xref: /dflybsd-src/sys/dev/netif/bge/if_bge.c (revision 46c7fb85d65f149352ad0eeb4d067b97ff57ce9e)
1 /*
2  * Copyright (c) 2001 Wind River Systems
3  * Copyright (c) 1997, 1998, 1999, 2001
4  *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. All advertising materials mentioning features or use of this software
15  *    must display the following acknowledgement:
16  *	This product includes software developed by Bill Paul.
17  * 4. Neither the name of the author nor the names of any co-contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
25  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
28  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
29  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
30  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
31  * THE POSSIBILITY OF SUCH DAMAGE.
32  *
33  * $FreeBSD: src/sys/dev/bge/if_bge.c,v 1.3.2.39 2005/07/03 03:41:18 silby Exp $
34  */
35 
36 /*
37  * Broadcom BCM570x family gigabit ethernet driver for FreeBSD.
38  *
39  * Written by Bill Paul <wpaul@windriver.com>
40  * Senior Engineer, Wind River Systems
41  */
42 
43 /*
44  * The Broadcom BCM5700 is based on technology originally developed by
45  * Alteon Networks as part of the Tigon I and Tigon II gigabit ethernet
46  * MAC chips. The BCM5700, sometimes refered to as the Tigon III, has
47  * two on-board MIPS R4000 CPUs and can have as much as 16MB of external
48  * SSRAM. The BCM5700 supports TCP, UDP and IP checksum offload, jumbo
49  * frames, highly configurable RX filtering, and 16 RX and TX queues
50  * (which, along with RX filter rules, can be used for QOS applications).
51  * Other features, such as TCP segmentation, may be available as part
52  * of value-added firmware updates. Unlike the Tigon I and Tigon II,
53  * firmware images can be stored in hardware and need not be compiled
54  * into the driver.
55  *
56  * The BCM5700 supports the PCI v2.2 and PCI-X v1.0 standards, and will
57  * function in a 32-bit/64-bit 33/66Mhz bus, or a 64-bit/133Mhz bus.
58  *
59  * The BCM5701 is a single-chip solution incorporating both the BCM5700
60  * MAC and a BCM5401 10/100/1000 PHY. Unlike the BCM5700, the BCM5701
61  * does not support external SSRAM.
62  *
63  * Broadcom also produces a variation of the BCM5700 under the "Altima"
64  * brand name, which is functionally similar but lacks PCI-X support.
65  *
66  * Without external SSRAM, you can only have at most 4 TX rings,
67  * and the use of the mini RX ring is disabled. This seems to imply
68  * that these features are simply not available on the BCM5701. As a
69  * result, this driver does not implement any support for the mini RX
70  * ring.
71  */
72 
73 #include "opt_ifpoll.h"
74 
75 #include <sys/param.h>
76 #include <sys/bus.h>
77 #include <sys/endian.h>
78 #include <sys/kernel.h>
79 #include <sys/ktr.h>
80 #include <sys/interrupt.h>
81 #include <sys/mbuf.h>
82 #include <sys/malloc.h>
83 #include <sys/queue.h>
84 #include <sys/rman.h>
85 #include <sys/serialize.h>
86 #include <sys/socket.h>
87 #include <sys/sockio.h>
88 #include <sys/sysctl.h>
89 
90 #include <netinet/ip.h>
91 #include <netinet/tcp.h>
92 
93 #include <net/bpf.h>
94 #include <net/ethernet.h>
95 #include <net/if.h>
96 #include <net/if_arp.h>
97 #include <net/if_dl.h>
98 #include <net/if_media.h>
99 #include <net/if_poll.h>
100 #include <net/if_types.h>
101 #include <net/ifq_var.h>
102 #include <net/vlan/if_vlan_var.h>
103 #include <net/vlan/if_vlan_ether.h>
104 
105 #include <dev/netif/mii_layer/mii.h>
106 #include <dev/netif/mii_layer/miivar.h>
107 #include <dev/netif/mii_layer/brgphyreg.h>
108 
109 #include <bus/pci/pcidevs.h>
110 #include <bus/pci/pcireg.h>
111 #include <bus/pci/pcivar.h>
112 
113 #include <dev/netif/bge/if_bgereg.h>
114 #include <dev/netif/bge/if_bgevar.h>
115 
116 /* "device miibus" required.  See GENERIC if you get errors here. */
117 #include "miibus_if.h"
118 
119 #define BGE_CSUM_FEATURES	(CSUM_IP | CSUM_TCP)
120 
121 static const struct bge_type {
122 	uint16_t		bge_vid;
123 	uint16_t		bge_did;
124 	char			*bge_name;
125 } bge_devs[] = {
126 	{ PCI_VENDOR_3COM, PCI_PRODUCT_3COM_3C996,
127 		"3COM 3C996 Gigabit Ethernet" },
128 
129 	{ PCI_VENDOR_ALTEON, PCI_PRODUCT_ALTEON_BCM5700,
130 		"Alteon BCM5700 Gigabit Ethernet" },
131 	{ PCI_VENDOR_ALTEON, PCI_PRODUCT_ALTEON_BCM5701,
132 		"Alteon BCM5701 Gigabit Ethernet" },
133 
134 	{ PCI_VENDOR_ALTIMA, PCI_PRODUCT_ALTIMA_AC1000,
135 		"Altima AC1000 Gigabit Ethernet" },
136 	{ PCI_VENDOR_ALTIMA, PCI_PRODUCT_ALTIMA_AC1001,
137 		"Altima AC1002 Gigabit Ethernet" },
138 	{ PCI_VENDOR_ALTIMA, PCI_PRODUCT_ALTIMA_AC9100,
139 		"Altima AC9100 Gigabit Ethernet" },
140 
141 	{ PCI_VENDOR_APPLE, PCI_PRODUCT_APPLE_BCM5701,
142 		"Apple BCM5701 Gigabit Ethernet" },
143 
144 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5700,
145 		"Broadcom BCM5700 Gigabit Ethernet" },
146 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5701,
147 		"Broadcom BCM5701 Gigabit Ethernet" },
148 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5702,
149 		"Broadcom BCM5702 Gigabit Ethernet" },
150 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5702X,
151 		"Broadcom BCM5702X Gigabit Ethernet" },
152 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5702_ALT,
153 		"Broadcom BCM5702 Gigabit Ethernet" },
154 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5703,
155 		"Broadcom BCM5703 Gigabit Ethernet" },
156 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5703X,
157 		"Broadcom BCM5703X Gigabit Ethernet" },
158 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5703A3,
159 		"Broadcom BCM5703 Gigabit Ethernet" },
160 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5704C,
161 		"Broadcom BCM5704C Dual Gigabit Ethernet" },
162 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5704S,
163 		"Broadcom BCM5704S Dual Gigabit Ethernet" },
164 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5704S_ALT,
165 		"Broadcom BCM5704S Dual Gigabit Ethernet" },
166 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5705,
167 		"Broadcom BCM5705 Gigabit Ethernet" },
168 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5705F,
169 		"Broadcom BCM5705F Gigabit Ethernet" },
170 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5705K,
171 		"Broadcom BCM5705K Gigabit Ethernet" },
172 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5705M,
173 		"Broadcom BCM5705M Gigabit Ethernet" },
174 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5705M_ALT,
175 		"Broadcom BCM5705M Gigabit Ethernet" },
176 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5714,
177 		"Broadcom BCM5714C Gigabit Ethernet" },
178 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5714S,
179 		"Broadcom BCM5714S Gigabit Ethernet" },
180 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5715,
181 		"Broadcom BCM5715 Gigabit Ethernet" },
182 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5715S,
183 		"Broadcom BCM5715S Gigabit Ethernet" },
184 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5720,
185 		"Broadcom BCM5720 Gigabit Ethernet" },
186 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5721,
187 		"Broadcom BCM5721 Gigabit Ethernet" },
188 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5722,
189 		"Broadcom BCM5722 Gigabit Ethernet" },
190 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5723,
191 		"Broadcom BCM5723 Gigabit Ethernet" },
192 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5750,
193 		"Broadcom BCM5750 Gigabit Ethernet" },
194 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5750M,
195 		"Broadcom BCM5750M Gigabit Ethernet" },
196 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5751,
197 		"Broadcom BCM5751 Gigabit Ethernet" },
198 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5751F,
199 		"Broadcom BCM5751F Gigabit Ethernet" },
200 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5751M,
201 		"Broadcom BCM5751M Gigabit Ethernet" },
202 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5752,
203 		"Broadcom BCM5752 Gigabit Ethernet" },
204 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5752M,
205 		"Broadcom BCM5752M Gigabit Ethernet" },
206 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5753,
207 		"Broadcom BCM5753 Gigabit Ethernet" },
208 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5753F,
209 		"Broadcom BCM5753F Gigabit Ethernet" },
210 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5753M,
211 		"Broadcom BCM5753M Gigabit Ethernet" },
212 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5754,
213 		"Broadcom BCM5754 Gigabit Ethernet" },
214 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5754M,
215 		"Broadcom BCM5754M Gigabit Ethernet" },
216 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5755,
217 		"Broadcom BCM5755 Gigabit Ethernet" },
218 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5755M,
219 		"Broadcom BCM5755M Gigabit Ethernet" },
220 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5756,
221 		"Broadcom BCM5756 Gigabit Ethernet" },
222 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5761,
223 		"Broadcom BCM5761 Gigabit Ethernet" },
224 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5761E,
225 		"Broadcom BCM5761E Gigabit Ethernet" },
226 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5761S,
227 		"Broadcom BCM5761S Gigabit Ethernet" },
228 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5761SE,
229 		"Broadcom BCM5761SE Gigabit Ethernet" },
230 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5764,
231 		"Broadcom BCM5764 Gigabit Ethernet" },
232 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5780,
233 		"Broadcom BCM5780 Gigabit Ethernet" },
234 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5780S,
235 		"Broadcom BCM5780S Gigabit Ethernet" },
236 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5781,
237 		"Broadcom BCM5781 Gigabit Ethernet" },
238 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5782,
239 		"Broadcom BCM5782 Gigabit Ethernet" },
240 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5784,
241 		"Broadcom BCM5784 Gigabit Ethernet" },
242 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5785F,
243 		"Broadcom BCM5785F Gigabit Ethernet" },
244 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5785G,
245 		"Broadcom BCM5785G Gigabit Ethernet" },
246 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5786,
247 		"Broadcom BCM5786 Gigabit Ethernet" },
248 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5787,
249 		"Broadcom BCM5787 Gigabit Ethernet" },
250 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5787F,
251 		"Broadcom BCM5787F Gigabit Ethernet" },
252 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5787M,
253 		"Broadcom BCM5787M Gigabit Ethernet" },
254 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5788,
255 		"Broadcom BCM5788 Gigabit Ethernet" },
256 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5789,
257 		"Broadcom BCM5789 Gigabit Ethernet" },
258 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5901,
259 		"Broadcom BCM5901 Fast Ethernet" },
260 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5901A2,
261 		"Broadcom BCM5901A2 Fast Ethernet" },
262 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5903M,
263 		"Broadcom BCM5903M Fast Ethernet" },
264 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5906,
265 		"Broadcom BCM5906 Fast Ethernet"},
266 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5906M,
267 		"Broadcom BCM5906M Fast Ethernet"},
268 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57760,
269 		"Broadcom BCM57760 Gigabit Ethernet"},
270 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57780,
271 		"Broadcom BCM57780 Gigabit Ethernet"},
272 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57788,
273 		"Broadcom BCM57788 Gigabit Ethernet"},
274 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57790,
275 		"Broadcom BCM57790 Gigabit Ethernet"},
276 	{ PCI_VENDOR_SCHNEIDERKOCH, PCI_PRODUCT_SCHNEIDERKOCH_SK_9DX1,
277 		"SysKonnect Gigabit Ethernet" },
278 
279 	{ 0, 0, NULL }
280 };
281 
282 #define BGE_IS_JUMBO_CAPABLE(sc)	((sc)->bge_flags & BGE_FLAG_JUMBO)
283 #define BGE_IS_5700_FAMILY(sc)		((sc)->bge_flags & BGE_FLAG_5700_FAMILY)
284 #define BGE_IS_5705_PLUS(sc)		((sc)->bge_flags & BGE_FLAG_5705_PLUS)
285 #define BGE_IS_5714_FAMILY(sc)		((sc)->bge_flags & BGE_FLAG_5714_FAMILY)
286 #define BGE_IS_575X_PLUS(sc)		((sc)->bge_flags & BGE_FLAG_575X_PLUS)
287 #define BGE_IS_5755_PLUS(sc)		((sc)->bge_flags & BGE_FLAG_5755_PLUS)
288 #define BGE_IS_5788(sc)			((sc)->bge_flags & BGE_FLAG_5788)
289 
290 #define BGE_IS_CRIPPLED(sc)		\
291 	(BGE_IS_5788((sc)) || (sc)->bge_asicrev == BGE_ASICREV_BCM5700)
292 
293 typedef int	(*bge_eaddr_fcn_t)(struct bge_softc *, uint8_t[]);
294 
295 static int	bge_probe(device_t);
296 static int	bge_attach(device_t);
297 static int	bge_detach(device_t);
298 static void	bge_txeof(struct bge_softc *, uint16_t);
299 static void	bge_rxeof(struct bge_softc *, uint16_t);
300 
301 static void	bge_tick(void *);
302 static void	bge_stats_update(struct bge_softc *);
303 static void	bge_stats_update_regs(struct bge_softc *);
304 static struct mbuf *
305 		bge_defrag_shortdma(struct mbuf *);
306 static int	bge_encap(struct bge_softc *, struct mbuf **, uint32_t *);
307 static int	bge_setup_tso(struct bge_softc *, struct mbuf **,
308 		    uint16_t *, uint16_t *);
309 
310 #ifdef IFPOLL_ENABLE
311 static void	bge_npoll(struct ifnet *, struct ifpoll_info *);
312 static void	bge_npoll_compat(struct ifnet *, void *, int );
313 #endif
314 static void	bge_intr_crippled(void *);
315 static void	bge_intr_legacy(void *);
316 static void	bge_msi(void *);
317 static void	bge_msi_oneshot(void *);
318 static void	bge_intr(struct bge_softc *);
319 static void	bge_enable_intr(struct bge_softc *);
320 static void	bge_disable_intr(struct bge_softc *);
321 static void	bge_start(struct ifnet *);
322 static int	bge_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
323 static void	bge_init(void *);
324 static void	bge_stop(struct bge_softc *);
325 static void	bge_watchdog(struct ifnet *);
326 static void	bge_shutdown(device_t);
327 static int	bge_suspend(device_t);
328 static int	bge_resume(device_t);
329 static int	bge_ifmedia_upd(struct ifnet *);
330 static void	bge_ifmedia_sts(struct ifnet *, struct ifmediareq *);
331 
332 static uint8_t	bge_nvram_getbyte(struct bge_softc *, int, uint8_t *);
333 static int	bge_read_nvram(struct bge_softc *, caddr_t, int, int);
334 
335 static uint8_t	bge_eeprom_getbyte(struct bge_softc *, uint32_t, uint8_t *);
336 static int	bge_read_eeprom(struct bge_softc *, caddr_t, uint32_t, size_t);
337 
338 static void	bge_setmulti(struct bge_softc *);
339 static void	bge_setpromisc(struct bge_softc *);
340 static void	bge_enable_msi(struct bge_softc *sc);
341 
342 static int	bge_alloc_jumbo_mem(struct bge_softc *);
343 static void	bge_free_jumbo_mem(struct bge_softc *);
344 static struct bge_jslot
345 		*bge_jalloc(struct bge_softc *);
346 static void	bge_jfree(void *);
347 static void	bge_jref(void *);
348 static int	bge_newbuf_std(struct bge_softc *, int, int);
349 static int	bge_newbuf_jumbo(struct bge_softc *, int, int);
350 static void	bge_setup_rxdesc_std(struct bge_softc *, int);
351 static void	bge_setup_rxdesc_jumbo(struct bge_softc *, int);
352 static int	bge_init_rx_ring_std(struct bge_softc *);
353 static void	bge_free_rx_ring_std(struct bge_softc *);
354 static int	bge_init_rx_ring_jumbo(struct bge_softc *);
355 static void	bge_free_rx_ring_jumbo(struct bge_softc *);
356 static void	bge_free_tx_ring(struct bge_softc *);
357 static int	bge_init_tx_ring(struct bge_softc *);
358 
359 static int	bge_chipinit(struct bge_softc *);
360 static int	bge_blockinit(struct bge_softc *);
361 static void	bge_stop_block(struct bge_softc *, bus_size_t, uint32_t);
362 
363 static uint32_t	bge_readmem_ind(struct bge_softc *, uint32_t);
364 static void	bge_writemem_ind(struct bge_softc *, uint32_t, uint32_t);
365 #ifdef notdef
366 static uint32_t	bge_readreg_ind(struct bge_softc *, uint32_t);
367 #endif
368 static void	bge_writereg_ind(struct bge_softc *, uint32_t, uint32_t);
369 static void	bge_writemem_direct(struct bge_softc *, uint32_t, uint32_t);
370 static void	bge_writembx(struct bge_softc *, int, int);
371 
372 static int	bge_miibus_readreg(device_t, int, int);
373 static int	bge_miibus_writereg(device_t, int, int, int);
374 static void	bge_miibus_statchg(device_t);
375 static void	bge_bcm5700_link_upd(struct bge_softc *, uint32_t);
376 static void	bge_tbi_link_upd(struct bge_softc *, uint32_t);
377 static void	bge_copper_link_upd(struct bge_softc *, uint32_t);
378 static void	bge_autopoll_link_upd(struct bge_softc *, uint32_t);
379 static void	bge_link_poll(struct bge_softc *);
380 
381 static void	bge_reset(struct bge_softc *);
382 
383 static int	bge_dma_alloc(struct bge_softc *);
384 static void	bge_dma_free(struct bge_softc *);
385 static int	bge_dma_block_alloc(struct bge_softc *, bus_size_t,
386 				    bus_dma_tag_t *, bus_dmamap_t *,
387 				    void **, bus_addr_t *);
388 static void	bge_dma_block_free(bus_dma_tag_t, bus_dmamap_t, void *);
389 
390 static int	bge_get_eaddr_mem(struct bge_softc *, uint8_t[]);
391 static int	bge_get_eaddr_nvram(struct bge_softc *, uint8_t[]);
392 static int	bge_get_eaddr_eeprom(struct bge_softc *, uint8_t[]);
393 static int	bge_get_eaddr(struct bge_softc *, uint8_t[]);
394 
395 static void	bge_coal_change(struct bge_softc *);
396 static int	bge_sysctl_rx_coal_ticks(SYSCTL_HANDLER_ARGS);
397 static int	bge_sysctl_tx_coal_ticks(SYSCTL_HANDLER_ARGS);
398 static int	bge_sysctl_rx_coal_bds(SYSCTL_HANDLER_ARGS);
399 static int	bge_sysctl_tx_coal_bds(SYSCTL_HANDLER_ARGS);
400 static int	bge_sysctl_rx_coal_ticks_int(SYSCTL_HANDLER_ARGS);
401 static int	bge_sysctl_tx_coal_ticks_int(SYSCTL_HANDLER_ARGS);
402 static int	bge_sysctl_rx_coal_bds_int(SYSCTL_HANDLER_ARGS);
403 static int	bge_sysctl_tx_coal_bds_int(SYSCTL_HANDLER_ARGS);
404 static int	bge_sysctl_coal_chg(SYSCTL_HANDLER_ARGS, uint32_t *,
405 		    int, int, uint32_t);
406 
407 /*
408  * Set following tunable to 1 for some IBM blade servers with the DNLK
409  * switch module. Auto negotiation is broken for those configurations.
410  */
411 static int	bge_fake_autoneg = 0;
412 TUNABLE_INT("hw.bge.fake_autoneg", &bge_fake_autoneg);
413 
414 static int	bge_msi_enable = 1;
415 TUNABLE_INT("hw.bge.msi.enable", &bge_msi_enable);
416 
417 #if !defined(KTR_IF_BGE)
418 #define KTR_IF_BGE	KTR_ALL
419 #endif
420 KTR_INFO_MASTER(if_bge);
421 KTR_INFO(KTR_IF_BGE, if_bge, intr, 0, "intr");
422 KTR_INFO(KTR_IF_BGE, if_bge, rx_pkt, 1, "rx_pkt");
423 KTR_INFO(KTR_IF_BGE, if_bge, tx_pkt, 2, "tx_pkt");
424 #define logif(name)	KTR_LOG(if_bge_ ## name)
425 
426 static device_method_t bge_methods[] = {
427 	/* Device interface */
428 	DEVMETHOD(device_probe,		bge_probe),
429 	DEVMETHOD(device_attach,	bge_attach),
430 	DEVMETHOD(device_detach,	bge_detach),
431 	DEVMETHOD(device_shutdown,	bge_shutdown),
432 	DEVMETHOD(device_suspend,	bge_suspend),
433 	DEVMETHOD(device_resume,	bge_resume),
434 
435 	/* bus interface */
436 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
437 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
438 
439 	/* MII interface */
440 	DEVMETHOD(miibus_readreg,	bge_miibus_readreg),
441 	DEVMETHOD(miibus_writereg,	bge_miibus_writereg),
442 	DEVMETHOD(miibus_statchg,	bge_miibus_statchg),
443 
444 	{ 0, 0 }
445 };
446 
447 static DEFINE_CLASS_0(bge, bge_driver, bge_methods, sizeof(struct bge_softc));
448 static devclass_t bge_devclass;
449 
450 DECLARE_DUMMY_MODULE(if_bge);
451 DRIVER_MODULE(if_bge, pci, bge_driver, bge_devclass, NULL, NULL);
452 DRIVER_MODULE(miibus, bge, miibus_driver, miibus_devclass, NULL, NULL);
453 
454 static uint32_t
455 bge_readmem_ind(struct bge_softc *sc, uint32_t off)
456 {
457 	device_t dev = sc->bge_dev;
458 	uint32_t val;
459 
460 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
461 	    off >= BGE_STATS_BLOCK && off < BGE_SEND_RING_1_TO_4)
462 		return 0;
463 
464 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4);
465 	val = pci_read_config(dev, BGE_PCI_MEMWIN_DATA, 4);
466 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4);
467 	return (val);
468 }
469 
470 static void
471 bge_writemem_ind(struct bge_softc *sc, uint32_t off, uint32_t val)
472 {
473 	device_t dev = sc->bge_dev;
474 
475 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
476 	    off >= BGE_STATS_BLOCK && off < BGE_SEND_RING_1_TO_4)
477 		return;
478 
479 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4);
480 	pci_write_config(dev, BGE_PCI_MEMWIN_DATA, val, 4);
481 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4);
482 }
483 
484 #ifdef notdef
485 static uint32_t
486 bge_readreg_ind(struct bge_softc *sc, uin32_t off)
487 {
488 	device_t dev = sc->bge_dev;
489 
490 	pci_write_config(dev, BGE_PCI_REG_BASEADDR, off, 4);
491 	return(pci_read_config(dev, BGE_PCI_REG_DATA, 4));
492 }
493 #endif
494 
495 static void
496 bge_writereg_ind(struct bge_softc *sc, uint32_t off, uint32_t val)
497 {
498 	device_t dev = sc->bge_dev;
499 
500 	pci_write_config(dev, BGE_PCI_REG_BASEADDR, off, 4);
501 	pci_write_config(dev, BGE_PCI_REG_DATA, val, 4);
502 }
503 
504 static void
505 bge_writemem_direct(struct bge_softc *sc, uint32_t off, uint32_t val)
506 {
507 	CSR_WRITE_4(sc, off, val);
508 }
509 
510 static void
511 bge_writembx(struct bge_softc *sc, int off, int val)
512 {
513 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
514 		off += BGE_LPMBX_IRQ0_HI - BGE_MBX_IRQ0_HI;
515 
516 	CSR_WRITE_4(sc, off, val);
517 	if (sc->bge_mbox_reorder)
518 		CSR_READ_4(sc, off);
519 }
520 
521 static uint8_t
522 bge_nvram_getbyte(struct bge_softc *sc, int addr, uint8_t *dest)
523 {
524 	uint32_t access, byte = 0;
525 	int i;
526 
527 	/* Lock. */
528 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_SET1);
529 	for (i = 0; i < 8000; i++) {
530 		if (CSR_READ_4(sc, BGE_NVRAM_SWARB) & BGE_NVRAMSWARB_GNT1)
531 			break;
532 		DELAY(20);
533 	}
534 	if (i == 8000)
535 		return (1);
536 
537 	/* Enable access. */
538 	access = CSR_READ_4(sc, BGE_NVRAM_ACCESS);
539 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access | BGE_NVRAMACC_ENABLE);
540 
541 	CSR_WRITE_4(sc, BGE_NVRAM_ADDR, addr & 0xfffffffc);
542 	CSR_WRITE_4(sc, BGE_NVRAM_CMD, BGE_NVRAM_READCMD);
543 	for (i = 0; i < BGE_TIMEOUT * 10; i++) {
544 		DELAY(10);
545 		if (CSR_READ_4(sc, BGE_NVRAM_CMD) & BGE_NVRAMCMD_DONE) {
546 			DELAY(10);
547 			break;
548 		}
549 	}
550 
551 	if (i == BGE_TIMEOUT * 10) {
552 		if_printf(&sc->arpcom.ac_if, "nvram read timed out\n");
553 		return (1);
554 	}
555 
556 	/* Get result. */
557 	byte = CSR_READ_4(sc, BGE_NVRAM_RDDATA);
558 
559 	*dest = (bswap32(byte) >> ((addr % 4) * 8)) & 0xFF;
560 
561 	/* Disable access. */
562 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access);
563 
564 	/* Unlock. */
565 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_CLR1);
566 	CSR_READ_4(sc, BGE_NVRAM_SWARB);
567 
568 	return (0);
569 }
570 
571 /*
572  * Read a sequence of bytes from NVRAM.
573  */
574 static int
575 bge_read_nvram(struct bge_softc *sc, caddr_t dest, int off, int cnt)
576 {
577 	int err = 0, i;
578 	uint8_t byte = 0;
579 
580 	if (sc->bge_asicrev != BGE_ASICREV_BCM5906)
581 		return (1);
582 
583 	for (i = 0; i < cnt; i++) {
584 		err = bge_nvram_getbyte(sc, off + i, &byte);
585 		if (err)
586 			break;
587 		*(dest + i) = byte;
588 	}
589 
590 	return (err ? 1 : 0);
591 }
592 
593 /*
594  * Read a byte of data stored in the EEPROM at address 'addr.' The
595  * BCM570x supports both the traditional bitbang interface and an
596  * auto access interface for reading the EEPROM. We use the auto
597  * access method.
