xref: /dflybsd-src/sys/dev/netif/bge/if_bge.c (revision 8bb2400d5e4f21e6ed6fe870515c2d9a1cdfb5c7)
1 /*
2  * Copyright (c) 2001 Wind River Systems
3  * Copyright (c) 1997, 1998, 1999, 2001
4  *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. All advertising materials mentioning features or use of this software
15  *    must display the following acknowledgement:
16  *	This product includes software developed by Bill Paul.
17  * 4. Neither the name of the author nor the names of any co-contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
25  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
28  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
29  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
30  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
31  * THE POSSIBILITY OF SUCH DAMAGE.
32  *
33  * $FreeBSD: src/sys/dev/bge/if_bge.c,v 1.3.2.39 2005/07/03 03:41:18 silby Exp $
34  */
35 
36 /*
37  * Broadcom BCM570x family gigabit ethernet driver for FreeBSD.
38  *
39  * Written by Bill Paul <wpaul@windriver.com>
40  * Senior Engineer, Wind River Systems
41  */
42 
43 /*
44  * The Broadcom BCM5700 is based on technology originally developed by
45  * Alteon Networks as part of the Tigon I and Tigon II gigabit ethernet
46  * MAC chips. The BCM5700, sometimes refered to as the Tigon III, has
47  * two on-board MIPS R4000 CPUs and can have as much as 16MB of external
48  * SSRAM. The BCM5700 supports TCP, UDP and IP checksum offload, jumbo
49  * frames, highly configurable RX filtering, and 16 RX and TX queues
50  * (which, along with RX filter rules, can be used for QOS applications).
51  * Other features, such as TCP segmentation, may be available as part
52  * of value-added firmware updates. Unlike the Tigon I and Tigon II,
53  * firmware images can be stored in hardware and need not be compiled
54  * into the driver.
55  *
56  * The BCM5700 supports the PCI v2.2 and PCI-X v1.0 standards, and will
57  * function in a 32-bit/64-bit 33/66Mhz bus, or a 64-bit/133Mhz bus.
58  *
59  * The BCM5701 is a single-chip solution incorporating both the BCM5700
60  * MAC and a BCM5401 10/100/1000 PHY. Unlike the BCM5700, the BCM5701
61  * does not support external SSRAM.
62  *
63  * Broadcom also produces a variation of the BCM5700 under the "Altima"
64  * brand name, which is functionally similar but lacks PCI-X support.
65  *
66  * Without external SSRAM, you can only have at most 4 TX rings,
67  * and the use of the mini RX ring is disabled. This seems to imply
68  * that these features are simply not available on the BCM5701. As a
69  * result, this driver does not implement any support for the mini RX
70  * ring.
71  */
72 
73 #include "opt_polling.h"
74 
75 #include <sys/param.h>
76 #include <sys/bus.h>
77 #include <sys/endian.h>
78 #include <sys/kernel.h>
79 #include <sys/ktr.h>
80 #include <sys/interrupt.h>
81 #include <sys/mbuf.h>
82 #include <sys/malloc.h>
83 #include <sys/queue.h>
84 #include <sys/rman.h>
85 #include <sys/serialize.h>
86 #include <sys/socket.h>
87 #include <sys/sockio.h>
88 #include <sys/sysctl.h>
89 
90 #include <net/bpf.h>
91 #include <net/ethernet.h>
92 #include <net/if.h>
93 #include <net/if_arp.h>
94 #include <net/if_dl.h>
95 #include <net/if_media.h>
96 #include <net/if_types.h>
97 #include <net/ifq_var.h>
98 #include <net/vlan/if_vlan_var.h>
99 #include <net/vlan/if_vlan_ether.h>
100 
101 #include <dev/netif/mii_layer/mii.h>
102 #include <dev/netif/mii_layer/miivar.h>
103 #include <dev/netif/mii_layer/brgphyreg.h>
104 
105 #include <bus/pci/pcidevs.h>
106 #include <bus/pci/pcireg.h>
107 #include <bus/pci/pcivar.h>
108 
109 #include <dev/netif/bge/if_bgereg.h>
110 #include <dev/netif/bge/if_bgevar.h>
111 
112 /* "device miibus" required.  See GENERIC if you get errors here. */
113 #include "miibus_if.h"
114 
115 #define BGE_CSUM_FEATURES	(CSUM_IP | CSUM_TCP)
116 
117 static const struct bge_type {
118 	uint16_t		bge_vid;
119 	uint16_t		bge_did;
120 	char			*bge_name;
121 } bge_devs[] = {
122 	{ PCI_VENDOR_3COM, PCI_PRODUCT_3COM_3C996,
123 		"3COM 3C996 Gigabit Ethernet" },
124 
125 	{ PCI_VENDOR_ALTEON, PCI_PRODUCT_ALTEON_BCM5700,
126 		"Alteon BCM5700 Gigabit Ethernet" },
127 	{ PCI_VENDOR_ALTEON, PCI_PRODUCT_ALTEON_BCM5701,
128 		"Alteon BCM5701 Gigabit Ethernet" },
129 
130 	{ PCI_VENDOR_ALTIMA, PCI_PRODUCT_ALTIMA_AC1000,
131 		"Altima AC1000 Gigabit Ethernet" },
132 	{ PCI_VENDOR_ALTIMA, PCI_PRODUCT_ALTIMA_AC1001,
133 		"Altima AC1002 Gigabit Ethernet" },
134 	{ PCI_VENDOR_ALTIMA, PCI_PRODUCT_ALTIMA_AC9100,
135 		"Altima AC9100 Gigabit Ethernet" },
136 
137 	{ PCI_VENDOR_APPLE, PCI_PRODUCT_APPLE_BCM5701,
138 		"Apple BCM5701 Gigabit Ethernet" },
139 
140 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5700,
141 		"Broadcom BCM5700 Gigabit Ethernet" },
142 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5701,
143 		"Broadcom BCM5701 Gigabit Ethernet" },
144 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5702,
145 		"Broadcom BCM5702 Gigabit Ethernet" },
146 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5702X,
147 		"Broadcom BCM5702X Gigabit Ethernet" },
148 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5702_ALT,
149 		"Broadcom BCM5702 Gigabit Ethernet" },
150 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5703,
151 		"Broadcom BCM5703 Gigabit Ethernet" },
152 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5703X,
153 		"Broadcom BCM5703X Gigabit Ethernet" },
154 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5703A3,
155 		"Broadcom BCM5703 Gigabit Ethernet" },
156 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5704C,
157 		"Broadcom BCM5704C Dual Gigabit Ethernet" },
158 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5704S,
159 		"Broadcom BCM5704S Dual Gigabit Ethernet" },
160 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5704S_ALT,
161 		"Broadcom BCM5704S Dual Gigabit Ethernet" },
162 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5705,
163 		"Broadcom BCM5705 Gigabit Ethernet" },
164 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5705F,
165 		"Broadcom BCM5705F Gigabit Ethernet" },
166 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5705K,
167 		"Broadcom BCM5705K Gigabit Ethernet" },
168 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5705M,
169 		"Broadcom BCM5705M Gigabit Ethernet" },
170 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5705M_ALT,
171 		"Broadcom BCM5705M Gigabit Ethernet" },
172 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5714,
173 		"Broadcom BCM5714C Gigabit Ethernet" },
174 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5714S,
175 		"Broadcom BCM5714S Gigabit Ethernet" },
176 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5715,
177 		"Broadcom BCM5715 Gigabit Ethernet" },
178 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5715S,
179 		"Broadcom BCM5715S Gigabit Ethernet" },
180 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5720,
181 		"Broadcom BCM5720 Gigabit Ethernet" },
182 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5721,
183 		"Broadcom BCM5721 Gigabit Ethernet" },
184 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5722,
185 		"Broadcom BCM5722 Gigabit Ethernet" },
186 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5723,
187 		"Broadcom BCM5723 Gigabit Ethernet" },
188 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5750,
189 		"Broadcom BCM5750 Gigabit Ethernet" },
190 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5750M,
191 		"Broadcom BCM5750M Gigabit Ethernet" },
192 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5751,
193 		"Broadcom BCM5751 Gigabit Ethernet" },
194 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5751F,
195 		"Broadcom BCM5751F Gigabit Ethernet" },
196 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5751M,
197 		"Broadcom BCM5751M Gigabit Ethernet" },
198 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5752,
199 		"Broadcom BCM5752 Gigabit Ethernet" },
200 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5752M,
201 		"Broadcom BCM5752M Gigabit Ethernet" },
202 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5753,
203 		"Broadcom BCM5753 Gigabit Ethernet" },
204 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5753F,
205 		"Broadcom BCM5753F Gigabit Ethernet" },
206 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5753M,
207 		"Broadcom BCM5753M Gigabit Ethernet" },
208 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5754,
209 		"Broadcom BCM5754 Gigabit Ethernet" },
210 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5754M,
211 		"Broadcom BCM5754M Gigabit Ethernet" },
212 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5755,
213 		"Broadcom BCM5755 Gigabit Ethernet" },
214 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5755M,
215 		"Broadcom BCM5755M Gigabit Ethernet" },
216 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5756,
217 		"Broadcom BCM5756 Gigabit Ethernet" },
218 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5761,
219 		"Broadcom BCM5761 Gigabit Ethernet" },
220 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5761E,
221 		"Broadcom BCM5761E Gigabit Ethernet" },
222 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5761S,
223 		"Broadcom BCM5761S Gigabit Ethernet" },
224 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5761SE,
225 		"Broadcom BCM5761SE Gigabit Ethernet" },
226 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5764,
227 		"Broadcom BCM5764 Gigabit Ethernet" },
228 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5780,
229 		"Broadcom BCM5780 Gigabit Ethernet" },
230 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5780S,
231 		"Broadcom BCM5780S Gigabit Ethernet" },
232 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5781,
233 		"Broadcom BCM5781 Gigabit Ethernet" },
234 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5782,
235 		"Broadcom BCM5782 Gigabit Ethernet" },
236 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5784,
237 		"Broadcom BCM5784 Gigabit Ethernet" },
238 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5785F,
239 		"Broadcom BCM5785F Gigabit Ethernet" },
240 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5785G,
241 		"Broadcom BCM5785G Gigabit Ethernet" },
242 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5786,
243 		"Broadcom BCM5786 Gigabit Ethernet" },
244 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5787,
245 		"Broadcom BCM5787 Gigabit Ethernet" },
246 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5787F,
247 		"Broadcom BCM5787F Gigabit Ethernet" },
248 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5787M,
249 		"Broadcom BCM5787M Gigabit Ethernet" },
250 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5788,
251 		"Broadcom BCM5788 Gigabit Ethernet" },
252 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5789,
253 		"Broadcom BCM5789 Gigabit Ethernet" },
254 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5901,
255 		"Broadcom BCM5901 Fast Ethernet" },
256 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5901A2,
257 		"Broadcom BCM5901A2 Fast Ethernet" },
258 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5903M,
259 		"Broadcom BCM5903M Fast Ethernet" },
260 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5906,
261 		"Broadcom BCM5906 Fast Ethernet"},
262 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5906M,
263 		"Broadcom BCM5906M Fast Ethernet"},
264 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57760,
265 		"Broadcom BCM57760 Gigabit Ethernet"},
266 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57780,
267 		"Broadcom BCM57780 Gigabit Ethernet"},
268 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57788,
269 		"Broadcom BCM57788 Gigabit Ethernet"},
270 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57790,
271 		"Broadcom BCM57790 Gigabit Ethernet"},
272 	{ PCI_VENDOR_SCHNEIDERKOCH, PCI_PRODUCT_SCHNEIDERKOCH_SK_9DX1,
273 		"SysKonnect Gigabit Ethernet" },
274 
275 	{ 0, 0, NULL }
276 };
277 
278 #define BGE_IS_JUMBO_CAPABLE(sc)	((sc)->bge_flags & BGE_FLAG_JUMBO)
279 #define BGE_IS_5700_FAMILY(sc)		((sc)->bge_flags & BGE_FLAG_5700_FAMILY)
280 #define BGE_IS_5705_PLUS(sc)		((sc)->bge_flags & BGE_FLAG_5705_PLUS)
281 #define BGE_IS_5714_FAMILY(sc)		((sc)->bge_flags & BGE_FLAG_5714_FAMILY)
282 #define BGE_IS_575X_PLUS(sc)		((sc)->bge_flags & BGE_FLAG_575X_PLUS)
283 #define BGE_IS_5755_PLUS(sc)		((sc)->bge_flags & BGE_FLAG_5755_PLUS)
284 #define BGE_IS_5788(sc)			((sc)->bge_flags & BGE_FLAG_5788)
285 
286 #define BGE_IS_CRIPPLED(sc)		\
287 	(BGE_IS_5788((sc)) || (sc)->bge_asicrev == BGE_ASICREV_BCM5700)
288 
289 typedef int	(*bge_eaddr_fcn_t)(struct bge_softc *, uint8_t[]);
290 
291 static int	bge_probe(device_t);
292 static int	bge_attach(device_t);
293 static int	bge_detach(device_t);
294 static void	bge_txeof(struct bge_softc *, uint16_t);
295 static void	bge_rxeof(struct bge_softc *, uint16_t);
296 
297 static void	bge_tick(void *);
298 static void	bge_stats_update(struct bge_softc *);
299 static void	bge_stats_update_regs(struct bge_softc *);
300 static struct mbuf *
301 		bge_defrag_shortdma(struct mbuf *);
302 static int	bge_encap(struct bge_softc *, struct mbuf **, uint32_t *);
303 
304 #ifdef DEVICE_POLLING
305 static void	bge_poll(struct ifnet *ifp, enum poll_cmd cmd, int count);
306 #endif
307 static void	bge_intr_crippled(void *);
308 static void	bge_intr_legacy(void *);
309 static void	bge_msi(void *);
310 static void	bge_msi_oneshot(void *);
311 static void	bge_intr(struct bge_softc *);
312 static void	bge_enable_intr(struct bge_softc *);
313 static void	bge_disable_intr(struct bge_softc *);
314 static void	bge_start(struct ifnet *);
315 static int	bge_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
316 static void	bge_init(void *);
317 static void	bge_stop(struct bge_softc *);
318 static void	bge_watchdog(struct ifnet *);
319 static void	bge_shutdown(device_t);
320 static int	bge_suspend(device_t);
321 static int	bge_resume(device_t);
322 static int	bge_ifmedia_upd(struct ifnet *);
323 static void	bge_ifmedia_sts(struct ifnet *, struct ifmediareq *);
324 
325 static uint8_t	bge_nvram_getbyte(struct bge_softc *, int, uint8_t *);
326 static int	bge_read_nvram(struct bge_softc *, caddr_t, int, int);
327 
328 static uint8_t	bge_eeprom_getbyte(struct bge_softc *, uint32_t, uint8_t *);
329 static int	bge_read_eeprom(struct bge_softc *, caddr_t, uint32_t, size_t);
330 
331 static void	bge_setmulti(struct bge_softc *);
332 static void	bge_setpromisc(struct bge_softc *);
333 static void	bge_enable_msi(struct bge_softc *sc);
334 
335 static int	bge_alloc_jumbo_mem(struct bge_softc *);
336 static void	bge_free_jumbo_mem(struct bge_softc *);
337 static struct bge_jslot
338 		*bge_jalloc(struct bge_softc *);
339 static void	bge_jfree(void *);
340 static void	bge_jref(void *);
341 static int	bge_newbuf_std(struct bge_softc *, int, int);
342 static int	bge_newbuf_jumbo(struct bge_softc *, int, int);
343 static void	bge_setup_rxdesc_std(struct bge_softc *, int);
344 static void	bge_setup_rxdesc_jumbo(struct bge_softc *, int);
345 static int	bge_init_rx_ring_std(struct bge_softc *);
346 static void	bge_free_rx_ring_std(struct bge_softc *);
347 static int	bge_init_rx_ring_jumbo(struct bge_softc *);
348 static void	bge_free_rx_ring_jumbo(struct bge_softc *);
349 static void	bge_free_tx_ring(struct bge_softc *);
350 static int	bge_init_tx_ring(struct bge_softc *);
351 
352 static int	bge_chipinit(struct bge_softc *);
353 static int	bge_blockinit(struct bge_softc *);
354 static void	bge_stop_block(struct bge_softc *, bus_size_t, uint32_t);
355 
356 static uint32_t	bge_readmem_ind(struct bge_softc *, uint32_t);
357 static void	bge_writemem_ind(struct bge_softc *, uint32_t, uint32_t);
358 #ifdef notdef
359 static uint32_t	bge_readreg_ind(struct bge_softc *, uint32_t);
360 #endif
361 static void	bge_writereg_ind(struct bge_softc *, uint32_t, uint32_t);
362 static void	bge_writemem_direct(struct bge_softc *, uint32_t, uint32_t);
363 static void	bge_writembx(struct bge_softc *, int, int);
364 
365 static int	bge_miibus_readreg(device_t, int, int);
366 static int	bge_miibus_writereg(device_t, int, int, int);
367 static void	bge_miibus_statchg(device_t);
368 static void	bge_bcm5700_link_upd(struct bge_softc *, uint32_t);
369 static void	bge_tbi_link_upd(struct bge_softc *, uint32_t);
370 static void	bge_copper_link_upd(struct bge_softc *, uint32_t);
371 static void	bge_autopoll_link_upd(struct bge_softc *, uint32_t);
372 static void	bge_link_poll(struct bge_softc *);
373 
374 static void	bge_reset(struct bge_softc *);
375 
376 static int	bge_dma_alloc(struct bge_softc *);
377 static void	bge_dma_free(struct bge_softc *);
378 static int	bge_dma_block_alloc(struct bge_softc *, bus_size_t,
379 				    bus_dma_tag_t *, bus_dmamap_t *,
380 				    void **, bus_addr_t *);
381 static void	bge_dma_block_free(bus_dma_tag_t, bus_dmamap_t, void *);
382 
383 static int	bge_get_eaddr_mem(struct bge_softc *, uint8_t[]);
384 static int	bge_get_eaddr_nvram(struct bge_softc *, uint8_t[]);
385 static int	bge_get_eaddr_eeprom(struct bge_softc *, uint8_t[]);
386 static int	bge_get_eaddr(struct bge_softc *, uint8_t[]);
387 
388 static void	bge_coal_change(struct bge_softc *);
389 static int	bge_sysctl_rx_coal_ticks(SYSCTL_HANDLER_ARGS);
390 static int	bge_sysctl_tx_coal_ticks(SYSCTL_HANDLER_ARGS);
391 static int	bge_sysctl_rx_coal_bds(SYSCTL_HANDLER_ARGS);
392 static int	bge_sysctl_tx_coal_bds(SYSCTL_HANDLER_ARGS);
393 static int	bge_sysctl_rx_coal_ticks_int(SYSCTL_HANDLER_ARGS);
394 static int	bge_sysctl_tx_coal_ticks_int(SYSCTL_HANDLER_ARGS);
395 static int	bge_sysctl_rx_coal_bds_int(SYSCTL_HANDLER_ARGS);
396 static int	bge_sysctl_tx_coal_bds_int(SYSCTL_HANDLER_ARGS);
397 static int	bge_sysctl_coal_chg(SYSCTL_HANDLER_ARGS, uint32_t *,
398 		    int, int, uint32_t);
399 
400 /*
401  * Set following tunable to 1 for some IBM blade servers with the DNLK
402  * switch module. Auto negotiation is broken for those configurations.
403  */
404 static int	bge_fake_autoneg = 0;
405 TUNABLE_INT("hw.bge.fake_autoneg", &bge_fake_autoneg);
406 
407 static int	bge_msi_enable = 1;
408 TUNABLE_INT("hw.bge.msi.enable", &bge_msi_enable);
409 
410 #if !defined(KTR_IF_BGE)
411 #define KTR_IF_BGE	KTR_ALL
412 #endif
413 KTR_INFO_MASTER(if_bge);
414 KTR_INFO(KTR_IF_BGE, if_bge, intr, 0, "intr");
415 KTR_INFO(KTR_IF_BGE, if_bge, rx_pkt, 1, "rx_pkt");
416 KTR_INFO(KTR_IF_BGE, if_bge, tx_pkt, 2, "tx_pkt");
417 #define logif(name)	KTR_LOG(if_bge_ ## name)
418 
419 static device_method_t bge_methods[] = {
420 	/* Device interface */
421 	DEVMETHOD(device_probe,		bge_probe),
422 	DEVMETHOD(device_attach,	bge_attach),
423 	DEVMETHOD(device_detach,	bge_detach),
424 	DEVMETHOD(device_shutdown,	bge_shutdown),
425 	DEVMETHOD(device_suspend,	bge_suspend),
426 	DEVMETHOD(device_resume,	bge_resume),
427 
428 	/* bus interface */
429 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
430 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
431 
432 	/* MII interface */
433 	DEVMETHOD(miibus_readreg,	bge_miibus_readreg),
434 	DEVMETHOD(miibus_writereg,	bge_miibus_writereg),
435 	DEVMETHOD(miibus_statchg,	bge_miibus_statchg),
436 
437 	{ 0, 0 }
438 };
439 
440 static DEFINE_CLASS_0(bge, bge_driver, bge_methods, sizeof(struct bge_softc));
441 static devclass_t bge_devclass;
442 
443 DECLARE_DUMMY_MODULE(if_bge);
444 DRIVER_MODULE(if_bge, pci, bge_driver, bge_devclass, NULL, NULL);
445 DRIVER_MODULE(miibus, bge, miibus_driver, miibus_devclass, NULL, NULL);
446 
447 static uint32_t
448 bge_readmem_ind(struct bge_softc *sc, uint32_t off)
449 {
450 	device_t dev = sc->bge_dev;
451 	uint32_t val;
452 
453 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
454 	    off >= BGE_STATS_BLOCK && off < BGE_SEND_RING_1_TO_4)
455 		return 0;
456 
457 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4);
458 	val = pci_read_config(dev, BGE_PCI_MEMWIN_DATA, 4);
459 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4);
460 	return (val);
461 }
462 
463 static void
464 bge_writemem_ind(struct bge_softc *sc, uint32_t off, uint32_t val)
465 {
466 	device_t dev = sc->bge_dev;
467 
468 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
469 	    off >= BGE_STATS_BLOCK && off < BGE_SEND_RING_1_TO_4)
470 		return;
471 
472 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4);
473 	pci_write_config(dev, BGE_PCI_MEMWIN_DATA, val, 4);
474 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4);
475 }
476 
477 #ifdef notdef
478 static uint32_t
479 bge_readreg_ind(struct bge_softc *sc, uin32_t off)
480 {
481 	device_t dev = sc->bge_dev;
482 
483 	pci_write_config(dev, BGE_PCI_REG_BASEADDR, off, 4);
484 	return(pci_read_config(dev, BGE_PCI_REG_DATA, 4));
485 }
486 #endif
487 
488 static void
489 bge_writereg_ind(struct bge_softc *sc, uint32_t off, uint32_t val)
490 {
491 	device_t dev = sc->bge_dev;
492 
493 	pci_write_config(dev, BGE_PCI_REG_BASEADDR, off, 4);
494 	pci_write_config(dev, BGE_PCI_REG_DATA, val, 4);
495 }
496 
497 static void
498 bge_writemem_direct(struct bge_softc *sc, uint32_t off, uint32_t val)
499 {
500 	CSR_WRITE_4(sc, off, val);
501 }
502 
503 static void
504 bge_writembx(struct bge_softc *sc, int off, int val)
505 {
506 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
507 		off += BGE_LPMBX_IRQ0_HI - BGE_MBX_IRQ0_HI;
508 
509 	CSR_WRITE_4(sc, off, val);
510 	if (sc->bge_mbox_reorder)
511 		CSR_READ_4(sc, off);
512 }
513 
514 static uint8_t
515 bge_nvram_getbyte(struct bge_softc *sc, int addr, uint8_t *dest)
516 {
517 	uint32_t access, byte = 0;
518 	int i;
519 
520 	/* Lock. */
521 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_SET1);
522 	for (i = 0; i < 8000; i++) {
523 		if (CSR_READ_4(sc, BGE_NVRAM_SWARB) & BGE_NVRAMSWARB_GNT1)
524 			break;
525 		DELAY(20);
526 	}
527 	if (i == 8000)
528 		return (1);
529 
530 	/* Enable access. */
531 	access = CSR_READ_4(sc, BGE_NVRAM_ACCESS);
532 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access | BGE_NVRAMACC_ENABLE);
533 
534 	CSR_WRITE_4(sc, BGE_NVRAM_ADDR, addr & 0xfffffffc);
535 	CSR_WRITE_4(sc, BGE_NVRAM_CMD, BGE_NVRAM_READCMD);
536 	for (i = 0; i < BGE_TIMEOUT * 10; i++) {
537 		DELAY(10);
538 		if (CSR_READ_4(sc, BGE_NVRAM_CMD) & BGE_NVRAMCMD_DONE) {
539 			DELAY(10);
540 			break;
541 		}
542 	}
543 
544 	if (i == BGE_TIMEOUT * 10) {
545 		if_printf(&sc->arpcom.ac_if, "nvram read timed out\n");
546 		return (1);
547 	}
548 
549 	/* Get result. */
550 	byte = CSR_READ_4(sc, BGE_NVRAM_RDDATA);
551 
552 	*dest = (bswap32(byte) >> ((addr % 4) * 8)) & 0xFF;
553 
554 	/* Disable access. */
555 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access);
556 
557 	/* Unlock. */
558 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_CLR1);
559 	CSR_READ_4(sc, BGE_NVRAM_SWARB);
560 
561 	return (0);
562 }
563 
564 /*
565  * Read a sequence of bytes from NVRAM.
