BCM5714 ||
2241 			    (sc->bge_asicrev == BGE_ASICREV_BCM5750 &&
2242 			     (sc->bge_chiprev == BGE_CHIPREV_5750_AX ||
2243 			      sc->bge_chiprev == BGE_CHIPREV_5750_BX)))
2244 				msi_enable = 0;
2245 			else if (BGE_IS_5755_PLUS(sc) ||
2246 			    sc->bge_asicrev == BGE_ASICREV_BCM5906)
2247 				sc->bge_flags |= BGE_FLAG_ONESHOT_MSI;
2248 		}
2249 	}
2250 	if (msi_enable) {
2251 		if (pci_find_extcap(dev, PCIY_MSI, &sc->bge_msicap)) {
2252 			device_printf(dev, "no MSI capability\n");
2253 			msi_enable = 0;
2254 		}
2255 	}
2256 
2257 	sc->bge_irq_type = pci_alloc_1intr(dev, msi_enable, &sc->bge_irq_rid,
2258 	    &intr_flags);
2259 
2260 	sc->bge_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &sc->bge_irq_rid,
2261 	    intr_flags);
2262 	if (sc->bge_irq == NULL) {
2263 		device_printf(dev, "couldn't map interrupt\n");
2264 		error = ENXIO;
2265 		goto fail;
2266 	}
2267 
2268 	if (sc->bge_irq_type == PCI_INTR_TYPE_MSI)
2269 		bge_enable_msi(sc);
2270 	else
2271 		sc->bge_flags &= ~BGE_FLAG_ONESHOT_MSI;
2272 
2273 	/* Initialize if_name earlier, so if_printf could be used */
2274 	ifp = &sc->arpcom.ac_if;
2275 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
2276 
2277 	/* Try to reset the chip. */
2278 	bge_reset(sc);
2279 
2280 	if (bge_chipinit(sc)) {
2281 		device_printf(dev, "chip initialization failed\n");
2282 		error = ENXIO;
2283 		goto fail;
2284 	}
2285 
2286 	/*
2287 	 * Get station address
2288 	 */
2289 	error = bge_get_eaddr(sc, ether_addr);
2290 	if (error) {
2291 		device_printf(dev, "failed to read station address\n");
2292 		goto fail;
2293 	}
2294 
2295 	/* 5705/5750 limits RX return ring to 512 entries. */
2296 	if (BGE_IS_5705_PLUS(sc))
2297 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT_5705;
2298 	else
2299 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT;
2300 
2301 	error = bge_dma_alloc(sc);
2302 	if (error)
2303 		goto fail;
2304 
2305 	/* Set default tuneable values. */
2306 	sc->bge_stat_ticks = BGE_TICKS_PER_SEC;
2307 	sc->bge_rx_coal_ticks = BGE_RX_COAL_TICKS_DEF;
2308 	sc->bge_tx_coal_ticks = BGE_TX_COAL_TICKS_DEF;
2309 	sc->bge_rx_coal_bds = BGE_RX_COAL_BDS_DEF;
2310 	sc->bge_tx_coal_bds = BGE_TX_COAL_BDS_DEF;
2311 	if (sc->bge_flags & BGE_FLAG_STATUS_TAG) {
2312 		sc->bge_rx_coal_ticks_int = BGE_RX_COAL_TICKS_DEF;
2313 		sc->bge_tx_coal_ticks_int = BGE_TX_COAL_TICKS_DEF;
2314 		sc->bge_rx_coal_bds_int = BGE_RX_COAL_BDS_DEF;
2315 		sc->bge_tx_coal_bds_int = BGE_TX_COAL_BDS_DEF;
2316 	} else {
2317 		sc->bge_rx_coal_ticks_int = BGE_RX_COAL_TICKS_MIN;
2318 		sc->bge_tx_coal_ticks_int = BGE_TX_COAL_TICKS_MIN;
2319 		sc->bge_rx_coal_bds_int = BGE_RX_COAL_BDS_MIN;
2320 		sc->bge_tx_coal_bds_int = BGE_TX_COAL_BDS_MIN;
2321 	}
2322 
2323 	/* Set up TX spare and reserved descriptor count */
2324 	if (sc->bge_flags & BGE_FLAG_TSO) {
2325 		sc->bge_txspare = BGE_NSEG_SPARE_TSO;
2326 		sc->bge_txrsvd = BGE_NSEG_RSVD_TSO;
2327 	} else {
2328 		sc->bge_txspare = BGE_NSEG_SPARE;
2329 		sc->bge_txrsvd = BGE_NSEG_RSVD;
2330 	}
2331 
2332 	/* Set up ifnet structure */
2333 	ifp->if_softc = sc;
2334 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2335 	ifp->if_ioctl = bge_ioctl;
2336 	ifp->if_start = bge_start;
2337 #ifdef DEVICE_POLLING
2338 	ifp->if_poll = bge_poll;
2339 #endif
2340 	ifp->if_watchdog = bge_watchdog;
2341 	ifp->if_init = bge_init;
2342 	ifp->if_mtu = ETHERMTU;
2343 	ifp->if_capabilities = IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU;
2344 	ifq_set_maxlen(&ifp->if_snd, BGE_TX_RING_CNT - 1);
2345 	ifq_set_ready(&ifp->if_snd);
2346 
2347 	/*
2348 	 * 5700 B0 chips do not support checksumming correctly due
2349 	 * to hardware bugs.
2350 	 */
2351 	if (sc->bge_chipid != BGE_CHIPID_BCM5700_B0) {
2352 		ifp->if_capabilities |= IFCAP_HWCSUM;
2353 		ifp->if_hwassist |= BGE_CSUM_FEATURES;
2354 	}
2355 	if (sc->bge_flags & BGE_FLAG_TSO) {
2356 		ifp->if_capabilities |= IFCAP_TSO;
2357 		ifp->if_hwassist |= CSUM_TSO;
2358 	}
2359 	ifp->if_capenable = ifp->if_capabilities;
2360 
2361 	/*
2362 	 * Figure out what sort of media we have by checking the
2363 	 * hardware config word in the first 32k of NIC internal memory,
2364 	 * or fall back to examining the EEPROM if necessary.
2365 	 * Note: on some BCM5700 cards, this value appears to be unset.
2366 	 * If that's the case, we have to rely on identifying the NIC
2367 	 * by its PCI subsystem ID, as we do below for the SysKonnect
2368 	 * SK-9D41.
2369 	 */
2370 	if (bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_SIG) == BGE_MAGIC_NUMBER) {
2371 		hwcfg = bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_NICCFG);
2372 	} else {
2373 		if (bge_read_eeprom(sc, (caddr_t)&hwcfg, BGE_EE_HWCFG_OFFSET,
2374 				    sizeof(hwcfg))) {
2375 			device_printf(dev, "failed to read EEPROM\n");
2376 			error = ENXIO;
2377 			goto fail;
2378 		}
2379 		hwcfg = ntohl(hwcfg);
2380 	}
2381 
2382 	/* The SysKonnect SK-9D41 is a 1000baseSX card. */
2383 	if (pci_get_subvendor(dev) == PCI_PRODUCT_SCHNEIDERKOCH_SK_9D41 ||
2384 	    (hwcfg & BGE_HWCFG_MEDIA) == BGE_MEDIA_FIBER) {
2385 		if (BGE_IS_5714_FAMILY(sc))
2386 			sc->bge_flags |= BGE_FLAG_MII_SERDES;
2387 		else
2388 			sc->bge_flags |= BGE_FLAG_TBI;
2389 	}
2390 
2391 	/* Setup MI MODE */
2392 	if (sc->bge_flags & BGE_FLAG_CPMU)
2393 		sc->bge_mi_mode = BGE_MIMODE_500KHZ_CONST;
2394 	else
2395 		sc->bge_mi_mode = BGE_MIMODE_BASE;
2396 	if (BGE_IS_5700_FAMILY(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5705) {
2397 		/* Enable auto polling for BCM570[0-5]. */
2398 		sc->bge_mi_mode |= BGE_MIMODE_AUTOPOLL;
2399 	}
2400 
2401 	/* Setup link status update stuffs */
2402 	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
2403 	    sc->bge_chipid != BGE_CHIPID_BCM5700_B2) {
2404 		sc->bge_link_upd = bge_bcm5700_link_upd;
2405 		sc->bge_link_chg = BGE_MACSTAT_MI_INTERRUPT;
2406 	} else if (sc->bge_flags & BGE_FLAG_TBI) {
2407 		sc->bge_link_upd = bge_tbi_link_upd;
2408 		sc->bge_link_chg = BGE_MACSTAT_LINK_CHANGED;
2409 	} else if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
2410 		sc->bge_link_upd = bge_autopoll_link_upd;
2411 		sc->bge_link_chg = BGE_MACSTAT_LINK_CHANGED;
2412 	} else {
2413 		sc->bge_link_upd = bge_copper_link_upd;
2414 		sc->bge_link_chg = BGE_MACSTAT_LINK_CHANGED;
2415 	}
2416 
2417 	/*
2418 	 * Broadcom's own driver always assumes the internal
2419 	 * PHY is at GMII address 1.  On some chips, the PHY responds
2420 	 * to accesses at all addresses, which could cause us to
2421 	 * bogusly attach the PHY 32 times at probe type.  Always
2422 	 * restricting the lookup to address 1 is simpler than
2423 	 * trying to figure out which chips revisions should be
2424 	 * special-cased.
2425 	 */
2426 	sc->bge_phyno = 1;
2427 
2428 	if (sc->bge_flags & BGE_FLAG_TBI) {
2429 		ifmedia_init(&sc->bge_ifmedia, IFM_IMASK,
2430 		    bge_ifmedia_upd, bge_ifmedia_sts);
2431 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER|IFM_1000_SX, 0, NULL);
2432 		ifmedia_add(&sc->bge_ifmedia,
2433 		    IFM_ETHER|IFM_1000_SX|IFM_FDX, 0, NULL);
2434 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER|IFM_AUTO, 0, NULL);
2435 		ifmedia_set(&sc->bge_ifmedia, IFM_ETHER|IFM_AUTO);
2436 		sc->bge_ifmedia.ifm_media = sc->bge_ifmedia.ifm_cur->ifm_media;
2437 	} else {
2438 		struct mii_probe_args mii_args;
2439 
2440 		mii_probe_args_init(&mii_args, bge_ifmedia_upd, bge_ifmedia_sts);
2441 		mii_args.mii_probemask = 1 << sc->bge_phyno;
2442 		mii_args.mii_capmask = capmask;
2443 		mii_args.mii_privtag = MII_PRIVTAG_BRGPHY;
2444 		mii_args.mii_priv = mii_priv;
2445 
2446 		error = mii_probe(dev, &sc->bge_miibus, &mii_args);
2447 		if (error) {
2448 			device_printf(dev, "MII without any PHY!\n");
2449 			goto fail;
2450 		}
2451 	}
2452 
2453 	/*
2454 	 * Create sysctl nodes.
2455 	 */
2456 	sysctl_ctx_init(&sc->bge_sysctl_ctx);
2457 	sc->bge_sysctl_tree = SYSCTL_ADD_NODE(&sc->bge_sysctl_ctx,
2458 					      SYSCTL_STATIC_CHILDREN(_hw),
2459 					      OID_AUTO,
2460 					      device_get_nameunit(dev),
2461 					      CTLFLAG_RD, 0, "");
2462 	if (sc->bge_sysctl_tree == NULL) {
2463 		device_printf(dev, "can't add sysctl node\n");
2464 		error = ENXIO;
2465 		goto fail;
2466 	}
2467 
2468 	SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2469 			SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2470 			OID_AUTO, "rx_coal_ticks",
2471 			CTLTYPE_INT | CTLFLAG_RW,
2472 			sc, 0, bge_sysctl_rx_coal_ticks, "I",
2473 			"Receive coalescing ticks (usec).");
2474 	SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2475 			SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2476 			OID_AUTO, "tx_coal_ticks",
2477 			CTLTYPE_INT | CTLFLAG_RW,
2478 			sc, 0, bge_sysctl_tx_coal_ticks, "I",
2479 			"Transmit coalescing ticks (usec).");
2480 	SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2481 			SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2482 			OID_AUTO, "rx_coal_bds",
2483 			CTLTYPE_INT | CTLFLAG_RW,
2484 			sc, 0, bge_sysctl_rx_coal_bds, "I",
2485 			"Receive max coalesced BD count.");
2486 	SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2487 			SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2488 			OID_AUTO, "tx_coal_bds",
2489 			CTLTYPE_INT | CTLFLAG_RW,
2490 			sc, 0, bge_sysctl_tx_coal_bds, "I",
2491 			"Transmit max coalesced BD count.");
2492 	if (sc->bge_flags & BGE_FLAG_PCIE) {
2493 		/*
2494 		 * A common design characteristic for many Broadcom
2495 		 * client controllers is that they only support a
2496 		 * single outstanding DMA read operation on the PCIe
2497 		 * bus. This means that it will take twice as long to
2498 		 * fetch a TX frame that is split into header and
2499 		 * payload buffers as it does to fetch a single,
2500 		 * contiguous TX frame (2 reads vs. 1 read). For these
2501 		 * controllers, coalescing buffers to reduce the number
2502 		 * of memory reads is effective way to get maximum
2503 		 * performance(about 940Mbps).  Without collapsing TX
2504 		 * buffers the maximum TCP bulk transfer performance
2505 		 * is about 850Mbps. However forcing coalescing mbufs
2506 		 * consumes a lot of CPU cycles, so leave it off by
2507 		 * default.
