xref: /dflybsd-src/sys/dev/netif/bge/if_bge.c (revision 75b4da16fa23b21d85fd03c59414b1503a5bae46)
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(void *);
308 static void	bge_intr_status_tag(void *);
309 static void	bge_enable_intr(struct bge_softc *);
310 static void	bge_disable_intr(struct bge_softc *);
311 static void	bge_start(struct ifnet *);
312 static int	bge_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
313 static void	bge_init(void *);
314 static void	bge_stop(struct bge_softc *);
315 static void	bge_watchdog(struct ifnet *);
316 static void	bge_shutdown(device_t);
317 static int	bge_suspend(device_t);
318 static int	bge_resume(device_t);
319 static int	bge_ifmedia_upd(struct ifnet *);
320 static void	bge_ifmedia_sts(struct ifnet *, struct ifmediareq *);
321 
322 static uint8_t	bge_nvram_getbyte(struct bge_softc *, int, uint8_t *);
323 static int	bge_read_nvram(struct bge_softc *, caddr_t, int, int);
324 
325 static uint8_t	bge_eeprom_getbyte(struct bge_softc *, uint32_t, uint8_t *);
326 static int	bge_read_eeprom(struct bge_softc *, caddr_t, uint32_t, size_t);
327 
328 static void	bge_setmulti(struct bge_softc *);
329 static void	bge_setpromisc(struct bge_softc *);
330 
331 static int	bge_alloc_jumbo_mem(struct bge_softc *);
332 static void	bge_free_jumbo_mem(struct bge_softc *);
333 static struct bge_jslot
334 		*bge_jalloc(struct bge_softc *);
335 static void	bge_jfree(void *);
336 static void	bge_jref(void *);
337 static int	bge_newbuf_std(struct bge_softc *, int, int);
338 static int	bge_newbuf_jumbo(struct bge_softc *, int, int);
339 static void	bge_setup_rxdesc_std(struct bge_softc *, int);
340 static void	bge_setup_rxdesc_jumbo(struct bge_softc *, int);
341 static int	bge_init_rx_ring_std(struct bge_softc *);
342 static void	bge_free_rx_ring_std(struct bge_softc *);
343 static int	bge_init_rx_ring_jumbo(struct bge_softc *);
344 static void	bge_free_rx_ring_jumbo(struct bge_softc *);
345 static void	bge_free_tx_ring(struct bge_softc *);
346 static int	bge_init_tx_ring(struct bge_softc *);
347 
348 static int	bge_chipinit(struct bge_softc *);
349 static int	bge_blockinit(struct bge_softc *);
350 static void	bge_stop_block(struct bge_softc *, bus_size_t, uint32_t);
351 
352 static uint32_t	bge_readmem_ind(struct bge_softc *, uint32_t);
353 static void	bge_writemem_ind(struct bge_softc *, uint32_t, uint32_t);
354 #ifdef notdef
355 static uint32_t	bge_readreg_ind(struct bge_softc *, uint32_t);
356 #endif
357 static void	bge_writereg_ind(struct bge_softc *, uint32_t, uint32_t);
358 static void	bge_writemem_direct(struct bge_softc *, uint32_t, uint32_t);
359 static void	bge_writembx(struct bge_softc *, int, int);
360 
361 static int	bge_miibus_readreg(device_t, int, int);
362 static int	bge_miibus_writereg(device_t, int, int, int);
363 static void	bge_miibus_statchg(device_t);
364 static void	bge_bcm5700_link_upd(struct bge_softc *, uint32_t);
365 static void	bge_tbi_link_upd(struct bge_softc *, uint32_t);
366 static void	bge_copper_link_upd(struct bge_softc *, uint32_t);
367 static void	bge_autopoll_link_upd(struct bge_softc *, uint32_t);
368 static void	bge_link_poll(struct bge_softc *);
369 
370 static void	bge_reset(struct bge_softc *);
371 
372 static int	bge_dma_alloc(struct bge_softc *);
373 static void	bge_dma_free(struct bge_softc *);
374 static int	bge_dma_block_alloc(struct bge_softc *, bus_size_t,
375 				    bus_dma_tag_t *, bus_dmamap_t *,
376 				    void **, bus_addr_t *);
377 static void	bge_dma_block_free(bus_dma_tag_t, bus_dmamap_t, void *);
378 
379 static int	bge_get_eaddr_mem(struct bge_softc *, uint8_t[]);
380 static int	bge_get_eaddr_nvram(struct bge_softc *, uint8_t[]);
381 static int	bge_get_eaddr_eeprom(struct bge_softc *, uint8_t[]);
382 static int	bge_get_eaddr(struct bge_softc *, uint8_t[]);
383 
384 static void	bge_coal_change(struct bge_softc *);
385 static int	bge_sysctl_rx_coal_ticks(SYSCTL_HANDLER_ARGS);
386 static int	bge_sysctl_tx_coal_ticks(SYSCTL_HANDLER_ARGS);
387 static int	bge_sysctl_rx_coal_bds(SYSCTL_HANDLER_ARGS);
388 static int	bge_sysctl_tx_coal_bds(SYSCTL_HANDLER_ARGS);
389 static int	bge_sysctl_rx_coal_ticks_int(SYSCTL_HANDLER_ARGS);
390 static int	bge_sysctl_tx_coal_ticks_int(SYSCTL_HANDLER_ARGS);
391 static int	bge_sysctl_rx_coal_bds_int(SYSCTL_HANDLER_ARGS);
392 static int	bge_sysctl_tx_coal_bds_int(SYSCTL_HANDLER_ARGS);
393 static int	bge_sysctl_coal_chg(SYSCTL_HANDLER_ARGS, uint32_t *,
394 		    int, int, uint32_t);
395 
396 /*
397  * Set following tunable to 1 for some IBM blade servers with the DNLK
398  * switch module. Auto negotiation is broken for those configurations.
399  */
400 static int	bge_fake_autoneg = 0;
401 TUNABLE_INT("hw.bge.fake_autoneg", &bge_fake_autoneg);
402 
403 #if !defined(KTR_IF_BGE)
404 #define KTR_IF_BGE	KTR_ALL
405 #endif
406 KTR_INFO_MASTER(if_bge);
407 KTR_INFO(KTR_IF_BGE, if_bge, intr, 0, "intr");
408 KTR_INFO(KTR_IF_BGE, if_bge, rx_pkt, 1, "rx_pkt");
409 KTR_INFO(KTR_IF_BGE, if_bge, tx_pkt, 2, "tx_pkt");
410 #define logif(name)	KTR_LOG(if_bge_ ## name)
411 
412 static device_method_t bge_methods[] = {
413 	/* Device interface */
414 	DEVMETHOD(device_probe,		bge_probe),
415 	DEVMETHOD(device_attach,	bge_attach),
416 	DEVMETHOD(device_detach,	bge_detach),
417 	DEVMETHOD(device_shutdown,	bge_shutdown),
418 	DEVMETHOD(device_suspend,	bge_suspend),
419 	DEVMETHOD(device_resume,	bge_resume),
420 
421 	/* bus interface */
422 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
423 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
424 
425 	/* MII interface */
426 	DEVMETHOD(miibus_readreg,	bge_miibus_readreg),
427 	DEVMETHOD(miibus_writereg,	bge_miibus_writereg),
428 	DEVMETHOD(miibus_statchg,	bge_miibus_statchg),
429 
430 	{ 0, 0 }
431 };
432 
433 static DEFINE_CLASS_0(bge, bge_driver, bge_methods, sizeof(struct bge_softc));
434 static devclass_t bge_devclass;
435 
436 DECLARE_DUMMY_MODULE(if_bge);
437 DRIVER_MODULE(if_bge, pci, bge_driver, bge_devclass, NULL, NULL);
438 DRIVER_MODULE(miibus, bge, miibus_driver, miibus_devclass, NULL, NULL);
439 
440 static uint32_t
441 bge_readmem_ind(struct bge_softc *sc, uint32_t off)
442 {
443 	device_t dev = sc->bge_dev;
444 	uint32_t val;
445 
446 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
447 	    off >= BGE_STATS_BLOCK && off < BGE_SEND_RING_1_TO_4)
448 		return 0;
449 
450 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4);
451 	val = pci_read_config(dev, BGE_PCI_MEMWIN_DATA, 4);
452 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4);
453 	return (val);
454 }
455 
456 static void
457 bge_writemem_ind(struct bge_softc *sc, uint32_t off, uint32_t val)
458 {
459 	device_t dev = sc->bge_dev;
460 
461 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
462 	    off >= BGE_STATS_BLOCK && off < BGE_SEND_RING_1_TO_4)
463 		return;
464 
465 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4);
466 	pci_write_config(dev, BGE_PCI_MEMWIN_DATA, val, 4);
467 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4);
468 }
469 
470 #ifdef notdef
471 static uint32_t
472 bge_readreg_ind(struct bge_softc *sc, uin32_t off)
473 {
474 	device_t dev = sc->bge_dev;
475 
476 	pci_write_config(dev, BGE_PCI_REG_BASEADDR, off, 4);
477 	return(pci_read_config(dev, BGE_PCI_REG_DATA, 4));
478 }
479 #endif
480 
481 static void
482 bge_writereg_ind(struct bge_softc *sc, uint32_t off, uint32_t val)
483 {
484 	device_t dev = sc->bge_dev;
485 
486 	pci_write_config(dev, BGE_PCI_REG_BASEADDR, off, 4);
487 	pci_write_config(dev, BGE_PCI_REG_DATA, val, 4);
488 }
489 
490 static void
491 bge_writemem_direct(struct bge_softc *sc, uint32_t off, uint32_t val)
492 {
493 	CSR_WRITE_4(sc, off, val);
494 }
495 
496 static void
497 bge_writembx(struct bge_softc *sc, int off, int val)
498 {
499 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
500 		off += BGE_LPMBX_IRQ0_HI - BGE_MBX_IRQ0_HI;
501 
502 	CSR_WRITE_4(sc, off, val);
503 	if (sc->bge_mbox_reorder)
504 		CSR_READ_4(sc, off);
505 }
506 
507 static uint8_t
508 bge_nvram_getbyte(struct bge_softc *sc, int addr, uint8_t *dest)
509 {
510 	uint32_t access, byte = 0;
511 	int i;
512 
513 	/* Lock. */
514 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_SET1);
515 	for (i = 0; i < 8000; i++) {
516 		if (CSR_READ_4(sc, BGE_NVRAM_SWARB) & BGE_NVRAMSWARB_GNT1)
517 			break;
518 		DELAY(20);
519 	}
520 	if (i == 8000)
521 		return (1);
522 
523 	/* Enable access. */
524 	access = CSR_READ_4(sc, BGE_NVRAM_ACCESS);
525 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access | BGE_NVRAMACC_ENABLE);
526 
527 	CSR_WRITE_4(sc, BGE_NVRAM_ADDR, addr & 0xfffffffc);
528 	CSR_WRITE_4(sc, BGE_NVRAM_CMD, BGE_NVRAM_READCMD);
529 	for (i = 0; i < BGE_TIMEOUT * 10; i++) {
530 		DELAY(10);
531 		if (CSR_READ_4(sc, BGE_NVRAM_CMD) & BGE_NVRAMCMD_DONE) {
532 			DELAY(10);
533 			break;
534 		}
535 	}
536 
537 	if (i == BGE_TIMEOUT * 10) {
538 		if_printf(&sc->arpcom.ac_if, "nvram read timed out\n");
539 		return (1);
540 	}
541 
542 	/* Get result. */
543 	byte = CSR_READ_4(sc, BGE_NVRAM_RDDATA);
544 
545 	*dest = (bswap32(byte) >> ((addr % 4) * 8)) & 0xFF;
546 
547 	/* Disable access. */
548 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access);
549 
550 	/* Unlock. */
551 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_CLR1);
552 	CSR_READ_4(sc, BGE_NVRAM_SWARB);
553 
554 	return (0);
555 }
556 
557 /*
558  * Read a sequence of bytes from NVRAM.
559  */
560 static int
561 bge_read_nvram(struct bge_softc *sc, caddr_t dest, int off, int cnt)
562 {
563 	int err = 0, i;
564 	uint8_t byte = 0;
565 
566 	if (sc->bge_asicrev != BGE_ASICREV_BCM5906)
567 		return (1);
568 
569 	for (i = 0; i < cnt; i++) {
570 		err = bge_nvram_getbyte(sc, off + i, &byte);
571 		if (err)
572 			break;
573 		*(dest + i) = byte;
574 	}
575 
576 	return (err ? 1 : 0);
577 }
578 
579 /*
580  * Read a byte of data stored in the EEPROM at address 'addr.' The
581  * BCM570x supports both the traditional bitbang interface and an
582  * auto access interface for reading the EEPROM. We use the auto
583  * access method.
584  */
585 static uint8_t
586 bge_eeprom_getbyte(struct bge_softc *sc, uint32_t addr, uint8_t *dest)
587 {
588 	int i;
589 	uint32_t byte = 0;
590 
591 	/*
592 	 * Enable use of auto EEPROM access so we can avoid
593 	 * having to use the bitbang method.
594 	 */
595 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_AUTO_EEPROM);
596 
597 	/* Reset the EEPROM, load the clock period. */
598 	CSR_WRITE_4(sc, BGE_EE_ADDR,
599 	    BGE_EEADDR_RESET|BGE_EEHALFCLK(BGE_HALFCLK_384SCL));
600 	DELAY(20);
601 
602 	/* Issue the read EEPROM command. */
603 	CSR_WRITE_4(sc, BGE_EE_ADDR, BGE_EE_READCMD | addr);
604 
605 	/* Wait for completion */
606 	for(i = 0; i < BGE_TIMEOUT * 10; i++) {
607 		DELAY(10);
608 		if (CSR_READ_4(sc, BGE_EE_ADDR) & BGE_EEADDR_DONE)
609 			break;
610 	}
611 
612 	if (i == BGE_TIMEOUT) {
613 		if_printf(&sc->arpcom.ac_if, "eeprom read timed out\n");
614 		return(1);
615 	}
616 
617 	/* Get result. */
618 	byte = CSR_READ_4(sc, BGE_EE_DATA);
619 
620         *dest = (byte >> ((addr % 4) * 8)) & 0xFF;
621 
622 	return(0);
623 }
624 
625 /*
626  * Read a sequence of bytes from the EEPROM.
627  */
628 static int
629 bge_read_eeprom(struct bge_softc *sc, caddr_t dest, uint32_t off, size_t len)
630 {
631 	size_t i;
632 	int err;
633 	uint8_t byte;
634 
635 	for (byte = 0, err = 0, i = 0; i < len; i++) {
636 		err = bge_eeprom_getbyte(sc, off + i, &byte);
637 		if (err)
638 			break;
639 		*(dest + i) = byte;
640 	}
641 
642 	return(err ? 1 : 0);
643 }
644 
645 static int
646 bge_miibus_readreg(device_t dev, int phy, int reg)
647 {
648 	struct bge_softc *sc = device_get_softc(dev);
649 	uint32_t val;
650 	int i;
651 
652 	KASSERT(phy == sc->bge_phyno,
653 	    ("invalid phyno %d, should be %d", phy, sc->bge_phyno));
654 
655 	/* Clear the autopoll bit if set, otherwise may trigger PCI errors. */
656 	if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
657 		CSR_WRITE_4(sc, BGE_MI_MODE,
658 		    sc->bge_mi_mode & ~BGE_MIMODE_AUTOPOLL);
659 		DELAY(80);
660 	}
661 
662 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_READ | BGE_MICOMM_BUSY |
663 	    BGE_MIPHY(phy) | BGE_MIREG(reg));
664 
665 	/* Poll for the PHY register access to complete. */
666 	for (i = 0; i < BGE_TIMEOUT; i++) {
667 		DELAY(10);
668 		val = CSR_READ_4(sc, BGE_MI_COMM);
669 		if ((val & BGE_MICOMM_BUSY) == 0) {
670 			DELAY(5);
671 			val = CSR_READ_4(sc, BGE_MI_COMM);
672 			break;
673 		}
674 	}
675 	if (i == BGE_TIMEOUT) {
676 		if_printf(&sc->arpcom.ac_if, "PHY read timed out "
677 		    "(phy %d, reg %d, val 0x%08x)\n", phy, reg, val);
678 		val = 0;
679 	}
680 
681 	/* Restore the autopoll bit if necessary. */
682 	if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
683 		CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode);
684 		DELAY(80);
685 	}
686 
687 	if (val & BGE_MICOMM_READFAIL)
688 		return 0;
689 
690 	return (val & 0xFFFF);
691 }
692 
693 static int
694 bge_miibus_writereg(device_t dev, int phy, int reg, int val)
695 {
696 	struct bge_softc *sc = device_get_softc(dev);
697 	int i;
698 
699 	KASSERT(phy == sc->bge_phyno,
700 	    ("invalid phyno %d, should be %d", phy, sc->bge_phyno));
701 
702 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
703 	    (reg == BRGPHY_MII_1000CTL || reg == BRGPHY_MII_AUXCTL))
704 	       return 0;
705 
706 	/* Clear the autopoll bit if set, otherwise may trigger PCI errors. */
707 	if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
708 		CSR_WRITE_4(sc, BGE_MI_MODE,
709 		    sc->bge_mi_mode & ~BGE_MIMODE_AUTOPOLL);
710 		DELAY(80);
711 	}
712 
713 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_WRITE | BGE_MICOMM_BUSY |
714 	    BGE_MIPHY(phy) | BGE_MIREG(reg) | val);
715 
716 	for (i = 0; i < BGE_TIMEOUT; i++) {
717 		DELAY(10);
718 		if (!(CSR_READ_4(sc, BGE_MI_COMM) & BGE_MICOMM_BUSY)) {
719 			DELAY(5);
720 			CSR_READ_4(sc, BGE_MI_COMM); /* dummy read */
721 			break;
722 		}
723 	}
724 	if (i == BGE_TIMEOUT) {
725 		if_printf(&sc->arpcom.ac_if, "PHY write timed out "
726 		    "(phy %d, reg %d, val %d)\n", phy, reg, val);
727 	}
728 
729 	/* Restore the autopoll bit if necessary. */
730 	if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
731 		CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode);
732 		DELAY(80);
733 	}
734 
735 	return 0;
736 }
737 
738 static void
739 bge_miibus_statchg(device_t dev)
740 {
741 	struct bge_softc *sc;
742 	struct mii_data *mii;
743 
744 	sc = device_get_softc(dev);
745 	mii = device_get_softc(sc->bge_miibus);
746 
747 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
748 	    (IFM_ACTIVE | IFM_AVALID)) {
749 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
750 		case IFM_10_T:
751 		case IFM_100_TX:
752 			sc->bge_link = 1;
753 			break;
754 		case IFM_1000_T:
755 		case IFM_1000_SX:
756 		case IFM_2500_SX:
757 			if (sc->bge_asicrev != BGE_ASICREV_BCM5906)
758 				sc->bge_link = 1;
759 			else
760 				sc->bge_link = 0;
761 			break;
762 		default:
763 			sc->bge_link = 0;
764 			break;
765 		}
766 	} else {
767 		sc->bge_link = 0;
768 	}
769 	if (sc->bge_link == 0)
770 		return;
771 
772 	BGE_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_PORTMODE);
773 	if (IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_T ||
774 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_SX) {
775 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_GMII);
776 	} else {
777 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_MII);
778 	}
779 
780 	if ((mii->mii_media_active & IFM_GMASK) == IFM_FDX) {
781 		BGE_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
782 	} else {
783 		BGE_SETBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
784 	}
785 }
786 
787 /*
788  * Memory management for jumbo frames.