566  */
567 static int
568 bge_read_nvram(struct bge_softc *sc, caddr_t dest, int off, int cnt)
569 {
570 	int err = 0, i;
571 	uint8_t byte = 0;
572 
573 	if (sc->bge_asicrev != BGE_ASICREV_BCM5906)
574 		return (1);
575 
576 	for (i = 0; i < cnt; i++) {
577 		err = bge_nvram_getbyte(sc, off + i, &byte);
578 		if (err)
579 			break;
580 		*(dest + i) = byte;
581 	}
582 
583 	return (err ? 1 : 0);
584 }
585 
586 /*
587  * Read a byte of data stored in the EEPROM at address 'addr.' The
588  * BCM570x supports both the traditional bitbang interface and an
589  * auto access interface for reading the EEPROM. We use the auto
590  * access method.
591  */
592 static uint8_t
593 bge_eeprom_getbyte(struct bge_softc *sc, uint32_t addr, uint8_t *dest)
594 {
595 	int i;
596 	uint32_t byte = 0;
597 
598 	/*
599 	 * Enable use of auto EEPROM access so we can avoid
600 	 * having to use the bitbang method.
601 	 */
602 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_AUTO_EEPROM);
603 
604 	/* Reset the EEPROM, load the clock period. */
605 	CSR_WRITE_4(sc, BGE_EE_ADDR,
606 	    BGE_EEADDR_RESET|BGE_EEHALFCLK(BGE_HALFCLK_384SCL));
607 	DELAY(20);
608 
609 	/* Issue the read EEPROM command. */
610 	CSR_WRITE_4(sc, BGE_EE_ADDR, BGE_EE_READCMD | addr);
611 
612 	/* Wait for completion */
613 	for(i = 0; i < BGE_TIMEOUT * 10; i++) {
614 		DELAY(10);
615 		if (CSR_READ_4(sc, BGE_EE_ADDR) & BGE_EEADDR_DONE)
616 			break;
617 	}
618 
619 	if (i == BGE_TIMEOUT) {
620 		if_printf(&sc->arpcom.ac_if, "eeprom read timed out\n");
621 		return(1);
622 	}
623 
624 	/* Get result. */
625 	byte = CSR_READ_4(sc, BGE_EE_DATA);
626 
627         *dest = (byte >> ((addr % 4) * 8)) & 0xFF;
628 
629 	return(0);
630 }
631 
632 /*
633  * Read a sequence of bytes from the EEPROM.
634  */
635 static int
636 bge_read_eeprom(struct bge_softc *sc, caddr_t dest, uint32_t off, size_t len)
637 {
638 	size_t i;
639 	int err;
640 	uint8_t byte;
641 
642 	for (byte = 0, err = 0, i = 0; i < len; i++) {
643 		err = bge_eeprom_getbyte(sc, off + i, &byte);
644 		if (err)
645 			break;
646 		*(dest + i) = byte;
647 	}
648 
649 	return(err ? 1 : 0);
650 }
651 
652 static int
653 bge_miibus_readreg(device_t dev, int phy, int reg)
654 {
655 	struct bge_softc *sc = device_get_softc(dev);
656 	uint32_t val;
657 	int i;
658 
659 	KASSERT(phy == sc->bge_phyno,
660 	    ("invalid phyno %d, should be %d", phy, sc->bge_phyno));
661 
662 	/* Clear the autopoll bit if set, otherwise may trigger PCI errors. */
663 	if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
664 		CSR_WRITE_4(sc, BGE_MI_MODE,
665 		    sc->bge_mi_mode & ~BGE_MIMODE_AUTOPOLL);
666 		DELAY(80);
667 	}
668 
669 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_READ | BGE_MICOMM_BUSY |
670 	    BGE_MIPHY(phy) | BGE_MIREG(reg));
671 
672 	/* Poll for the PHY register access to complete. */
673 	for (i = 0; i < BGE_TIMEOUT; i++) {
674 		DELAY(10);
675 		val = CSR_READ_4(sc, BGE_MI_COMM);
676 		if ((val & BGE_MICOMM_BUSY) == 0) {
677 			DELAY(5);
678 			val = CSR_READ_4(sc, BGE_MI_COMM);
679 			break;
680 		}
681 	}
682 	if (i == BGE_TIMEOUT) {
683 		if_printf(&sc->arpcom.ac_if, "PHY read timed out "
684 		    "(phy %d, reg %d, val 0x%08x)\n", phy, reg, val);
685 		val = 0;
686 	}
687 
688 	/* Restore the autopoll bit if necessary. */
689 	if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
690 		CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode);
691 		DELAY(80);
692 	}
693 
694 	if (val & BGE_MICOMM_READFAIL)
695 		return 0;
696 
697 	return (val & 0xFFFF);
698 }
699 
700 static int
701 bge_miibus_writereg(device_t dev, int phy, int reg, int val)
702 {
703 	struct bge_softc *sc = device_get_softc(dev);
704 	int i;
705 
706 	KASSERT(phy == sc->bge_phyno,
707 	    ("invalid phyno %d, should be %d", phy, sc->bge_phyno));
708 
709 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
710 	    (reg == BRGPHY_MII_1000CTL || reg == BRGPHY_MII_AUXCTL))
711 	       return 0;
712 
713 	/* Clear the autopoll bit if set, otherwise may trigger PCI errors. */
714 	if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
715 		CSR_WRITE_4(sc, BGE_MI_MODE,
716 		    sc->bge_mi_mode & ~BGE_MIMODE_AUTOPOLL);
717 		DELAY(80);
718 	}
719 
720 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_WRITE | BGE_MICOMM_BUSY |
721 	    BGE_MIPHY(phy) | BGE_MIREG(reg) | val);
722 
723 	for (i = 0; i < BGE_TIMEOUT; i++) {
724 		DELAY(10);
725 		if (!(CSR_READ_4(sc, BGE_MI_COMM) & BGE_MICOMM_BUSY)) {
726 			DELAY(5);
727 			CSR_READ_4(sc, BGE_MI_COMM); /* dummy read */
728 			break;
729 		}
730 	}
731 	if (i == BGE_TIMEOUT) {
732 		if_printf(&sc->arpcom.ac_if, "PHY write timed out "
733 		    "(phy %d, reg %d, val %d)\n", phy, reg, val);
734 	}
735 
736 	/* Restore the autopoll bit if necessary. */
737 	if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
738 		CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode);
739 		DELAY(80);
740 	}
741 
742 	return 0;
743 }
744 
745 static void
746 bge_miibus_statchg(device_t dev)
747 {
748 	struct bge_softc *sc;
749 	struct mii_data *mii;
750 
751 	sc = device_get_softc(dev);
752 	mii = device_get_softc(sc->bge_miibus);
753 
754 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
755 	    (IFM_ACTIVE | IFM_AVALID)) {
756 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
757 		case IFM_10_T:
758 		case IFM_100_TX:
759 			sc->bge_link = 1;
760 			break;
761 		case IFM_1000_T:
762 		case IFM_1000_SX:
763 		case IFM_2500_SX:
764 			if (sc->bge_asicrev != BGE_ASICREV_BCM5906)
765 				sc->bge_link = 1;
766 			else
767 				sc->bge_link = 0;
768 			break;
769 		default:
770 			sc->bge_link = 0;
771 			break;
772 		}
773 	} else {
774 		sc->bge_link = 0;
775 	}
776 	if (sc->bge_link == 0)
777 		return;
778 
779 	BGE_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_PORTMODE);
780 	if (IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_T ||
781 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_SX) {
782 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_GMII);
783 	} else {
784 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_MII);
785 	}
786 
787 	if ((mii->mii_media_active & IFM_GMASK) == IFM_FDX) {
788 		BGE_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
789 	} else {
790 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
791 	}
792 }
793 
794 /*
795  * Memory management for jumbo frames.
796  */
797 static int
798 bge_alloc_jumbo_mem(struct bge_softc *sc)
799 {
800 	struct ifnet *ifp = &sc->arpcom.ac_if;
801 	struct bge_jslot *entry;
802 	uint8_t *ptr;
803 	bus_addr_t paddr;
804 	int i, error;
805 
806 	/*
807 	 * Create tag for jumbo mbufs.
808 	 * This is really a bit of a kludge. We allocate a special
809 	 * jumbo buffer pool which (thanks to the way our DMA
810 	 * memory allocation works) will consist of contiguous
811 	 * pages. This means that even though a jumbo buffer might
812 	 * be larger than a page size, we don't really need to
813 	 * map it into more than one DMA segment. However, the
814 	 * default mbuf tag will result in multi-segment mappings,
815 	 * so we have to create a special jumbo mbuf tag that
816 	 * lets us get away with mapping the jumbo buffers as
817 	 * a single segment. I think eventually the driver should
818 	 * be changed so that it uses ordinary mbufs and cluster
819 	 * buffers, i.e. jumbo frames can span multiple DMA
820 	 * descriptors. But that's a project for another day.
821 	 */
822 
823 	/*
824 	 * Create DMA stuffs for jumbo RX ring.
825 	 */
826 	error = bge_dma_block_alloc(sc, BGE_JUMBO_RX_RING_SZ,
827 				    &sc->bge_cdata.bge_rx_jumbo_ring_tag,
828 				    &sc->bge_cdata.bge_rx_jumbo_ring_map,
829 				    (void *)&sc->bge_ldata.bge_rx_jumbo_ring,
830 				    &sc->bge_ldata.bge_rx_jumbo_ring_paddr);
831 	if (error) {
832 		if_printf(ifp, "could not create jumbo RX ring\n");
833 		return error;
834 	}
835 
836 	/*
837 	 * Create DMA stuffs for jumbo buffer block.
838 	 */
839 	error = bge_dma_block_alloc(sc, BGE_JMEM,
840 				    &sc->bge_cdata.bge_jumbo_tag,
841 				    &sc->bge_cdata.bge_jumbo_map,
842 				    (void **)&sc->bge_ldata.bge_jumbo_buf,
843 				    &paddr);
844 	if (error) {
845 		if_printf(ifp, "could not create jumbo buffer\n");
846 		return error;
847 	}
848 
849 	SLIST_INIT(&sc->bge_jfree_listhead);
850 
851 	/*
852 	 * Now divide it up into 9K pieces and save the addresses
853 	 * in an array. Note that we play an evil trick here by using
854 	 * the first few bytes in the buffer to hold the the address
855 	 * of the softc structure for this interface. This is because
856 	 * bge_jfree() needs it, but it is called by the mbuf management
857 	 * code which will not pass it to us explicitly.
858 	 */
859 	for (i = 0, ptr = sc->bge_ldata.bge_jumbo_buf; i < BGE_JSLOTS; i++) {
860 		entry = &sc->bge_cdata.bge_jslots[i];
861 		entry->bge_sc = sc;
862 		entry->bge_buf = ptr;
863 		entry->bge_paddr = paddr;
864 		entry->bge_inuse = 0;
865 		entry->bge_slot = i;
866 		SLIST_INSERT_HEAD(&sc->bge_jfree_listhead, entry, jslot_link);
867 
868 		ptr += BGE_JLEN;
869 		paddr += BGE_JLEN;
870 	}
871 	return 0;
872 }
873 
874 static void
875 bge_free_jumbo_mem(struct bge_softc *sc)
876 {
877 	/* Destroy jumbo RX ring. */
878 	bge_dma_block_free(sc->bge_cdata.bge_rx_jumbo_ring_tag,
879 			   sc->bge_cdata.bge_rx_jumbo_ring_map,
880 			   sc->bge_ldata.bge_rx_jumbo_ring);
881 
882 	/* Destroy jumbo buffer block. */
883 	bge_dma_block_free(sc->bge_cdata.bge_jumbo_tag,
884 			   sc->bge_cdata.bge_jumbo_map,
885 			   sc->bge_ldata.bge_jumbo_buf);
886 }
887 
888 /*
889  * Allocate a jumbo buffer.
890  */
891 static struct bge_jslot *
892 bge_jalloc(struct bge_softc *sc)
893 {
894 	struct bge_jslot *entry;
895 
896 	lwkt_serialize_enter(&sc->bge_jslot_serializer);
897 	entry = SLIST_FIRST(&sc->bge_jfree_listhead);
898 	if (entry) {
899 		SLIST_REMOVE_HEAD(&sc->bge_jfree_listhead, jslot_link);
900 		entry->bge_inuse = 1;
901 	} else {
902 		if_printf(&sc->arpcom.ac_if, "no free jumbo buffers\n");
903 	}
904 	lwkt_serialize_exit(&sc->bge_jslot_serializer);
905 	return(entry);
906 }
907 
908 /*
909  * Adjust usage count on a jumbo buffer.
910  */
911 static void
912 bge_jref(void *arg)
913 {
914 	struct bge_jslot *entry = (struct bge_jslot *)arg;
915 	struct bge_softc *sc = entry->bge_sc;
916 
917 	if (sc == NULL)
918 		panic("bge_jref: can't find softc pointer!");
919 
920 	if (&sc->bge_cdata.bge_jslots[entry->bge_slot] != entry) {
921 		panic("bge_jref: asked to reference buffer "
922 		    "that we don't manage!");
923 	} else if (entry->bge_inuse == 0) {
924 		panic("bge_jref: buffer already free!");
925 	} else {
926 		atomic_add_int(&entry->bge_inuse, 1);
927 	}
928 }
929 
930 /*
931  * Release a jumbo buffer.
932  */
933 static void
934 bge_jfree(void *arg)
935 {
936 	struct bge_jslot *entry = (struct bge_jslot *)arg;
937 	struct bge_softc *sc = entry->bge_sc;
938 
939 	if (sc == NULL)
940 		panic("bge_jfree: can't find softc pointer!");
941 
942 	if (&sc->bge_cdata.bge_jslots[entry->bge_slot] != entry) {
943 		panic("bge_jfree: asked to free buffer that we don't manage!");
944 	} else if (entry->bge_inuse == 0) {
945 		panic("bge_jfree: buffer already free!");
946 	} else {
947 		/*
948 		 * Possible MP race to 0, use the serializer.  The atomic insn
949 		 * is still needed for races against bge_jref().
950 		 */
951 		lwkt_serialize_enter(&sc->bge_jslot_serializer);
952 		atomic_subtract_int(&entry->bge_inuse, 1);
953 		if (entry->bge_inuse == 0) {
954 			SLIST_INSERT_HEAD(&sc->bge_jfree_listhead,
955 					  entry, jslot_link);
956 		}
957 		lwkt_serialize_exit(&sc->bge_jslot_serializer);
958 	}
959 }
960 
961 
962 /*
963  * Intialize a standard receive ring descriptor.
964  */
965 static int
966 bge_newbuf_std(struct bge_softc *sc, int i, int init)
967 {
968 	struct mbuf *m_new = NULL;
969 	bus_dma_segment_t seg;
970 	bus_dmamap_t map;
971 	int error, nsegs;
972 
973 	m_new = m_getcl(init ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
974 	if (m_new == NULL)
975 		return ENOBUFS;
976 	m_new->m_len = m_new->m_pkthdr.len = MCLBYTES;
977 
978 	if ((sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) == 0)
979 		m_adj(m_new, ETHER_ALIGN);
980 
981 	error = bus_dmamap_load_mbuf_segment(sc->bge_cdata.bge_rx_mtag,
982 			sc->bge_cdata.bge_rx_tmpmap, m_new,
983 			&seg, 1, &nsegs, BUS_DMA_NOWAIT);
984 	if (error) {
985 		m_freem(m_new);
986 		return error;
987 	}
988 
989 	if (!init) {
990 		bus_dmamap_sync(sc->bge_cdata.bge_rx_mtag,
991 				sc->bge_cdata.bge_rx_std_dmamap[i],
992 				BUS_DMASYNC_POSTREAD);
993 		bus_dmamap_unload(sc->bge_cdata.bge_rx_mtag,
994 			sc->bge_cdata.bge_rx_std_dmamap[i]);
995 	}
996 
997 	map = sc->bge_cdata.bge_rx_tmpmap;
998 	sc->bge_cdata.bge_rx_tmpmap = sc->bge_cdata.bge_rx_std_dmamap[i];
999 	sc->bge_cdata.bge_rx_std_dmamap[i] = map;
1000 
1001 	sc->bge_cdata.bge_rx_std_chain[i].bge_mbuf = m_new;
1002 	sc->bge_cdata.bge_rx_std_chain[i].bge_paddr = seg.ds_addr;
1003 
1004 	bge_setup_rxdesc_std(sc, i);
1005 	return 0;
1006 }
1007 
1008 static void
1009 bge_setup_rxdesc_std(struct bge_softc *sc, int i)
1010 {
1011 	struct bge_rxchain *rc;
1012 	struct bge_rx_bd *r;
1013 
1014 	rc = &sc->bge_cdata.bge_rx_std_chain[i];
1015 	r = &sc->bge_ldata.bge_rx_std_ring[i];
1016 
1017 	r->bge_addr.bge_addr_lo = BGE_ADDR_LO(rc->bge_paddr);
1018 	r->bge_addr.bge_addr_hi = BGE_ADDR_HI(rc->bge_paddr);
1019 	r->bge_len = rc->bge_mbuf->m_len;
1020 	r->bge_idx = i;
1021 	r->bge_flags = BGE_RXBDFLAG_END;
1022 }
1023 
1024 /*
1025  * Initialize a jumbo receive ring descriptor. This allocates
1026  * a jumbo buffer from the pool managed internally by the driver.
1027  */
1028 static int
1029 bge_newbuf_jumbo(struct bge_softc *sc, int i, int init)
1030 {
1031 	struct mbuf *m_new = NULL;
1032 	struct bge_jslot *buf;
1033 	bus_addr_t paddr;
1034 
1035 	/* Allocate the mbuf. */
1036 	MGETHDR(m_new, init ? MB_WAIT : MB_DONTWAIT, MT_DATA);
1037 	if (m_new == NULL)
1038 		return ENOBUFS;
1039 
1040 	/* Allocate the jumbo buffer */
1041 	buf = bge_jalloc(sc);
1042 	if (buf == NULL) {
1043 		m_freem(m_new);
1044 		return ENOBUFS;
1045 	}
1046 
1047 	/* Attach the buffer to the mbuf. */
1048 	m_new->m_ext.ext_arg = buf;
1049 	m_new->m_ext.ext_buf = buf->bge_buf;
1050 	m_new->m_ext.ext_free = bge_jfree;
1051 	m_new->m_ext.ext_ref = bge_jref;
1052 	m_new->m_ext.ext_size = BGE_JUMBO_FRAMELEN;
1053 
1054 	m_new->m_flags |= M_EXT;
1055 
1056 	m_new->m_data = m_new->m_ext.ext_buf;
1057 	m_new->m_len = m_new->m_pkthdr.len = m_new->m_ext.ext_size;
1058 
1059 	paddr = buf->bge_paddr;
1060 	if ((sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) == 0) {
1061 		m_adj(m_new, ETHER_ALIGN);
1062 		paddr += ETHER_ALIGN;
1063 	}
1064 
1065 	/* Save necessary information */
1066 	sc->bge_cdata.bge_rx_jumbo_chain[i].bge_mbuf = m_new;
1067 	sc->bge_cdata.bge_rx_jumbo_chain[i].bge_paddr = paddr;
1068 
1069 	/* Set up the descriptor. */
1070 	bge_setup_rxdesc_jumbo(sc, i);
1071 	return 0;
1072 }
1073 
1074 static void
1075 bge_setup_rxdesc_jumbo(struct bge_softc *sc, int i)
1076 {
1077 	struct bge_rx_bd *r;
1078 	struct bge_rxchain *rc;
1079 
1080 	r = &sc->bge_ldata.bge_rx_jumbo_ring[i];
1081 	rc = &sc->bge_cdata.bge_rx_jumbo_chain[i];
1082 
1083 	r->bge_addr.bge_addr_lo = BGE_ADDR_LO(rc->bge_paddr);
1084 	r->bge_addr.bge_addr_hi = BGE_ADDR_HI(rc->bge_paddr);
1085 	r->bge_len = rc->bge_mbuf->m_len;
1086 	r->bge_idx = i;
1087 	r->bge_flags = BGE_RXBDFLAG_END|BGE_RXBDFLAG_JUMBO_RING;
1088 }
1089 
1090 static int
1091 bge_init_rx_ring_std(struct bge_softc *sc)
1092 {
1093 	int i, error;
1094 
1095 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1096 		error = bge_newbuf_std(sc, i, 1);
1097 		if (error)
1098 			return error;
1099 	};
1100 
1101 	sc->bge_std = BGE_STD_RX_RING_CNT - 1;
1102 	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bge_std);
1103 
1104 	return(0);
1105 }
1106 
1107 static void
1108 bge_free_rx_ring_std(struct bge_softc *sc)
1109 {
1110 	int i;
1111 
1112 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1113 		struct bge_rxchain *rc = &sc->bge_cdata.bge_rx_std_chain[i];
1114 
1115 		if (rc->bge_mbuf != NULL) {
1116 			bus_dmamap_unload(sc->bge_cdata.bge_rx_mtag,
1117 					  sc->bge_cdata.bge_rx_std_dmamap[i]);
1118 			m_freem(rc->bge_mbuf);
1119 			rc->bge_mbuf = NULL;
1120 		}
1121 		bzero(&sc->bge_ldata.bge_rx_std_ring[i],
1122 		    sizeof(struct bge_rx_bd));
1123 	}
1124 }
1125 
1126 static int
1127 bge_init_rx_ring_jumbo(struct bge_softc *sc)
1128 {
1129 	struct bge_rcb *rcb;
1130 	int i, error;
1131 
1132 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1133 		error = bge_newbuf_jumbo(sc, i, 1);
1134 		if (error)
1135 			return error;
1136 	};
1137 
1138 	sc->bge_jumbo = BGE_JUMBO_RX_RING_CNT - 1;
1139 
1140 	rcb = &sc->bge_ldata.bge_info.bge_jumbo_rx_rcb;
1141 	rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(0, 0);
1142 	CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1143 
1144 	bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
1145 
1146 	return(0);
1147 }
1148 
1149 static void
1150 bge_free_rx_ring_jumbo(struct bge_softc *sc)
1151 {
1152 	int i;
1153 
1154 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1155 		struct bge_rxchain *rc = &sc->bge_cdata.bge_rx_jumbo_chain[i];
1156 
1157 		if (rc->bge_mbuf != NULL) {
1158 			m_freem(rc->bge_mbuf);
1159 			rc->bge_mbuf = NULL;
1160 		}
1161 		bzero(&sc->bge_ldata.bge_rx_jumbo_ring[i],
1162 		    sizeof(struct bge_rx_bd));
1163 	}
1164 }
1165 
1166 static void
1167 bge_free_tx_ring(struct bge_softc *sc)
1168 {
1169 	int i;
1170 
1171 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
1172 		if (sc->bge_cdata.bge_tx_chain[i] != NULL) {
1173 			bus_dmamap_unload(sc->bge_cdata.bge_tx_mtag,
1174 					  sc->bge_cdata.bge_tx_dmamap[i]);
1175 			m_freem(sc->bge_cdata.bge_tx_chain[i]);
1176 			sc->bge_cdata.bge_tx_chain[i] = NULL;
1177 		}
1178 		bzero(&sc->bge_ldata.bge_tx_ring[i],
1179 		    sizeof(struct bge_tx_bd));
1180 	}
1181 }
1182 
1183 static int
1184 bge_init_tx_ring(struct bge_softc *sc)
1185 {
1186 	sc->bge_txcnt = 0;
1187 	sc->bge_tx_saved_considx = 0;
1188 	sc->bge_tx_prodidx = 0;
1189 
1190 	/* Initialize transmit producer index for host-memory send ring. */
1191 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1192 
1193 	/* 5700 b2 errata */
1194 	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
1195 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1196 
1197 	bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1198 	/* 5700 b2 errata */
1199 	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
1200 		bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1201 
1202 	return(0);
1203 }
1204 
1205 static void
1206 bge_setmulti(struct bge_softc *sc)
1207 {
1208 	struct ifnet *ifp;
1209 	struct ifmultiaddr *ifma;
1210 	uint32_t hashes[4] = { 0, 0, 0, 0 };
1211 	int h, i;
1212 
1213 	ifp = &sc->arpcom.ac_if;
1214 
1215 	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
1216 		for (i = 0; i < 4; i++)
1217 			CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0xFFFFFFFF);
1218 		return;
1219 	}
1220 
1221 	/* First, zot all the existing filters. */
1222 	for (i = 0; i < 4; i++)
1223 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0);
1224 
1225 	/* Now program new ones. */
1226 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1227 		if (ifma->ifma_addr->sa_family != AF_LINK)
1228 			continue;
1229 		h = ether_crc32_le(
1230 		    LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1231 		    ETHER_ADDR_LEN) & 0x7f;
1232 		hashes[(h & 0x60) >> 5] |= 1 << (h & 0x1F);
1233 	}
1234 
1235 	for (i = 0; i < 4; i++)
1236 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), hashes[i]);
1237 }
1238 
1239 /*
1240  * Do endian, PCI and DMA initialization. Also check the on-board ROM
1241  * self-test results.
1242  */
1243 static int
1244 bge_chipinit(struct bge_softc *sc)
1245 {
1246 	int i;
1247 	uint32_t dma_rw_ctl;
1248 	uint16_t val;
1249 
1250 	/* Set endian type before we access any non-PCI registers. */
1251 	pci_write_config(sc->bge_dev, BGE_PCI_MISC_CTL,
1252 	    BGE_INIT | sc->bge_pci_miscctl, 4);
1253 
1254 	/* Clear the MAC control register */
1255 	CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
1256 
1257 	/*
1258 	 * Clear the MAC statistics block in the NIC's
1259 	 * internal memory.
1260 	 */
1261 	for (i = BGE_STATS_BLOCK;
1262 	    i < BGE_STATS_BLOCK_END + 1; i += sizeof(uint32_t))
1263 		BGE_MEMWIN_WRITE(sc, i, 0);
1264 
1265 	for (i = BGE_STATUS_BLOCK;
1266 	    i < BGE_STATUS_BLOCK_END + 1; i += sizeof(uint32_t))
1267 		BGE_MEMWIN_WRITE(sc, i, 0);
1268 
1269 	if (sc->bge_chiprev == BGE_CHIPREV_5704_BX) {
1270 		/*
1271 		 * Fix data corruption caused by non-qword write with WB.