2508 		 */
2509 		SYSCTL_ADD_INT(&sc->bge_sysctl_ctx,
2510 			       SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2511 			       OID_AUTO, "force_defrag", CTLFLAG_RW,
2512 			       &sc->bge_force_defrag, 0,
2513 			       "Force defragment on TX path");
2514 	}
2515 	if (sc->bge_flags & BGE_FLAG_STATUS_TAG) {
2516 		if (!BGE_IS_5705_PLUS(sc)) {
2517 			SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2518 			    SYSCTL_CHILDREN(sc->bge_sysctl_tree), OID_AUTO,
2519 			    "rx_coal_ticks_int", CTLTYPE_INT | CTLFLAG_RW,
2520 			    sc, 0, bge_sysctl_rx_coal_ticks_int, "I",
2521 			    "Receive coalescing ticks "
2522 			    "during interrupt (usec).");
2523 			SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2524 			    SYSCTL_CHILDREN(sc->bge_sysctl_tree), OID_AUTO,
2525 			    "tx_coal_ticks_int", CTLTYPE_INT | CTLFLAG_RW,
2526 			    sc, 0, bge_sysctl_tx_coal_ticks_int, "I",
2527 			    "Transmit coalescing ticks "
2528 			    "during interrupt (usec).");
2529 		}
2530 		SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2531 		    SYSCTL_CHILDREN(sc->bge_sysctl_tree), OID_AUTO,
2532 		    "rx_coal_bds_int", CTLTYPE_INT | CTLFLAG_RW,
2533 		    sc, 0, bge_sysctl_rx_coal_bds_int, "I",
2534 		    "Receive max coalesced BD count during interrupt.");
2535 		SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2536 		    SYSCTL_CHILDREN(sc->bge_sysctl_tree), OID_AUTO,
2537 		    "tx_coal_bds_int", CTLTYPE_INT | CTLFLAG_RW,
2538 		    sc, 0, bge_sysctl_tx_coal_bds_int, "I",
2539 		    "Transmit max coalesced BD count during interrupt.");
2540 	}
2541 
2542 	/*
2543 	 * Call MI attach routine.
2544 	 */
2545 	ether_ifattach(ifp, ether_addr, NULL);
2546 
2547 	if (sc->bge_irq_type == PCI_INTR_TYPE_MSI) {
2548 		if (sc->bge_flags & BGE_FLAG_ONESHOT_MSI) {
2549 			intr_func = bge_msi_oneshot;
2550 			if (bootverbose)
2551 				device_printf(dev, "oneshot MSI\n");
2552 		} else {
2553 			intr_func = bge_msi;
2554 		}
2555 	} else if (sc->bge_flags & BGE_FLAG_STATUS_TAG) {
2556 		intr_func = bge_intr_legacy;
2557 	} else {
2558 		intr_func = bge_intr_crippled;
2559 	}
2560 	error = bus_setup_intr(dev, sc->bge_irq, INTR_MPSAFE, intr_func, sc,
2561 	    &sc->bge_intrhand, ifp->if_serializer);
2562 	if (error) {
2563 		ether_ifdetach(ifp);
2564 		device_printf(dev, "couldn't set up irq\n");
2565 		goto fail;
2566 	}
2567 
2568 	ifp->if_cpuid = rman_get_cpuid(sc->bge_irq);
2569 	KKASSERT(ifp->if_cpuid >= 0 && ifp->if_cpuid < ncpus);
2570 
2571 	return(0);
2572 fail:
2573 	bge_detach(dev);
2574 	return(error);
2575 }
2576 
2577 static int
2578 bge_detach(device_t dev)
2579 {
2580 	struct bge_softc *sc = device_get_softc(dev);
2581 
2582 	if (device_is_attached(dev)) {
2583 		struct ifnet *ifp = &sc->arpcom.ac_if;
2584 
2585 		lwkt_serialize_enter(ifp->if_serializer);
2586 		bge_stop(sc);
2587 		bge_reset(sc);
2588 		bus_teardown_intr(dev, sc->bge_irq, sc->bge_intrhand);
2589 		lwkt_serialize_exit(ifp->if_serializer);
2590 
2591 		ether_ifdetach(ifp);
2592 	}
2593 
2594 	if (sc->bge_flags & BGE_FLAG_TBI)
2595 		ifmedia_removeall(&sc->bge_ifmedia);
2596 	if (sc->bge_miibus)
2597 		device_delete_child(dev, sc->bge_miibus);
2598 	bus_generic_detach(dev);
2599 
2600 	if (sc->bge_irq != NULL) {
2601 		bus_release_resource(dev, SYS_RES_IRQ, sc->bge_irq_rid,
2602 		    sc->bge_irq);
2603 	}
2604 	if (sc->bge_irq_type == PCI_INTR_TYPE_MSI)
2605 		pci_release_msi(dev);
2606 
2607 	if (sc->bge_res != NULL) {
2608 		bus_release_resource(dev, SYS_RES_MEMORY,
2609 		    BGE_PCI_BAR0, sc->bge_res);
2610 	}
2611 
2612 	if (sc->bge_sysctl_tree != NULL)
2613 		sysctl_ctx_free(&sc->bge_sysctl_ctx);
2614 
2615 	bge_dma_free(sc);
2616 
2617 	return 0;
2618 }
2619 
2620 static void
2621 bge_reset(struct bge_softc *sc)
2622 {
2623 	device_t dev;
2624 	uint32_t cachesize, command, pcistate, reset;
2625 	void (*write_op)(struct bge_softc *, uint32_t, uint32_t);
2626 	int i, val = 0;
2627 
2628 	dev = sc->bge_dev;
2629 
2630 	if (BGE_IS_575X_PLUS(sc) && !BGE_IS_5714_FAMILY(sc) &&
2631 	    sc->bge_asicrev != BGE_ASICREV_BCM5906) {
2632 		if (sc->bge_flags & BGE_FLAG_PCIE)
2633 			write_op = bge_writemem_direct;
2634 		else
2635 			write_op = bge_writemem_ind;
2636 	} else {
2637 		write_op = bge_writereg_ind;
2638 	}
2639 
2640 	/* Save some important PCI state. */
2641 	cachesize = pci_read_config(dev, BGE_PCI_CACHESZ, 4);
2642 	command = pci_read_config(dev, BGE_PCI_CMD, 4);
2643 	pcistate = pci_read_config(dev, BGE_PCI_PCISTATE, 4);
2644 
2645 	pci_write_config(dev, BGE_PCI_MISC_CTL,
2646 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2647 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW|
2648 	    sc->bge_pci_miscctl, 4);
2649 
2650 	/* Disable fastboot on controllers that support it. */
2651 	if (sc->bge_asicrev == BGE_ASICREV_BCM5752 ||
2652 	    BGE_IS_5755_PLUS(sc)) {
2653 		if (bootverbose)
2654 			if_printf(&sc->arpcom.ac_if, "Disabling fastboot\n");
2655 		CSR_WRITE_4(sc, BGE_FASTBOOT_PC, 0x0);
2656 	}
2657 
2658 	/*
2659 	 * Write the magic number to SRAM at offset 0xB50.
2660 	 * When firmware finishes its initialization it will
2661 	 * write ~BGE_MAGIC_NUMBER to the same location.
2662 	 */
2663 	bge_writemem_ind(sc, BGE_SOFTWARE_GENCOMM, BGE_MAGIC_NUMBER);
2664 
2665 	reset = BGE_MISCCFG_RESET_CORE_CLOCKS|(65<<1);
2666 
2667 	/* XXX: Broadcom Linux driver. */
2668 	if (sc->bge_flags & BGE_FLAG_PCIE) {
2669 		/* Force PCI-E 1.0a mode */
2670 		if (sc->bge_asicrev != BGE_ASICREV_BCM5785 &&
2671 		    CSR_READ_4(sc, BGE_PCIE_PHY_TSTCTL) ==
2672 		    (BGE_PCIE_PHY_TSTCTL_PSCRAM |
2673 		     BGE_PCIE_PHY_TSTCTL_PCIE10)) {
2674 			CSR_WRITE_4(sc, BGE_PCIE_PHY_TSTCTL,
2675 			    BGE_PCIE_PHY_TSTCTL_PSCRAM);
2676 		}
2677 		if (sc->bge_chipid != BGE_CHIPID_BCM5750_A0) {
2678 			/* Prevent PCIE link training during global reset */
2679 			CSR_WRITE_4(sc, BGE_MISC_CFG, (1<<29));
2680 			reset |= (1<<29);
2681 		}
2682 	}
2683 
2684 	/*
2685 	 * Set GPHY Power Down Override to leave GPHY
2686 	 * powered up in D0 uninitialized.
2687 	 */
2688 	if (BGE_IS_5705_PLUS(sc) && (sc->bge_flags & BGE_FLAG_CPMU) == 0)
2689 		reset |= BGE_MISCCFG_GPHY_PD_OVERRIDE;
2690 
2691 	/* Issue global reset */
2692 	write_op(sc, BGE_MISC_CFG, reset);
2693 
2694 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
2695 		uint32_t status, ctrl;
2696 
2697 		status = CSR_READ_4(sc, BGE_VCPU_STATUS);
2698 		CSR_WRITE_4(sc, BGE_VCPU_STATUS,
2699 		    status | BGE_VCPU_STATUS_DRV_RESET);
2700 		ctrl = CSR_READ_4(sc, BGE_VCPU_EXT_CTRL);
2701 		CSR_WRITE_4(sc, BGE_VCPU_EXT_CTRL,
2702 		    ctrl & ~BGE_VCPU_EXT_CTRL_HALT_CPU);
2703 	}
2704 
2705 	DELAY(1000);
2706 
2707 	/* XXX: Broadcom Linux driver. */
2708 	if (sc->bge_flags & BGE_FLAG_PCIE) {
2709 		uint16_t devctl;
2710 
2711 		if (sc->bge_chipid == BGE_CHIPID_BCM5750_A0) {
2712 			uint32_t v;
2713 
2714 			DELAY(500000); /* wait for link training to complete */
2715 			v = pci_read_config(dev, 0xc4, 4);
2716 			pci_write_config(dev, 0xc4, v | (1<<15), 4);
2717 		}
2718 
2719 		devctl = pci_read_config(dev,
2720 		    sc->bge_pciecap + PCIER_DEVCTRL, 2);
2721 
2722 		/* Disable no snoop and disable relaxed ordering. */
2723 		devctl &= ~(PCIEM_DEVCTL_RELAX_ORDER | PCIEM_DEVCTL_NOSNOOP);
2724 
2725 		/* Old PCI-E chips only support 128 bytes Max PayLoad Size. */
2726 		if ((sc->bge_flags & BGE_FLAG_CPMU) == 0) {
2727 			devctl &= ~PCIEM_DEVCTL_MAX_PAYLOAD_MASK;
2728 			devctl |= PCIEM_DEVCTL_MAX_PAYLOAD_128;
2729 		}
2730 
2731 		pci_write_config(dev, sc->bge_pciecap + PCIER_DEVCTRL,
2732 		    devctl, 2);
2733 
2734 		/* Clear error status. */
2735 		pci_write_config(dev, sc->bge_pciecap + PCIER_DEVSTS,
2736 		    PCIEM_DEVSTS_CORR_ERR |
2737 		    PCIEM_DEVSTS_NFATAL_ERR |
2738 		    PCIEM_DEVSTS_FATAL_ERR |
2739 		    PCIEM_DEVSTS_UNSUPP_REQ, 2);
2740 	}
2741 
2742 	/* Reset some of the PCI state that got zapped by reset */
2743 	pci_write_config(dev, BGE_PCI_MISC_CTL,
2744 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2745 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW|
2746 	    sc->bge_pci_miscctl, 4);
2747 	pci_write_config(dev, BGE_PCI_CACHESZ, cachesize, 4);
2748 	pci_write_config(dev, BGE_PCI_CMD, command, 4);
2749 	write_op(sc, BGE_MISC_CFG, (65 << 1));
2750 
2751 	/*
2752 	 * Disable PCI-X relaxed ordering to ensure status block update
2753 	 * comes first then packet buffer DMA. Otherwise driver may
2754 	 * read stale status block.
2755 	 */
2756 	if (sc->bge_flags & BGE_FLAG_PCIX) {
2757 		uint16_t devctl;
2758 
2759 		devctl = pci_read_config(dev,
2760 		    sc->bge_pcixcap + PCIXR_COMMAND, 2);
2761 		devctl &= ~PCIXM_COMMAND_ERO;
2762 		if (sc->bge_asicrev == BGE_ASICREV_BCM5703) {
2763 			devctl &= ~PCIXM_COMMAND_MAX_READ;
2764 			devctl |= PCIXM_COMMAND_MAX_READ_2048;
2765 		} else if (sc->bge_asicrev == BGE_ASICREV_BCM5704) {
2766 			devctl &= ~(PCIXM_COMMAND_MAX_SPLITS |
2767 			    PCIXM_COMMAND_MAX_READ);
2768 			devctl |= PCIXM_COMMAND_MAX_READ_2048;
2769 		}
2770 		pci_write_config(dev, sc->bge_pcixcap + PCIXR_COMMAND,
2771 		    devctl, 2);
2772 	}
2773 
2774 	/*
2775 	 * Enable memory arbiter and re-enable MSI if necessary.
2776 	 */
2777 	if (BGE_IS_5714_FAMILY(sc)) {
2778 		uint32_t val;
2779 
2780 		if (sc->bge_irq_type == PCI_INTR_TYPE_MSI) {
2781 			/*
2782 			 * Resetting BCM5714 family will clear MSI
2783 			 * enable bit; restore it after resetting.
2784 			 */
2785 			PCI_SETBIT(sc->bge_dev, sc->bge_msicap + PCIR_MSI_CTRL,
2786 			    PCIM_MSICTRL_MSI_ENABLE, 2);
2787 			BGE_SETBIT(sc, BGE_MSI_MODE, BGE_MSIMODE_ENABLE);
2788 		}
2789 		val = CSR_READ_4(sc, BGE_MARB_MODE);
2790 		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE | val);
2791 	} else {
2792 		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
2793 	}
2794 
2795 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
2796 		for (i = 0; i < BGE_TIMEOUT; i++) {
2797 			val = CSR_READ_4(sc, BGE_VCPU_STATUS);
2798 			if (val & BGE_VCPU_STATUS_INIT_DONE)
2799 				break;
2800 			DELAY(100);
2801 		}
2802 		if (i == BGE_TIMEOUT) {
2803 			if_printf(&sc->arpcom.ac_if, "reset timed out\n");
2804 			return;
2805 		}
2806 	} else {
2807 		/*
2808 		 * Poll until we see the 1's complement of the magic number.
2809 		 * This indicates that the firmware initialization
2810 		 * is complete.