598  */
599 static uint8_t
600 bge_eeprom_getbyte(struct bge_softc *sc, uint32_t addr, uint8_t *dest)
601 {
602 	int i;
603 	uint32_t byte = 0;
604 
605 	/*
606 	 * Enable use of auto EEPROM access so we can avoid
607 	 * having to use the bitbang method.
608 	 */
609 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_AUTO_EEPROM);
610 
611 	/* Reset the EEPROM, load the clock period. */
612 	CSR_WRITE_4(sc, BGE_EE_ADDR,
613 	    BGE_EEADDR_RESET|BGE_EEHALFCLK(BGE_HALFCLK_384SCL));
614 	DELAY(20);
615 
616 	/* Issue the read EEPROM command. */
617 	CSR_WRITE_4(sc, BGE_EE_ADDR, BGE_EE_READCMD | addr);
618 
619 	/* Wait for completion */
620 	for(i = 0; i < BGE_TIMEOUT * 10; i++) {
621 		DELAY(10);
622 		if (CSR_READ_4(sc, BGE_EE_ADDR) & BGE_EEADDR_DONE)
623 			break;
624 	}
625 
626 	if (i == BGE_TIMEOUT) {
627 		if_printf(&sc->arpcom.ac_if, "eeprom read timed out\n");
628 		return(1);
629 	}
630 
631 	/* Get result. */
632 	byte = CSR_READ_4(sc, BGE_EE_DATA);
633 
634         *dest = (byte >> ((addr % 4) * 8)) & 0xFF;
635 
636 	return(0);
637 }
638 
639 /*
640  * Read a sequence of bytes from the EEPROM.
641  */
642 static int
643 bge_read_eeprom(struct bge_softc *sc, caddr_t dest, uint32_t off, size_t len)
644 {
645 	size_t i;
646 	int err;
647 	uint8_t byte;
648 
649 	for (byte = 0, err = 0, i = 0; i < len; i++) {
650 		err = bge_eeprom_getbyte(sc, off + i, &byte);
651 		if (err)
652 			break;
653 		*(dest + i) = byte;
654 	}
655 
656 	return(err ? 1 : 0);
657 }
658 
659 static int
660 bge_miibus_readreg(device_t dev, int phy, int reg)
661 {
662 	struct bge_softc *sc = device_get_softc(dev);
663 	uint32_t val;
664 	int i;
665 
666 	KASSERT(phy == sc->bge_phyno,
667 	    ("invalid phyno %d, should be %d", phy, sc->bge_phyno));
668 
669 	/* Clear the autopoll bit if set, otherwise may trigger PCI errors. */
670 	if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
671 		CSR_WRITE_4(sc, BGE_MI_MODE,
672 		    sc->bge_mi_mode & ~BGE_MIMODE_AUTOPOLL);
673 		DELAY(80);
674 	}
675 
676 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_READ | BGE_MICOMM_BUSY |
677 	    BGE_MIPHY(phy) | BGE_MIREG(reg));
678 
679 	/* Poll for the PHY register access to complete. */
680 	for (i = 0; i < BGE_TIMEOUT; i++) {
681 		DELAY(10);
682 		val = CSR_READ_4(sc, BGE_MI_COMM);
683 		if ((val & BGE_MICOMM_BUSY) == 0) {
684 			DELAY(5);
685 			val = CSR_READ_4(sc, BGE_MI_COMM);
686 			break;
687 		}
688 	}
689 	if (i == BGE_TIMEOUT) {
690 		if_printf(&sc->arpcom.ac_if, "PHY read timed out "
691 		    "(phy %d, reg %d, val 0x%08x)\n", phy, reg, val);
692 		val = 0;
693 	}
694 
695 	/* Restore the autopoll bit if necessary. */
696 	if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
697 		CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode);
698 		DELAY(80);
699 	}
700 
701 	if (val & BGE_MICOMM_READFAIL)
702 		return 0;
703 
704 	return (val & 0xFFFF);
705 }
706 
707 static int
708 bge_miibus_writereg(device_t dev, int phy, int reg, int val)
709 {
710 	struct bge_softc *sc = device_get_softc(dev);
711 	int i;
712 
713 	KASSERT(phy == sc->bge_phyno,
714 	    ("invalid phyno %d, should be %d", phy, sc->bge_phyno));
715 
716 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
717 	    (reg == BRGPHY_MII_1000CTL || reg == BRGPHY_MII_AUXCTL))
718 	       return 0;
719 
720 	/* Clear the autopoll bit if set, otherwise may trigger PCI errors. */
721 	if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
722 		CSR_WRITE_4(sc, BGE_MI_MODE,
723 		    sc->bge_mi_mode & ~BGE_MIMODE_AUTOPOLL);
724 		DELAY(80);
725 	}
726 
727 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_WRITE | BGE_MICOMM_BUSY |
728 	    BGE_MIPHY(phy) | BGE_MIREG(reg) | val);
729 
730 	for (i = 0; i < BGE_TIMEOUT; i++) {
731 		DELAY(10);
732 		if (!(CSR_READ_4(sc, BGE_MI_COMM) & BGE_MICOMM_BUSY)) {
733 			DELAY(5);
734 			CSR_READ_4(sc, BGE_MI_COMM); /* dummy read */
735 			break;
736 		}
737 	}
738 	if (i == BGE_TIMEOUT) {
739 		if_printf(&sc->arpcom.ac_if, "PHY write timed out "
740 		    "(phy %d, reg %d, val %d)\n", phy, reg, val);
741 	}
742 
743 	/* Restore the autopoll bit if necessary. */
744 	if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
745 		CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode);
746 		DELAY(80);
747 	}
748 
749 	return 0;
750 }
751 
752 static void
753 bge_miibus_statchg(device_t dev)
754 {
755 	struct bge_softc *sc;
756 	struct mii_data *mii;
757 
758 	sc = device_get_softc(dev);
759 	mii = device_get_softc(sc->bge_miibus);
760 
761 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
762 	    (IFM_ACTIVE | IFM_AVALID)) {
763 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
764 		case IFM_10_T:
765 		case IFM_100_TX:
766 			sc->bge_link = 1;
767 			break;
768 		case IFM_1000_T:
769 		case IFM_1000_SX:
770 		case IFM_2500_SX:
771 			if (sc->bge_asicrev != BGE_ASICREV_BCM5906)
772 				sc->bge_link = 1;
773 			else
774 				sc->bge_link = 0;
775 			break;
776 		default:
777 			sc->bge_link = 0;
778 			break;
779 		}
780 	} else {
781 		sc->bge_link = 0;
782 	}
783 	if (sc->bge_link == 0)
784 		return;
785 
786 	BGE_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_PORTMODE);
787 	if (IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_T ||
788 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_SX) {
789 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_GMII);
790 	} else {
791 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_MII);
792 	}
793 
794 	if ((mii->mii_media_active & IFM_GMASK) == IFM_FDX) {
795 		BGE_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
796 	} else {
797 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
798 	}
799 }
800 
801 /*
802  * Memory management for jumbo frames.
803  */
804 static int
805 bge_alloc_jumbo_mem(struct bge_softc *sc)
806 {
807 	struct ifnet *ifp = &sc->arpcom.ac_if;
808 	struct bge_jslot *entry;
809 	uint8_t *ptr;
810 	bus_addr_t paddr;
811 	int i, error;
812 
813 	/*
814 	 * Create tag for jumbo mbufs.
815 	 * This is really a bit of a kludge. We allocate a special
816 	 * jumbo buffer pool which (thanks to the way our DMA
817 	 * memory allocation works) will consist of contiguous
818 	 * pages. This means that even though a jumbo buffer might
819 	 * be larger than a page size, we don't really need to
820 	 * map it into more than one DMA segment. However, the
821 	 * default mbuf tag will result in multi-segment mappings,
822 	 * so we have to create a special jumbo mbuf tag that
823 	 * lets us get away with mapping the jumbo buffers as
824 	 * a single segment. I think eventually the driver should
825 	 * be changed so that it uses ordinary mbufs and cluster
826 	 * buffers, i.e. jumbo frames can span multiple DMA
827 	 * descriptors. But that's a project for another day.
828 	 */
829 
830 	/*
831 	 * Create DMA stuffs for jumbo RX ring.
832 	 */
833 	error = bge_dma_block_alloc(sc, BGE_JUMBO_RX_RING_SZ,
834 				    &sc->bge_cdata.bge_rx_jumbo_ring_tag,
835 				    &sc->bge_cdata.bge_rx_jumbo_ring_map,
836 				    (void *)&sc->bge_ldata.bge_rx_jumbo_ring,
837 				    &sc->bge_ldata.bge_rx_jumbo_ring_paddr);
838 	if (error) {
839 		if_printf(ifp, "could not create jumbo RX ring\n");
840 		return error;
841 	}
842 
843 	/*
844 	 * Create DMA stuffs for jumbo buffer block.
845 	 */
846 	error = bge_dma_block_alloc(sc, BGE_JMEM,
847 				    &sc->bge_cdata.bge_jumbo_tag,
848 				    &sc->bge_cdata.bge_jumbo_map,
849 				    (void **)&sc->bge_ldata.bge_jumbo_buf,
850 				    &paddr);
851 	if (error) {
852 		if_printf(ifp, "could not create jumbo buffer\n");
853 		return error;
854 	}
855 
856 	SLIST_INIT(&sc->bge_jfree_listhead);
857 
858 	/*
859 	 * Now divide it up into 9K pieces and save the addresses
860 	 * in an array. Note that we play an evil trick here by using
861 	 * the first few bytes in the buffer to hold the the address
862 	 * of the softc structure for this interface. This is because
863 	 * bge_jfree() needs it, but it is called by the mbuf management
864 	 * code which will not pass it to us explicitly.
865 	 */
866 	for (i = 0, ptr = sc->bge_ldata.bge_jumbo_buf; i < BGE_JSLOTS; i++) {
867 		entry = &sc->bge_cdata.bge_jslots[i];
868 		entry->bge_sc = sc;
869 		entry->bge_buf = ptr;
870 		entry->bge_paddr = paddr;
871 		entry->bge_inuse = 0;
872 		entry->bge_slot = i;
873 		SLIST_INSERT_HEAD(&sc->bge_jfree_listhead, entry, jslot_link);
874 
875 		ptr += BGE_JLEN;
876 		paddr += BGE_JLEN;
877 	}
878 	return 0;
879 }
880 
881 static void
882 bge_free_jumbo_mem(struct bge_softc *sc)
883 {
884 	/* Destroy jumbo RX ring. */
885 	bge_dma_block_free(sc->bge_cdata.bge_rx_jumbo_ring_tag,
886 			   sc->bge_cdata.bge_rx_jumbo_ring_map,
887 			   sc->bge_ldata.bge_rx_jumbo_ring);
888 
889 	/* Destroy jumbo buffer block. */
890 	bge_dma_block_free(sc->bge_cdata.bge_jumbo_tag,
891 			   sc->bge_cdata.bge_jumbo_map,
892 			   sc->bge_ldata.bge_jumbo_buf);
893 }
894 
895 /*
896  * Allocate a jumbo buffer.
897  */
898 static struct bge_jslot *
899 bge_jalloc(struct bge_softc *sc)
900 {
901 	struct bge_jslot *entry;
902 
903 	lwkt_serialize_enter(&sc->bge_jslot_serializer);
904 	entry = SLIST_FIRST(&sc->bge_jfree_listhead);
905 	if (entry) {
906 		SLIST_REMOVE_HEAD(&sc->bge_jfree_listhead, jslot_link);
907 		entry->bge_inuse = 1;
908 	} else {
909 		if_printf(&sc->arpcom.ac_if, "no free jumbo buffers\n");
910 	}
911 	lwkt_serialize_exit(&sc->bge_jslot_serializer);
912 	return(entry);
913 }
914 
915 /*
916  * Adjust usage count on a jumbo buffer.
917  */
918 static void
919 bge_jref(void *arg)
920 {
921 	struct bge_jslot *entry = (struct bge_jslot *)arg;
922 	struct bge_softc *sc = entry->bge_sc;
923 
924 	if (sc == NULL)
925 		panic("bge_jref: can't find softc pointer!");
926 
927 	if (&sc->bge_cdata.bge_jslots[entry->bge_slot] != entry) {
928 		panic("bge_jref: asked to reference buffer "
929 		    "that we don't manage!");
930 	} else if (entry->bge_inuse == 0) {
931 		panic("bge_jref: buffer already free!");
932 	} else {
933 		atomic_add_int(&entry->bge_inuse, 1);
934 	}
935 }
936 
937 /*
938  * Release a jumbo buffer.
939  */
940 static void
941 bge_jfree(void *arg)
942 {
943 	struct bge_jslot *entry = (struct bge_jslot *)arg;
944 	struct bge_softc *sc = entry->bge_sc;
945 
946 	if (sc == NULL)
947 		panic("bge_jfree: can't find softc pointer!");
948 
949 	if (&sc->bge_cdata.bge_jslots[entry->bge_slot] != entry) {
950 		panic("bge_jfree: asked to free buffer that we don't manage!");
951 	} else if (entry->bge_inuse == 0) {
952 		panic("bge_jfree: buffer already free!");
953 	} else {
954 		/*
955 		 * Possible MP race to 0, use the serializer.  The atomic insn
956 		 * is still needed for races against bge_jref().
957 		 */
958 		lwkt_serialize_enter(&sc->bge_jslot_serializer);
959 		atomic_subtract_int(&entry->bge_inuse, 1);
960 		if (entry->bge_inuse == 0) {
961 			SLIST_INSERT_HEAD(&sc->bge_jfree_listhead,
962 					  entry, jslot_link);
963 		}
964 		lwkt_serialize_exit(&sc->bge_jslot_serializer);
965 	}
966 }
967 
968 
969 /*
970  * Intialize a standard receive ring descriptor.
971  */
972 static int
973 bge_newbuf_std(struct bge_softc *sc, int i, int init)
974 {
975 	struct mbuf *m_new = NULL;
976 	bus_dma_segment_t seg;
977 	bus_dmamap_t map;
978 	int error, nsegs;
979 
980 	m_new = m_getcl(init ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
981 	if (m_new == NULL)
982 		return ENOBUFS;
983 	m_new->m_len = m_new->m_pkthdr.len = MCLBYTES;
984 
985 	if ((sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) == 0)
986 		m_adj(m_new, ETHER_ALIGN);
987 
988 	error = bus_dmamap_load_mbuf_segment(sc->bge_cdata.bge_rx_mtag,
989 			sc->bge_cdata.bge_rx_tmpmap, m_new,
990 			&seg, 1, &nsegs, BUS_DMA_NOWAIT);
991 	if (error) {
992 		m_freem(m_new);
993 		return error;
994 	}
995 
996 	if (!init) {
997 		bus_dmamap_sync(sc->bge_cdata.bge_rx_mtag,
998 				sc->bge_cdata.bge_rx_std_dmamap[i],
999 				BUS_DMASYNC_POSTREAD);
1000 		bus_dmamap_unload(sc->bge_cdata.bge_rx_mtag,
1001 			sc->bge_cdata.bge_rx_std_dmamap[i]);
1002 	}
1003 
1004 	map = sc->bge_cdata.bge_rx_tmpmap;
1005 	sc->bge_cdata.bge_rx_tmpmap = sc->bge_cdata.bge_rx_std_dmamap[i];
1006 	sc->bge_cdata.bge_rx_std_dmamap[i] = map;
1007 
1008 	sc->bge_cdata.bge_rx_std_chain[i].bge_mbuf = m_new;
1009 	sc->bge_cdata.bge_rx_std_chain[i].bge_paddr = seg.ds_addr;
1010 
1011 	bge_setup_rxdesc_std(sc, i);
1012 	return 0;
1013 }
1014 
1015 static void
1016 bge_setup_rxdesc_std(struct bge_softc *sc, int i)
1017 {
1018 	struct bge_rxchain *rc;
1019 	struct bge_rx_bd *r;
1020 
1021 	rc = &sc->bge_cdata.bge_rx_std_chain[i];
1022 	r = &sc->bge_ldata.bge_rx_std_ring[i];
1023 
1024 	r->bge_addr.bge_addr_lo = BGE_ADDR_LO(rc->bge_paddr);
1025 	r->bge_addr.bge_addr_hi = BGE_ADDR_HI(rc->bge_paddr);
1026 	r->bge_len = rc->bge_mbuf->m_len;
1027 	r->bge_idx = i;
1028 	r->bge_flags = BGE_RXBDFLAG_END;
1029 }
1030 
1031 /*
1032  * Initialize a jumbo receive ring descriptor. This allocates
1033  * a jumbo buffer from the pool managed internally by the driver.
1034  */
1035 static int
1036 bge_newbuf_jumbo(struct bge_softc *sc, int i, int init)
1037 {
1038 	struct mbuf *m_new = NULL;
1039 	struct bge_jslot *buf;
1040 	bus_addr_t paddr;
1041 
1042 	/* Allocate the mbuf. */
1043 	MGETHDR(m_new, init ? MB_WAIT : MB_DONTWAIT, MT_DATA);
1044 	if (m_new == NULL)
1045 		return ENOBUFS;
1046 
1047 	/* Allocate the jumbo buffer */
1048 	buf = bge_jalloc(sc);
1049 	if (buf == NULL) {
1050 		m_freem(m_new);
1051 		return ENOBUFS;
1052 	}
1053 
1054 	/* Attach the buffer to the mbuf. */
1055 	m_new->m_ext.ext_arg = buf;
1056 	m_new->m_ext.ext_buf = buf->bge_buf;
1057 	m_new->m_ext.ext_free = bge_jfree;
1058 	m_new->m_ext.ext_ref = bge_jref;
1059 	m_new->m_ext.ext_size = BGE_JUMBO_FRAMELEN;
1060 
1061 	m_new->m_flags |= M_EXT;
1062 
1063 	m_new->m_data = m_new->m_ext.ext_buf;
1064 	m_new->m_len = m_new->m_pkthdr.len = m_new->m_ext.ext_size;
1065 
1066 	paddr = buf->bge_paddr;
1067 	if ((sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) == 0) {
1068 		m_adj(m_new, ETHER_ALIGN);
1069 		paddr += ETHER_ALIGN;
1070 	}
1071 
1072 	/* Save necessary information */
1073 	sc->bge_cdata.bge_rx_jumbo_chain[i].bge_mbuf = m_new;
1074 	sc->bge_cdata.bge_rx_jumbo_chain[i].bge_paddr = paddr;
1075 
1076 	/* Set up the descriptor. */
1077 	bge_setup_rxdesc_jumbo(sc, i);
1078 	return 0;
1079 }
1080 
1081 static void
1082 bge_setup_rxdesc_jumbo(struct bge_softc *sc, int i)
1083 {
1084 	struct bge_rx_bd *r;
1085 	struct bge_rxchain *rc;
1086 
1087 	r = &sc->bge_ldata.bge_rx_jumbo_ring[i];
1088 	rc = &sc->bge_cdata.bge_rx_jumbo_chain[i];
1089 
1090 	r->bge_addr.bge_addr_lo = BGE_ADDR_LO(rc->bge_paddr);
1091 	r->bge_addr.bge_addr_hi = BGE_ADDR_HI(rc->bge_paddr);
1092 	r->bge_len = rc->bge_mbuf->m_len;
1093 	r->bge_idx = i;
1094 	r->bge_flags = BGE_RXBDFLAG_END|BGE_RXBDFLAG_JUMBO_RING;
1095 }
1096 
1097 static int
1098 bge_init_rx_ring_std(struct bge_softc *sc)
1099 {
1100 	int i, error;
1101 
1102 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1103 		error = bge_newbuf_std(sc, i, 1);
1104 		if (error)
1105 			return error;
1106 	};
1107 
1108 	sc->bge_std = BGE_STD_RX_RING_CNT - 1;
1109 	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bge_std);
1110 
1111 	return(0);
1112 }
1113 
1114 static void
1115 bge_free_rx_ring_std(struct bge_softc *sc)
1116 {
1117 	int i;
1118 
1119 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1120 		struct bge_rxchain *rc = &sc->bge_cdata.bge_rx_std_chain[i];
1121 
1122 		if (rc->bge_mbuf != NULL) {
1123 			bus_dmamap_unload(sc->bge_cdata.bge_rx_mtag,
1124 					  sc->bge_cdata.bge_rx_std_dmamap[i]);
1125 			m_freem(rc->bge_mbuf);
1126 			rc->bge_mbuf = NULL;
1127 		}
1128 		bzero(&sc->bge_ldata.bge_rx_std_ring[i],
1129 		    sizeof(struct bge_rx_bd));
1130 	}
1131 }
1132 
1133 static int
1134 bge_init_rx_ring_jumbo(struct bge_softc *sc)
1135 {
1136 	struct bge_rcb *rcb;
1137 	int i, error;
1138 
1139 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1140 		error = bge_newbuf_jumbo(sc, i, 1);
1141 		if (error)
1142 			return error;
1143 	};
1144 
1145 	sc->bge_jumbo = BGE_JUMBO_RX_RING_CNT - 1;
1146 
1147 	rcb = &sc->bge_ldata.bge_info.bge_jumbo_rx_rcb;
1148 	rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(0, 0);
1149 	CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1150 
1151 	bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
1152 
1153 	return(0);
1154 }
1155 
1156 static void
1157 bge_free_rx_ring_jumbo(struct bge_softc *sc)
1158 {
1159 	int i;
1160 
1161 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1162 		struct bge_rxchain *rc = &sc->bge_cdata.bge_rx_jumbo_chain[i];
1163 
1164 		if (rc->bge_mbuf != NULL) {
1165 			m_freem(rc->bge_mbuf);
1166 			rc->bge_mbuf = NULL;
1167 		}
1168 		bzero(&sc->bge_ldata.bge_rx_jumbo_ring[i],
1169 		    sizeof(struct bge_rx_bd));
1170 	}
1171 }
1172 
1173 static void
1174 bge_free_tx_ring(struct bge_softc *sc)
1175 {
1176 	int i;
1177 
1178 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
1179 		if (sc->bge_cdata.bge_tx_chain[i] != NULL) {
1180 			bus_dmamap_unload(sc->bge_cdata.bge_tx_mtag,
1181 					  sc->bge_cdata.bge_tx_dmamap[i]);
1182 			m_freem(sc->bge_cdata.bge_tx_chain[i]);
1183 			sc->bge_cdata.bge_tx_chain[i] = NULL;
1184 		}
1185 		bzero(&sc->bge_ldata.bge_tx_ring[i],
1186 		    sizeof(struct bge_tx_bd));
1187 	}
1188 }
1189 
1190 static int
1191 bge_init_tx_ring(struct bge_softc *sc)
1192 {
1193 	sc->bge_txcnt = 0;
1194 	sc->bge_tx_saved_considx = 0;
1195 	sc->bge_tx_prodidx = 0;
1196 
1197 	/* Initialize transmit producer index for host-memory send ring. */
1198 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1199 
1200 	/* 5700 b2 errata */
1201 	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
1202 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1203 
1204 	bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1205 	/* 5700 b2 errata */
1206 	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
1207 		bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1208 
1209 	return(0);
1210 }
1211 
1212 static void
1213 bge_setmulti(struct bge_softc *sc)
1214 {
1215 	struct ifnet *ifp;
1216 	struct ifmultiaddr *ifma;
1217 	uint32_t hashes[4] = { 0, 0, 0, 0 };
1218 	int h, i;
1219 
1220 	ifp = &sc->arpcom.ac_if;
1221 
1222 	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
1223 		for (i = 0; i < 4; i++)
1224 			CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0xFFFFFFFF);
1225 		return;
1226 	}
1227 
1228 	/* First, zot all the existing filters. */
1229 	for (i = 0; i < 4; i++)
1230 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0);
1231 
1232 	/* Now program new ones. */
1233 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1234 		if (ifma->ifma_addr->sa_family != AF_LINK)
1235 			continue;
1236 		h = ether_crc32_le(
1237 		    LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1238 		    ETHER_ADDR_LEN) & 0x7f;
1239 		hashes[(h & 0x60) >> 5] |= 1 << (h & 0x1F);
1240 	}
1241 
1242 	for (i = 0; i < 4; i++)
1243 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), hashes[i]);
1244 }
1245 
1246 /*
1247  * Do endian, PCI and DMA initialization. Also check the on-board ROM
1248  * self-test results.
1249  */
1250 static int
1251 bge_chipinit(struct bge_softc *sc)
1252 {
1253 	int i;
1254 	uint32_t dma_rw_ctl;
1255 	uint16_t val;
1256 
1257 	/* Set endian type before we access any non-PCI registers. */
1258 	pci_write_config(sc->bge_dev, BGE_PCI_MISC_CTL,
1259 	    BGE_INIT | sc->bge_pci_miscctl, 4);
1260 
1261 	/* Clear the MAC control register */
1262 	CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
1263 
1264 	/*
1265 	 * Clear the MAC statistics block in the NIC's
1266 	 * internal memory.
1267 	 */
1268 	for (i = BGE_STATS_BLOCK;
1269 	    i < BGE_STATS_BLOCK_END + 1; i += sizeof(uint32_t))
1270 		BGE_MEMWIN_WRITE(sc, i, 0);
1271 
1272 	for (i = BGE_STATUS_BLOCK;
1273 	    i < BGE_STATUS_BLOCK_END + 1; i += sizeof(uint32_t))
1274 		BGE_MEMWIN_WRITE(sc, i, 0);
1275 
1276 	if (sc->bge_chiprev == BGE_CHIPREV_5704_BX) {
1277 		/*
1278 		 * Fix data corruption caused by non-qword write with WB.
1279 		 * Fix master abort in PCI mode.
1280 		 * Fix PCI latency timer.