1272 		 * Fix master abort in PCI mode.
1273 		 * Fix PCI latency timer.
1274 		 */
1275 		val = pci_read_config(sc->bge_dev, BGE_PCI_MSI_DATA + 2, 2);
1276 		val |= (1 << 10) | (1 << 12) | (1 << 13);
1277 		pci_write_config(sc->bge_dev, BGE_PCI_MSI_DATA + 2, val, 2);
1278 	}
1279 
1280 	/* Set up the PCI DMA control register. */
1281 	if (sc->bge_flags & BGE_FLAG_PCIE) {
1282 		/* PCI Express */
1283 		dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD |
1284 		    (0xf << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1285 		    (0x2 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1286 	} else if (sc->bge_flags & BGE_FLAG_PCIX) {
1287 		/* PCI-X bus */
1288 		if (BGE_IS_5714_FAMILY(sc)) {
1289 			dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD;
1290 			dma_rw_ctl &= ~BGE_PCIDMARWCTL_ONEDMA_ATONCE; /* XXX */
1291 			/* XXX magic values, Broadcom-supplied Linux driver */
1292 			if (sc->bge_asicrev == BGE_ASICREV_BCM5780) {
1293 				dma_rw_ctl |= (1 << 20) | (1 << 18) |
1294 				    BGE_PCIDMARWCTL_ONEDMA_ATONCE;
1295 			} else {
1296 				dma_rw_ctl |= (1 << 20) | (1 << 18) | (1 << 15);
1297 			}
1298 		} else if (sc->bge_asicrev == BGE_ASICREV_BCM5703) {
1299 			/*
1300 			 * In the BCM5703, the DMA read watermark should
1301 			 * be set to less than or equal to the maximum
1302 			 * memory read byte count of the PCI-X command
1303 			 * register.
1304 			 */
1305 			dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD |
1306 			    (0x4 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1307 			    (0x3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1308 		} else if (sc->bge_asicrev == BGE_ASICREV_BCM5704) {
1309 			/*
1310 			 * The 5704 uses a different encoding of read/write
1311 			 * watermarks.
1312 			 */
1313 			dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD |
1314 			    (0x7 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1315 			    (0x3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1316 		} else {
1317 			dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD |
1318 			    (0x3 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1319 			    (0x3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT) |
1320 			    (0x0F);
1321 		}
1322 
1323 		/*
1324 		 * 5703 and 5704 need ONEDMA_AT_ONCE as a workaround
1325 		 * for hardware bugs.
1326 		 */
1327 		if (sc->bge_asicrev == BGE_ASICREV_BCM5703 ||
1328 		    sc->bge_asicrev == BGE_ASICREV_BCM5704) {
1329 			uint32_t tmp;
1330 
1331 			tmp = CSR_READ_4(sc, BGE_PCI_CLKCTL) & 0x1f;
1332 			if (tmp == 0x6 || tmp == 0x7)
1333 				dma_rw_ctl |= BGE_PCIDMARWCTL_ONEDMA_ATONCE;
1334 		}
1335 	} else {
1336 		/* Conventional PCI bus */
1337 		dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD |
1338 		    (0x7 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1339 		    (0x7 << BGE_PCIDMARWCTL_WR_WAT_SHIFT) |
1340 		    (0x0F);
1341 	}
1342 
1343 	if (sc->bge_asicrev == BGE_ASICREV_BCM5703 ||
1344 	    sc->bge_asicrev == BGE_ASICREV_BCM5704 ||
1345 	    sc->bge_asicrev == BGE_ASICREV_BCM5705)
1346 		dma_rw_ctl &= ~BGE_PCIDMARWCTL_MINDMA;
1347 	pci_write_config(sc->bge_dev, BGE_PCI_DMA_RW_CTL, dma_rw_ctl, 4);
1348 
1349 	/*
1350 	 * Set up general mode register.
1351 	 */
1352 	CSR_WRITE_4(sc, BGE_MODE_CTL, BGE_DMA_SWAP_OPTIONS|
1353 	    BGE_MODECTL_MAC_ATTN_INTR|BGE_MODECTL_HOST_SEND_BDS|
1354 	    BGE_MODECTL_TX_NO_PHDR_CSUM);
1355 
1356 	/*
1357 	 * BCM5701 B5 have a bug causing data corruption when using
1358 	 * 64-bit DMA reads, which can be terminated early and then
1359 	 * completed later as 32-bit accesses, in combination with
1360 	 * certain bridges.
1361 	 */
1362 	if (sc->bge_asicrev == BGE_ASICREV_BCM5701 &&
1363 	    sc->bge_chipid == BGE_CHIPID_BCM5701_B5)
1364 		BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_FORCE_PCI32);
1365 
1366 	/*
1367 	 * Disable memory write invalidate.  Apparently it is not supported
1368 	 * properly by these devices.  Also ensure that INTx isn't disabled,
1369 	 * as these chips need it even when using MSI.
1370 	 */
1371 	PCI_CLRBIT(sc->bge_dev, BGE_PCI_CMD,
1372 	    (PCIM_CMD_MWRICEN | PCIM_CMD_INTxDIS), 4);
1373 
1374 	/* Set the timer prescaler (always 66Mhz) */
1375 	CSR_WRITE_4(sc, BGE_MISC_CFG, 65 << 1/*BGE_32BITTIME_66MHZ*/);
1376 
1377 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
1378 		DELAY(40);	/* XXX */
1379 
1380 		/* Put PHY into ready state */
1381 		BGE_CLRBIT(sc, BGE_MISC_CFG, BGE_MISCCFG_EPHY_IDDQ);
1382 		CSR_READ_4(sc, BGE_MISC_CFG); /* Flush */
1383 		DELAY(40);
1384 	}
1385 
1386 	return(0);
1387 }
1388 
1389 static int
1390 bge_blockinit(struct bge_softc *sc)
1391 {
1392 	struct bge_rcb *rcb;
1393 	bus_size_t vrcb;
1394 	bge_hostaddr taddr;
1395 	uint32_t val;
1396 	int i, limit;
1397 
1398 	/*
1399 	 * Initialize the memory window pointer register so that
1400 	 * we can access the first 32K of internal NIC RAM. This will
1401 	 * allow us to set up the TX send ring RCBs and the RX return
1402 	 * ring RCBs, plus other things which live in NIC memory.
1403 	 */
1404 	CSR_WRITE_4(sc, BGE_PCI_MEMWIN_BASEADDR, 0);
1405 
1406 	/* Note: the BCM5704 has a smaller mbuf space than other chips. */
1407 
1408 	if (!BGE_IS_5705_PLUS(sc)) {
1409 		/* Configure mbuf memory pool */
1410 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_BASEADDR, BGE_BUFFPOOL_1);
1411 		if (sc->bge_asicrev == BGE_ASICREV_BCM5704)
1412 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x10000);
1413 		else
1414 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x18000);
1415 
1416 		/* Configure DMA resource pool */
1417 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_BASEADDR,
1418 		    BGE_DMA_DESCRIPTORS);
1419 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LEN, 0x2000);
1420 	}
1421 
1422 	/* Configure mbuf pool watermarks */
1423 	if (!BGE_IS_5705_PLUS(sc)) {
1424 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x50);
1425 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x20);
1426 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1427 	} else if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
1428 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1429 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x04);
1430 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x10);
1431 	} else {
1432 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1433 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x10);
1434 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1435 	}
1436 
1437 	/* Configure DMA resource watermarks */
1438 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LOWAT, 5);
1439 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_HIWAT, 10);
1440 
1441 	/* Enable buffer manager */
1442 	CSR_WRITE_4(sc, BGE_BMAN_MODE,
1443 	    BGE_BMANMODE_ENABLE|BGE_BMANMODE_LOMBUF_ATTN);
1444 
1445 	/* Poll for buffer manager start indication */
1446 	for (i = 0; i < BGE_TIMEOUT; i++) {
1447 		if (CSR_READ_4(sc, BGE_BMAN_MODE) & BGE_BMANMODE_ENABLE)
1448 			break;
1449 		DELAY(10);
1450 	}
1451 
1452 	if (i == BGE_TIMEOUT) {
1453 		if_printf(&sc->arpcom.ac_if,
1454 			  "buffer manager failed to start\n");
1455 		return(ENXIO);
1456 	}
1457 
1458 	/* Enable flow-through queues */
1459 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
1460 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
1461 
1462 	/* Wait until queue initialization is complete */
1463 	for (i = 0; i < BGE_TIMEOUT; i++) {
1464 		if (CSR_READ_4(sc, BGE_FTQ_RESET) == 0)
1465 			break;
1466 		DELAY(10);
1467 	}
1468 
1469 	if (i == BGE_TIMEOUT) {
1470 		if_printf(&sc->arpcom.ac_if,
1471 			  "flow-through queue init failed\n");
1472 		return(ENXIO);
1473 	}
1474 
1475 	/*
1476 	 * Summary of rings supported by the controller:
1477 	 *
1478 	 * Standard Receive Producer Ring
1479 	 * - This ring is used to feed receive buffers for "standard"
1480 	 *   sized frames (typically 1536 bytes) to the controller.
1481 	 *
1482 	 * Jumbo Receive Producer Ring
1483 	 * - This ring is used to feed receive buffers for jumbo sized
1484 	 *   frames (i.e. anything bigger than the "standard" frames)
1485 	 *   to the controller.
1486 	 *
1487 	 * Mini Receive Producer Ring
1488 	 * - This ring is used to feed receive buffers for "mini"
1489 	 *   sized frames to the controller.
1490 	 * - This feature required external memory for the controller
1491 	 *   but was never used in a production system.  Should always
1492 	 *   be disabled.
1493 	 *
1494 	 * Receive Return Ring
1495 	 * - After the controller has placed an incoming frame into a
1496 	 *   receive buffer that buffer is moved into a receive return
1497 	 *   ring.  The driver is then responsible to passing the
1498 	 *   buffer up to the stack.  Many versions of the controller
1499 	 *   support multiple RR rings.
1500 	 *
1501 	 * Send Ring
1502 	 * - This ring is used for outgoing frames.  Many versions of
1503 	 *   the controller support multiple send rings.
1504 	 */
1505 
1506 	/* Initialize the standard receive producer ring control block. */
1507 	rcb = &sc->bge_ldata.bge_info.bge_std_rx_rcb;
1508 	rcb->bge_hostaddr.bge_addr_lo =
1509 	    BGE_ADDR_LO(sc->bge_ldata.bge_rx_std_ring_paddr);
1510 	rcb->bge_hostaddr.bge_addr_hi =
1511 	    BGE_ADDR_HI(sc->bge_ldata.bge_rx_std_ring_paddr);
1512 	if (BGE_IS_5705_PLUS(sc)) {
1513 		/*
1514 		 * Bits 31-16: Programmable ring size (512, 256, 128, 64, 32)
1515 		 * Bits 15-2 : Reserved (should be 0)
1516 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
1517 		 * Bit 0     : Reserved
1518 		 */
1519 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(512, 0);
1520 	} else {
1521 		/*
1522 		 * Ring size is always XXX entries
1523 		 * Bits 31-16: Maximum RX frame size
1524 		 * Bits 15-2 : Reserved (should be 0)
1525 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
1526 		 * Bit 0     : Reserved
1527 		 */
1528 		rcb->bge_maxlen_flags =
1529 		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN, 0);
1530 	}
1531 	rcb->bge_nicaddr = BGE_STD_RX_RINGS;
1532 	/* Write the standard receive producer ring control block. */
1533 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_HI, rcb->bge_hostaddr.bge_addr_hi);
1534 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_LO, rcb->bge_hostaddr.bge_addr_lo);
1535 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1536 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_NICADDR, rcb->bge_nicaddr);
1537 	/* Reset the standard receive producer ring producer index. */
1538 	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, 0);
1539 
1540 	/*
1541 	 * Initialize the jumbo RX producer ring control
1542 	 * block.  We set the 'ring disabled' bit in the
1543 	 * flags field until we're actually ready to start
1544 	 * using this ring (i.e. once we set the MTU
1545 	 * high enough to require it).
1546 	 */
1547 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
1548 		rcb = &sc->bge_ldata.bge_info.bge_jumbo_rx_rcb;
1549 		/* Get the jumbo receive producer ring RCB parameters. */
1550 		rcb->bge_hostaddr.bge_addr_lo =
1551 		    BGE_ADDR_LO(sc->bge_ldata.bge_rx_jumbo_ring_paddr);
1552 		rcb->bge_hostaddr.bge_addr_hi =
1553 		    BGE_ADDR_HI(sc->bge_ldata.bge_rx_jumbo_ring_paddr);
1554 		rcb->bge_maxlen_flags =
1555 		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN,
1556 		    BGE_RCB_FLAG_RING_DISABLED);
1557 		rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS;
1558 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_HI,
1559 		    rcb->bge_hostaddr.bge_addr_hi);
1560 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_LO,
1561 		    rcb->bge_hostaddr.bge_addr_lo);
1562 		/* Program the jumbo receive producer ring RCB parameters. */
1563 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS,
1564 		    rcb->bge_maxlen_flags);
1565 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_NICADDR, rcb->bge_nicaddr);
1566 		/* Reset the jumbo receive producer ring producer index. */
1567 		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, 0);
1568 	}
1569 
1570 	/* Disable the mini receive producer ring RCB. */
1571 	if (BGE_IS_5700_FAMILY(sc)) {
1572 		rcb = &sc->bge_ldata.bge_info.bge_mini_rx_rcb;
1573 		rcb->bge_maxlen_flags =
1574 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED);
1575 		CSR_WRITE_4(sc, BGE_RX_MINI_RCB_MAXLEN_FLAGS,
1576 		    rcb->bge_maxlen_flags);
1577 		/* Reset the mini receive producer ring producer index. */
1578 		bge_writembx(sc, BGE_MBX_RX_MINI_PROD_LO, 0);
1579 	}
1580 
1581 	/* Choose de-pipeline mode for BCM5906 A0, A1 and A2. */
1582 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
1583 	    (sc->bge_chipid == BGE_CHIPID_BCM5906_A0 ||
1584 	     sc->bge_chipid == BGE_CHIPID_BCM5906_A1 ||
1585 	     sc->bge_chipid == BGE_CHIPID_BCM5906_A2)) {
1586 		CSR_WRITE_4(sc, BGE_ISO_PKT_TX,
1587 		    (CSR_READ_4(sc, BGE_ISO_PKT_TX) & ~3) | 2);
1588 	}
1589 
1590 	/*
1591 	 * The BD ring replenish thresholds control how often the
1592 	 * hardware fetches new BD's from the producer rings in host
1593 	 * memory.  Setting the value too low on a busy system can
1594 	 * starve the hardware and recue the throughpout.
1595 	 *
1596 	 * Set the BD ring replentish thresholds. The recommended
1597 	 * values are 1/8th the number of descriptors allocated to
1598 	 * each ring.
1599 	 */
1600 	if (BGE_IS_5705_PLUS(sc))
1601 		val = 8;
1602 	else
1603 		val = BGE_STD_RX_RING_CNT / 8;
1604 	CSR_WRITE_4(sc, BGE_RBDI_STD_REPL_THRESH, val);
1605 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
1606 		CSR_WRITE_4(sc, BGE_RBDI_JUMBO_REPL_THRESH,
1607 		    BGE_JUMBO_RX_RING_CNT/8);
1608 	}
1609 
1610 	/*
1611 	 * Disable all send rings by setting the 'ring disabled' bit
1612 	 * in the flags field of all the TX send ring control blocks,
1613 	 * located in NIC memory.
1614 	 */
1615 	if (!BGE_IS_5705_PLUS(sc)) {
1616 		/* 5700 to 5704 had 16 send rings. */
1617 		limit = BGE_TX_RINGS_EXTSSRAM_MAX;
1618 	} else {
1619 		limit = 1;
1620 	}
1621 	vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1622 	for (i = 0; i < limit; i++) {
1623 		RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1624 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED));
1625 		RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1626 		vrcb += sizeof(struct bge_rcb);
1627 	}
1628 
1629 	/* Configure send ring RCB 0 (we use only the first ring) */
1630 	vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1631 	BGE_HOSTADDR(taddr, sc->bge_ldata.bge_tx_ring_paddr);
1632 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1633 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1634 	RCB_WRITE_4(sc, vrcb, bge_nicaddr,
1635 	    BGE_NIC_TXRING_ADDR(0, BGE_TX_RING_CNT));
1636 	RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1637 	    BGE_RCB_MAXLEN_FLAGS(BGE_TX_RING_CNT, 0));
1638 
1639 	/*
1640 	 * Disable all receive return rings by setting the
1641 	 * 'ring diabled' bit in the flags field of all the receive
1642 	 * return ring control blocks, located in NIC memory.
1643 	 */
1644 	if (!BGE_IS_5705_PLUS(sc))
1645 		limit = BGE_RX_RINGS_MAX;
1646 	else if (sc->bge_asicrev == BGE_ASICREV_BCM5755)
1647 		limit = 4;
1648 	else
1649 		limit = 1;
1650 	/* Disable all receive return rings. */
1651 	vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1652 	for (i = 0; i < limit; i++) {
1653 		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, 0);
1654 		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, 0);
1655 		RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1656 		    BGE_RCB_FLAG_RING_DISABLED);
1657 		RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1658 		bge_writembx(sc, BGE_MBX_RX_CONS0_LO +
1659 		    (i * (sizeof(uint64_t))), 0);
1660 		vrcb += sizeof(struct bge_rcb);
1661 	}
1662 
1663 	/*
1664 	 * Set up receive return ring 0.  Note that the NIC address
1665 	 * for RX return rings is 0x0.  The return rings live entirely
1666 	 * within the host, so the nicaddr field in the RCB isn't used.
1667 	 */
1668 	vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1669 	BGE_HOSTADDR(taddr, sc->bge_ldata.bge_rx_return_ring_paddr);
1670 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1671 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1672 	RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1673 	RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1674 	    BGE_RCB_MAXLEN_FLAGS(sc->bge_return_ring_cnt, 0));
1675 
1676 	/* Set random backoff seed for TX */
1677 	CSR_WRITE_4(sc, BGE_TX_RANDOM_BACKOFF,
1678 	    sc->arpcom.ac_enaddr[0] + sc->arpcom.ac_enaddr[1] +
1679 	    sc->arpcom.ac_enaddr[2] + sc->arpcom.ac_enaddr[3] +
1680 	    sc->arpcom.ac_enaddr[4] + sc->arpcom.ac_enaddr[5] +
1681 	    BGE_TX_BACKOFF_SEED_MASK);
1682 
1683 	/* Set inter-packet gap */
1684 	CSR_WRITE_4(sc, BGE_TX_LENGTHS, 0x2620);
1685 
1686 	/*
1687 	 * Specify which ring to use for packets that don't match
1688 	 * any RX rules.
1689 	 */
1690 	CSR_WRITE_4(sc, BGE_RX_RULES_CFG, 0x08);
1691 
1692 	/*
1693 	 * Configure number of RX lists. One interrupt distribution
1694 	 * list, sixteen active lists, one bad frames class.
1695 	 */
1696 	CSR_WRITE_4(sc, BGE_RXLP_CFG, 0x181);
1697 
1698 	/* Inialize RX list placement stats mask. */
1699 	CSR_WRITE_4(sc, BGE_RXLP_STATS_ENABLE_MASK, 0x007FFFFF);
1700 	CSR_WRITE_4(sc, BGE_RXLP_STATS_CTL, 0x1);
1701 
1702 	/* Disable host coalescing until we get it set up */
1703 	CSR_WRITE_4(sc, BGE_HCC_MODE, 0x00000000);
1704 
1705 	/* Poll to make sure it's shut down. */
1706 	for (i = 0; i < BGE_TIMEOUT; i++) {
1707 		if (!(CSR_READ_4(sc, BGE_HCC_MODE) & BGE_HCCMODE_ENABLE))
1708 			break;
1709 		DELAY(10);
1710 	}
1711 
1712 	if (i == BGE_TIMEOUT) {
1713 		if_printf(&sc->arpcom.ac_if,
1714 			  "host coalescing engine failed to idle\n");
1715 		return(ENXIO);
1716 	}
1717 
1718 	/* Set up host coalescing defaults */
1719 	CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, sc->bge_rx_coal_ticks);
1720 	CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS, sc->bge_tx_coal_ticks);
1721 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, sc->bge_rx_coal_bds);
1722 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS, sc->bge_tx_coal_bds);
1723 	if (!BGE_IS_5705_PLUS(sc)) {
1724 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS_INT,
1725 		    sc->bge_rx_coal_ticks_int);
1726 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS_INT,
1727 		    sc->bge_tx_coal_ticks_int);
1728 	}
1729 	/*
1730 	 * NOTE:
1731 	 * The datasheet (57XX-PG105-R) says BCM5705+ do not
1732 	 * have following two registers; obviously it is wrong.
1733 	 */
1734 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT, sc->bge_rx_coal_bds_int);
1735 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT, sc->bge_tx_coal_bds_int);
1736 
1737 	/* Set up address of statistics block */
1738 	if (!BGE_IS_5705_PLUS(sc)) {
1739 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_HI,
1740 		    BGE_ADDR_HI(sc->bge_ldata.bge_stats_paddr));
1741 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_LO,
1742 		    BGE_ADDR_LO(sc->bge_ldata.bge_stats_paddr));
1743 
1744 		CSR_WRITE_4(sc, BGE_HCC_STATS_BASEADDR, BGE_STATS_BLOCK);
1745 		CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_BASEADDR, BGE_STATUS_BLOCK);
1746 		CSR_WRITE_4(sc, BGE_HCC_STATS_TICKS, sc->bge_stat_ticks);
1747 	}
1748 
1749 	/* Set up address of status block */
1750 	bzero(sc->bge_ldata.bge_status_block, BGE_STATUS_BLK_SZ);
1751 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_HI,
1752 	    BGE_ADDR_HI(sc->bge_ldata.bge_status_block_paddr));
1753 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_LO,
1754 	    BGE_ADDR_LO(sc->bge_ldata.bge_status_block_paddr));
1755 
1756 	/*
1757 	 * Set up status block partail update size.
1758 	 *
1759 	 * Because only single TX ring, RX produce ring and Rx return ring
1760 	 * are used, ask device to update only minimum part of status block
1761 	 * except for BCM5700 AX/BX, whose status block partial update size
1762 	 * can't be configured.
1763 	 */
1764 	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
1765 	    sc->bge_chipid != BGE_CHIPID_BCM5700_C0) {
1766 		/* XXX Actually reserved on BCM5700 AX/BX */
1767 		val = BGE_STATBLKSZ_FULL;
1768 	} else {
1769 		val = BGE_STATBLKSZ_32BYTE;
1770 	}
1771 #if 0
1772 	/*
1773 	 * Does not seem to have visible effect in both
1774 	 * bulk data (1472B UDP datagram) and tiny data
1775 	 * (18B UDP datagram) TX tests.
1776 	 */
1777 	if (!BGE_IS_CRIPPLED(sc))
1778 		val |= BGE_HCCMODE_CLRTICK_TX;
1779 #endif
1780 
1781 	/* Turn on host coalescing state machine */
1782 	CSR_WRITE_4(sc, BGE_HCC_MODE, val | BGE_HCCMODE_ENABLE);
1783 
1784 	/* Turn on RX BD completion state machine and enable attentions */
1785 	CSR_WRITE_4(sc, BGE_RBDC_MODE,
1786 	    BGE_RBDCMODE_ENABLE|BGE_RBDCMODE_ATTN);
1787 
1788 	/* Turn on RX list placement state machine */
1789 	CSR_WRITE_4(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
1790 
1791 	/* Turn on RX list selector state machine. */
1792 	if (!BGE_IS_5705_PLUS(sc))
1793 		CSR_WRITE_4(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
1794 
1795 	val = BGE_MACMODE_TXDMA_ENB | BGE_MACMODE_RXDMA_ENB |
1796 	    BGE_MACMODE_RX_STATS_CLEAR | BGE_MACMODE_TX_STATS_CLEAR |
1797 	    BGE_MACMODE_RX_STATS_ENB | BGE_MACMODE_TX_STATS_ENB |
1798 	    BGE_MACMODE_FRMHDR_DMA_ENB;
1799 
1800 	if (sc->bge_flags & BGE_FLAG_TBI)
1801 		val |= BGE_PORTMODE_TBI;
1802 	else if (sc->bge_flags & BGE_FLAG_MII_SERDES)
1803 		val |= BGE_PORTMODE_GMII;
1804 	else
1805 		val |= BGE_PORTMODE_MII;
1806 
1807 	/* Turn on DMA, clear stats */
1808 	CSR_WRITE_4(sc, BGE_MAC_MODE, val);
1809 
1810 	/* Set misc. local control, enable interrupts on attentions */
1811 	CSR_WRITE_4(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_ONATTN);
1812 
1813 #ifdef notdef
1814 	/* Assert GPIO pins for PHY reset */
1815 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUT0|
1816 	    BGE_MLC_MISCIO_OUT1|BGE_MLC_MISCIO_OUT2);
1817 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUTEN0|
1818 	    BGE_MLC_MISCIO_OUTEN1|BGE_MLC_MISCIO_OUTEN2);
1819 #endif
1820 
1821 	/* Turn on DMA completion state machine */
1822 	if (!BGE_IS_5705_PLUS(sc))
1823 		CSR_WRITE_4(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
1824 
1825 	/* Turn on write DMA state machine */
1826 	val = BGE_WDMAMODE_ENABLE|BGE_WDMAMODE_ALL_ATTNS;
1827 	if (BGE_IS_5755_PLUS(sc)) {
1828 		/* Enable host coalescing bug fix. */
1829 		val |= BGE_WDMAMODE_STATUS_TAG_FIX;
1830 	}
1831 	if (sc->bge_asicrev == BGE_ASICREV_BCM5785) {
1832 		/* Request larger DMA burst size to get better performance. */
1833 		val |= BGE_WDMAMODE_BURST_ALL_DATA;
1834 	}
1835 	CSR_WRITE_4(sc, BGE_WDMA_MODE, val);
1836 	DELAY(40);
1837 
1838 	if (sc->bge_asicrev == BGE_ASICREV_BCM5761 ||
1839 	    sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
1840 	    sc->bge_asicrev == BGE_ASICREV_BCM5785 ||
1841 	    sc->bge_asicrev == BGE_ASICREV_BCM57780) {
1842 		/*
1843 		 * Enable fix for read DMA FIFO overruns.