2811 		 */
2812 		for (i = 0; i < BGE_FIRMWARE_TIMEOUT; i++) {
2813 			val = bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM);
2814 			if (val == ~BGE_MAGIC_NUMBER)
2815 				break;
2816 			DELAY(10);
2817 		}
2818 		if (i == BGE_FIRMWARE_TIMEOUT) {
2819 			if_printf(&sc->arpcom.ac_if, "firmware handshake "
2820 				  "timed out, found 0x%08x\n", val);
2821 		}
2822 	}
2823 
2824 	/*
2825 	 * XXX Wait for the value of the PCISTATE register to
2826 	 * return to its original pre-reset state. This is a
2827 	 * fairly good indicator of reset completion. If we don't
2828 	 * wait for the reset to fully complete, trying to read
2829 	 * from the device's non-PCI registers may yield garbage
2830 	 * results.
2831 	 */
2832 	for (i = 0; i < BGE_TIMEOUT; i++) {
2833 		if (pci_read_config(dev, BGE_PCI_PCISTATE, 4) == pcistate)
2834 			break;
2835 		DELAY(10);
2836 	}
2837 
2838 	/* Fix up byte swapping */
2839 	CSR_WRITE_4(sc, BGE_MODE_CTL, BGE_DMA_SWAP_OPTIONS |
2840 	    BGE_MODECTL_BYTESWAP_DATA);
2841 
2842 	CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
2843 
2844 	/*
2845 	 * The 5704 in TBI mode apparently needs some special
2846 	 * adjustment to insure the SERDES drive level is set
2847 	 * to 1.2V.
2848 	 */
2849 	if (sc->bge_asicrev == BGE_ASICREV_BCM5704 &&
2850 	    (sc->bge_flags & BGE_FLAG_TBI)) {
2851 		uint32_t serdescfg;
2852 
2853 		serdescfg = CSR_READ_4(sc, BGE_SERDES_CFG);
2854 		serdescfg = (serdescfg & ~0xFFF) | 0x880;
2855 		CSR_WRITE_4(sc, BGE_SERDES_CFG, serdescfg);
2856 	}
2857 
2858 	CSR_WRITE_4(sc, BGE_MI_MODE,
2859 	    sc->bge_mi_mode & ~BGE_MIMODE_AUTOPOLL);
2860 	DELAY(80);
2861 
2862 	/* XXX: Broadcom Linux driver. */
2863 	if ((sc->bge_flags & BGE_FLAG_PCIE) &&
2864 	    sc->bge_chipid != BGE_CHIPID_BCM5750_A0 &&
2865 	    sc->bge_asicrev != BGE_ASICREV_BCM5785) {
2866 		uint32_t v;
2867 
2868 		/* Enable Data FIFO protection. */
2869 		v = CSR_READ_4(sc, BGE_PCIE_TLDLPL_PORT);
2870 		CSR_WRITE_4(sc, BGE_PCIE_TLDLPL_PORT, v | (1 << 25));
2871 	}
2872 
2873 	DELAY(10000);
2874 }
2875 
2876 /*
2877  * Frame reception handling. This is called if there's a frame
2878  * on the receive return list.
2879  *
2880  * Note: we have to be able to handle two possibilities here:
2881  * 1) the frame is from the jumbo recieve ring
2882  * 2) the frame is from the standard receive ring
2883  */
2884 
2885 static void
2886 bge_rxeof(struct bge_softc *sc, uint16_t rx_prod)
2887 {
2888 	struct ifnet *ifp;
2889 	int stdcnt = 0, jumbocnt = 0;
2890 
2891 	ifp = &sc->arpcom.ac_if;
2892 
2893 	while (sc->bge_rx_saved_considx != rx_prod) {
2894 		struct bge_rx_bd	*cur_rx;
2895 		uint32_t		rxidx;
2896 		struct mbuf		*m = NULL;
2897 		uint16_t		vlan_tag = 0;
2898 		int			have_tag = 0;
2899 
2900 		cur_rx =
2901 	    &sc->bge_ldata.bge_rx_return_ring[sc->bge_rx_saved_considx];
2902 
2903 		rxidx = cur_rx->bge_idx;
2904 		BGE_INC(sc->bge_rx_saved_considx, sc->bge_return_ring_cnt);
2905 		logif(rx_pkt);
2906 
2907 		if (cur_rx->bge_flags & BGE_RXBDFLAG_VLAN_TAG) {
2908 			have_tag = 1;
2909 			vlan_tag = cur_rx->bge_vlan_tag;
2910 		}
2911 
2912 		if (cur_rx->bge_flags & BGE_RXBDFLAG_JUMBO_RING) {
2913 			BGE_INC(sc->bge_jumbo, BGE_JUMBO_RX_RING_CNT);
2914 			jumbocnt++;
2915 
2916 			if (rxidx != sc->bge_jumbo) {
2917 				ifp->if_ierrors++;
2918 				if_printf(ifp, "sw jumbo index(%d) "
2919 				    "and hw jumbo index(%d) mismatch, drop!\n",
2920 				    sc->bge_jumbo, rxidx);
2921 				bge_setup_rxdesc_jumbo(sc, rxidx);
2922 				continue;
2923 			}
2924 
2925 			m = sc->bge_cdata.bge_rx_jumbo_chain[rxidx].bge_mbuf;
2926 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
2927 				ifp->if_ierrors++;
2928 				bge_setup_rxdesc_jumbo(sc, sc->bge_jumbo);
2929 				continue;
2930 			}
2931 			if (bge_newbuf_jumbo(sc, sc->bge_jumbo, 0)) {
2932 				ifp->if_ierrors++;
2933 				bge_setup_rxdesc_jumbo(sc, sc->bge_jumbo);
2934 				continue;
2935 			}
2936 		} else {
2937 			BGE_INC(sc->bge_std, BGE_STD_RX_RING_CNT);
2938 			stdcnt++;
2939 
2940 			if (rxidx != sc->bge_std) {
2941 				ifp->if_ierrors++;
2942 				if_printf(ifp, "sw std index(%d) "
2943 				    "and hw std index(%d) mismatch, drop!\n",
2944 				    sc->bge_std, rxidx);
2945 				bge_setup_rxdesc_std(sc, rxidx);
2946 				continue;
2947 			}
2948 
2949 			m = sc->bge_cdata.bge_rx_std_chain[rxidx].bge_mbuf;
2950 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
2951 				ifp->if_ierrors++;
2952 				bge_setup_rxdesc_std(sc, sc->bge_std);
2953 				continue;
2954 			}
2955 			if (bge_newbuf_std(sc, sc->bge_std, 0)) {
2956 				ifp->if_ierrors++;
2957 				bge_setup_rxdesc_std(sc, sc->bge_std);
2958 				continue;
2959 			}
2960 		}
2961 
2962 		ifp->if_ipackets++;
2963 #if !defined(__i386__) && !defined(__x86_64__)
2964 		/*
2965 		 * The x86 allows unaligned accesses, but for other
2966 		 * platforms we must make sure the payload is aligned.
2967 		 */
2968 		if (sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) {
2969 			bcopy(m->m_data, m->m_data + ETHER_ALIGN,
2970 			    cur_rx->bge_len);
2971 			m->m_data += ETHER_ALIGN;
2972 		}
2973 #endif
2974 		m->m_pkthdr.len = m->m_len = cur_rx->bge_len - ETHER_CRC_LEN;
2975 		m->m_pkthdr.rcvif = ifp;
2976 
2977 		if (ifp->if_capenable & IFCAP_RXCSUM) {
2978 			if (cur_rx->bge_flags & BGE_RXBDFLAG_IP_CSUM) {
2979 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2980 				if ((cur_rx->bge_ip_csum ^ 0xffff) == 0)
2981 					m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2982 			}
2983 			if ((cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM) &&
2984 			    m->m_pkthdr.len >= BGE_MIN_FRAMELEN) {
2985 				m->m_pkthdr.csum_data =
2986 					cur_rx->bge_tcp_udp_csum;
2987 				m->m_pkthdr.csum_flags |=
2988 					CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2989 			}
2990 		}
2991 
2992 		/*
2993 		 * If we received a packet with a vlan tag, pass it
2994 		 * to vlan_input() instead of ether_input().
2995 		 */
2996 		if (have_tag) {
2997 			m->m_flags |= M_VLANTAG;
2998 			m->m_pkthdr.ether_vlantag = vlan_tag;
2999 			have_tag = vlan_tag = 0;
3000 		}
3001 		ifp->if_input(ifp, m);
3002 	}
3003 
3004 	bge_writembx(sc, BGE_MBX_RX_CONS0_LO, sc->bge_rx_saved_considx);
3005 	if (stdcnt)
3006 		bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bge_std);
3007 	if (jumbocnt)
3008 		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
3009 }
3010 
3011 static void
3012 bge_txeof(struct bge_softc *sc, uint16_t tx_cons)
3013 {
3014 	struct ifnet *ifp;
3015 
3016 	ifp = &sc->arpcom.ac_if;
3017 
3018 	/*
3019 	 * Go through our tx ring and free mbufs for those
3020 	 * frames that have been sent.
3021 	 */
3022 	while (sc->bge_tx_saved_considx != tx_cons) {
3023 		uint32_t idx = 0;
3024 
3025 		idx = sc->bge_tx_saved_considx;
3026 		if (sc->bge_cdata.bge_tx_chain[idx] != NULL) {
3027 			ifp->if_opackets++;
3028 			bus_dmamap_unload(sc->bge_cdata.bge_tx_mtag,
3029 			    sc->bge_cdata.bge_tx_dmamap[idx]);
3030 			m_freem(sc->bge_cdata.bge_tx_chain[idx]);
3031 			sc->bge_cdata.bge_tx_chain[idx] = NULL;
3032 		}
3033 		sc->bge_txcnt--;
3034 		BGE_INC(sc->bge_tx_saved_considx, BGE_TX_RING_CNT);
3035 		logif(tx_pkt);
3036 	}
3037 
3038 	if ((BGE_TX_RING_CNT - sc->bge_txcnt) >=
3039 	    (sc->bge_txrsvd + sc->bge_txspare))
3040 		ifp->if_flags &= ~IFF_OACTIVE;
3041 
3042 	if (sc->bge_txcnt == 0)
3043 		ifp->if_timer = 0;
3044 
3045 	if (!ifq_is_empty(&ifp->if_snd))
3046 		if_devstart(ifp);
3047 }
3048 
3049 #ifdef DEVICE_POLLING
3050 
3051 static void
3052 bge_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
3053 {
3054 	struct bge_softc *sc = ifp->if_softc;
3055 	struct bge_status_block *sblk = sc->bge_ldata.bge_status_block;
3056 	uint16_t rx_prod, tx_cons;
3057 
3058 	switch(cmd) {
3059 	case POLL_REGISTER:
3060 		bge_disable_intr(sc);
3061 		break;
3062 	case POLL_DEREGISTER:
3063 		bge_enable_intr(sc);
3064 		break;
3065 	case POLL_AND_CHECK_STATUS:
3066 		/*
3067 		 * Process link state changes.
3068 		 */
3069 		bge_link_poll(sc);
3070 		/* Fall through */
3071 	case POLL_ONLY:
3072 		if (sc->bge_flags & BGE_FLAG_STATUS_TAG) {
3073 			sc->bge_status_tag = sblk->bge_status_tag;
3074 			/*
3075 			 * Use a load fence to ensure that status_tag
3076 			 * is saved  before rx_prod and tx_cons.
3077 			 */
3078 			cpu_lfence();
3079 		}
3080 		rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
3081 		tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
3082 		if (ifp->if_flags & IFF_RUNNING) {
3083 			rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
3084 			if (sc->bge_rx_saved_considx != rx_prod)
3085 				bge_rxeof(sc, rx_prod);
3086 
3087 			tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
3088 			if (sc->bge_tx_saved_considx != tx_cons)
3089 				bge_txeof(sc, tx_cons);
3090 		}
3091 		break;
3092 	}
3093 }
3094 
3095 #endif
3096 
3097 static void
3098 bge_intr_crippled(void *xsc)
3099 {
3100 	struct bge_softc *sc = xsc;
3101 	struct ifnet *ifp = &sc->arpcom.ac_if;
3102 
3103 	logif(intr);
3104 
3105  	/*
3106 	 * Ack the interrupt by writing something to BGE_MBX_IRQ0_LO.  Don't
3107 	 * disable interrupts by writing nonzero like we used to, since with
3108 	 * our current organization this just gives complications and
3109 	 * pessimizations for re-enabling interrupts.  We used to have races
3110 	 * instead of the necessary complications.  Disabling interrupts
3111 	 * would just reduce the chance of a status update while we are
3112 	 * running (by switching to the interrupt-mode coalescence
3113 	 * parameters), but this chance is already very low so it is more
3114 	 * efficient to get another interrupt than prevent it.
3115 	 *
3116 	 * We do the ack first to ensure another interrupt if there is a
3117 	 * status update after the ack.  We don't check for the status
3118 	 * changing later because it is more efficient to get another
3119 	 * interrupt than prevent it, not quite as above (not checking is
3120 	 * a smaller optimization than not toggling the interrupt enable,
3121 	 * since checking doesn't involve PCI accesses and toggling require
3122 	 * the status check).  So toggling would probably be a pessimization
3123 	 * even with MSI.  It would only be needed for using a task queue.
3124 	 */
3125 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 0);
3126 
3127 	/*
3128 	 * Process link state changes.
3129 	 */
3130 	bge_link_poll(sc);
3131 
3132 	if (ifp->if_flags & IFF_RUNNING) {
3133 		struct bge_status_block *sblk = sc->bge_ldata.bge_status_block;
3134 		uint16_t rx_prod, tx_cons;
3135 
3136 		rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
3137 		if (sc->bge_rx_saved_considx != rx_prod)
3138 			bge_rxeof(sc, rx_prod);
3139 
3140 		tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
3141 		if (sc->bge_tx_saved_considx != tx_cons)
3142 			bge_txeof(sc, tx_cons);
3143 	}
3144 
3145 	if (sc->bge_coal_chg)
3146 		bge_coal_change(sc);
3147 }
3148 
3149 static void
3150 bge_intr_legacy(void *xsc)
3151 {
3152 	struct bge_softc *sc = xsc;
3153 	struct bge_status_block *sblk = sc->bge_ldata.bge_status_block;
3154 
3155 	if (sc->bge_status_tag == sblk->bge_status_tag) {
3156 		uint32_t val;
3157 
3158 		val = pci_read_config(sc->bge_dev, BGE_PCI_PCISTATE, 4);
3159 		if (val & BGE_PCISTAT_INTR_NOTACT)
3160 			return;
3161 	}
3162 
3163 	/*
3164 	 * NOTE:
3165 	 * Interrupt will have to be disabled if tagged status
3166 	 * is used, else interrupt will always be asserted on
3167 	 * certain chips (at least on BCM5750 AX/BX).