1281 		 */
1282 		val = pci_read_config(sc->bge_dev, BGE_PCI_MSI_DATA + 2, 2);
1283 		val |= (1 << 10) | (1 << 12) | (1 << 13);
1284 		pci_write_config(sc->bge_dev, BGE_PCI_MSI_DATA + 2, val, 2);
1285 	}
1286 
1287 	/* Set up the PCI DMA control register. */
1288 	dma_rw_ctl = BGE_PCI_READ_CMD | BGE_PCI_WRITE_CMD;
1289 	if (sc->bge_flags & BGE_FLAG_PCIE) {
1290 		/* PCI-E bus */
1291 		/* DMA read watermark not used on PCI-E */
1292 		dma_rw_ctl |= (0x3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1293 	} else if (sc->bge_flags & BGE_FLAG_PCIX) {
1294 		/* PCI-X bus */
1295 		if (sc->bge_asicrev == BGE_ASICREV_BCM5780) {
1296 			dma_rw_ctl |= (0x4 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1297 			    (0x2 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1298 			dma_rw_ctl |= BGE_PCIDMARWCTL_ONEDMA_ATONCE_GLOBAL;
1299 		} else if (sc->bge_asicrev == BGE_ASICREV_BCM5714) {
1300 			dma_rw_ctl |= (0x4 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1301 			    (0x2 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1302 			dma_rw_ctl |= BGE_PCIDMARWCTL_ONEDMA_ATONCE_LOCAL;
1303 		} else if (sc->bge_asicrev == BGE_ASICREV_BCM5703 ||
1304 		    sc->bge_asicrev == BGE_ASICREV_BCM5704) {
1305 			uint32_t rd_wat = 0x7;
1306 			uint32_t clkctl;
1307 
1308 			clkctl = CSR_READ_4(sc, BGE_PCI_CLKCTL) & 0x1f;
1309 			if ((sc->bge_flags & BGE_FLAG_MAXADDR_40BIT) &&
1310 			    sc->bge_asicrev == BGE_ASICREV_BCM5704) {
1311 				dma_rw_ctl |=
1312 				    BGE_PCIDMARWCTL_ONEDMA_ATONCE_LOCAL;
1313 			} else if (clkctl == 0x6 || clkctl == 0x7) {
1314 				dma_rw_ctl |=
1315 				    BGE_PCIDMARWCTL_ONEDMA_ATONCE_GLOBAL;
1316 			}
1317 			if (sc->bge_asicrev == BGE_ASICREV_BCM5703)
1318 				rd_wat = 0x4;
1319 
1320 			dma_rw_ctl |= (rd_wat << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1321 			    (3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1322 			dma_rw_ctl |= BGE_PCIDMARWCTL_ASRT_ALL_BE;
1323 		} else {
1324 			dma_rw_ctl |= (0x3 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1325 			    (0x3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1326 			dma_rw_ctl |= 0xf;
1327 		}
1328 	} else {
1329 		/* Conventional PCI bus */
1330 		dma_rw_ctl |= (0x7 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1331 		    (0x7 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1332 		if (sc->bge_asicrev != BGE_ASICREV_BCM5705 &&
1333 		    sc->bge_asicrev != BGE_ASICREV_BCM5750)
1334 			dma_rw_ctl |= 0xf;
1335 	}
1336 
1337 	if (sc->bge_asicrev == BGE_ASICREV_BCM5703 ||
1338 	    sc->bge_asicrev == BGE_ASICREV_BCM5704) {
1339 		dma_rw_ctl &= ~BGE_PCIDMARWCTL_MINDMA;
1340 	} else if (sc->bge_asicrev == BGE_ASICREV_BCM5700 ||
1341 	    sc->bge_asicrev == BGE_ASICREV_BCM5701) {
1342 		dma_rw_ctl |= BGE_PCIDMARWCTL_USE_MRM |
1343 		    BGE_PCIDMARWCTL_ASRT_ALL_BE;
1344 	}
1345 	pci_write_config(sc->bge_dev, BGE_PCI_DMA_RW_CTL, dma_rw_ctl, 4);
1346 
1347 	/*
1348 	 * Set up general mode register.
1349 	 */
1350 	CSR_WRITE_4(sc, BGE_MODE_CTL, BGE_DMA_SWAP_OPTIONS|
1351 	    BGE_MODECTL_MAC_ATTN_INTR|BGE_MODECTL_HOST_SEND_BDS|
1352 	    BGE_MODECTL_TX_NO_PHDR_CSUM);
1353 
1354 	/*
1355 	 * BCM5701 B5 have a bug causing data corruption when using
1356 	 * 64-bit DMA reads, which can be terminated early and then
1357 	 * completed later as 32-bit accesses, in combination with
1358 	 * certain bridges.
1359 	 */
1360 	if (sc->bge_asicrev == BGE_ASICREV_BCM5701 &&
1361 	    sc->bge_chipid == BGE_CHIPID_BCM5701_B5)
1362 		BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_FORCE_PCI32);
1363 
1364 	/*
1365 	 * Disable memory write invalidate.  Apparently it is not supported
1366 	 * properly by these devices.  Also ensure that INTx isn't disabled,
1367 	 * as these chips need it even when using MSI.
1368 	 */
1369 	PCI_CLRBIT(sc->bge_dev, BGE_PCI_CMD,
1370 	    (PCIM_CMD_MWRICEN | PCIM_CMD_INTxDIS), 4);
1371 
1372 	/* Set the timer prescaler (always 66Mhz) */
1373 	CSR_WRITE_4(sc, BGE_MISC_CFG, 65 << 1/*BGE_32BITTIME_66MHZ*/);
1374 
1375 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
1376 		DELAY(40);	/* XXX */
1377 
1378 		/* Put PHY into ready state */
1379 		BGE_CLRBIT(sc, BGE_MISC_CFG, BGE_MISCCFG_EPHY_IDDQ);
1380 		CSR_READ_4(sc, BGE_MISC_CFG); /* Flush */
1381 		DELAY(40);
1382 	}
1383 
1384 	return(0);
1385 }
1386 
1387 static int
1388 bge_blockinit(struct bge_softc *sc)
1389 {
1390 	struct bge_rcb *rcb;
1391 	bus_size_t vrcb;
1392 	bge_hostaddr taddr;
1393 	uint32_t val;
1394 	int i, limit;
1395 
1396 	/*
1397 	 * Initialize the memory window pointer register so that
1398 	 * we can access the first 32K of internal NIC RAM. This will
1399 	 * allow us to set up the TX send ring RCBs and the RX return
1400 	 * ring RCBs, plus other things which live in NIC memory.
1401 	 */
1402 	CSR_WRITE_4(sc, BGE_PCI_MEMWIN_BASEADDR, 0);
1403 
1404 	/* Note: the BCM5704 has a smaller mbuf space than other chips. */
1405 
1406 	if (!BGE_IS_5705_PLUS(sc)) {
1407 		/* Configure mbuf memory pool */
1408 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_BASEADDR, BGE_BUFFPOOL_1);
1409 		if (sc->bge_asicrev == BGE_ASICREV_BCM5704)
1410 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x10000);
1411 		else
1412 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x18000);
1413 
1414 		/* Configure DMA resource pool */
1415 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_BASEADDR,
1416 		    BGE_DMA_DESCRIPTORS);
1417 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LEN, 0x2000);
1418 	}
1419 
1420 	/* Configure mbuf pool watermarks */
1421 	if (!BGE_IS_5705_PLUS(sc)) {
1422 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x50);
1423 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x20);
1424 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1425 	} else if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
1426 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1427 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x04);
1428 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x10);
1429 	} else {
1430 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1431 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x10);
1432 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1433 	}
1434 
1435 	/* Configure DMA resource watermarks */
1436 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LOWAT, 5);
1437 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_HIWAT, 10);
1438 
1439 	/* Enable buffer manager */
1440 	CSR_WRITE_4(sc, BGE_BMAN_MODE,
1441 	    BGE_BMANMODE_ENABLE|BGE_BMANMODE_LOMBUF_ATTN);
1442 
1443 	/* Poll for buffer manager start indication */
1444 	for (i = 0; i < BGE_TIMEOUT; i++) {
1445 		if (CSR_READ_4(sc, BGE_BMAN_MODE) & BGE_BMANMODE_ENABLE)
1446 			break;
1447 		DELAY(10);
1448 	}
1449 
1450 	if (i == BGE_TIMEOUT) {
1451 		if_printf(&sc->arpcom.ac_if,
1452 			  "buffer manager failed to start\n");
1453 		return(ENXIO);
1454 	}
1455 
1456 	/* Enable flow-through queues */
1457 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
1458 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
1459 
1460 	/* Wait until queue initialization is complete */
1461 	for (i = 0; i < BGE_TIMEOUT; i++) {
1462 		if (CSR_READ_4(sc, BGE_FTQ_RESET) == 0)
1463 			break;
1464 		DELAY(10);
1465 	}
1466 
1467 	if (i == BGE_TIMEOUT) {
1468 		if_printf(&sc->arpcom.ac_if,
1469 			  "flow-through queue init failed\n");
1470 		return(ENXIO);
1471 	}
1472 
1473 	/*
1474 	 * Summary of rings supported by the controller:
1475 	 *
1476 	 * Standard Receive Producer Ring
1477 	 * - This ring is used to feed receive buffers for "standard"
1478 	 *   sized frames (typically 1536 bytes) to the controller.
1479 	 *
1480 	 * Jumbo Receive Producer Ring
1481 	 * - This ring is used to feed receive buffers for jumbo sized
1482 	 *   frames (i.e. anything bigger than the "standard" frames)
1483 	 *   to the controller.
1484 	 *
1485 	 * Mini Receive Producer Ring
1486 	 * - This ring is used to feed receive buffers for "mini"
1487 	 *   sized frames to the controller.
1488 	 * - This feature required external memory for the controller
1489 	 *   but was never used in a production system.  Should always
1490 	 *   be disabled.
1491 	 *
1492 	 * Receive Return Ring
1493 	 * - After the controller has placed an incoming frame into a
1494 	 *   receive buffer that buffer is moved into a receive return
1495 	 *   ring.  The driver is then responsible to passing the
1496 	 *   buffer up to the stack.  Many versions of the controller
1497 	 *   support multiple RR rings.
1498 	 *
1499 	 * Send Ring
1500 	 * - This ring is used for outgoing frames.  Many versions of
1501 	 *   the controller support multiple send rings.
1502 	 */
1503 
1504 	/* Initialize the standard receive producer ring control block. */
1505 	rcb = &sc->bge_ldata.bge_info.bge_std_rx_rcb;
1506 	rcb->bge_hostaddr.bge_addr_lo =
1507 	    BGE_ADDR_LO(sc->bge_ldata.bge_rx_std_ring_paddr);
1508 	rcb->bge_hostaddr.bge_addr_hi =
1509 	    BGE_ADDR_HI(sc->bge_ldata.bge_rx_std_ring_paddr);
1510 	if (BGE_IS_5705_PLUS(sc)) {
1511 		/*
1512 		 * Bits 31-16: Programmable ring size (512, 256, 128, 64, 32)
1513 		 * Bits 15-2 : Reserved (should be 0)
1514 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
1515 		 * Bit 0     : Reserved
1516 		 */
1517 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(512, 0);
1518 	} else {
1519 		/*
1520 		 * Ring size is always XXX entries
1521 		 * Bits 31-16: Maximum RX frame size
1522 		 * Bits 15-2 : Reserved (should be 0)
1523 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
1524 		 * Bit 0     : Reserved
1525 		 */
1526 		rcb->bge_maxlen_flags =
1527 		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN, 0);
1528 	}
1529 	rcb->bge_nicaddr = BGE_STD_RX_RINGS;
1530 	/* Write the standard receive producer ring control block. */
1531 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_HI, rcb->bge_hostaddr.bge_addr_hi);
1532 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_LO, rcb->bge_hostaddr.bge_addr_lo);
1533 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1534 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_NICADDR, rcb->bge_nicaddr);
1535 	/* Reset the standard receive producer ring producer index. */
1536 	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, 0);
1537 
1538 	/*
1539 	 * Initialize the jumbo RX producer ring control
1540 	 * block.  We set the 'ring disabled' bit in the
1541 	 * flags field until we're actually ready to start
1542 	 * using this ring (i.e. once we set the MTU
1543 	 * high enough to require it).
1544 	 */
1545 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
1546 		rcb = &sc->bge_ldata.bge_info.bge_jumbo_rx_rcb;
1547 		/* Get the jumbo receive producer ring RCB parameters. */
1548 		rcb->bge_hostaddr.bge_addr_lo =
1549 		    BGE_ADDR_LO(sc->bge_ldata.bge_rx_jumbo_ring_paddr);
1550 		rcb->bge_hostaddr.bge_addr_hi =
1551 		    BGE_ADDR_HI(sc->bge_ldata.bge_rx_jumbo_ring_paddr);
1552 		rcb->bge_maxlen_flags =
1553 		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN,
1554 		    BGE_RCB_FLAG_RING_DISABLED);
1555 		rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS;
1556 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_HI,
1557 		    rcb->bge_hostaddr.bge_addr_hi);
1558 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_LO,
1559 		    rcb->bge_hostaddr.bge_addr_lo);
1560 		/* Program the jumbo receive producer ring RCB parameters. */
1561 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS,
1562 		    rcb->bge_maxlen_flags);
1563 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_NICADDR, rcb->bge_nicaddr);
1564 		/* Reset the jumbo receive producer ring producer index. */
1565 		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, 0);
1566 	}
1567 
1568 	/* Disable the mini receive producer ring RCB. */
1569 	if (BGE_IS_5700_FAMILY(sc)) {
1570 		rcb = &sc->bge_ldata.bge_info.bge_mini_rx_rcb;
1571 		rcb->bge_maxlen_flags =
1572 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED);
1573 		CSR_WRITE_4(sc, BGE_RX_MINI_RCB_MAXLEN_FLAGS,
1574 		    rcb->bge_maxlen_flags);
1575 		/* Reset the mini receive producer ring producer index. */
1576 		bge_writembx(sc, BGE_MBX_RX_MINI_PROD_LO, 0);
1577 	}
1578 
1579 	/* Choose de-pipeline mode for BCM5906 A0, A1 and A2. */
1580 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
1581 	    (sc->bge_chipid == BGE_CHIPID_BCM5906_A0 ||
1582 	     sc->bge_chipid == BGE_CHIPID_BCM5906_A1 ||
1583 	     sc->bge_chipid == BGE_CHIPID_BCM5906_A2)) {
1584 		CSR_WRITE_4(sc, BGE_ISO_PKT_TX,
1585 		    (CSR_READ_4(sc, BGE_ISO_PKT_TX) & ~3) | 2);
1586 	}
1587 
1588 	/*
1589 	 * The BD ring replenish thresholds control how often the
1590 	 * hardware fetches new BD's from the producer rings in host
1591 	 * memory.  Setting the value too low on a busy system can
1592 	 * starve the hardware and recue the throughpout.
1593 	 *
1594 	 * Set the BD ring replentish thresholds. The recommended
1595 	 * values are 1/8th the number of descriptors allocated to
1596 	 * each ring.
1597 	 */
1598 	if (BGE_IS_5705_PLUS(sc))
1599 		val = 8;
1600 	else
1601 		val = BGE_STD_RX_RING_CNT / 8;
1602 	CSR_WRITE_4(sc, BGE_RBDI_STD_REPL_THRESH, val);
1603 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
1604 		CSR_WRITE_4(sc, BGE_RBDI_JUMBO_REPL_THRESH,
1605 		    BGE_JUMBO_RX_RING_CNT/8);
1606 	}
1607 
1608 	/*
1609 	 * Disable all send rings by setting the 'ring disabled' bit
1610 	 * in the flags field of all the TX send ring control blocks,
1611 	 * located in NIC memory.
1612 	 */
1613 	if (!BGE_IS_5705_PLUS(sc)) {
1614 		/* 5700 to 5704 had 16 send rings. */
1615 		limit = BGE_TX_RINGS_EXTSSRAM_MAX;
1616 	} else {
1617 		limit = 1;
1618 	}
1619 	vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1620 	for (i = 0; i < limit; i++) {
1621 		RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1622 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED));
1623 		RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1624 		vrcb += sizeof(struct bge_rcb);
1625 	}
1626 
1627 	/* Configure send ring RCB 0 (we use only the first ring) */
1628 	vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1629 	BGE_HOSTADDR(taddr, sc->bge_ldata.bge_tx_ring_paddr);
1630 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1631 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1632 	RCB_WRITE_4(sc, vrcb, bge_nicaddr,
1633 	    BGE_NIC_TXRING_ADDR(0, BGE_TX_RING_CNT));
1634 	RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1635 	    BGE_RCB_MAXLEN_FLAGS(BGE_TX_RING_CNT, 0));
1636 
1637 	/*
1638 	 * Disable all receive return rings by setting the
1639 	 * 'ring diabled' bit in the flags field of all the receive
1640 	 * return ring control blocks, located in NIC memory.
1641 	 */
1642 	if (!BGE_IS_5705_PLUS(sc))
1643 		limit = BGE_RX_RINGS_MAX;
1644 	else if (sc->bge_asicrev == BGE_ASICREV_BCM5755)
1645 		limit = 4;
1646 	else
1647 		limit = 1;
1648 	/* Disable all receive return rings. */
1649 	vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1650 	for (i = 0; i < limit; i++) {
1651 		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, 0);
1652 		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, 0);
1653 		RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1654 		    BGE_RCB_FLAG_RING_DISABLED);
1655 		RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1656 		bge_writembx(sc, BGE_MBX_RX_CONS0_LO +
1657 		    (i * (sizeof(uint64_t))), 0);
1658 		vrcb += sizeof(struct bge_rcb);
1659 	}
1660 
1661 	/*
1662 	 * Set up receive return ring 0.  Note that the NIC address
1663 	 * for RX return rings is 0x0.  The return rings live entirely
1664 	 * within the host, so the nicaddr field in the RCB isn't used.
1665 	 */
1666 	vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1667 	BGE_HOSTADDR(taddr, sc->bge_ldata.bge_rx_return_ring_paddr);
1668 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1669 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1670 	RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1671 	RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1672 	    BGE_RCB_MAXLEN_FLAGS(sc->bge_return_ring_cnt, 0));
1673 
1674 	/* Set random backoff seed for TX */
1675 	CSR_WRITE_4(sc, BGE_TX_RANDOM_BACKOFF,
1676 	    sc->arpcom.ac_enaddr[0] + sc->arpcom.ac_enaddr[1] +
1677 	    sc->arpcom.ac_enaddr[2] + sc->arpcom.ac_enaddr[3] +
1678 	    sc->arpcom.ac_enaddr[4] + sc->arpcom.ac_enaddr[5] +
1679 	    BGE_TX_BACKOFF_SEED_MASK);
1680 
1681 	/* Set inter-packet gap */
1682 	CSR_WRITE_4(sc, BGE_TX_LENGTHS, 0x2620);
1683 
1684 	/*
1685 	 * Specify which ring to use for packets that don't match
1686 	 * any RX rules.
1687 	 */
1688 	CSR_WRITE_4(sc, BGE_RX_RULES_CFG, 0x08);
1689 
1690 	/*
1691 	 * Configure number of RX lists. One interrupt distribution
1692 	 * list, sixteen active lists, one bad frames class.
1693 	 */
1694 	CSR_WRITE_4(sc, BGE_RXLP_CFG, 0x181);
1695 
1696 	/* Inialize RX list placement stats mask. */
1697 	CSR_WRITE_4(sc, BGE_RXLP_STATS_ENABLE_MASK, 0x007FFFFF);
1698 	CSR_WRITE_4(sc, BGE_RXLP_STATS_CTL, 0x1);
1699 
1700 	/* Disable host coalescing until we get it set up */
1701 	CSR_WRITE_4(sc, BGE_HCC_MODE, 0x00000000);
1702 
1703 	/* Poll to make sure it's shut down. */
1704 	for (i = 0; i < BGE_TIMEOUT; i++) {
1705 		if (!(CSR_READ_4(sc, BGE_HCC_MODE) & BGE_HCCMODE_ENABLE))
1706 			break;
1707 		DELAY(10);
1708 	}
1709 
1710 	if (i == BGE_TIMEOUT) {
1711 		if_printf(&sc->arpcom.ac_if,
1712 			  "host coalescing engine failed to idle\n");
1713 		return(ENXIO);
1714 	}
1715 
1716 	/* Set up host coalescing defaults */
1717 	CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, sc->bge_rx_coal_ticks);
1718 	CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS, sc->bge_tx_coal_ticks);
1719 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, sc->bge_rx_coal_bds);
1720 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS, sc->bge_tx_coal_bds);
1721 	if (!BGE_IS_5705_PLUS(sc)) {
1722 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS_INT,
1723 		    sc->bge_rx_coal_ticks_int);
1724 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS_INT,
1725 		    sc->bge_tx_coal_ticks_int);
1726 	}
1727 	/*
1728 	 * NOTE:
1729 	 * The datasheet (57XX-PG105-R) says BCM5705+ do not
1730 	 * have following two registers; obviously it is wrong.
1731 	 */
1732 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT, sc->bge_rx_coal_bds_int);
1733 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT, sc->bge_tx_coal_bds_int);
1734 
1735 	/* Set up address of statistics block */
1736 	if (!BGE_IS_5705_PLUS(sc)) {
1737 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_HI,
1738 		    BGE_ADDR_HI(sc->bge_ldata.bge_stats_paddr));
1739 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_LO,
1740 		    BGE_ADDR_LO(sc->bge_ldata.bge_stats_paddr));
1741 
1742 		CSR_WRITE_4(sc, BGE_HCC_STATS_BASEADDR, BGE_STATS_BLOCK);
1743 		CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_BASEADDR, BGE_STATUS_BLOCK);
1744 		CSR_WRITE_4(sc, BGE_HCC_STATS_TICKS, sc->bge_stat_ticks);
1745 	}
1746 
1747 	/* Set up address of status block */
1748 	bzero(sc->bge_ldata.bge_status_block, BGE_STATUS_BLK_SZ);
1749 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_HI,
1750 	    BGE_ADDR_HI(sc->bge_ldata.bge_status_block_paddr));
1751 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_LO,
1752 	    BGE_ADDR_LO(sc->bge_ldata.bge_status_block_paddr));
1753 
1754 	/*
1755 	 * Set up status block partail update size.
1756 	 *
1757 	 * Because only single TX ring, RX produce ring and Rx return ring
1758 	 * are used, ask device to update only minimum part of status block
1759 	 * except for BCM5700 AX/BX, whose status block partial update size
1760 	 * can't be configured.
1761 	 */
1762 	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
1763 	    sc->bge_chipid != BGE_CHIPID_BCM5700_C0) {
1764 		/* XXX Actually reserved on BCM5700 AX/BX */
1765 		val = BGE_STATBLKSZ_FULL;
1766 	} else {
1767 		val = BGE_STATBLKSZ_32BYTE;
1768 	}
1769 #if 0
1770 	/*
1771 	 * Does not seem to have visible effect in both
1772 	 * bulk data (1472B UDP datagram) and tiny data
1773 	 * (18B UDP datagram) TX tests.
1774 	 */
1775 	if (!BGE_IS_CRIPPLED(sc))
1776 		val |= BGE_HCCMODE_CLRTICK_TX;
1777 #endif
1778 
1779 	/* Turn on host coalescing state machine */
1780 	CSR_WRITE_4(sc, BGE_HCC_MODE, val | BGE_HCCMODE_ENABLE);
1781 
1782 	/* Turn on RX BD completion state machine and enable attentions */
1783 	CSR_WRITE_4(sc, BGE_RBDC_MODE,
1784 	    BGE_RBDCMODE_ENABLE|BGE_RBDCMODE_ATTN);
1785 
1786 	/* Turn on RX list placement state machine */
1787 	CSR_WRITE_4(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
1788 
1789 	/* Turn on RX list selector state machine. */
1790 	if (!BGE_IS_5705_PLUS(sc))
1791 		CSR_WRITE_4(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
1792 
1793 	val = BGE_MACMODE_TXDMA_ENB | BGE_MACMODE_RXDMA_ENB |
1794 	    BGE_MACMODE_RX_STATS_CLEAR | BGE_MACMODE_TX_STATS_CLEAR |
1795 	    BGE_MACMODE_RX_STATS_ENB | BGE_MACMODE_TX_STATS_ENB |
1796 	    BGE_MACMODE_FRMHDR_DMA_ENB;
1797 
1798 	if (sc->bge_flags & BGE_FLAG_TBI)
1799 		val |= BGE_PORTMODE_TBI;
1800 	else if (sc->bge_flags & BGE_FLAG_MII_SERDES)
1801 		val |= BGE_PORTMODE_GMII;
1802 	else
1803 		val |= BGE_PORTMODE_MII;
1804 
1805 	/* Turn on DMA, clear stats */
1806 	CSR_WRITE_4(sc, BGE_MAC_MODE, val);
1807 
1808 	/* Set misc. local control, enable interrupts on attentions */
1809 	CSR_WRITE_4(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_ONATTN);
1810 
1811 #ifdef notdef
1812 	/* Assert GPIO pins for PHY reset */
1813 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUT0|
1814 	    BGE_MLC_MISCIO_OUT1|BGE_MLC_MISCIO_OUT2);
1815 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUTEN0|
1816 	    BGE_MLC_MISCIO_OUTEN1|BGE_MLC_MISCIO_OUTEN2);
1817 #endif
1818 
1819 	/* Turn on DMA completion state machine */
1820 	if (!BGE_IS_5705_PLUS(sc))
1821 		CSR_WRITE_4(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
1822 
1823 	/* Turn on write DMA state machine */
1824 	val = BGE_WDMAMODE_ENABLE|BGE_WDMAMODE_ALL_ATTNS;
1825 	if (BGE_IS_5755_PLUS(sc)) {
1826 		/* Enable host coalescing bug fix. */
1827 		val |= BGE_WDMAMODE_STATUS_TAG_FIX;
1828 	}
1829 	if (sc->bge_asicrev == BGE_ASICREV_BCM5785) {
1830 		/* Request larger DMA burst size to get better performance. */
1831 		val |= BGE_WDMAMODE_BURST_ALL_DATA;
1832 	}
1833 	CSR_WRITE_4(sc, BGE_WDMA_MODE, val);
1834 	DELAY(40);
1835 
1836 	if (sc->bge_asicrev == BGE_ASICREV_BCM5761 ||
1837 	    sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
1838 	    sc->bge_asicrev == BGE_ASICREV_BCM5785 ||
1839 	    sc->bge_asicrev == BGE_ASICREV_BCM57780) {
1840 		/*
1841 		 * Enable fix for read DMA FIFO overruns.
1842 		 * The fix is to limit the number of RX BDs
1843 		 * the hardware would fetch at a fime.