789  */
790 static int
791 bge_alloc_jumbo_mem(struct bge_softc *sc)
792 {
793 	struct ifnet *ifp = &sc->arpcom.ac_if;
794 	struct bge_jslot *entry;
795 	uint8_t *ptr;
796 	bus_addr_t paddr;
797 	int i, error;
798 
799 	/*
800 	 * Create tag for jumbo mbufs.
801 	 * This is really a bit of a kludge. We allocate a special
802 	 * jumbo buffer pool which (thanks to the way our DMA
803 	 * memory allocation works) will consist of contiguous
804 	 * pages. This means that even though a jumbo buffer might
805 	 * be larger than a page size, we don't really need to
806 	 * map it into more than one DMA segment. However, the
807 	 * default mbuf tag will result in multi-segment mappings,
808 	 * so we have to create a special jumbo mbuf tag that
809 	 * lets us get away with mapping the jumbo buffers as
810 	 * a single segment. I think eventually the driver should
811 	 * be changed so that it uses ordinary mbufs and cluster
812 	 * buffers, i.e. jumbo frames can span multiple DMA
813 	 * descriptors. But that's a project for another day.
814 	 */
815 
816 	/*
817 	 * Create DMA stuffs for jumbo RX ring.
818 	 */
819 	error = bge_dma_block_alloc(sc, BGE_JUMBO_RX_RING_SZ,
820 				    &sc->bge_cdata.bge_rx_jumbo_ring_tag,
821 				    &sc->bge_cdata.bge_rx_jumbo_ring_map,
822 				    (void *)&sc->bge_ldata.bge_rx_jumbo_ring,
823 				    &sc->bge_ldata.bge_rx_jumbo_ring_paddr);
824 	if (error) {
825 		if_printf(ifp, "could not create jumbo RX ring\n");
826 		return error;
827 	}
828 
829 	/*
830 	 * Create DMA stuffs for jumbo buffer block.
831 	 */
832 	error = bge_dma_block_alloc(sc, BGE_JMEM,
833 				    &sc->bge_cdata.bge_jumbo_tag,
834 				    &sc->bge_cdata.bge_jumbo_map,
835 				    (void **)&sc->bge_ldata.bge_jumbo_buf,
836 				    &paddr);
837 	if (error) {
838 		if_printf(ifp, "could not create jumbo buffer\n");
839 		return error;
840 	}
841 
842 	SLIST_INIT(&sc->bge_jfree_listhead);
843 
844 	/*
845 	 * Now divide it up into 9K pieces and save the addresses
846 	 * in an array. Note that we play an evil trick here by using
847 	 * the first few bytes in the buffer to hold the the address
848 	 * of the softc structure for this interface. This is because
849 	 * bge_jfree() needs it, but it is called by the mbuf management
850 	 * code which will not pass it to us explicitly.
851 	 */
852 	for (i = 0, ptr = sc->bge_ldata.bge_jumbo_buf; i < BGE_JSLOTS; i++) {
853 		entry = &sc->bge_cdata.bge_jslots[i];
854 		entry->bge_sc = sc;
855 		entry->bge_buf = ptr;
856 		entry->bge_paddr = paddr;
857 		entry->bge_inuse = 0;
858 		entry->bge_slot = i;
859 		SLIST_INSERT_HEAD(&sc->bge_jfree_listhead, entry, jslot_link);
860 
861 		ptr += BGE_JLEN;
862 		paddr += BGE_JLEN;
863 	}
864 	return 0;
865 }
866 
867 static void
868 bge_free_jumbo_mem(struct bge_softc *sc)
869 {
870 	/* Destroy jumbo RX ring. */
871 	bge_dma_block_free(sc->bge_cdata.bge_rx_jumbo_ring_tag,
872 			   sc->bge_cdata.bge_rx_jumbo_ring_map,
873 			   sc->bge_ldata.bge_rx_jumbo_ring);
874 
875 	/* Destroy jumbo buffer block. */
876 	bge_dma_block_free(sc->bge_cdata.bge_jumbo_tag,
877 			   sc->bge_cdata.bge_jumbo_map,
878 			   sc->bge_ldata.bge_jumbo_buf);
879 }
880 
881 /*
882  * Allocate a jumbo buffer.
883  */
884 static struct bge_jslot *
885 bge_jalloc(struct bge_softc *sc)
886 {
887 	struct bge_jslot *entry;
888 
889 	lwkt_serialize_enter(&sc->bge_jslot_serializer);
890 	entry = SLIST_FIRST(&sc->bge_jfree_listhead);
891 	if (entry) {
892 		SLIST_REMOVE_HEAD(&sc->bge_jfree_listhead, jslot_link);
893 		entry->bge_inuse = 1;
894 	} else {
895 		if_printf(&sc->arpcom.ac_if, "no free jumbo buffers\n");
896 	}
897 	lwkt_serialize_exit(&sc->bge_jslot_serializer);
898 	return(entry);
899 }
900 
901 /*
902  * Adjust usage count on a jumbo buffer.
903  */
904 static void
905 bge_jref(void *arg)
906 {
907 	struct bge_jslot *entry = (struct bge_jslot *)arg;
908 	struct bge_softc *sc = entry->bge_sc;
909 
910 	if (sc == NULL)
911 		panic("bge_jref: can't find softc pointer!");
912 
913 	if (&sc->bge_cdata.bge_jslots[entry->bge_slot] != entry) {
914 		panic("bge_jref: asked to reference buffer "
915 		    "that we don't manage!");
916 	} else if (entry->bge_inuse == 0) {
917 		panic("bge_jref: buffer already free!");
918 	} else {
919 		atomic_add_int(&entry->bge_inuse, 1);
920 	}
921 }
922 
923 /*
924  * Release a jumbo buffer.
925  */
926 static void
927 bge_jfree(void *arg)
928 {
929 	struct bge_jslot *entry = (struct bge_jslot *)arg;
930 	struct bge_softc *sc = entry->bge_sc;
931 
932 	if (sc == NULL)
933 		panic("bge_jfree: can't find softc pointer!");
934 
935 	if (&sc->bge_cdata.bge_jslots[entry->bge_slot] != entry) {
936 		panic("bge_jfree: asked to free buffer that we don't manage!");
937 	} else if (entry->bge_inuse == 0) {
938 		panic("bge_jfree: buffer already free!");
939 	} else {
940 		/*
941 		 * Possible MP race to 0, use the serializer.  The atomic insn
942 		 * is still needed for races against bge_jref().
943 		 */
944 		lwkt_serialize_enter(&sc->bge_jslot_serializer);
945 		atomic_subtract_int(&entry->bge_inuse, 1);
946 		if (entry->bge_inuse == 0) {
947 			SLIST_INSERT_HEAD(&sc->bge_jfree_listhead,
948 					  entry, jslot_link);
949 		}
950 		lwkt_serialize_exit(&sc->bge_jslot_serializer);
951 	}
952 }
953 
954 
955 /*
956  * Intialize a standard receive ring descriptor.
957  */
958 static int
959 bge_newbuf_std(struct bge_softc *sc, int i, int init)
960 {
961 	struct mbuf *m_new = NULL;
962 	bus_dma_segment_t seg;
963 	bus_dmamap_t map;
964 	int error, nsegs;
965 
966 	m_new = m_getcl(init ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
967 	if (m_new == NULL)
968 		return ENOBUFS;
969 	m_new->m_len = m_new->m_pkthdr.len = MCLBYTES;
970 
971 	if ((sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) == 0)
972 		m_adj(m_new, ETHER_ALIGN);
973 
974 	error = bus_dmamap_load_mbuf_segment(sc->bge_cdata.bge_rx_mtag,
975 			sc->bge_cdata.bge_rx_tmpmap, m_new,
976 			&seg, 1, &nsegs, BUS_DMA_NOWAIT);
977 	if (error) {
978 		m_freem(m_new);
979 		return error;
980 	}
981 
982 	if (!init) {
983 		bus_dmamap_sync(sc->bge_cdata.bge_rx_mtag,
984 				sc->bge_cdata.bge_rx_std_dmamap[i],
985 				BUS_DMASYNC_POSTREAD);
986 		bus_dmamap_unload(sc->bge_cdata.bge_rx_mtag,
987 			sc->bge_cdata.bge_rx_std_dmamap[i]);
988 	}
989 
990 	map = sc->bge_cdata.bge_rx_tmpmap;
991 	sc->bge_cdata.bge_rx_tmpmap = sc->bge_cdata.bge_rx_std_dmamap[i];
992 	sc->bge_cdata.bge_rx_std_dmamap[i] = map;
993 
994 	sc->bge_cdata.bge_rx_std_chain[i].bge_mbuf = m_new;
995 	sc->bge_cdata.bge_rx_std_chain[i].bge_paddr = seg.ds_addr;
996 
997 	bge_setup_rxdesc_std(sc, i);
998 	return 0;
999 }
1000 
1001 static void
1002 bge_setup_rxdesc_std(struct bge_softc *sc, int i)
1003 {
1004 	struct bge_rxchain *rc;
1005 	struct bge_rx_bd *r;
1006 
1007 	rc = &sc->bge_cdata.bge_rx_std_chain[i];
1008 	r = &sc->bge_ldata.bge_rx_std_ring[i];
1009 
1010 	r->bge_addr.bge_addr_lo = BGE_ADDR_LO(rc->bge_paddr);
1011 	r->bge_addr.bge_addr_hi = BGE_ADDR_HI(rc->bge_paddr);
1012 	r->bge_len = rc->bge_mbuf->m_len;
1013 	r->bge_idx = i;
1014 	r->bge_flags = BGE_RXBDFLAG_END;
1015 }
1016 
1017 /*
1018  * Initialize a jumbo receive ring descriptor. This allocates
1019  * a jumbo buffer from the pool managed internally by the driver.
1020  */
1021 static int
1022 bge_newbuf_jumbo(struct bge_softc *sc, int i, int init)
1023 {
1024 	struct mbuf *m_new = NULL;
1025 	struct bge_jslot *buf;
1026 	bus_addr_t paddr;
1027 
1028 	/* Allocate the mbuf. */
1029 	MGETHDR(m_new, init ? MB_WAIT : MB_DONTWAIT, MT_DATA);
1030 	if (m_new == NULL)
1031 		return ENOBUFS;
1032 
1033 	/* Allocate the jumbo buffer */
1034 	buf = bge_jalloc(sc);
1035 	if (buf == NULL) {
1036 		m_freem(m_new);
1037 		return ENOBUFS;
1038 	}
1039 
1040 	/* Attach the buffer to the mbuf. */
1041 	m_new->m_ext.ext_arg = buf;
1042 	m_new->m_ext.ext_buf = buf->bge_buf;
1043 	m_new->m_ext.ext_free = bge_jfree;
1044 	m_new->m_ext.ext_ref = bge_jref;
1045 	m_new->m_ext.ext_size = BGE_JUMBO_FRAMELEN;
1046 
1047 	m_new->m_flags |= M_EXT;
1048 
1049 	m_new->m_data = m_new->m_ext.ext_buf;
1050 	m_new->m_len = m_new->m_pkthdr.len = m_new->m_ext.ext_size;
1051 
1052 	paddr = buf->bge_paddr;
1053 	if ((sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) == 0) {
1054 		m_adj(m_new, ETHER_ALIGN);
1055 		paddr += ETHER_ALIGN;
1056 	}
1057 
1058 	/* Save necessary information */
1059 	sc->bge_cdata.bge_rx_jumbo_chain[i].bge_mbuf = m_new;
1060 	sc->bge_cdata.bge_rx_jumbo_chain[i].bge_paddr = paddr;
1061 
1062 	/* Set up the descriptor. */
1063 	bge_setup_rxdesc_jumbo(sc, i);
1064 	return 0;
1065 }
1066 
1067 static void
1068 bge_setup_rxdesc_jumbo(struct bge_softc *sc, int i)
1069 {
1070 	struct bge_rx_bd *r;
1071 	struct bge_rxchain *rc;
1072 
1073 	r = &sc->bge_ldata.bge_rx_jumbo_ring[i];
1074 	rc = &sc->bge_cdata.bge_rx_jumbo_chain[i];
1075 
1076 	r->bge_addr.bge_addr_lo = BGE_ADDR_LO(rc->bge_paddr);
1077 	r->bge_addr.bge_addr_hi = BGE_ADDR_HI(rc->bge_paddr);
1078 	r->bge_len = rc->bge_mbuf->m_len;
1079 	r->bge_idx = i;
1080 	r->bge_flags = BGE_RXBDFLAG_END|BGE_RXBDFLAG_JUMBO_RING;
1081 }
1082 
1083 static int
1084 bge_init_rx_ring_std(struct bge_softc *sc)
1085 {
1086 	int i, error;
1087 
1088 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1089 		error = bge_newbuf_std(sc, i, 1);
1090 		if (error)
1091 			return error;
1092 	};
1093 
1094 	sc->bge_std = BGE_STD_RX_RING_CNT - 1;
1095 	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bge_std);
1096 
1097 	return(0);
1098 }
1099 
1100 static void
1101 bge_free_rx_ring_std(struct bge_softc *sc)
1102 {
1103 	int i;
1104 
1105 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
1106 		struct bge_rxchain *rc = &sc->bge_cdata.bge_rx_std_chain[i];
1107 
1108 		if (rc->bge_mbuf != NULL) {
1109 			bus_dmamap_unload(sc->bge_cdata.bge_rx_mtag,
1110 					  sc->bge_cdata.bge_rx_std_dmamap[i]);
1111 			m_freem(rc->bge_mbuf);
1112 			rc->bge_mbuf = NULL;
1113 		}
1114 		bzero(&sc->bge_ldata.bge_rx_std_ring[i],
1115 		    sizeof(struct bge_rx_bd));
1116 	}
1117 }
1118 
1119 static int
1120 bge_init_rx_ring_jumbo(struct bge_softc *sc)
1121 {
1122 	struct bge_rcb *rcb;
1123 	int i, error;
1124 
1125 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1126 		error = bge_newbuf_jumbo(sc, i, 1);
1127 		if (error)
1128 			return error;
1129 	};
1130 
1131 	sc->bge_jumbo = BGE_JUMBO_RX_RING_CNT - 1;
1132 
1133 	rcb = &sc->bge_ldata.bge_info.bge_jumbo_rx_rcb;
1134 	rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(0, 0);
1135 	CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1136 
1137 	bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
1138 
1139 	return(0);
1140 }
1141 
1142 static void
1143 bge_free_rx_ring_jumbo(struct bge_softc *sc)
1144 {
1145 	int i;
1146 
1147 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1148 		struct bge_rxchain *rc = &sc->bge_cdata.bge_rx_jumbo_chain[i];
1149 
1150 		if (rc->bge_mbuf != NULL) {
1151 			m_freem(rc->bge_mbuf);
1152 			rc->bge_mbuf = NULL;
1153 		}
1154 		bzero(&sc->bge_ldata.bge_rx_jumbo_ring[i],
1155 		    sizeof(struct bge_rx_bd));
1156 	}
1157 }
1158 
1159 static void
1160 bge_free_tx_ring(struct bge_softc *sc)
1161 {
1162 	int i;
1163 
1164 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
1165 		if (sc->bge_cdata.bge_tx_chain[i] != NULL) {
1166 			bus_dmamap_unload(sc->bge_cdata.bge_tx_mtag,
1167 					  sc->bge_cdata.bge_tx_dmamap[i]);
1168 			m_freem(sc->bge_cdata.bge_tx_chain[i]);
1169 			sc->bge_cdata.bge_tx_chain[i] = NULL;
1170 		}
1171 		bzero(&sc->bge_ldata.bge_tx_ring[i],
1172 		    sizeof(struct bge_tx_bd));
1173 	}
1174 }
1175 
1176 static int
1177 bge_init_tx_ring(struct bge_softc *sc)
1178 {
1179 	sc->bge_txcnt = 0;
1180 	sc->bge_tx_saved_considx = 0;
1181 	sc->bge_tx_prodidx = 0;
1182 
1183 	/* Initialize transmit producer index for host-memory send ring. */
1184 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1185 
1186 	/* 5700 b2 errata */
1187 	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
1188 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
1189 
1190 	bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1191 	/* 5700 b2 errata */
1192 	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
1193 		bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1194 
1195 	return(0);
1196 }
1197 
1198 static void
1199 bge_setmulti(struct bge_softc *sc)
1200 {
1201 	struct ifnet *ifp;
1202 	struct ifmultiaddr *ifma;
1203 	uint32_t hashes[4] = { 0, 0, 0, 0 };
1204 	int h, i;
1205 
1206 	ifp = &sc->arpcom.ac_if;
1207 
1208 	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
1209 		for (i = 0; i < 4; i++)
1210 			CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0xFFFFFFFF);
1211 		return;
1212 	}
1213 
1214 	/* First, zot all the existing filters. */
1215 	for (i = 0; i < 4; i++)
1216 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0);
1217 
1218 	/* Now program new ones. */
1219 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1220 		if (ifma->ifma_addr->sa_family != AF_LINK)
1221 			continue;
1222 		h = ether_crc32_le(
1223 		    LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1224 		    ETHER_ADDR_LEN) & 0x7f;
1225 		hashes[(h & 0x60) >> 5] |= 1 << (h & 0x1F);
1226 	}
1227 
1228 	for (i = 0; i < 4; i++)
1229 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), hashes[i]);
1230 }
1231 
1232 /*
1233  * Do endian, PCI and DMA initialization. Also check the on-board ROM
1234  * self-test results.