3168 	 */
3169 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
3170 
3171 	bge_intr(sc);
3172 }
3173 
3174 static void
3175 bge_msi(void *xsc)
3176 {
3177 	struct bge_softc *sc = xsc;
3178 
3179 	/* Disable interrupt first */
3180 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
3181 	bge_intr(sc);
3182 }
3183 
3184 static void
3185 bge_msi_oneshot(void *xsc)
3186 {
3187 	bge_intr(xsc);
3188 }
3189 
3190 static void
3191 bge_intr(struct bge_softc *sc)
3192 {
3193 	struct ifnet *ifp = &sc->arpcom.ac_if;
3194 	struct bge_status_block *sblk = sc->bge_ldata.bge_status_block;
3195 	uint16_t rx_prod, tx_cons;
3196 	uint32_t status;
3197 
3198 	sc->bge_status_tag = sblk->bge_status_tag;
3199 	/*
3200 	 * Use a load fence to ensure that status_tag is saved
3201 	 * before rx_prod, tx_cons and status.
3202 	 */
3203 	cpu_lfence();
3204 
3205 	rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
3206 	tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
3207 	status = sblk->bge_status;
3208 
3209 	if ((status & BGE_STATFLAG_LINKSTATE_CHANGED) || sc->bge_link_evt)
3210 		bge_link_poll(sc);
3211 
3212 	if (ifp->if_flags & IFF_RUNNING) {
3213 		if (sc->bge_rx_saved_considx != rx_prod)
3214 			bge_rxeof(sc, rx_prod);
3215 
3216 		if (sc->bge_tx_saved_considx != tx_cons)
3217 			bge_txeof(sc, tx_cons);
3218 	}
3219 
3220 	bge_writembx(sc, BGE_MBX_IRQ0_LO, sc->bge_status_tag << 24);
3221 
3222 	if (sc->bge_coal_chg)
3223 		bge_coal_change(sc);
3224 }
3225 
3226 static void
3227 bge_tick(void *xsc)
3228 {
3229 	struct bge_softc *sc = xsc;
3230 	struct ifnet *ifp = &sc->arpcom.ac_if;
3231 
3232 	lwkt_serialize_enter(ifp->if_serializer);
3233 
3234 	if (BGE_IS_5705_PLUS(sc))
3235 		bge_stats_update_regs(sc);
3236 	else
3237 		bge_stats_update(sc);
3238 
3239 	if (sc->bge_flags & BGE_FLAG_TBI) {
3240 		/*
3241 		 * Since in TBI mode auto-polling can't be used we should poll
3242 		 * link status manually. Here we register pending link event
3243 		 * and trigger interrupt.
3244 		 */
3245 		sc->bge_link_evt++;
3246 		if (BGE_IS_CRIPPLED(sc))
3247 			BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
3248 		else
3249 			BGE_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
3250 	} else if (!sc->bge_link) {
3251 		mii_tick(device_get_softc(sc->bge_miibus));
3252 	}
3253 
3254 	callout_reset(&sc->bge_stat_timer, hz, bge_tick, sc);
3255 
3256 	lwkt_serialize_exit(ifp->if_serializer);
3257 }
3258 
3259 static void
3260 bge_stats_update_regs(struct bge_softc *sc)
3261 {
3262 	struct ifnet *ifp = &sc->arpcom.ac_if;
3263 	struct bge_mac_stats_regs stats;
3264 	uint32_t *s;
3265 	int i;
3266 
3267 	s = (uint32_t *)&stats;
3268 	for (i = 0; i < sizeof(struct bge_mac_stats_regs); i += 4) {
3269 		*s = CSR_READ_4(sc, BGE_RX_STATS + i);
3270 		s++;
3271 	}
3272 
3273 	ifp->if_collisions +=
3274 	   (stats.dot3StatsSingleCollisionFrames +
3275 	   stats.dot3StatsMultipleCollisionFrames +
3276 	   stats.dot3StatsExcessiveCollisions +
3277 	   stats.dot3StatsLateCollisions) -
3278 	   ifp->if_collisions;
3279 }
3280 
3281 static void
3282 bge_stats_update(struct bge_softc *sc)
3283 {
3284 	struct ifnet *ifp = &sc->arpcom.ac_if;
3285 	bus_size_t stats;
3286 
3287 	stats = BGE_MEMWIN_START + BGE_STATS_BLOCK;
3288 
3289 #define READ_STAT(sc, stats, stat)	\
3290 	CSR_READ_4(sc, stats + offsetof(struct bge_stats, stat))
3291 
3292 	ifp->if_collisions +=
3293 	   (READ_STAT(sc, stats,
3294 		txstats.dot3StatsSingleCollisionFrames.bge_addr_lo) +
3295 	    READ_STAT(sc, stats,
3296 		txstats.dot3StatsMultipleCollisionFrames.bge_addr_lo) +
3297 	    READ_STAT(sc, stats,
3298 		txstats.dot3StatsExcessiveCollisions.bge_addr_lo) +
3299 	    READ_STAT(sc, stats,
3300 		txstats.dot3StatsLateCollisions.bge_addr_lo)) -
3301 	   ifp->if_collisions;
3302 
3303 #undef READ_STAT
3304 
3305 #ifdef notdef
3306 	ifp->if_collisions +=
3307 	   (sc->bge_rdata->bge_info.bge_stats.dot3StatsSingleCollisionFrames +
3308 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsMultipleCollisionFrames +
3309 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsExcessiveCollisions +
3310 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsLateCollisions) -
3311 	   ifp->if_collisions;
3312 #endif
3313 }
3314 
3315 /*
3316  * Encapsulate an mbuf chain in the tx ring  by coupling the mbuf data
3317  * pointers to descriptors.
3318  */
3319 static int
3320 bge_encap(struct bge_softc *sc, struct mbuf **m_head0, uint32_t *txidx)
3321 {
3322 	struct bge_tx_bd *d = NULL, *last_d;
3323 	uint16_t csum_flags = 0, mss = 0;
3324 	bus_dma_segment_t segs[BGE_NSEG_NEW];
3325 	bus_dmamap_t map;
3326 	int error, maxsegs, nsegs, idx, i;
3327 	struct mbuf *m_head = *m_head0, *m_new;
3328 
3329 	if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
3330 		error = bge_setup_tso(sc, m_head0, &mss, &csum_flags);
3331 		if (error)
3332 			return ENOBUFS;
3333 		m_head = *m_head0;
3334 	} else if (m_head->m_pkthdr.csum_flags & BGE_CSUM_FEATURES) {
3335 		if (m_head->m_pkthdr.csum_flags & CSUM_IP)
3336 			csum_flags |= BGE_TXBDFLAG_IP_CSUM;
3337 		if (m_head->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP))
3338 			csum_flags |= BGE_TXBDFLAG_TCP_UDP_CSUM;
3339 		if (m_head->m_flags & M_LASTFRAG)
3340 			csum_flags |= BGE_TXBDFLAG_IP_FRAG_END;
3341 		else if (m_head->m_flags & M_FRAG)
3342 			csum_flags |= BGE_TXBDFLAG_IP_FRAG;
3343 	}
3344 
3345 	idx = *txidx;
3346 	map = sc->bge_cdata.bge_tx_dmamap[idx];
3347 
3348 	maxsegs = (BGE_TX_RING_CNT - sc->bge_txcnt) - sc->bge_txrsvd;
3349 	KASSERT(maxsegs >= sc->bge_txspare,
3350 		("not enough segments %d", maxsegs));
3351 
3352 	if (maxsegs > BGE_NSEG_NEW)
3353 		maxsegs = BGE_NSEG_NEW;
3354 
3355 	/*
3356 	 * Pad outbound frame to BGE_MIN_FRAMELEN for an unusual reason.
3357 	 * The bge hardware will pad out Tx runts to BGE_MIN_FRAMELEN,
3358 	 * but when such padded frames employ the bge IP/TCP checksum
3359 	 * offload, the hardware checksum assist gives incorrect results
3360 	 * (possibly from incorporating its own padding into the UDP/TCP
3361 	 * checksum; who knows).  If we pad such runts with zeros, the
3362 	 * onboard checksum comes out correct.
3363 	 */
3364 	if ((csum_flags & BGE_TXBDFLAG_TCP_UDP_CSUM) &&
3365 	    m_head->m_pkthdr.len < BGE_MIN_FRAMELEN) {
3366 		error = m_devpad(m_head, BGE_MIN_FRAMELEN);
3367 		if (error)
3368 			goto back;
3369 	}
3370 
3371 	if ((sc->bge_flags & BGE_FLAG_SHORTDMA) && m_head->m_next != NULL) {
3372 		m_new = bge_defrag_shortdma(m_head);
3373 		if (m_new == NULL) {
3374 			error = ENOBUFS;
3375 			goto back;
3376 		}
3377 		*m_head0 = m_head = m_new;
3378 	}
3379 	if ((m_head->m_pkthdr.csum_flags & CSUM_TSO) == 0 &&
3380 	    sc->bge_force_defrag && (sc->bge_flags & BGE_FLAG_PCIE) &&
3381 	    m_head->m_next != NULL) {
3382 		/*
3383 		 * Forcefully defragment mbuf chain to overcome hardware
3384 		 * limitation which only support a single outstanding
3385 		 * DMA read operation.  If it fails, keep moving on using
3386 		 * the original mbuf chain.
3387 		 */
3388 		m_new = m_defrag(m_head, MB_DONTWAIT);
3389 		if (m_new != NULL)
3390 			*m_head0 = m_head = m_new;
3391 	}
3392 
3393 	error = bus_dmamap_load_mbuf_defrag(sc->bge_cdata.bge_tx_mtag, map,
3394 			m_head0, segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
3395 	if (error)
3396 		goto back;
3397 
3398 	m_head = *m_head0;
3399 	bus_dmamap_sync(sc->bge_cdata.bge_tx_mtag, map, BUS_DMASYNC_PREWRITE);
3400 
3401 	for (i = 0; ; i++) {
3402 		d = &sc->bge_ldata.bge_tx_ring[idx];
3403 
3404 		d->bge_addr.bge_addr_lo = BGE_ADDR_LO(segs[i].ds_addr);
3405 		d->bge_addr.bge_addr_hi = BGE_ADDR_HI(segs[i].ds_addr);
3406 		d->bge_len = segs[i].ds_len;
3407 		d->bge_flags = csum_flags;
3408 		d->bge_mss = mss;
3409 
3410 		if (i == nsegs - 1)
3411 			break;
3412 		BGE_INC(idx, BGE_TX_RING_CNT);
3413 	}
3414 	last_d = d;
3415 
3416 	/* Set vlan tag to the first segment of the packet. */
3417 	d = &sc->bge_ldata.bge_tx_ring[*txidx];
3418 	if (m_head->m_flags & M_VLANTAG) {
3419 		d->bge_flags |= BGE_TXBDFLAG_VLAN_TAG;
3420 		d->bge_vlan_tag = m_head->m_pkthdr.ether_vlantag;
3421 	} else {
3422 		d->bge_vlan_tag = 0;
3423 	}
3424 
3425 	/* Mark the last segment as end of packet... */
3426 	last_d->bge_flags |= BGE_TXBDFLAG_END;
3427 
3428 	/*
3429 	 * Insure that the map for this transmission is placed at
3430 	 * the array index of the last descriptor in this chain.
3431 	 */
3432 	sc->bge_cdata.bge_tx_dmamap[*txidx] = sc->bge_cdata.bge_tx_dmamap[idx];
3433 	sc->bge_cdata.bge_tx_dmamap[idx] = map;
3434 	sc->bge_cdata.bge_tx_chain[idx] = m_head;
3435 	sc->bge_txcnt += nsegs;
3436 
3437 	BGE_INC(idx, BGE_TX_RING_CNT);
3438 	*txidx = idx;
3439 back:
3440 	if (error) {
3441 		m_freem(*m_head0);
3442 		*m_head0 = NULL;
3443 	}
3444 	return error;
3445 }
3446 
3447 /*
3448  * Main transmit routine. To avoid having to do mbuf copies, we put pointers
3449  * to the mbuf data regions directly in the transmit descriptors.
3450  */
3451 static void
3452 bge_start(struct ifnet *ifp)
3453 {
3454 	struct bge_softc *sc = ifp->if_softc;
3455 	struct mbuf *m_head = NULL;
3456 	uint32_t prodidx;
3457 	int need_trans;
3458 
3459 	if ((ifp->if_flags & (IFF_RUNNING | IFF_OACTIVE)) != IFF_RUNNING)
3460 		return;
3461 
3462 	prodidx = sc->bge_tx_prodidx;
3463 
3464 	need_trans = 0;
3465 	while (sc->bge_cdata.bge_tx_chain[prodidx] == NULL) {
3466 		m_head = ifq_dequeue(&ifp->if_snd, NULL);
3467 		if (m_head == NULL)
3468 			break;
3469 
3470 		/*
3471 		 * XXX
3472 		 * The code inside the if() block is never reached since we
3473 		 * must mark CSUM_IP_FRAGS in our if_hwassist to start getting
3474 		 * requests to checksum TCP/UDP in a fragmented packet.
3475 		 *
3476 		 * XXX
3477 		 * safety overkill.  If this is a fragmented packet chain
3478 		 * with delayed TCP/UDP checksums, then only encapsulate
3479 		 * it if we have enough descriptors to handle the entire
3480 		 * chain at once.
3481 		 * (paranoia -- may not actually be needed)
3482 		 */
3483 		if ((m_head->m_flags & M_FIRSTFRAG) &&
3484 		    (m_head->m_pkthdr.csum_flags & CSUM_DELAY_DATA)) {
3485 			if ((BGE_TX_RING_CNT - sc->bge_txcnt) <
3486 			    m_head->m_pkthdr.csum_data + sc->bge_txrsvd) {
3487 				ifp->if_flags |= IFF_OACTIVE;
3488 				ifq_prepend(&ifp->if_snd, m_head);
3489 				break;
3490 			}
3491 		}
3492 
3493 		/*
3494 		 * Sanity check: avoid coming within bge_txrsvd
3495 		 * descriptors of the end of the ring.  Also make
3496 		 * sure there are bge_txspare descriptors for
3497 		 * jumbo buffers' defragmentation.