1844 		 */
1845 		val = CSR_READ_4(sc, BGE_RDMA_RSRVCTRL);
1846 		CSR_WRITE_4(sc, BGE_RDMA_RSRVCTRL,
1847 		    val| BGE_RDMA_RSRVCTRL_FIFO_OFLW_FIX);
1848 	}
1849 
1850 	/* Turn on read DMA state machine */
1851 	val = BGE_RDMAMODE_ENABLE | BGE_RDMAMODE_ALL_ATTNS;
1852         if (sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
1853             sc->bge_asicrev == BGE_ASICREV_BCM5785 ||
1854             sc->bge_asicrev == BGE_ASICREV_BCM57780)
1855 		val |= BGE_RDMAMODE_BD_SBD_CRPT_ATTN |
1856                   BGE_RDMAMODE_MBUF_RBD_CRPT_ATTN |
1857                   BGE_RDMAMODE_MBUF_SBD_CRPT_ATTN;
1858 	if (sc->bge_flags & BGE_FLAG_PCIE)
1859 		val |= BGE_RDMAMODE_FIFO_LONG_BURST;
1860 	if (sc->bge_flags & BGE_FLAG_TSO)
1861 		val |= BGE_RDMAMODE_TSO4_ENABLE;
1862 	CSR_WRITE_4(sc, BGE_RDMA_MODE, val);
1863 	DELAY(40);
1864 
1865 	/* Turn on RX data completion state machine */
1866 	CSR_WRITE_4(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
1867 
1868 	/* Turn on RX BD initiator state machine */
1869 	CSR_WRITE_4(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
1870 
1871 	/* Turn on RX data and RX BD initiator state machine */
1872 	CSR_WRITE_4(sc, BGE_RDBDI_MODE, BGE_RDBDIMODE_ENABLE);
1873 
1874 	/* Turn on Mbuf cluster free state machine */
1875 	if (!BGE_IS_5705_PLUS(sc))
1876 		CSR_WRITE_4(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
1877 
1878 	/* Turn on send BD completion state machine */
1879 	CSR_WRITE_4(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
1880 
1881 	/* Turn on send data completion state machine */
1882 	val = BGE_SDCMODE_ENABLE;
1883 	if (sc->bge_asicrev == BGE_ASICREV_BCM5761)
1884 		val |= BGE_SDCMODE_CDELAY;
1885 	CSR_WRITE_4(sc, BGE_SDC_MODE, val);
1886 
1887 	/* Turn on send data initiator state machine */
1888 	if (sc->bge_flags & BGE_FLAG_TSO)
1889 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE |
1890 		    BGE_SDIMODE_HW_LSO_PRE_DMA);
1891 	else
1892 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
1893 
1894 	/* Turn on send BD initiator state machine */
1895 	CSR_WRITE_4(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
1896 
1897 	/* Turn on send BD selector state machine */
1898 	CSR_WRITE_4(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
1899 
1900 	CSR_WRITE_4(sc, BGE_SDI_STATS_ENABLE_MASK, 0x007FFFFF);
1901 	CSR_WRITE_4(sc, BGE_SDI_STATS_CTL,
1902 	    BGE_SDISTATSCTL_ENABLE|BGE_SDISTATSCTL_FASTER);
1903 
1904 	/* ack/clear link change events */
1905 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
1906 	    BGE_MACSTAT_CFG_CHANGED|BGE_MACSTAT_MI_COMPLETE|
1907 	    BGE_MACSTAT_LINK_CHANGED);
1908 	CSR_WRITE_4(sc, BGE_MI_STS, 0);
1909 
1910 	/*
1911 	 * Enable attention when the link has changed state for
1912 	 * devices that use auto polling.
1913 	 */
1914 	if (sc->bge_flags & BGE_FLAG_TBI) {
1915 		CSR_WRITE_4(sc, BGE_MI_STS, BGE_MISTS_LINK);
1916  	} else {
1917 		if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
1918 			CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode);
1919 			DELAY(80);
1920 		}
1921 		if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
1922 		    sc->bge_chipid != BGE_CHIPID_BCM5700_B2) {
1923 			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
1924 			    BGE_EVTENB_MI_INTERRUPT);
1925 		}
1926 	}
1927 
1928 	/*
1929 	 * Clear any pending link state attention.
1930 	 * Otherwise some link state change events may be lost until attention
1931 	 * is cleared by bge_intr() -> bge_softc.bge_link_upd() sequence.
1932 	 * It's not necessary on newer BCM chips - perhaps enabling link
1933 	 * state change attentions implies clearing pending attention.
1934 	 */
1935 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
1936 	    BGE_MACSTAT_CFG_CHANGED|BGE_MACSTAT_MI_COMPLETE|
1937 	    BGE_MACSTAT_LINK_CHANGED);
1938 
1939 	/* Enable link state change attentions. */
1940 	BGE_SETBIT(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_LINK_CHANGED);
1941 
1942 	return(0);
1943 }
1944 
1945 /*
1946  * Probe for a Broadcom chip. Check the PCI vendor and device IDs
1947  * against our list and return its name if we find a match. Note
1948  * that since the Broadcom controller contains VPD support, we
1949  * can get the device name string from the controller itself instead
1950  * of the compiled-in string. This is a little slow, but it guarantees
1951  * we'll always announce the right product name.
1952  */
1953 static int
1954 bge_probe(device_t dev)
1955 {
1956 	const struct bge_type *t;
1957 	uint16_t product, vendor;
1958 
1959 	product = pci_get_device(dev);
1960 	vendor = pci_get_vendor(dev);
1961 
1962 	for (t = bge_devs; t->bge_name != NULL; t++) {
1963 		if (vendor == t->bge_vid && product == t->bge_did)
1964 			break;
1965 	}
1966 	if (t->bge_name == NULL)
1967 		return(ENXIO);
1968 
1969 	device_set_desc(dev, t->bge_name);
1970 	return(0);
1971 }
1972 
1973 static int
1974 bge_attach(device_t dev)
1975 {
1976 	struct ifnet *ifp;
1977 	struct bge_softc *sc;
1978 	uint32_t hwcfg = 0, misccfg;
1979 	int error = 0, rid, capmask;
1980 	uint8_t ether_addr[ETHER_ADDR_LEN];
1981 	uint16_t product, vendor;
1982 	driver_intr_t *intr_func;
1983 	uintptr_t mii_priv = 0;
1984 	u_int intr_flags;
1985 	int msi_enable;
1986 
1987 	sc = device_get_softc(dev);
1988 	sc->bge_dev = dev;
1989 	callout_init_mp(&sc->bge_stat_timer);
1990 	lwkt_serialize_init(&sc->bge_jslot_serializer);
1991 
1992 	product = pci_get_device(dev);
1993 	vendor = pci_get_vendor(dev);
1994 
1995 #ifndef BURN_BRIDGES
1996 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1997 		uint32_t irq, mem;
1998 
1999 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
2000 		mem = pci_read_config(dev, BGE_PCI_BAR0, 4);
2001 
2002 		device_printf(dev, "chip is in D%d power mode "
2003 		    "-- setting to D0\n", pci_get_powerstate(dev));
2004 
2005 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
2006 
2007 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
2008 		pci_write_config(dev, BGE_PCI_BAR0, mem, 4);
2009 	}
2010 #endif	/* !BURN_BRIDGE */
2011 
2012 	/*
2013 	 * Map control/status registers.
2014 	 */
2015 	pci_enable_busmaster(dev);
2016 
2017 	rid = BGE_PCI_BAR0;
2018 	sc->bge_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
2019 	    RF_ACTIVE);
2020 
2021 	if (sc->bge_res == NULL) {
2022 		device_printf(dev, "couldn't map memory\n");
2023 		return ENXIO;
2024 	}
2025 
2026 	sc->bge_btag = rman_get_bustag(sc->bge_res);
2027 	sc->bge_bhandle = rman_get_bushandle(sc->bge_res);
2028 
2029 	/* Save various chip information */
2030 	sc->bge_chipid =
2031 	    pci_read_config(dev, BGE_PCI_MISC_CTL, 4) >>
2032 	    BGE_PCIMISCCTL_ASICREV_SHIFT;
2033 	if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_USE_PRODID_REG) {
2034 		/* All chips, which use BGE_PCI_PRODID_ASICREV, have CPMU */
2035 		sc->bge_flags |= BGE_FLAG_CPMU;
2036 		sc->bge_chipid = pci_read_config(dev, BGE_PCI_PRODID_ASICREV, 4);
2037 	}
2038 	sc->bge_asicrev = BGE_ASICREV(sc->bge_chipid);
2039 	sc->bge_chiprev = BGE_CHIPREV(sc->bge_chipid);
2040 
2041 	/* Save chipset family. */
2042 	switch (sc->bge_asicrev) {
2043 	case BGE_ASICREV_BCM5755:
2044 	case BGE_ASICREV_BCM5761:
2045 	case BGE_ASICREV_BCM5784:
2046 	case BGE_ASICREV_BCM5785:
2047 	case BGE_ASICREV_BCM5787:
2048 	case BGE_ASICREV_BCM57780:
2049 	    sc->bge_flags |= BGE_FLAG_5755_PLUS | BGE_FLAG_575X_PLUS |
2050 		BGE_FLAG_5705_PLUS;
2051 	    break;
2052 
2053 	case BGE_ASICREV_BCM5700:
2054 	case BGE_ASICREV_BCM5701:
2055 	case BGE_ASICREV_BCM5703:
2056 	case BGE_ASICREV_BCM5704:
2057 		sc->bge_flags |= BGE_FLAG_5700_FAMILY | BGE_FLAG_JUMBO;
2058 		break;
2059 
2060 	case BGE_ASICREV_BCM5714_A0:
2061 	case BGE_ASICREV_BCM5780:
2062 	case BGE_ASICREV_BCM5714:
2063 		sc->bge_flags |= BGE_FLAG_5714_FAMILY;
2064 		/* Fall through */
2065 
2066 	case BGE_ASICREV_BCM5750:
2067 	case BGE_ASICREV_BCM5752:
2068 	case BGE_ASICREV_BCM5906:
2069 		sc->bge_flags |= BGE_FLAG_575X_PLUS;
2070 		/* Fall through */
2071 
2072 	case BGE_ASICREV_BCM5705:
2073 		sc->bge_flags |= BGE_FLAG_5705_PLUS;
2074 		break;
2075 	}
2076 
2077 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
2078 		sc->bge_flags |= BGE_FLAG_NO_EEPROM;
2079 
2080 	misccfg = CSR_READ_4(sc, BGE_MISC_CFG) & BGE_MISCCFG_BOARD_ID_MASK;
2081 	if (sc->bge_asicrev == BGE_ASICREV_BCM5705 &&
2082 	    (misccfg == BGE_MISCCFG_BOARD_ID_5788 ||
2083 	     misccfg == BGE_MISCCFG_BOARD_ID_5788M))
2084 		sc->bge_flags |= BGE_FLAG_5788;
2085 
2086 	/* BCM5755 or higher and BCM5906 have short DMA bug. */
2087 	if (BGE_IS_5755_PLUS(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5906)
2088 		sc->bge_flags |= BGE_FLAG_SHORTDMA;
2089 
2090   	/*
2091 	 * Check if this is a PCI-X or PCI Express device.
2092   	 */
2093 	if (BGE_IS_5705_PLUS(sc)) {
2094 		if (pci_is_pcie(dev)) {
2095 			sc->bge_flags |= BGE_FLAG_PCIE;
2096 			sc->bge_pciecap = pci_get_pciecap_ptr(sc->bge_dev);
2097 			pcie_set_max_readrq(dev, PCIEM_DEVCTL_MAX_READRQ_4096);
2098 		}
2099 	} else {
2100 		/*
2101 		 * Check if the device is in PCI-X Mode.
2102 		 * (This bit is not valid on PCI Express controllers.)
2103 		 */
2104 		if ((pci_read_config(sc->bge_dev, BGE_PCI_PCISTATE, 4) &
2105 		    BGE_PCISTATE_PCI_BUSMODE) == 0) {
2106 			sc->bge_flags |= BGE_FLAG_PCIX;
2107 			sc->bge_pcixcap = pci_get_pcixcap_ptr(sc->bge_dev);
2108 			sc->bge_mbox_reorder = device_getenv_int(sc->bge_dev,
2109 			    "mbox_reorder", 0);
2110 		}
2111  	}
2112 	device_printf(dev, "CHIP ID 0x%08x; "
2113 		      "ASIC REV 0x%02x; CHIP REV 0x%02x; %s\n",
2114 		      sc->bge_chipid, sc->bge_asicrev, sc->bge_chiprev,
2115 		      (sc->bge_flags & BGE_FLAG_PCIX) ? "PCI-X"
2116 		      : ((sc->bge_flags & BGE_FLAG_PCIE) ?
2117 			"PCI-E" : "PCI"));
2118 
2119 	/*
2120 	 * The 40bit DMA bug applies to the 5714/5715 controllers and is
2121 	 * not actually a MAC controller bug but an issue with the embedded
2122 	 * PCIe to PCI-X bridge in the device. Use 40bit DMA workaround.
2123 	 */
2124 	if ((sc->bge_flags & BGE_FLAG_PCIX) &&
2125 	    (BGE_IS_5714_FAMILY(sc) || device_getenv_int(dev, "dma40b", 0)))
2126 		sc->bge_flags |= BGE_FLAG_MAXADDR_40BIT;
2127 
2128 	/*
2129 	 * When using the BCM5701 in PCI-X mode, data corruption has
2130 	 * been observed in the first few bytes of some received packets.
2131 	 * Aligning the packet buffer in memory eliminates the corruption.
2132 	 * Unfortunately, this misaligns the packet payloads.  On platforms
2133 	 * which do not support unaligned accesses, we will realign the
2134 	 * payloads by copying the received packets.
2135 	 */
2136 	if (sc->bge_asicrev == BGE_ASICREV_BCM5701 &&
2137 	    (sc->bge_flags & BGE_FLAG_PCIX))
2138 		sc->bge_flags |= BGE_FLAG_RX_ALIGNBUG;
2139 
2140 	if (!BGE_IS_CRIPPLED(sc)) {
2141 		if (device_getenv_int(dev, "status_tag", 1)) {
2142 			sc->bge_flags |= BGE_FLAG_STATUS_TAG;
2143 			sc->bge_pci_miscctl = BGE_PCIMISCCTL_TAGGED_STATUS;
2144 			if (bootverbose)
2145 				device_printf(dev, "enable status tag\n");
2146 		}
2147 	}
2148 
2149 	if (BGE_IS_5755_PLUS(sc)) {
2150 		/*
2151 		 * BCM5754 and BCM5787 shares the same ASIC id so
2152 		 * explicit device id check is required.
2153 		 * Due to unknown reason TSO does not work on BCM5755M.
2154 		 */
2155 		if (product != PCI_PRODUCT_BROADCOM_BCM5754 &&
2156 		    product != PCI_PRODUCT_BROADCOM_BCM5754M &&
2157 		    product != PCI_PRODUCT_BROADCOM_BCM5755M)
2158 			sc->bge_flags |= BGE_FLAG_TSO;
2159 	}
2160 
2161 	/*
2162 	 * Set various PHY quirk flags.
2163 	 */
2164 
2165 	if ((sc->bge_asicrev == BGE_ASICREV_BCM5700 ||
2166 	     sc->bge_asicrev == BGE_ASICREV_BCM5701) &&
2167 	    pci_get_subvendor(dev) == PCI_VENDOR_DELL)
2168 		mii_priv |= BRGPHY_FLAG_NO_3LED;
2169 
2170 	capmask = MII_CAPMASK_DEFAULT;
2171 	if ((sc->bge_asicrev == BGE_ASICREV_BCM5703 &&
2172 	     (misccfg == 0x4000 || misccfg == 0x8000)) ||
2173 	    (sc->bge_asicrev == BGE_ASICREV_BCM5705 &&
2174 	     vendor == PCI_VENDOR_BROADCOM &&
2175 	     (product == PCI_PRODUCT_BROADCOM_BCM5901 ||
2176 	      product == PCI_PRODUCT_BROADCOM_BCM5901A2 ||
2177 	      product == PCI_PRODUCT_BROADCOM_BCM5705F)) ||
2178 	    (vendor == PCI_VENDOR_BROADCOM &&
2179 	     (product == PCI_PRODUCT_BROADCOM_BCM5751F ||
2180 	      product == PCI_PRODUCT_BROADCOM_BCM5753F ||
2181 	      product == PCI_PRODUCT_BROADCOM_BCM5787F)) ||
2182 	    product == PCI_PRODUCT_BROADCOM_BCM57790 ||
2183 	    sc->bge_asicrev == BGE_ASICREV_BCM5906) {
2184 		/* 10/100 only */
2185 		capmask &= ~BMSR_EXTSTAT;
2186 	}
2187 
2188 	mii_priv |= BRGPHY_FLAG_WIRESPEED;
2189 	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 ||
2190 	    (sc->bge_asicrev == BGE_ASICREV_BCM5705 &&
2191 	     (sc->bge_chipid != BGE_CHIPID_BCM5705_A0 &&
2192 	      sc->bge_chipid != BGE_CHIPID_BCM5705_A1)) ||
2193 	    sc->bge_asicrev == BGE_ASICREV_BCM5906)
2194 		mii_priv &= ~BRGPHY_FLAG_WIRESPEED;
2195 
2196 	if (sc->bge_chipid == BGE_CHIPID_BCM5701_A0 ||
2197 	    sc->bge_chipid == BGE_CHIPID_BCM5701_B0)
2198 		mii_priv |= BRGPHY_FLAG_CRC_BUG;
2199 
2200 	if (sc->bge_chiprev == BGE_CHIPREV_5703_AX ||
2201 	    sc->bge_chiprev == BGE_CHIPREV_5704_AX)
2202 		mii_priv |= BRGPHY_FLAG_ADC_BUG;
2203 
2204 	if (sc->bge_chipid == BGE_CHIPID_BCM5704_A0)
2205 		mii_priv |= BRGPHY_FLAG_5704_A0;
2206 
2207 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
2208 		mii_priv |= BRGPHY_FLAG_5906;
2209 
2210 	if (BGE_IS_5705_PLUS(sc) &&
2211 	    sc->bge_asicrev != BGE_ASICREV_BCM5906 &&
2212 	    /* sc->bge_asicrev != BGE_ASICREV_BCM5717 && */
2213 	    sc->bge_asicrev != BGE_ASICREV_BCM5785 &&
2214 	    /* sc->bge_asicrev != BGE_ASICREV_BCM57765 && */
2215 	    sc->bge_asicrev != BGE_ASICREV_BCM57780) {
2216 		if (sc->bge_asicrev == BGE_ASICREV_BCM5755 ||
2217 		    sc->bge_asicrev == BGE_ASICREV_BCM5761 ||
2218 		    sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
2219 		    sc->bge_asicrev == BGE_ASICREV_BCM5787) {
2220 			if (product != PCI_PRODUCT_BROADCOM_BCM5722 &&
2221 			    product != PCI_PRODUCT_BROADCOM_BCM5756)
2222 				mii_priv |= BRGPHY_FLAG_JITTER_BUG;
2223 			if (product == PCI_PRODUCT_BROADCOM_BCM5755M)
2224 				mii_priv |= BRGPHY_FLAG_ADJUST_TRIM;
2225 		} else {
2226 			mii_priv |= BRGPHY_FLAG_BER_BUG;
2227 		}
2228 	}
2229 
2230 	/*
2231 	 * Allocate interrupt
2232 	 */
2233 	msi_enable = bge_msi_enable;
2234 	if ((sc->bge_flags & BGE_FLAG_STATUS_TAG) == 0) {
2235 		/* If "tagged status" is disabled, don't enable MSI */
2236 		msi_enable = 0;
2237 	} else if (msi_enable) {
2238 		msi_enable = 0; /* Disable by default */
2239 		if (BGE_IS_575X_PLUS(sc)) {
2240 			msi_enable = 1;
2241 			/* XXX we filter all 5714 chips */
2242 			if (sc->bge_asicrev == BGE_ASICREV_BCM5714 ||
2243 			    (sc->bge_asicrev == BGE_ASICREV_BCM5750 &&
2244 			     (sc->bge_chiprev == BGE_CHIPREV_5750_AX ||
2245 			      sc->bge_chiprev == BGE_CHIPREV_5750_BX)))
2246 				msi_enable = 0;
2247 			else if (BGE_IS_5755_PLUS(sc) ||
2248 			    sc->bge_asicrev == BGE_ASICREV_BCM5906)
2249 				sc->bge_flags |= BGE_FLAG_ONESHOT_MSI;
2250 		}
2251 	}
2252 	if (msi_enable) {
2253 		if (pci_find_extcap(dev, PCIY_MSI, &sc->bge_msicap)) {
2254 			device_printf(dev, "no MSI capability\n");
2255 			msi_enable = 0;
2256 		}
2257 	}
2258 
2259 	sc->bge_irq_type = pci_alloc_1intr(dev, msi_enable, &sc->bge_irq_rid,
2260 	    &intr_flags);
2261 
2262 	sc->bge_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &sc->bge_irq_rid,
2263 	    intr_flags);
2264 	if (sc->bge_irq == NULL) {
2265 		device_printf(dev, "couldn't map interrupt\n");
2266 		error = ENXIO;
2267 		goto fail;
2268 	}
2269 
2270 	if (sc->bge_irq_type == PCI_INTR_TYPE_MSI)
2271 		bge_enable_msi(sc);
2272 	else
2273 		sc->bge_flags &= ~BGE_FLAG_ONESHOT_MSI;
2274 
2275 	/* Initialize if_name earlier, so if_printf could be used */
2276 	ifp = &sc->arpcom.ac_if;
2277 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
2278 
2279 	/* Try to reset the chip. */
2280 	bge_reset(sc);
2281 
2282 	if (bge_chipinit(sc)) {
2283 		device_printf(dev, "chip initialization failed\n");
2284 		error = ENXIO;
2285 		goto fail;
2286 	}
2287 
2288 	/*
2289 	 * Get station address
2290 	 */
2291 	error = bge_get_eaddr(sc, ether_addr);
2292 	if (error) {
2293 		device_printf(dev, "failed to read station address\n");
2294 		goto fail;
2295 	}
2296 
2297 	/* 5705/5750 limits RX return ring to 512 entries. */
2298 	if (BGE_IS_5705_PLUS(sc))
2299 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT_5705;
2300 	else
2301 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT;
2302 
2303 	error = bge_dma_alloc(sc);
2304 	if (error)
2305 		goto fail;
2306 
2307 	/* Set default tuneable values. */
2308 	sc->bge_stat_ticks = BGE_TICKS_PER_SEC;
2309 	sc->bge_rx_coal_ticks = BGE_RX_COAL_TICKS_DEF;
2310 	sc->bge_tx_coal_ticks = BGE_TX_COAL_TICKS_DEF;
2311 	sc->bge_rx_coal_bds = BGE_RX_COAL_BDS_DEF;
2312 	sc->bge_tx_coal_bds = BGE_TX_COAL_BDS_DEF;
2313 	if (sc->bge_flags & BGE_FLAG_STATUS_TAG) {
2314 		sc->bge_rx_coal_ticks_int = BGE_RX_COAL_TICKS_DEF;
2315 		sc->bge_tx_coal_ticks_int = BGE_TX_COAL_TICKS_DEF;
2316 		sc->bge_rx_coal_bds_int = BGE_RX_COAL_BDS_DEF;
2317 		sc->bge_tx_coal_bds_int = BGE_TX_COAL_BDS_DEF;
2318 	} else {
2319 		sc->bge_rx_coal_ticks_int = BGE_RX_COAL_TICKS_MIN;
2320 		sc->bge_tx_coal_ticks_int = BGE_TX_COAL_TICKS_MIN;
2321 		sc->bge_rx_coal_bds_int = BGE_RX_COAL_BDS_MIN;
2322 		sc->bge_tx_coal_bds_int = BGE_TX_COAL_BDS_MIN;
2323 	}
2324 
2325 	/* Set up TX spare and reserved descriptor count */
2326 	if (sc->bge_flags & BGE_FLAG_TSO) {
2327 		sc->bge_txspare = BGE_NSEG_SPARE_TSO;
2328 		sc->bge_txrsvd = BGE_NSEG_RSVD_TSO;
2329 	} else {
2330 		sc->bge_txspare = BGE_NSEG_SPARE;
2331 		sc->bge_txrsvd = BGE_NSEG_RSVD;
2332 	}
2333 
2334 	/* Set up ifnet structure */
2335 	ifp->if_softc = sc;
2336 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2337 	ifp->if_ioctl = bge_ioctl;
2338 	ifp->if_start = bge_start;
2339 #ifdef IFPOLL_ENABLE
2340 	ifp->if_npoll = bge_npoll;
2341 #endif
2342 	ifp->if_watchdog = bge_watchdog;
2343 	ifp->if_init = bge_init;
2344 	ifp->if_mtu = ETHERMTU;
2345 	ifp->if_capabilities = IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU;
2346 	ifq_set_maxlen(&ifp->if_snd, BGE_TX_RING_CNT - 1);
2347 	ifq_set_ready(&ifp->if_snd);
2348 
2349 	/*
2350 	 * 5700 B0 chips do not support checksumming correctly due
2351 	 * to hardware bugs.
2352 	 */
2353 	if (sc->bge_chipid != BGE_CHIPID_BCM5700_B0) {
2354 		ifp->if_capabilities |= IFCAP_HWCSUM;
2355 		ifp->if_hwassist |= BGE_CSUM_FEATURES;
2356 	}
2357 	if (sc->bge_flags & BGE_FLAG_TSO) {
2358 		ifp->if_capabilities |= IFCAP_TSO;
2359 		ifp->if_hwassist |= CSUM_TSO;
2360 	}
2361 	ifp->if_capenable = ifp->if_capabilities;
2362 
2363 	/*
2364 	 * Figure out what sort of media we have by checking the
2365 	 * hardware config word in the first 32k of NIC internal memory,
2366 	 * or fall back to examining the EEPROM if necessary.
2367 	 * Note: on some BCM5700 cards, this value appears to be unset.
2368 	 * If that's the case, we have to rely on identifying the NIC
2369 	 * by its PCI subsystem ID, as we do below for the SysKonnect
2370 	 * SK-9D41.
2371 	 */
2372 	if (bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_SIG) == BGE_MAGIC_NUMBER) {
2373 		hwcfg = bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_NICCFG);
2374 	} else {
2375 		if (bge_read_eeprom(sc, (caddr_t)&hwcfg, BGE_EE_HWCFG_OFFSET,
2376 				    sizeof(hwcfg))) {
2377 			device_printf(dev, "failed to read EEPROM\n");
2378 			error = ENXIO;
2379 			goto fail;
2380 		}
2381 		hwcfg = ntohl(hwcfg);
2382 	}
2383 
2384 	/* The SysKonnect SK-9D41 is a 1000baseSX card. */
2385 	if (pci_get_subvendor(dev) == PCI_PRODUCT_SCHNEIDERKOCH_SK_9D41 ||
2386 	    (hwcfg & BGE_HWCFG_MEDIA) == BGE_MEDIA_FIBER) {
2387 		if (BGE_IS_5714_FAMILY(sc))
2388 			sc->bge_flags |= BGE_FLAG_MII_SERDES;
2389 		else
2390 			sc->bge_flags |= BGE_FLAG_TBI;
2391 	}
2392 
2393 	/* Setup MI MODE */
2394 	if (sc->bge_flags & BGE_FLAG_CPMU)
2395 		sc->bge_mi_mode = BGE_MIMODE_500KHZ_CONST;
2396 	else
2397 		sc->bge_mi_mode = BGE_MIMODE_BASE;
2398 	if (BGE_IS_5700_FAMILY(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5705) {
2399 		/* Enable auto polling for BCM570[0-5]. */
2400 		sc->bge_mi_mode |= BGE_MIMODE_AUTOPOLL;
2401 	}
2402 
2403 	/* Setup link status update stuffs */
2404 	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
2405 	    sc->bge_chipid != BGE_CHIPID_BCM5700_B2) {
2406 		sc->bge_link_upd = bge_bcm5700_link_upd;
2407 		sc->bge_link_chg = BGE_MACSTAT_MI_INTERRUPT;
2408 	} else if (sc->bge_flags & BGE_FLAG_TBI) {
2409 		sc->bge_link_upd = bge_tbi_link_upd;
2410 		sc->bge_link_chg = BGE_MACSTAT_LINK_CHANGED;
2411 	} else if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
2412 		sc->bge_link_upd = bge_autopoll_link_upd;
2413 		sc->bge_link_chg = BGE_MACSTAT_LINK_CHANGED;
2414 	} else {
2415 		sc->bge_link_upd = bge_copper_link_upd;
2416 		sc->bge_link_chg = BGE_MACSTAT_LINK_CHANGED;
2417 	}
2418 
2419 	/*
2420 	 * Broadcom's own driver always assumes the internal
2421 	 * PHY is at GMII address 1.  On some chips, the PHY responds
2422 	 * to accesses at all addresses, which could cause us to
2423 	 * bogusly attach the PHY 32 times at probe type.  Always
2424 	 * restricting the lookup to address 1 is simpler than
2425 	 * trying to figure out which chips revisions should be
2426 	 * special-cased.