1844 		 * The fix is to limit the number of RX BDs
1845 		 * the hardware would fetch at a fime.
1846 		 */
1847 		val = CSR_READ_4(sc, BGE_RDMA_RSRVCTRL);
1848 		CSR_WRITE_4(sc, BGE_RDMA_RSRVCTRL,
1849 		    val| BGE_RDMA_RSRVCTRL_FIFO_OFLW_FIX);
1850 	}
1851 
1852 	/* Turn on read DMA state machine */
1853 	val = BGE_RDMAMODE_ENABLE | BGE_RDMAMODE_ALL_ATTNS;
1854         if (sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
1855             sc->bge_asicrev == BGE_ASICREV_BCM5785 ||
1856             sc->bge_asicrev == BGE_ASICREV_BCM57780)
1857 		val |= BGE_RDMAMODE_BD_SBD_CRPT_ATTN |
1858                   BGE_RDMAMODE_MBUF_RBD_CRPT_ATTN |
1859                   BGE_RDMAMODE_MBUF_SBD_CRPT_ATTN;
1860 	if (sc->bge_flags & BGE_FLAG_PCIE)
1861 		val |= BGE_RDMAMODE_FIFO_LONG_BURST;
1862 	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 	CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
1889 
1890 	/* Turn on send BD initiator state machine */
1891 	CSR_WRITE_4(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
1892 
1893 	/* Turn on send BD selector state machine */
1894 	CSR_WRITE_4(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
1895 
1896 	CSR_WRITE_4(sc, BGE_SDI_STATS_ENABLE_MASK, 0x007FFFFF);
1897 	CSR_WRITE_4(sc, BGE_SDI_STATS_CTL,
1898 	    BGE_SDISTATSCTL_ENABLE|BGE_SDISTATSCTL_FASTER);
1899 
1900 	/* ack/clear link change events */
1901 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
1902 	    BGE_MACSTAT_CFG_CHANGED|BGE_MACSTAT_MI_COMPLETE|
1903 	    BGE_MACSTAT_LINK_CHANGED);
1904 	CSR_WRITE_4(sc, BGE_MI_STS, 0);
1905 
1906 	/*
1907 	 * Enable attention when the link has changed state for
1908 	 * devices that use auto polling.
1909 	 */
1910 	if (sc->bge_flags & BGE_FLAG_TBI) {
1911 		CSR_WRITE_4(sc, BGE_MI_STS, BGE_MISTS_LINK);
1912  	} else {
1913 		if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
1914 			CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode);
1915 			DELAY(80);
1916 		}
1917 		if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
1918 		    sc->bge_chipid != BGE_CHIPID_BCM5700_B2) {
1919 			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
1920 			    BGE_EVTENB_MI_INTERRUPT);
1921 		}
1922 	}
1923 
1924 	/*
1925 	 * Clear any pending link state attention.
1926 	 * Otherwise some link state change events may be lost until attention
1927 	 * is cleared by bge_intr() -> bge_softc.bge_link_upd() sequence.
1928 	 * It's not necessary on newer BCM chips - perhaps enabling link
1929 	 * state change attentions implies clearing pending attention.
1930 	 */
1931 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
1932 	    BGE_MACSTAT_CFG_CHANGED|BGE_MACSTAT_MI_COMPLETE|
1933 	    BGE_MACSTAT_LINK_CHANGED);
1934 
1935 	/* Enable link state change attentions. */
1936 	BGE_SETBIT(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_LINK_CHANGED);
1937 
1938 	return(0);
1939 }
1940 
1941 /*
1942  * Probe for a Broadcom chip. Check the PCI vendor and device IDs
1943  * against our list and return its name if we find a match. Note
1944  * that since the Broadcom controller contains VPD support, we
1945  * can get the device name string from the controller itself instead
1946  * of the compiled-in string. This is a little slow, but it guarantees
1947  * we'll always announce the right product name.
1948  */
1949 static int
1950 bge_probe(device_t dev)
1951 {
1952 	const struct bge_type *t;
1953 	uint16_t product, vendor;
1954 
1955 	product = pci_get_device(dev);
1956 	vendor = pci_get_vendor(dev);
1957 
1958 	for (t = bge_devs; t->bge_name != NULL; t++) {
1959 		if (vendor == t->bge_vid && product == t->bge_did)
1960 			break;
1961 	}
1962 	if (t->bge_name == NULL)
1963 		return(ENXIO);
1964 
1965 	device_set_desc(dev, t->bge_name);
1966 	return(0);
1967 }
1968 
1969 static int
1970 bge_attach(device_t dev)
1971 {
1972 	struct ifnet *ifp;
1973 	struct bge_softc *sc;
1974 	uint32_t hwcfg = 0, misccfg;
1975 	int error = 0, rid, capmask;
1976 	uint8_t ether_addr[ETHER_ADDR_LEN];
1977 	uint16_t product, vendor;
1978 	driver_intr_t *intr_func;
1979 	uintptr_t mii_priv = 0;
1980 	u_int intr_flags;
1981 	int msi_enable;
1982 
1983 	sc = device_get_softc(dev);
1984 	sc->bge_dev = dev;
1985 	callout_init(&sc->bge_stat_timer);
1986 	lwkt_serialize_init(&sc->bge_jslot_serializer);
1987 
1988 #ifndef BURN_BRIDGES
1989 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1990 		uint32_t irq, mem;
1991 
1992 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
1993 		mem = pci_read_config(dev, BGE_PCI_BAR0, 4);
1994 
1995 		device_printf(dev, "chip is in D%d power mode "
1996 		    "-- setting to D0\n", pci_get_powerstate(dev));
1997 
1998 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
1999 
2000 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
2001 		pci_write_config(dev, BGE_PCI_BAR0, mem, 4);
2002 	}
2003 #endif	/* !BURN_BRIDGE */
2004 
2005 	/*
2006 	 * Map control/status registers.
2007 	 */
2008 	pci_enable_busmaster(dev);
2009 
2010 	rid = BGE_PCI_BAR0;
2011 	sc->bge_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
2012 	    RF_ACTIVE);
2013 
2014 	if (sc->bge_res == NULL) {
2015 		device_printf(dev, "couldn't map memory\n");
2016 		return ENXIO;
2017 	}
2018 
2019 	sc->bge_btag = rman_get_bustag(sc->bge_res);
2020 	sc->bge_bhandle = rman_get_bushandle(sc->bge_res);
2021 
2022 	/* Save various chip information */
2023 	sc->bge_chipid =
2024 	    pci_read_config(dev, BGE_PCI_MISC_CTL, 4) >>
2025 	    BGE_PCIMISCCTL_ASICREV_SHIFT;
2026         if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_USE_PRODID_REG)
2027 		sc->bge_chipid = pci_read_config(dev, BGE_PCI_PRODID_ASICREV, 4);
2028 	sc->bge_asicrev = BGE_ASICREV(sc->bge_chipid);
2029 	sc->bge_chiprev = BGE_CHIPREV(sc->bge_chipid);
2030 
2031 	/* Save chipset family. */
2032 	switch (sc->bge_asicrev) {
2033 	case BGE_ASICREV_BCM5755:
2034 	case BGE_ASICREV_BCM5761:
2035 	case BGE_ASICREV_BCM5784:
2036 	case BGE_ASICREV_BCM5785:
2037 	case BGE_ASICREV_BCM5787:
2038 	case BGE_ASICREV_BCM57780:
2039 	    sc->bge_flags |= BGE_FLAG_5755_PLUS | BGE_FLAG_575X_PLUS |
2040 		BGE_FLAG_5705_PLUS;
2041 	    break;
2042 
2043 	case BGE_ASICREV_BCM5700:
2044 	case BGE_ASICREV_BCM5701:
2045 	case BGE_ASICREV_BCM5703:
2046 	case BGE_ASICREV_BCM5704:
2047 		sc->bge_flags |= BGE_FLAG_5700_FAMILY | BGE_FLAG_JUMBO;
2048 		break;
2049 
2050 	case BGE_ASICREV_BCM5714_A0:
2051 	case BGE_ASICREV_BCM5780:
2052 	case BGE_ASICREV_BCM5714:
2053 		sc->bge_flags |= BGE_FLAG_5714_FAMILY;
2054 		/* Fall through */
2055 
2056 	case BGE_ASICREV_BCM5750:
2057 	case BGE_ASICREV_BCM5752:
2058 	case BGE_ASICREV_BCM5906:
2059 		sc->bge_flags |= BGE_FLAG_575X_PLUS;
2060 		/* Fall through */
2061 
2062 	case BGE_ASICREV_BCM5705:
2063 		sc->bge_flags |= BGE_FLAG_5705_PLUS;
2064 		break;
2065 	}
2066 
2067 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
2068 		sc->bge_flags |= BGE_FLAG_NO_EEPROM;
2069 
2070 	misccfg = CSR_READ_4(sc, BGE_MISC_CFG) & BGE_MISCCFG_BOARD_ID_MASK;
2071 	if (sc->bge_asicrev == BGE_ASICREV_BCM5705 &&
2072 	    (misccfg == BGE_MISCCFG_BOARD_ID_5788 ||
2073 	     misccfg == BGE_MISCCFG_BOARD_ID_5788M))
2074 		sc->bge_flags |= BGE_FLAG_5788;
2075 
2076 	/* BCM5755 or higher and BCM5906 have short DMA bug. */
2077 	if (BGE_IS_5755_PLUS(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5906)
2078 		sc->bge_flags |= BGE_FLAG_SHORTDMA;
2079 
2080   	/*
2081 	 * Check if this is a PCI-X or PCI Express device.
2082   	 */
2083 	if (BGE_IS_5705_PLUS(sc)) {
2084 		if (pci_is_pcie(dev)) {
2085 			sc->bge_flags |= BGE_FLAG_PCIE;
2086 			sc->bge_pciecap = pci_get_pciecap_ptr(sc->bge_dev);
2087 			pcie_set_max_readrq(dev, PCIEM_DEVCTL_MAX_READRQ_4096);
2088 		}
2089 	} else {
2090 		/*
2091 		 * Check if the device is in PCI-X Mode.
2092 		 * (This bit is not valid on PCI Express controllers.)
2093 		 */
2094 		if ((pci_read_config(sc->bge_dev, BGE_PCI_PCISTATE, 4) &
2095 		    BGE_PCISTATE_PCI_BUSMODE) == 0) {
2096 			sc->bge_flags |= BGE_FLAG_PCIX;
2097 			sc->bge_pcixcap = pci_get_pcixcap_ptr(sc->bge_dev);
2098 			sc->bge_mbox_reorder = device_getenv_int(sc->bge_dev,
2099 			    "mbox_reorder", 0);
2100 		}
2101  	}
2102 	device_printf(dev, "CHIP ID 0x%08x; "
2103 		      "ASIC REV 0x%02x; CHIP REV 0x%02x; %s\n",
2104 		      sc->bge_chipid, sc->bge_asicrev, sc->bge_chiprev,
2105 		      (sc->bge_flags & BGE_FLAG_PCIX) ? "PCI-X"
2106 		      : ((sc->bge_flags & BGE_FLAG_PCIE) ?
2107 			"PCI-E" : "PCI"));
2108 
2109 	/*
2110 	 * The 40bit DMA bug applies to the 5714/5715 controllers and is
2111 	 * not actually a MAC controller bug but an issue with the embedded
2112 	 * PCIe to PCI-X bridge in the device. Use 40bit DMA workaround.
2113 	 */
2114 	if (BGE_IS_5714_FAMILY(sc) && (sc->bge_flags & BGE_FLAG_PCIX))
2115 		sc->bge_flags |= BGE_FLAG_MAXADDR_40BIT;
2116 
2117 	/* Identify the chips that use an CPMU. */
2118 	if (sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
2119 	    sc->bge_asicrev == BGE_ASICREV_BCM5761 ||
2120 	    sc->bge_asicrev == BGE_ASICREV_BCM5785 ||
2121 	    sc->bge_asicrev == BGE_ASICREV_BCM57780)
2122 		sc->bge_flags |= BGE_FLAG_CPMU;
2123 
2124 	/*
2125 	 * When using the BCM5701 in PCI-X mode, data corruption has
2126 	 * been observed in the first few bytes of some received packets.
2127 	 * Aligning the packet buffer in memory eliminates the corruption.
2128 	 * Unfortunately, this misaligns the packet payloads.  On platforms
2129 	 * which do not support unaligned accesses, we will realign the
2130 	 * payloads by copying the received packets.
2131 	 */
2132 	if (sc->bge_asicrev == BGE_ASICREV_BCM5701 &&
2133 	    (sc->bge_flags & BGE_FLAG_PCIX))
2134 		sc->bge_flags |= BGE_FLAG_RX_ALIGNBUG;
2135 
2136 	if (!BGE_IS_CRIPPLED(sc)) {
2137 		if (device_getenv_int(dev, "status_tag", 1)) {
2138 			sc->bge_flags |= BGE_FLAG_STATUS_TAG;
2139 			sc->bge_pci_miscctl = BGE_PCIMISCCTL_TAGGED_STATUS;
2140 			if (bootverbose)
2141 				device_printf(dev, "enable status tag\n");
2142 		}
2143 	}
2144 
2145 	/*
2146 	 * Set various PHY quirk flags.
2147 	 */
2148 	product = pci_get_device(dev);
2149 	vendor = pci_get_vendor(dev);
2150 
2151 	if ((sc->bge_asicrev == BGE_ASICREV_BCM5700 ||
2152 	     sc->bge_asicrev == BGE_ASICREV_BCM5701) &&
2153 	    pci_get_subvendor(dev) == PCI_VENDOR_DELL)
2154 		mii_priv |= BRGPHY_FLAG_NO_3LED;
2155 
2156 	capmask = MII_CAPMASK_DEFAULT;
2157 	if ((sc->bge_asicrev == BGE_ASICREV_BCM5703 &&
2158 	     (misccfg == 0x4000 || misccfg == 0x8000)) ||
2159 	    (sc->bge_asicrev == BGE_ASICREV_BCM5705 &&
2160 	     vendor == PCI_VENDOR_BROADCOM &&
2161 	     (product == PCI_PRODUCT_BROADCOM_BCM5901 ||
2162 	      product == PCI_PRODUCT_BROADCOM_BCM5901A2 ||
2163 	      product == PCI_PRODUCT_BROADCOM_BCM5705F)) ||
2164 	    (vendor == PCI_VENDOR_BROADCOM &&
2165 	     (product == PCI_PRODUCT_BROADCOM_BCM5751F ||
2166 	      product == PCI_PRODUCT_BROADCOM_BCM5753F ||
2167 	      product == PCI_PRODUCT_BROADCOM_BCM5787F)) ||
2168 	    product == PCI_PRODUCT_BROADCOM_BCM57790 ||
2169 	    sc->bge_asicrev == BGE_ASICREV_BCM5906) {
2170 		/* 10/100 only */
2171 		capmask &= ~BMSR_EXTSTAT;
2172 	}
2173 
2174 	mii_priv |= BRGPHY_FLAG_WIRESPEED;
2175 	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 ||
2176 	    (sc->bge_asicrev == BGE_ASICREV_BCM5705 &&
2177 	     (sc->bge_chipid != BGE_CHIPID_BCM5705_A0 &&
2178 	      sc->bge_chipid != BGE_CHIPID_BCM5705_A1)) ||
2179 	    sc->bge_asicrev == BGE_ASICREV_BCM5906)
2180 		mii_priv &= ~BRGPHY_FLAG_WIRESPEED;
2181 
2182 	if (sc->bge_chipid == BGE_CHIPID_BCM5701_A0 ||
2183 	    sc->bge_chipid == BGE_CHIPID_BCM5701_B0)
2184 		mii_priv |= BRGPHY_FLAG_CRC_BUG;
2185 
2186 	if (sc->bge_chiprev == BGE_CHIPREV_5703_AX ||
2187 	    sc->bge_chiprev == BGE_CHIPREV_5704_AX)
2188 		mii_priv |= BRGPHY_FLAG_ADC_BUG;
2189 
2190 	if (sc->bge_chipid == BGE_CHIPID_BCM5704_A0)
2191 		mii_priv |= BRGPHY_FLAG_5704_A0;
2192 
2193 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
2194 		mii_priv |= BRGPHY_FLAG_5906;
2195 
2196 	if (BGE_IS_5705_PLUS(sc) &&
2197 	    sc->bge_asicrev != BGE_ASICREV_BCM5906 &&
2198 	    /* sc->bge_asicrev != BGE_ASICREV_BCM5717 && */
2199 	    sc->bge_asicrev != BGE_ASICREV_BCM5785 &&
2200 	    /* sc->bge_asicrev != BGE_ASICREV_BCM57765 && */
2201 	    sc->bge_asicrev != BGE_ASICREV_BCM57780) {
2202 		if (sc->bge_asicrev == BGE_ASICREV_BCM5755 ||
2203 		    sc->bge_asicrev == BGE_ASICREV_BCM5761 ||
2204 		    sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
2205 		    sc->bge_asicrev == BGE_ASICREV_BCM5787) {
2206 			if (product != PCI_PRODUCT_BROADCOM_BCM5722 &&
2207 			    product != PCI_PRODUCT_BROADCOM_BCM5756)
2208 				mii_priv |= BRGPHY_FLAG_JITTER_BUG;
2209 			if (product == PCI_PRODUCT_BROADCOM_BCM5755M)
2210 				mii_priv |= BRGPHY_FLAG_ADJUST_TRIM;
2211 		} else {
2212 			mii_priv |= BRGPHY_FLAG_BER_BUG;
2213 		}
2214 	}
2215 
2216 	/*
2217 	 * Allocate interrupt
2218 	 */
2219 	msi_enable = bge_msi_enable;
2220 	if ((sc->bge_flags & BGE_FLAG_STATUS_TAG) == 0) {
2221 		/* If "tagged status" is disabled, don't enable MSI */
2222 		msi_enable = 0;
2223 	} else if (msi_enable) {
2224 		msi_enable = 0; /* Disable by default */
2225 		if (BGE_IS_575X_PLUS(sc)) {
2226 			msi_enable = 1;
2227 			/* XXX we filter all 5714 chips */
2228 			if (sc->bge_asicrev == BGE_ASICREV_BCM5714 ||
2229 			    (sc->bge_asicrev == BGE_ASICREV_BCM5750 &&
2230 			     (sc->bge_chiprev == BGE_CHIPREV_5750_AX ||
2231 			      sc->bge_chiprev == BGE_CHIPREV_5750_BX)))
2232 				msi_enable = 0;
2233 			else if (BGE_IS_5755_PLUS(sc) ||
2234 			    sc->bge_asicrev == BGE_ASICREV_BCM5906)
2235 				sc->bge_flags |= BGE_FLAG_ONESHOT_MSI;
2236 		}
2237 	}
2238 	if (msi_enable) {
2239 		if (pci_find_extcap(dev, PCIY_MSI, &sc->bge_msicap)) {
2240 			device_printf(dev, "no MSI capability\n");
2241 			msi_enable = 0;
2242 		}
2243 	}
2244 
2245 	sc->bge_irq_type = pci_alloc_1intr(dev, msi_enable, &sc->bge_irq_rid,
2246 	    &intr_flags);
2247 
2248 	sc->bge_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &sc->bge_irq_rid,
2249 	    intr_flags);
2250 	if (sc->bge_irq == NULL) {
2251 		device_printf(dev, "couldn't map interrupt\n");
2252 		error = ENXIO;
2253 		goto fail;
2254 	}
2255 
2256 	if (sc->bge_irq_type == PCI_INTR_TYPE_MSI)
2257 		bge_enable_msi(sc);
2258 	else
2259 		sc->bge_flags &= ~BGE_FLAG_ONESHOT_MSI;
2260 
2261 	/* Initialize if_name earlier, so if_printf could be used */
2262 	ifp = &sc->arpcom.ac_if;
2263 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
2264 
2265 	/* Try to reset the chip. */
2266 	bge_reset(sc);
2267 
2268 	if (bge_chipinit(sc)) {
2269 		device_printf(dev, "chip initialization failed\n");
2270 		error = ENXIO;
2271 		goto fail;
2272 	}
2273 
2274 	/*
2275 	 * Get station address
2276 	 */
2277 	error = bge_get_eaddr(sc, ether_addr);
2278 	if (error) {
2279 		device_printf(dev, "failed to read station address\n");
2280 		goto fail;
2281 	}
2282 
2283 	/* 5705/5750 limits RX return ring to 512 entries. */
2284 	if (BGE_IS_5705_PLUS(sc))
2285 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT_5705;
2286 	else
2287 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT;
2288 
2289 	error = bge_dma_alloc(sc);
2290 	if (error)
2291 		goto fail;
2292 
2293 	/* Set default tuneable values. */
2294 	sc->bge_stat_ticks = BGE_TICKS_PER_SEC;
2295 	sc->bge_rx_coal_ticks = BGE_RX_COAL_TICKS_DEF;
2296 	sc->bge_tx_coal_ticks = BGE_TX_COAL_TICKS_DEF;
2297 	sc->bge_rx_coal_bds = BGE_RX_COAL_BDS_DEF;
2298 	sc->bge_tx_coal_bds = BGE_TX_COAL_BDS_DEF;
2299 	if (sc->bge_flags & BGE_FLAG_STATUS_TAG) {
2300 		sc->bge_rx_coal_ticks_int = BGE_RX_COAL_TICKS_DEF;
2301 		sc->bge_tx_coal_ticks_int = BGE_TX_COAL_TICKS_DEF;
2302 		sc->bge_rx_coal_bds_int = BGE_RX_COAL_BDS_DEF;
2303 		sc->bge_tx_coal_bds_int = BGE_TX_COAL_BDS_DEF;
2304 	} else {
2305 		sc->bge_rx_coal_ticks_int = BGE_RX_COAL_TICKS_MIN;
2306 		sc->bge_tx_coal_ticks_int = BGE_TX_COAL_TICKS_MIN;
2307 		sc->bge_rx_coal_bds_int = BGE_RX_COAL_BDS_MIN;
2308 		sc->bge_tx_coal_bds_int = BGE_TX_COAL_BDS_MIN;
2309 	}
2310 
2311 	/* Set up ifnet structure */
2312 	ifp->if_softc = sc;
2313 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2314 	ifp->if_ioctl = bge_ioctl;
2315 	ifp->if_start = bge_start;
2316 #ifdef DEVICE_POLLING
2317 	ifp->if_poll = bge_poll;
2318 #endif
2319 	ifp->if_watchdog = bge_watchdog;
2320 	ifp->if_init = bge_init;
2321 	ifp->if_mtu = ETHERMTU;
2322 	ifp->if_capabilities = IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU;
2323 	ifq_set_maxlen(&ifp->if_snd, BGE_TX_RING_CNT - 1);
2324 	ifq_set_ready(&ifp->if_snd);
2325 
2326 	/*
2327 	 * 5700 B0 chips do not support checksumming correctly due
2328 	 * to hardware bugs.
2329 	 */
2330 	if (sc->bge_chipid != BGE_CHIPID_BCM5700_B0) {
2331 		ifp->if_capabilities |= IFCAP_HWCSUM;
2332 		ifp->if_hwassist = BGE_CSUM_FEATURES;
2333 	}
2334 	ifp->if_capenable = ifp->if_capabilities;
2335 
2336 	/*
2337 	 * Figure out what sort of media we have by checking the
2338 	 * hardware config word in the first 32k of NIC internal memory,
2339 	 * or fall back to examining the EEPROM if necessary.
2340 	 * Note: on some BCM5700 cards, this value appears to be unset.
2341 	 * If that's the case, we have to rely on identifying the NIC
2342 	 * by its PCI subsystem ID, as we do below for the SysKonnect
2343 	 * SK-9D41.