1235  */
1236 static int
1237 bge_chipinit(struct bge_softc *sc)
1238 {
1239 	int i;
1240 	uint32_t dma_rw_ctl;
1241 	uint16_t val;
1242 
1243 	/* Set endian type before we access any non-PCI registers. */
1244 	pci_write_config(sc->bge_dev, BGE_PCI_MISC_CTL,
1245 	    BGE_INIT | sc->bge_pci_miscctl, 4);
1246 
1247 	/* Clear the MAC control register */
1248 	CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
1249 
1250 	/*
1251 	 * Clear the MAC statistics block in the NIC's
1252 	 * internal memory.
1253 	 */
1254 	for (i = BGE_STATS_BLOCK;
1255 	    i < BGE_STATS_BLOCK_END + 1; i += sizeof(uint32_t))
1256 		BGE_MEMWIN_WRITE(sc, i, 0);
1257 
1258 	for (i = BGE_STATUS_BLOCK;
1259 	    i < BGE_STATUS_BLOCK_END + 1; i += sizeof(uint32_t))
1260 		BGE_MEMWIN_WRITE(sc, i, 0);
1261 
1262 	if (sc->bge_chiprev == BGE_CHIPREV_5704_BX) {
1263 		/*
1264 		 * Fix data corruption caused by non-qword write with WB.
1265 		 * Fix master abort in PCI mode.
1266 		 * Fix PCI latency timer.
1267 		 */
1268 		val = pci_read_config(sc->bge_dev, BGE_PCI_MSI_DATA + 2, 2);
1269 		val |= (1 << 10) | (1 << 12) | (1 << 13);
1270 		pci_write_config(sc->bge_dev, BGE_PCI_MSI_DATA + 2, val, 2);
1271 	}
1272 
1273 	/* Set up the PCI DMA control register. */
1274 	if (sc->bge_flags & BGE_FLAG_PCIE) {
1275 		/* PCI Express */
1276 		dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD |
1277 		    (0xf << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1278 		    (0x2 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1279 	} else if (sc->bge_flags & BGE_FLAG_PCIX) {
1280 		/* PCI-X bus */
1281 		if (BGE_IS_5714_FAMILY(sc)) {
1282 			dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD;
1283 			dma_rw_ctl &= ~BGE_PCIDMARWCTL_ONEDMA_ATONCE; /* XXX */
1284 			/* XXX magic values, Broadcom-supplied Linux driver */
1285 			if (sc->bge_asicrev == BGE_ASICREV_BCM5780) {
1286 				dma_rw_ctl |= (1 << 20) | (1 << 18) |
1287 				    BGE_PCIDMARWCTL_ONEDMA_ATONCE;
1288 			} else {
1289 				dma_rw_ctl |= (1 << 20) | (1 << 18) | (1 << 15);
1290 			}
1291 		} else if (sc->bge_asicrev == BGE_ASICREV_BCM5703) {
1292 			/*
1293 			 * In the BCM5703, the DMA read watermark should
1294 			 * be set to less than or equal to the maximum
1295 			 * memory read byte count of the PCI-X command
1296 			 * register.
1297 			 */
1298 			dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD |
1299 			    (0x4 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1300 			    (0x3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1301 		} else if (sc->bge_asicrev == BGE_ASICREV_BCM5704) {
1302 			/*
1303 			 * The 5704 uses a different encoding of read/write
1304 			 * watermarks.
1305 			 */
1306 			dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD |
1307 			    (0x7 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1308 			    (0x3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT);
1309 		} else {
1310 			dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD |
1311 			    (0x3 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1312 			    (0x3 << BGE_PCIDMARWCTL_WR_WAT_SHIFT) |
1313 			    (0x0F);
1314 		}
1315 
1316 		/*
1317 		 * 5703 and 5704 need ONEDMA_AT_ONCE as a workaround
1318 		 * for hardware bugs.
1319 		 */
1320 		if (sc->bge_asicrev == BGE_ASICREV_BCM5703 ||
1321 		    sc->bge_asicrev == BGE_ASICREV_BCM5704) {
1322 			uint32_t tmp;
1323 
1324 			tmp = CSR_READ_4(sc, BGE_PCI_CLKCTL) & 0x1f;
1325 			if (tmp == 0x6 || tmp == 0x7)
1326 				dma_rw_ctl |= BGE_PCIDMARWCTL_ONEDMA_ATONCE;
1327 		}
1328 	} else {
1329 		/* Conventional PCI bus */
1330 		dma_rw_ctl = BGE_PCI_READ_CMD|BGE_PCI_WRITE_CMD |
1331 		    (0x7 << BGE_PCIDMARWCTL_RD_WAT_SHIFT) |
1332 		    (0x7 << BGE_PCIDMARWCTL_WR_WAT_SHIFT) |
1333 		    (0x0F);
1334 	}
1335 
1336 	if (sc->bge_asicrev == BGE_ASICREV_BCM5703 ||
1337 	    sc->bge_asicrev == BGE_ASICREV_BCM5704 ||
1338 	    sc->bge_asicrev == BGE_ASICREV_BCM5705)
1339 		dma_rw_ctl &= ~BGE_PCIDMARWCTL_MINDMA;
1340 	pci_write_config(sc->bge_dev, BGE_PCI_DMA_RW_CTL, dma_rw_ctl, 4);
1341 
1342 	/*
1343 	 * Set up general mode register.
1344 	 */
1345 	CSR_WRITE_4(sc, BGE_MODE_CTL, BGE_DMA_SWAP_OPTIONS|
1346 	    BGE_MODECTL_MAC_ATTN_INTR|BGE_MODECTL_HOST_SEND_BDS|
1347 	    BGE_MODECTL_TX_NO_PHDR_CSUM);
1348 
1349 	/*
1350 	 * BCM5701 B5 have a bug causing data corruption when using
1351 	 * 64-bit DMA reads, which can be terminated early and then
1352 	 * completed later as 32-bit accesses, in combination with
1353 	 * certain bridges.
1354 	 */
1355 	if (sc->bge_asicrev == BGE_ASICREV_BCM5701 &&
1356 	    sc->bge_chipid == BGE_CHIPID_BCM5701_B5)
1357 		BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_FORCE_PCI32);
1358 
1359 	/*
1360 	 * Disable memory write invalidate.  Apparently it is not supported
1361 	 * properly by these devices.
1362 	 */
1363 	PCI_CLRBIT(sc->bge_dev, BGE_PCI_CMD, PCIM_CMD_MWIEN, 4);
1364 
1365 	/* Set the timer prescaler (always 66Mhz) */
1366 	CSR_WRITE_4(sc, BGE_MISC_CFG, 65 << 1/*BGE_32BITTIME_66MHZ*/);
1367 
1368 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
1369 		DELAY(40);	/* XXX */
1370 
1371 		/* Put PHY into ready state */
1372 		BGE_CLRBIT(sc, BGE_MISC_CFG, BGE_MISCCFG_EPHY_IDDQ);
1373 		CSR_READ_4(sc, BGE_MISC_CFG); /* Flush */
1374 		DELAY(40);
1375 	}
1376 
1377 	return(0);
1378 }
1379 
1380 static int
1381 bge_blockinit(struct bge_softc *sc)
1382 {
1383 	struct bge_rcb *rcb;
1384 	bus_size_t vrcb;
1385 	bge_hostaddr taddr;
1386 	uint32_t val;
1387 	int i, limit;
1388 
1389 	/*
1390 	 * Initialize the memory window pointer register so that
1391 	 * we can access the first 32K of internal NIC RAM. This will
1392 	 * allow us to set up the TX send ring RCBs and the RX return
1393 	 * ring RCBs, plus other things which live in NIC memory.
1394 	 */
1395 	CSR_WRITE_4(sc, BGE_PCI_MEMWIN_BASEADDR, 0);
1396 
1397 	/* Note: the BCM5704 has a smaller mbuf space than other chips. */
1398 
1399 	if (!BGE_IS_5705_PLUS(sc)) {
1400 		/* Configure mbuf memory pool */
1401 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_BASEADDR, BGE_BUFFPOOL_1);
1402 		if (sc->bge_asicrev == BGE_ASICREV_BCM5704)
1403 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x10000);
1404 		else
1405 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x18000);
1406 
1407 		/* Configure DMA resource pool */
1408 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_BASEADDR,
1409 		    BGE_DMA_DESCRIPTORS);
1410 		CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LEN, 0x2000);
1411 	}
1412 
1413 	/* Configure mbuf pool watermarks */
1414 	if (!BGE_IS_5705_PLUS(sc)) {
1415 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x50);
1416 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x20);
1417 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1418 	} else if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
1419 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1420 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x04);
1421 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x10);
1422 	} else {
1423 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1424 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x10);
1425 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1426 	}
1427 
1428 	/* Configure DMA resource watermarks */
1429 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LOWAT, 5);
1430 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_HIWAT, 10);
1431 
1432 	/* Enable buffer manager */
1433 	CSR_WRITE_4(sc, BGE_BMAN_MODE,
1434 	    BGE_BMANMODE_ENABLE|BGE_BMANMODE_LOMBUF_ATTN);
1435 
1436 	/* Poll for buffer manager start indication */
1437 	for (i = 0; i < BGE_TIMEOUT; i++) {
1438 		if (CSR_READ_4(sc, BGE_BMAN_MODE) & BGE_BMANMODE_ENABLE)
1439 			break;
1440 		DELAY(10);
1441 	}
1442 
1443 	if (i == BGE_TIMEOUT) {
1444 		if_printf(&sc->arpcom.ac_if,
1445 			  "buffer manager failed to start\n");
1446 		return(ENXIO);
1447 	}
1448 
1449 	/* Enable flow-through queues */
1450 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
1451 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
1452 
1453 	/* Wait until queue initialization is complete */
1454 	for (i = 0; i < BGE_TIMEOUT; i++) {
1455 		if (CSR_READ_4(sc, BGE_FTQ_RESET) == 0)
1456 			break;
1457 		DELAY(10);
1458 	}
1459 
1460 	if (i == BGE_TIMEOUT) {
1461 		if_printf(&sc->arpcom.ac_if,
1462 			  "flow-through queue init failed\n");
1463 		return(ENXIO);
1464 	}
1465 
1466 	/*
1467 	 * Summary of rings supported by the controller:
1468 	 *
1469 	 * Standard Receive Producer Ring
1470 	 * - This ring is used to feed receive buffers for "standard"
1471 	 *   sized frames (typically 1536 bytes) to the controller.
1472 	 *
1473 	 * Jumbo Receive Producer Ring
1474 	 * - This ring is used to feed receive buffers for jumbo sized
1475 	 *   frames (i.e. anything bigger than the "standard" frames)
1476 	 *   to the controller.
1477 	 *
1478 	 * Mini Receive Producer Ring
1479 	 * - This ring is used to feed receive buffers for "mini"
1480 	 *   sized frames to the controller.
1481 	 * - This feature required external memory for the controller
1482 	 *   but was never used in a production system.  Should always
1483 	 *   be disabled.
1484 	 *
1485 	 * Receive Return Ring
1486 	 * - After the controller has placed an incoming frame into a
1487 	 *   receive buffer that buffer is moved into a receive return
1488 	 *   ring.  The driver is then responsible to passing the
1489 	 *   buffer up to the stack.  Many versions of the controller
1490 	 *   support multiple RR rings.
1491 	 *
1492 	 * Send Ring
1493 	 * - This ring is used for outgoing frames.  Many versions of
1494 	 *   the controller support multiple send rings.
1495 	 */
1496 
1497 	/* Initialize the standard receive producer ring control block. */
1498 	rcb = &sc->bge_ldata.bge_info.bge_std_rx_rcb;
1499 	rcb->bge_hostaddr.bge_addr_lo =
1500 	    BGE_ADDR_LO(sc->bge_ldata.bge_rx_std_ring_paddr);
1501 	rcb->bge_hostaddr.bge_addr_hi =
1502 	    BGE_ADDR_HI(sc->bge_ldata.bge_rx_std_ring_paddr);
1503 	if (BGE_IS_5705_PLUS(sc)) {
1504 		/*
1505 		 * Bits 31-16: Programmable ring size (512, 256, 128, 64, 32)
1506 		 * Bits 15-2 : Reserved (should be 0)
1507 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
1508 		 * Bit 0     : Reserved
1509 		 */
1510 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(512, 0);
1511 	} else {
1512 		/*
1513 		 * Ring size is always XXX entries
1514 		 * Bits 31-16: Maximum RX frame size
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 =
1520 		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN, 0);
1521 	}
1522 	rcb->bge_nicaddr = BGE_STD_RX_RINGS;
1523 	/* Write the standard receive producer ring control block. */
1524 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_HI, rcb->bge_hostaddr.bge_addr_hi);
1525 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_LO, rcb->bge_hostaddr.bge_addr_lo);
1526 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1527 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_NICADDR, rcb->bge_nicaddr);
1528 	/* Reset the standard receive producer ring producer index. */
1529 	bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, 0);
1530 
1531 	/*
1532 	 * Initialize the jumbo RX producer ring control
1533 	 * block.  We set the 'ring disabled' bit in the
1534 	 * flags field until we're actually ready to start
1535 	 * using this ring (i.e. once we set the MTU
1536 	 * high enough to require it).
1537 	 */
1538 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
1539 		rcb = &sc->bge_ldata.bge_info.bge_jumbo_rx_rcb;
1540 		/* Get the jumbo receive producer ring RCB parameters. */
1541 		rcb->bge_hostaddr.bge_addr_lo =
1542 		    BGE_ADDR_LO(sc->bge_ldata.bge_rx_jumbo_ring_paddr);
1543 		rcb->bge_hostaddr.bge_addr_hi =
1544 		    BGE_ADDR_HI(sc->bge_ldata.bge_rx_jumbo_ring_paddr);
1545 		rcb->bge_maxlen_flags =
1546 		    BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN,
1547 		    BGE_RCB_FLAG_RING_DISABLED);
1548 		rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS;
1549 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_HI,
1550 		    rcb->bge_hostaddr.bge_addr_hi);
1551 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_LO,
1552 		    rcb->bge_hostaddr.bge_addr_lo);
1553 		/* Program the jumbo receive producer ring RCB parameters. */
1554 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS,
1555 		    rcb->bge_maxlen_flags);
1556 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_NICADDR, rcb->bge_nicaddr);
1557 		/* Reset the jumbo receive producer ring producer index. */
1558 		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, 0);
1559 	}
1560 
1561 	/* Disable the mini receive producer ring RCB. */
1562 	if (BGE_IS_5700_FAMILY(sc)) {
1563 		rcb = &sc->bge_ldata.bge_info.bge_mini_rx_rcb;
1564 		rcb->bge_maxlen_flags =
1565 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED);
1566 		CSR_WRITE_4(sc, BGE_RX_MINI_RCB_MAXLEN_FLAGS,
1567 		    rcb->bge_maxlen_flags);
1568 		/* Reset the mini receive producer ring producer index. */
1569 		bge_writembx(sc, BGE_MBX_RX_MINI_PROD_LO, 0);
1570 	}
1571 
1572 	/* Choose de-pipeline mode for BCM5906 A0, A1 and A2. */
1573 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906 &&
1574 	    (sc->bge_chipid == BGE_CHIPID_BCM5906_A0 ||
1575 	     sc->bge_chipid == BGE_CHIPID_BCM5906_A1 ||
1576 	     sc->bge_chipid == BGE_CHIPID_BCM5906_A2)) {
1577 		CSR_WRITE_4(sc, BGE_ISO_PKT_TX,
1578 		    (CSR_READ_4(sc, BGE_ISO_PKT_TX) & ~3) | 2);
1579 	}
1580 
1581 	/*
1582 	 * The BD ring replenish thresholds control how often the
1583 	 * hardware fetches new BD's from the producer rings in host
1584 	 * memory.  Setting the value too low on a busy system can
1585 	 * starve the hardware and recue the throughpout.
1586 	 *
1587 	 * Set the BD ring replentish thresholds. The recommended
1588 	 * values are 1/8th the number of descriptors allocated to
1589 	 * each ring.
1590 	 */
1591 	if (BGE_IS_5705_PLUS(sc))
1592 		val = 8;
1593 	else
1594 		val = BGE_STD_RX_RING_CNT / 8;
1595 	CSR_WRITE_4(sc, BGE_RBDI_STD_REPL_THRESH, val);
1596 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
1597 		CSR_WRITE_4(sc, BGE_RBDI_JUMBO_REPL_THRESH,
1598 		    BGE_JUMBO_RX_RING_CNT/8);
1599 	}
1600 
1601 	/*
1602 	 * Disable all send rings by setting the 'ring disabled' bit
1603 	 * in the flags field of all the TX send ring control blocks,
1604 	 * located in NIC memory.
1605 	 */
1606 	if (!BGE_IS_5705_PLUS(sc)) {
1607 		/* 5700 to 5704 had 16 send rings. */
1608 		limit = BGE_TX_RINGS_EXTSSRAM_MAX;
1609 	} else {
1610 		limit = 1;
1611 	}
1612 	vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1613 	for (i = 0; i < limit; i++) {
1614 		RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1615 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED));
1616 		RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1617 		vrcb += sizeof(struct bge_rcb);
1618 	}
1619 
1620 	/* Configure send ring RCB 0 (we use only the first ring) */
1621 	vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1622 	BGE_HOSTADDR(taddr, sc->bge_ldata.bge_tx_ring_paddr);
1623 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1624 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1625 	RCB_WRITE_4(sc, vrcb, bge_nicaddr,
1626 	    BGE_NIC_TXRING_ADDR(0, BGE_TX_RING_CNT));
1627 	RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1628 	    BGE_RCB_MAXLEN_FLAGS(BGE_TX_RING_CNT, 0));
1629 
1630 	/*
1631 	 * Disable all receive return rings by setting the
1632 	 * 'ring diabled' bit in the flags field of all the receive
1633 	 * return ring control blocks, located in NIC memory.
1634 	 */
1635 	if (!BGE_IS_5705_PLUS(sc))
1636 		limit = BGE_RX_RINGS_MAX;
1637 	else if (sc->bge_asicrev == BGE_ASICREV_BCM5755)
1638 		limit = 4;
1639 	else
1640 		limit = 1;
1641 	/* Disable all receive return rings. */
1642 	vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1643 	for (i = 0; i < limit; i++) {
1644 		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, 0);
1645 		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, 0);
1646 		RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1647 		    BGE_RCB_FLAG_RING_DISABLED);
1648 		RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1649 		bge_writembx(sc, BGE_MBX_RX_CONS0_LO +
1650 		    (i * (sizeof(uint64_t))), 0);
1651 		vrcb += sizeof(struct bge_rcb);
1652 	}
1653 
1654 	/*
1655 	 * Set up receive return ring 0.  Note that the NIC address
1656 	 * for RX return rings is 0x0.  The return rings live entirely
1657 	 * within the host, so the nicaddr field in the RCB isn't used.