2427 	 */
2428 	sc->bge_phyno = 1;
2429 
2430 	if (sc->bge_flags & BGE_FLAG_TBI) {
2431 		ifmedia_init(&sc->bge_ifmedia, IFM_IMASK,
2432 		    bge_ifmedia_upd, bge_ifmedia_sts);
2433 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER|IFM_1000_SX, 0, NULL);
2434 		ifmedia_add(&sc->bge_ifmedia,
2435 		    IFM_ETHER|IFM_1000_SX|IFM_FDX, 0, NULL);
2436 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER|IFM_AUTO, 0, NULL);
2437 		ifmedia_set(&sc->bge_ifmedia, IFM_ETHER|IFM_AUTO);
2438 		sc->bge_ifmedia.ifm_media = sc->bge_ifmedia.ifm_cur->ifm_media;
2439 	} else {
2440 		struct mii_probe_args mii_args;
2441 
2442 		mii_probe_args_init(&mii_args, bge_ifmedia_upd, bge_ifmedia_sts);
2443 		mii_args.mii_probemask = 1 << sc->bge_phyno;
2444 		mii_args.mii_capmask = capmask;
2445 		mii_args.mii_privtag = MII_PRIVTAG_BRGPHY;
2446 		mii_args.mii_priv = mii_priv;
2447 
2448 		error = mii_probe(dev, &sc->bge_miibus, &mii_args);
2449 		if (error) {
2450 			device_printf(dev, "MII without any PHY!\n");
2451 			goto fail;
2452 		}
2453 	}
2454 
2455 	/*
2456 	 * Create sysctl nodes.
2457 	 */
2458 	sysctl_ctx_init(&sc->bge_sysctl_ctx);
2459 	sc->bge_sysctl_tree = SYSCTL_ADD_NODE(&sc->bge_sysctl_ctx,
2460 					      SYSCTL_STATIC_CHILDREN(_hw),
2461 					      OID_AUTO,
2462 					      device_get_nameunit(dev),
2463 					      CTLFLAG_RD, 0, "");
2464 	if (sc->bge_sysctl_tree == NULL) {
2465 		device_printf(dev, "can't add sysctl node\n");
2466 		error = ENXIO;
2467 		goto fail;
2468 	}
2469 
2470 	SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2471 			SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2472 			OID_AUTO, "rx_coal_ticks",
2473 			CTLTYPE_INT | CTLFLAG_RW,
2474 			sc, 0, bge_sysctl_rx_coal_ticks, "I",
2475 			"Receive coalescing ticks (usec).");
2476 	SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2477 			SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2478 			OID_AUTO, "tx_coal_ticks",
2479 			CTLTYPE_INT | CTLFLAG_RW,
2480 			sc, 0, bge_sysctl_tx_coal_ticks, "I",
2481 			"Transmit coalescing ticks (usec).");
2482 	SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2483 			SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2484 			OID_AUTO, "rx_coal_bds",
2485 			CTLTYPE_INT | CTLFLAG_RW,
2486 			sc, 0, bge_sysctl_rx_coal_bds, "I",
2487 			"Receive max coalesced BD count.");
2488 	SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2489 			SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2490 			OID_AUTO, "tx_coal_bds",
2491 			CTLTYPE_INT | CTLFLAG_RW,
2492 			sc, 0, bge_sysctl_tx_coal_bds, "I",
2493 			"Transmit max coalesced BD count.");
2494 	if (sc->bge_flags & BGE_FLAG_PCIE) {
2495 		/*
2496 		 * A common design characteristic for many Broadcom
2497 		 * client controllers is that they only support a
2498 		 * single outstanding DMA read operation on the PCIe
2499 		 * bus. This means that it will take twice as long to
2500 		 * fetch a TX frame that is split into header and
2501 		 * payload buffers as it does to fetch a single,
2502 		 * contiguous TX frame (2 reads vs. 1 read). For these
2503 		 * controllers, coalescing buffers to reduce the number
2504 		 * of memory reads is effective way to get maximum
2505 		 * performance(about 940Mbps).  Without collapsing TX
2506 		 * buffers the maximum TCP bulk transfer performance
2507 		 * is about 850Mbps. However forcing coalescing mbufs
2508 		 * consumes a lot of CPU cycles, so leave it off by
2509 		 * default.
2510 		 */
2511 		SYSCTL_ADD_INT(&sc->bge_sysctl_ctx,
2512 			       SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2513 			       OID_AUTO, "force_defrag", CTLFLAG_RW,
2514 			       &sc->bge_force_defrag, 0,
2515 			       "Force defragment on TX path");
2516 	}
2517 	if (sc->bge_flags & BGE_FLAG_STATUS_TAG) {
2518 		if (!BGE_IS_5705_PLUS(sc)) {
2519 			SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2520 			    SYSCTL_CHILDREN(sc->bge_sysctl_tree), OID_AUTO,
2521 			    "rx_coal_ticks_int", CTLTYPE_INT | CTLFLAG_RW,
2522 			    sc, 0, bge_sysctl_rx_coal_ticks_int, "I",
2523 			    "Receive coalescing ticks "
2524 			    "during interrupt (usec).");
2525 			SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2526 			    SYSCTL_CHILDREN(sc->bge_sysctl_tree), OID_AUTO,
2527 			    "tx_coal_ticks_int", CTLTYPE_INT | CTLFLAG_RW,
2528 			    sc, 0, bge_sysctl_tx_coal_ticks_int, "I",
2529 			    "Transmit coalescing ticks "
2530 			    "during interrupt (usec).");
2531 		}
2532 		SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2533 		    SYSCTL_CHILDREN(sc->bge_sysctl_tree), OID_AUTO,
2534 		    "rx_coal_bds_int", CTLTYPE_INT | CTLFLAG_RW,
2535 		    sc, 0, bge_sysctl_rx_coal_bds_int, "I",
2536 		    "Receive max coalesced BD count during interrupt.");
2537 		SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2538 		    SYSCTL_CHILDREN(sc->bge_sysctl_tree), OID_AUTO,
2539 		    "tx_coal_bds_int", CTLTYPE_INT | CTLFLAG_RW,
2540 		    sc, 0, bge_sysctl_tx_coal_bds_int, "I",
2541 		    "Transmit max coalesced BD count during interrupt.");
2542 	}
2543 
2544 	/*
2545 	 * Call MI attach routine.
2546 	 */
2547 	ether_ifattach(ifp, ether_addr, NULL);
2548 
2549 #ifdef IFPOLL_ENABLE
2550 	/* Polling setup */
2551 	ifpoll_compat_setup(&sc->bge_npoll,
2552 	    &sc->bge_sysctl_ctx, sc->bge_sysctl_tree, device_get_unit(dev),
2553 	    ifp->if_serializer);
2554 #endif
2555 
2556 	if (sc->bge_irq_type == PCI_INTR_TYPE_MSI) {
2557 		if (sc->bge_flags & BGE_FLAG_ONESHOT_MSI) {
2558 			intr_func = bge_msi_oneshot;
2559 			if (bootverbose)
2560 				device_printf(dev, "oneshot MSI\n");
2561 		} else {
2562 			intr_func = bge_msi;
2563 		}
2564 	} else if (sc->bge_flags & BGE_FLAG_STATUS_TAG) {
2565 		intr_func = bge_intr_legacy;
2566 	} else {
2567 		intr_func = bge_intr_crippled;
2568 	}
2569 	error = bus_setup_intr(dev, sc->bge_irq, INTR_MPSAFE, intr_func, sc,
2570 	    &sc->bge_intrhand, ifp->if_serializer);
2571 	if (error) {
2572 		ether_ifdetach(ifp);
2573 		device_printf(dev, "couldn't set up irq\n");
2574 		goto fail;
2575 	}
2576 
2577 	ifp->if_cpuid = rman_get_cpuid(sc->bge_irq);
2578 	KKASSERT(ifp->if_cpuid >= 0 && ifp->if_cpuid < ncpus);
2579 
2580 	return(0);
2581 fail:
2582 	bge_detach(dev);
2583 	return(error);
2584 }
2585 
2586 static int
2587 bge_detach(device_t dev)
2588 {
2589 	struct bge_softc *sc = device_get_softc(dev);
2590 
2591 	if (device_is_attached(dev)) {
2592 		struct ifnet *ifp = &sc->arpcom.ac_if;
2593 
2594 		lwkt_serialize_enter(ifp->if_serializer);
2595 		bge_stop(sc);
2596 		bge_reset(sc);
2597 		bus_teardown_intr(dev, sc->bge_irq, sc->bge_intrhand);
2598 		lwkt_serialize_exit(ifp->if_serializer);
2599 
2600 		ether_ifdetach(ifp);
2601 	}
2602 
2603 	if (sc->bge_flags & BGE_FLAG_TBI)
2604 		ifmedia_removeall(&sc->bge_ifmedia);
2605 	if (sc->bge_miibus)
2606 		device_delete_child(dev, sc->bge_miibus);
2607 	bus_generic_detach(dev);
2608 
2609 	if (sc->bge_irq != NULL) {
2610 		bus_release_resource(dev, SYS_RES_IRQ, sc->bge_irq_rid,
2611 		    sc->bge_irq);
2612 	}
2613 	if (sc->bge_irq_type == PCI_INTR_TYPE_MSI)
2614 		pci_release_msi(dev);
2615 
2616 	if (sc->bge_res != NULL) {
2617 		bus_release_resource(dev, SYS_RES_MEMORY,
2618 		    BGE_PCI_BAR0, sc->bge_res);
2619 	}
2620 
2621 	if (sc->bge_sysctl_tree != NULL)
2622 		sysctl_ctx_free(&sc->bge_sysctl_ctx);
2623 
2624 	bge_dma_free(sc);
2625 
2626 	return 0;
2627 }
2628 
2629 static void
2630 bge_reset(struct bge_softc *sc)
2631 {
2632 	device_t dev;
2633 	uint32_t cachesize, command, pcistate, reset;
2634 	void (*write_op)(struct bge_softc *, uint32_t, uint32_t);
2635 	int i, val = 0;
2636 
2637 	dev = sc->bge_dev;
2638 
2639 	if (BGE_IS_575X_PLUS(sc) && !BGE_IS_5714_FAMILY(sc) &&
2640 	    sc->bge_asicrev != BGE_ASICREV_BCM5906) {
2641 		if (sc->bge_flags & BGE_FLAG_PCIE)
2642 			write_op = bge_writemem_direct;
2643 		else
2644 			write_op = bge_writemem_ind;
2645 	} else {
2646 		write_op = bge_writereg_ind;
2647 	}
2648 
2649 	/* Save some important PCI state. */
2650 	cachesize = pci_read_config(dev, BGE_PCI_CACHESZ, 4);
2651 	command = pci_read_config(dev, BGE_PCI_CMD, 4);
2652 	pcistate = pci_read_config(dev, BGE_PCI_PCISTATE, 4);
2653 
2654 	pci_write_config(dev, BGE_PCI_MISC_CTL,
2655 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2656 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW|
2657 	    sc->bge_pci_miscctl, 4);
2658 
2659 	/* Disable fastboot on controllers that support it. */
2660 	if (sc->bge_asicrev == BGE_ASICREV_BCM5752 ||
2661 	    BGE_IS_5755_PLUS(sc)) {
2662 		if (bootverbose)
2663 			if_printf(&sc->arpcom.ac_if, "Disabling fastboot\n");
2664 		CSR_WRITE_4(sc, BGE_FASTBOOT_PC, 0x0);
2665 	}
2666 
2667 	/*
2668 	 * Write the magic number to SRAM at offset 0xB50.
2669 	 * When firmware finishes its initialization it will
2670 	 * write ~BGE_MAGIC_NUMBER to the same location.
2671 	 */
2672 	bge_writemem_ind(sc, BGE_SOFTWARE_GENCOMM, BGE_MAGIC_NUMBER);
2673 
2674 	reset = BGE_MISCCFG_RESET_CORE_CLOCKS|(65<<1);
2675 
2676 	/* XXX: Broadcom Linux driver. */
2677 	if (sc->bge_flags & BGE_FLAG_PCIE) {
2678 		/* Force PCI-E 1.0a mode */
2679 		if (sc->bge_asicrev != BGE_ASICREV_BCM5785 &&
2680 		    CSR_READ_4(sc, BGE_PCIE_PHY_TSTCTL) ==
2681 		    (BGE_PCIE_PHY_TSTCTL_PSCRAM |
2682 		     BGE_PCIE_PHY_TSTCTL_PCIE10)) {
2683 			CSR_WRITE_4(sc, BGE_PCIE_PHY_TSTCTL,
2684 			    BGE_PCIE_PHY_TSTCTL_PSCRAM);
2685 		}
2686 		if (sc->bge_chipid != BGE_CHIPID_BCM5750_A0) {
2687 			/* Prevent PCIE link training during global reset */
2688 			CSR_WRITE_4(sc, BGE_MISC_CFG, (1<<29));
2689 			reset |= (1<<29);
2690 		}
2691 	}
2692 
2693 	/*
2694 	 * Set GPHY Power Down Override to leave GPHY
2695 	 * powered up in D0 uninitialized.
2696 	 */
2697 	if (BGE_IS_5705_PLUS(sc) && (sc->bge_flags & BGE_FLAG_CPMU) == 0)
2698 		reset |= BGE_MISCCFG_GPHY_PD_OVERRIDE;
2699 
2700 	/* Issue global reset */
2701 	write_op(sc, BGE_MISC_CFG, reset);
2702 
2703 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
2704 		uint32_t status, ctrl;
2705 
2706 		status = CSR_READ_4(sc, BGE_VCPU_STATUS);
2707 		CSR_WRITE_4(sc, BGE_VCPU_STATUS,
2708 		    status | BGE_VCPU_STATUS_DRV_RESET);
2709 		ctrl = CSR_READ_4(sc, BGE_VCPU_EXT_CTRL);
2710 		CSR_WRITE_4(sc, BGE_VCPU_EXT_CTRL,
2711 		    ctrl & ~BGE_VCPU_EXT_CTRL_HALT_CPU);
2712 	}
2713 
2714 	DELAY(1000);
2715 
2716 	/* XXX: Broadcom Linux driver. */
2717 	if (sc->bge_flags & BGE_FLAG_PCIE) {
2718 		uint16_t devctl;
2719 
2720 		if (sc->bge_chipid == BGE_CHIPID_BCM5750_A0) {
2721 			uint32_t v;
2722 
2723 			DELAY(500000); /* wait for link training to complete */
2724 			v = pci_read_config(dev, 0xc4, 4);
2725 			pci_write_config(dev, 0xc4, v | (1<<15), 4);
2726 		}
2727 
2728 		devctl = pci_read_config(dev,
2729 		    sc->bge_pciecap + PCIER_DEVCTRL, 2);
2730 
2731 		/* Disable no snoop and disable relaxed ordering. */
2732 		devctl &= ~(PCIEM_DEVCTL_RELAX_ORDER | PCIEM_DEVCTL_NOSNOOP);
2733 
2734 		/* Old PCI-E chips only support 128 bytes Max PayLoad Size. */
2735 		if ((sc->bge_flags & BGE_FLAG_CPMU) == 0) {
2736 			devctl &= ~PCIEM_DEVCTL_MAX_PAYLOAD_MASK;
2737 			devctl |= PCIEM_DEVCTL_MAX_PAYLOAD_128;
2738 		}
2739 
2740 		pci_write_config(dev, sc->bge_pciecap + PCIER_DEVCTRL,
2741 		    devctl, 2);
2742 
2743 		/* Clear error status. */
2744 		pci_write_config(dev, sc->bge_pciecap + PCIER_DEVSTS,
2745 		    PCIEM_DEVSTS_CORR_ERR |
2746 		    PCIEM_DEVSTS_NFATAL_ERR |
2747 		    PCIEM_DEVSTS_FATAL_ERR |
2748 		    PCIEM_DEVSTS_UNSUPP_REQ, 2);
2749 	}
2750 
2751 	/* Reset some of the PCI state that got zapped by reset */
2752 	pci_write_config(dev, BGE_PCI_MISC_CTL,
2753 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2754 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW|
2755 	    sc->bge_pci_miscctl, 4);
2756 	pci_write_config(dev, BGE_PCI_CACHESZ, cachesize, 4);
2757 	pci_write_config(dev, BGE_PCI_CMD, command, 4);
2758 	write_op(sc, BGE_MISC_CFG, (65 << 1));
2759 
2760 	/*
2761 	 * Disable PCI-X relaxed ordering to ensure status block update
2762 	 * comes first then packet buffer DMA. Otherwise driver may
2763 	 * read stale status block.
2764 	 */
2765 	if (sc->bge_flags & BGE_FLAG_PCIX) {
2766 		uint16_t devctl;
2767 
2768 		devctl = pci_read_config(dev,
2769 		    sc->bge_pcixcap + PCIXR_COMMAND, 2);
2770 		devctl &= ~PCIXM_COMMAND_ERO;
2771 		if (sc->bge_asicrev == BGE_ASICREV_BCM5703) {
2772 			devctl &= ~PCIXM_COMMAND_MAX_READ;
2773 			devctl |= PCIXM_COMMAND_MAX_READ_2048;
2774 		} else if (sc->bge_asicrev == BGE_ASICREV_BCM5704) {
2775 			devctl &= ~(PCIXM_COMMAND_MAX_SPLITS |
2776 			    PCIXM_COMMAND_MAX_READ);
2777 			devctl |= PCIXM_COMMAND_MAX_READ_2048;
2778 		}
2779 		pci_write_config(dev, sc->bge_pcixcap + PCIXR_COMMAND,
2780 		    devctl, 2);
2781 	}
2782 
2783 	/*
2784 	 * Enable memory arbiter and re-enable MSI if necessary.
2785 	 */
2786 	if (BGE_IS_5714_FAMILY(sc)) {
2787 		uint32_t val;
2788 
2789 		if (sc->bge_irq_type == PCI_INTR_TYPE_MSI) {
2790 			/*
2791 			 * Resetting BCM5714 family will clear MSI
2792 			 * enable bit; restore it after resetting.
2793 			 */
2794 			PCI_SETBIT(sc->bge_dev, sc->bge_msicap + PCIR_MSI_CTRL,
2795 			    PCIM_MSICTRL_MSI_ENABLE, 2);
2796 			BGE_SETBIT(sc, BGE_MSI_MODE, BGE_MSIMODE_ENABLE);
2797 		}
2798 		val = CSR_READ_4(sc, BGE_MARB_MODE);
2799 		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE | val);
2800 	} else {
2801 		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
2802 	}
2803 
2804 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
2805 		for (i = 0; i < BGE_TIMEOUT; i++) {
2806 			val = CSR_READ_4(sc, BGE_VCPU_STATUS);
2807 			if (val & BGE_VCPU_STATUS_INIT_DONE)
2808 				break;
2809 			DELAY(100);
2810 		}
2811 		if (i == BGE_TIMEOUT) {
2812 			if_printf(&sc->arpcom.ac_if, "reset timed out\n");
2813 			return;
2814 		}
2815 	} else {
2816 		/*
2817 		 * Poll until we see the 1's complement of the magic number.
2818 		 * This indicates that the firmware initialization
2819 		 * is complete.
2820 		 */
2821 		for (i = 0; i < BGE_FIRMWARE_TIMEOUT; i++) {
2822 			val = bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM);
2823 			if (val == ~BGE_MAGIC_NUMBER)
2824 				break;
2825 			DELAY(10);
2826 		}
2827 		if (i == BGE_FIRMWARE_TIMEOUT) {
2828 			if_printf(&sc->arpcom.ac_if, "firmware handshake "
2829 				  "timed out, found 0x%08x\n", val);
2830 		}
2831 	}
2832 
2833 	/*
2834 	 * XXX Wait for the value of the PCISTATE register to
2835 	 * return to its original pre-reset state. This is a
2836 	 * fairly good indicator of reset completion. If we don't
2837 	 * wait for the reset to fully complete, trying to read
2838 	 * from the device's non-PCI registers may yield garbage
2839 	 * results.
2840 	 */
2841 	for (i = 0; i < BGE_TIMEOUT; i++) {
2842 		if (pci_read_config(dev, BGE_PCI_PCISTATE, 4) == pcistate)
2843 			break;
2844 		DELAY(10);
2845 	}
2846 
2847 	/* Fix up byte swapping */
2848 	CSR_WRITE_4(sc, BGE_MODE_CTL, BGE_DMA_SWAP_OPTIONS |
2849 	    BGE_MODECTL_BYTESWAP_DATA);
2850 
2851 	CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
2852 
2853 	/*
2854 	 * The 5704 in TBI mode apparently needs some special
2855 	 * adjustment to insure the SERDES drive level is set
2856 	 * to 1.2V.
2857 	 */
2858 	if (sc->bge_asicrev == BGE_ASICREV_BCM5704 &&
2859 	    (sc->bge_flags & BGE_FLAG_TBI)) {
2860 		uint32_t serdescfg;
2861 
2862 		serdescfg = CSR_READ_4(sc, BGE_SERDES_CFG);
2863 		serdescfg = (serdescfg & ~0xFFF) | 0x880;
2864 		CSR_WRITE_4(sc, BGE_SERDES_CFG, serdescfg);
2865 	}
2866 
2867 	/* XXX: Broadcom Linux driver. */
2868 	if ((sc->bge_flags & BGE_FLAG_PCIE) &&
2869 	    sc->bge_chipid != BGE_CHIPID_BCM5750_A0 &&
2870 	    sc->bge_asicrev != BGE_ASICREV_BCM5785) {
2871 		uint32_t v;
2872 
2873 		/* Enable Data FIFO protection. */
2874 		v = CSR_READ_4(sc, BGE_PCIE_TLDLPL_PORT);
2875 		CSR_WRITE_4(sc, BGE_PCIE_TLDLPL_PORT, v | (1 << 25));
2876 	}
2877 
2878 	DELAY(10000);
2879 }
2880 
2881 /*
2882  * Frame reception handling. This is called if there's a frame
2883  * on the receive return list.
2884  *
2885  * Note: we have to be able to handle two possibilities here:
2886  * 1) the frame is from the jumbo recieve ring
2887  * 2) the frame is from the standard receive ring
2888  */
2889 
2890 static void
2891 bge_rxeof(struct bge_softc *sc, uint16_t rx_prod)
2892 {
2893 	struct ifnet *ifp;
2894 	int stdcnt = 0, jumbocnt = 0;
2895 
2896 	ifp = &sc->arpcom.ac_if;
2897 
2898 	while (sc->bge_rx_saved_considx != rx_prod) {
2899 		struct bge_rx_bd	*cur_rx;
2900 		uint32_t		rxidx;
2901 		struct mbuf		*m = NULL;
2902 		uint16_t		vlan_tag = 0;
2903 		int			have_tag = 0;
2904 
2905 		cur_rx =
2906 	    &sc->bge_ldata.bge_rx_return_ring[sc->bge_rx_saved_considx];
2907 
2908 		rxidx = cur_rx->bge_idx;
2909 		BGE_INC(sc->bge_rx_saved_considx, sc->bge_return_ring_cnt);
2910 		logif(rx_pkt);
2911 
2912 		if (cur_rx->bge_flags & BGE_RXBDFLAG_VLAN_TAG) {
2913 			have_tag = 1;
2914 			vlan_tag = cur_rx->bge_vlan_tag;
2915 		}
2916 
2917 		if (cur_rx->bge_flags & BGE_RXBDFLAG_JUMBO_RING) {
2918 			BGE_INC(sc->bge_jumbo, BGE_JUMBO_RX_RING_CNT);
2919 			jumbocnt++;
2920 
2921 			if (rxidx != sc->bge_jumbo) {
2922 				ifp->if_ierrors++;
2923 				if_printf(ifp, "sw jumbo index(%d) "
2924 				    "and hw jumbo index(%d) mismatch, drop!\n",
2925 				    sc->bge_jumbo, rxidx);
2926 				bge_setup_rxdesc_jumbo(sc, rxidx);
2927 				continue;
2928 			}
2929 
2930 			m = sc->bge_cdata.bge_rx_jumbo_chain[rxidx].bge_mbuf;
2931 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
2932 				ifp->if_ierrors++;
2933 				bge_setup_rxdesc_jumbo(sc, sc->bge_jumbo);
2934 				continue;
2935 			}
2936 			if (bge_newbuf_jumbo(sc, sc->bge_jumbo, 0)) {
2937 				ifp->if_ierrors++;
2938 				bge_setup_rxdesc_jumbo(sc, sc->bge_jumbo);
2939 				continue;
2940 			}
2941 		} else {
2942 			BGE_INC(sc->bge_std, BGE_STD_RX_RING_CNT);
2943 			stdcnt++;
2944 
2945 			if (rxidx != sc->bge_std) {
2946 				ifp->if_ierrors++;
2947 				if_printf(ifp, "sw std index(%d) "
2948 				    "and hw std index(%d) mismatch, drop!\n",
2949 				    sc->bge_std, rxidx);
2950 				bge_setup_rxdesc_std(sc, rxidx);
2951 				continue;
2952 			}
2953 
2954 			m = sc->bge_cdata.bge_rx_std_chain[rxidx].bge_mbuf;
2955 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
2956 				ifp->if_ierrors++;
2957 				bge_setup_rxdesc_std(sc, sc->bge_std);
2958 				continue;
2959 			}
2960 			if (bge_newbuf_std(sc, sc->bge_std, 0)) {
2961 				ifp->if_ierrors++;
2962 				bge_setup_rxdesc_std(sc, sc->bge_std);
2963 				continue;
2964 			}
2965 		}
2966 
2967 		ifp->if_ipackets++;
2968 #if !defined(__i386__) && !defined(__x86_64__)
2969 		/*
2970 		 * The x86 allows unaligned accesses, but for other
2971 		 * platforms we must make sure the payload is aligned.