2344 	 */
2345 	if (bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_SIG) == BGE_MAGIC_NUMBER) {
2346 		hwcfg = bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_NICCFG);
2347 	} else {
2348 		if (bge_read_eeprom(sc, (caddr_t)&hwcfg, BGE_EE_HWCFG_OFFSET,
2349 				    sizeof(hwcfg))) {
2350 			device_printf(dev, "failed to read EEPROM\n");
2351 			error = ENXIO;
2352 			goto fail;
2353 		}
2354 		hwcfg = ntohl(hwcfg);
2355 	}
2356 
2357 	/* The SysKonnect SK-9D41 is a 1000baseSX card. */
2358 	if (pci_get_subvendor(dev) == PCI_PRODUCT_SCHNEIDERKOCH_SK_9D41 ||
2359 	    (hwcfg & BGE_HWCFG_MEDIA) == BGE_MEDIA_FIBER) {
2360 		if (BGE_IS_5714_FAMILY(sc))
2361 			sc->bge_flags |= BGE_FLAG_MII_SERDES;
2362 		else
2363 			sc->bge_flags |= BGE_FLAG_TBI;
2364 	}
2365 
2366 	/* Setup MI MODE */
2367 	if (sc->bge_flags & BGE_FLAG_CPMU)
2368 		sc->bge_mi_mode = BGE_MIMODE_500KHZ_CONST;
2369 	else
2370 		sc->bge_mi_mode = BGE_MIMODE_BASE;
2371 	if (BGE_IS_5700_FAMILY(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5705) {
2372 		/* Enable auto polling for BCM570[0-5]. */
2373 		sc->bge_mi_mode |= BGE_MIMODE_AUTOPOLL;
2374 	}
2375 
2376 	/* Setup link status update stuffs */
2377 	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
2378 	    sc->bge_chipid != BGE_CHIPID_BCM5700_B2) {
2379 		sc->bge_link_upd = bge_bcm5700_link_upd;
2380 		sc->bge_link_chg = BGE_MACSTAT_MI_INTERRUPT;
2381 	} else if (sc->bge_flags & BGE_FLAG_TBI) {
2382 		sc->bge_link_upd = bge_tbi_link_upd;
2383 		sc->bge_link_chg = BGE_MACSTAT_LINK_CHANGED;
2384 	} else if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
2385 		sc->bge_link_upd = bge_autopoll_link_upd;
2386 		sc->bge_link_chg = BGE_MACSTAT_LINK_CHANGED;
2387 	} else {
2388 		sc->bge_link_upd = bge_copper_link_upd;
2389 		sc->bge_link_chg = BGE_MACSTAT_LINK_CHANGED;
2390 	}
2391 
2392 	/*
2393 	 * Broadcom's own driver always assumes the internal
2394 	 * PHY is at GMII address 1.  On some chips, the PHY responds
2395 	 * to accesses at all addresses, which could cause us to
2396 	 * bogusly attach the PHY 32 times at probe type.  Always
2397 	 * restricting the lookup to address 1 is simpler than
2398 	 * trying to figure out which chips revisions should be
2399 	 * special-cased.
2400 	 */
2401 	sc->bge_phyno = 1;
2402 
2403 	if (sc->bge_flags & BGE_FLAG_TBI) {
2404 		ifmedia_init(&sc->bge_ifmedia, IFM_IMASK,
2405 		    bge_ifmedia_upd, bge_ifmedia_sts);
2406 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER|IFM_1000_SX, 0, NULL);
2407 		ifmedia_add(&sc->bge_ifmedia,
2408 		    IFM_ETHER|IFM_1000_SX|IFM_FDX, 0, NULL);
2409 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER|IFM_AUTO, 0, NULL);
2410 		ifmedia_set(&sc->bge_ifmedia, IFM_ETHER|IFM_AUTO);
2411 		sc->bge_ifmedia.ifm_media = sc->bge_ifmedia.ifm_cur->ifm_media;
2412 	} else {
2413 		struct mii_probe_args mii_args;
2414 
2415 		mii_probe_args_init(&mii_args, bge_ifmedia_upd, bge_ifmedia_sts);
2416 		mii_args.mii_probemask = 1 << sc->bge_phyno;
2417 		mii_args.mii_capmask = capmask;
2418 		mii_args.mii_privtag = MII_PRIVTAG_BRGPHY;
2419 		mii_args.mii_priv = mii_priv;
2420 
2421 		error = mii_probe(dev, &sc->bge_miibus, &mii_args);
2422 		if (error) {
2423 			device_printf(dev, "MII without any PHY!\n");
2424 			goto fail;
2425 		}
2426 	}
2427 
2428 	/*
2429 	 * Create sysctl nodes.
2430 	 */
2431 	sysctl_ctx_init(&sc->bge_sysctl_ctx);
2432 	sc->bge_sysctl_tree = SYSCTL_ADD_NODE(&sc->bge_sysctl_ctx,
2433 					      SYSCTL_STATIC_CHILDREN(_hw),
2434 					      OID_AUTO,
2435 					      device_get_nameunit(dev),
2436 					      CTLFLAG_RD, 0, "");
2437 	if (sc->bge_sysctl_tree == NULL) {
2438 		device_printf(dev, "can't add sysctl node\n");
2439 		error = ENXIO;
2440 		goto fail;
2441 	}
2442 
2443 	SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2444 			SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2445 			OID_AUTO, "rx_coal_ticks",
2446 			CTLTYPE_INT | CTLFLAG_RW,
2447 			sc, 0, bge_sysctl_rx_coal_ticks, "I",
2448 			"Receive coalescing ticks (usec).");
2449 	SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2450 			SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2451 			OID_AUTO, "tx_coal_ticks",
2452 			CTLTYPE_INT | CTLFLAG_RW,
2453 			sc, 0, bge_sysctl_tx_coal_ticks, "I",
2454 			"Transmit coalescing ticks (usec).");
2455 	SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2456 			SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2457 			OID_AUTO, "rx_coal_bds",
2458 			CTLTYPE_INT | CTLFLAG_RW,
2459 			sc, 0, bge_sysctl_rx_coal_bds, "I",
2460 			"Receive max coalesced BD count.");
2461 	SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2462 			SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2463 			OID_AUTO, "tx_coal_bds",
2464 			CTLTYPE_INT | CTLFLAG_RW,
2465 			sc, 0, bge_sysctl_tx_coal_bds, "I",
2466 			"Transmit max coalesced BD count.");
2467 	if (sc->bge_flags & BGE_FLAG_PCIE) {
2468 		/*
2469 		 * A common design characteristic for many Broadcom
2470 		 * client controllers is that they only support a
2471 		 * single outstanding DMA read operation on the PCIe
2472 		 * bus. This means that it will take twice as long to
2473 		 * fetch a TX frame that is split into header and
2474 		 * payload buffers as it does to fetch a single,
2475 		 * contiguous TX frame (2 reads vs. 1 read). For these
2476 		 * controllers, coalescing buffers to reduce the number
2477 		 * of memory reads is effective way to get maximum
2478 		 * performance(about 940Mbps).  Without collapsing TX
2479 		 * buffers the maximum TCP bulk transfer performance
2480 		 * is about 850Mbps. However forcing coalescing mbufs
2481 		 * consumes a lot of CPU cycles, so leave it off by
2482 		 * default.
2483 		 */
2484 		SYSCTL_ADD_INT(&sc->bge_sysctl_ctx,
2485 			       SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2486 			       OID_AUTO, "force_defrag", CTLFLAG_RW,
2487 			       &sc->bge_force_defrag, 0,
2488 			       "Force defragment on TX path");
2489 	}
2490 	if (sc->bge_flags & BGE_FLAG_STATUS_TAG) {
2491 		if (!BGE_IS_5705_PLUS(sc)) {
2492 			SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2493 			    SYSCTL_CHILDREN(sc->bge_sysctl_tree), OID_AUTO,
2494 			    "rx_coal_ticks_int", CTLTYPE_INT | CTLFLAG_RW,
2495 			    sc, 0, bge_sysctl_rx_coal_ticks_int, "I",
2496 			    "Receive coalescing ticks "
2497 			    "during interrupt (usec).");
2498 			SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2499 			    SYSCTL_CHILDREN(sc->bge_sysctl_tree), OID_AUTO,
2500 			    "tx_coal_ticks_int", CTLTYPE_INT | CTLFLAG_RW,
2501 			    sc, 0, bge_sysctl_tx_coal_ticks_int, "I",
2502 			    "Transmit coalescing ticks "
2503 			    "during interrupt (usec).");
2504 		}
2505 		SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2506 		    SYSCTL_CHILDREN(sc->bge_sysctl_tree), OID_AUTO,
2507 		    "rx_coal_bds_int", CTLTYPE_INT | CTLFLAG_RW,
2508 		    sc, 0, bge_sysctl_rx_coal_bds_int, "I",
2509 		    "Receive max coalesced BD count during interrupt.");
2510 		SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2511 		    SYSCTL_CHILDREN(sc->bge_sysctl_tree), OID_AUTO,
2512 		    "tx_coal_bds_int", CTLTYPE_INT | CTLFLAG_RW,
2513 		    sc, 0, bge_sysctl_tx_coal_bds_int, "I",
2514 		    "Transmit max coalesced BD count during interrupt.");
2515 	}
2516 
2517 	/*
2518 	 * Call MI attach routine.
2519 	 */
2520 	ether_ifattach(ifp, ether_addr, NULL);
2521 
2522 	if (sc->bge_irq_type == PCI_INTR_TYPE_MSI) {
2523 		if (sc->bge_flags & BGE_FLAG_ONESHOT_MSI) {
2524 			intr_func = bge_msi_oneshot;
2525 			if (bootverbose)
2526 				device_printf(dev, "oneshot MSI\n");
2527 		} else {
2528 			intr_func = bge_msi;
2529 		}
2530 	} else if (sc->bge_flags & BGE_FLAG_STATUS_TAG) {
2531 		intr_func = bge_intr_legacy;
2532 	} else {
2533 		intr_func = bge_intr_crippled;
2534 	}
2535 	error = bus_setup_intr(dev, sc->bge_irq, INTR_MPSAFE, intr_func, sc,
2536 	    &sc->bge_intrhand, ifp->if_serializer);
2537 	if (error) {
2538 		ether_ifdetach(ifp);
2539 		device_printf(dev, "couldn't set up irq\n");
2540 		goto fail;
2541 	}
2542 
2543 	ifp->if_cpuid = rman_get_cpuid(sc->bge_irq);
2544 	KKASSERT(ifp->if_cpuid >= 0 && ifp->if_cpuid < ncpus);
2545 
2546 	return(0);
2547 fail:
2548 	bge_detach(dev);
2549 	return(error);
2550 }
2551 
2552 static int
2553 bge_detach(device_t dev)
2554 {
2555 	struct bge_softc *sc = device_get_softc(dev);
2556 
2557 	if (device_is_attached(dev)) {
2558 		struct ifnet *ifp = &sc->arpcom.ac_if;
2559 
2560 		lwkt_serialize_enter(ifp->if_serializer);
2561 		bge_stop(sc);
2562 		bge_reset(sc);
2563 		bus_teardown_intr(dev, sc->bge_irq, sc->bge_intrhand);
2564 		lwkt_serialize_exit(ifp->if_serializer);
2565 
2566 		ether_ifdetach(ifp);
2567 	}
2568 
2569 	if (sc->bge_flags & BGE_FLAG_TBI)
2570 		ifmedia_removeall(&sc->bge_ifmedia);
2571 	if (sc->bge_miibus)
2572 		device_delete_child(dev, sc->bge_miibus);
2573 	bus_generic_detach(dev);
2574 
2575 	if (sc->bge_irq != NULL) {
2576 		bus_release_resource(dev, SYS_RES_IRQ, sc->bge_irq_rid,
2577 		    sc->bge_irq);
2578 	}
2579 	if (sc->bge_irq_type == PCI_INTR_TYPE_MSI)
2580 		pci_release_msi(dev);
2581 
2582 	if (sc->bge_res != NULL) {
2583 		bus_release_resource(dev, SYS_RES_MEMORY,
2584 		    BGE_PCI_BAR0, sc->bge_res);
2585 	}
2586 
2587 	if (sc->bge_sysctl_tree != NULL)
2588 		sysctl_ctx_free(&sc->bge_sysctl_ctx);
2589 
2590 	bge_dma_free(sc);
2591 
2592 	return 0;
2593 }
2594 
2595 static void
2596 bge_reset(struct bge_softc *sc)
2597 {
2598 	device_t dev;
2599 	uint32_t cachesize, command, pcistate, reset;
2600 	void (*write_op)(struct bge_softc *, uint32_t, uint32_t);
2601 	int i, val = 0;
2602 
2603 	dev = sc->bge_dev;
2604 
2605 	if (BGE_IS_575X_PLUS(sc) && !BGE_IS_5714_FAMILY(sc) &&
2606 	    sc->bge_asicrev != BGE_ASICREV_BCM5906) {
2607 		if (sc->bge_flags & BGE_FLAG_PCIE)
2608 			write_op = bge_writemem_direct;
2609 		else
2610 			write_op = bge_writemem_ind;
2611 	} else {
2612 		write_op = bge_writereg_ind;
2613 	}
2614 
2615 	/* Save some important PCI state. */
2616 	cachesize = pci_read_config(dev, BGE_PCI_CACHESZ, 4);
2617 	command = pci_read_config(dev, BGE_PCI_CMD, 4);
2618 	pcistate = pci_read_config(dev, BGE_PCI_PCISTATE, 4);
2619 
2620 	pci_write_config(dev, BGE_PCI_MISC_CTL,
2621 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2622 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW|
2623 	    sc->bge_pci_miscctl, 4);
2624 
2625 	/* Disable fastboot on controllers that support it. */
2626 	if (sc->bge_asicrev == BGE_ASICREV_BCM5752 ||
2627 	    BGE_IS_5755_PLUS(sc)) {
2628 		if (bootverbose)
2629 			if_printf(&sc->arpcom.ac_if, "Disabling fastboot\n");
2630 		CSR_WRITE_4(sc, BGE_FASTBOOT_PC, 0x0);
2631 	}
2632 
2633 	/*
2634 	 * Write the magic number to SRAM at offset 0xB50.
2635 	 * When firmware finishes its initialization it will
2636 	 * write ~BGE_MAGIC_NUMBER to the same location.
2637 	 */
2638 	bge_writemem_ind(sc, BGE_SOFTWARE_GENCOMM, BGE_MAGIC_NUMBER);
2639 
2640 	reset = BGE_MISCCFG_RESET_CORE_CLOCKS|(65<<1);
2641 
2642 	/* XXX: Broadcom Linux driver. */
2643 	if (sc->bge_flags & BGE_FLAG_PCIE) {
2644 		/* Force PCI-E 1.0a mode */
2645 		if (sc->bge_asicrev != BGE_ASICREV_BCM5785 &&
2646 		    CSR_READ_4(sc, BGE_PCIE_PHY_TSTCTL) ==
2647 		    (BGE_PCIE_PHY_TSTCTL_PSCRAM |
2648 		     BGE_PCIE_PHY_TSTCTL_PCIE10)) {
2649 			CSR_WRITE_4(sc, BGE_PCIE_PHY_TSTCTL,
2650 			    BGE_PCIE_PHY_TSTCTL_PSCRAM);
2651 		}
2652 		if (sc->bge_chipid != BGE_CHIPID_BCM5750_A0) {
2653 			/* Prevent PCIE link training during global reset */
2654 			CSR_WRITE_4(sc, BGE_MISC_CFG, (1<<29));
2655 			reset |= (1<<29);
2656 		}
2657 	}
2658 
2659 	/*
2660 	 * Set GPHY Power Down Override to leave GPHY
2661 	 * powered up in D0 uninitialized.
2662 	 */
2663 	if (BGE_IS_5705_PLUS(sc) && (sc->bge_flags & BGE_FLAG_CPMU) == 0)
2664 		reset |= BGE_MISCCFG_GPHY_PD_OVERRIDE;
2665 
2666 	/* Issue global reset */
2667 	write_op(sc, BGE_MISC_CFG, reset);
2668 
2669 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
2670 		uint32_t status, ctrl;
2671 
2672 		status = CSR_READ_4(sc, BGE_VCPU_STATUS);
2673 		CSR_WRITE_4(sc, BGE_VCPU_STATUS,
2674 		    status | BGE_VCPU_STATUS_DRV_RESET);
2675 		ctrl = CSR_READ_4(sc, BGE_VCPU_EXT_CTRL);
2676 		CSR_WRITE_4(sc, BGE_VCPU_EXT_CTRL,
2677 		    ctrl & ~BGE_VCPU_EXT_CTRL_HALT_CPU);
2678 	}
2679 
2680 	DELAY(1000);
2681 
2682 	/* XXX: Broadcom Linux driver. */
2683 	if (sc->bge_flags & BGE_FLAG_PCIE) {
2684 		uint16_t devctl;
2685 
2686 		if (sc->bge_chipid == BGE_CHIPID_BCM5750_A0) {
2687 			uint32_t v;
2688 
2689 			DELAY(500000); /* wait for link training to complete */
2690 			v = pci_read_config(dev, 0xc4, 4);
2691 			pci_write_config(dev, 0xc4, v | (1<<15), 4);
2692 		}
2693 
2694 		devctl = pci_read_config(dev,
2695 		    sc->bge_pciecap + PCIER_DEVCTRL, 2);
2696 
2697 		/* Disable no snoop and disable relaxed ordering. */
2698 		devctl &= ~(PCIEM_DEVCTL_RELAX_ORDER | PCIEM_DEVCTL_NOSNOOP);
2699 
2700 		/* Old PCI-E chips only support 128 bytes Max PayLoad Size. */
2701 		if ((sc->bge_flags & BGE_FLAG_CPMU) == 0) {
2702 			devctl &= ~PCIEM_DEVCTL_MAX_PAYLOAD_MASK;
2703 			devctl |= PCIEM_DEVCTL_MAX_PAYLOAD_128;
2704 		}
2705 
2706 		pci_write_config(dev, sc->bge_pciecap + PCIER_DEVCTRL,
2707 		    devctl, 2);
2708 
2709 		/* Clear error status. */
2710 		pci_write_config(dev, sc->bge_pciecap + PCIER_DEVSTS,
2711 		    PCIEM_DEVSTS_CORR_ERR |
2712 		    PCIEM_DEVSTS_NFATAL_ERR |
2713 		    PCIEM_DEVSTS_FATAL_ERR |
2714 		    PCIEM_DEVSTS_UNSUPP_REQ, 2);
2715 	}
2716 
2717 	/* Reset some of the PCI state that got zapped by reset */
2718 	pci_write_config(dev, BGE_PCI_MISC_CTL,
2719 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2720 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW|
2721 	    sc->bge_pci_miscctl, 4);
2722 	pci_write_config(dev, BGE_PCI_CACHESZ, cachesize, 4);
2723 	pci_write_config(dev, BGE_PCI_CMD, command, 4);
2724 	write_op(sc, BGE_MISC_CFG, (65 << 1));
2725 
2726 	/*
2727 	 * Disable PCI-X relaxed ordering to ensure status block update
2728 	 * comes first then packet buffer DMA. Otherwise driver may
2729 	 * read stale status block.
2730 	 */
2731 	if (sc->bge_flags & BGE_FLAG_PCIX) {
2732 		uint16_t devctl;
2733 
2734 		devctl = pci_read_config(dev,
2735 		    sc->bge_pcixcap + PCIXR_COMMAND, 2);
2736 		devctl &= ~PCIXM_COMMAND_ERO;
2737 		if (sc->bge_asicrev == BGE_ASICREV_BCM5703) {
2738 			devctl &= ~PCIXM_COMMAND_MAX_READ;
2739 			devctl |= PCIXM_COMMAND_MAX_READ_2048;
2740 		} else if (sc->bge_asicrev == BGE_ASICREV_BCM5704) {
2741 			devctl &= ~(PCIXM_COMMAND_MAX_SPLITS |
2742 			    PCIXM_COMMAND_MAX_READ);
2743 			devctl |= PCIXM_COMMAND_MAX_READ_2048;
2744 		}
2745 		pci_write_config(dev, sc->bge_pcixcap + PCIXR_COMMAND,
2746 		    devctl, 2);
2747 	}
2748 
2749 	/*
2750 	 * Enable memory arbiter and re-enable MSI if necessary.
2751 	 */
2752 	if (BGE_IS_5714_FAMILY(sc)) {
2753 		uint32_t val;
2754 
2755 		if (sc->bge_irq_type == PCI_INTR_TYPE_MSI) {
2756 			/*
2757 			 * Resetting BCM5714 family will clear MSI
2758 			 * enable bit; restore it after resetting.
2759 			 */
2760 			PCI_SETBIT(sc->bge_dev, sc->bge_msicap + PCIR_MSI_CTRL,
2761 			    PCIM_MSICTRL_MSI_ENABLE, 2);
2762 			BGE_SETBIT(sc, BGE_MSI_MODE, BGE_MSIMODE_ENABLE);
2763 		}
2764 		val = CSR_READ_4(sc, BGE_MARB_MODE);
2765 		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE | val);
2766 	} else {
2767 		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
2768 	}
2769 
2770 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
2771 		for (i = 0; i < BGE_TIMEOUT; i++) {
2772 			val = CSR_READ_4(sc, BGE_VCPU_STATUS);
2773 			if (val & BGE_VCPU_STATUS_INIT_DONE)
2774 				break;
2775 			DELAY(100);
2776 		}
2777 		if (i == BGE_TIMEOUT) {
2778 			if_printf(&sc->arpcom.ac_if, "reset timed out\n");
2779 			return;
2780 		}
2781 	} else {
2782 		/*
2783 		 * Poll until we see the 1's complement of the magic number.
2784 		 * This indicates that the firmware initialization
2785 		 * is complete.
2786 		 */
2787 		for (i = 0; i < BGE_FIRMWARE_TIMEOUT; i++) {
2788 			val = bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM);
2789 			if (val == ~BGE_MAGIC_NUMBER)
2790 				break;
2791 			DELAY(10);
2792 		}
2793 		if (i == BGE_FIRMWARE_TIMEOUT) {
2794 			if_printf(&sc->arpcom.ac_if, "firmware handshake "
2795 				  "timed out, found 0x%08x\n", val);
2796 			return;
2797 		}
2798 	}
2799 
2800 	/*
2801 	 * XXX Wait for the value of the PCISTATE register to
2802 	 * return to its original pre-reset state. This is a
2803 	 * fairly good indicator of reset completion. If we don't
2804 	 * wait for the reset to fully complete, trying to read
2805 	 * from the device's non-PCI registers may yield garbage
2806 	 * results.
2807 	 */
2808 	for (i = 0; i < BGE_TIMEOUT; i++) {
2809 		if (pci_read_config(dev, BGE_PCI_PCISTATE, 4) == pcistate)
2810 			break;
2811 		DELAY(10);
2812 	}
2813 
2814 	/* Fix up byte swapping */
2815 	CSR_WRITE_4(sc, BGE_MODE_CTL, BGE_DMA_SWAP_OPTIONS |
2816 	    BGE_MODECTL_BYTESWAP_DATA);
2817 
2818 	CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
2819 
2820 	/*
2821 	 * The 5704 in TBI mode apparently needs some special
2822 	 * adjustment to insure the SERDES drive level is set
2823 	 * to 1.2V.
2824 	 */
2825 	if (sc->bge_asicrev == BGE_ASICREV_BCM5704 &&
2826 	    (sc->bge_flags & BGE_FLAG_TBI)) {
2827 		uint32_t serdescfg;
2828 
2829 		serdescfg = CSR_READ_4(sc, BGE_SERDES_CFG);
2830 		serdescfg = (serdescfg & ~0xFFF) | 0x880;
2831 		CSR_WRITE_4(sc, BGE_SERDES_CFG, serdescfg);
2832 	}
2833 
2834 	/* XXX: Broadcom Linux driver. */
2835 	if ((sc->bge_flags & BGE_FLAG_PCIE) &&
2836 	    sc->bge_chipid != BGE_CHIPID_BCM5750_A0 &&
2837 	    sc->bge_asicrev != BGE_ASICREV_BCM5785) {
2838 		uint32_t v;
2839 
2840 		/* Enable Data FIFO protection. */
2841 		v = CSR_READ_4(sc, 0x7c00);
2842 		CSR_WRITE_4(sc, 0x7c00, v | (1<<25));
2843 	}
2844 
2845 	DELAY(10000);
2846 }
2847 
2848 /*
2849  * Frame reception handling. This is called if there's a frame
2850  * on the receive return list.
2851  *
2852  * Note: we have to be able to handle two possibilities here:
2853  * 1) the frame is from the jumbo recieve ring
2854  * 2) the frame is from the standard receive ring
2855  */
2856 
2857 static void
2858 bge_rxeof(struct bge_softc *sc, uint16_t rx_prod)
2859 {
2860 	struct ifnet *ifp;
2861 	int stdcnt = 0, jumbocnt = 0;
2862 
2863 	ifp = &sc->arpcom.ac_if;
2864 
2865 	while (sc->bge_rx_saved_considx != rx_prod) {
2866 		struct bge_rx_bd	*cur_rx;
2867 		uint32_t		rxidx;
2868 		struct mbuf		*m = NULL;
2869 		uint16_t		vlan_tag = 0;
2870 		int			have_tag = 0;
2871 
2872 		cur_rx =
2873 	    &sc->bge_ldata.bge_rx_return_ring[sc->bge_rx_saved_considx];
2874 
2875 		rxidx = cur_rx->bge_idx;
2876 		BGE_INC(sc->bge_rx_saved_considx, sc->bge_return_ring_cnt);
2877 		logif(rx_pkt);
2878 
2879 		if (cur_rx->bge_flags & BGE_RXBDFLAG_VLAN_TAG) {
2880 			have_tag = 1;
2881 			vlan_tag = cur_rx->bge_vlan_tag;
2882 		}
2883 
2884 		if (cur_rx->bge_flags & BGE_RXBDFLAG_JUMBO_RING) {
2885 			BGE_INC(sc->bge_jumbo, BGE_JUMBO_RX_RING_CNT);
2886 			jumbocnt++;
2887 
2888 			if (rxidx != sc->bge_jumbo) {
2889 				ifp->if_ierrors++;
2890 				if_printf(ifp, "sw jumbo index(%d) "
2891 				    "and hw jumbo index(%d) mismatch, drop!\n",
2892 				    sc->bge_jumbo, rxidx);
2893 				bge_setup_rxdesc_jumbo(sc, rxidx);
2894 				continue;
2895 			}
2896 
2897 			m = sc->bge_cdata.bge_rx_jumbo_chain[rxidx].bge_mbuf;
2898 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
2899 				ifp->if_ierrors++;
2900 				bge_setup_rxdesc_jumbo(sc, sc->bge_jumbo);
2901 				continue;
2902 			}
2903 			if (bge_newbuf_jumbo(sc, sc->bge_jumbo, 0)) {
2904 				ifp->if_ierrors++;
2905 				bge_setup_rxdesc_jumbo(sc, sc->bge_jumbo);
2906 				continue;
2907 			}
2908 		} else {
2909 			BGE_INC(sc->bge_std, BGE_STD_RX_RING_CNT);
2910 			stdcnt++;
2911 
2912 			if (rxidx != sc->bge_std) {
2913 				ifp->if_ierrors++;
2914 				if_printf(ifp, "sw std index(%d) "
2915 				    "and hw std index(%d) mismatch, drop!\n",
2916 				    sc->bge_std, rxidx);
2917 				bge_setup_rxdesc_std(sc, rxidx);
2918 				continue;
2919 			}
2920 
2921 			m = sc->bge_cdata.bge_rx_std_chain[rxidx].bge_mbuf;
2922 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
2923 				ifp->if_ierrors++;
2924 				bge_setup_rxdesc_std(sc, sc->bge_std);
2925 				continue;
2926 			}
2927 			if (bge_newbuf_std(sc, sc->bge_std, 0)) {
2928 				ifp->if_ierrors++;
2929 				bge_setup_rxdesc_std(sc, sc->bge_std);
2930 				continue;
2931 			}
2932 		}
2933 
2934 		ifp->if_ipackets++;
2935 #if !defined(__i386__) && !defined(__x86_64__)
2936 		/*
2937 		 * The x86 allows unaligned accesses, but for other
2938 		 * platforms we must make sure the payload is aligned.