3498 		 */
3499 		if ((BGE_TX_RING_CNT - sc->bge_txcnt) <
3500 		    (sc->bge_txrsvd + sc->bge_txspare)) {
3501 			ifp->if_flags |= IFF_OACTIVE;
3502 			ifq_prepend(&ifp->if_snd, m_head);
3503 			break;
3504 		}
3505 
3506 		/*
3507 		 * Pack the data into the transmit ring. If we
3508 		 * don't have room, set the OACTIVE flag and wait
3509 		 * for the NIC to drain the ring.
3510 		 */
3511 		if (bge_encap(sc, &m_head, &prodidx)) {
3512 			ifp->if_flags |= IFF_OACTIVE;
3513 			ifp->if_oerrors++;
3514 			break;
3515 		}
3516 		need_trans = 1;
3517 
3518 		ETHER_BPF_MTAP(ifp, m_head);
3519 	}
3520 
3521 	if (!need_trans)
3522 		return;
3523 
3524 	/* Transmit */
3525 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
3526 	/* 5700 b2 errata */
3527 	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
3528 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
3529 
3530 	sc->bge_tx_prodidx = prodidx;
3531 
3532 	/*
3533 	 * Set a timeout in case the chip goes out to lunch.
3534 	 */
3535 	ifp->if_timer = 5;
3536 }
3537 
3538 static void
3539 bge_init(void *xsc)
3540 {
3541 	struct bge_softc *sc = xsc;
3542 	struct ifnet *ifp = &sc->arpcom.ac_if;
3543 	uint16_t *m;
3544 	uint32_t mode;
3545 
3546 	ASSERT_SERIALIZED(ifp->if_serializer);
3547 
3548 	/* Cancel pending I/O and flush buffers. */
3549 	bge_stop(sc);
3550 	bge_reset(sc);
3551 	bge_chipinit(sc);
3552 
3553 	/*
3554 	 * Init the various state machines, ring
3555 	 * control blocks and firmware.
3556 	 */
3557 	if (bge_blockinit(sc)) {
3558 		if_printf(ifp, "initialization failure\n");
3559 		bge_stop(sc);
3560 		return;
3561 	}
3562 
3563 	/* Specify MTU. */
3564 	CSR_WRITE_4(sc, BGE_RX_MTU, ifp->if_mtu +
3565 	    ETHER_HDR_LEN + ETHER_CRC_LEN + EVL_ENCAPLEN);
3566 
3567 	/* Load our MAC address. */
3568 	m = (uint16_t *)&sc->arpcom.ac_enaddr[0];
3569 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_LO, htons(m[0]));
3570 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_HI, (htons(m[1]) << 16) | htons(m[2]));
3571 
3572 	/* Enable or disable promiscuous mode as needed. */
3573 	bge_setpromisc(sc);
3574 
3575 	/* Program multicast filter. */
3576 	bge_setmulti(sc);
3577 
3578 	/* Init RX ring. */
3579 	if (bge_init_rx_ring_std(sc)) {
3580 		if_printf(ifp, "RX ring initialization failed\n");
3581 		bge_stop(sc);
3582 		return;
3583 	}
3584 
3585 	/*
3586 	 * Workaround for a bug in 5705 ASIC rev A0. Poll the NIC's
3587 	 * memory to insure that the chip has in fact read the first
3588 	 * entry of the ring.
3589 	 */
3590 	if (sc->bge_chipid == BGE_CHIPID_BCM5705_A0) {
3591 		uint32_t		v, i;
3592 		for (i = 0; i < 10; i++) {
3593 			DELAY(20);
3594 			v = bge_readmem_ind(sc, BGE_STD_RX_RINGS + 8);
3595 			if (v == (MCLBYTES - ETHER_ALIGN))
3596 				break;
3597 		}
3598 		if (i == 10)
3599 			if_printf(ifp, "5705 A0 chip failed to load RX ring\n");
3600 	}
3601 
3602 	/* Init jumbo RX ring. */
3603 	if (ifp->if_mtu > (ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN)) {
3604 		if (bge_init_rx_ring_jumbo(sc)) {
3605 			if_printf(ifp, "Jumbo RX ring initialization failed\n");
3606 			bge_stop(sc);
3607 			return;
3608 		}
3609 	}
3610 
3611 	/* Init our RX return ring index */
3612 	sc->bge_rx_saved_considx = 0;
3613 
3614 	/* Init TX ring. */
3615 	bge_init_tx_ring(sc);
3616 
3617 	/* Enable TX MAC state machine lockup fix. */
3618 	mode = CSR_READ_4(sc, BGE_TX_MODE);
3619 	if (BGE_IS_5755_PLUS(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5906)
3620 		mode |= BGE_TXMODE_MBUF_LOCKUP_FIX;
3621 	/* Turn on transmitter */
3622 	CSR_WRITE_4(sc, BGE_TX_MODE, mode | BGE_TXMODE_ENABLE);
3623 
3624 	/* Turn on receiver */
3625 	BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
3626 
3627 	/*
3628 	 * Set the number of good frames to receive after RX MBUF
3629 	 * Low Watermark has been reached.  After the RX MAC receives
3630 	 * this number of frames, it will drop subsequent incoming
3631 	 * frames until the MBUF High Watermark is reached.
3632 	 */
3633 	CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 2);
3634 
3635 	if (sc->bge_irq_type == PCI_INTR_TYPE_MSI) {
3636 		if (bootverbose) {
3637 			if_printf(ifp, "MSI_MODE: %#x\n",
3638 			    CSR_READ_4(sc, BGE_MSI_MODE));
3639 		}
3640 
3641 		/*
3642 		 * XXX
3643 		 * Linux driver turns it on for all chips supporting MSI?!
3644 		 */
3645 		if (sc->bge_flags & BGE_FLAG_ONESHOT_MSI) {
3646 			/*
3647 			 * XXX
3648 			 * According to 5722-PG101-R,
3649 			 * BGE_PCIE_TRANSACT_ONESHOT_MSI applies only to
3650 			 * BCM5906.
3651 			 */
3652 			BGE_SETBIT(sc, BGE_PCIE_TRANSACT,
3653 			    BGE_PCIE_TRANSACT_ONESHOT_MSI);
3654 		}
3655 	}
3656 
3657 	/* Tell firmware we're alive. */
3658 	BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3659 
3660 	/* Enable host interrupts if polling(4) is not enabled. */
3661 	PCI_SETBIT(sc->bge_dev, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_CLEAR_INTA, 4);
3662 #ifdef DEVICE_POLLING
3663 	if (ifp->if_flags & IFF_POLLING)
3664 		bge_disable_intr(sc);
3665 	else
3666 #endif
3667 	bge_enable_intr(sc);
3668 
3669 	bge_ifmedia_upd(ifp);
3670 
3671 	ifp->if_flags |= IFF_RUNNING;
3672 	ifp->if_flags &= ~IFF_OACTIVE;
3673 
3674 	callout_reset(&sc->bge_stat_timer, hz, bge_tick, sc);
3675 }
3676 
3677 /*
3678  * Set media options.
3679  */
3680 static int
3681 bge_ifmedia_upd(struct ifnet *ifp)
3682 {
3683 	struct bge_softc *sc = ifp->if_softc;
3684 
3685 	/* If this is a 1000baseX NIC, enable the TBI port. */
3686 	if (sc->bge_flags & BGE_FLAG_TBI) {
3687 		struct ifmedia *ifm = &sc->bge_ifmedia;
3688 
3689 		if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
3690 			return(EINVAL);
3691 
3692 		switch(IFM_SUBTYPE(ifm->ifm_media)) {
3693 		case IFM_AUTO:
3694 			/*
3695 			 * The BCM5704 ASIC appears to have a special
3696 			 * mechanism for programming the autoneg
3697 			 * advertisement registers in TBI mode.
3698 			 */
3699 			if (!bge_fake_autoneg &&
3700 			    sc->bge_asicrev == BGE_ASICREV_BCM5704) {
3701 				uint32_t sgdig;
3702 
3703 				CSR_WRITE_4(sc, BGE_TX_TBI_AUTONEG, 0);
3704 				sgdig = CSR_READ_4(sc, BGE_SGDIG_CFG);
3705 				sgdig |= BGE_SGDIGCFG_AUTO |
3706 					 BGE_SGDIGCFG_PAUSE_CAP |
3707 					 BGE_SGDIGCFG_ASYM_PAUSE;
3708 				CSR_WRITE_4(sc, BGE_SGDIG_CFG,
3709 					    sgdig | BGE_SGDIGCFG_SEND);
3710 				DELAY(5);
3711 				CSR_WRITE_4(sc, BGE_SGDIG_CFG, sgdig);
3712 			}
3713 			break;
3714 		case IFM_1000_SX:
3715 			if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX) {
3716 				BGE_CLRBIT(sc, BGE_MAC_MODE,
3717 				    BGE_MACMODE_HALF_DUPLEX);
3718 			} else {
3719 				BGE_SETBIT(sc, BGE_MAC_MODE,
3720 				    BGE_MACMODE_HALF_DUPLEX);
3721 			}
3722 			break;
3723 		default:
3724 			return(EINVAL);
3725 		}
3726 	} else {
3727 		struct mii_data *mii = device_get_softc(sc->bge_miibus);
3728 
3729 		sc->bge_link_evt++;
3730 		sc->bge_link = 0;
3731 		if (mii->mii_instance) {
3732 			struct mii_softc *miisc;
3733 
3734 			LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
3735 				mii_phy_reset(miisc);
3736 		}
3737 		mii_mediachg(mii);
3738 
3739 		/*
3740 		 * Force an interrupt so that we will call bge_link_upd
3741 		 * if needed and clear any pending link state attention.
3742 		 * Without this we are not getting any further interrupts
3743 		 * for link state changes and thus will not UP the link and
3744 		 * not be able to send in bge_start.  The only way to get
3745 		 * things working was to receive a packet and get an RX
3746 		 * intr.
3747 		 *
3748 		 * bge_tick should help for fiber cards and we might not
3749 		 * need to do this here if BGE_FLAG_TBI is set but as
3750 		 * we poll for fiber anyway it should not harm.
3751 		 */
3752 		if (BGE_IS_CRIPPLED(sc))
3753 			BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
3754 		else
3755 			BGE_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
3756 	}
3757 	return(0);
3758 }
3759 
3760 /*
3761  * Report current media status.
3762  */
3763 static void
3764 bge_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3765 {
3766 	struct bge_softc *sc = ifp->if_softc;
3767 
3768 	if (sc->bge_flags & BGE_FLAG_TBI) {
3769 		ifmr->ifm_status = IFM_AVALID;
3770 		ifmr->ifm_active = IFM_ETHER;
3771 		if (CSR_READ_4(sc, BGE_MAC_STS) &
3772 		    BGE_MACSTAT_TBI_PCS_SYNCHED) {
3773 			ifmr->ifm_status |= IFM_ACTIVE;
3774 		} else {
3775 			ifmr->ifm_active |= IFM_NONE;
3776 			return;
3777 		}
3778 
3779 		ifmr->ifm_active |= IFM_1000_SX;
3780 		if (CSR_READ_4(sc, BGE_MAC_MODE) & BGE_MACMODE_HALF_DUPLEX)
3781 			ifmr->ifm_active |= IFM_HDX;
3782 		else
3783 			ifmr->ifm_active |= IFM_FDX;
3784 	} else {
3785 		struct mii_data *mii = device_get_softc(sc->bge_miibus);
3786 
3787 		mii_pollstat(mii);
3788 		ifmr->ifm_active = mii->mii_media_active;
3789 		ifmr->ifm_status = mii->mii_media_status;
3790 	}
3791 }
3792 
3793 static int
3794 bge_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
3795 {
3796 	struct bge_softc *sc = ifp->if_softc;
3797 	struct ifreq *ifr = (struct ifreq *)data;
3798 	int mask, error = 0;
3799 
3800 	ASSERT_SERIALIZED(ifp->if_serializer);
3801 
3802 	switch (command) {
3803 	case SIOCSIFMTU:
3804 		if ((!BGE_IS_JUMBO_CAPABLE(sc) && ifr->ifr_mtu > ETHERMTU) ||
3805 		    (BGE_IS_JUMBO_CAPABLE(sc) &&
3806 		     ifr->ifr_mtu > BGE_JUMBO_MTU)) {
3807 			error = EINVAL;
3808 		} else if (ifp->if_mtu != ifr->ifr_mtu) {
3809 			ifp->if_mtu = ifr->ifr_mtu;
3810 			if (ifp->if_flags & IFF_RUNNING)
3811 				bge_init(sc);
3812 		}
3813 		break;
3814 	case SIOCSIFFLAGS:
3815 		if (ifp->if_flags & IFF_UP) {
3816 			if (ifp->if_flags & IFF_RUNNING) {
3817 				mask = ifp->if_flags ^ sc->bge_if_flags;
3818 
3819 				/*
3820 				 * If only the state of the PROMISC flag
3821 				 * changed, then just use the 'set promisc
3822 				 * mode' command instead of reinitializing
3823 				 * the entire NIC. Doing a full re-init
3824 				 * means reloading the firmware and waiting
3825 				 * for it to start up, which may take a
3826 				 * second or two.  Similarly for ALLMULTI.