2972 		 */
2973 		if (sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) {
2974 			bcopy(m->m_data, m->m_data + ETHER_ALIGN,
2975 			    cur_rx->bge_len);
2976 			m->m_data += ETHER_ALIGN;
2977 		}
2978 #endif
2979 		m->m_pkthdr.len = m->m_len = cur_rx->bge_len - ETHER_CRC_LEN;
2980 		m->m_pkthdr.rcvif = ifp;
2981 
2982 		if (ifp->if_capenable & IFCAP_RXCSUM) {
2983 			if (cur_rx->bge_flags & BGE_RXBDFLAG_IP_CSUM) {
2984 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2985 				if ((cur_rx->bge_ip_csum ^ 0xffff) == 0)
2986 					m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2987 			}
2988 			if ((cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM) &&
2989 			    m->m_pkthdr.len >= BGE_MIN_FRAMELEN) {
2990 				m->m_pkthdr.csum_data =
2991 					cur_rx->bge_tcp_udp_csum;
2992 				m->m_pkthdr.csum_flags |=
2993 					CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2994 			}
2995 		}
2996 
2997 		/*
2998 		 * If we received a packet with a vlan tag, pass it
2999 		 * to vlan_input() instead of ether_input().
3000 		 */
3001 		if (have_tag) {
3002 			m->m_flags |= M_VLANTAG;
3003 			m->m_pkthdr.ether_vlantag = vlan_tag;
3004 			have_tag = vlan_tag = 0;
3005 		}
3006 		ifp->if_input(ifp, m);
3007 	}
3008 
3009 	bge_writembx(sc, BGE_MBX_RX_CONS0_LO, sc->bge_rx_saved_considx);
3010 	if (stdcnt)
3011 		bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bge_std);
3012 	if (jumbocnt)
3013 		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
3014 }
3015 
3016 static void
3017 bge_txeof(struct bge_softc *sc, uint16_t tx_cons)
3018 {
3019 	struct ifnet *ifp;
3020 
3021 	ifp = &sc->arpcom.ac_if;
3022 
3023 	/*
3024 	 * Go through our tx ring and free mbufs for those
3025 	 * frames that have been sent.
3026 	 */
3027 	while (sc->bge_tx_saved_considx != tx_cons) {
3028 		uint32_t idx = 0;
3029 
3030 		idx = sc->bge_tx_saved_considx;
3031 		if (sc->bge_cdata.bge_tx_chain[idx] != NULL) {
3032 			ifp->if_opackets++;
3033 			bus_dmamap_unload(sc->bge_cdata.bge_tx_mtag,
3034 			    sc->bge_cdata.bge_tx_dmamap[idx]);
3035 			m_freem(sc->bge_cdata.bge_tx_chain[idx]);
3036 			sc->bge_cdata.bge_tx_chain[idx] = NULL;
3037 		}
3038 		sc->bge_txcnt--;
3039 		BGE_INC(sc->bge_tx_saved_considx, BGE_TX_RING_CNT);
3040 		logif(tx_pkt);
3041 	}
3042 
3043 	if ((BGE_TX_RING_CNT - sc->bge_txcnt) >=
3044 	    (sc->bge_txrsvd + sc->bge_txspare))
3045 		ifp->if_flags &= ~IFF_OACTIVE;
3046 
3047 	if (sc->bge_txcnt == 0)
3048 		ifp->if_timer = 0;
3049 
3050 	if (!ifq_is_empty(&ifp->if_snd))
3051 		if_devstart(ifp);
3052 }
3053 
3054 #ifdef IFPOLL_ENABLE
3055 
3056 static void
3057 bge_npoll_compat(struct ifnet *ifp, void *arg __unused, int cycles __unused)
3058 {
3059 	struct bge_softc *sc = ifp->if_softc;
3060 	struct bge_status_block *sblk = sc->bge_ldata.bge_status_block;
3061 	uint16_t rx_prod, tx_cons;
3062 
3063 	ASSERT_SERIALIZED(ifp->if_serializer);
3064 
3065 	if (sc->bge_npoll.ifpc_stcount-- == 0) {
3066 		sc->bge_npoll.ifpc_stcount = sc->bge_npoll.ifpc_stfrac;
3067 		/*
3068 		 * Process link state changes.
3069 		 */
3070 		bge_link_poll(sc);
3071 	}
3072 
3073 	if (sc->bge_flags & BGE_FLAG_STATUS_TAG) {
3074 		sc->bge_status_tag = sblk->bge_status_tag;
3075 		/*
3076 		 * Use a load fence to ensure that status_tag
3077 		 * is saved  before rx_prod and tx_cons.
3078 		 */
3079 		cpu_lfence();
3080 	}
3081 
3082 	rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
3083 	if (sc->bge_rx_saved_considx != rx_prod)
3084 		bge_rxeof(sc, rx_prod);
3085 
3086 	tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
3087 	if (sc->bge_tx_saved_considx != tx_cons)
3088 		bge_txeof(sc, tx_cons);
3089 
3090 	if (sc->bge_flags & BGE_FLAG_STATUS_TAG)
3091 		bge_writembx(sc, BGE_MBX_IRQ0_LO, sc->bge_status_tag << 24);
3092 }
3093 
3094 static void
3095 bge_npoll(struct ifnet *ifp, struct ifpoll_info *info)
3096 {
3097 	struct bge_softc *sc = ifp->if_softc;
3098 
3099 	ASSERT_SERIALIZED(ifp->if_serializer);
3100 
3101 	if (info != NULL) {
3102 		int cpuid = sc->bge_npoll.ifpc_cpuid;
3103 
3104 		info->ifpi_rx[cpuid].poll_func = bge_npoll_compat;
3105 		info->ifpi_rx[cpuid].arg = NULL;
3106 		info->ifpi_rx[cpuid].serializer = ifp->if_serializer;
3107 
3108 		if (ifp->if_flags & IFF_RUNNING)
3109 			bge_disable_intr(sc);
3110 		ifp->if_npoll_cpuid = cpuid;
3111 	} else {
3112 		if (ifp->if_flags & IFF_RUNNING)
3113 			bge_enable_intr(sc);
3114 		ifp->if_npoll_cpuid = -1;
3115 	}
3116 }
3117 
3118 #endif	/* IFPOLL_ENABLE */
3119 
3120 static void
3121 bge_intr_crippled(void *xsc)
3122 {
3123 	struct bge_softc *sc = xsc;
3124 	struct ifnet *ifp = &sc->arpcom.ac_if;
3125 
3126 	logif(intr);
3127 
3128  	/*
3129 	 * Ack the interrupt by writing something to BGE_MBX_IRQ0_LO.  Don't
3130 	 * disable interrupts by writing nonzero like we used to, since with
3131 	 * our current organization this just gives complications and
3132 	 * pessimizations for re-enabling interrupts.  We used to have races
3133 	 * instead of the necessary complications.  Disabling interrupts
3134 	 * would just reduce the chance of a status update while we are
3135 	 * running (by switching to the interrupt-mode coalescence
3136 	 * parameters), but this chance is already very low so it is more
3137 	 * efficient to get another interrupt than prevent it.
3138 	 *
3139 	 * We do the ack first to ensure another interrupt if there is a
3140 	 * status update after the ack.  We don't check for the status
3141 	 * changing later because it is more efficient to get another
3142 	 * interrupt than prevent it, not quite as above (not checking is
3143 	 * a smaller optimization than not toggling the interrupt enable,
3144 	 * since checking doesn't involve PCI accesses and toggling require
3145 	 * the status check).  So toggling would probably be a pessimization
3146 	 * even with MSI.  It would only be needed for using a task queue.
3147 	 */
3148 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 0);
3149 
3150 	/*
3151 	 * Process link state changes.
3152 	 */
3153 	bge_link_poll(sc);
3154 
3155 	if (ifp->if_flags & IFF_RUNNING) {
3156 		struct bge_status_block *sblk = sc->bge_ldata.bge_status_block;
3157 		uint16_t rx_prod, tx_cons;
3158 
3159 		rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
3160 		if (sc->bge_rx_saved_considx != rx_prod)
3161 			bge_rxeof(sc, rx_prod);
3162 
3163 		tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
3164 		if (sc->bge_tx_saved_considx != tx_cons)
3165 			bge_txeof(sc, tx_cons);
3166 	}
3167 
3168 	if (sc->bge_coal_chg)
3169 		bge_coal_change(sc);
3170 }
3171 
3172 static void
3173 bge_intr_legacy(void *xsc)
3174 {
3175 	struct bge_softc *sc = xsc;
3176 	struct bge_status_block *sblk = sc->bge_ldata.bge_status_block;
3177 
3178 	if (sc->bge_status_tag == sblk->bge_status_tag) {
3179 		uint32_t val;
3180 
3181 		val = pci_read_config(sc->bge_dev, BGE_PCI_PCISTATE, 4);
3182 		if (val & BGE_PCISTAT_INTR_NOTACT)
3183 			return;
3184 	}
3185 
3186 	/*
3187 	 * NOTE:
3188 	 * Interrupt will have to be disabled if tagged status
3189 	 * is used, else interrupt will always be asserted on
3190 	 * certain chips (at least on BCM5750 AX/BX).
3191 	 */
3192 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
3193 
3194 	bge_intr(sc);
3195 }
3196 
3197 static void
3198 bge_msi(void *xsc)
3199 {
3200 	struct bge_softc *sc = xsc;
3201 
3202 	/* Disable interrupt first */
3203 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
3204 	bge_intr(sc);
3205 }
3206 
3207 static void
3208 bge_msi_oneshot(void *xsc)
3209 {
3210 	bge_intr(xsc);
3211 }
3212 
3213 static void
3214 bge_intr(struct bge_softc *sc)
3215 {
3216 	struct ifnet *ifp = &sc->arpcom.ac_if;
3217 	struct bge_status_block *sblk = sc->bge_ldata.bge_status_block;
3218 	uint16_t rx_prod, tx_cons;
3219 	uint32_t status;
3220 
3221 	sc->bge_status_tag = sblk->bge_status_tag;
3222 	/*
3223 	 * Use a load fence to ensure that status_tag is saved
3224 	 * before rx_prod, tx_cons and status.
3225 	 */
3226 	cpu_lfence();
3227 
3228 	rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
3229 	tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
3230 	status = sblk->bge_status;
3231 
3232 	if ((status & BGE_STATFLAG_LINKSTATE_CHANGED) || sc->bge_link_evt)
3233 		bge_link_poll(sc);
3234 
3235 	if (ifp->if_flags & IFF_RUNNING) {
3236 		if (sc->bge_rx_saved_considx != rx_prod)
3237 			bge_rxeof(sc, rx_prod);
3238 
3239 		if (sc->bge_tx_saved_considx != tx_cons)
3240 			bge_txeof(sc, tx_cons);
3241 	}
3242 
3243 	bge_writembx(sc, BGE_MBX_IRQ0_LO, sc->bge_status_tag << 24);
3244 
3245 	if (sc->bge_coal_chg)
3246 		bge_coal_change(sc);
3247 }
3248 
3249 static void
3250 bge_tick(void *xsc)
3251 {
3252 	struct bge_softc *sc = xsc;
3253 	struct ifnet *ifp = &sc->arpcom.ac_if;
3254 
3255 	lwkt_serialize_enter(ifp->if_serializer);
3256 
3257 	if (BGE_IS_5705_PLUS(sc))
3258 		bge_stats_update_regs(sc);
3259 	else
3260 		bge_stats_update(sc);
3261 
3262 	if (sc->bge_flags & BGE_FLAG_TBI) {
3263 		/*
3264 		 * Since in TBI mode auto-polling can't be used we should poll
3265 		 * link status manually. Here we register pending link event
3266 		 * and trigger interrupt.
3267 		 */
3268 		sc->bge_link_evt++;
3269 		if (BGE_IS_CRIPPLED(sc))
3270 			BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
3271 		else
3272 			BGE_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
3273 	} else if (!sc->bge_link) {
3274 		mii_tick(device_get_softc(sc->bge_miibus));
3275 	}
3276 
3277 	callout_reset(&sc->bge_stat_timer, hz, bge_tick, sc);
3278 
3279 	lwkt_serialize_exit(ifp->if_serializer);
3280 }
3281 
3282 static void
3283 bge_stats_update_regs(struct bge_softc *sc)
3284 {
3285 	struct ifnet *ifp = &sc->arpcom.ac_if;
3286 	struct bge_mac_stats_regs stats;
3287 	uint32_t *s;
3288 	int i;
3289 
3290 	s = (uint32_t *)&stats;
3291 	for (i = 0; i < sizeof(struct bge_mac_stats_regs); i += 4) {
3292 		*s = CSR_READ_4(sc, BGE_RX_STATS + i);
3293 		s++;
3294 	}
3295 
3296 	ifp->if_collisions +=
3297 	   (stats.dot3StatsSingleCollisionFrames +
3298 	   stats.dot3StatsMultipleCollisionFrames +
3299 	   stats.dot3StatsExcessiveCollisions +
3300 	   stats.dot3StatsLateCollisions) -
3301 	   ifp->if_collisions;
3302 }
3303 
3304 static void
3305 bge_stats_update(struct bge_softc *sc)
3306 {
3307 	struct ifnet *ifp = &sc->arpcom.ac_if;
3308 	bus_size_t stats;
3309 
3310 	stats = BGE_MEMWIN_START + BGE_STATS_BLOCK;
3311 
3312 #define READ_STAT(sc, stats, stat)	\
3313 	CSR_READ_4(sc, stats + offsetof(struct bge_stats, stat))
3314 
3315 	ifp->if_collisions +=
3316 	   (READ_STAT(sc, stats,
3317 		txstats.dot3StatsSingleCollisionFrames.bge_addr_lo) +
3318 	    READ_STAT(sc, stats,
3319 		txstats.dot3StatsMultipleCollisionFrames.bge_addr_lo) +
3320 	    READ_STAT(sc, stats,
3321 		txstats.dot3StatsExcessiveCollisions.bge_addr_lo) +
3322 	    READ_STAT(sc, stats,
3323 		txstats.dot3StatsLateCollisions.bge_addr_lo)) -
3324 	   ifp->if_collisions;
3325 
3326 #undef READ_STAT
3327 
3328 #ifdef notdef
3329 	ifp->if_collisions +=
3330 	   (sc->bge_rdata->bge_info.bge_stats.dot3StatsSingleCollisionFrames +
3331 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsMultipleCollisionFrames +
3332 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsExcessiveCollisions +
3333 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsLateCollisions) -
3334 	   ifp->if_collisions;
3335 #endif
3336 }
3337 
3338 /*
3339  * Encapsulate an mbuf chain in the tx ring  by coupling the mbuf data
3340  * pointers to descriptors.
3341  */
3342 static int
3343 bge_encap(struct bge_softc *sc, struct mbuf **m_head0, uint32_t *txidx)
3344 {
3345 	struct bge_tx_bd *d = NULL, *last_d;
3346 	uint16_t csum_flags = 0, mss = 0;
3347 	bus_dma_segment_t segs[BGE_NSEG_NEW];
3348 	bus_dmamap_t map;
3349 	int error, maxsegs, nsegs, idx, i;
3350 	struct mbuf *m_head = *m_head0, *m_new;
3351 
3352 	if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
3353 		error = bge_setup_tso(sc, m_head0, &mss, &csum_flags);
3354 		if (error)
3355 			return ENOBUFS;
3356 		m_head = *m_head0;
3357 	} else if (m_head->m_pkthdr.csum_flags & BGE_CSUM_FEATURES) {
3358 		if (m_head->m_pkthdr.csum_flags & CSUM_IP)
3359 			csum_flags |= BGE_TXBDFLAG_IP_CSUM;
3360 		if (m_head->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP))
3361 			csum_flags |= BGE_TXBDFLAG_TCP_UDP_CSUM;
3362 		if (m_head->m_flags & M_LASTFRAG)
3363 			csum_flags |= BGE_TXBDFLAG_IP_FRAG_END;
3364 		else if (m_head->m_flags & M_FRAG)
3365 			csum_flags |= BGE_TXBDFLAG_IP_FRAG;
3366 	}
3367 
3368 	idx = *txidx;
3369 	map = sc->bge_cdata.bge_tx_dmamap[idx];
3370 
3371 	maxsegs = (BGE_TX_RING_CNT - sc->bge_txcnt) - sc->bge_txrsvd;
3372 	KASSERT(maxsegs >= sc->bge_txspare,
3373 		("not enough segments %d", maxsegs));
3374 
3375 	if (maxsegs > BGE_NSEG_NEW)
3376 		maxsegs = BGE_NSEG_NEW;
3377 
3378 	/*
3379 	 * Pad outbound frame to BGE_MIN_FRAMELEN for an unusual reason.
3380 	 * The bge hardware will pad out Tx runts to BGE_MIN_FRAMELEN,
3381 	 * but when such padded frames employ the bge IP/TCP checksum
3382 	 * offload, the hardware checksum assist gives incorrect results
3383 	 * (possibly from incorporating its own padding into the UDP/TCP
3384 	 * checksum; who knows).  If we pad such runts with zeros, the
3385 	 * onboard checksum comes out correct.
3386 	 */
3387 	if ((csum_flags & BGE_TXBDFLAG_TCP_UDP_CSUM) &&
3388 	    m_head->m_pkthdr.len < BGE_MIN_FRAMELEN) {
3389 		error = m_devpad(m_head, BGE_MIN_FRAMELEN);
3390 		if (error)
3391 			goto back;
3392 	}
3393 
3394 	if ((sc->bge_flags & BGE_FLAG_SHORTDMA) && m_head->m_next != NULL) {
3395 		m_new = bge_defrag_shortdma(m_head);
3396 		if (m_new == NULL) {
3397 			error = ENOBUFS;
3398 			goto back;
3399 		}
3400 		*m_head0 = m_head = m_new;
3401 	}
3402 	if ((m_head->m_pkthdr.csum_flags & CSUM_TSO) == 0 &&
3403 	    sc->bge_force_defrag && (sc->bge_flags & BGE_FLAG_PCIE) &&
3404 	    m_head->m_next != NULL) {
3405 		/*
3406 		 * Forcefully defragment mbuf chain to overcome hardware
3407 		 * limitation which only support a single outstanding
3408 		 * DMA read operation.  If it fails, keep moving on using
3409 		 * the original mbuf chain.
3410 		 */
3411 		m_new = m_defrag(m_head, MB_DONTWAIT);
3412 		if (m_new != NULL)
3413 			*m_head0 = m_head = m_new;
3414 	}
3415 
3416 	error = bus_dmamap_load_mbuf_defrag(sc->bge_cdata.bge_tx_mtag, map,
3417 			m_head0, segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
3418 	if (error)
3419 		goto back;
3420 
3421 	m_head = *m_head0;
3422 	bus_dmamap_sync(sc->bge_cdata.bge_tx_mtag, map, BUS_DMASYNC_PREWRITE);
3423 
3424 	for (i = 0; ; i++) {
3425 		d = &sc->bge_ldata.bge_tx_ring[idx];
3426 
3427 		d->bge_addr.bge_addr_lo = BGE_ADDR_LO(segs[i].ds_addr);
3428 		d->bge_addr.bge_addr_hi = BGE_ADDR_HI(segs[i].ds_addr);
3429 		d->bge_len = segs[i].ds_len;
3430 		d->bge_flags = csum_flags;
3431 		d->bge_mss = mss;
3432 
3433 		if (i == nsegs - 1)
3434 			break;
3435 		BGE_INC(idx, BGE_TX_RING_CNT);
3436 	}
3437 	last_d = d;
3438 
3439 	/* Set vlan tag to the first segment of the packet. */
3440 	d = &sc->bge_ldata.bge_tx_ring[*txidx];
3441 	if (m_head->m_flags & M_VLANTAG) {
3442 		d->bge_flags |= BGE_TXBDFLAG_VLAN_TAG;
3443 		d->bge_vlan_tag = m_head->m_pkthdr.ether_vlantag;
3444 	} else {
3445 		d->bge_vlan_tag = 0;
3446 	}
3447 
3448 	/* Mark the last segment as end of packet... */
3449 	last_d->bge_flags |= BGE_TXBDFLAG_END;
3450 
3451 	/*
3452 	 * Insure that the map for this transmission is placed at
3453 	 * the array index of the last descriptor in this chain.
3454 	 */
3455 	sc->bge_cdata.bge_tx_dmamap[*txidx] = sc->bge_cdata.bge_tx_dmamap[idx];
3456 	sc->bge_cdata.bge_tx_dmamap[idx] = map;
3457 	sc->bge_cdata.bge_tx_chain[idx] = m_head;
3458 	sc->bge_txcnt += nsegs;
3459 
3460 	BGE_INC(idx, BGE_TX_RING_CNT);
3461 	*txidx = idx;
3462 back:
3463 	if (error) {
3464 		m_freem(*m_head0);
3465 		*m_head0 = NULL;
3466 	}
3467 	return error;
3468 }
3469 
3470 /*
3471  * Main transmit routine. To avoid having to do mbuf copies, we put pointers
3472  * to the mbuf data regions directly in the transmit descriptors.
3473  */
3474 static void
3475 bge_start(struct ifnet *ifp)
3476 {
3477 	struct bge_softc *sc = ifp->if_softc;
3478 	struct mbuf *m_head = NULL;
3479 	uint32_t prodidx;
3480 	int need_trans;
3481 
3482 	if ((ifp->if_flags & (IFF_RUNNING | IFF_OACTIVE)) != IFF_RUNNING)
3483 		return;
3484 
3485 	prodidx = sc->bge_tx_prodidx;
3486 
3487 	need_trans = 0;
3488 	while (sc->bge_cdata.bge_tx_chain[prodidx] == NULL) {
3489 		m_head = ifq_dequeue(&ifp->if_snd, NULL);
3490 		if (m_head == NULL)
3491 			break;
3492 
3493 		/*
3494 		 * XXX
3495 		 * The code inside the if() block is never reached since we
3496 		 * must mark CSUM_IP_FRAGS in our if_hwassist to start getting
3497 		 * requests to checksum TCP/UDP in a fragmented packet.
3498 		 *
3499 		 * XXX
3500 		 * safety overkill.  If this is a fragmented packet chain
3501 		 * with delayed TCP/UDP checksums, then only encapsulate
3502 		 * it if we have enough descriptors to handle the entire
3503 		 * chain at once.
3504 		 * (paranoia -- may not actually be needed)
3505 		 */
3506 		if ((m_head->m_flags & M_FIRSTFRAG) &&
3507 		    (m_head->m_pkthdr.csum_flags & CSUM_DELAY_DATA)) {
3508 			if ((BGE_TX_RING_CNT - sc->bge_txcnt) <
3509 			    m_head->m_pkthdr.csum_data + sc->bge_txrsvd) {
3510 				ifp->if_flags |= IFF_OACTIVE;
3511 				ifq_prepend(&ifp->if_snd, m_head);
3512 				break;
3513 			}
3514 		}
3515 
3516 		/*
3517 		 * Sanity check: avoid coming within bge_txrsvd
3518 		 * descriptors of the end of the ring.  Also make
3519 		 * sure there are bge_txspare descriptors for
3520 		 * jumbo buffers' defragmentation.
3521 		 */
3522 		if ((BGE_TX_RING_CNT - sc->bge_txcnt) <
3523 		    (sc->bge_txrsvd + sc->bge_txspare)) {
3524 			ifp->if_flags |= IFF_OACTIVE;
3525 			ifq_prepend(&ifp->if_snd, m_head);
3526 			break;
3527 		}
3528 
3529 		/*
3530 		 * Pack the data into the transmit ring. If we
3531 		 * don't have room, set the OACTIVE flag and wait
3532 		 * for the NIC to drain the ring.
3533 		 */
3534 		if (bge_encap(sc, &m_head, &prodidx)) {
3535 			ifp->if_flags |= IFF_OACTIVE;
3536 			ifp->if_oerrors++;
3537 			break;
3538 		}
3539 		need_trans = 1;
3540 
3541 		ETHER_BPF_MTAP(ifp, m_head);
3542 	}
3543 
3544 	if (!need_trans)
3545 		return;
3546 
3547 	/* Transmit */
3548 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
3549 	/* 5700 b2 errata */
3550 	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
3551 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
3552 
3553 	sc->bge_tx_prodidx = prodidx;
3554 
3555 	/*
3556 	 * Set a timeout in case the chip goes out to lunch.
3557 	 */
3558 	ifp->if_timer = 5;
3559 }
3560 
3561 static void
3562 bge_init(void *xsc)
3563 {
3564 	struct bge_softc *sc = xsc;
3565 	struct ifnet *ifp = &sc->arpcom.ac_if;
3566 	uint16_t *m;
3567 	uint32_t mode;
3568 
3569 	ASSERT_SERIALIZED(ifp->if_serializer);
3570 
3571 	/* Cancel pending I/O and flush buffers. */
3572 	bge_stop(sc);
3573 	bge_reset(sc);
3574 	bge_chipinit(sc);
3575 
3576 	/*
3577 	 * Init the various state machines, ring
3578 	 * control blocks and firmware.
3579 	 */
3580 	if (bge_blockinit(sc)) {
3581 		if_printf(ifp, "initialization failure\n");
3582 		bge_stop(sc);
3583 		return;
3584 	}
3585 
3586 	/* Specify MTU. */
3587 	CSR_WRITE_4(sc, BGE_RX_MTU, ifp->if_mtu +
3588 	    ETHER_HDR_LEN + ETHER_CRC_LEN + EVL_ENCAPLEN);
3589 
3590 	/* Load our MAC address. */
3591 	m = (uint16_t *)&sc->arpcom.ac_enaddr[0];
3592 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_LO, htons(m[0]));
3593 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_HI, (htons(m[1]) << 16) | htons(m[2]));
3594 
3595 	/* Enable or disable promiscuous mode as needed. */
3596 	bge_setpromisc(sc);
3597 
3598 	/* Program multicast filter. */
3599 	bge_setmulti(sc);
3600 
3601 	/* Init RX ring. */
3602 	if (bge_init_rx_ring_std(sc)) {
3603 		if_printf(ifp, "RX ring initialization failed\n");
3604 		bge_stop(sc);
3605 		return;
3606 	}
3607 
3608 	/*
3609 	 * Workaround for a bug in 5705 ASIC rev A0. Poll the NIC's
3610 	 * memory to insure that the chip has in fact read the first
3611 	 * entry of the ring.