2939 		 */
2940 		if (sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) {
2941 			bcopy(m->m_data, m->m_data + ETHER_ALIGN,
2942 			    cur_rx->bge_len);
2943 			m->m_data += ETHER_ALIGN;
2944 		}
2945 #endif
2946 		m->m_pkthdr.len = m->m_len = cur_rx->bge_len - ETHER_CRC_LEN;
2947 		m->m_pkthdr.rcvif = ifp;
2948 
2949 		if (ifp->if_capenable & IFCAP_RXCSUM) {
2950 			if (cur_rx->bge_flags & BGE_RXBDFLAG_IP_CSUM) {
2951 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2952 				if ((cur_rx->bge_ip_csum ^ 0xffff) == 0)
2953 					m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2954 			}
2955 			if ((cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM) &&
2956 			    m->m_pkthdr.len >= BGE_MIN_FRAMELEN) {
2957 				m->m_pkthdr.csum_data =
2958 					cur_rx->bge_tcp_udp_csum;
2959 				m->m_pkthdr.csum_flags |=
2960 					CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2961 			}
2962 		}
2963 
2964 		/*
2965 		 * If we received a packet with a vlan tag, pass it
2966 		 * to vlan_input() instead of ether_input().
2967 		 */
2968 		if (have_tag) {
2969 			m->m_flags |= M_VLANTAG;
2970 			m->m_pkthdr.ether_vlantag = vlan_tag;
2971 			have_tag = vlan_tag = 0;
2972 		}
2973 		ifp->if_input(ifp, m);
2974 	}
2975 
2976 	bge_writembx(sc, BGE_MBX_RX_CONS0_LO, sc->bge_rx_saved_considx);
2977 	if (stdcnt)
2978 		bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bge_std);
2979 	if (jumbocnt)
2980 		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
2981 }
2982 
2983 static void
2984 bge_txeof(struct bge_softc *sc, uint16_t tx_cons)
2985 {
2986 	struct bge_tx_bd *cur_tx = NULL;
2987 	struct ifnet *ifp;
2988 
2989 	ifp = &sc->arpcom.ac_if;
2990 
2991 	/*
2992 	 * Go through our tx ring and free mbufs for those
2993 	 * frames that have been sent.
2994 	 */
2995 	while (sc->bge_tx_saved_considx != tx_cons) {
2996 		uint32_t idx = 0;
2997 
2998 		idx = sc->bge_tx_saved_considx;
2999 		cur_tx = &sc->bge_ldata.bge_tx_ring[idx];
3000 		if (cur_tx->bge_flags & BGE_TXBDFLAG_END)
3001 			ifp->if_opackets++;
3002 		if (sc->bge_cdata.bge_tx_chain[idx] != NULL) {
3003 			bus_dmamap_unload(sc->bge_cdata.bge_tx_mtag,
3004 			    sc->bge_cdata.bge_tx_dmamap[idx]);
3005 			m_freem(sc->bge_cdata.bge_tx_chain[idx]);
3006 			sc->bge_cdata.bge_tx_chain[idx] = NULL;
3007 		}
3008 		sc->bge_txcnt--;
3009 		BGE_INC(sc->bge_tx_saved_considx, BGE_TX_RING_CNT);
3010 		logif(tx_pkt);
3011 	}
3012 
3013 	if (cur_tx != NULL &&
3014 	    (BGE_TX_RING_CNT - sc->bge_txcnt) >=
3015 	    (BGE_NSEG_RSVD + BGE_NSEG_SPARE))
3016 		ifp->if_flags &= ~IFF_OACTIVE;
3017 
3018 	if (sc->bge_txcnt == 0)
3019 		ifp->if_timer = 0;
3020 
3021 	if (!ifq_is_empty(&ifp->if_snd))
3022 		if_devstart(ifp);
3023 }
3024 
3025 #ifdef DEVICE_POLLING
3026 
3027 static void
3028 bge_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
3029 {
3030 	struct bge_softc *sc = ifp->if_softc;
3031 	struct bge_status_block *sblk = sc->bge_ldata.bge_status_block;
3032 	uint16_t rx_prod, tx_cons;
3033 
3034 	switch(cmd) {
3035 	case POLL_REGISTER:
3036 		bge_disable_intr(sc);
3037 		break;
3038 	case POLL_DEREGISTER:
3039 		bge_enable_intr(sc);
3040 		break;
3041 	case POLL_AND_CHECK_STATUS:
3042 		/*
3043 		 * Process link state changes.
3044 		 */
3045 		bge_link_poll(sc);
3046 		/* Fall through */
3047 	case POLL_ONLY:
3048 		if (sc->bge_flags & BGE_FLAG_STATUS_TAG) {
3049 			sc->bge_status_tag = sblk->bge_status_tag;
3050 			/*
3051 			 * Use a load fence to ensure that status_tag
3052 			 * is saved  before rx_prod and tx_cons.
3053 			 */
3054 			cpu_lfence();
3055 		}
3056 		rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
3057 		tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
3058 		if (ifp->if_flags & IFF_RUNNING) {
3059 			rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
3060 			if (sc->bge_rx_saved_considx != rx_prod)
3061 				bge_rxeof(sc, rx_prod);
3062 
3063 			tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
3064 			if (sc->bge_tx_saved_considx != tx_cons)
3065 				bge_txeof(sc, tx_cons);
3066 		}
3067 		break;
3068 	}
3069 }
3070 
3071 #endif
3072 
3073 static void
3074 bge_intr_crippled(void *xsc)
3075 {
3076 	struct bge_softc *sc = xsc;
3077 	struct ifnet *ifp = &sc->arpcom.ac_if;
3078 
3079 	logif(intr);
3080 
3081  	/*
3082 	 * Ack the interrupt by writing something to BGE_MBX_IRQ0_LO.  Don't
3083 	 * disable interrupts by writing nonzero like we used to, since with
3084 	 * our current organization this just gives complications and
3085 	 * pessimizations for re-enabling interrupts.  We used to have races
3086 	 * instead of the necessary complications.  Disabling interrupts
3087 	 * would just reduce the chance of a status update while we are
3088 	 * running (by switching to the interrupt-mode coalescence
3089 	 * parameters), but this chance is already very low so it is more
3090 	 * efficient to get another interrupt than prevent it.
3091 	 *
3092 	 * We do the ack first to ensure another interrupt if there is a
3093 	 * status update after the ack.  We don't check for the status
3094 	 * changing later because it is more efficient to get another
3095 	 * interrupt than prevent it, not quite as above (not checking is
3096 	 * a smaller optimization than not toggling the interrupt enable,
3097 	 * since checking doesn't involve PCI accesses and toggling require
3098 	 * the status check).  So toggling would probably be a pessimization
3099 	 * even with MSI.  It would only be needed for using a task queue.
3100 	 */
3101 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 0);
3102 
3103 	/*
3104 	 * Process link state changes.
3105 	 */
3106 	bge_link_poll(sc);
3107 
3108 	if (ifp->if_flags & IFF_RUNNING) {
3109 		struct bge_status_block *sblk = sc->bge_ldata.bge_status_block;
3110 		uint16_t rx_prod, tx_cons;
3111 
3112 		rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
3113 		if (sc->bge_rx_saved_considx != rx_prod)
3114 			bge_rxeof(sc, rx_prod);
3115 
3116 		tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
3117 		if (sc->bge_tx_saved_considx != tx_cons)
3118 			bge_txeof(sc, tx_cons);
3119 	}
3120 
3121 	if (sc->bge_coal_chg)
3122 		bge_coal_change(sc);
3123 }
3124 
3125 static void
3126 bge_intr_legacy(void *xsc)
3127 {
3128 	struct bge_softc *sc = xsc;
3129 	struct bge_status_block *sblk = sc->bge_ldata.bge_status_block;
3130 
3131 	if (sc->bge_status_tag == sblk->bge_status_tag) {
3132 		uint32_t val;
3133 
3134 		val = pci_read_config(sc->bge_dev, BGE_PCI_PCISTATE, 4);
3135 		if (val & BGE_PCISTAT_INTR_NOTACT)
3136 			return;
3137 	}
3138 
3139 	/*
3140 	 * NOTE:
3141 	 * Interrupt will have to be disabled if tagged status
3142 	 * is used, else interrupt will always be asserted on
3143 	 * certain chips (at least on BCM5750 AX/BX).
3144 	 */
3145 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
3146 
3147 	bge_intr(sc);
3148 }
3149 
3150 static void
3151 bge_msi(void *xsc)
3152 {
3153 	struct bge_softc *sc = xsc;
3154 
3155 	/* Disable interrupt first */
3156 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
3157 	bge_intr(sc);
3158 }
3159 
3160 static void
3161 bge_msi_oneshot(void *xsc)
3162 {
3163 	bge_intr(xsc);
3164 }
3165 
3166 static void
3167 bge_intr(struct bge_softc *sc)
3168 {
3169 	struct ifnet *ifp = &sc->arpcom.ac_if;
3170 	struct bge_status_block *sblk = sc->bge_ldata.bge_status_block;
3171 	uint16_t rx_prod, tx_cons;
3172 	uint32_t status;
3173 
3174 	sc->bge_status_tag = sblk->bge_status_tag;
3175 	/*
3176 	 * Use a load fence to ensure that status_tag is saved
3177 	 * before rx_prod, tx_cons and status.
3178 	 */
3179 	cpu_lfence();
3180 
3181 	rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
3182 	tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
3183 	status = sblk->bge_status;
3184 
3185 	if ((status & BGE_STATFLAG_LINKSTATE_CHANGED) || sc->bge_link_evt)
3186 		bge_link_poll(sc);
3187 
3188 	if (ifp->if_flags & IFF_RUNNING) {
3189 		if (sc->bge_rx_saved_considx != rx_prod)
3190 			bge_rxeof(sc, rx_prod);
3191 
3192 		if (sc->bge_tx_saved_considx != tx_cons)
3193 			bge_txeof(sc, tx_cons);
3194 	}
3195 
3196 	bge_writembx(sc, BGE_MBX_IRQ0_LO, sc->bge_status_tag << 24);
3197 
3198 	if (sc->bge_coal_chg)
3199 		bge_coal_change(sc);
3200 }
3201 
3202 static void
3203 bge_tick(void *xsc)
3204 {
3205 	struct bge_softc *sc = xsc;
3206 	struct ifnet *ifp = &sc->arpcom.ac_if;
3207 
3208 	lwkt_serialize_enter(ifp->if_serializer);
3209 
3210 	if (BGE_IS_5705_PLUS(sc))
3211 		bge_stats_update_regs(sc);
3212 	else
3213 		bge_stats_update(sc);
3214 
3215 	if (sc->bge_flags & BGE_FLAG_TBI) {
3216 		/*
3217 		 * Since in TBI mode auto-polling can't be used we should poll
3218 		 * link status manually. Here we register pending link event
3219 		 * and trigger interrupt.
3220 		 */
3221 		sc->bge_link_evt++;
3222 		if (BGE_IS_CRIPPLED(sc))
3223 			BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
3224 		else
3225 			BGE_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
3226 	} else if (!sc->bge_link) {
3227 		mii_tick(device_get_softc(sc->bge_miibus));
3228 	}
3229 
3230 	callout_reset(&sc->bge_stat_timer, hz, bge_tick, sc);
3231 
3232 	lwkt_serialize_exit(ifp->if_serializer);
3233 }
3234 
3235 static void
3236 bge_stats_update_regs(struct bge_softc *sc)
3237 {
3238 	struct ifnet *ifp = &sc->arpcom.ac_if;
3239 	struct bge_mac_stats_regs stats;
3240 	uint32_t *s;
3241 	int i;
3242 
3243 	s = (uint32_t *)&stats;
3244 	for (i = 0; i < sizeof(struct bge_mac_stats_regs); i += 4) {
3245 		*s = CSR_READ_4(sc, BGE_RX_STATS + i);
3246 		s++;
3247 	}
3248 
3249 	ifp->if_collisions +=
3250 	   (stats.dot3StatsSingleCollisionFrames +
3251 	   stats.dot3StatsMultipleCollisionFrames +
3252 	   stats.dot3StatsExcessiveCollisions +
3253 	   stats.dot3StatsLateCollisions) -
3254 	   ifp->if_collisions;
3255 }
3256 
3257 static void
3258 bge_stats_update(struct bge_softc *sc)
3259 {
3260 	struct ifnet *ifp = &sc->arpcom.ac_if;
3261 	bus_size_t stats;
3262 
3263 	stats = BGE_MEMWIN_START + BGE_STATS_BLOCK;
3264 
3265 #define READ_STAT(sc, stats, stat)	\
3266 	CSR_READ_4(sc, stats + offsetof(struct bge_stats, stat))
3267 
3268 	ifp->if_collisions +=
3269 	   (READ_STAT(sc, stats,
3270 		txstats.dot3StatsSingleCollisionFrames.bge_addr_lo) +
3271 	    READ_STAT(sc, stats,
3272 		txstats.dot3StatsMultipleCollisionFrames.bge_addr_lo) +
3273 	    READ_STAT(sc, stats,
3274 		txstats.dot3StatsExcessiveCollisions.bge_addr_lo) +
3275 	    READ_STAT(sc, stats,
3276 		txstats.dot3StatsLateCollisions.bge_addr_lo)) -
3277 	   ifp->if_collisions;
3278 
3279 #undef READ_STAT
3280 
3281 #ifdef notdef
3282 	ifp->if_collisions +=
3283 	   (sc->bge_rdata->bge_info.bge_stats.dot3StatsSingleCollisionFrames +
3284 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsMultipleCollisionFrames +
3285 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsExcessiveCollisions +
3286 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsLateCollisions) -
3287 	   ifp->if_collisions;
3288 #endif
3289 }
3290 
3291 /*
3292  * Encapsulate an mbuf chain in the tx ring  by coupling the mbuf data
3293  * pointers to descriptors.
3294  */
3295 static int
3296 bge_encap(struct bge_softc *sc, struct mbuf **m_head0, uint32_t *txidx)
3297 {
3298 	struct bge_tx_bd *d = NULL;
3299 	uint16_t csum_flags = 0;
3300 	bus_dma_segment_t segs[BGE_NSEG_NEW];
3301 	bus_dmamap_t map;
3302 	int error, maxsegs, nsegs, idx, i;
3303 	struct mbuf *m_head = *m_head0, *m_new;
3304 
3305 	if (m_head->m_pkthdr.csum_flags) {
3306 		if (m_head->m_pkthdr.csum_flags & CSUM_IP)
3307 			csum_flags |= BGE_TXBDFLAG_IP_CSUM;
3308 		if (m_head->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP))
3309 			csum_flags |= BGE_TXBDFLAG_TCP_UDP_CSUM;
3310 		if (m_head->m_flags & M_LASTFRAG)
3311 			csum_flags |= BGE_TXBDFLAG_IP_FRAG_END;
3312 		else if (m_head->m_flags & M_FRAG)
3313 			csum_flags |= BGE_TXBDFLAG_IP_FRAG;
3314 	}
3315 
3316 	idx = *txidx;
3317 	map = sc->bge_cdata.bge_tx_dmamap[idx];
3318 
3319 	maxsegs = (BGE_TX_RING_CNT - sc->bge_txcnt) - BGE_NSEG_RSVD;
3320 	KASSERT(maxsegs >= BGE_NSEG_SPARE,
3321 		("not enough segments %d", maxsegs));
3322 
3323 	if (maxsegs > BGE_NSEG_NEW)
3324 		maxsegs = BGE_NSEG_NEW;
3325 
3326 	/*
3327 	 * Pad outbound frame to BGE_MIN_FRAMELEN for an unusual reason.
3328 	 * The bge hardware will pad out Tx runts to BGE_MIN_FRAMELEN,
3329 	 * but when such padded frames employ the bge IP/TCP checksum
3330 	 * offload, the hardware checksum assist gives incorrect results
3331 	 * (possibly from incorporating its own padding into the UDP/TCP
3332 	 * checksum; who knows).  If we pad such runts with zeros, the
3333 	 * onboard checksum comes out correct.
3334 	 */
3335 	if ((csum_flags & BGE_TXBDFLAG_TCP_UDP_CSUM) &&
3336 	    m_head->m_pkthdr.len < BGE_MIN_FRAMELEN) {
3337 		error = m_devpad(m_head, BGE_MIN_FRAMELEN);
3338 		if (error)
3339 			goto back;
3340 	}
3341 
3342 	if ((sc->bge_flags & BGE_FLAG_SHORTDMA) && m_head->m_next != NULL) {
3343 		m_new = bge_defrag_shortdma(m_head);
3344 		if (m_new == NULL) {
3345 			error = ENOBUFS;
3346 			goto back;
3347 		}
3348 		*m_head0 = m_head = m_new;
3349 	}
3350 	if (sc->bge_force_defrag && (sc->bge_flags & BGE_FLAG_PCIE) &&
3351 	    m_head->m_next != NULL) {
3352 		/*
3353 		 * Forcefully defragment mbuf chain to overcome hardware
3354 		 * limitation which only support a single outstanding
3355 		 * DMA read operation.  If it fails, keep moving on using
3356 		 * the original mbuf chain.
3357 		 */
3358 		m_new = m_defrag(m_head, MB_DONTWAIT);
3359 		if (m_new != NULL)
3360 			*m_head0 = m_head = m_new;
3361 	}
3362 
3363 	error = bus_dmamap_load_mbuf_defrag(sc->bge_cdata.bge_tx_mtag, map,
3364 			m_head0, segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
3365 	if (error)
3366 		goto back;
3367 
3368 	m_head = *m_head0;
3369 	bus_dmamap_sync(sc->bge_cdata.bge_tx_mtag, map, BUS_DMASYNC_PREWRITE);
3370 
3371 	for (i = 0; ; i++) {
3372 		d = &sc->bge_ldata.bge_tx_ring[idx];
3373 
3374 		d->bge_addr.bge_addr_lo = BGE_ADDR_LO(segs[i].ds_addr);
3375 		d->bge_addr.bge_addr_hi = BGE_ADDR_HI(segs[i].ds_addr);
3376 		d->bge_len = segs[i].ds_len;
3377 		d->bge_flags = csum_flags;
3378 
3379 		if (i == nsegs - 1)
3380 			break;
3381 		BGE_INC(idx, BGE_TX_RING_CNT);
3382 	}
3383 	/* Mark the last segment as end of packet... */
3384 	d->bge_flags |= BGE_TXBDFLAG_END;
3385 
3386 	/* Set vlan tag to the first segment of the packet. */
3387 	d = &sc->bge_ldata.bge_tx_ring[*txidx];
3388 	if (m_head->m_flags & M_VLANTAG) {
3389 		d->bge_flags |= BGE_TXBDFLAG_VLAN_TAG;
3390 		d->bge_vlan_tag = m_head->m_pkthdr.ether_vlantag;
3391 	} else {
3392 		d->bge_vlan_tag = 0;
3393 	}
3394 
3395 	/*
3396 	 * Insure that the map for this transmission is placed at
3397 	 * the array index of the last descriptor in this chain.
3398 	 */
3399 	sc->bge_cdata.bge_tx_dmamap[*txidx] = sc->bge_cdata.bge_tx_dmamap[idx];
3400 	sc->bge_cdata.bge_tx_dmamap[idx] = map;
3401 	sc->bge_cdata.bge_tx_chain[idx] = m_head;
3402 	sc->bge_txcnt += nsegs;
3403 
3404 	BGE_INC(idx, BGE_TX_RING_CNT);
3405 	*txidx = idx;
3406 back:
3407 	if (error) {
3408 		m_freem(*m_head0);
3409 		*m_head0 = NULL;
3410 	}
3411 	return error;
3412 }
3413 
3414 /*
3415  * Main transmit routine. To avoid having to do mbuf copies, we put pointers
3416  * to the mbuf data regions directly in the transmit descriptors.
3417  */
3418 static void
3419 bge_start(struct ifnet *ifp)
3420 {
3421 	struct bge_softc *sc = ifp->if_softc;
3422 	struct mbuf *m_head = NULL;
3423 	uint32_t prodidx;
3424 	int need_trans;
3425 
3426 	if ((ifp->if_flags & (IFF_RUNNING | IFF_OACTIVE)) != IFF_RUNNING)
3427 		return;
3428 
3429 	prodidx = sc->bge_tx_prodidx;
3430 
3431 	need_trans = 0;
3432 	while (sc->bge_cdata.bge_tx_chain[prodidx] == NULL) {
3433 		m_head = ifq_dequeue(&ifp->if_snd, NULL);
3434 		if (m_head == NULL)
3435 			break;
3436 
3437 		/*
3438 		 * XXX
3439 		 * The code inside the if() block is never reached since we
3440 		 * must mark CSUM_IP_FRAGS in our if_hwassist to start getting
3441 		 * requests to checksum TCP/UDP in a fragmented packet.
3442 		 *
3443 		 * XXX
3444 		 * safety overkill.  If this is a fragmented packet chain
3445 		 * with delayed TCP/UDP checksums, then only encapsulate
3446 		 * it if we have enough descriptors to handle the entire
3447 		 * chain at once.
3448 		 * (paranoia -- may not actually be needed)
3449 		 */
3450 		if ((m_head->m_flags & M_FIRSTFRAG) &&
3451 		    (m_head->m_pkthdr.csum_flags & CSUM_DELAY_DATA)) {
3452 			if ((BGE_TX_RING_CNT - sc->bge_txcnt) <
3453 			    m_head->m_pkthdr.csum_data + BGE_NSEG_RSVD) {
3454 				ifp->if_flags |= IFF_OACTIVE;
3455 				ifq_prepend(&ifp->if_snd, m_head);
3456 				break;
3457 			}
3458 		}
3459 
3460 		/*
3461 		 * Sanity check: avoid coming within BGE_NSEG_RSVD
3462 		 * descriptors of the end of the ring.  Also make
3463 		 * sure there are BGE_NSEG_SPARE descriptors for
3464 		 * jumbo buffers' defragmentation.
3465 		 */
3466 		if ((BGE_TX_RING_CNT - sc->bge_txcnt) <
3467 		    (BGE_NSEG_RSVD + BGE_NSEG_SPARE)) {
3468 			ifp->if_flags |= IFF_OACTIVE;
3469 			ifq_prepend(&ifp->if_snd, m_head);
3470 			break;
3471 		}
3472 
3473 		/*
3474 		 * Pack the data into the transmit ring. If we
3475 		 * don't have room, set the OACTIVE flag and wait
3476 		 * for the NIC to drain the ring.
3477 		 */
3478 		if (bge_encap(sc, &m_head, &prodidx)) {
3479 			ifp->if_flags |= IFF_OACTIVE;
3480 			ifp->if_oerrors++;
3481 			break;
3482 		}
3483 		need_trans = 1;
3484 
3485 		ETHER_BPF_MTAP(ifp, m_head);
3486 	}
3487 
3488 	if (!need_trans)
3489 		return;
3490 
3491 	/* Transmit */
3492 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
3493 	/* 5700 b2 errata */
3494 	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
3495 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
3496 
3497 	sc->bge_tx_prodidx = prodidx;
3498 
3499 	/*
3500 	 * Set a timeout in case the chip goes out to lunch.
3501 	 */
3502 	ifp->if_timer = 5;
3503 }
3504 
3505 static void
3506 bge_init(void *xsc)
3507 {
3508 	struct bge_softc *sc = xsc;
3509 	struct ifnet *ifp = &sc->arpcom.ac_if;
3510 	uint16_t *m;
3511 	uint32_t mode;
3512 
3513 	ASSERT_SERIALIZED(ifp->if_serializer);
3514 
3515 	/* Cancel pending I/O and flush buffers. */
3516 	bge_stop(sc);
3517 	bge_reset(sc);
3518 	bge_chipinit(sc);
3519 
3520 	/*
3521 	 * Init the various state machines, ring
3522 	 * control blocks and firmware.
3523 	 */
3524 	if (bge_blockinit(sc)) {
3525 		if_printf(ifp, "initialization failure\n");
3526 		bge_stop(sc);
3527 		return;
3528 	}
3529 
3530 	/* Specify MTU. */
3531 	CSR_WRITE_4(sc, BGE_RX_MTU, ifp->if_mtu +
3532 	    ETHER_HDR_LEN + ETHER_CRC_LEN + EVL_ENCAPLEN);
3533 
3534 	/* Load our MAC address. */
3535 	m = (uint16_t *)&sc->arpcom.ac_enaddr[0];
3536 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_LO, htons(m[0]));
3537 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_HI, (htons(m[1]) << 16) | htons(m[2]));
3538 
3539 	/* Enable or disable promiscuous mode as needed. */
3540 	bge_setpromisc(sc);
3541 
3542 	/* Program multicast filter. */
3543 	bge_setmulti(sc);
3544 
3545 	/* Init RX ring. */
3546 	if (bge_init_rx_ring_std(sc)) {
3547 		if_printf(ifp, "RX ring initialization failed\n");
3548 		bge_stop(sc);
3549 		return;
3550 	}
3551 
3552 	/*
3553 	 * Workaround for a bug in 5705 ASIC rev A0. Poll the NIC's
3554 	 * memory to insure that the chip has in fact read the first
3555 	 * entry of the ring.