1658 	 */
1659 	vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1660 	BGE_HOSTADDR(taddr, sc->bge_ldata.bge_rx_return_ring_paddr);
1661 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1662 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1663 	RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1664 	RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1665 	    BGE_RCB_MAXLEN_FLAGS(sc->bge_return_ring_cnt, 0));
1666 
1667 	/* Set random backoff seed for TX */
1668 	CSR_WRITE_4(sc, BGE_TX_RANDOM_BACKOFF,
1669 	    sc->arpcom.ac_enaddr[0] + sc->arpcom.ac_enaddr[1] +
1670 	    sc->arpcom.ac_enaddr[2] + sc->arpcom.ac_enaddr[3] +
1671 	    sc->arpcom.ac_enaddr[4] + sc->arpcom.ac_enaddr[5] +
1672 	    BGE_TX_BACKOFF_SEED_MASK);
1673 
1674 	/* Set inter-packet gap */
1675 	CSR_WRITE_4(sc, BGE_TX_LENGTHS, 0x2620);
1676 
1677 	/*
1678 	 * Specify which ring to use for packets that don't match
1679 	 * any RX rules.
1680 	 */
1681 	CSR_WRITE_4(sc, BGE_RX_RULES_CFG, 0x08);
1682 
1683 	/*
1684 	 * Configure number of RX lists. One interrupt distribution
1685 	 * list, sixteen active lists, one bad frames class.
1686 	 */
1687 	CSR_WRITE_4(sc, BGE_RXLP_CFG, 0x181);
1688 
1689 	/* Inialize RX list placement stats mask. */
1690 	CSR_WRITE_4(sc, BGE_RXLP_STATS_ENABLE_MASK, 0x007FFFFF);
1691 	CSR_WRITE_4(sc, BGE_RXLP_STATS_CTL, 0x1);
1692 
1693 	/* Disable host coalescing until we get it set up */
1694 	CSR_WRITE_4(sc, BGE_HCC_MODE, 0x00000000);
1695 
1696 	/* Poll to make sure it's shut down. */
1697 	for (i = 0; i < BGE_TIMEOUT; i++) {
1698 		if (!(CSR_READ_4(sc, BGE_HCC_MODE) & BGE_HCCMODE_ENABLE))
1699 			break;
1700 		DELAY(10);
1701 	}
1702 
1703 	if (i == BGE_TIMEOUT) {
1704 		if_printf(&sc->arpcom.ac_if,
1705 			  "host coalescing engine failed to idle\n");
1706 		return(ENXIO);
1707 	}
1708 
1709 	/* Set up host coalescing defaults */
1710 	CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, sc->bge_rx_coal_ticks);
1711 	CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS, sc->bge_tx_coal_ticks);
1712 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, sc->bge_rx_coal_bds);
1713 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS, sc->bge_tx_coal_bds);
1714 	if (!BGE_IS_5705_PLUS(sc)) {
1715 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS_INT,
1716 		    sc->bge_rx_coal_ticks_int);
1717 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS_INT,
1718 		    sc->bge_tx_coal_ticks_int);
1719 	}
1720 	/*
1721 	 * NOTE:
1722 	 * The datasheet (57XX-PG105-R) says BCM5705+ do not
1723 	 * have following two registers; obviously it is wrong.
1724 	 */
1725 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT, sc->bge_rx_coal_bds_int);
1726 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT, sc->bge_tx_coal_bds_int);
1727 
1728 	/* Set up address of statistics block */
1729 	if (!BGE_IS_5705_PLUS(sc)) {
1730 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_HI,
1731 		    BGE_ADDR_HI(sc->bge_ldata.bge_stats_paddr));
1732 		CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_LO,
1733 		    BGE_ADDR_LO(sc->bge_ldata.bge_stats_paddr));
1734 
1735 		CSR_WRITE_4(sc, BGE_HCC_STATS_BASEADDR, BGE_STATS_BLOCK);
1736 		CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_BASEADDR, BGE_STATUS_BLOCK);
1737 		CSR_WRITE_4(sc, BGE_HCC_STATS_TICKS, sc->bge_stat_ticks);
1738 	}
1739 
1740 	/* Set up address of status block */
1741 	bzero(sc->bge_ldata.bge_status_block, BGE_STATUS_BLK_SZ);
1742 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_HI,
1743 	    BGE_ADDR_HI(sc->bge_ldata.bge_status_block_paddr));
1744 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_LO,
1745 	    BGE_ADDR_LO(sc->bge_ldata.bge_status_block_paddr));
1746 
1747 	/*
1748 	 * Set up status block partail update size.
1749 	 *
1750 	 * Because only single TX ring, RX produce ring and Rx return ring
1751 	 * are used, ask device to update only minimum part of status block
1752 	 * except for BCM5700 AX/BX, whose status block partial update size
1753 	 * can't be configured.
1754 	 */
1755 	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
1756 	    sc->bge_chipid != BGE_CHIPID_BCM5700_C0) {
1757 		/* XXX Actually reserved on BCM5700 AX/BX */
1758 		val = BGE_STATBLKSZ_FULL;
1759 	} else {
1760 		val = BGE_STATBLKSZ_32BYTE;
1761 	}
1762 #if 0
1763 	/*
1764 	 * Does not seem to have visible effect in both
1765 	 * bulk data (1472B UDP datagram) and tiny data
1766 	 * (18B UDP datagram) TX tests.
1767 	 */
1768 	if (!BGE_IS_CRIPPLED(sc))
1769 		val |= BGE_HCCMODE_CLRTICK_TX;
1770 #endif
1771 
1772 	/* Turn on host coalescing state machine */
1773 	CSR_WRITE_4(sc, BGE_HCC_MODE, val | BGE_HCCMODE_ENABLE);
1774 
1775 	/* Turn on RX BD completion state machine and enable attentions */
1776 	CSR_WRITE_4(sc, BGE_RBDC_MODE,
1777 	    BGE_RBDCMODE_ENABLE|BGE_RBDCMODE_ATTN);
1778 
1779 	/* Turn on RX list placement state machine */
1780 	CSR_WRITE_4(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
1781 
1782 	/* Turn on RX list selector state machine. */
1783 	if (!BGE_IS_5705_PLUS(sc))
1784 		CSR_WRITE_4(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
1785 
1786 	val = BGE_MACMODE_TXDMA_ENB | BGE_MACMODE_RXDMA_ENB |
1787 	    BGE_MACMODE_RX_STATS_CLEAR | BGE_MACMODE_TX_STATS_CLEAR |
1788 	    BGE_MACMODE_RX_STATS_ENB | BGE_MACMODE_TX_STATS_ENB |
1789 	    BGE_MACMODE_FRMHDR_DMA_ENB;
1790 
1791 	if (sc->bge_flags & BGE_FLAG_TBI)
1792 		val |= BGE_PORTMODE_TBI;
1793 	else if (sc->bge_flags & BGE_FLAG_MII_SERDES)
1794 		val |= BGE_PORTMODE_GMII;
1795 	else
1796 		val |= BGE_PORTMODE_MII;
1797 
1798 	/* Turn on DMA, clear stats */
1799 	CSR_WRITE_4(sc, BGE_MAC_MODE, val);
1800 
1801 	/* Set misc. local control, enable interrupts on attentions */
1802 	CSR_WRITE_4(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_ONATTN);
1803 
1804 #ifdef notdef
1805 	/* Assert GPIO pins for PHY reset */
1806 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUT0|
1807 	    BGE_MLC_MISCIO_OUT1|BGE_MLC_MISCIO_OUT2);
1808 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUTEN0|
1809 	    BGE_MLC_MISCIO_OUTEN1|BGE_MLC_MISCIO_OUTEN2);
1810 #endif
1811 
1812 	/* Turn on DMA completion state machine */
1813 	if (!BGE_IS_5705_PLUS(sc))
1814 		CSR_WRITE_4(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
1815 
1816 	/* Turn on write DMA state machine */
1817 	val = BGE_WDMAMODE_ENABLE|BGE_WDMAMODE_ALL_ATTNS;
1818 	if (BGE_IS_5755_PLUS(sc)) {
1819 		/* Enable host coalescing bug fix. */
1820 		val |= BGE_WDMAMODE_STATUS_TAG_FIX;
1821 	}
1822 	if (sc->bge_asicrev == BGE_ASICREV_BCM5785) {
1823 		/* Request larger DMA burst size to get better performance. */
1824 		val |= BGE_WDMAMODE_BURST_ALL_DATA;
1825 	}
1826 	CSR_WRITE_4(sc, BGE_WDMA_MODE, val);
1827 	DELAY(40);
1828 
1829 	if (sc->bge_asicrev == BGE_ASICREV_BCM5761 ||
1830 	    sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
1831 	    sc->bge_asicrev == BGE_ASICREV_BCM5785 ||
1832 	    sc->bge_asicrev == BGE_ASICREV_BCM57780) {
1833 		/*
1834 		 * Enable fix for read DMA FIFO overruns.
1835 		 * The fix is to limit the number of RX BDs
1836 		 * the hardware would fetch at a fime.
1837 		 */
1838 		val = CSR_READ_4(sc, BGE_RDMA_RSRVCTRL);
1839 		CSR_WRITE_4(sc, BGE_RDMA_RSRVCTRL,
1840 		    val| BGE_RDMA_RSRVCTRL_FIFO_OFLW_FIX);
1841 	}
1842 
1843 	/* Turn on read DMA state machine */
1844 	val = BGE_RDMAMODE_ENABLE | BGE_RDMAMODE_ALL_ATTNS;
1845         if (sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
1846             sc->bge_asicrev == BGE_ASICREV_BCM5785 ||
1847             sc->bge_asicrev == BGE_ASICREV_BCM57780)
1848 		val |= BGE_RDMAMODE_BD_SBD_CRPT_ATTN |
1849                   BGE_RDMAMODE_MBUF_RBD_CRPT_ATTN |
1850                   BGE_RDMAMODE_MBUF_SBD_CRPT_ATTN;
1851 	if (sc->bge_flags & BGE_FLAG_PCIE)
1852 		val |= BGE_RDMAMODE_FIFO_LONG_BURST;
1853 	CSR_WRITE_4(sc, BGE_RDMA_MODE, val);
1854 	DELAY(40);
1855 
1856 	/* Turn on RX data completion state machine */
1857 	CSR_WRITE_4(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
1858 
1859 	/* Turn on RX BD initiator state machine */
1860 	CSR_WRITE_4(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
1861 
1862 	/* Turn on RX data and RX BD initiator state machine */
1863 	CSR_WRITE_4(sc, BGE_RDBDI_MODE, BGE_RDBDIMODE_ENABLE);
1864 
1865 	/* Turn on Mbuf cluster free state machine */
1866 	if (!BGE_IS_5705_PLUS(sc))
1867 		CSR_WRITE_4(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
1868 
1869 	/* Turn on send BD completion state machine */
1870 	CSR_WRITE_4(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
1871 
1872 	/* Turn on send data completion state machine */
1873 	val = BGE_SDCMODE_ENABLE;
1874 	if (sc->bge_asicrev == BGE_ASICREV_BCM5761)
1875 		val |= BGE_SDCMODE_CDELAY;
1876 	CSR_WRITE_4(sc, BGE_SDC_MODE, val);
1877 
1878 	/* Turn on send data initiator state machine */
1879 	CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
1880 
1881 	/* Turn on send BD initiator state machine */
1882 	CSR_WRITE_4(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
1883 
1884 	/* Turn on send BD selector state machine */
1885 	CSR_WRITE_4(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
1886 
1887 	CSR_WRITE_4(sc, BGE_SDI_STATS_ENABLE_MASK, 0x007FFFFF);
1888 	CSR_WRITE_4(sc, BGE_SDI_STATS_CTL,
1889 	    BGE_SDISTATSCTL_ENABLE|BGE_SDISTATSCTL_FASTER);
1890 
1891 	/* ack/clear link change events */
1892 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
1893 	    BGE_MACSTAT_CFG_CHANGED|BGE_MACSTAT_MI_COMPLETE|
1894 	    BGE_MACSTAT_LINK_CHANGED);
1895 	CSR_WRITE_4(sc, BGE_MI_STS, 0);
1896 
1897 	/*
1898 	 * Enable attention when the link has changed state for
1899 	 * devices that use auto polling.
1900 	 */
1901 	if (sc->bge_flags & BGE_FLAG_TBI) {
1902 		CSR_WRITE_4(sc, BGE_MI_STS, BGE_MISTS_LINK);
1903  	} else {
1904 		if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
1905 			CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode);
1906 			DELAY(80);
1907 		}
1908 		if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
1909 		    sc->bge_chipid != BGE_CHIPID_BCM5700_B2) {
1910 			CSR_WRITE_4(sc, BGE_MAC_EVT_ENB,
1911 			    BGE_EVTENB_MI_INTERRUPT);
1912 		}
1913 	}
1914 
1915 	/*
1916 	 * Clear any pending link state attention.
1917 	 * Otherwise some link state change events may be lost until attention
1918 	 * is cleared by bge_intr() -> bge_softc.bge_link_upd() sequence.
1919 	 * It's not necessary on newer BCM chips - perhaps enabling link
1920 	 * state change attentions implies clearing pending attention.
1921 	 */
1922 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
1923 	    BGE_MACSTAT_CFG_CHANGED|BGE_MACSTAT_MI_COMPLETE|
1924 	    BGE_MACSTAT_LINK_CHANGED);
1925 
1926 	/* Enable link state change attentions. */
1927 	BGE_SETBIT(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_LINK_CHANGED);
1928 
1929 	return(0);
1930 }
1931 
1932 /*
1933  * Probe for a Broadcom chip. Check the PCI vendor and device IDs
1934  * against our list and return its name if we find a match. Note
1935  * that since the Broadcom controller contains VPD support, we
1936  * can get the device name string from the controller itself instead
1937  * of the compiled-in string. This is a little slow, but it guarantees
1938  * we'll always announce the right product name.
1939  */
1940 static int
1941 bge_probe(device_t dev)
1942 {
1943 	const struct bge_type *t;
1944 	uint16_t product, vendor;
1945 
1946 	product = pci_get_device(dev);
1947 	vendor = pci_get_vendor(dev);
1948 
1949 	for (t = bge_devs; t->bge_name != NULL; t++) {
1950 		if (vendor == t->bge_vid && product == t->bge_did)
1951 			break;
1952 	}
1953 	if (t->bge_name == NULL)
1954 		return(ENXIO);
1955 
1956 	device_set_desc(dev, t->bge_name);
1957 	return(0);
1958 }
1959 
1960 static int
1961 bge_attach(device_t dev)
1962 {
1963 	struct ifnet *ifp;
1964 	struct bge_softc *sc;
1965 	uint32_t hwcfg = 0, misccfg;
1966 	int error = 0, rid, capmask;
1967 	uint8_t ether_addr[ETHER_ADDR_LEN];
1968 	uint16_t product, vendor;
1969 	driver_intr_t *intr_func;
1970 	uintptr_t mii_priv = 0;
1971 
1972 	sc = device_get_softc(dev);
1973 	sc->bge_dev = dev;
1974 	callout_init(&sc->bge_stat_timer);
1975 	lwkt_serialize_init(&sc->bge_jslot_serializer);
1976 
1977 #ifndef BURN_BRIDGES
1978 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1979 		uint32_t irq, mem;
1980 
1981 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
1982 		mem = pci_read_config(dev, BGE_PCI_BAR0, 4);
1983 
1984 		device_printf(dev, "chip is in D%d power mode "
1985 		    "-- setting to D0\n", pci_get_powerstate(dev));
1986 
1987 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
1988 
1989 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
1990 		pci_write_config(dev, BGE_PCI_BAR0, mem, 4);
1991 	}
1992 #endif	/* !BURN_BRIDGE */
1993 
1994 	/*
1995 	 * Map control/status registers.
1996 	 */
1997 	pci_enable_busmaster(dev);
1998 
1999 	rid = BGE_PCI_BAR0;
2000 	sc->bge_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
2001 	    RF_ACTIVE);
2002 
2003 	if (sc->bge_res == NULL) {
2004 		device_printf(dev, "couldn't map memory\n");
2005 		return ENXIO;
2006 	}
2007 
2008 	sc->bge_btag = rman_get_bustag(sc->bge_res);
2009 	sc->bge_bhandle = rman_get_bushandle(sc->bge_res);
2010 
2011 	/* Save various chip information */
2012 	sc->bge_chipid =
2013 	    pci_read_config(dev, BGE_PCI_MISC_CTL, 4) >>
2014 	    BGE_PCIMISCCTL_ASICREV_SHIFT;
2015         if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_USE_PRODID_REG)
2016 		sc->bge_chipid = pci_read_config(dev, BGE_PCI_PRODID_ASICREV, 4);
2017 	sc->bge_asicrev = BGE_ASICREV(sc->bge_chipid);
2018 	sc->bge_chiprev = BGE_CHIPREV(sc->bge_chipid);
2019 
2020 	/* Save chipset family. */
2021 	switch (sc->bge_asicrev) {
2022 	case BGE_ASICREV_BCM5755:
2023 	case BGE_ASICREV_BCM5761:
2024 	case BGE_ASICREV_BCM5784:
2025 	case BGE_ASICREV_BCM5785:
2026 	case BGE_ASICREV_BCM5787:
2027 	case BGE_ASICREV_BCM57780:
2028 	    sc->bge_flags |= BGE_FLAG_5755_PLUS | BGE_FLAG_575X_PLUS |
2029 		BGE_FLAG_5705_PLUS;
2030 	    break;
2031 
2032 	case BGE_ASICREV_BCM5700:
2033 	case BGE_ASICREV_BCM5701:
2034 	case BGE_ASICREV_BCM5703:
2035 	case BGE_ASICREV_BCM5704:
2036 		sc->bge_flags |= BGE_FLAG_5700_FAMILY | BGE_FLAG_JUMBO;
2037 		break;
2038 
2039 	case BGE_ASICREV_BCM5714_A0:
2040 	case BGE_ASICREV_BCM5780:
2041 	case BGE_ASICREV_BCM5714:
2042 		sc->bge_flags |= BGE_FLAG_5714_FAMILY;
2043 		/* Fall through */
2044 
2045 	case BGE_ASICREV_BCM5750:
2046 	case BGE_ASICREV_BCM5752:
2047 	case BGE_ASICREV_BCM5906:
2048 		sc->bge_flags |= BGE_FLAG_575X_PLUS;
2049 		/* Fall through */
2050 
2051 	case BGE_ASICREV_BCM5705:
2052 		sc->bge_flags |= BGE_FLAG_5705_PLUS;
2053 		break;
2054 	}
2055 
2056 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
2057 		sc->bge_flags |= BGE_FLAG_NO_EEPROM;
2058 
2059 	misccfg = CSR_READ_4(sc, BGE_MISC_CFG) & BGE_MISCCFG_BOARD_ID_MASK;
2060 	if (sc->bge_asicrev == BGE_ASICREV_BCM5705 &&
2061 	    (misccfg == BGE_MISCCFG_BOARD_ID_5788 ||
2062 	     misccfg == BGE_MISCCFG_BOARD_ID_5788M))
2063 		sc->bge_flags |= BGE_FLAG_5788;
2064 
2065 	/* BCM5755 or higher and BCM5906 have short DMA bug. */
2066 	if (BGE_IS_5755_PLUS(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5906)
2067 		sc->bge_flags |= BGE_FLAG_SHORTDMA;
2068 
2069   	/*
2070 	 * Check if this is a PCI-X or PCI Express device.