3827 				 */
3828 				if (mask & IFF_PROMISC)
3829 					bge_setpromisc(sc);
3830 				if (mask & IFF_ALLMULTI)
3831 					bge_setmulti(sc);
3832 			} else {
3833 				bge_init(sc);
3834 			}
3835 		} else if (ifp->if_flags & IFF_RUNNING) {
3836 			bge_stop(sc);
3837 		}
3838 		sc->bge_if_flags = ifp->if_flags;
3839 		break;
3840 	case SIOCADDMULTI:
3841 	case SIOCDELMULTI:
3842 		if (ifp->if_flags & IFF_RUNNING)
3843 			bge_setmulti(sc);
3844 		break;
3845 	case SIOCSIFMEDIA:
3846 	case SIOCGIFMEDIA:
3847 		if (sc->bge_flags & BGE_FLAG_TBI) {
3848 			error = ifmedia_ioctl(ifp, ifr,
3849 			    &sc->bge_ifmedia, command);
3850 		} else {
3851 			struct mii_data *mii;
3852 
3853 			mii = device_get_softc(sc->bge_miibus);
3854 			error = ifmedia_ioctl(ifp, ifr,
3855 					      &mii->mii_media, command);
3856 		}
3857 		break;
3858         case SIOCSIFCAP:
3859 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3860 		if (mask & IFCAP_HWCSUM) {
3861 			ifp->if_capenable ^= (mask & IFCAP_HWCSUM);
3862 			if (ifp->if_capenable & IFCAP_TXCSUM)
3863 				ifp->if_hwassist |= BGE_CSUM_FEATURES;
3864 			else
3865 				ifp->if_hwassist &= ~BGE_CSUM_FEATURES;
3866 		}
3867 		if (mask & IFCAP_TSO) {
3868 			ifp->if_capenable ^= IFCAP_TSO;
3869 			if (ifp->if_capenable & IFCAP_TSO)
3870 				ifp->if_hwassist |= CSUM_TSO;
3871 			else
3872 				ifp->if_hwassist &= ~CSUM_TSO;
3873 		}
3874 		break;
3875 	default:
3876 		error = ether_ioctl(ifp, command, data);
3877 		break;
3878 	}
3879 	return error;
3880 }
3881 
3882 static void
3883 bge_watchdog(struct ifnet *ifp)
3884 {
3885 	struct bge_softc *sc = ifp->if_softc;
3886 
3887 	if_printf(ifp, "watchdog timeout -- resetting\n");
3888 
3889 	bge_init(sc);
3890 
3891 	ifp->if_oerrors++;
3892 
3893 	if (!ifq_is_empty(&ifp->if_snd))
3894 		if_devstart(ifp);
3895 }
3896 
3897 /*
3898  * Stop the adapter and free any mbufs allocated to the
3899  * RX and TX lists.
3900  */
3901 static void
3902 bge_stop(struct bge_softc *sc)
3903 {
3904 	struct ifnet *ifp = &sc->arpcom.ac_if;
3905 
3906 	ASSERT_SERIALIZED(ifp->if_serializer);
3907 
3908 	callout_stop(&sc->bge_stat_timer);
3909 
3910 	/*
3911 	 * Disable all of the receiver blocks
3912 	 */
3913 	bge_stop_block(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
3914 	bge_stop_block(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
3915 	bge_stop_block(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
3916 	if (BGE_IS_5700_FAMILY(sc))
3917 		bge_stop_block(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
3918 	bge_stop_block(sc, BGE_RDBDI_MODE, BGE_RBDIMODE_ENABLE);
3919 	bge_stop_block(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
3920 	bge_stop_block(sc, BGE_RBDC_MODE, BGE_RBDCMODE_ENABLE);
3921 
3922 	/*
3923 	 * Disable all of the transmit blocks
3924 	 */
3925 	bge_stop_block(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
3926 	bge_stop_block(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
3927 	bge_stop_block(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
3928 	bge_stop_block(sc, BGE_RDMA_MODE, BGE_RDMAMODE_ENABLE);
3929 	bge_stop_block(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE);
3930 	if (BGE_IS_5700_FAMILY(sc))
3931 		bge_stop_block(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
3932 	bge_stop_block(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
3933 
3934 	/*
3935 	 * Shut down all of the memory managers and related
3936 	 * state machines.
3937 	 */
3938 	bge_stop_block(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE);
3939 	bge_stop_block(sc, BGE_WDMA_MODE, BGE_WDMAMODE_ENABLE);
3940 	if (BGE_IS_5700_FAMILY(sc))
3941 		bge_stop_block(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
3942 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
3943 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
3944 	if (!BGE_IS_5705_PLUS(sc)) {
3945 		BGE_CLRBIT(sc, BGE_BMAN_MODE, BGE_BMANMODE_ENABLE);
3946 		BGE_CLRBIT(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
3947 	}
3948 
3949 	/* Disable host interrupts. */
3950 	bge_disable_intr(sc);
3951 
3952 	/*
3953 	 * Tell firmware we're shutting down.
3954 	 */
3955 	BGE_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3956 
3957 	/* Free the RX lists. */
3958 	bge_free_rx_ring_std(sc);
3959 
3960 	/* Free jumbo RX list. */
3961 	if (BGE_IS_JUMBO_CAPABLE(sc))
3962 		bge_free_rx_ring_jumbo(sc);
3963 
3964 	/* Free TX buffers. */
3965 	bge_free_tx_ring(sc);
3966 
3967 	sc->bge_status_tag = 0;
3968 	sc->bge_link = 0;
3969 	sc->bge_coal_chg = 0;
3970 
3971 	sc->bge_tx_saved_considx = BGE_TXCONS_UNSET;
3972 
3973 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
3974 	ifp->if_timer = 0;
3975 }
3976 
3977 /*
3978  * Stop all chip I/O so that the kernel's probe routines don't
3979  * get confused by errant DMAs when rebooting.
3980  */
3981 static void
3982 bge_shutdown(device_t dev)
3983 {
3984 	struct bge_softc *sc = device_get_softc(dev);
3985 	struct ifnet *ifp = &sc->arpcom.ac_if;
3986 
3987 	lwkt_serialize_enter(ifp->if_serializer);
3988 	bge_stop(sc);
3989 	bge_reset(sc);
3990 	lwkt_serialize_exit(ifp->if_serializer);
3991 }
3992 
3993 static int
3994 bge_suspend(device_t dev)
3995 {
3996 	struct bge_softc *sc = device_get_softc(dev);
3997 	struct ifnet *ifp = &sc->arpcom.ac_if;
3998 
3999 	lwkt_serialize_enter(ifp->if_serializer);
4000 	bge_stop(sc);
4001 	lwkt_serialize_exit(ifp->if_serializer);
4002 
4003 	return 0;
4004 }
4005 
4006 static int
4007 bge_resume(device_t dev)
4008 {
4009 	struct bge_softc *sc = device_get_softc(dev);
4010 	struct ifnet *ifp = &sc->arpcom.ac_if;
4011 
4012 	lwkt_serialize_enter(ifp->if_serializer);
4013 
4014 	if (ifp->if_flags & IFF_UP) {
4015 		bge_init(sc);
4016 
4017 		if (!ifq_is_empty(&ifp->if_snd))
4018 			if_devstart(ifp);
4019 	}
4020 
4021 	lwkt_serialize_exit(ifp->if_serializer);
4022 
4023 	return 0;
4024 }
4025 
4026 static void
4027 bge_setpromisc(struct bge_softc *sc)
4028 {
4029 	struct ifnet *ifp = &sc->arpcom.ac_if;
4030 
4031 	if (ifp->if_flags & IFF_PROMISC)
4032 		BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
4033 	else
4034 		BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
4035 }
4036 
4037 static void
4038 bge_dma_free(struct bge_softc *sc)
4039 {
4040 	int i;
4041 
4042 	/* Destroy RX mbuf DMA stuffs. */
4043 	if (sc->bge_cdata.bge_rx_mtag != NULL) {
4044 		for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
4045 			bus_dmamap_destroy(sc->bge_cdata.bge_rx_mtag,
4046 			    sc->bge_cdata.bge_rx_std_dmamap[i]);
4047 		}
4048 		bus_dmamap_destroy(sc->bge_cdata.bge_rx_mtag,
4049 				   sc->bge_cdata.bge_rx_tmpmap);
4050 		bus_dma_tag_destroy(sc->bge_cdata.bge_rx_mtag);
4051 	}
4052 
4053 	/* Destroy TX mbuf DMA stuffs. */
4054 	if (sc->bge_cdata.bge_tx_mtag != NULL) {
4055 		for (i = 0; i < BGE_TX_RING_CNT; i++) {
4056 			bus_dmamap_destroy(sc->bge_cdata.bge_tx_mtag,
4057 			    sc->bge_cdata.bge_tx_dmamap[i]);
4058 		}
4059 		bus_dma_tag_destroy(sc->bge_cdata.bge_tx_mtag);
4060 	}
4061 
4062 	/* Destroy standard RX ring */
4063 	bge_dma_block_free(sc->bge_cdata.bge_rx_std_ring_tag,
4064 			   sc->bge_cdata.bge_rx_std_ring_map,
4065 			   sc->bge_ldata.bge_rx_std_ring);
4066 
4067 	if (BGE_IS_JUMBO_CAPABLE(sc))
4068 		bge_free_jumbo_mem(sc);
4069 
4070 	/* Destroy RX return ring */
4071 	bge_dma_block_free(sc->bge_cdata.bge_rx_return_ring_tag,
4072 			   sc->bge_cdata.bge_rx_return_ring_map,
4073 			   sc->bge_ldata.bge_rx_return_ring);
4074 
4075 	/* Destroy TX ring */
4076 	bge_dma_block_free(sc->bge_cdata.bge_tx_ring_tag,
4077 			   sc->bge_cdata.bge_tx_ring_map,
4078 			   sc->bge_ldata.bge_tx_ring);
4079 
4080 	/* Destroy status block */
4081 	bge_dma_block_free(sc->bge_cdata.bge_status_tag,
4082 			   sc->bge_cdata.bge_status_map,
4083 			   sc->bge_ldata.bge_status_block);
4084 
4085 	/* Destroy statistics block */
4086 	bge_dma_block_free(sc->bge_cdata.bge_stats_tag,
4087 			   sc->bge_cdata.bge_stats_map,
4088 			   sc->bge_ldata.bge_stats);
4089 
4090 	/* Destroy the parent tag */
4091 	if (sc->bge_cdata.bge_parent_tag != NULL)
4092 		bus_dma_tag_destroy(sc->bge_cdata.bge_parent_tag);
4093 }
4094 
4095 static int
4096 bge_dma_alloc(struct bge_softc *sc)
4097 {
4098 	struct ifnet *ifp = &sc->arpcom.ac_if;
4099 	int i, error;
4100 	bus_addr_t lowaddr;
4101 	bus_size_t txmaxsz;
4102 
4103 	lowaddr = BUS_SPACE_MAXADDR;
4104 	if (sc->bge_flags & BGE_FLAG_MAXADDR_40BIT)
4105 		lowaddr = BGE_DMA_MAXADDR_40BIT;
4106 
4107 	/*
4108 	 * Allocate the parent bus DMA tag appropriate for PCI.
4109 	 *
4110 	 * All of the NetExtreme/NetLink controllers have 4GB boundary
4111 	 * DMA bug.
4112 	 * Whenever an address crosses a multiple of the 4GB boundary
4113 	 * (including 4GB, 8Gb, 12Gb, etc.) and makes the transition
4114 	 * from 0xX_FFFF_FFFF to 0x(X+1)_0000_0000 an internal DMA
4115 	 * state machine will lockup and cause the device to hang.
4116 	 */
4117 	error = bus_dma_tag_create(NULL, 1, BGE_DMA_BOUNDARY_4G,
4118 				   lowaddr, BUS_SPACE_MAXADDR,
4119 				   NULL, NULL,
4120 				   BUS_SPACE_MAXSIZE_32BIT, 0,
4121 				   BUS_SPACE_MAXSIZE_32BIT,
4122 				   0, &sc->bge_cdata.bge_parent_tag);
4123 	if (error) {
4124 		if_printf(ifp, "could not allocate parent dma tag\n");
4125 		return error;
4126 	}
4127 
4128 	/*
4129 	 * Create DMA tag and maps for RX mbufs.
4130 	 */
4131 	error = bus_dma_tag_create(sc->bge_cdata.bge_parent_tag, 1, 0,
4132 				   BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
4133 				   NULL, NULL, MCLBYTES, 1, MCLBYTES,
4134 				   BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK,
4135 				   &sc->bge_cdata.bge_rx_mtag);
4136 	if (error) {
4137 		if_printf(ifp, "could not allocate RX mbuf dma tag\n");
4138 		return error;
4139 	}
4140 
4141 	error = bus_dmamap_create(sc->bge_cdata.bge_rx_mtag,
4142 				  BUS_DMA_WAITOK, &sc->bge_cdata.bge_rx_tmpmap);
4143 	if (error) {
4144 		bus_dma_tag_destroy(sc->bge_cdata.bge_rx_mtag);
4145 		sc->bge_cdata.bge_rx_mtag = NULL;
4146 		return error;
4147 	}
4148 
4149 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
4150 		error = bus_dmamap_create(sc->bge_cdata.bge_rx_mtag,
4151 					  BUS_DMA_WAITOK,
4152 					  &sc->bge_cdata.bge_rx_std_dmamap[i]);
4153 		if (error) {
4154 			int j;
4155 
4156 			for (j = 0; j < i; ++j) {
4157 				bus_dmamap_destroy(sc->bge_cdata.bge_rx_mtag,
4158 					sc->bge_cdata.bge_rx_std_dmamap[j]);
4159 			}
4160 			bus_dma_tag_destroy(sc->bge_cdata.bge_rx_mtag);
4161 			sc->bge_cdata.bge_rx_mtag = NULL;
4162 
4163 			if_printf(ifp, "could not create DMA map for RX\n");
4164 			return error;
4165 		}
4166 	}
4167 
4168 	/*
4169 	 * Create DMA tag and maps for TX mbufs.
4170 	 */
4171 	if (sc->bge_flags & BGE_FLAG_TSO)
4172 		txmaxsz = IP_MAXPACKET + sizeof(struct ether_vlan_header);
4173 	else
4174 		txmaxsz = BGE_JUMBO_FRAMELEN;
4175 	error = bus_dma_tag_create(sc->bge_cdata.bge_parent_tag, 1, 0,
4176 				   BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
4177 				   NULL, NULL,
4178 				   txmaxsz, BGE_NSEG_NEW, PAGE_SIZE,
4179 				   BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK |
4180 				   BUS_DMA_ONEBPAGE,
4181 				   &sc->bge_cdata.bge_tx_mtag);
4182 	if (error) {
4183 		if_printf(ifp, "could not allocate TX mbuf dma tag\n");
4184 		return error;
4185 	}
4186 
4187 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
4188 		error = bus_dmamap_create(sc->bge_cdata.bge_tx_mtag,
4189 					  BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
4190 					  &sc->bge_cdata.bge_tx_dmamap[i]);
4191 		if (error) {
4192 			int j;
4193 
4194 			for (j = 0; j < i; ++j) {
4195 				bus_dmamap_destroy(sc->bge_cdata.bge_tx_mtag,
4196 					sc->bge_cdata.bge_tx_dmamap[j]);
4197 			}
4198 			bus_dma_tag_destroy(sc->bge_cdata.bge_tx_mtag);
4199 			sc->bge_cdata.bge_tx_mtag = NULL;
4200 
4201 			if_printf(ifp, "could not create DMA map for TX\n");
4202 			return error;
4203 		}
4204 	}
4205 
4206 	/*
4207 	 * Create DMA stuffs for standard RX ring.