3556 	 */
3557 	if (sc->bge_chipid == BGE_CHIPID_BCM5705_A0) {
3558 		uint32_t		v, i;
3559 		for (i = 0; i < 10; i++) {
3560 			DELAY(20);
3561 			v = bge_readmem_ind(sc, BGE_STD_RX_RINGS + 8);
3562 			if (v == (MCLBYTES - ETHER_ALIGN))
3563 				break;
3564 		}
3565 		if (i == 10)
3566 			if_printf(ifp, "5705 A0 chip failed to load RX ring\n");
3567 	}
3568 
3569 	/* Init jumbo RX ring. */
3570 	if (ifp->if_mtu > (ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN)) {
3571 		if (bge_init_rx_ring_jumbo(sc)) {
3572 			if_printf(ifp, "Jumbo RX ring initialization failed\n");
3573 			bge_stop(sc);
3574 			return;
3575 		}
3576 	}
3577 
3578 	/* Init our RX return ring index */
3579 	sc->bge_rx_saved_considx = 0;
3580 
3581 	/* Init TX ring. */
3582 	bge_init_tx_ring(sc);
3583 
3584 	/* Enable TX MAC state machine lockup fix. */
3585 	mode = CSR_READ_4(sc, BGE_TX_MODE);
3586 	if (BGE_IS_5755_PLUS(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5906)
3587 		mode |= BGE_TXMODE_MBUF_LOCKUP_FIX;
3588 	/* Turn on transmitter */
3589 	CSR_WRITE_4(sc, BGE_TX_MODE, mode | BGE_TXMODE_ENABLE);
3590 
3591 	/* Turn on receiver */
3592 	BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
3593 
3594 	/*
3595 	 * Set the number of good frames to receive after RX MBUF
3596 	 * Low Watermark has been reached.  After the RX MAC receives
3597 	 * this number of frames, it will drop subsequent incoming
3598 	 * frames until the MBUF High Watermark is reached.
3599 	 */
3600 	CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 2);
3601 
3602 	if (sc->bge_irq_type == PCI_INTR_TYPE_MSI) {
3603 		if (bootverbose) {
3604 			if_printf(ifp, "MSI_MODE: %#x\n",
3605 			    CSR_READ_4(sc, BGE_MSI_MODE));
3606 		}
3607 
3608 		/*
3609 		 * XXX
3610 		 * Linux driver turns it on for all chips supporting MSI?!
3611 		 */
3612 		if (sc->bge_flags & BGE_FLAG_ONESHOT_MSI) {
3613 			/*
3614 			 * XXX
3615 			 * According to 5722-PG101-R,
3616 			 * BGE_PCIE_TRANSACT_ONESHOT_MSI applies only to
3617 			 * BCM5906.
3618 			 */
3619 			BGE_SETBIT(sc, BGE_PCIE_TRANSACT,
3620 			    BGE_PCIE_TRANSACT_ONESHOT_MSI);
3621 		}
3622 	}
3623 
3624 	/* Tell firmware we're alive. */
3625 	BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3626 
3627 	/* Enable host interrupts if polling(4) is not enabled. */
3628 	PCI_SETBIT(sc->bge_dev, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_CLEAR_INTA, 4);
3629 #ifdef DEVICE_POLLING
3630 	if (ifp->if_flags & IFF_POLLING)
3631 		bge_disable_intr(sc);
3632 	else
3633 #endif
3634 	bge_enable_intr(sc);
3635 
3636 	bge_ifmedia_upd(ifp);
3637 
3638 	ifp->if_flags |= IFF_RUNNING;
3639 	ifp->if_flags &= ~IFF_OACTIVE;
3640 
3641 	callout_reset(&sc->bge_stat_timer, hz, bge_tick, sc);
3642 }
3643 
3644 /*
3645  * Set media options.
3646  */
3647 static int
3648 bge_ifmedia_upd(struct ifnet *ifp)
3649 {
3650 	struct bge_softc *sc = ifp->if_softc;
3651 
3652 	/* If this is a 1000baseX NIC, enable the TBI port. */
3653 	if (sc->bge_flags & BGE_FLAG_TBI) {
3654 		struct ifmedia *ifm = &sc->bge_ifmedia;
3655 
3656 		if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
3657 			return(EINVAL);
3658 
3659 		switch(IFM_SUBTYPE(ifm->ifm_media)) {
3660 		case IFM_AUTO:
3661 			/*
3662 			 * The BCM5704 ASIC appears to have a special
3663 			 * mechanism for programming the autoneg
3664 			 * advertisement registers in TBI mode.
3665 			 */
3666 			if (!bge_fake_autoneg &&
3667 			    sc->bge_asicrev == BGE_ASICREV_BCM5704) {
3668 				uint32_t sgdig;
3669 
3670 				CSR_WRITE_4(sc, BGE_TX_TBI_AUTONEG, 0);
3671 				sgdig = CSR_READ_4(sc, BGE_SGDIG_CFG);
3672 				sgdig |= BGE_SGDIGCFG_AUTO |
3673 					 BGE_SGDIGCFG_PAUSE_CAP |
3674 					 BGE_SGDIGCFG_ASYM_PAUSE;
3675 				CSR_WRITE_4(sc, BGE_SGDIG_CFG,
3676 					    sgdig | BGE_SGDIGCFG_SEND);
3677 				DELAY(5);
3678 				CSR_WRITE_4(sc, BGE_SGDIG_CFG, sgdig);
3679 			}
3680 			break;
3681 		case IFM_1000_SX:
3682 			if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX) {
3683 				BGE_CLRBIT(sc, BGE_MAC_MODE,
3684 				    BGE_MACMODE_HALF_DUPLEX);
3685 			} else {
3686 				BGE_SETBIT(sc, BGE_MAC_MODE,
3687 				    BGE_MACMODE_HALF_DUPLEX);
3688 			}
3689 			break;
3690 		default:
3691 			return(EINVAL);
3692 		}
3693 	} else {
3694 		struct mii_data *mii = device_get_softc(sc->bge_miibus);
3695 
3696 		sc->bge_link_evt++;
3697 		sc->bge_link = 0;
3698 		if (mii->mii_instance) {
3699 			struct mii_softc *miisc;
3700 
3701 			LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
3702 				mii_phy_reset(miisc);
3703 		}
3704 		mii_mediachg(mii);
3705 
3706 		/*
3707 		 * Force an interrupt so that we will call bge_link_upd
3708 		 * if needed and clear any pending link state attention.
3709 		 * Without this we are not getting any further interrupts
3710 		 * for link state changes and thus will not UP the link and
3711 		 * not be able to send in bge_start.  The only way to get
3712 		 * things working was to receive a packet and get an RX
3713 		 * intr.
3714 		 *
3715 		 * bge_tick should help for fiber cards and we might not
3716 		 * need to do this here if BGE_FLAG_TBI is set but as
3717 		 * we poll for fiber anyway it should not harm.
3718 		 */
3719 		if (BGE_IS_CRIPPLED(sc))
3720 			BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
3721 		else
3722 			BGE_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
3723 	}
3724 	return(0);
3725 }
3726 
3727 /*
3728  * Report current media status.
3729  */
3730 static void
3731 bge_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3732 {
3733 	struct bge_softc *sc = ifp->if_softc;
3734 
3735 	if (sc->bge_flags & BGE_FLAG_TBI) {
3736 		ifmr->ifm_status = IFM_AVALID;
3737 		ifmr->ifm_active = IFM_ETHER;
3738 		if (CSR_READ_4(sc, BGE_MAC_STS) &
3739 		    BGE_MACSTAT_TBI_PCS_SYNCHED) {
3740 			ifmr->ifm_status |= IFM_ACTIVE;
3741 		} else {
3742 			ifmr->ifm_active |= IFM_NONE;
3743 			return;
3744 		}
3745 
3746 		ifmr->ifm_active |= IFM_1000_SX;
3747 		if (CSR_READ_4(sc, BGE_MAC_MODE) & BGE_MACMODE_HALF_DUPLEX)
3748 			ifmr->ifm_active |= IFM_HDX;
3749 		else
3750 			ifmr->ifm_active |= IFM_FDX;
3751 	} else {
3752 		struct mii_data *mii = device_get_softc(sc->bge_miibus);
3753 
3754 		mii_pollstat(mii);
3755 		ifmr->ifm_active = mii->mii_media_active;
3756 		ifmr->ifm_status = mii->mii_media_status;
3757 	}
3758 }
3759 
3760 static int
3761 bge_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
3762 {
3763 	struct bge_softc *sc = ifp->if_softc;
3764 	struct ifreq *ifr = (struct ifreq *)data;
3765 	int mask, error = 0;
3766 
3767 	ASSERT_SERIALIZED(ifp->if_serializer);
3768 
3769 	switch (command) {
3770 	case SIOCSIFMTU:
3771 		if ((!BGE_IS_JUMBO_CAPABLE(sc) && ifr->ifr_mtu > ETHERMTU) ||
3772 		    (BGE_IS_JUMBO_CAPABLE(sc) &&
3773 		     ifr->ifr_mtu > BGE_JUMBO_MTU)) {
3774 			error = EINVAL;
3775 		} else if (ifp->if_mtu != ifr->ifr_mtu) {
3776 			ifp->if_mtu = ifr->ifr_mtu;
3777 			if (ifp->if_flags & IFF_RUNNING)
3778 				bge_init(sc);
3779 		}
3780 		break;
3781 	case SIOCSIFFLAGS:
3782 		if (ifp->if_flags & IFF_UP) {
3783 			if (ifp->if_flags & IFF_RUNNING) {
3784 				mask = ifp->if_flags ^ sc->bge_if_flags;
3785 
3786 				/*
3787 				 * If only the state of the PROMISC flag
3788 				 * changed, then just use the 'set promisc
3789 				 * mode' command instead of reinitializing
3790 				 * the entire NIC. Doing a full re-init
3791 				 * means reloading the firmware and waiting
3792 				 * for it to start up, which may take a
3793 				 * second or two.  Similarly for ALLMULTI.
3794 				 */
3795 				if (mask & IFF_PROMISC)
3796 					bge_setpromisc(sc);
3797 				if (mask & IFF_ALLMULTI)
3798 					bge_setmulti(sc);
3799 			} else {
3800 				bge_init(sc);
3801 			}
3802 		} else if (ifp->if_flags & IFF_RUNNING) {
3803 			bge_stop(sc);
3804 		}
3805 		sc->bge_if_flags = ifp->if_flags;
3806 		break;
3807 	case SIOCADDMULTI:
3808 	case SIOCDELMULTI:
3809 		if (ifp->if_flags & IFF_RUNNING)
3810 			bge_setmulti(sc);
3811 		break;
3812 	case SIOCSIFMEDIA:
3813 	case SIOCGIFMEDIA:
3814 		if (sc->bge_flags & BGE_FLAG_TBI) {
3815 			error = ifmedia_ioctl(ifp, ifr,
3816 			    &sc->bge_ifmedia, command);
3817 		} else {
3818 			struct mii_data *mii;
3819 
3820 			mii = device_get_softc(sc->bge_miibus);
3821 			error = ifmedia_ioctl(ifp, ifr,
3822 					      &mii->mii_media, command);
3823 		}
3824 		break;
3825         case SIOCSIFCAP:
3826 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3827 		if (mask & IFCAP_HWCSUM) {
3828 			ifp->if_capenable ^= (mask & IFCAP_HWCSUM);
3829 			if (IFCAP_HWCSUM & ifp->if_capenable)
3830 				ifp->if_hwassist = BGE_CSUM_FEATURES;
3831 			else
3832 				ifp->if_hwassist = 0;
3833 		}
3834 		break;
3835 	default:
3836 		error = ether_ioctl(ifp, command, data);
3837 		break;
3838 	}
3839 	return error;
3840 }
3841 
3842 static void
3843 bge_watchdog(struct ifnet *ifp)
3844 {
3845 	struct bge_softc *sc = ifp->if_softc;
3846 
3847 	if_printf(ifp, "watchdog timeout -- resetting\n");
3848 
3849 	bge_init(sc);
3850 
3851 	ifp->if_oerrors++;
3852 
3853 	if (!ifq_is_empty(&ifp->if_snd))
3854 		if_devstart(ifp);
3855 }
3856 
3857 /*
3858  * Stop the adapter and free any mbufs allocated to the
3859  * RX and TX lists.
3860  */
3861 static void
3862 bge_stop(struct bge_softc *sc)
3863 {
3864 	struct ifnet *ifp = &sc->arpcom.ac_if;
3865 
3866 	ASSERT_SERIALIZED(ifp->if_serializer);
3867 
3868 	callout_stop(&sc->bge_stat_timer);
3869 
3870 	/*
3871 	 * Disable all of the receiver blocks
3872 	 */
3873 	bge_stop_block(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
3874 	bge_stop_block(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
3875 	bge_stop_block(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
3876 	if (BGE_IS_5700_FAMILY(sc))
3877 		bge_stop_block(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
3878 	bge_stop_block(sc, BGE_RDBDI_MODE, BGE_RBDIMODE_ENABLE);
3879 	bge_stop_block(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
3880 	bge_stop_block(sc, BGE_RBDC_MODE, BGE_RBDCMODE_ENABLE);
3881 
3882 	/*
3883 	 * Disable all of the transmit blocks
3884 	 */
3885 	bge_stop_block(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
3886 	bge_stop_block(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
3887 	bge_stop_block(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
3888 	bge_stop_block(sc, BGE_RDMA_MODE, BGE_RDMAMODE_ENABLE);
3889 	bge_stop_block(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE);
3890 	if (BGE_IS_5700_FAMILY(sc))
3891 		bge_stop_block(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
3892 	bge_stop_block(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
3893 
3894 	/*
3895 	 * Shut down all of the memory managers and related
3896 	 * state machines.
3897 	 */
3898 	bge_stop_block(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE);
3899 	bge_stop_block(sc, BGE_WDMA_MODE, BGE_WDMAMODE_ENABLE);
3900 	if (BGE_IS_5700_FAMILY(sc))
3901 		bge_stop_block(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
3902 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
3903 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
3904 	if (!BGE_IS_5705_PLUS(sc)) {
3905 		BGE_CLRBIT(sc, BGE_BMAN_MODE, BGE_BMANMODE_ENABLE);
3906 		BGE_CLRBIT(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
3907 	}
3908 
3909 	/* Disable host interrupts. */
3910 	bge_disable_intr(sc);
3911 
3912 	/*
3913 	 * Tell firmware we're shutting down.
3914 	 */
3915 	BGE_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3916 
3917 	/* Free the RX lists. */
3918 	bge_free_rx_ring_std(sc);
3919 
3920 	/* Free jumbo RX list. */
3921 	if (BGE_IS_JUMBO_CAPABLE(sc))
3922 		bge_free_rx_ring_jumbo(sc);
3923 
3924 	/* Free TX buffers. */
3925 	bge_free_tx_ring(sc);
3926 
3927 	sc->bge_status_tag = 0;
3928 	sc->bge_link = 0;
3929 	sc->bge_coal_chg = 0;
3930 
3931 	sc->bge_tx_saved_considx = BGE_TXCONS_UNSET;
3932 
3933 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
3934 	ifp->if_timer = 0;
3935 }
3936 
3937 /*
3938  * Stop all chip I/O so that the kernel's probe routines don't
3939  * get confused by errant DMAs when rebooting.
3940  */
3941 static void
3942 bge_shutdown(device_t dev)
3943 {
3944 	struct bge_softc *sc = device_get_softc(dev);
3945 	struct ifnet *ifp = &sc->arpcom.ac_if;
3946 
3947 	lwkt_serialize_enter(ifp->if_serializer);
3948 	bge_stop(sc);
3949 	bge_reset(sc);
3950 	lwkt_serialize_exit(ifp->if_serializer);
3951 }
3952 
3953 static int
3954 bge_suspend(device_t dev)
3955 {
3956 	struct bge_softc *sc = device_get_softc(dev);
3957 	struct ifnet *ifp = &sc->arpcom.ac_if;
3958 
3959 	lwkt_serialize_enter(ifp->if_serializer);
3960 	bge_stop(sc);
3961 	lwkt_serialize_exit(ifp->if_serializer);
3962 
3963 	return 0;
3964 }
3965 
3966 static int
3967 bge_resume(device_t dev)
3968 {
3969 	struct bge_softc *sc = device_get_softc(dev);
3970 	struct ifnet *ifp = &sc->arpcom.ac_if;
3971 
3972 	lwkt_serialize_enter(ifp->if_serializer);
3973 
3974 	if (ifp->if_flags & IFF_UP) {
3975 		bge_init(sc);
3976 
3977 		if (!ifq_is_empty(&ifp->if_snd))
3978 			if_devstart(ifp);
3979 	}
3980 
3981 	lwkt_serialize_exit(ifp->if_serializer);
3982 
3983 	return 0;
3984 }
3985 
3986 static void
3987 bge_setpromisc(struct bge_softc *sc)
3988 {
3989 	struct ifnet *ifp = &sc->arpcom.ac_if;
3990 
3991 	if (ifp->if_flags & IFF_PROMISC)
3992 		BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
3993 	else
3994 		BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
3995 }
3996 
3997 static void
3998 bge_dma_free(struct bge_softc *sc)
3999 {
4000 	int i;
4001 
4002 	/* Destroy RX mbuf DMA stuffs. */
4003 	if (sc->bge_cdata.bge_rx_mtag != NULL) {
4004 		for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
4005 			bus_dmamap_destroy(sc->bge_cdata.bge_rx_mtag,
4006 			    sc->bge_cdata.bge_rx_std_dmamap[i]);
4007 		}
4008 		bus_dmamap_destroy(sc->bge_cdata.bge_rx_mtag,
4009 				   sc->bge_cdata.bge_rx_tmpmap);
4010 		bus_dma_tag_destroy(sc->bge_cdata.bge_rx_mtag);
4011 	}
4012 
4013 	/* Destroy TX mbuf DMA stuffs. */
4014 	if (sc->bge_cdata.bge_tx_mtag != NULL) {
4015 		for (i = 0; i < BGE_TX_RING_CNT; i++) {
4016 			bus_dmamap_destroy(sc->bge_cdata.bge_tx_mtag,
4017 			    sc->bge_cdata.bge_tx_dmamap[i]);
4018 		}
4019 		bus_dma_tag_destroy(sc->bge_cdata.bge_tx_mtag);
4020 	}
4021 
4022 	/* Destroy standard RX ring */
4023 	bge_dma_block_free(sc->bge_cdata.bge_rx_std_ring_tag,
4024 			   sc->bge_cdata.bge_rx_std_ring_map,
4025 			   sc->bge_ldata.bge_rx_std_ring);
4026 
4027 	if (BGE_IS_JUMBO_CAPABLE(sc))
4028 		bge_free_jumbo_mem(sc);
4029 
4030 	/* Destroy RX return ring */
4031 	bge_dma_block_free(sc->bge_cdata.bge_rx_return_ring_tag,
4032 			   sc->bge_cdata.bge_rx_return_ring_map,
4033 			   sc->bge_ldata.bge_rx_return_ring);
4034 
4035 	/* Destroy TX ring */
4036 	bge_dma_block_free(sc->bge_cdata.bge_tx_ring_tag,
4037 			   sc->bge_cdata.bge_tx_ring_map,
4038 			   sc->bge_ldata.bge_tx_ring);
4039 
4040 	/* Destroy status block */
4041 	bge_dma_block_free(sc->bge_cdata.bge_status_tag,
4042 			   sc->bge_cdata.bge_status_map,
4043 			   sc->bge_ldata.bge_status_block);
4044 
4045 	/* Destroy statistics block */
4046 	bge_dma_block_free(sc->bge_cdata.bge_stats_tag,
4047 			   sc->bge_cdata.bge_stats_map,
4048 			   sc->bge_ldata.bge_stats);
4049 
4050 	/* Destroy the parent tag */
4051 	if (sc->bge_cdata.bge_parent_tag != NULL)
4052 		bus_dma_tag_destroy(sc->bge_cdata.bge_parent_tag);
4053 }
4054 
4055 static int
4056 bge_dma_alloc(struct bge_softc *sc)
4057 {
4058 	struct ifnet *ifp = &sc->arpcom.ac_if;
4059 	int i, error;
4060 	bus_addr_t lowaddr;
4061 
4062 	lowaddr = BUS_SPACE_MAXADDR;
4063 	if (sc->bge_flags & BGE_FLAG_MAXADDR_40BIT)
4064 		lowaddr = BGE_DMA_MAXADDR_40BIT;
4065 
4066 	/*
4067 	 * Allocate the parent bus DMA tag appropriate for PCI.
4068 	 *
4069 	 * All of the NetExtreme/NetLink controllers have 4GB boundary
4070 	 * DMA bug.
4071 	 * Whenever an address crosses a multiple of the 4GB boundary
4072 	 * (including 4GB, 8Gb, 12Gb, etc.) and makes the transition
4073 	 * from 0xX_FFFF_FFFF to 0x(X+1)_0000_0000 an internal DMA
4074 	 * state machine will lockup and cause the device to hang.
4075 	 */
4076 	error = bus_dma_tag_create(NULL, 1, BGE_DMA_BOUNDARY_4G,
4077 				   lowaddr, BUS_SPACE_MAXADDR,
4078 				   NULL, NULL,
4079 				   BUS_SPACE_MAXSIZE_32BIT, 0,
4080 				   BUS_SPACE_MAXSIZE_32BIT,
4081 				   0, &sc->bge_cdata.bge_parent_tag);
4082 	if (error) {
4083 		if_printf(ifp, "could not allocate parent dma tag\n");
4084 		return error;
4085 	}
4086 
4087 	/*
4088 	 * Create DMA tag and maps for RX mbufs.
4089 	 */
4090 	error = bus_dma_tag_create(sc->bge_cdata.bge_parent_tag, 1, 0,
4091 				   BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
4092 				   NULL, NULL, MCLBYTES, 1, MCLBYTES,
4093 				   BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK,
4094 				   &sc->bge_cdata.bge_rx_mtag);
4095 	if (error) {
4096 		if_printf(ifp, "could not allocate RX mbuf dma tag\n");
4097 		return error;
4098 	}
4099 
4100 	error = bus_dmamap_create(sc->bge_cdata.bge_rx_mtag,
4101 				  BUS_DMA_WAITOK, &sc->bge_cdata.bge_rx_tmpmap);
4102 	if (error) {
4103 		bus_dma_tag_destroy(sc->bge_cdata.bge_rx_mtag);
4104 		sc->bge_cdata.bge_rx_mtag = NULL;
4105 		return error;
4106 	}
4107 
4108 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
4109 		error = bus_dmamap_create(sc->bge_cdata.bge_rx_mtag,
4110 					  BUS_DMA_WAITOK,
4111 					  &sc->bge_cdata.bge_rx_std_dmamap[i]);
4112 		if (error) {
4113 			int j;
4114 
4115 			for (j = 0; j < i; ++j) {
4116 				bus_dmamap_destroy(sc->bge_cdata.bge_rx_mtag,
4117 					sc->bge_cdata.bge_rx_std_dmamap[j]);
4118 			}
4119 			bus_dma_tag_destroy(sc->bge_cdata.bge_rx_mtag);
4120 			sc->bge_cdata.bge_rx_mtag = NULL;
4121 
4122 			if_printf(ifp, "could not create DMA map for RX\n");
4123 			return error;
4124 		}
4125 	}
4126 
4127 	/*
4128 	 * Create DMA tag and maps for TX mbufs.
4129 	 */
4130 	error = bus_dma_tag_create(sc->bge_cdata.bge_parent_tag, 1, 0,
4131 				   BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
4132 				   NULL, NULL,
4133 				   BGE_JUMBO_FRAMELEN, BGE_NSEG_NEW, MCLBYTES,
4134 				   BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK |
4135 				   BUS_DMA_ONEBPAGE,
4136 				   &sc->bge_cdata.bge_tx_mtag);
4137 	if (error) {
4138 		if_printf(ifp, "could not allocate TX mbuf dma tag\n");
4139 		return error;
4140 	}
4141 
4142 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
4143 		error = bus_dmamap_create(sc->bge_cdata.bge_tx_mtag,
4144 					  BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
4145 					  &sc->bge_cdata.bge_tx_dmamap[i]);
4146 		if (error) {
4147 			int j;
4148 
4149 			for (j = 0; j < i; ++j) {
4150 				bus_dmamap_destroy(sc->bge_cdata.bge_tx_mtag,
4151 					sc->bge_cdata.bge_tx_dmamap[j]);
4152 			}
4153 			bus_dma_tag_destroy(sc->bge_cdata.bge_tx_mtag);
4154 			sc->bge_cdata.bge_tx_mtag = NULL;
4155 
4156 			if_printf(ifp, "could not create DMA map for TX\n");
4157 			return error;
4158 		}
4159 	}
4160 
4161 	/*
4162 	 * Create DMA stuffs for standard RX ring.
4163 	 */
4164 	error = bge_dma_block_alloc(sc, BGE_STD_RX_RING_SZ,
4165 				    &sc->bge_cdata.bge_rx_std_ring_tag,
4166 				    &sc->bge_cdata.bge_rx_std_ring_map,
4167 				    (void *)&sc->bge_ldata.bge_rx_std_ring,
4168 				    &sc->bge_ldata.bge_rx_std_ring_paddr);
4169 	if (error) {
4170 		if_printf(ifp, "could not create std RX ring\n");
4171 		return error;
4172 	}
4173 
4174 	/*
4175 	 * Create jumbo buffer pool.