2071   	 */
2072 	if (BGE_IS_5705_PLUS(sc)) {
2073 		if (pci_is_pcie(dev)) {
2074 			sc->bge_flags |= BGE_FLAG_PCIE;
2075 			sc->bge_pciecap = pci_get_pciecap_ptr(sc->bge_dev);
2076 			pcie_set_max_readrq(dev, PCIEM_DEVCTL_MAX_READRQ_4096);
2077 		}
2078 	} else {
2079 		/*
2080 		 * Check if the device is in PCI-X Mode.
2081 		 * (This bit is not valid on PCI Express controllers.)
2082 		 */
2083 		if ((pci_read_config(sc->bge_dev, BGE_PCI_PCISTATE, 4) &
2084 		    BGE_PCISTATE_PCI_BUSMODE) == 0) {
2085 			sc->bge_flags |= BGE_FLAG_PCIX;
2086 			sc->bge_pcixcap = pci_get_pcixcap_ptr(sc->bge_dev);
2087 			sc->bge_mbox_reorder = device_getenv_int(sc->bge_dev,
2088 			    "mbox_reorder", 0);
2089 		}
2090  	}
2091 	device_printf(dev, "CHIP ID 0x%08x; "
2092 		      "ASIC REV 0x%02x; CHIP REV 0x%02x; %s\n",
2093 		      sc->bge_chipid, sc->bge_asicrev, sc->bge_chiprev,
2094 		      (sc->bge_flags & BGE_FLAG_PCIX) ? "PCI-X"
2095 		      : ((sc->bge_flags & BGE_FLAG_PCIE) ?
2096 			"PCI-E" : "PCI"));
2097 
2098 	/*
2099 	 * The 40bit DMA bug applies to the 5714/5715 controllers and is
2100 	 * not actually a MAC controller bug but an issue with the embedded
2101 	 * PCIe to PCI-X bridge in the device. Use 40bit DMA workaround.
2102 	 */
2103 	if (BGE_IS_5714_FAMILY(sc) && (sc->bge_flags & BGE_FLAG_PCIX))
2104 		sc->bge_flags |= BGE_FLAG_MAXADDR_40BIT;
2105 
2106 	/* Identify the chips that use an CPMU. */
2107 	if (sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
2108 	    sc->bge_asicrev == BGE_ASICREV_BCM5761 ||
2109 	    sc->bge_asicrev == BGE_ASICREV_BCM5785 ||
2110 	    sc->bge_asicrev == BGE_ASICREV_BCM57780)
2111 		sc->bge_flags |= BGE_FLAG_CPMU;
2112 
2113 	/*
2114 	 * When using the BCM5701 in PCI-X mode, data corruption has
2115 	 * been observed in the first few bytes of some received packets.
2116 	 * Aligning the packet buffer in memory eliminates the corruption.
2117 	 * Unfortunately, this misaligns the packet payloads.  On platforms
2118 	 * which do not support unaligned accesses, we will realign the
2119 	 * payloads by copying the received packets.
2120 	 */
2121 	if (sc->bge_asicrev == BGE_ASICREV_BCM5701 &&
2122 	    (sc->bge_flags & BGE_FLAG_PCIX))
2123 		sc->bge_flags |= BGE_FLAG_RX_ALIGNBUG;
2124 
2125 	if (!BGE_IS_CRIPPLED(sc)) {
2126 		if (device_getenv_int(dev, "status_tag", 1)) {
2127 			sc->bge_flags |= BGE_FLAG_STATUS_TAG;
2128 			sc->bge_pci_miscctl = BGE_PCIMISCCTL_TAGGED_STATUS;
2129 			if (bootverbose)
2130 				device_printf(dev, "enable status tag\n");
2131 		}
2132 	}
2133 
2134 	/*
2135 	 * Set various PHY quirk flags.
2136 	 */
2137 	product = pci_get_device(dev);
2138 	vendor = pci_get_vendor(dev);
2139 
2140 	if ((sc->bge_asicrev == BGE_ASICREV_BCM5700 ||
2141 	     sc->bge_asicrev == BGE_ASICREV_BCM5701) &&
2142 	    pci_get_subvendor(dev) == PCI_VENDOR_DELL)
2143 		mii_priv |= BRGPHY_FLAG_NO_3LED;
2144 
2145 	capmask = MII_CAPMASK_DEFAULT;
2146 	if ((sc->bge_asicrev == BGE_ASICREV_BCM5703 &&
2147 	     (misccfg == 0x4000 || misccfg == 0x8000)) ||
2148 	    (sc->bge_asicrev == BGE_ASICREV_BCM5705 &&
2149 	     vendor == PCI_VENDOR_BROADCOM &&
2150 	     (product == PCI_PRODUCT_BROADCOM_BCM5901 ||
2151 	      product == PCI_PRODUCT_BROADCOM_BCM5901A2 ||
2152 	      product == PCI_PRODUCT_BROADCOM_BCM5705F)) ||
2153 	    (vendor == PCI_VENDOR_BROADCOM &&
2154 	     (product == PCI_PRODUCT_BROADCOM_BCM5751F ||
2155 	      product == PCI_PRODUCT_BROADCOM_BCM5753F ||
2156 	      product == PCI_PRODUCT_BROADCOM_BCM5787F)) ||
2157 	    product == PCI_PRODUCT_BROADCOM_BCM57790 ||
2158 	    sc->bge_asicrev == BGE_ASICREV_BCM5906) {
2159 		/* 10/100 only */
2160 		capmask &= ~BMSR_EXTSTAT;
2161 	}
2162 
2163 	mii_priv |= BRGPHY_FLAG_WIRESPEED;
2164 	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 ||
2165 	    (sc->bge_asicrev == BGE_ASICREV_BCM5705 &&
2166 	     (sc->bge_chipid != BGE_CHIPID_BCM5705_A0 &&
2167 	      sc->bge_chipid != BGE_CHIPID_BCM5705_A1)) ||
2168 	    sc->bge_asicrev == BGE_ASICREV_BCM5906)
2169 		mii_priv &= ~BRGPHY_FLAG_WIRESPEED;
2170 
2171 	if (sc->bge_chipid == BGE_CHIPID_BCM5701_A0 ||
2172 	    sc->bge_chipid == BGE_CHIPID_BCM5701_B0)
2173 		mii_priv |= BRGPHY_FLAG_CRC_BUG;
2174 
2175 	if (sc->bge_chiprev == BGE_CHIPREV_5703_AX ||
2176 	    sc->bge_chiprev == BGE_CHIPREV_5704_AX)
2177 		mii_priv |= BRGPHY_FLAG_ADC_BUG;
2178 
2179 	if (sc->bge_chipid == BGE_CHIPID_BCM5704_A0)
2180 		mii_priv |= BRGPHY_FLAG_5704_A0;
2181 
2182 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
2183 		mii_priv |= BRGPHY_FLAG_5906;
2184 
2185 	if (BGE_IS_5705_PLUS(sc) &&
2186 	    sc->bge_asicrev != BGE_ASICREV_BCM5906 &&
2187 	    /* sc->bge_asicrev != BGE_ASICREV_BCM5717 && */
2188 	    sc->bge_asicrev != BGE_ASICREV_BCM5785 &&
2189 	    /* sc->bge_asicrev != BGE_ASICREV_BCM57765 && */
2190 	    sc->bge_asicrev != BGE_ASICREV_BCM57780) {
2191 		if (sc->bge_asicrev == BGE_ASICREV_BCM5755 ||
2192 		    sc->bge_asicrev == BGE_ASICREV_BCM5761 ||
2193 		    sc->bge_asicrev == BGE_ASICREV_BCM5784 ||
2194 		    sc->bge_asicrev == BGE_ASICREV_BCM5787) {
2195 			if (product != PCI_PRODUCT_BROADCOM_BCM5722 &&
2196 			    product != PCI_PRODUCT_BROADCOM_BCM5756)
2197 				mii_priv |= BRGPHY_FLAG_JITTER_BUG;
2198 			if (product == PCI_PRODUCT_BROADCOM_BCM5755M)
2199 				mii_priv |= BRGPHY_FLAG_ADJUST_TRIM;
2200 		} else {
2201 			mii_priv |= BRGPHY_FLAG_BER_BUG;
2202 		}
2203 	}
2204 
2205 	/* Allocate interrupt */
2206 	rid = 0;
2207 	sc->bge_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
2208 	    RF_SHAREABLE | RF_ACTIVE);
2209 	if (sc->bge_irq == NULL) {
2210 		device_printf(dev, "couldn't map interrupt\n");
2211 		error = ENXIO;
2212 		goto fail;
2213 	}
2214 
2215 	/* Initialize if_name earlier, so if_printf could be used */
2216 	ifp = &sc->arpcom.ac_if;
2217 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
2218 
2219 	/* Try to reset the chip. */
2220 	bge_reset(sc);
2221 
2222 	if (bge_chipinit(sc)) {
2223 		device_printf(dev, "chip initialization failed\n");
2224 		error = ENXIO;
2225 		goto fail;
2226 	}
2227 
2228 	/*
2229 	 * Get station address
2230 	 */
2231 	error = bge_get_eaddr(sc, ether_addr);
2232 	if (error) {
2233 		device_printf(dev, "failed to read station address\n");
2234 		goto fail;
2235 	}
2236 
2237 	/* 5705/5750 limits RX return ring to 512 entries. */
2238 	if (BGE_IS_5705_PLUS(sc))
2239 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT_5705;
2240 	else
2241 		sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT;
2242 
2243 	error = bge_dma_alloc(sc);
2244 	if (error)
2245 		goto fail;
2246 
2247 	/* Set default tuneable values. */
2248 	sc->bge_stat_ticks = BGE_TICKS_PER_SEC;
2249 	sc->bge_rx_coal_ticks = BGE_RX_COAL_TICKS_DEF;
2250 	sc->bge_tx_coal_ticks = BGE_TX_COAL_TICKS_DEF;
2251 	sc->bge_rx_coal_bds = BGE_RX_COAL_BDS_DEF;
2252 	sc->bge_tx_coal_bds = BGE_TX_COAL_BDS_DEF;
2253 	if (sc->bge_flags & BGE_FLAG_STATUS_TAG) {
2254 		sc->bge_rx_coal_ticks_int = BGE_RX_COAL_TICKS_DEF;
2255 		sc->bge_tx_coal_ticks_int = BGE_TX_COAL_TICKS_DEF;
2256 		sc->bge_rx_coal_bds_int = BGE_RX_COAL_BDS_DEF;
2257 		sc->bge_tx_coal_bds_int = BGE_TX_COAL_BDS_DEF;
2258 	} else {
2259 		sc->bge_rx_coal_ticks_int = BGE_RX_COAL_TICKS_MIN;
2260 		sc->bge_tx_coal_ticks_int = BGE_TX_COAL_TICKS_MIN;
2261 		sc->bge_rx_coal_bds_int = BGE_RX_COAL_BDS_MIN;
2262 		sc->bge_tx_coal_bds_int = BGE_TX_COAL_BDS_MIN;
2263 	}
2264 
2265 	/* Set up ifnet structure */
2266 	ifp->if_softc = sc;
2267 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2268 	ifp->if_ioctl = bge_ioctl;
2269 	ifp->if_start = bge_start;
2270 #ifdef DEVICE_POLLING
2271 	ifp->if_poll = bge_poll;
2272 #endif
2273 	ifp->if_watchdog = bge_watchdog;
2274 	ifp->if_init = bge_init;
2275 	ifp->if_mtu = ETHERMTU;
2276 	ifp->if_capabilities = IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU;
2277 	ifq_set_maxlen(&ifp->if_snd, BGE_TX_RING_CNT - 1);
2278 	ifq_set_ready(&ifp->if_snd);
2279 
2280 	/*
2281 	 * 5700 B0 chips do not support checksumming correctly due
2282 	 * to hardware bugs.
2283 	 */
2284 	if (sc->bge_chipid != BGE_CHIPID_BCM5700_B0) {
2285 		ifp->if_capabilities |= IFCAP_HWCSUM;
2286 		ifp->if_hwassist = BGE_CSUM_FEATURES;
2287 	}
2288 	ifp->if_capenable = ifp->if_capabilities;
2289 
2290 	/*
2291 	 * Figure out what sort of media we have by checking the
2292 	 * hardware config word in the first 32k of NIC internal memory,
2293 	 * or fall back to examining the EEPROM if necessary.
2294 	 * Note: on some BCM5700 cards, this value appears to be unset.
2295 	 * If that's the case, we have to rely on identifying the NIC
2296 	 * by its PCI subsystem ID, as we do below for the SysKonnect
2297 	 * SK-9D41.
2298 	 */
2299 	if (bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_SIG) == BGE_MAGIC_NUMBER) {
2300 		hwcfg = bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_NICCFG);
2301 	} else {
2302 		if (bge_read_eeprom(sc, (caddr_t)&hwcfg, BGE_EE_HWCFG_OFFSET,
2303 				    sizeof(hwcfg))) {
2304 			device_printf(dev, "failed to read EEPROM\n");
2305 			error = ENXIO;
2306 			goto fail;
2307 		}
2308 		hwcfg = ntohl(hwcfg);
2309 	}
2310 
2311 	/* The SysKonnect SK-9D41 is a 1000baseSX card. */
2312 	if (pci_get_subvendor(dev) == PCI_PRODUCT_SCHNEIDERKOCH_SK_9D41 ||
2313 	    (hwcfg & BGE_HWCFG_MEDIA) == BGE_MEDIA_FIBER) {
2314 		if (BGE_IS_5714_FAMILY(sc))
2315 			sc->bge_flags |= BGE_FLAG_MII_SERDES;
2316 		else
2317 			sc->bge_flags |= BGE_FLAG_TBI;
2318 	}
2319 
2320 	/* Setup MI MODE */
2321 	if (sc->bge_flags & BGE_FLAG_CPMU)
2322 		sc->bge_mi_mode = BGE_MIMODE_500KHZ_CONST;
2323 	else
2324 		sc->bge_mi_mode = BGE_MIMODE_BASE;
2325 	if (BGE_IS_5700_FAMILY(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5705) {
2326 		/* Enable auto polling for BCM570[0-5]. */
2327 		sc->bge_mi_mode |= BGE_MIMODE_AUTOPOLL;
2328 	}
2329 
2330 	/* Setup link status update stuffs */
2331 	if (sc->bge_asicrev == BGE_ASICREV_BCM5700 &&
2332 	    sc->bge_chipid != BGE_CHIPID_BCM5700_B2) {
2333 		sc->bge_link_upd = bge_bcm5700_link_upd;
2334 		sc->bge_link_chg = BGE_MACSTAT_MI_INTERRUPT;
2335 	} else if (sc->bge_flags & BGE_FLAG_TBI) {
2336 		sc->bge_link_upd = bge_tbi_link_upd;
2337 		sc->bge_link_chg = BGE_MACSTAT_LINK_CHANGED;
2338 	} else if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) {
2339 		sc->bge_link_upd = bge_autopoll_link_upd;
2340 		sc->bge_link_chg = BGE_MACSTAT_LINK_CHANGED;
2341 	} else {
2342 		sc->bge_link_upd = bge_copper_link_upd;
2343 		sc->bge_link_chg = BGE_MACSTAT_LINK_CHANGED;
2344 	}
2345 
2346 	/*
2347 	 * Broadcom's own driver always assumes the internal
2348 	 * PHY is at GMII address 1.  On some chips, the PHY responds
2349 	 * to accesses at all addresses, which could cause us to
2350 	 * bogusly attach the PHY 32 times at probe type.  Always
2351 	 * restricting the lookup to address 1 is simpler than
2352 	 * trying to figure out which chips revisions should be
2353 	 * special-cased.
2354 	 */
2355 	sc->bge_phyno = 1;
2356 
2357 	if (sc->bge_flags & BGE_FLAG_TBI) {
2358 		ifmedia_init(&sc->bge_ifmedia, IFM_IMASK,
2359 		    bge_ifmedia_upd, bge_ifmedia_sts);
2360 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER|IFM_1000_SX, 0, NULL);
2361 		ifmedia_add(&sc->bge_ifmedia,
2362 		    IFM_ETHER|IFM_1000_SX|IFM_FDX, 0, NULL);
2363 		ifmedia_add(&sc->bge_ifmedia, IFM_ETHER|IFM_AUTO, 0, NULL);
2364 		ifmedia_set(&sc->bge_ifmedia, IFM_ETHER|IFM_AUTO);
2365 		sc->bge_ifmedia.ifm_media = sc->bge_ifmedia.ifm_cur->ifm_media;
2366 	} else {
2367 		struct mii_probe_args mii_args;
2368 
2369 		mii_probe_args_init(&mii_args, bge_ifmedia_upd, bge_ifmedia_sts);
2370 		mii_args.mii_probemask = 1 << sc->bge_phyno;
2371 		mii_args.mii_capmask = capmask;
2372 		mii_args.mii_privtag = MII_PRIVTAG_BRGPHY;
2373 		mii_args.mii_priv = mii_priv;
2374 
2375 		error = mii_probe(dev, &sc->bge_miibus, &mii_args);
2376 		if (error) {
2377 			device_printf(dev, "MII without any PHY!\n");
2378 			goto fail;
2379 		}
2380 	}
2381 
2382 	/*
2383 	 * Create sysctl nodes.