4208 	 */
4209 	error = bge_dma_block_alloc(sc, BGE_STD_RX_RING_SZ,
4210 				    &sc->bge_cdata.bge_rx_std_ring_tag,
4211 				    &sc->bge_cdata.bge_rx_std_ring_map,
4212 				    (void *)&sc->bge_ldata.bge_rx_std_ring,
4213 				    &sc->bge_ldata.bge_rx_std_ring_paddr);
4214 	if (error) {
4215 		if_printf(ifp, "could not create std RX ring\n");
4216 		return error;
4217 	}
4218 
4219 	/*
4220 	 * Create jumbo buffer pool.
4221 	 */
4222 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
4223 		error = bge_alloc_jumbo_mem(sc);
4224 		if (error) {
4225 			if_printf(ifp, "could not create jumbo buffer pool\n");
4226 			return error;
4227 		}
4228 	}
4229 
4230 	/*
4231 	 * Create DMA stuffs for RX return ring.
4232 	 */
4233 	error = bge_dma_block_alloc(sc,
4234 	    BGE_RX_RTN_RING_SZ(sc->bge_return_ring_cnt),
4235 	    &sc->bge_cdata.bge_rx_return_ring_tag,
4236 	    &sc->bge_cdata.bge_rx_return_ring_map,
4237 	    (void *)&sc->bge_ldata.bge_rx_return_ring,
4238 	    &sc->bge_ldata.bge_rx_return_ring_paddr);
4239 	if (error) {
4240 		if_printf(ifp, "could not create RX ret ring\n");
4241 		return error;
4242 	}
4243 
4244 	/*
4245 	 * Create DMA stuffs for TX ring.
4246 	 */
4247 	error = bge_dma_block_alloc(sc, BGE_TX_RING_SZ,
4248 				    &sc->bge_cdata.bge_tx_ring_tag,
4249 				    &sc->bge_cdata.bge_tx_ring_map,
4250 				    (void *)&sc->bge_ldata.bge_tx_ring,
4251 				    &sc->bge_ldata.bge_tx_ring_paddr);
4252 	if (error) {
4253 		if_printf(ifp, "could not create TX ring\n");
4254 		return error;
4255 	}
4256 
4257 	/*
4258 	 * Create DMA stuffs for status block.
4259 	 */
4260 	error = bge_dma_block_alloc(sc, BGE_STATUS_BLK_SZ,
4261 				    &sc->bge_cdata.bge_status_tag,
4262 				    &sc->bge_cdata.bge_status_map,
4263 				    (void *)&sc->bge_ldata.bge_status_block,
4264 				    &sc->bge_ldata.bge_status_block_paddr);
4265 	if (error) {
4266 		if_printf(ifp, "could not create status block\n");
4267 		return error;
4268 	}
4269 
4270 	/*
4271 	 * Create DMA stuffs for statistics block.
4272 	 */
4273 	error = bge_dma_block_alloc(sc, BGE_STATS_SZ,
4274 				    &sc->bge_cdata.bge_stats_tag,
4275 				    &sc->bge_cdata.bge_stats_map,
4276 				    (void *)&sc->bge_ldata.bge_stats,
4277 				    &sc->bge_ldata.bge_stats_paddr);
4278 	if (error) {
4279 		if_printf(ifp, "could not create stats block\n");
4280 		return error;
4281 	}
4282 	return 0;
4283 }
4284 
4285 static int
4286 bge_dma_block_alloc(struct bge_softc *sc, bus_size_t size, bus_dma_tag_t *tag,
4287 		    bus_dmamap_t *map, void **addr, bus_addr_t *paddr)
4288 {
4289 	bus_dmamem_t dmem;
4290 	int error;
4291 
4292 	error = bus_dmamem_coherent(sc->bge_cdata.bge_parent_tag, PAGE_SIZE, 0,
4293 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
4294 				    size, BUS_DMA_WAITOK | BUS_DMA_ZERO, &dmem);
4295 	if (error)
4296 		return error;
4297 
4298 	*tag = dmem.dmem_tag;
4299 	*map = dmem.dmem_map;
4300 	*addr = dmem.dmem_addr;
4301 	*paddr = dmem.dmem_busaddr;
4302 
4303 	return 0;
4304 }
4305 
4306 static void
4307 bge_dma_block_free(bus_dma_tag_t tag, bus_dmamap_t map, void *addr)
4308 {
4309 	if (tag != NULL) {
4310 		bus_dmamap_unload(tag, map);
4311 		bus_dmamem_free(tag, addr, map);
4312 		bus_dma_tag_destroy(tag);
4313 	}
4314 }
4315 
4316 /*
4317  * Grrr. The link status word in the status block does
4318  * not work correctly on the BCM5700 rev AX and BX chips,
4319  * according to all available information. Hence, we have
4320  * to enable MII interrupts in order to properly obtain
4321  * async link changes. Unfortunately, this also means that
4322  * we have to read the MAC status register to detect link
4323  * changes, thereby adding an additional register access to
4324  * the interrupt handler.
4325  *
4326  * XXX: perhaps link state detection procedure used for
4327  * BGE_CHIPID_BCM5700_B2 can be used for others BCM5700 revisions.
4328  */
4329 static void
4330 bge_bcm5700_link_upd(struct bge_softc *sc, uint32_t status __unused)
4331 {
4332 	struct ifnet *ifp = &sc->arpcom.ac_if;
4333 	struct mii_data *mii = device_get_softc(sc->bge_miibus);
4334 
4335 	mii_pollstat(mii);
4336 
4337 	if (!sc->bge_link &&
4338 	    (mii->mii_media_status & IFM_ACTIVE) &&
4339 	    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
4340 		sc->bge_link++;
4341 		if (bootverbose)
4342 			if_printf(ifp, "link UP\n");
4343 	} else if (sc->bge_link &&
4344 	    (!(mii->mii_media_status & IFM_ACTIVE) ||
4345 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE)) {
4346 		sc->bge_link = 0;
4347 		if (bootverbose)
4348 			if_printf(ifp, "link DOWN\n");
4349 	}
4350 
4351 	/* Clear the interrupt. */
4352 	CSR_WRITE_4(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_MI_INTERRUPT);
4353 	bge_miibus_readreg(sc->bge_dev, 1, BRGPHY_MII_ISR);
4354 	bge_miibus_writereg(sc->bge_dev, 1, BRGPHY_MII_IMR, BRGPHY_INTRS);
4355 }
4356 
4357 static void
4358 bge_tbi_link_upd(struct bge_softc *sc, uint32_t status)
4359 {
4360 	struct ifnet *ifp = &sc->arpcom.ac_if;
4361 
4362 #define PCS_ENCODE_ERR	(BGE_MACSTAT_PORT_DECODE_ERROR|BGE_MACSTAT_MI_COMPLETE)
4363 
4364 	/*
4365 	 * Sometimes PCS encoding errors are detected in
4366 	 * TBI mode (on fiber NICs), and for some reason
4367 	 * the chip will signal them as link changes.
4368 	 * If we get a link change event, but the 'PCS
4369 	 * encoding error' bit in the MAC status register
4370 	 * is set, don't bother doing a link check.
4371 	 * This avoids spurious "gigabit link up" messages
4372 	 * that sometimes appear on fiber NICs during
4373 	 * periods of heavy traffic.
4374 	 */
4375 	if (status & BGE_MACSTAT_TBI_PCS_SYNCHED) {
4376 		if (!sc->bge_link) {
4377 			sc->bge_link++;
4378 			if (sc->bge_asicrev == BGE_ASICREV_BCM5704) {
4379 				BGE_CLRBIT(sc, BGE_MAC_MODE,
4380 				    BGE_MACMODE_TBI_SEND_CFGS);
4381 			}
4382 			CSR_WRITE_4(sc, BGE_MAC_STS, 0xFFFFFFFF);
4383 
4384 			if (bootverbose)
4385 				if_printf(ifp, "link UP\n");
4386 
4387 			ifp->if_link_state = LINK_STATE_UP;
4388 			if_link_state_change(ifp);
4389 		}
4390 	} else if ((status & PCS_ENCODE_ERR) != PCS_ENCODE_ERR) {
4391 		if (sc->bge_link) {
4392 			sc->bge_link = 0;
4393 
4394 			if (bootverbose)
4395 				if_printf(ifp, "link DOWN\n");
4396 
4397 			ifp->if_link_state = LINK_STATE_DOWN;
4398 			if_link_state_change(ifp);
4399 		}
4400 	}
4401 
4402 #undef PCS_ENCODE_ERR
4403 
4404 	/* Clear the attention. */
4405 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
4406 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
4407 	    BGE_MACSTAT_LINK_CHANGED);
4408 }
4409 
4410 static void
4411 bge_copper_link_upd(struct bge_softc *sc, uint32_t status __unused)
4412 {
4413 	struct ifnet *ifp = &sc->arpcom.ac_if;
4414 	struct mii_data *mii = device_get_softc(sc->bge_miibus);
4415 
4416 	mii_pollstat(mii);
4417 	bge_miibus_statchg(sc->bge_dev);
4418 
4419 	if (bootverbose) {
4420 		if (sc->bge_link)
4421 			if_printf(ifp, "link UP\n");
4422 		else
4423 			if_printf(ifp, "link DOWN\n");
4424 	}
4425 
4426 	/* Clear the attention. */
4427 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
4428 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
4429 	    BGE_MACSTAT_LINK_CHANGED);
4430 }
4431 
4432 static void
4433 bge_autopoll_link_upd(struct bge_softc *sc, uint32_t status __unused)
4434 {
4435 	struct ifnet *ifp = &sc->arpcom.ac_if;
4436 	struct mii_data *mii = device_get_softc(sc->bge_miibus);
4437 
4438 	mii_pollstat(mii);
4439 
4440 	if (!sc->bge_link &&
4441 	    (mii->mii_media_status & IFM_ACTIVE) &&
4442 	    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
4443 		sc->bge_link++;
4444 		if (bootverbose)
4445 			if_printf(ifp, "link UP\n");
4446 	} else if (sc->bge_link &&
4447 	    (!(mii->mii_media_status & IFM_ACTIVE) ||
4448 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE)) {
4449 		sc->bge_link = 0;
4450 		if (bootverbose)
4451 			if_printf(ifp, "link DOWN\n");
4452 	}
4453 
4454 	/* Clear the attention. */
4455 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
4456 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
4457 	    BGE_MACSTAT_LINK_CHANGED);
4458 }
4459 
4460 static int
4461 bge_sysctl_rx_coal_ticks(SYSCTL_HANDLER_ARGS)
4462 {
4463 	struct bge_softc *sc = arg1;
4464 
4465 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4466 	    &sc->bge_rx_coal_ticks,
4467 	    BGE_RX_COAL_TICKS_MIN, BGE_RX_COAL_TICKS_MAX,
4468 	    BGE_RX_COAL_TICKS_CHG);
4469 }
4470 
4471 static int
4472 bge_sysctl_tx_coal_ticks(SYSCTL_HANDLER_ARGS)
4473 {
4474 	struct bge_softc *sc = arg1;
4475 
4476 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4477 	    &sc->bge_tx_coal_ticks,
4478 	    BGE_TX_COAL_TICKS_MIN, BGE_TX_COAL_TICKS_MAX,
4479 	    BGE_TX_COAL_TICKS_CHG);
4480 }
4481 
4482 static int
4483 bge_sysctl_rx_coal_bds(SYSCTL_HANDLER_ARGS)
4484 {
4485 	struct bge_softc *sc = arg1;
4486 
4487 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4488 	    &sc->bge_rx_coal_bds,
4489 	    BGE_RX_COAL_BDS_MIN, BGE_RX_COAL_BDS_MAX,
4490 	    BGE_RX_COAL_BDS_CHG);
4491 }
4492 
4493 static int
4494 bge_sysctl_tx_coal_bds(SYSCTL_HANDLER_ARGS)
4495 {
4496 	struct bge_softc *sc = arg1;
4497 
4498 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4499 	    &sc->bge_tx_coal_bds,
4500 	    BGE_TX_COAL_BDS_MIN, BGE_TX_COAL_BDS_MAX,
4501 	    BGE_TX_COAL_BDS_CHG);
4502 }
4503 
4504 static int
4505 bge_sysctl_rx_coal_ticks_int(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_int,
4511 	    BGE_RX_COAL_TICKS_MIN, BGE_RX_COAL_TICKS_MAX,
4512 	    BGE_RX_COAL_TICKS_INT_CHG);
4513 }
4514 
4515 static int
4516 bge_sysctl_tx_coal_ticks_int(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_int,
4522 	    BGE_TX_COAL_TICKS_MIN, BGE_TX_COAL_TICKS_MAX,
4523 	    BGE_TX_COAL_TICKS_INT_CHG);
4524 }
4525 
4526 static int
4527 bge_sysctl_rx_coal_bds_int(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_int,
4533 	    BGE_RX_COAL_BDS_MIN, BGE_RX_COAL_BDS_MAX,
4534 	    BGE_RX_COAL_BDS_INT_CHG);
4535 }
4536 
4537 static int
4538 bge_sysctl_tx_coal_bds_int(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_int,
4544 	    BGE_TX_COAL_BDS_MIN, BGE_TX_COAL_BDS_MAX,
4545 	    BGE_TX_COAL_BDS_INT_CHG);
4546 }
4547 
4548 static int
4549 bge_sysctl_coal_chg(SYSCTL_HANDLER_ARGS, uint32_t *coal,