3612 	 */
3613 	if (sc->bge_chipid == BGE_CHIPID_BCM5705_A0) {
3614 		uint32_t		v, i;
3615 		for (i = 0; i < 10; i++) {
3616 			DELAY(20);
3617 			v = bge_readmem_ind(sc, BGE_STD_RX_RINGS + 8);
3618 			if (v == (MCLBYTES - ETHER_ALIGN))
3619 				break;
3620 		}
3621 		if (i == 10)
3622 			if_printf(ifp, "5705 A0 chip failed to load RX ring\n");
3623 	}
3624 
3625 	/* Init jumbo RX ring. */
3626 	if (ifp->if_mtu > (ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN)) {
3627 		if (bge_init_rx_ring_jumbo(sc)) {
3628 			if_printf(ifp, "Jumbo RX ring initialization failed\n");
3629 			bge_stop(sc);
3630 			return;
3631 		}
3632 	}
3633 
3634 	/* Init our RX return ring index */
3635 	sc->bge_rx_saved_considx = 0;
3636 
3637 	/* Init TX ring. */
3638 	bge_init_tx_ring(sc);
3639 
3640 	/* Enable TX MAC state machine lockup fix. */
3641 	mode = CSR_READ_4(sc, BGE_TX_MODE);
3642 	if (BGE_IS_5755_PLUS(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5906)
3643 		mode |= BGE_TXMODE_MBUF_LOCKUP_FIX;
3644 	/* Turn on transmitter */
3645 	CSR_WRITE_4(sc, BGE_TX_MODE, mode | BGE_TXMODE_ENABLE);
3646 
3647 	/* Turn on receiver */
3648 	BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
3649 
3650 	/*
3651 	 * Set the number of good frames to receive after RX MBUF
3652 	 * Low Watermark has been reached.  After the RX MAC receives
3653 	 * this number of frames, it will drop subsequent incoming
3654 	 * frames until the MBUF High Watermark is reached.
3655 	 */
3656 	CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 2);
3657 
3658 	if (sc->bge_irq_type == PCI_INTR_TYPE_MSI) {
3659 		if (bootverbose) {
3660 			if_printf(ifp, "MSI_MODE: %#x\n",
3661 			    CSR_READ_4(sc, BGE_MSI_MODE));
3662 		}
3663 
3664 		/*
3665 		 * XXX
3666 		 * Linux driver turns it on for all chips supporting MSI?!
3667 		 */
3668 		if (sc->bge_flags & BGE_FLAG_ONESHOT_MSI) {
3669 			/*
3670 			 * XXX
3671 			 * According to 5722-PG101-R,
3672 			 * BGE_PCIE_TRANSACT_ONESHOT_MSI applies only to
3673 			 * BCM5906.
3674 			 */
3675 			BGE_SETBIT(sc, BGE_PCIE_TRANSACT,
3676 			    BGE_PCIE_TRANSACT_ONESHOT_MSI);
3677 		}
3678 	}
3679 
3680 	/* Tell firmware we're alive. */
3681 	BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3682 
3683 	/* Enable host interrupts if polling(4) is not enabled. */
3684 	PCI_SETBIT(sc->bge_dev, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_CLEAR_INTA, 4);
3685 #ifdef IFPOLL_ENABLE
3686 	if (ifp->if_flags & IFF_NPOLLING)
3687 		bge_disable_intr(sc);
3688 	else
3689 #endif
3690 	bge_enable_intr(sc);
3691 
3692 	bge_ifmedia_upd(ifp);
3693 
3694 	ifp->if_flags |= IFF_RUNNING;
3695 	ifp->if_flags &= ~IFF_OACTIVE;
3696 
3697 	callout_reset(&sc->bge_stat_timer, hz, bge_tick, sc);
3698 }
3699 
3700 /*
3701  * Set media options.
3702  */
3703 static int
3704 bge_ifmedia_upd(struct ifnet *ifp)
3705 {
3706 	struct bge_softc *sc = ifp->if_softc;
3707 
3708 	/* If this is a 1000baseX NIC, enable the TBI port. */
3709 	if (sc->bge_flags & BGE_FLAG_TBI) {
3710 		struct ifmedia *ifm = &sc->bge_ifmedia;
3711 
3712 		if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
3713 			return(EINVAL);
3714 
3715 		switch(IFM_SUBTYPE(ifm->ifm_media)) {
3716 		case IFM_AUTO:
3717 			/*
3718 			 * The BCM5704 ASIC appears to have a special
3719 			 * mechanism for programming the autoneg
3720 			 * advertisement registers in TBI mode.
3721 			 */
3722 			if (!bge_fake_autoneg &&
3723 			    sc->bge_asicrev == BGE_ASICREV_BCM5704) {
3724 				uint32_t sgdig;
3725 
3726 				CSR_WRITE_4(sc, BGE_TX_TBI_AUTONEG, 0);
3727 				sgdig = CSR_READ_4(sc, BGE_SGDIG_CFG);
3728 				sgdig |= BGE_SGDIGCFG_AUTO |
3729 					 BGE_SGDIGCFG_PAUSE_CAP |
3730 					 BGE_SGDIGCFG_ASYM_PAUSE;
3731 				CSR_WRITE_4(sc, BGE_SGDIG_CFG,
3732 					    sgdig | BGE_SGDIGCFG_SEND);
3733 				DELAY(5);
3734 				CSR_WRITE_4(sc, BGE_SGDIG_CFG, sgdig);
3735 			}
3736 			break;
3737 		case IFM_1000_SX:
3738 			if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX) {
3739 				BGE_CLRBIT(sc, BGE_MAC_MODE,
3740 				    BGE_MACMODE_HALF_DUPLEX);
3741 			} else {
3742 				BGE_SETBIT(sc, BGE_MAC_MODE,
3743 				    BGE_MACMODE_HALF_DUPLEX);
3744 			}
3745 			break;
3746 		default:
3747 			return(EINVAL);
3748 		}
3749 	} else {
3750 		struct mii_data *mii = device_get_softc(sc->bge_miibus);
3751 
3752 		sc->bge_link_evt++;
3753 		sc->bge_link = 0;
3754 		if (mii->mii_instance) {
3755 			struct mii_softc *miisc;
3756 
3757 			LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
3758 				mii_phy_reset(miisc);
3759 		}
3760 		mii_mediachg(mii);
3761 
3762 		/*
3763 		 * Force an interrupt so that we will call bge_link_upd
3764 		 * if needed and clear any pending link state attention.
3765 		 * Without this we are not getting any further interrupts
3766 		 * for link state changes and thus will not UP the link and
3767 		 * not be able to send in bge_start.  The only way to get
3768 		 * things working was to receive a packet and get an RX
3769 		 * intr.
3770 		 *
3771 		 * bge_tick should help for fiber cards and we might not
3772 		 * need to do this here if BGE_FLAG_TBI is set but as
3773 		 * we poll for fiber anyway it should not harm.
3774 		 */
3775 		if (BGE_IS_CRIPPLED(sc))
3776 			BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
3777 		else
3778 			BGE_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
3779 	}
3780 	return(0);
3781 }
3782 
3783 /*
3784  * Report current media status.
3785  */
3786 static void
3787 bge_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3788 {
3789 	struct bge_softc *sc = ifp->if_softc;
3790 
3791 	if (sc->bge_flags & BGE_FLAG_TBI) {
3792 		ifmr->ifm_status = IFM_AVALID;
3793 		ifmr->ifm_active = IFM_ETHER;
3794 		if (CSR_READ_4(sc, BGE_MAC_STS) &
3795 		    BGE_MACSTAT_TBI_PCS_SYNCHED) {
3796 			ifmr->ifm_status |= IFM_ACTIVE;
3797 		} else {
3798 			ifmr->ifm_active |= IFM_NONE;
3799 			return;
3800 		}
3801 
3802 		ifmr->ifm_active |= IFM_1000_SX;
3803 		if (CSR_READ_4(sc, BGE_MAC_MODE) & BGE_MACMODE_HALF_DUPLEX)
3804 			ifmr->ifm_active |= IFM_HDX;
3805 		else
3806 			ifmr->ifm_active |= IFM_FDX;
3807 	} else {
3808 		struct mii_data *mii = device_get_softc(sc->bge_miibus);
3809 
3810 		mii_pollstat(mii);
3811 		ifmr->ifm_active = mii->mii_media_active;
3812 		ifmr->ifm_status = mii->mii_media_status;
3813 	}
3814 }
3815 
3816 static int
3817 bge_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
3818 {
3819 	struct bge_softc *sc = ifp->if_softc;
3820 	struct ifreq *ifr = (struct ifreq *)data;
3821 	int mask, error = 0;
3822 
3823 	ASSERT_SERIALIZED(ifp->if_serializer);
3824 
3825 	switch (command) {
3826 	case SIOCSIFMTU:
3827 		if ((!BGE_IS_JUMBO_CAPABLE(sc) && ifr->ifr_mtu > ETHERMTU) ||
3828 		    (BGE_IS_JUMBO_CAPABLE(sc) &&
3829 		     ifr->ifr_mtu > BGE_JUMBO_MTU)) {
3830 			error = EINVAL;
3831 		} else if (ifp->if_mtu != ifr->ifr_mtu) {
3832 			ifp->if_mtu = ifr->ifr_mtu;
3833 			if (ifp->if_flags & IFF_RUNNING)
3834 				bge_init(sc);
3835 		}
3836 		break;
3837 	case SIOCSIFFLAGS:
3838 		if (ifp->if_flags & IFF_UP) {
3839 			if (ifp->if_flags & IFF_RUNNING) {
3840 				mask = ifp->if_flags ^ sc->bge_if_flags;
3841 
3842 				/*
3843 				 * If only the state of the PROMISC flag
3844 				 * changed, then just use the 'set promisc
3845 				 * mode' command instead of reinitializing
3846 				 * the entire NIC. Doing a full re-init
3847 				 * means reloading the firmware and waiting
3848 				 * for it to start up, which may take a
3849 				 * second or two.  Similarly for ALLMULTI.
3850 				 */
3851 				if (mask & IFF_PROMISC)
3852 					bge_setpromisc(sc);
3853 				if (mask & IFF_ALLMULTI)
3854 					bge_setmulti(sc);
3855 			} else {
3856 				bge_init(sc);
3857 			}
3858 		} else if (ifp->if_flags & IFF_RUNNING) {
3859 			bge_stop(sc);
3860 		}
3861 		sc->bge_if_flags = ifp->if_flags;
3862 		break;
3863 	case SIOCADDMULTI:
3864 	case SIOCDELMULTI:
3865 		if (ifp->if_flags & IFF_RUNNING)
3866 			bge_setmulti(sc);
3867 		break;
3868 	case SIOCSIFMEDIA:
3869 	case SIOCGIFMEDIA:
3870 		if (sc->bge_flags & BGE_FLAG_TBI) {
3871 			error = ifmedia_ioctl(ifp, ifr,
3872 			    &sc->bge_ifmedia, command);
3873 		} else {
3874 			struct mii_data *mii;
3875 
3876 			mii = device_get_softc(sc->bge_miibus);
3877 			error = ifmedia_ioctl(ifp, ifr,
3878 					      &mii->mii_media, command);
3879 		}
3880 		break;
3881         case SIOCSIFCAP:
3882 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3883 		if (mask & IFCAP_HWCSUM) {
3884 			ifp->if_capenable ^= (mask & IFCAP_HWCSUM);
3885 			if (ifp->if_capenable & IFCAP_TXCSUM)
3886 				ifp->if_hwassist |= BGE_CSUM_FEATURES;
3887 			else
3888 				ifp->if_hwassist &= ~BGE_CSUM_FEATURES;
3889 		}
3890 		if (mask & IFCAP_TSO) {
3891 			ifp->if_capenable ^= IFCAP_TSO;
3892 			if (ifp->if_capenable & IFCAP_TSO)
3893 				ifp->if_hwassist |= CSUM_TSO;
3894 			else
3895 				ifp->if_hwassist &= ~CSUM_TSO;
3896 		}
3897 		break;
3898 	default:
3899 		error = ether_ioctl(ifp, command, data);
3900 		break;
3901 	}
3902 	return error;
3903 }
3904 
3905 static void
3906 bge_watchdog(struct ifnet *ifp)
3907 {
3908 	struct bge_softc *sc = ifp->if_softc;
3909 
3910 	if_printf(ifp, "watchdog timeout -- resetting\n");
3911 
3912 	bge_init(sc);
3913 
3914 	ifp->if_oerrors++;
3915 
3916 	if (!ifq_is_empty(&ifp->if_snd))
3917 		if_devstart(ifp);
3918 }
3919 
3920 /*
3921  * Stop the adapter and free any mbufs allocated to the
3922  * RX and TX lists.
3923  */
3924 static void
3925 bge_stop(struct bge_softc *sc)
3926 {
3927 	struct ifnet *ifp = &sc->arpcom.ac_if;
3928 
3929 	ASSERT_SERIALIZED(ifp->if_serializer);
3930 
3931 	callout_stop(&sc->bge_stat_timer);
3932 
3933 	/*
3934 	 * Disable all of the receiver blocks
3935 	 */
3936 	bge_stop_block(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
3937 	bge_stop_block(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
3938 	bge_stop_block(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
3939 	if (BGE_IS_5700_FAMILY(sc))
3940 		bge_stop_block(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
3941 	bge_stop_block(sc, BGE_RDBDI_MODE, BGE_RBDIMODE_ENABLE);
3942 	bge_stop_block(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
3943 	bge_stop_block(sc, BGE_RBDC_MODE, BGE_RBDCMODE_ENABLE);
3944 
3945 	/*
3946 	 * Disable all of the transmit blocks
3947 	 */
3948 	bge_stop_block(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
3949 	bge_stop_block(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
3950 	bge_stop_block(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
3951 	bge_stop_block(sc, BGE_RDMA_MODE, BGE_RDMAMODE_ENABLE);
3952 	bge_stop_block(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE);
3953 	if (BGE_IS_5700_FAMILY(sc))
3954 		bge_stop_block(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
3955 	bge_stop_block(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
3956 
3957 	/*
3958 	 * Shut down all of the memory managers and related
3959 	 * state machines.
3960 	 */
3961 	bge_stop_block(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE);
3962 	bge_stop_block(sc, BGE_WDMA_MODE, BGE_WDMAMODE_ENABLE);
3963 	if (BGE_IS_5700_FAMILY(sc))
3964 		bge_stop_block(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
3965 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
3966 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
3967 	if (!BGE_IS_5705_PLUS(sc)) {
3968 		BGE_CLRBIT(sc, BGE_BMAN_MODE, BGE_BMANMODE_ENABLE);
3969 		BGE_CLRBIT(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
3970 	}
3971 
3972 	/* Disable host interrupts. */
3973 	bge_disable_intr(sc);
3974 
3975 	/*
3976 	 * Tell firmware we're shutting down.
3977 	 */
3978 	BGE_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3979 
3980 	/* Free the RX lists. */
3981 	bge_free_rx_ring_std(sc);
3982 
3983 	/* Free jumbo RX list. */
3984 	if (BGE_IS_JUMBO_CAPABLE(sc))
3985 		bge_free_rx_ring_jumbo(sc);
3986 
3987 	/* Free TX buffers. */
3988 	bge_free_tx_ring(sc);
3989 
3990 	sc->bge_status_tag = 0;
3991 	sc->bge_link = 0;
3992 	sc->bge_coal_chg = 0;
3993 
3994 	sc->bge_tx_saved_considx = BGE_TXCONS_UNSET;
3995 
3996 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
3997 	ifp->if_timer = 0;
3998 }
3999 
4000 /*
4001  * Stop all chip I/O so that the kernel's probe routines don't
4002  * get confused by errant DMAs when rebooting.
4003  */
4004 static void
4005 bge_shutdown(device_t dev)
4006 {
4007 	struct bge_softc *sc = device_get_softc(dev);
4008 	struct ifnet *ifp = &sc->arpcom.ac_if;
4009 
4010 	lwkt_serialize_enter(ifp->if_serializer);
4011 	bge_stop(sc);
4012 	bge_reset(sc);
4013 	lwkt_serialize_exit(ifp->if_serializer);
4014 }
4015 
4016 static int
4017 bge_suspend(device_t dev)
4018 {
4019 	struct bge_softc *sc = device_get_softc(dev);
4020 	struct ifnet *ifp = &sc->arpcom.ac_if;
4021 
4022 	lwkt_serialize_enter(ifp->if_serializer);
4023 	bge_stop(sc);
4024 	lwkt_serialize_exit(ifp->if_serializer);
4025 
4026 	return 0;
4027 }
4028 
4029 static int
4030 bge_resume(device_t dev)
4031 {
4032 	struct bge_softc *sc = device_get_softc(dev);
4033 	struct ifnet *ifp = &sc->arpcom.ac_if;
4034 
4035 	lwkt_serialize_enter(ifp->if_serializer);
4036 
4037 	if (ifp->if_flags & IFF_UP) {
4038 		bge_init(sc);
4039 
4040 		if (!ifq_is_empty(&ifp->if_snd))
4041 			if_devstart(ifp);
4042 	}
4043 
4044 	lwkt_serialize_exit(ifp->if_serializer);
4045 
4046 	return 0;
4047 }
4048 
4049 static void
4050 bge_setpromisc(struct bge_softc *sc)
4051 {
4052 	struct ifnet *ifp = &sc->arpcom.ac_if;
4053 
4054 	if (ifp->if_flags & IFF_PROMISC)
4055 		BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
4056 	else
4057 		BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
4058 }
4059 
4060 static void
4061 bge_dma_free(struct bge_softc *sc)
4062 {
4063 	int i;
4064 
4065 	/* Destroy RX mbuf DMA stuffs. */
4066 	if (sc->bge_cdata.bge_rx_mtag != NULL) {
4067 		for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
4068 			bus_dmamap_destroy(sc->bge_cdata.bge_rx_mtag,
4069 			    sc->bge_cdata.bge_rx_std_dmamap[i]);
4070 		}
4071 		bus_dmamap_destroy(sc->bge_cdata.bge_rx_mtag,
4072 				   sc->bge_cdata.bge_rx_tmpmap);
4073 		bus_dma_tag_destroy(sc->bge_cdata.bge_rx_mtag);
4074 	}
4075 
4076 	/* Destroy TX mbuf DMA stuffs. */
4077 	if (sc->bge_cdata.bge_tx_mtag != NULL) {
4078 		for (i = 0; i < BGE_TX_RING_CNT; i++) {
4079 			bus_dmamap_destroy(sc->bge_cdata.bge_tx_mtag,
4080 			    sc->bge_cdata.bge_tx_dmamap[i]);
4081 		}
4082 		bus_dma_tag_destroy(sc->bge_cdata.bge_tx_mtag);
4083 	}
4084 
4085 	/* Destroy standard RX ring */
4086 	bge_dma_block_free(sc->bge_cdata.bge_rx_std_ring_tag,
4087 			   sc->bge_cdata.bge_rx_std_ring_map,
4088 			   sc->bge_ldata.bge_rx_std_ring);
4089 
4090 	if (BGE_IS_JUMBO_CAPABLE(sc))
4091 		bge_free_jumbo_mem(sc);
4092 
4093 	/* Destroy RX return ring */
4094 	bge_dma_block_free(sc->bge_cdata.bge_rx_return_ring_tag,
4095 			   sc->bge_cdata.bge_rx_return_ring_map,
4096 			   sc->bge_ldata.bge_rx_return_ring);
4097 
4098 	/* Destroy TX ring */
4099 	bge_dma_block_free(sc->bge_cdata.bge_tx_ring_tag,
4100 			   sc->bge_cdata.bge_tx_ring_map,
4101 			   sc->bge_ldata.bge_tx_ring);
4102 
4103 	/* Destroy status block */
4104 	bge_dma_block_free(sc->bge_cdata.bge_status_tag,
4105 			   sc->bge_cdata.bge_status_map,
4106 			   sc->bge_ldata.bge_status_block);
4107 
4108 	/* Destroy statistics block */
4109 	bge_dma_block_free(sc->bge_cdata.bge_stats_tag,
4110 			   sc->bge_cdata.bge_stats_map,
4111 			   sc->bge_ldata.bge_stats);
4112 
4113 	/* Destroy the parent tag */
4114 	if (sc->bge_cdata.bge_parent_tag != NULL)
4115 		bus_dma_tag_destroy(sc->bge_cdata.bge_parent_tag);
4116 }
4117 
4118 static int
4119 bge_dma_alloc(struct bge_softc *sc)
4120 {
4121 	struct ifnet *ifp = &sc->arpcom.ac_if;
4122 	int i, error;
4123 	bus_addr_t lowaddr;
4124 	bus_size_t txmaxsz;
4125 
4126 	lowaddr = BUS_SPACE_MAXADDR;
4127 	if (sc->bge_flags & BGE_FLAG_MAXADDR_40BIT)
4128 		lowaddr = BGE_DMA_MAXADDR_40BIT;
4129 
4130 	/*
4131 	 * Allocate the parent bus DMA tag appropriate for PCI.
4132 	 *
4133 	 * All of the NetExtreme/NetLink controllers have 4GB boundary
4134 	 * DMA bug.
4135 	 * Whenever an address crosses a multiple of the 4GB boundary
4136 	 * (including 4GB, 8Gb, 12Gb, etc.) and makes the transition
4137 	 * from 0xX_FFFF_FFFF to 0x(X+1)_0000_0000 an internal DMA
4138 	 * state machine will lockup and cause the device to hang.
4139 	 */
4140 	error = bus_dma_tag_create(NULL, 1, BGE_DMA_BOUNDARY_4G,
4141 				   lowaddr, BUS_SPACE_MAXADDR,
4142 				   NULL, NULL,
4143 				   BUS_SPACE_MAXSIZE_32BIT, 0,
4144 				   BUS_SPACE_MAXSIZE_32BIT,
4145 				   0, &sc->bge_cdata.bge_parent_tag);
4146 	if (error) {
4147 		if_printf(ifp, "could not allocate parent dma tag\n");
4148 		return error;
4149 	}
4150 
4151 	/*
4152 	 * Create DMA tag and maps for RX mbufs.
4153 	 */
4154 	error = bus_dma_tag_create(sc->bge_cdata.bge_parent_tag, 1, 0,
4155 				   BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
4156 				   NULL, NULL, MCLBYTES, 1, MCLBYTES,
4157 				   BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK,
4158 				   &sc->bge_cdata.bge_rx_mtag);
4159 	if (error) {
4160 		if_printf(ifp, "could not allocate RX mbuf dma tag\n");
4161 		return error;
4162 	}
4163 
4164 	error = bus_dmamap_create(sc->bge_cdata.bge_rx_mtag,
4165 				  BUS_DMA_WAITOK, &sc->bge_cdata.bge_rx_tmpmap);
4166 	if (error) {
4167 		bus_dma_tag_destroy(sc->bge_cdata.bge_rx_mtag);
4168 		sc->bge_cdata.bge_rx_mtag = NULL;
4169 		return error;
4170 	}
4171 
4172 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
4173 		error = bus_dmamap_create(sc->bge_cdata.bge_rx_mtag,
4174 					  BUS_DMA_WAITOK,
4175 					  &sc->bge_cdata.bge_rx_std_dmamap[i]);
4176 		if (error) {
4177 			int j;
4178 
4179 			for (j = 0; j < i; ++j) {
4180 				bus_dmamap_destroy(sc->bge_cdata.bge_rx_mtag,
4181 					sc->bge_cdata.bge_rx_std_dmamap[j]);
4182 			}
4183 			bus_dma_tag_destroy(sc->bge_cdata.bge_rx_mtag);
4184 			sc->bge_cdata.bge_rx_mtag = NULL;
4185 
4186 			if_printf(ifp, "could not create DMA map for RX\n");
4187 			return error;
4188 		}
4189 	}
4190 
4191 	/*
4192 	 * Create DMA tag and maps for TX mbufs.
4193 	 */
4194 	if (sc->bge_flags & BGE_FLAG_TSO)
4195 		txmaxsz = IP_MAXPACKET + sizeof(struct ether_vlan_header);
4196 	else
4197 		txmaxsz = BGE_JUMBO_FRAMELEN;
4198 	error = bus_dma_tag_create(sc->bge_cdata.bge_parent_tag, 1, 0,
4199 				   BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
4200 				   NULL, NULL,
4201 				   txmaxsz, BGE_NSEG_NEW, PAGE_SIZE,
4202 				   BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK |
4203 				   BUS_DMA_ONEBPAGE,
4204 				   &sc->bge_cdata.bge_tx_mtag);
4205 	if (error) {
4206 		if_printf(ifp, "could not allocate TX mbuf dma tag\n");
4207 		return error;
4208 	}
4209 
4210 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
4211 		error = bus_dmamap_create(sc->bge_cdata.bge_tx_mtag,
4212 					  BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
4213 					  &sc->bge_cdata.bge_tx_dmamap[i]);
4214 		if (error) {
4215 			int j;
4216 
4217 			for (j = 0; j < i; ++j) {
4218 				bus_dmamap_destroy(sc->bge_cdata.bge_tx_mtag,
4219 					sc->bge_cdata.bge_tx_dmamap[j]);
4220 			}
4221 			bus_dma_tag_destroy(sc->bge_cdata.bge_tx_mtag);
4222 			sc->bge_cdata.bge_tx_mtag = NULL;
4223 
4224 			if_printf(ifp, "could not create DMA map for TX\n");
4225 			return error;
4226 		}
4227 	}
4228 
4229 	/*
4230 	 * Create DMA stuffs for standard RX ring.
4231 	 */
4232 	error = bge_dma_block_alloc(sc, BGE_STD_RX_RING_SZ,
4233 				    &sc->bge_cdata.bge_rx_std_ring_tag,
4234 				    &sc->bge_cdata.bge_rx_std_ring_map,
4235 				    (void *)&sc->bge_ldata.bge_rx_std_ring,
4236 				    &sc->bge_ldata.bge_rx_std_ring_paddr);
4237 	if (error) {
4238 		if_printf(ifp, "could not create std RX ring\n");
4239 		return error;
4240 	}
4241 
4242 	/*
4243 	 * Create jumbo buffer pool.
4244 	 */
4245 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
4246 		error = bge_alloc_jumbo_mem(sc);
4247 		if (error) {
4248 			if_printf(ifp, "could not create jumbo buffer pool\n");
4249 			return error;
4250 		}
4251 	}
4252 
4253 	/*
4254 	 * Create DMA stuffs for RX return ring.
4255 	 */
4256 	error = bge_dma_block_alloc(sc,
4257 	    BGE_RX_RTN_RING_SZ(sc->bge_return_ring_cnt),
4258 	    &sc->bge_cdata.bge_rx_return_ring_tag,
4259 	    &sc->bge_cdata.bge_rx_return_ring_map,
4260 	    (void *)&sc->bge_ldata.bge_rx_return_ring,
4261 	    &sc->bge_ldata.bge_rx_return_ring_paddr);
4262 	if (error) {
4263 		if_printf(ifp, "could not create RX ret ring\n");
4264 		return error;
4265 	}
4266 
4267 	/*
4268 	 * Create DMA stuffs for TX ring.