2384 	 */
2385 	sysctl_ctx_init(&sc->bge_sysctl_ctx);
2386 	sc->bge_sysctl_tree = SYSCTL_ADD_NODE(&sc->bge_sysctl_ctx,
2387 					      SYSCTL_STATIC_CHILDREN(_hw),
2388 					      OID_AUTO,
2389 					      device_get_nameunit(dev),
2390 					      CTLFLAG_RD, 0, "");
2391 	if (sc->bge_sysctl_tree == NULL) {
2392 		device_printf(dev, "can't add sysctl node\n");
2393 		error = ENXIO;
2394 		goto fail;
2395 	}
2396 
2397 	SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2398 			SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2399 			OID_AUTO, "rx_coal_ticks",
2400 			CTLTYPE_INT | CTLFLAG_RW,
2401 			sc, 0, bge_sysctl_rx_coal_ticks, "I",
2402 			"Receive coalescing ticks (usec).");
2403 	SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2404 			SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2405 			OID_AUTO, "tx_coal_ticks",
2406 			CTLTYPE_INT | CTLFLAG_RW,
2407 			sc, 0, bge_sysctl_tx_coal_ticks, "I",
2408 			"Transmit coalescing ticks (usec).");
2409 	SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2410 			SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2411 			OID_AUTO, "rx_coal_bds",
2412 			CTLTYPE_INT | CTLFLAG_RW,
2413 			sc, 0, bge_sysctl_rx_coal_bds, "I",
2414 			"Receive max coalesced BD count.");
2415 	SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2416 			SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2417 			OID_AUTO, "tx_coal_bds",
2418 			CTLTYPE_INT | CTLFLAG_RW,
2419 			sc, 0, bge_sysctl_tx_coal_bds, "I",
2420 			"Transmit max coalesced BD count.");
2421 	if (sc->bge_flags & BGE_FLAG_PCIE) {
2422 		/*
2423 		 * A common design characteristic for many Broadcom
2424 		 * client controllers is that they only support a
2425 		 * single outstanding DMA read operation on the PCIe
2426 		 * bus. This means that it will take twice as long to
2427 		 * fetch a TX frame that is split into header and
2428 		 * payload buffers as it does to fetch a single,
2429 		 * contiguous TX frame (2 reads vs. 1 read). For these
2430 		 * controllers, coalescing buffers to reduce the number
2431 		 * of memory reads is effective way to get maximum
2432 		 * performance(about 940Mbps).  Without collapsing TX
2433 		 * buffers the maximum TCP bulk transfer performance
2434 		 * is about 850Mbps. However forcing coalescing mbufs
2435 		 * consumes a lot of CPU cycles, so leave it off by
2436 		 * default.
2437 		 */
2438 		SYSCTL_ADD_INT(&sc->bge_sysctl_ctx,
2439 			       SYSCTL_CHILDREN(sc->bge_sysctl_tree),
2440 			       OID_AUTO, "force_defrag", CTLFLAG_RW,
2441 			       &sc->bge_force_defrag, 0,
2442 			       "Force defragment on TX path");
2443 	}
2444 	if (sc->bge_flags & BGE_FLAG_STATUS_TAG) {
2445 		if (!BGE_IS_5705_PLUS(sc)) {
2446 			SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2447 			    SYSCTL_CHILDREN(sc->bge_sysctl_tree), OID_AUTO,
2448 			    "rx_coal_ticks_int", CTLTYPE_INT | CTLFLAG_RW,
2449 			    sc, 0, bge_sysctl_rx_coal_ticks_int, "I",
2450 			    "Receive coalescing ticks "
2451 			    "during interrupt (usec).");
2452 			SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2453 			    SYSCTL_CHILDREN(sc->bge_sysctl_tree), OID_AUTO,
2454 			    "tx_coal_ticks_int", CTLTYPE_INT | CTLFLAG_RW,
2455 			    sc, 0, bge_sysctl_tx_coal_ticks_int, "I",
2456 			    "Transmit coalescing ticks "
2457 			    "during interrupt (usec).");
2458 		}
2459 		SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2460 		    SYSCTL_CHILDREN(sc->bge_sysctl_tree), OID_AUTO,
2461 		    "rx_coal_bds_int", CTLTYPE_INT | CTLFLAG_RW,
2462 		    sc, 0, bge_sysctl_rx_coal_bds_int, "I",
2463 		    "Receive max coalesced BD count during interrupt.");
2464 		SYSCTL_ADD_PROC(&sc->bge_sysctl_ctx,
2465 		    SYSCTL_CHILDREN(sc->bge_sysctl_tree), OID_AUTO,
2466 		    "tx_coal_bds_int", CTLTYPE_INT | CTLFLAG_RW,
2467 		    sc, 0, bge_sysctl_tx_coal_bds_int, "I",
2468 		    "Transmit max coalesced BD count during interrupt.");
2469 	}
2470 
2471 	/*
2472 	 * Call MI attach routine.
2473 	 */
2474 	ether_ifattach(ifp, ether_addr, NULL);
2475 
2476 	if (sc->bge_flags & BGE_FLAG_STATUS_TAG)
2477 		intr_func = bge_intr_status_tag;
2478 	else
2479 		intr_func = bge_intr;
2480 
2481 	error = bus_setup_intr(dev, sc->bge_irq, INTR_MPSAFE, intr_func, sc,
2482 	    &sc->bge_intrhand, ifp->if_serializer);
2483 	if (error) {
2484 		ether_ifdetach(ifp);
2485 		device_printf(dev, "couldn't set up irq\n");
2486 		goto fail;
2487 	}
2488 
2489 	ifp->if_cpuid = rman_get_cpuid(sc->bge_irq);
2490 	KKASSERT(ifp->if_cpuid >= 0 && ifp->if_cpuid < ncpus);
2491 
2492 	return(0);
2493 fail:
2494 	bge_detach(dev);
2495 	return(error);
2496 }
2497 
2498 static int
2499 bge_detach(device_t dev)
2500 {
2501 	struct bge_softc *sc = device_get_softc(dev);
2502 
2503 	if (device_is_attached(dev)) {
2504 		struct ifnet *ifp = &sc->arpcom.ac_if;
2505 
2506 		lwkt_serialize_enter(ifp->if_serializer);
2507 		bge_stop(sc);
2508 		bge_reset(sc);
2509 		bus_teardown_intr(dev, sc->bge_irq, sc->bge_intrhand);
2510 		lwkt_serialize_exit(ifp->if_serializer);
2511 
2512 		ether_ifdetach(ifp);
2513 	}
2514 
2515 	if (sc->bge_flags & BGE_FLAG_TBI)
2516 		ifmedia_removeall(&sc->bge_ifmedia);
2517 	if (sc->bge_miibus)
2518 		device_delete_child(dev, sc->bge_miibus);
2519 	bus_generic_detach(dev);
2520 
2521         if (sc->bge_irq != NULL)
2522 		bus_release_resource(dev, SYS_RES_IRQ, 0, sc->bge_irq);
2523 
2524         if (sc->bge_res != NULL)
2525 		bus_release_resource(dev, SYS_RES_MEMORY,
2526 		    BGE_PCI_BAR0, sc->bge_res);
2527 
2528 	if (sc->bge_sysctl_tree != NULL)
2529 		sysctl_ctx_free(&sc->bge_sysctl_ctx);
2530 
2531 	bge_dma_free(sc);
2532 
2533 	return 0;
2534 }
2535 
2536 static void
2537 bge_reset(struct bge_softc *sc)
2538 {
2539 	device_t dev;
2540 	uint32_t cachesize, command, pcistate, reset;
2541 	void (*write_op)(struct bge_softc *, uint32_t, uint32_t);
2542 	int i, val = 0;
2543 
2544 	dev = sc->bge_dev;
2545 
2546 	if (BGE_IS_575X_PLUS(sc) && !BGE_IS_5714_FAMILY(sc) &&
2547 	    sc->bge_asicrev != BGE_ASICREV_BCM5906) {
2548 		if (sc->bge_flags & BGE_FLAG_PCIE)
2549 			write_op = bge_writemem_direct;
2550 		else
2551 			write_op = bge_writemem_ind;
2552 	} else {
2553 		write_op = bge_writereg_ind;
2554 	}
2555 
2556 	/* Save some important PCI state. */
2557 	cachesize = pci_read_config(dev, BGE_PCI_CACHESZ, 4);
2558 	command = pci_read_config(dev, BGE_PCI_CMD, 4);
2559 	pcistate = pci_read_config(dev, BGE_PCI_PCISTATE, 4);
2560 
2561 	pci_write_config(dev, BGE_PCI_MISC_CTL,
2562 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2563 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW|
2564 	    sc->bge_pci_miscctl, 4);
2565 
2566 	/* Disable fastboot on controllers that support it. */
2567 	if (sc->bge_asicrev == BGE_ASICREV_BCM5752 ||
2568 	    BGE_IS_5755_PLUS(sc)) {
2569 		if (bootverbose)
2570 			if_printf(&sc->arpcom.ac_if, "Disabling fastboot\n");
2571 		CSR_WRITE_4(sc, BGE_FASTBOOT_PC, 0x0);
2572 	}
2573 
2574 	/*
2575 	 * Write the magic number to SRAM at offset 0xB50.
2576 	 * When firmware finishes its initialization it will
2577 	 * write ~BGE_MAGIC_NUMBER to the same location.
2578 	 */
2579 	bge_writemem_ind(sc, BGE_SOFTWARE_GENCOMM, BGE_MAGIC_NUMBER);
2580 
2581 	reset = BGE_MISCCFG_RESET_CORE_CLOCKS|(65<<1);
2582 
2583 	/* XXX: Broadcom Linux driver. */
2584 	if (sc->bge_flags & BGE_FLAG_PCIE) {
2585 		if (CSR_READ_4(sc, 0x7e2c) == 0x60)	/* PCIE 1.0 */
2586 			CSR_WRITE_4(sc, 0x7e2c, 0x20);
2587 		if (sc->bge_chipid != BGE_CHIPID_BCM5750_A0) {
2588 			/* Prevent PCIE link training during global reset */
2589 			CSR_WRITE_4(sc, BGE_MISC_CFG, (1<<29));
2590 			reset |= (1<<29);
2591 		}
2592 	}
2593 
2594 	/*
2595 	 * Set GPHY Power Down Override to leave GPHY
2596 	 * powered up in D0 uninitialized.
2597 	 */
2598 	if (BGE_IS_5705_PLUS(sc) && (sc->bge_flags & BGE_FLAG_CPMU) == 0)
2599 		reset |= BGE_MISCCFG_GPHY_PD_OVERRIDE;
2600 
2601 	/* Issue global reset */
2602 	write_op(sc, BGE_MISC_CFG, reset);
2603 
2604 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
2605 		uint32_t status, ctrl;
2606 
2607 		status = CSR_READ_4(sc, BGE_VCPU_STATUS);
2608 		CSR_WRITE_4(sc, BGE_VCPU_STATUS,
2609 		    status | BGE_VCPU_STATUS_DRV_RESET);
2610 		ctrl = CSR_READ_4(sc, BGE_VCPU_EXT_CTRL);
2611 		CSR_WRITE_4(sc, BGE_VCPU_EXT_CTRL,
2612 		    ctrl & ~BGE_VCPU_EXT_CTRL_HALT_CPU);
2613 	}
2614 
2615 	DELAY(1000);
2616 
2617 	/* XXX: Broadcom Linux driver. */
2618 	if (sc->bge_flags & BGE_FLAG_PCIE) {
2619 		uint16_t devctl;
2620 
2621 		if (sc->bge_chipid == BGE_CHIPID_BCM5750_A0) {
2622 			uint32_t v;
2623 
2624 			DELAY(500000); /* wait for link training to complete */
2625 			v = pci_read_config(dev, 0xc4, 4);
2626 			pci_write_config(dev, 0xc4, v | (1<<15), 4);
2627 		}
2628 
2629 		/* Clear enable no snoop and disable relaxed ordering. */
2630 		devctl = pci_read_config(dev,
2631 		    sc->bge_pciecap + PCIER_DEVCTRL, 2);
2632 		devctl &= ~(PCIEM_DEVCTL_RELAX_ORDER | PCIEM_DEVCTL_NOSNOOP);
2633 		pci_write_config(dev, sc->bge_pciecap + PCIER_DEVCTRL,
2634 		    devctl, 2);
2635 
2636 		/* Clear error status. */
2637 		pci_write_config(dev, sc->bge_pciecap + PCIER_DEVSTS,
2638 		    PCIEM_DEVSTS_CORR_ERR |
2639 		    PCIEM_DEVSTS_NFATAL_ERR |
2640 		    PCIEM_DEVSTS_FATAL_ERR |
2641 		    PCIEM_DEVSTS_UNSUPP_REQ, 2);
2642 	}
2643 
2644 	/* Reset some of the PCI state that got zapped by reset */
2645 	pci_write_config(dev, BGE_PCI_MISC_CTL,
2646 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2647 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW|
2648 	    sc->bge_pci_miscctl, 4);
2649 	pci_write_config(dev, BGE_PCI_CACHESZ, cachesize, 4);
2650 	pci_write_config(dev, BGE_PCI_CMD, command, 4);
2651 	write_op(sc, BGE_MISC_CFG, (65 << 1));
2652 
2653 	/*
2654 	 * Disable PCI-X relaxed ordering to ensure status block update
2655 	 * comes first then packet buffer DMA. Otherwise driver may
2656 	 * read stale status block.
2657 	 */
2658 	if (sc->bge_flags & BGE_FLAG_PCIX) {
2659 		uint16_t devctl;
2660 
2661 		devctl = pci_read_config(dev,
2662 		    sc->bge_pcixcap + PCIXR_COMMAND, 2);
2663 		devctl &= ~PCIXM_COMMAND_ERO;
2664 		if (sc->bge_asicrev == BGE_ASICREV_BCM5703) {
2665 			devctl &= ~PCIXM_COMMAND_MAX_READ;
2666 			devctl |= PCIXM_COMMAND_MAX_READ_2048;
2667 		} else if (sc->bge_asicrev == BGE_ASICREV_BCM5704) {
2668 			devctl &= ~(PCIXM_COMMAND_MAX_SPLITS |
2669 			    PCIXM_COMMAND_MAX_READ);
2670 			devctl |= PCIXM_COMMAND_MAX_READ_2048;
2671 		}
2672 		pci_write_config(dev, sc->bge_pcixcap + PCIXR_COMMAND,
2673 		    devctl, 2);
2674 	}
2675 
2676 	/* Enable memory arbiter. */
2677 	if (BGE_IS_5714_FAMILY(sc)) {
2678 		uint32_t val;
2679 
2680 		val = CSR_READ_4(sc, BGE_MARB_MODE);
2681 		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE | val);
2682 	} else {
2683 		CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
2684 	}
2685 
2686 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906) {
2687 		for (i = 0; i < BGE_TIMEOUT; i++) {
2688 			val = CSR_READ_4(sc, BGE_VCPU_STATUS);
2689 			if (val & BGE_VCPU_STATUS_INIT_DONE)
2690 				break;
2691 			DELAY(100);
2692 		}
2693 		if (i == BGE_TIMEOUT) {
2694 			if_printf(&sc->arpcom.ac_if, "reset timed out\n");
2695 			return;
2696 		}
2697 	} else {
2698 		/*
2699 		 * Poll until we see the 1's complement of the magic number.
2700 		 * This indicates that the firmware initialization
2701 		 * is complete.
2702 		 */
2703 		for (i = 0; i < BGE_FIRMWARE_TIMEOUT; i++) {
2704 			val = bge_readmem_ind(sc, BGE_SOFTWARE_GENCOMM);
2705 			if (val == ~BGE_MAGIC_NUMBER)
2706 				break;
2707 			DELAY(10);
2708 		}
2709 		if (i == BGE_FIRMWARE_TIMEOUT) {
2710 			if_printf(&sc->arpcom.ac_if, "firmware handshake "
2711 				  "timed out, found 0x%08x\n", val);
2712 			return;
2713 		}
2714 	}
2715 
2716 	/*
2717 	 * XXX Wait for the value of the PCISTATE register to
2718 	 * return to its original pre-reset state. This is a
2719 	 * fairly good indicator of reset completion. If we don't
2720 	 * wait for the reset to fully complete, trying to read
2721 	 * from the device's non-PCI registers may yield garbage
2722 	 * results.
2723 	 */
2724 	for (i = 0; i < BGE_TIMEOUT; i++) {
2725 		if (pci_read_config(dev, BGE_PCI_PCISTATE, 4) == pcistate)
2726 			break;
2727 		DELAY(10);
2728 	}
2729 
2730 	/* Fix up byte swapping */
2731 	CSR_WRITE_4(sc, BGE_MODE_CTL, BGE_DMA_SWAP_OPTIONS |
2732 	    BGE_MODECTL_BYTESWAP_DATA);
2733 
2734 	CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
2735 
2736 	/*
2737 	 * The 5704 in TBI mode apparently needs some special
2738 	 * adjustment to insure the SERDES drive level is set
2739 	 * to 1.2V.
2740 	 */
2741 	if (sc->bge_asicrev == BGE_ASICREV_BCM5704 &&
2742 	    (sc->bge_flags & BGE_FLAG_TBI)) {
2743 		uint32_t serdescfg;
2744 
2745 		serdescfg = CSR_READ_4(sc, BGE_SERDES_CFG);
2746 		serdescfg = (serdescfg & ~0xFFF) | 0x880;
2747 		CSR_WRITE_4(sc, BGE_SERDES_CFG, serdescfg);
2748 	}
2749 
2750 	/* XXX: Broadcom Linux driver. */
2751 	if ((sc->bge_flags & BGE_FLAG_PCIE) &&
2752 	    sc->bge_chipid != BGE_CHIPID_BCM5750_A0 &&
2753 	    sc->bge_asicrev != BGE_ASICREV_BCM5785) {
2754 		uint32_t v;
2755 
2756 		/* Enable Data FIFO protection. */
2757 		v = CSR_READ_4(sc, 0x7c00);
2758 		CSR_WRITE_4(sc, 0x7c00, v | (1<<25));
2759 	}
2760 
2761 	DELAY(10000);
2762 }
2763 
2764 /*
2765  * Frame reception handling. This is called if there's a frame
2766  * on the receive return list.