4550     int coal_min, int coal_max, uint32_t coal_chg_mask)
4551 {
4552 	struct bge_softc *sc = arg1;
4553 	struct ifnet *ifp = &sc->arpcom.ac_if;
4554 	int error = 0, v;
4555 
4556 	lwkt_serialize_enter(ifp->if_serializer);
4557 
4558 	v = *coal;
4559 	error = sysctl_handle_int(oidp, &v, 0, req);
4560 	if (!error && req->newptr != NULL) {
4561 		if (v < coal_min || v > coal_max) {
4562 			error = EINVAL;
4563 		} else {
4564 			*coal = v;
4565 			sc->bge_coal_chg |= coal_chg_mask;
4566 		}
4567 	}
4568 
4569 	lwkt_serialize_exit(ifp->if_serializer);
4570 	return error;
4571 }
4572 
4573 static void
4574 bge_coal_change(struct bge_softc *sc)
4575 {
4576 	struct ifnet *ifp = &sc->arpcom.ac_if;
4577 	uint32_t val;
4578 
4579 	ASSERT_SERIALIZED(ifp->if_serializer);
4580 
4581 	if (sc->bge_coal_chg & BGE_RX_COAL_TICKS_CHG) {
4582 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS,
4583 			    sc->bge_rx_coal_ticks);
4584 		DELAY(10);
4585 		val = CSR_READ_4(sc, BGE_HCC_RX_COAL_TICKS);
4586 
4587 		if (bootverbose) {
4588 			if_printf(ifp, "rx_coal_ticks -> %u\n",
4589 				  sc->bge_rx_coal_ticks);
4590 		}
4591 	}
4592 
4593 	if (sc->bge_coal_chg & BGE_TX_COAL_TICKS_CHG) {
4594 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS,
4595 			    sc->bge_tx_coal_ticks);
4596 		DELAY(10);
4597 		val = CSR_READ_4(sc, BGE_HCC_TX_COAL_TICKS);
4598 
4599 		if (bootverbose) {
4600 			if_printf(ifp, "tx_coal_ticks -> %u\n",
4601 				  sc->bge_tx_coal_ticks);
4602 		}
4603 	}
4604 
4605 	if (sc->bge_coal_chg & BGE_RX_COAL_BDS_CHG) {
4606 		CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS,
4607 			    sc->bge_rx_coal_bds);
4608 		DELAY(10);
4609 		val = CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS);
4610 
4611 		if (bootverbose) {
4612 			if_printf(ifp, "rx_coal_bds -> %u\n",
4613 				  sc->bge_rx_coal_bds);
4614 		}
4615 	}
4616 
4617 	if (sc->bge_coal_chg & BGE_TX_COAL_BDS_CHG) {
4618 		CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS,
4619 			    sc->bge_tx_coal_bds);
4620 		DELAY(10);
4621 		val = CSR_READ_4(sc, BGE_HCC_TX_MAX_COAL_BDS);
4622 
4623 		if (bootverbose) {
4624 			if_printf(ifp, "tx_max_coal_bds -> %u\n",
4625 				  sc->bge_tx_coal_bds);
4626 		}
4627 	}
4628 
4629 	if (sc->bge_coal_chg & BGE_RX_COAL_TICKS_INT_CHG) {
4630 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS_INT,
4631 		    sc->bge_rx_coal_ticks_int);
4632 		DELAY(10);
4633 		val = CSR_READ_4(sc, BGE_HCC_RX_COAL_TICKS_INT);
4634 
4635 		if (bootverbose) {
4636 			if_printf(ifp, "rx_coal_ticks_int -> %u\n",
4637 			    sc->bge_rx_coal_ticks_int);
4638 		}
4639 	}
4640 
4641 	if (sc->bge_coal_chg & BGE_TX_COAL_TICKS_INT_CHG) {
4642 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS_INT,
4643 		    sc->bge_tx_coal_ticks_int);
4644 		DELAY(10);
4645 		val = CSR_READ_4(sc, BGE_HCC_TX_COAL_TICKS_INT);
4646 
4647 		if (bootverbose) {
4648 			if_printf(ifp, "tx_coal_ticks_int -> %u\n",
4649 			    sc->bge_tx_coal_ticks_int);
4650 		}
4651 	}
4652 
4653 	if (sc->bge_coal_chg & BGE_RX_COAL_BDS_INT_CHG) {
4654 		CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT,
4655 		    sc->bge_rx_coal_bds_int);
4656 		DELAY(10);
4657 		val = CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT);
4658 
4659 		if (bootverbose) {
4660 			if_printf(ifp, "rx_coal_bds_int -> %u\n",
4661 			    sc->bge_rx_coal_bds_int);
4662 		}
4663 	}
4664 
4665 	if (sc->bge_coal_chg & BGE_TX_COAL_BDS_INT_CHG) {
4666 		CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT,
4667 		    sc->bge_tx_coal_bds_int);
4668 		DELAY(10);
4669 		val = CSR_READ_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT);
4670 
4671 		if (bootverbose) {
4672 			if_printf(ifp, "tx_coal_bds_int -> %u\n",
4673 			    sc->bge_tx_coal_bds_int);
4674 		}
4675 	}
4676 
4677 	sc->bge_coal_chg = 0;
4678 }
4679 
4680 static void
4681 bge_enable_intr(struct bge_softc *sc)
4682 {
4683 	struct ifnet *ifp = &sc->arpcom.ac_if;
4684 
4685 	lwkt_serialize_handler_enable(ifp->if_serializer);
4686 
4687 	/*
4688 	 * Enable interrupt.
4689 	 */
4690 	bge_writembx(sc, BGE_MBX_IRQ0_LO, sc->bge_status_tag << 24);
4691 	if (sc->bge_flags & BGE_FLAG_ONESHOT_MSI) {
4692 		/* XXX Linux driver */
4693 		bge_writembx(sc, BGE_MBX_IRQ0_LO, sc->bge_status_tag << 24);
4694 	}
4695 
4696 	/*
4697 	 * Unmask the interrupt when we stop polling.
4698 	 */
4699 	PCI_CLRBIT(sc->bge_dev, BGE_PCI_MISC_CTL,
4700 	    BGE_PCIMISCCTL_MASK_PCI_INTR, 4);
4701 
4702 	/*
4703 	 * Trigger another interrupt, since above writing
4704 	 * to interrupt mailbox0 may acknowledge pending
4705 	 * interrupt.
4706 	 */
4707 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
4708 }
4709 
4710 static void
4711 bge_disable_intr(struct bge_softc *sc)
4712 {
4713 	struct ifnet *ifp = &sc->arpcom.ac_if;
4714 
4715 	/*
4716 	 * Mask the interrupt when we start polling.
4717 	 */
4718 	PCI_SETBIT(sc->bge_dev, BGE_PCI_MISC_CTL,
4719 	    BGE_PCIMISCCTL_MASK_PCI_INTR, 4);
4720 
4721 	/*
4722 	 * Acknowledge possible asserted interrupt.
4723 	 */
4724 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
4725 
4726 	lwkt_serialize_handler_disable(ifp->if_serializer);
4727 }
4728 
4729 static int
4730 bge_get_eaddr_mem(struct bge_softc *sc, uint8_t ether_addr[])
4731 {
4732 	uint32_t mac_addr;
4733 	int ret = 1;
4734 
4735 	mac_addr = bge_readmem_ind(sc, 0x0c14);
4736 	if ((mac_addr >> 16) == 0x484b) {
4737 		ether_addr[0] = (uint8_t)(mac_addr >> 8);
4738 		ether_addr[1] = (uint8_t)mac_addr;
4739 		mac_addr = bge_readmem_ind(sc, 0x0c18);
4740 		ether_addr[2] = (uint8_t)(mac_addr >> 24);
4741 		ether_addr[3] = (uint8_t)(mac_addr >> 16);
4742 		ether_addr[4] = (uint8_t)(mac_addr >> 8);
4743 		ether_addr[5] = (uint8_t)mac_addr;
4744 		ret = 0;
4745 	}
4746 	return ret;
4747 }
4748 
4749 static int
4750 bge_get_eaddr_nvram(struct bge_softc *sc, uint8_t ether_addr[])
4751 {
4752 	int mac_offset = BGE_EE_MAC_OFFSET;
4753 
4754 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
4755 		mac_offset = BGE_EE_MAC_OFFSET_5906;
4756 
4757 	return bge_read_nvram(sc, ether_addr, mac_offset + 2, ETHER_ADDR_LEN);
4758 }
4759 
4760 static int
4761 bge_get_eaddr_eeprom(struct bge_softc *sc, uint8_t ether_addr[])
4762 {
4763 	if (sc->bge_flags & BGE_FLAG_NO_EEPROM)
4764 		return 1;
4765 
4766 	return bge_read_eeprom(sc, ether_addr, BGE_EE_MAC_OFFSET + 2,
4767 			       ETHER_ADDR_LEN);
4768 }
4769 
4770 static int
4771 bge_get_eaddr(struct bge_softc *sc, uint8_t eaddr[])
4772 {
4773 	static const bge_eaddr_fcn_t bge_eaddr_funcs[] = {
4774 		/* NOTE: Order is critical */
4775 		bge_get_eaddr_mem,
4776 		bge_get_eaddr_nvram,
4777 		bge_get_eaddr_eeprom,
4778 		NULL
4779 	};
4780 	const bge_eaddr_fcn_t *func;
4781 
4782 	for (func = bge_eaddr_funcs; *func != NULL; ++func) {
4783 		if ((*func)(sc, eaddr) == 0)
4784 			break;
4785 	}
4786 	return (*func == NULL ? ENXIO : 0);
4787 }
4788 
4789 /*
4790  * NOTE: 'm' is not freed upon failure
4791  */
4792 struct mbuf *
4793 bge_defrag_shortdma(struct mbuf *m)
4794 {
4795 	struct mbuf *n;
4796 	int found;
4797 
4798 	/*
4799 	 * If device receive two back-to-back send BDs with less than
4800 	 * or equal to 8 total bytes then the device may hang.  The two
4801 	 * back-to-back send BDs must in the same frame for this failure
4802 	 * to occur.  Scan mbuf chains and see whether two back-to-back
4803 	 * send BDs are there.  If this is the case, allocate new mbuf
4804 	 * and copy the frame to workaround the silicon bug.
4805 	 */
4806 	for (n = m, found = 0; n != NULL; n = n->m_next) {
4807 		if (n->m_len < 8) {
4808 			found++;
4809 			if (found > 1)
4810 				break;
4811 			continue;
4812 		}
4813 		found = 0;
4814 	}
4815 
4816 	if (found > 1)
4817 		n = m_defrag(m, MB_DONTWAIT);
4818 	else
4819 		n = m;
4820 	return n;
4821 }
4822 
4823 static void
4824 bge_stop_block(struct bge_softc *sc, bus_size_t reg, uint32_t bit)
4825 {
4826 	int i;
4827 
4828 	BGE_CLRBIT(sc, reg, bit);
4829 	for (i = 0; i < BGE_TIMEOUT; i++) {
4830 		if ((CSR_READ_4(sc, reg) & bit) == 0)
4831 			return;
4832 		DELAY(100);
4833 	}
4834 }
4835 
4836 static void
4837 bge_link_poll(struct bge_softc *sc)
4838 {
4839 	uint32_t status;
4840 
4841 	status = CSR_READ_4(sc, BGE_MAC_STS);
4842 	if ((status & sc->bge_link_chg) || sc->bge_link_evt) {
4843 		sc->bge_link_evt = 0;
4844 		sc->bge_link_upd(sc, status);
4845 	}
4846 }
4847 
4848 static void
4849 bge_enable_msi(struct bge_softc *sc)
4850 {
4851 	uint32_t msi_mode;
4852 
4853 	msi_mode = CSR_READ_4(sc, BGE_MSI_MODE);
4854 	msi_mode |= BGE_MSIMODE_ENABLE;
4855 	if (sc->bge_flags & BGE_FLAG_ONESHOT_MSI) {
4856 		/*
4857 		 * According to all of the datasheets that are publicly
4858 		 * available, bit 5 of the MSI_MODE is defined to be
4859 		 * "MSI FIFO Underrun Attn" for BCM5755+ and BCM5906, on
4860 		 * which "oneshot MSI" is enabled.  However, it is always
4861 		 * safe to clear it here.
4862 		 */
4863 		msi_mode &= ~BGE_MSIMODE_ONESHOT_DISABLE;
4864 	}
4865 	CSR_WRITE_4(sc, BGE_MSI_MODE, msi_mode);
4866 }
4867 
4868 static int
4869 bge_setup_tso(struct bge_softc *sc, struct mbuf **mp,
4870     uint16_t *mss0, uint16_t *flags0)
4871 {
4872 	struct mbuf *m;
4873 	struct ip *ip;
4874 	struct tcphdr *th;
4875 	int thoff, iphlen, hoff, hlen;
4876 	uint16_t flags, mss;
4877 
4878 	m = *mp;
4879 	KASSERT(M_WRITABLE(m), ("TSO mbuf not writable"));
4880 
4881 	hoff = m->m_pkthdr.csum_lhlen;
4882 	iphlen = m->m_pkthdr.csum_iphlen;
4883 	thoff = m->m_pkthdr.csum_thlen;
4884 
4885 	KASSERT(hoff > 0, ("invalid ether header len"));
4886 	KASSERT(iphlen > 0, ("invalid ip header len"));
4887 	KASSERT(thoff > 0, ("invalid tcp header len"));
4888 
4889 	if (__predict_false(m->m_len < hoff + iphlen + thoff)) {
4890 		m = m_pullup(m, hoff + iphlen + thoff);
4891 		if (m == NULL) {
4892 			*mp = NULL;
4893 			return ENOBUFS;
4894 		}
4895 		*mp = m;
4896 	}
4897 	ip = mtodoff(m, struct ip *, hoff);
4898 	th = mtodoff(m, struct tcphdr *, hoff + iphlen);
4899 
4900 	mss = m->m_pkthdr.tso_segsz;
4901 	flags = BGE_TXBDFLAG_CPU_PRE_DMA | BGE_TXBDFLAG_CPU_POST_DMA;
4902 
4903 	ip->ip_len = htons(mss + iphlen + thoff);
4904 	th->th_sum = 0;
4905 
4906 	hlen = (iphlen + thoff) >> 2;
4907 	mss |= (hlen << 11);
4908 
4909 	*mss0 = mss;
4910 	*flags0 = flags;
4911 
4912 	return 0;
4913 }
4914