4176 	 */
4177 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
4178 		error = bge_alloc_jumbo_mem(sc);
4179 		if (error) {
4180 			if_printf(ifp, "could not create jumbo buffer pool\n");
4181 			return error;
4182 		}
4183 	}
4184 
4185 	/*
4186 	 * Create DMA stuffs for RX return ring.
4187 	 */
4188 	error = bge_dma_block_alloc(sc,
4189 	    BGE_RX_RTN_RING_SZ(sc->bge_return_ring_cnt),
4190 	    &sc->bge_cdata.bge_rx_return_ring_tag,
4191 	    &sc->bge_cdata.bge_rx_return_ring_map,
4192 	    (void *)&sc->bge_ldata.bge_rx_return_ring,
4193 	    &sc->bge_ldata.bge_rx_return_ring_paddr);
4194 	if (error) {
4195 		if_printf(ifp, "could not create RX ret ring\n");
4196 		return error;
4197 	}
4198 
4199 	/*
4200 	 * Create DMA stuffs for TX ring.
4201 	 */
4202 	error = bge_dma_block_alloc(sc, BGE_TX_RING_SZ,
4203 				    &sc->bge_cdata.bge_tx_ring_tag,
4204 				    &sc->bge_cdata.bge_tx_ring_map,
4205 				    (void *)&sc->bge_ldata.bge_tx_ring,
4206 				    &sc->bge_ldata.bge_tx_ring_paddr);
4207 	if (error) {
4208 		if_printf(ifp, "could not create TX ring\n");
4209 		return error;
4210 	}
4211 
4212 	/*
4213 	 * Create DMA stuffs for status block.
4214 	 */
4215 	error = bge_dma_block_alloc(sc, BGE_STATUS_BLK_SZ,
4216 				    &sc->bge_cdata.bge_status_tag,
4217 				    &sc->bge_cdata.bge_status_map,
4218 				    (void *)&sc->bge_ldata.bge_status_block,
4219 				    &sc->bge_ldata.bge_status_block_paddr);
4220 	if (error) {
4221 		if_printf(ifp, "could not create status block\n");
4222 		return error;
4223 	}
4224 
4225 	/*
4226 	 * Create DMA stuffs for statistics block.
4227 	 */
4228 	error = bge_dma_block_alloc(sc, BGE_STATS_SZ,
4229 				    &sc->bge_cdata.bge_stats_tag,
4230 				    &sc->bge_cdata.bge_stats_map,
4231 				    (void *)&sc->bge_ldata.bge_stats,
4232 				    &sc->bge_ldata.bge_stats_paddr);
4233 	if (error) {
4234 		if_printf(ifp, "could not create stats block\n");
4235 		return error;
4236 	}
4237 	return 0;
4238 }
4239 
4240 static int
4241 bge_dma_block_alloc(struct bge_softc *sc, bus_size_t size, bus_dma_tag_t *tag,
4242 		    bus_dmamap_t *map, void **addr, bus_addr_t *paddr)
4243 {
4244 	bus_dmamem_t dmem;
4245 	int error;
4246 
4247 	error = bus_dmamem_coherent(sc->bge_cdata.bge_parent_tag, PAGE_SIZE, 0,
4248 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
4249 				    size, BUS_DMA_WAITOK | BUS_DMA_ZERO, &dmem);
4250 	if (error)
4251 		return error;
4252 
4253 	*tag = dmem.dmem_tag;
4254 	*map = dmem.dmem_map;
4255 	*addr = dmem.dmem_addr;
4256 	*paddr = dmem.dmem_busaddr;
4257 
4258 	return 0;
4259 }
4260 
4261 static void
4262 bge_dma_block_free(bus_dma_tag_t tag, bus_dmamap_t map, void *addr)
4263 {
4264 	if (tag != NULL) {
4265 		bus_dmamap_unload(tag, map);
4266 		bus_dmamem_free(tag, addr, map);
4267 		bus_dma_tag_destroy(tag);
4268 	}
4269 }
4270 
4271 /*
4272  * Grrr. The link status word in the status block does
4273  * not work correctly on the BCM5700 rev AX and BX chips,
4274  * according to all available information. Hence, we have
4275  * to enable MII interrupts in order to properly obtain
4276  * async link changes. Unfortunately, this also means that
4277  * we have to read the MAC status register to detect link
4278  * changes, thereby adding an additional register access to
4279  * the interrupt handler.
4280  *
4281  * XXX: perhaps link state detection procedure used for
4282  * BGE_CHIPID_BCM5700_B2 can be used for others BCM5700 revisions.
4283  */
4284 static void
4285 bge_bcm5700_link_upd(struct bge_softc *sc, uint32_t status __unused)
4286 {
4287 	struct ifnet *ifp = &sc->arpcom.ac_if;
4288 	struct mii_data *mii = device_get_softc(sc->bge_miibus);
4289 
4290 	mii_pollstat(mii);
4291 
4292 	if (!sc->bge_link &&
4293 	    (mii->mii_media_status & IFM_ACTIVE) &&
4294 	    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
4295 		sc->bge_link++;
4296 		if (bootverbose)
4297 			if_printf(ifp, "link UP\n");
4298 	} else if (sc->bge_link &&
4299 	    (!(mii->mii_media_status & IFM_ACTIVE) ||
4300 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE)) {
4301 		sc->bge_link = 0;
4302 		if (bootverbose)
4303 			if_printf(ifp, "link DOWN\n");
4304 	}
4305 
4306 	/* Clear the interrupt. */
4307 	CSR_WRITE_4(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_MI_INTERRUPT);
4308 	bge_miibus_readreg(sc->bge_dev, 1, BRGPHY_MII_ISR);
4309 	bge_miibus_writereg(sc->bge_dev, 1, BRGPHY_MII_IMR, BRGPHY_INTRS);
4310 }
4311 
4312 static void
4313 bge_tbi_link_upd(struct bge_softc *sc, uint32_t status)
4314 {
4315 	struct ifnet *ifp = &sc->arpcom.ac_if;
4316 
4317 #define PCS_ENCODE_ERR	(BGE_MACSTAT_PORT_DECODE_ERROR|BGE_MACSTAT_MI_COMPLETE)
4318 
4319 	/*
4320 	 * Sometimes PCS encoding errors are detected in
4321 	 * TBI mode (on fiber NICs), and for some reason
4322 	 * the chip will signal them as link changes.
4323 	 * If we get a link change event, but the 'PCS
4324 	 * encoding error' bit in the MAC status register
4325 	 * is set, don't bother doing a link check.
4326 	 * This avoids spurious "gigabit link up" messages
4327 	 * that sometimes appear on fiber NICs during
4328 	 * periods of heavy traffic.
4329 	 */
4330 	if (status & BGE_MACSTAT_TBI_PCS_SYNCHED) {
4331 		if (!sc->bge_link) {
4332 			sc->bge_link++;
4333 			if (sc->bge_asicrev == BGE_ASICREV_BCM5704) {
4334 				BGE_CLRBIT(sc, BGE_MAC_MODE,
4335 				    BGE_MACMODE_TBI_SEND_CFGS);
4336 			}
4337 			CSR_WRITE_4(sc, BGE_MAC_STS, 0xFFFFFFFF);
4338 
4339 			if (bootverbose)
4340 				if_printf(ifp, "link UP\n");
4341 
4342 			ifp->if_link_state = LINK_STATE_UP;
4343 			if_link_state_change(ifp);
4344 		}
4345 	} else if ((status & PCS_ENCODE_ERR) != PCS_ENCODE_ERR) {
4346 		if (sc->bge_link) {
4347 			sc->bge_link = 0;
4348 
4349 			if (bootverbose)
4350 				if_printf(ifp, "link DOWN\n");
4351 
4352 			ifp->if_link_state = LINK_STATE_DOWN;
4353 			if_link_state_change(ifp);
4354 		}
4355 	}
4356 
4357 #undef PCS_ENCODE_ERR
4358 
4359 	/* Clear the attention. */
4360 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
4361 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
4362 	    BGE_MACSTAT_LINK_CHANGED);
4363 }
4364 
4365 static void
4366 bge_copper_link_upd(struct bge_softc *sc, uint32_t status __unused)
4367 {
4368 	struct ifnet *ifp = &sc->arpcom.ac_if;
4369 	struct mii_data *mii = device_get_softc(sc->bge_miibus);
4370 
4371 	mii_pollstat(mii);
4372 	bge_miibus_statchg(sc->bge_dev);
4373 
4374 	if (bootverbose) {
4375 		if (sc->bge_link)
4376 			if_printf(ifp, "link UP\n");
4377 		else
4378 			if_printf(ifp, "link DOWN\n");
4379 	}
4380 
4381 	/* Clear the attention. */
4382 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
4383 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
4384 	    BGE_MACSTAT_LINK_CHANGED);
4385 }
4386 
4387 static void
4388 bge_autopoll_link_upd(struct bge_softc *sc, uint32_t status __unused)
4389 {
4390 	struct ifnet *ifp = &sc->arpcom.ac_if;
4391 	struct mii_data *mii = device_get_softc(sc->bge_miibus);
4392 
4393 	mii_pollstat(mii);
4394 
4395 	if (!sc->bge_link &&
4396 	    (mii->mii_media_status & IFM_ACTIVE) &&
4397 	    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
4398 		sc->bge_link++;
4399 		if (bootverbose)
4400 			if_printf(ifp, "link UP\n");
4401 	} else if (sc->bge_link &&
4402 	    (!(mii->mii_media_status & IFM_ACTIVE) ||
4403 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE)) {
4404 		sc->bge_link = 0;
4405 		if (bootverbose)
4406 			if_printf(ifp, "link DOWN\n");
4407 	}
4408 
4409 	/* Clear the attention. */
4410 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
4411 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
4412 	    BGE_MACSTAT_LINK_CHANGED);
4413 }
4414 
4415 static int
4416 bge_sysctl_rx_coal_ticks(SYSCTL_HANDLER_ARGS)
4417 {
4418 	struct bge_softc *sc = arg1;
4419 
4420 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4421 	    &sc->bge_rx_coal_ticks,
4422 	    BGE_RX_COAL_TICKS_MIN, BGE_RX_COAL_TICKS_MAX,
4423 	    BGE_RX_COAL_TICKS_CHG);
4424 }
4425 
4426 static int
4427 bge_sysctl_tx_coal_ticks(SYSCTL_HANDLER_ARGS)
4428 {
4429 	struct bge_softc *sc = arg1;
4430 
4431 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4432 	    &sc->bge_tx_coal_ticks,
4433 	    BGE_TX_COAL_TICKS_MIN, BGE_TX_COAL_TICKS_MAX,
4434 	    BGE_TX_COAL_TICKS_CHG);
4435 }
4436 
4437 static int
4438 bge_sysctl_rx_coal_bds(SYSCTL_HANDLER_ARGS)
4439 {
4440 	struct bge_softc *sc = arg1;
4441 
4442 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4443 	    &sc->bge_rx_coal_bds,
4444 	    BGE_RX_COAL_BDS_MIN, BGE_RX_COAL_BDS_MAX,
4445 	    BGE_RX_COAL_BDS_CHG);
4446 }
4447 
4448 static int
4449 bge_sysctl_tx_coal_bds(SYSCTL_HANDLER_ARGS)
4450 {
4451 	struct bge_softc *sc = arg1;
4452 
4453 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4454 	    &sc->bge_tx_coal_bds,
4455 	    BGE_TX_COAL_BDS_MIN, BGE_TX_COAL_BDS_MAX,
4456 	    BGE_TX_COAL_BDS_CHG);
4457 }
4458 
4459 static int
4460 bge_sysctl_rx_coal_ticks_int(SYSCTL_HANDLER_ARGS)
4461 {
4462 	struct bge_softc *sc = arg1;
4463 
4464 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4465 	    &sc->bge_rx_coal_ticks_int,
4466 	    BGE_RX_COAL_TICKS_MIN, BGE_RX_COAL_TICKS_MAX,
4467 	    BGE_RX_COAL_TICKS_INT_CHG);
4468 }
4469 
4470 static int
4471 bge_sysctl_tx_coal_ticks_int(SYSCTL_HANDLER_ARGS)
4472 {
4473 	struct bge_softc *sc = arg1;
4474 
4475 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4476 	    &sc->bge_tx_coal_ticks_int,
4477 	    BGE_TX_COAL_TICKS_MIN, BGE_TX_COAL_TICKS_MAX,
4478 	    BGE_TX_COAL_TICKS_INT_CHG);
4479 }
4480 
4481 static int
4482 bge_sysctl_rx_coal_bds_int(SYSCTL_HANDLER_ARGS)
4483 {
4484 	struct bge_softc *sc = arg1;
4485 
4486 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4487 	    &sc->bge_rx_coal_bds_int,
4488 	    BGE_RX_COAL_BDS_MIN, BGE_RX_COAL_BDS_MAX,
4489 	    BGE_RX_COAL_BDS_INT_CHG);
4490 }
4491 
4492 static int
4493 bge_sysctl_tx_coal_bds_int(SYSCTL_HANDLER_ARGS)
4494 {
4495 	struct bge_softc *sc = arg1;
4496 
4497 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4498 	    &sc->bge_tx_coal_bds_int,
4499 	    BGE_TX_COAL_BDS_MIN, BGE_TX_COAL_BDS_MAX,
4500 	    BGE_TX_COAL_BDS_INT_CHG);
4501 }
4502 
4503 static int
4504 bge_sysctl_coal_chg(SYSCTL_HANDLER_ARGS, uint32_t *coal,
4505     int coal_min, int coal_max, uint32_t coal_chg_mask)
4506 {
4507 	struct bge_softc *sc = arg1;
4508 	struct ifnet *ifp = &sc->arpcom.ac_if;
4509 	int error = 0, v;
4510 
4511 	lwkt_serialize_enter(ifp->if_serializer);
4512 
4513 	v = *coal;
4514 	error = sysctl_handle_int(oidp, &v, 0, req);
4515 	if (!error && req->newptr != NULL) {
4516 		if (v < coal_min || v > coal_max) {
4517 			error = EINVAL;
4518 		} else {
4519 			*coal = v;
4520 			sc->bge_coal_chg |= coal_chg_mask;
4521 		}
4522 	}
4523 
4524 	lwkt_serialize_exit(ifp->if_serializer);
4525 	return error;
4526 }
4527 
4528 static void
4529 bge_coal_change(struct bge_softc *sc)
4530 {
4531 	struct ifnet *ifp = &sc->arpcom.ac_if;
4532 	uint32_t val;
4533 
4534 	ASSERT_SERIALIZED(ifp->if_serializer);
4535 
4536 	if (sc->bge_coal_chg & BGE_RX_COAL_TICKS_CHG) {
4537 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS,
4538 			    sc->bge_rx_coal_ticks);
4539 		DELAY(10);
4540 		val = CSR_READ_4(sc, BGE_HCC_RX_COAL_TICKS);
4541 
4542 		if (bootverbose) {
4543 			if_printf(ifp, "rx_coal_ticks -> %u\n",
4544 				  sc->bge_rx_coal_ticks);
4545 		}
4546 	}
4547 
4548 	if (sc->bge_coal_chg & BGE_TX_COAL_TICKS_CHG) {
4549 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS,
4550 			    sc->bge_tx_coal_ticks);
4551 		DELAY(10);
4552 		val = CSR_READ_4(sc, BGE_HCC_TX_COAL_TICKS);
4553 
4554 		if (bootverbose) {
4555 			if_printf(ifp, "tx_coal_ticks -> %u\n",
4556 				  sc->bge_tx_coal_ticks);
4557 		}
4558 	}
4559 
4560 	if (sc->bge_coal_chg & BGE_RX_COAL_BDS_CHG) {
4561 		CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS,
4562 			    sc->bge_rx_coal_bds);
4563 		DELAY(10);
4564 		val = CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS);
4565 
4566 		if (bootverbose) {
4567 			if_printf(ifp, "rx_coal_bds -> %u\n",
4568 				  sc->bge_rx_coal_bds);
4569 		}
4570 	}
4571 
4572 	if (sc->bge_coal_chg & BGE_TX_COAL_BDS_CHG) {
4573 		CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS,
4574 			    sc->bge_tx_coal_bds);
4575 		DELAY(10);
4576 		val = CSR_READ_4(sc, BGE_HCC_TX_MAX_COAL_BDS);
4577 
4578 		if (bootverbose) {
4579 			if_printf(ifp, "tx_max_coal_bds -> %u\n",
4580 				  sc->bge_tx_coal_bds);
4581 		}
4582 	}
4583 
4584 	if (sc->bge_coal_chg & BGE_RX_COAL_TICKS_INT_CHG) {
4585 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS_INT,
4586 		    sc->bge_rx_coal_ticks_int);
4587 		DELAY(10);
4588 		val = CSR_READ_4(sc, BGE_HCC_RX_COAL_TICKS_INT);
4589 
4590 		if (bootverbose) {
4591 			if_printf(ifp, "rx_coal_ticks_int -> %u\n",
4592 			    sc->bge_rx_coal_ticks_int);
4593 		}
4594 	}
4595 
4596 	if (sc->bge_coal_chg & BGE_TX_COAL_TICKS_INT_CHG) {
4597 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS_INT,
4598 		    sc->bge_tx_coal_ticks_int);
4599 		DELAY(10);
4600 		val = CSR_READ_4(sc, BGE_HCC_TX_COAL_TICKS_INT);
4601 
4602 		if (bootverbose) {
4603 			if_printf(ifp, "tx_coal_ticks_int -> %u\n",
4604 			    sc->bge_tx_coal_ticks_int);
4605 		}
4606 	}
4607 
4608 	if (sc->bge_coal_chg & BGE_RX_COAL_BDS_INT_CHG) {
4609 		CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT,
4610 		    sc->bge_rx_coal_bds_int);
4611 		DELAY(10);
4612 		val = CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT);
4613 
4614 		if (bootverbose) {
4615 			if_printf(ifp, "rx_coal_bds_int -> %u\n",
4616 			    sc->bge_rx_coal_bds_int);
4617 		}
4618 	}
4619 
4620 	if (sc->bge_coal_chg & BGE_TX_COAL_BDS_INT_CHG) {
4621 		CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT,
4622 		    sc->bge_tx_coal_bds_int);
4623 		DELAY(10);
4624 		val = CSR_READ_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT);
4625 
4626 		if (bootverbose) {
4627 			if_printf(ifp, "tx_coal_bds_int -> %u\n",
4628 			    sc->bge_tx_coal_bds_int);
4629 		}
4630 	}
4631 
4632 	sc->bge_coal_chg = 0;
4633 }
4634 
4635 static void
4636 bge_enable_intr(struct bge_softc *sc)
4637 {
4638 	struct ifnet *ifp = &sc->arpcom.ac_if;
4639 
4640 	lwkt_serialize_handler_enable(ifp->if_serializer);
4641 
4642 	/*
4643 	 * Enable interrupt.
4644 	 */
4645 	bge_writembx(sc, BGE_MBX_IRQ0_LO, sc->bge_status_tag << 24);
4646 	if (sc->bge_flags & BGE_FLAG_ONESHOT_MSI) {
4647 		/* XXX Linux driver */
4648 		bge_writembx(sc, BGE_MBX_IRQ0_LO, sc->bge_status_tag << 24);
4649 	}
4650 
4651 	/*
4652 	 * Unmask the interrupt when we stop polling.
4653 	 */
4654 	PCI_CLRBIT(sc->bge_dev, BGE_PCI_MISC_CTL,
4655 	    BGE_PCIMISCCTL_MASK_PCI_INTR, 4);
4656 
4657 	/*
4658 	 * Trigger another interrupt, since above writing
4659 	 * to interrupt mailbox0 may acknowledge pending
4660 	 * interrupt.
4661 	 */
4662 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
4663 }
4664 
4665 static void
4666 bge_disable_intr(struct bge_softc *sc)
4667 {
4668 	struct ifnet *ifp = &sc->arpcom.ac_if;
4669 
4670 	/*
4671 	 * Mask the interrupt when we start polling.
4672 	 */
4673 	PCI_SETBIT(sc->bge_dev, BGE_PCI_MISC_CTL,
4674 	    BGE_PCIMISCCTL_MASK_PCI_INTR, 4);
4675 
4676 	/*
4677 	 * Acknowledge possible asserted interrupt.
4678 	 */
4679 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
4680 
4681 	lwkt_serialize_handler_disable(ifp->if_serializer);
4682 }
4683 
4684 static int
4685 bge_get_eaddr_mem(struct bge_softc *sc, uint8_t ether_addr[])
4686 {
4687 	uint32_t mac_addr;
4688 	int ret = 1;
4689 
4690 	mac_addr = bge_readmem_ind(sc, 0x0c14);
4691 	if ((mac_addr >> 16) == 0x484b) {
4692 		ether_addr[0] = (uint8_t)(mac_addr >> 8);
4693 		ether_addr[1] = (uint8_t)mac_addr;
4694 		mac_addr = bge_readmem_ind(sc, 0x0c18);
4695 		ether_addr[2] = (uint8_t)(mac_addr >> 24);
4696 		ether_addr[3] = (uint8_t)(mac_addr >> 16);
4697 		ether_addr[4] = (uint8_t)(mac_addr >> 8);
4698 		ether_addr[5] = (uint8_t)mac_addr;
4699 		ret = 0;
4700 	}
4701 	return ret;
4702 }
4703 
4704 static int
4705 bge_get_eaddr_nvram(struct bge_softc *sc, uint8_t ether_addr[])
4706 {
4707 	int mac_offset = BGE_EE_MAC_OFFSET;
4708 
4709 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
4710 		mac_offset = BGE_EE_MAC_OFFSET_5906;
4711 
4712 	return bge_read_nvram(sc, ether_addr, mac_offset + 2, ETHER_ADDR_LEN);
4713 }
4714 
4715 static int
4716 bge_get_eaddr_eeprom(struct bge_softc *sc, uint8_t ether_addr[])
4717 {
4718 	if (sc->bge_flags & BGE_FLAG_NO_EEPROM)
4719 		return 1;
4720 
4721 	return bge_read_eeprom(sc, ether_addr, BGE_EE_MAC_OFFSET + 2,
4722 			       ETHER_ADDR_LEN);
4723 }
4724 
4725 static int
4726 bge_get_eaddr(struct bge_softc *sc, uint8_t eaddr[])
4727 {
4728 	static const bge_eaddr_fcn_t bge_eaddr_funcs[] = {
4729 		/* NOTE: Order is critical */
4730 		bge_get_eaddr_mem,
4731 		bge_get_eaddr_nvram,
4732 		bge_get_eaddr_eeprom,
4733 		NULL
4734 	};
4735 	const bge_eaddr_fcn_t *func;
4736 
4737 	for (func = bge_eaddr_funcs; *func != NULL; ++func) {
4738 		if ((*func)(sc, eaddr) == 0)
4739 			break;
4740 	}
4741 	return (*func == NULL ? ENXIO : 0);
4742 }
4743 
4744 /*
4745  * NOTE: 'm' is not freed upon failure
4746  */
4747 struct mbuf *
4748 bge_defrag_shortdma(struct mbuf *m)
4749 {
4750 	struct mbuf *n;
4751 	int found;
4752 
4753 	/*
4754 	 * If device receive two back-to-back send BDs with less than
4755 	 * or equal to 8 total bytes then the device may hang.  The two
4756 	 * back-to-back send BDs must in the same frame for this failure
4757 	 * to occur.  Scan mbuf chains and see whether two back-to-back
4758 	 * send BDs are there.  If this is the case, allocate new mbuf
4759 	 * and copy the frame to workaround the silicon bug.
4760 	 */
4761 	for (n = m, found = 0; n != NULL; n = n->m_next) {
4762 		if (n->m_len < 8) {
4763 			found++;
4764 			if (found > 1)
4765 				break;
4766 			continue;
4767 		}
4768 		found = 0;
4769 	}
4770 
4771 	if (found > 1)
4772 		n = m_defrag(m, MB_DONTWAIT);
4773 	else
4774 		n = m;
4775 	return n;
4776 }
4777 
4778 static void
4779 bge_stop_block(struct bge_softc *sc, bus_size_t reg, uint32_t bit)
4780 {
4781 	int i;
4782 
4783 	BGE_CLRBIT(sc, reg, bit);
4784 	for (i = 0; i < BGE_TIMEOUT; i++) {
4785 		if ((CSR_READ_4(sc, reg) & bit) == 0)
4786 			return;
4787 		DELAY(100);
4788 	}
4789 }
4790 
4791 static void
4792 bge_link_poll(struct bge_softc *sc)
4793 {
4794 	uint32_t status;
4795 
4796 	status = CSR_READ_4(sc, BGE_MAC_STS);
4797 	if ((status & sc->bge_link_chg) || sc->bge_link_evt) {
4798 		sc->bge_link_evt = 0;
4799 		sc->bge_link_upd(sc, status);
4800 	}
4801 }
4802 
4803 static void
4804 bge_enable_msi(struct bge_softc *sc)
4805 {
4806 	uint32_t msi_mode;
4807 
4808 	msi_mode = CSR_READ_4(sc, BGE_MSI_MODE);
4809 	msi_mode |= BGE_MSIMODE_ENABLE;
4810 	if (sc->bge_flags & BGE_FLAG_ONESHOT_MSI) {
4811 		/*
4812 		 * According to all of the datasheets that are publicly
4813 		 * available, bit 5 of the MSI_MODE is defined to be
4814 		 * "MSI FIFO Underrun Attn" for BCM5755+ and BCM5906, on
4815 		 * which "oneshot MSI" is enabled.  However, it is always
4816 		 * safe to clear it here.
4817 		 */
4818 		msi_mode &= ~BGE_MSIMODE_ONESHOT_DISABLE;
4819 	}
4820 	CSR_WRITE_4(sc, BGE_MSI_MODE, msi_mode);
4821 }
4822