2767  *
2768  * Note: we have to be able to handle two possibilities here:
2769  * 1) the frame is from the jumbo recieve ring
2770  * 2) the frame is from the standard receive ring
2771  */
2772 
2773 static void
2774 bge_rxeof(struct bge_softc *sc, uint16_t rx_prod)
2775 {
2776 	struct ifnet *ifp;
2777 	int stdcnt = 0, jumbocnt = 0;
2778 
2779 	ifp = &sc->arpcom.ac_if;
2780 
2781 	while (sc->bge_rx_saved_considx != rx_prod) {
2782 		struct bge_rx_bd	*cur_rx;
2783 		uint32_t		rxidx;
2784 		struct mbuf		*m = NULL;
2785 		uint16_t		vlan_tag = 0;
2786 		int			have_tag = 0;
2787 
2788 		cur_rx =
2789 	    &sc->bge_ldata.bge_rx_return_ring[sc->bge_rx_saved_considx];
2790 
2791 		rxidx = cur_rx->bge_idx;
2792 		BGE_INC(sc->bge_rx_saved_considx, sc->bge_return_ring_cnt);
2793 		logif(rx_pkt);
2794 
2795 		if (cur_rx->bge_flags & BGE_RXBDFLAG_VLAN_TAG) {
2796 			have_tag = 1;
2797 			vlan_tag = cur_rx->bge_vlan_tag;
2798 		}
2799 
2800 		if (cur_rx->bge_flags & BGE_RXBDFLAG_JUMBO_RING) {
2801 			BGE_INC(sc->bge_jumbo, BGE_JUMBO_RX_RING_CNT);
2802 			jumbocnt++;
2803 
2804 			if (rxidx != sc->bge_jumbo) {
2805 				ifp->if_ierrors++;
2806 				if_printf(ifp, "sw jumbo index(%d) "
2807 				    "and hw jumbo index(%d) mismatch, drop!\n",
2808 				    sc->bge_jumbo, rxidx);
2809 				bge_setup_rxdesc_jumbo(sc, rxidx);
2810 				continue;
2811 			}
2812 
2813 			m = sc->bge_cdata.bge_rx_jumbo_chain[rxidx].bge_mbuf;
2814 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
2815 				ifp->if_ierrors++;
2816 				bge_setup_rxdesc_jumbo(sc, sc->bge_jumbo);
2817 				continue;
2818 			}
2819 			if (bge_newbuf_jumbo(sc, sc->bge_jumbo, 0)) {
2820 				ifp->if_ierrors++;
2821 				bge_setup_rxdesc_jumbo(sc, sc->bge_jumbo);
2822 				continue;
2823 			}
2824 		} else {
2825 			BGE_INC(sc->bge_std, BGE_STD_RX_RING_CNT);
2826 			stdcnt++;
2827 
2828 			if (rxidx != sc->bge_std) {
2829 				ifp->if_ierrors++;
2830 				if_printf(ifp, "sw std index(%d) "
2831 				    "and hw std index(%d) mismatch, drop!\n",
2832 				    sc->bge_std, rxidx);
2833 				bge_setup_rxdesc_std(sc, rxidx);
2834 				continue;
2835 			}
2836 
2837 			m = sc->bge_cdata.bge_rx_std_chain[rxidx].bge_mbuf;
2838 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
2839 				ifp->if_ierrors++;
2840 				bge_setup_rxdesc_std(sc, sc->bge_std);
2841 				continue;
2842 			}
2843 			if (bge_newbuf_std(sc, sc->bge_std, 0)) {
2844 				ifp->if_ierrors++;
2845 				bge_setup_rxdesc_std(sc, sc->bge_std);
2846 				continue;
2847 			}
2848 		}
2849 
2850 		ifp->if_ipackets++;
2851 #if !defined(__i386__) && !defined(__x86_64__)
2852 		/*
2853 		 * The x86 allows unaligned accesses, but for other
2854 		 * platforms we must make sure the payload is aligned.
2855 		 */
2856 		if (sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) {
2857 			bcopy(m->m_data, m->m_data + ETHER_ALIGN,
2858 			    cur_rx->bge_len);
2859 			m->m_data += ETHER_ALIGN;
2860 		}
2861 #endif
2862 		m->m_pkthdr.len = m->m_len = cur_rx->bge_len - ETHER_CRC_LEN;
2863 		m->m_pkthdr.rcvif = ifp;
2864 
2865 		if (ifp->if_capenable & IFCAP_RXCSUM) {
2866 			if (cur_rx->bge_flags & BGE_RXBDFLAG_IP_CSUM) {
2867 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2868 				if ((cur_rx->bge_ip_csum ^ 0xffff) == 0)
2869 					m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2870 			}
2871 			if ((cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM) &&
2872 			    m->m_pkthdr.len >= BGE_MIN_FRAMELEN) {
2873 				m->m_pkthdr.csum_data =
2874 					cur_rx->bge_tcp_udp_csum;
2875 				m->m_pkthdr.csum_flags |=
2876 					CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2877 			}
2878 		}
2879 
2880 		/*
2881 		 * If we received a packet with a vlan tag, pass it
2882 		 * to vlan_input() instead of ether_input().
2883 		 */
2884 		if (have_tag) {
2885 			m->m_flags |= M_VLANTAG;
2886 			m->m_pkthdr.ether_vlantag = vlan_tag;
2887 			have_tag = vlan_tag = 0;
2888 		}
2889 		ifp->if_input(ifp, m);
2890 	}
2891 
2892 	bge_writembx(sc, BGE_MBX_RX_CONS0_LO, sc->bge_rx_saved_considx);
2893 	if (stdcnt)
2894 		bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bge_std);
2895 	if (jumbocnt)
2896 		bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bge_jumbo);
2897 }
2898 
2899 static void
2900 bge_txeof(struct bge_softc *sc, uint16_t tx_cons)
2901 {
2902 	struct bge_tx_bd *cur_tx = NULL;
2903 	struct ifnet *ifp;
2904 
2905 	ifp = &sc->arpcom.ac_if;
2906 
2907 	/*
2908 	 * Go through our tx ring and free mbufs for those
2909 	 * frames that have been sent.
2910 	 */
2911 	while (sc->bge_tx_saved_considx != tx_cons) {
2912 		uint32_t idx = 0;
2913 
2914 		idx = sc->bge_tx_saved_considx;
2915 		cur_tx = &sc->bge_ldata.bge_tx_ring[idx];
2916 		if (cur_tx->bge_flags & BGE_TXBDFLAG_END)
2917 			ifp->if_opackets++;
2918 		if (sc->bge_cdata.bge_tx_chain[idx] != NULL) {
2919 			bus_dmamap_unload(sc->bge_cdata.bge_tx_mtag,
2920 			    sc->bge_cdata.bge_tx_dmamap[idx]);
2921 			m_freem(sc->bge_cdata.bge_tx_chain[idx]);
2922 			sc->bge_cdata.bge_tx_chain[idx] = NULL;
2923 		}
2924 		sc->bge_txcnt--;
2925 		BGE_INC(sc->bge_tx_saved_considx, BGE_TX_RING_CNT);
2926 		logif(tx_pkt);
2927 	}
2928 
2929 	if (cur_tx != NULL &&
2930 	    (BGE_TX_RING_CNT - sc->bge_txcnt) >=
2931 	    (BGE_NSEG_RSVD + BGE_NSEG_SPARE))
2932 		ifp->if_flags &= ~IFF_OACTIVE;
2933 
2934 	if (sc->bge_txcnt == 0)
2935 		ifp->if_timer = 0;
2936 
2937 	if (!ifq_is_empty(&ifp->if_snd))
2938 		if_devstart(ifp);
2939 }
2940 
2941 #ifdef DEVICE_POLLING
2942 
2943 static void
2944 bge_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2945 {
2946 	struct bge_softc *sc = ifp->if_softc;
2947 	struct bge_status_block *sblk = sc->bge_ldata.bge_status_block;
2948 	uint16_t rx_prod, tx_cons;
2949 
2950 	switch(cmd) {
2951 	case POLL_REGISTER:
2952 		bge_disable_intr(sc);
2953 		break;
2954 	case POLL_DEREGISTER:
2955 		bge_enable_intr(sc);
2956 		break;
2957 	case POLL_AND_CHECK_STATUS:
2958 		/*
2959 		 * Process link state changes.
2960 		 */
2961 		bge_link_poll(sc);
2962 		/* Fall through */
2963 	case POLL_ONLY:
2964 		if (sc->bge_flags & BGE_FLAG_STATUS_TAG) {
2965 			sc->bge_status_tag = sblk->bge_status_tag;
2966 			/*
2967 			 * Use a load fence to ensure that status_tag
2968 			 * is saved  before rx_prod and tx_cons.
2969 			 */
2970 			cpu_lfence();
2971 		}
2972 		rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
2973 		tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
2974 		if (ifp->if_flags & IFF_RUNNING) {
2975 			rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
2976 			if (sc->bge_rx_saved_considx != rx_prod)
2977 				bge_rxeof(sc, rx_prod);
2978 
2979 			tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
2980 			if (sc->bge_tx_saved_considx != tx_cons)
2981 				bge_txeof(sc, tx_cons);
2982 		}
2983 		break;
2984 	}
2985 }
2986 
2987 #endif
2988 
2989 static void
2990 bge_intr(void *xsc)
2991 {
2992 	struct bge_softc *sc = xsc;
2993 	struct ifnet *ifp = &sc->arpcom.ac_if;
2994 
2995 	logif(intr);
2996 
2997  	/*
2998 	 * Ack the interrupt by writing something to BGE_MBX_IRQ0_LO.  Don't
2999 	 * disable interrupts by writing nonzero like we used to, since with
3000 	 * our current organization this just gives complications and
3001 	 * pessimizations for re-enabling interrupts.  We used to have races
3002 	 * instead of the necessary complications.  Disabling interrupts
3003 	 * would just reduce the chance of a status update while we are
3004 	 * running (by switching to the interrupt-mode coalescence
3005 	 * parameters), but this chance is already very low so it is more
3006 	 * efficient to get another interrupt than prevent it.
3007 	 *
3008 	 * We do the ack first to ensure another interrupt if there is a
3009 	 * status update after the ack.  We don't check for the status
3010 	 * changing later because it is more efficient to get another
3011 	 * interrupt than prevent it, not quite as above (not checking is
3012 	 * a smaller optimization than not toggling the interrupt enable,
3013 	 * since checking doesn't involve PCI accesses and toggling require
3014 	 * the status check).  So toggling would probably be a pessimization
3015 	 * even with MSI.  It would only be needed for using a task queue.
3016 	 */
3017 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 0);
3018 
3019 	/*
3020 	 * Process link state changes.
3021 	 */
3022 	bge_link_poll(sc);
3023 
3024 	if (ifp->if_flags & IFF_RUNNING) {
3025 		struct bge_status_block *sblk = sc->bge_ldata.bge_status_block;
3026 		uint16_t rx_prod, tx_cons;
3027 
3028 		rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
3029 		if (sc->bge_rx_saved_considx != rx_prod)
3030 			bge_rxeof(sc, rx_prod);
3031 
3032 		tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
3033 		if (sc->bge_tx_saved_considx != tx_cons)
3034 			bge_txeof(sc, tx_cons);
3035 	}
3036 
3037 	if (sc->bge_coal_chg)
3038 		bge_coal_change(sc);
3039 }
3040 
3041 static void
3042 bge_intr_status_tag(void *xsc)
3043 {
3044 	struct bge_softc *sc = xsc;
3045 	struct ifnet *ifp = &sc->arpcom.ac_if;
3046 	struct bge_status_block *sblk = sc->bge_ldata.bge_status_block;
3047 	uint16_t rx_prod, tx_cons;
3048 	uint32_t status;
3049 
3050 	if (sc->bge_status_tag == sblk->bge_status_tag) {
3051 		uint32_t val;
3052 
3053 		val = pci_read_config(sc->bge_dev, BGE_PCI_PCISTATE, 4);
3054 		if (val & BGE_PCISTAT_INTR_NOTACT)
3055 			return;
3056 	}
3057 
3058 	/*
3059 	 * NOTE:
3060 	 * Interrupt will have to be disabled if tagged status
3061 	 * is used, else interrupt will always be asserted on
3062 	 * certain chips (at least on BCM5750 AX/BX).
3063 	 */
3064 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
3065 
3066 	sc->bge_status_tag = sblk->bge_status_tag;
3067 	/*
3068 	 * Use a load fence to ensure that status_tag is saved
3069 	 * before rx_prod, tx_cons and status.
3070 	 */
3071 	cpu_lfence();
3072 
3073 	rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
3074 	tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
3075 	status = sblk->bge_status;
3076 
3077 	if ((status & BGE_STATFLAG_LINKSTATE_CHANGED) || sc->bge_link_evt)
3078 		bge_link_poll(sc);
3079 
3080 	if (ifp->if_flags & IFF_RUNNING) {
3081 		if (sc->bge_rx_saved_considx != rx_prod)
3082 			bge_rxeof(sc, rx_prod);
3083 
3084 		if (sc->bge_tx_saved_considx != tx_cons)
3085 			bge_txeof(sc, tx_cons);
3086 	}
3087 
3088 	bge_writembx(sc, BGE_MBX_IRQ0_LO, sc->bge_status_tag << 24);
3089 
3090 	if (sc->bge_coal_chg)
3091 		bge_coal_change(sc);
3092 }
3093 
3094 static void
3095 bge_tick(void *xsc)
3096 {
3097 	struct bge_softc *sc = xsc;
3098 	struct ifnet *ifp = &sc->arpcom.ac_if;
3099 
3100 	lwkt_serialize_enter(ifp->if_serializer);
3101 
3102 	if (BGE_IS_5705_PLUS(sc))
3103 		bge_stats_update_regs(sc);
3104 	else
3105 		bge_stats_update(sc);
3106 
3107 	if (sc->bge_flags & BGE_FLAG_TBI) {
3108 		/*
3109 		 * Since in TBI mode auto-polling can't be used we should poll
3110 		 * link status manually. Here we register pending link event
3111 		 * and trigger interrupt.
3112 		 */
3113 		sc->bge_link_evt++;
3114 		if (BGE_IS_CRIPPLED(sc))
3115 			BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
3116 		else
3117 			BGE_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
3118 	} else if (!sc->bge_link) {
3119 		mii_tick(device_get_softc(sc->bge_miibus));
3120 	}
3121 
3122 	callout_reset(&sc->bge_stat_timer, hz, bge_tick, sc);
3123 
3124 	lwkt_serialize_exit(ifp->if_serializer);
3125 }
3126 
3127 static void
3128 bge_stats_update_regs(struct bge_softc *sc)
3129 {
3130 	struct ifnet *ifp = &sc->arpcom.ac_if;
3131 	struct bge_mac_stats_regs stats;
3132 	uint32_t *s;
3133 	int i;
3134 
3135 	s = (uint32_t *)&stats;
3136 	for (i = 0; i < sizeof(struct bge_mac_stats_regs); i += 4) {
3137 		*s = CSR_READ_4(sc, BGE_RX_STATS + i);
3138 		s++;
3139 	}
3140 
3141 	ifp->if_collisions +=
3142 	   (stats.dot3StatsSingleCollisionFrames +
3143 	   stats.dot3StatsMultipleCollisionFrames +
3144 	   stats.dot3StatsExcessiveCollisions +
3145 	   stats.dot3StatsLateCollisions) -
3146 	   ifp->if_collisions;
3147 }
3148 
3149 static void
3150 bge_stats_update(struct bge_softc *sc)
3151 {
3152 	struct ifnet *ifp = &sc->arpcom.ac_if;
3153 	bus_size_t stats;
3154 
3155 	stats = BGE_MEMWIN_START + BGE_STATS_BLOCK;
3156 
3157 #define READ_STAT(sc, stats, stat)	\
3158 	CSR_READ_4(sc, stats + offsetof(struct bge_stats, stat))
3159 
3160 	ifp->if_collisions +=
3161 	   (READ_STAT(sc, stats,
3162 		txstats.dot3StatsSingleCollisionFrames.bge_addr_lo) +
3163 	    READ_STAT(sc, stats,
3164 		txstats.dot3StatsMultipleCollisionFrames.bge_addr_lo) +
3165 	    READ_STAT(sc, stats,
3166 		txstats.dot3StatsExcessiveCollisions.bge_addr_lo) +
3167 	    READ_STAT(sc, stats,
3168 		txstats.dot3StatsLateCollisions.bge_addr_lo)) -
3169 	   ifp->if_collisions;
3170 
3171 #undef READ_STAT
3172 
3173 #ifdef notdef
3174 	ifp->if_collisions +=
3175 	   (sc->bge_rdata->bge_info.bge_stats.dot3StatsSingleCollisionFrames +
3176 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsMultipleCollisionFrames +
3177 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsExcessiveCollisions +
3178 	   sc->bge_rdata->bge_info.bge_stats.dot3StatsLateCollisions) -
3179 	   ifp->if_collisions;
3180 #endif
3181 }
3182 
3183 /*
3184  * Encapsulate an mbuf chain in the tx ring  by coupling the mbuf data
3185  * pointers to descriptors.
3186  */
3187 static int
3188 bge_encap(struct bge_softc *sc, struct mbuf **m_head0, uint32_t *txidx)
3189 {
3190 	struct bge_tx_bd *d = NULL;
3191 	uint16_t csum_flags = 0;
3192 	bus_dma_segment_t segs[BGE_NSEG_NEW];
3193 	bus_dmamap_t map;
3194 	int error, maxsegs, nsegs, idx, i;
3195 	struct mbuf *m_head = *m_head0, *m_new;
3196 
3197 	if (m_head->m_pkthdr.csum_flags) {
3198 		if (m_head->m_pkthdr.csum_flags & CSUM_IP)
3199 			csum_flags |= BGE_TXBDFLAG_IP_CSUM;
3200 		if (m_head->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP))
3201 			csum_flags |= BGE_TXBDFLAG_TCP_UDP_CSUM;
3202 		if (m_head->m_flags & M_LASTFRAG)
3203 			csum_flags |= BGE_TXBDFLAG_IP_FRAG_END;
3204 		else if (m_head->m_flags & M_FRAG)
3205 			csum_flags |= BGE_TXBDFLAG_IP_FRAG;
3206 	}
3207 
3208 	idx = *txidx;
3209 	map = sc->bge_cdata.bge_tx_dmamap[idx];
3210 
3211 	maxsegs = (BGE_TX_RING_CNT - sc->bge_txcnt) - BGE_NSEG_RSVD;
3212 	KASSERT(maxsegs >= BGE_NSEG_SPARE,
3213 		("not enough segments %d", maxsegs));
3214 
3215 	if (maxsegs > BGE_NSEG_NEW)
3216 		maxsegs = BGE_NSEG_NEW;
3217 
3218 	/*
3219 	 * Pad outbound frame to BGE_MIN_FRAMELEN for an unusual reason.