4269 	 */
4270 	error = bge_dma_block_alloc(sc, BGE_TX_RING_SZ,
4271 				    &sc->bge_cdata.bge_tx_ring_tag,
4272 				    &sc->bge_cdata.bge_tx_ring_map,
4273 				    (void *)&sc->bge_ldata.bge_tx_ring,
4274 				    &sc->bge_ldata.bge_tx_ring_paddr);
4275 	if (error) {
4276 		if_printf(ifp, "could not create TX ring\n");
4277 		return error;
4278 	}
4279 
4280 	/*
4281 	 * Create DMA stuffs for status block.
4282 	 */
4283 	error = bge_dma_block_alloc(sc, BGE_STATUS_BLK_SZ,
4284 				    &sc->bge_cdata.bge_status_tag,
4285 				    &sc->bge_cdata.bge_status_map,
4286 				    (void *)&sc->bge_ldata.bge_status_block,
4287 				    &sc->bge_ldata.bge_status_block_paddr);
4288 	if (error) {
4289 		if_printf(ifp, "could not create status block\n");
4290 		return error;
4291 	}
4292 
4293 	/*
4294 	 * Create DMA stuffs for statistics block.
4295 	 */
4296 	error = bge_dma_block_alloc(sc, BGE_STATS_SZ,
4297 				    &sc->bge_cdata.bge_stats_tag,
4298 				    &sc->bge_cdata.bge_stats_map,
4299 				    (void *)&sc->bge_ldata.bge_stats,
4300 				    &sc->bge_ldata.bge_stats_paddr);
4301 	if (error) {
4302 		if_printf(ifp, "could not create stats block\n");
4303 		return error;
4304 	}
4305 	return 0;
4306 }
4307 
4308 static int
4309 bge_dma_block_alloc(struct bge_softc *sc, bus_size_t size, bus_dma_tag_t *tag,
4310 		    bus_dmamap_t *map, void **addr, bus_addr_t *paddr)
4311 {
4312 	bus_dmamem_t dmem;
4313 	int error;
4314 
4315 	error = bus_dmamem_coherent(sc->bge_cdata.bge_parent_tag, PAGE_SIZE, 0,
4316 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
4317 				    size, BUS_DMA_WAITOK | BUS_DMA_ZERO, &dmem);
4318 	if (error)
4319 		return error;
4320 
4321 	*tag = dmem.dmem_tag;
4322 	*map = dmem.dmem_map;
4323 	*addr = dmem.dmem_addr;
4324 	*paddr = dmem.dmem_busaddr;
4325 
4326 	return 0;
4327 }
4328 
4329 static void
4330 bge_dma_block_free(bus_dma_tag_t tag, bus_dmamap_t map, void *addr)
4331 {
4332 	if (tag != NULL) {
4333 		bus_dmamap_unload(tag, map);
4334 		bus_dmamem_free(tag, addr, map);
4335 		bus_dma_tag_destroy(tag);
4336 	}
4337 }
4338 
4339 /*
4340  * Grrr. The link status word in the status block does
4341  * not work correctly on the BCM5700 rev AX and BX chips,
4342  * according to all available information. Hence, we have
4343  * to enable MII interrupts in order to properly obtain
4344  * async link changes. Unfortunately, this also means that
4345  * we have to read the MAC status register to detect link
4346  * changes, thereby adding an additional register access to
4347  * the interrupt handler.
4348  *
4349  * XXX: perhaps link state detection procedure used for
4350  * BGE_CHIPID_BCM5700_B2 can be used for others BCM5700 revisions.
4351  */
4352 static void
4353 bge_bcm5700_link_upd(struct bge_softc *sc, uint32_t status __unused)
4354 {
4355 	struct ifnet *ifp = &sc->arpcom.ac_if;
4356 	struct mii_data *mii = device_get_softc(sc->bge_miibus);
4357 
4358 	mii_pollstat(mii);
4359 
4360 	if (!sc->bge_link &&
4361 	    (mii->mii_media_status & IFM_ACTIVE) &&
4362 	    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
4363 		sc->bge_link++;
4364 		if (bootverbose)
4365 			if_printf(ifp, "link UP\n");
4366 	} else if (sc->bge_link &&
4367 	    (!(mii->mii_media_status & IFM_ACTIVE) ||
4368 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE)) {
4369 		sc->bge_link = 0;
4370 		if (bootverbose)
4371 			if_printf(ifp, "link DOWN\n");
4372 	}
4373 
4374 	/* Clear the interrupt. */
4375 	CSR_WRITE_4(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_MI_INTERRUPT);
4376 	bge_miibus_readreg(sc->bge_dev, 1, BRGPHY_MII_ISR);
4377 	bge_miibus_writereg(sc->bge_dev, 1, BRGPHY_MII_IMR, BRGPHY_INTRS);
4378 }
4379 
4380 static void
4381 bge_tbi_link_upd(struct bge_softc *sc, uint32_t status)
4382 {
4383 	struct ifnet *ifp = &sc->arpcom.ac_if;
4384 
4385 #define PCS_ENCODE_ERR	(BGE_MACSTAT_PORT_DECODE_ERROR|BGE_MACSTAT_MI_COMPLETE)
4386 
4387 	/*
4388 	 * Sometimes PCS encoding errors are detected in
4389 	 * TBI mode (on fiber NICs), and for some reason
4390 	 * the chip will signal them as link changes.
4391 	 * If we get a link change event, but the 'PCS
4392 	 * encoding error' bit in the MAC status register
4393 	 * is set, don't bother doing a link check.
4394 	 * This avoids spurious "gigabit link up" messages
4395 	 * that sometimes appear on fiber NICs during
4396 	 * periods of heavy traffic.
4397 	 */
4398 	if (status & BGE_MACSTAT_TBI_PCS_SYNCHED) {
4399 		if (!sc->bge_link) {
4400 			sc->bge_link++;
4401 			if (sc->bge_asicrev == BGE_ASICREV_BCM5704) {
4402 				BGE_CLRBIT(sc, BGE_MAC_MODE,
4403 				    BGE_MACMODE_TBI_SEND_CFGS);
4404 			}
4405 			CSR_WRITE_4(sc, BGE_MAC_STS, 0xFFFFFFFF);
4406 
4407 			if (bootverbose)
4408 				if_printf(ifp, "link UP\n");
4409 
4410 			ifp->if_link_state = LINK_STATE_UP;
4411 			if_link_state_change(ifp);
4412 		}
4413 	} else if ((status & PCS_ENCODE_ERR) != PCS_ENCODE_ERR) {
4414 		if (sc->bge_link) {
4415 			sc->bge_link = 0;
4416 
4417 			if (bootverbose)
4418 				if_printf(ifp, "link DOWN\n");
4419 
4420 			ifp->if_link_state = LINK_STATE_DOWN;
4421 			if_link_state_change(ifp);
4422 		}
4423 	}
4424 
4425 #undef PCS_ENCODE_ERR
4426 
4427 	/* Clear the attention. */
4428 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
4429 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
4430 	    BGE_MACSTAT_LINK_CHANGED);
4431 }
4432 
4433 static void
4434 bge_copper_link_upd(struct bge_softc *sc, uint32_t status __unused)
4435 {
4436 	struct ifnet *ifp = &sc->arpcom.ac_if;
4437 	struct mii_data *mii = device_get_softc(sc->bge_miibus);
4438 
4439 	mii_pollstat(mii);
4440 	bge_miibus_statchg(sc->bge_dev);
4441 
4442 	if (bootverbose) {
4443 		if (sc->bge_link)
4444 			if_printf(ifp, "link UP\n");
4445 		else
4446 			if_printf(ifp, "link DOWN\n");
4447 	}
4448 
4449 	/* Clear the attention. */
4450 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
4451 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
4452 	    BGE_MACSTAT_LINK_CHANGED);
4453 }
4454 
4455 static void
4456 bge_autopoll_link_upd(struct bge_softc *sc, uint32_t status __unused)
4457 {
4458 	struct ifnet *ifp = &sc->arpcom.ac_if;
4459 	struct mii_data *mii = device_get_softc(sc->bge_miibus);
4460 
4461 	mii_pollstat(mii);
4462 
4463 	if (!sc->bge_link &&
4464 	    (mii->mii_media_status & IFM_ACTIVE) &&
4465 	    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
4466 		sc->bge_link++;
4467 		if (bootverbose)
4468 			if_printf(ifp, "link UP\n");
4469 	} else if (sc->bge_link &&
4470 	    (!(mii->mii_media_status & IFM_ACTIVE) ||
4471 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE)) {
4472 		sc->bge_link = 0;
4473 		if (bootverbose)
4474 			if_printf(ifp, "link DOWN\n");
4475 	}
4476 
4477 	/* Clear the attention. */
4478 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
4479 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
4480 	    BGE_MACSTAT_LINK_CHANGED);
4481 }
4482 
4483 static int
4484 bge_sysctl_rx_coal_ticks(SYSCTL_HANDLER_ARGS)
4485 {
4486 	struct bge_softc *sc = arg1;
4487 
4488 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4489 	    &sc->bge_rx_coal_ticks,
4490 	    BGE_RX_COAL_TICKS_MIN, BGE_RX_COAL_TICKS_MAX,
4491 	    BGE_RX_COAL_TICKS_CHG);
4492 }
4493 
4494 static int
4495 bge_sysctl_tx_coal_ticks(SYSCTL_HANDLER_ARGS)
4496 {
4497 	struct bge_softc *sc = arg1;
4498 
4499 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4500 	    &sc->bge_tx_coal_ticks,
4501 	    BGE_TX_COAL_TICKS_MIN, BGE_TX_COAL_TICKS_MAX,
4502 	    BGE_TX_COAL_TICKS_CHG);
4503 }
4504 
4505 static int
4506 bge_sysctl_rx_coal_bds(SYSCTL_HANDLER_ARGS)
4507 {
4508 	struct bge_softc *sc = arg1;
4509 
4510 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4511 	    &sc->bge_rx_coal_bds,
4512 	    BGE_RX_COAL_BDS_MIN, BGE_RX_COAL_BDS_MAX,
4513 	    BGE_RX_COAL_BDS_CHG);
4514 }
4515 
4516 static int
4517 bge_sysctl_tx_coal_bds(SYSCTL_HANDLER_ARGS)
4518 {
4519 	struct bge_softc *sc = arg1;
4520 
4521 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4522 	    &sc->bge_tx_coal_bds,
4523 	    BGE_TX_COAL_BDS_MIN, BGE_TX_COAL_BDS_MAX,
4524 	    BGE_TX_COAL_BDS_CHG);
4525 }
4526 
4527 static int
4528 bge_sysctl_rx_coal_ticks_int(SYSCTL_HANDLER_ARGS)
4529 {
4530 	struct bge_softc *sc = arg1;
4531 
4532 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4533 	    &sc->bge_rx_coal_ticks_int,
4534 	    BGE_RX_COAL_TICKS_MIN, BGE_RX_COAL_TICKS_MAX,
4535 	    BGE_RX_COAL_TICKS_INT_CHG);
4536 }
4537 
4538 static int
4539 bge_sysctl_tx_coal_ticks_int(SYSCTL_HANDLER_ARGS)
4540 {
4541 	struct bge_softc *sc = arg1;
4542 
4543 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4544 	    &sc->bge_tx_coal_ticks_int,
4545 	    BGE_TX_COAL_TICKS_MIN, BGE_TX_COAL_TICKS_MAX,
4546 	    BGE_TX_COAL_TICKS_INT_CHG);
4547 }
4548 
4549 static int
4550 bge_sysctl_rx_coal_bds_int(SYSCTL_HANDLER_ARGS)
4551 {
4552 	struct bge_softc *sc = arg1;
4553 
4554 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4555 	    &sc->bge_rx_coal_bds_int,
4556 	    BGE_RX_COAL_BDS_MIN, BGE_RX_COAL_BDS_MAX,
4557 	    BGE_RX_COAL_BDS_INT_CHG);
4558 }
4559 
4560 static int
4561 bge_sysctl_tx_coal_bds_int(SYSCTL_HANDLER_ARGS)
4562 {
4563 	struct bge_softc *sc = arg1;
4564 
4565 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4566 	    &sc->bge_tx_coal_bds_int,
4567 	    BGE_TX_COAL_BDS_MIN, BGE_TX_COAL_BDS_MAX,
4568 	    BGE_TX_COAL_BDS_INT_CHG);
4569 }
4570 
4571 static int
4572 bge_sysctl_coal_chg(SYSCTL_HANDLER_ARGS, uint32_t *coal,
4573     int coal_min, int coal_max, uint32_t coal_chg_mask)
4574 {
4575 	struct bge_softc *sc = arg1;
4576 	struct ifnet *ifp = &sc->arpcom.ac_if;
4577 	int error = 0, v;
4578 
4579 	lwkt_serialize_enter(ifp->if_serializer);
4580 
4581 	v = *coal;
4582 	error = sysctl_handle_int(oidp, &v, 0, req);
4583 	if (!error && req->newptr != NULL) {
4584 		if (v < coal_min || v > coal_max) {
4585 			error = EINVAL;
4586 		} else {
4587 			*coal = v;
4588 			sc->bge_coal_chg |= coal_chg_mask;
4589 		}
4590 	}
4591 
4592 	lwkt_serialize_exit(ifp->if_serializer);
4593 	return error;
4594 }
4595 
4596 static void
4597 bge_coal_change(struct bge_softc *sc)
4598 {
4599 	struct ifnet *ifp = &sc->arpcom.ac_if;
4600 	uint32_t val;
4601 
4602 	ASSERT_SERIALIZED(ifp->if_serializer);
4603 
4604 	if (sc->bge_coal_chg & BGE_RX_COAL_TICKS_CHG) {
4605 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS,
4606 			    sc->bge_rx_coal_ticks);
4607 		DELAY(10);
4608 		val = CSR_READ_4(sc, BGE_HCC_RX_COAL_TICKS);
4609 
4610 		if (bootverbose) {
4611 			if_printf(ifp, "rx_coal_ticks -> %u\n",
4612 				  sc->bge_rx_coal_ticks);
4613 		}
4614 	}
4615 
4616 	if (sc->bge_coal_chg & BGE_TX_COAL_TICKS_CHG) {
4617 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS,
4618 			    sc->bge_tx_coal_ticks);
4619 		DELAY(10);
4620 		val = CSR_READ_4(sc, BGE_HCC_TX_COAL_TICKS);
4621 
4622 		if (bootverbose) {
4623 			if_printf(ifp, "tx_coal_ticks -> %u\n",
4624 				  sc->bge_tx_coal_ticks);
4625 		}
4626 	}
4627 
4628 	if (sc->bge_coal_chg & BGE_RX_COAL_BDS_CHG) {
4629 		CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS,
4630 			    sc->bge_rx_coal_bds);
4631 		DELAY(10);
4632 		val = CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS);
4633 
4634 		if (bootverbose) {
4635 			if_printf(ifp, "rx_coal_bds -> %u\n",
4636 				  sc->bge_rx_coal_bds);
4637 		}
4638 	}
4639 
4640 	if (sc->bge_coal_chg & BGE_TX_COAL_BDS_CHG) {
4641 		CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS,
4642 			    sc->bge_tx_coal_bds);
4643 		DELAY(10);
4644 		val = CSR_READ_4(sc, BGE_HCC_TX_MAX_COAL_BDS);
4645 
4646 		if (bootverbose) {
4647 			if_printf(ifp, "tx_max_coal_bds -> %u\n",
4648 				  sc->bge_tx_coal_bds);
4649 		}
4650 	}
4651 
4652 	if (sc->bge_coal_chg & BGE_RX_COAL_TICKS_INT_CHG) {
4653 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS_INT,
4654 		    sc->bge_rx_coal_ticks_int);
4655 		DELAY(10);
4656 		val = CSR_READ_4(sc, BGE_HCC_RX_COAL_TICKS_INT);
4657 
4658 		if (bootverbose) {
4659 			if_printf(ifp, "rx_coal_ticks_int -> %u\n",
4660 			    sc->bge_rx_coal_ticks_int);
4661 		}
4662 	}
4663 
4664 	if (sc->bge_coal_chg & BGE_TX_COAL_TICKS_INT_CHG) {
4665 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS_INT,
4666 		    sc->bge_tx_coal_ticks_int);
4667 		DELAY(10);
4668 		val = CSR_READ_4(sc, BGE_HCC_TX_COAL_TICKS_INT);
4669 
4670 		if (bootverbose) {
4671 			if_printf(ifp, "tx_coal_ticks_int -> %u\n",
4672 			    sc->bge_tx_coal_ticks_int);
4673 		}
4674 	}
4675 
4676 	if (sc->bge_coal_chg & BGE_RX_COAL_BDS_INT_CHG) {
4677 		CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT,
4678 		    sc->bge_rx_coal_bds_int);
4679 		DELAY(10);
4680 		val = CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT);
4681 
4682 		if (bootverbose) {
4683 			if_printf(ifp, "rx_coal_bds_int -> %u\n",
4684 			    sc->bge_rx_coal_bds_int);
4685 		}
4686 	}
4687 
4688 	if (sc->bge_coal_chg & BGE_TX_COAL_BDS_INT_CHG) {
4689 		CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT,
4690 		    sc->bge_tx_coal_bds_int);
4691 		DELAY(10);
4692 		val = CSR_READ_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT);
4693 
4694 		if (bootverbose) {
4695 			if_printf(ifp, "tx_coal_bds_int -> %u\n",
4696 			    sc->bge_tx_coal_bds_int);
4697 		}
4698 	}
4699 
4700 	sc->bge_coal_chg = 0;
4701 }
4702 
4703 static void
4704 bge_enable_intr(struct bge_softc *sc)
4705 {
4706 	struct ifnet *ifp = &sc->arpcom.ac_if;
4707 
4708 	lwkt_serialize_handler_enable(ifp->if_serializer);
4709 
4710 	/*
4711 	 * Enable interrupt.
4712 	 */
4713 	bge_writembx(sc, BGE_MBX_IRQ0_LO, sc->bge_status_tag << 24);
4714 	if (sc->bge_flags & BGE_FLAG_ONESHOT_MSI) {
4715 		/* XXX Linux driver */
4716 		bge_writembx(sc, BGE_MBX_IRQ0_LO, sc->bge_status_tag << 24);
4717 	}
4718 
4719 	/*
4720 	 * Unmask the interrupt when we stop polling.
4721 	 */
4722 	PCI_CLRBIT(sc->bge_dev, BGE_PCI_MISC_CTL,
4723 	    BGE_PCIMISCCTL_MASK_PCI_INTR, 4);
4724 
4725 	/*
4726 	 * Trigger another interrupt, since above writing
4727 	 * to interrupt mailbox0 may acknowledge pending
4728 	 * interrupt.
4729 	 */
4730 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
4731 }
4732 
4733 static void
4734 bge_disable_intr(struct bge_softc *sc)
4735 {
4736 	struct ifnet *ifp = &sc->arpcom.ac_if;
4737 
4738 	/*
4739 	 * Mask the interrupt when we start polling.
4740 	 */
4741 	PCI_SETBIT(sc->bge_dev, BGE_PCI_MISC_CTL,
4742 	    BGE_PCIMISCCTL_MASK_PCI_INTR, 4);
4743 
4744 	/*
4745 	 * Acknowledge possible asserted interrupt.
4746 	 */
4747 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
4748 
4749 	sc->bge_npoll.ifpc_stcount = 0;
4750 
4751 	lwkt_serialize_handler_disable(ifp->if_serializer);
4752 }
4753 
4754 static int
4755 bge_get_eaddr_mem(struct bge_softc *sc, uint8_t ether_addr[])
4756 {
4757 	uint32_t mac_addr;
4758 	int ret = 1;
4759 
4760 	mac_addr = bge_readmem_ind(sc, 0x0c14);
4761 	if ((mac_addr >> 16) == 0x484b) {
4762 		ether_addr[0] = (uint8_t)(mac_addr >> 8);
4763 		ether_addr[1] = (uint8_t)mac_addr;
4764 		mac_addr = bge_readmem_ind(sc, 0x0c18);
4765 		ether_addr[2] = (uint8_t)(mac_addr >> 24);
4766 		ether_addr[3] = (uint8_t)(mac_addr >> 16);
4767 		ether_addr[4] = (uint8_t)(mac_addr >> 8);
4768 		ether_addr[5] = (uint8_t)mac_addr;
4769 		ret = 0;
4770 	}
4771 	return ret;
4772 }
4773 
4774 static int
4775 bge_get_eaddr_nvram(struct bge_softc *sc, uint8_t ether_addr[])
4776 {
4777 	int mac_offset = BGE_EE_MAC_OFFSET;
4778 
4779 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
4780 		mac_offset = BGE_EE_MAC_OFFSET_5906;
4781 
4782 	return bge_read_nvram(sc, ether_addr, mac_offset + 2, ETHER_ADDR_LEN);
4783 }
4784 
4785 static int
4786 bge_get_eaddr_eeprom(struct bge_softc *sc, uint8_t ether_addr[])
4787 {
4788 	if (sc->bge_flags & BGE_FLAG_NO_EEPROM)
4789 		return 1;
4790 
4791 	return bge_read_eeprom(sc, ether_addr, BGE_EE_MAC_OFFSET + 2,
4792 			       ETHER_ADDR_LEN);
4793 }
4794 
4795 static int
4796 bge_get_eaddr(struct bge_softc *sc, uint8_t eaddr[])
4797 {
4798 	static const bge_eaddr_fcn_t bge_eaddr_funcs[] = {
4799 		/* NOTE: Order is critical */
4800 		bge_get_eaddr_mem,
4801 		bge_get_eaddr_nvram,
4802 		bge_get_eaddr_eeprom,
4803 		NULL
4804 	};
4805 	const bge_eaddr_fcn_t *func;
4806 
4807 	for (func = bge_eaddr_funcs; *func != NULL; ++func) {
4808 		if ((*func)(sc, eaddr) == 0)
4809 			break;
4810 	}
4811 	return (*func == NULL ? ENXIO : 0);
4812 }
4813 
4814 /*
4815  * NOTE: 'm' is not freed upon failure
4816  */
4817 struct mbuf *
4818 bge_defrag_shortdma(struct mbuf *m)
4819 {
4820 	struct mbuf *n;
4821 	int found;
4822 
4823 	/*
4824 	 * If device receive two back-to-back send BDs with less than
4825 	 * or equal to 8 total bytes then the device may hang.  The two
4826 	 * back-to-back send BDs must in the same frame for this failure
4827 	 * to occur.  Scan mbuf chains and see whether two back-to-back
4828 	 * send BDs are there.  If this is the case, allocate new mbuf
4829 	 * and copy the frame to workaround the silicon bug.
4830 	 */
4831 	for (n = m, found = 0; n != NULL; n = n->m_next) {
4832 		if (n->m_len < 8) {
4833 			found++;
4834 			if (found > 1)
4835 				break;
4836 			continue;
4837 		}
4838 		found = 0;
4839 	}
4840 
4841 	if (found > 1)
4842 		n = m_defrag(m, MB_DONTWAIT);
4843 	else
4844 		n = m;
4845 	return n;
4846 }
4847 
4848 static void
4849 bge_stop_block(struct bge_softc *sc, bus_size_t reg, uint32_t bit)
4850 {
4851 	int i;
4852 
4853 	BGE_CLRBIT(sc, reg, bit);
4854 	for (i = 0; i < BGE_TIMEOUT; i++) {
4855 		if ((CSR_READ_4(sc, reg) & bit) == 0)
4856 			return;
4857 		DELAY(100);
4858 	}
4859 }
4860 
4861 static void
4862 bge_link_poll(struct bge_softc *sc)
4863 {
4864 	uint32_t status;
4865 
4866 	status = CSR_READ_4(sc, BGE_MAC_STS);
4867 	if ((status & sc->bge_link_chg) || sc->bge_link_evt) {
4868 		sc->bge_link_evt = 0;
4869 		sc->bge_link_upd(sc, status);
4870 	}
4871 }
4872 
4873 static void
4874 bge_enable_msi(struct bge_softc *sc)
4875 {
4876 	uint32_t msi_mode;
4877 
4878 	msi_mode = CSR_READ_4(sc, BGE_MSI_MODE);
4879 	msi_mode |= BGE_MSIMODE_ENABLE;
4880 	if (sc->bge_flags & BGE_FLAG_ONESHOT_MSI) {
4881 		/*
4882 		 * According to all of the datasheets that are publicly
4883 		 * available, bit 5 of the MSI_MODE is defined to be
4884 		 * "MSI FIFO Underrun Attn" for BCM5755+ and BCM5906, on
4885 		 * which "oneshot MSI" is enabled.  However, it is always
4886 		 * safe to clear it here.
4887 		 */
4888 		msi_mode &= ~BGE_MSIMODE_ONESHOT_DISABLE;
4889 	}
4890 	CSR_WRITE_4(sc, BGE_MSI_MODE, msi_mode);
4891 }
4892 
4893 static int
4894 bge_setup_tso(struct bge_softc *sc, struct mbuf **mp,
4895     uint16_t *mss0, uint16_t *flags0)
4896 {
4897 	struct mbuf *m;
4898 	struct ip *ip;
4899 	struct tcphdr *th;
4900 	int thoff, iphlen, hoff, hlen;
4901 	uint16_t flags, mss;
4902 
4903 	m = *mp;
4904 	KASSERT(M_WRITABLE(m), ("TSO mbuf not writable"));
4905 
4906 	hoff = m->m_pkthdr.csum_lhlen;
4907 	iphlen = m->m_pkthdr.csum_iphlen;
4908 	thoff = m->m_pkthdr.csum_thlen;
4909 
4910 	KASSERT(hoff > 0, ("invalid ether header len"));
4911 	KASSERT(iphlen > 0, ("invalid ip header len"));
4912 	KASSERT(thoff > 0, ("invalid tcp header len"));
4913 
4914 	if (__predict_false(m->m_len < hoff + iphlen + thoff)) {
4915 		m = m_pullup(m, hoff + iphlen + thoff);
4916 		if (m == NULL) {
4917 			*mp = NULL;
4918 			return ENOBUFS;
4919 		}
4920 		*mp = m;
4921 	}
4922 	ip = mtodoff(m, struct ip *, hoff);
4923 	th = mtodoff(m, struct tcphdr *, hoff + iphlen);
4924 
4925 	mss = m->m_pkthdr.tso_segsz;
4926 	flags = BGE_TXBDFLAG_CPU_PRE_DMA | BGE_TXBDFLAG_CPU_POST_DMA;
4927 
4928 	ip->ip_len = htons(mss + iphlen + thoff);
4929 	th->th_sum = 0;
4930 
4931 	hlen = (iphlen + thoff) >> 2;
4932 	mss |= (hlen << 11);
4933 
4934 	*mss0 = mss;
4935 	*flags0 = flags;
4936 
4937 	return 0;
4938 }
4939