3220 	 * The bge hardware will pad out Tx runts to BGE_MIN_FRAMELEN,
3221 	 * but when such padded frames employ the bge IP/TCP checksum
3222 	 * offload, the hardware checksum assist gives incorrect results
3223 	 * (possibly from incorporating its own padding into the UDP/TCP
3224 	 * checksum; who knows).  If we pad such runts with zeros, the
3225 	 * onboard checksum comes out correct.
3226 	 */
3227 	if ((csum_flags & BGE_TXBDFLAG_TCP_UDP_CSUM) &&
3228 	    m_head->m_pkthdr.len < BGE_MIN_FRAMELEN) {
3229 		error = m_devpad(m_head, BGE_MIN_FRAMELEN);
3230 		if (error)
3231 			goto back;
3232 	}
3233 
3234 	if ((sc->bge_flags & BGE_FLAG_SHORTDMA) && m_head->m_next != NULL) {
3235 		m_new = bge_defrag_shortdma(m_head);
3236 		if (m_new == NULL) {
3237 			error = ENOBUFS;
3238 			goto back;
3239 		}
3240 		*m_head0 = m_head = m_new;
3241 	}
3242 	if (sc->bge_force_defrag && (sc->bge_flags & BGE_FLAG_PCIE) &&
3243 	    m_head->m_next != NULL) {
3244 		/*
3245 		 * Forcefully defragment mbuf chain to overcome hardware
3246 		 * limitation which only support a single outstanding
3247 		 * DMA read operation.  If it fails, keep moving on using
3248 		 * the original mbuf chain.
3249 		 */
3250 		m_new = m_defrag(m_head, MB_DONTWAIT);
3251 		if (m_new != NULL)
3252 			*m_head0 = m_head = m_new;
3253 	}
3254 
3255 	error = bus_dmamap_load_mbuf_defrag(sc->bge_cdata.bge_tx_mtag, map,
3256 			m_head0, segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
3257 	if (error)
3258 		goto back;
3259 
3260 	m_head = *m_head0;
3261 	bus_dmamap_sync(sc->bge_cdata.bge_tx_mtag, map, BUS_DMASYNC_PREWRITE);
3262 
3263 	for (i = 0; ; i++) {
3264 		d = &sc->bge_ldata.bge_tx_ring[idx];
3265 
3266 		d->bge_addr.bge_addr_lo = BGE_ADDR_LO(segs[i].ds_addr);
3267 		d->bge_addr.bge_addr_hi = BGE_ADDR_HI(segs[i].ds_addr);
3268 		d->bge_len = segs[i].ds_len;
3269 		d->bge_flags = csum_flags;
3270 
3271 		if (i == nsegs - 1)
3272 			break;
3273 		BGE_INC(idx, BGE_TX_RING_CNT);
3274 	}
3275 	/* Mark the last segment as end of packet... */
3276 	d->bge_flags |= BGE_TXBDFLAG_END;
3277 
3278 	/* Set vlan tag to the first segment of the packet. */
3279 	d = &sc->bge_ldata.bge_tx_ring[*txidx];
3280 	if (m_head->m_flags & M_VLANTAG) {
3281 		d->bge_flags |= BGE_TXBDFLAG_VLAN_TAG;
3282 		d->bge_vlan_tag = m_head->m_pkthdr.ether_vlantag;
3283 	} else {
3284 		d->bge_vlan_tag = 0;
3285 	}
3286 
3287 	/*
3288 	 * Insure that the map for this transmission is placed at
3289 	 * the array index of the last descriptor in this chain.
3290 	 */
3291 	sc->bge_cdata.bge_tx_dmamap[*txidx] = sc->bge_cdata.bge_tx_dmamap[idx];
3292 	sc->bge_cdata.bge_tx_dmamap[idx] = map;
3293 	sc->bge_cdata.bge_tx_chain[idx] = m_head;
3294 	sc->bge_txcnt += nsegs;
3295 
3296 	BGE_INC(idx, BGE_TX_RING_CNT);
3297 	*txidx = idx;
3298 back:
3299 	if (error) {
3300 		m_freem(*m_head0);
3301 		*m_head0 = NULL;
3302 	}
3303 	return error;
3304 }
3305 
3306 /*
3307  * Main transmit routine. To avoid having to do mbuf copies, we put pointers
3308  * to the mbuf data regions directly in the transmit descriptors.
3309  */
3310 static void
3311 bge_start(struct ifnet *ifp)
3312 {
3313 	struct bge_softc *sc = ifp->if_softc;
3314 	struct mbuf *m_head = NULL;
3315 	uint32_t prodidx;
3316 	int need_trans;
3317 
3318 	if ((ifp->if_flags & (IFF_RUNNING | IFF_OACTIVE)) != IFF_RUNNING)
3319 		return;
3320 
3321 	prodidx = sc->bge_tx_prodidx;
3322 
3323 	need_trans = 0;
3324 	while (sc->bge_cdata.bge_tx_chain[prodidx] == NULL) {
3325 		m_head = ifq_dequeue(&ifp->if_snd, NULL);
3326 		if (m_head == NULL)
3327 			break;
3328 
3329 		/*
3330 		 * XXX
3331 		 * The code inside the if() block is never reached since we
3332 		 * must mark CSUM_IP_FRAGS in our if_hwassist to start getting
3333 		 * requests to checksum TCP/UDP in a fragmented packet.
3334 		 *
3335 		 * XXX
3336 		 * safety overkill.  If this is a fragmented packet chain
3337 		 * with delayed TCP/UDP checksums, then only encapsulate
3338 		 * it if we have enough descriptors to handle the entire
3339 		 * chain at once.
3340 		 * (paranoia -- may not actually be needed)
3341 		 */
3342 		if ((m_head->m_flags & M_FIRSTFRAG) &&
3343 		    (m_head->m_pkthdr.csum_flags & CSUM_DELAY_DATA)) {
3344 			if ((BGE_TX_RING_CNT - sc->bge_txcnt) <
3345 			    m_head->m_pkthdr.csum_data + BGE_NSEG_RSVD) {
3346 				ifp->if_flags |= IFF_OACTIVE;
3347 				ifq_prepend(&ifp->if_snd, m_head);
3348 				break;
3349 			}
3350 		}
3351 
3352 		/*
3353 		 * Sanity check: avoid coming within BGE_NSEG_RSVD
3354 		 * descriptors of the end of the ring.  Also make
3355 		 * sure there are BGE_NSEG_SPARE descriptors for
3356 		 * jumbo buffers' defragmentation.
3357 		 */
3358 		if ((BGE_TX_RING_CNT - sc->bge_txcnt) <
3359 		    (BGE_NSEG_RSVD + BGE_NSEG_SPARE)) {
3360 			ifp->if_flags |= IFF_OACTIVE;
3361 			ifq_prepend(&ifp->if_snd, m_head);
3362 			break;
3363 		}
3364 
3365 		/*
3366 		 * Pack the data into the transmit ring. If we
3367 		 * don't have room, set the OACTIVE flag and wait
3368 		 * for the NIC to drain the ring.
3369 		 */
3370 		if (bge_encap(sc, &m_head, &prodidx)) {
3371 			ifp->if_flags |= IFF_OACTIVE;
3372 			ifp->if_oerrors++;
3373 			break;
3374 		}
3375 		need_trans = 1;
3376 
3377 		ETHER_BPF_MTAP(ifp, m_head);
3378 	}
3379 
3380 	if (!need_trans)
3381 		return;
3382 
3383 	/* Transmit */
3384 	bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
3385 	/* 5700 b2 errata */
3386 	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
3387 		bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
3388 
3389 	sc->bge_tx_prodidx = prodidx;
3390 
3391 	/*
3392 	 * Set a timeout in case the chip goes out to lunch.
3393 	 */
3394 	ifp->if_timer = 5;
3395 }
3396 
3397 static void
3398 bge_init(void *xsc)
3399 {
3400 	struct bge_softc *sc = xsc;
3401 	struct ifnet *ifp = &sc->arpcom.ac_if;
3402 	uint16_t *m;
3403 	uint32_t mode;
3404 
3405 	ASSERT_SERIALIZED(ifp->if_serializer);
3406 
3407 	/* Cancel pending I/O and flush buffers. */
3408 	bge_stop(sc);
3409 	bge_reset(sc);
3410 	bge_chipinit(sc);
3411 
3412 	/*
3413 	 * Init the various state machines, ring
3414 	 * control blocks and firmware.
3415 	 */
3416 	if (bge_blockinit(sc)) {
3417 		if_printf(ifp, "initialization failure\n");
3418 		bge_stop(sc);
3419 		return;
3420 	}
3421 
3422 	/* Specify MTU. */
3423 	CSR_WRITE_4(sc, BGE_RX_MTU, ifp->if_mtu +
3424 	    ETHER_HDR_LEN + ETHER_CRC_LEN + EVL_ENCAPLEN);
3425 
3426 	/* Load our MAC address. */
3427 	m = (uint16_t *)&sc->arpcom.ac_enaddr[0];
3428 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_LO, htons(m[0]));
3429 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_HI, (htons(m[1]) << 16) | htons(m[2]));
3430 
3431 	/* Enable or disable promiscuous mode as needed. */
3432 	bge_setpromisc(sc);
3433 
3434 	/* Program multicast filter. */
3435 	bge_setmulti(sc);
3436 
3437 	/* Init RX ring. */
3438 	if (bge_init_rx_ring_std(sc)) {
3439 		if_printf(ifp, "RX ring initialization failed\n");
3440 		bge_stop(sc);
3441 		return;
3442 	}
3443 
3444 	/*
3445 	 * Workaround for a bug in 5705 ASIC rev A0. Poll the NIC's
3446 	 * memory to insure that the chip has in fact read the first
3447 	 * entry of the ring.
3448 	 */
3449 	if (sc->bge_chipid == BGE_CHIPID_BCM5705_A0) {
3450 		uint32_t		v, i;
3451 		for (i = 0; i < 10; i++) {
3452 			DELAY(20);
3453 			v = bge_readmem_ind(sc, BGE_STD_RX_RINGS + 8);
3454 			if (v == (MCLBYTES - ETHER_ALIGN))
3455 				break;
3456 		}
3457 		if (i == 10)
3458 			if_printf(ifp, "5705 A0 chip failed to load RX ring\n");
3459 	}
3460 
3461 	/* Init jumbo RX ring. */
3462 	if (ifp->if_mtu > (ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN)) {
3463 		if (bge_init_rx_ring_jumbo(sc)) {
3464 			if_printf(ifp, "Jumbo RX ring initialization failed\n");
3465 			bge_stop(sc);
3466 			return;
3467 		}
3468 	}
3469 
3470 	/* Init our RX return ring index */
3471 	sc->bge_rx_saved_considx = 0;
3472 
3473 	/* Init TX ring. */
3474 	bge_init_tx_ring(sc);
3475 
3476 	/* Enable TX MAC state machine lockup fix. */
3477 	mode = CSR_READ_4(sc, BGE_TX_MODE);
3478 	if (BGE_IS_5755_PLUS(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5906)
3479 		mode |= BGE_TXMODE_MBUF_LOCKUP_FIX;
3480 	/* Turn on transmitter */
3481 	CSR_WRITE_4(sc, BGE_TX_MODE, mode | BGE_TXMODE_ENABLE);
3482 
3483 	/* Turn on receiver */
3484 	BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
3485 
3486 	/*
3487 	 * Set the number of good frames to receive after RX MBUF
3488 	 * Low Watermark has been reached.  After the RX MAC receives
3489 	 * this number of frames, it will drop subsequent incoming
3490 	 * frames until the MBUF High Watermark is reached.
3491 	 */
3492 	CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 2);
3493 
3494 	/* Tell firmware we're alive. */
3495 	BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3496 
3497 	/* Enable host interrupts if polling(4) is not enabled. */
3498 	PCI_SETBIT(sc->bge_dev, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_CLEAR_INTA, 4);
3499 #ifdef DEVICE_POLLING
3500 	if (ifp->if_flags & IFF_POLLING)
3501 		bge_disable_intr(sc);
3502 	else
3503 #endif
3504 	bge_enable_intr(sc);
3505 
3506 	bge_ifmedia_upd(ifp);
3507 
3508 	ifp->if_flags |= IFF_RUNNING;
3509 	ifp->if_flags &= ~IFF_OACTIVE;
3510 
3511 	callout_reset(&sc->bge_stat_timer, hz, bge_tick, sc);
3512 }
3513 
3514 /*
3515  * Set media options.
3516  */
3517 static int
3518 bge_ifmedia_upd(struct ifnet *ifp)
3519 {
3520 	struct bge_softc *sc = ifp->if_softc;
3521 
3522 	/* If this is a 1000baseX NIC, enable the TBI port. */
3523 	if (sc->bge_flags & BGE_FLAG_TBI) {
3524 		struct ifmedia *ifm = &sc->bge_ifmedia;
3525 
3526 		if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
3527 			return(EINVAL);
3528 
3529 		switch(IFM_SUBTYPE(ifm->ifm_media)) {
3530 		case IFM_AUTO:
3531 			/*
3532 			 * The BCM5704 ASIC appears to have a special
3533 			 * mechanism for programming the autoneg
3534 			 * advertisement registers in TBI mode.
3535 			 */
3536 			if (!bge_fake_autoneg &&
3537 			    sc->bge_asicrev == BGE_ASICREV_BCM5704) {
3538 				uint32_t sgdig;
3539 
3540 				CSR_WRITE_4(sc, BGE_TX_TBI_AUTONEG, 0);
3541 				sgdig = CSR_READ_4(sc, BGE_SGDIG_CFG);
3542 				sgdig |= BGE_SGDIGCFG_AUTO |
3543 					 BGE_SGDIGCFG_PAUSE_CAP |
3544 					 BGE_SGDIGCFG_ASYM_PAUSE;
3545 				CSR_WRITE_4(sc, BGE_SGDIG_CFG,
3546 					    sgdig | BGE_SGDIGCFG_SEND);
3547 				DELAY(5);
3548 				CSR_WRITE_4(sc, BGE_SGDIG_CFG, sgdig);
3549 			}
3550 			break;
3551 		case IFM_1000_SX:
3552 			if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX) {
3553 				BGE_CLRBIT(sc, BGE_MAC_MODE,
3554 				    BGE_MACMODE_HALF_DUPLEX);
3555 			} else {
3556 				BGE_SETBIT(sc, BGE_MAC_MODE,
3557 				    BGE_MACMODE_HALF_DUPLEX);
3558 			}
3559 			break;
3560 		default:
3561 			return(EINVAL);
3562 		}
3563 	} else {
3564 		struct mii_data *mii = device_get_softc(sc->bge_miibus);
3565 
3566 		sc->bge_link_evt++;
3567 		sc->bge_link = 0;
3568 		if (mii->mii_instance) {
3569 			struct mii_softc *miisc;
3570 
3571 			LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
3572 				mii_phy_reset(miisc);
3573 		}
3574 		mii_mediachg(mii);
3575 
3576 		/*
3577 		 * Force an interrupt so that we will call bge_link_upd
3578 		 * if needed and clear any pending link state attention.
3579 		 * Without this we are not getting any further interrupts
3580 		 * for link state changes and thus will not UP the link and
3581 		 * not be able to send in bge_start.  The only way to get
3582 		 * things working was to receive a packet and get an RX
3583 		 * intr.
3584 		 *
3585 		 * bge_tick should help for fiber cards and we might not
3586 		 * need to do this here if BGE_FLAG_TBI is set but as
3587 		 * we poll for fiber anyway it should not harm.
3588 		 */
3589 		if (BGE_IS_CRIPPLED(sc))
3590 			BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
3591 		else
3592 			BGE_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
3593 	}
3594 	return(0);
3595 }
3596 
3597 /*
3598  * Report current media status.
3599  */
3600 static void
3601 bge_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3602 {
3603 	struct bge_softc *sc = ifp->if_softc;
3604 
3605 	if (sc->bge_flags & BGE_FLAG_TBI) {
3606 		ifmr->ifm_status = IFM_AVALID;
3607 		ifmr->ifm_active = IFM_ETHER;
3608 		if (CSR_READ_4(sc, BGE_MAC_STS) &
3609 		    BGE_MACSTAT_TBI_PCS_SYNCHED) {
3610 			ifmr->ifm_status |= IFM_ACTIVE;
3611 		} else {
3612 			ifmr->ifm_active |= IFM_NONE;
3613 			return;
3614 		}
3615 
3616 		ifmr->ifm_active |= IFM_1000_SX;
3617 		if (CSR_READ_4(sc, BGE_MAC_MODE) & BGE_MACMODE_HALF_DUPLEX)
3618 			ifmr->ifm_active |= IFM_HDX;
3619 		else
3620 			ifmr->ifm_active |= IFM_FDX;
3621 	} else {
3622 		struct mii_data *mii = device_get_softc(sc->bge_miibus);
3623 
3624 		mii_pollstat(mii);
3625 		ifmr->ifm_active = mii->mii_media_active;
3626 		ifmr->ifm_status = mii->mii_media_status;
3627 	}
3628 }
3629 
3630 static int
3631 bge_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
3632 {
3633 	struct bge_softc *sc = ifp->if_softc;
3634 	struct ifreq *ifr = (struct ifreq *)data;
3635 	int mask, error = 0;
3636 
3637 	ASSERT_SERIALIZED(ifp->if_serializer);
3638 
3639 	switch (command) {
3640 	case SIOCSIFMTU:
3641 		if ((!BGE_IS_JUMBO_CAPABLE(sc) && ifr->ifr_mtu > ETHERMTU) ||
3642 		    (BGE_IS_JUMBO_CAPABLE(sc) &&
3643 		     ifr->ifr_mtu > BGE_JUMBO_MTU)) {
3644 			error = EINVAL;
3645 		} else if (ifp->if_mtu != ifr->ifr_mtu) {
3646 			ifp->if_mtu = ifr->ifr_mtu;
3647 			if (ifp->if_flags & IFF_RUNNING)
3648 				bge_init(sc);
3649 		}
3650 		break;
3651 	case SIOCSIFFLAGS:
3652 		if (ifp->if_flags & IFF_UP) {
3653 			if (ifp->if_flags & IFF_RUNNING) {
3654 				mask = ifp->if_flags ^ sc->bge_if_flags;
3655 
3656 				/*
3657 				 * If only the state of the PROMISC flag
3658 				 * changed, then just use the 'set promisc
3659 				 * mode' command instead of reinitializing
3660 				 * the entire NIC. Doing a full re-init
3661 				 * means reloading the firmware and waiting
3662 				 * for it to start up, which may take a
3663 				 * second or two.  Similarly for ALLMULTI.
3664 				 */
3665 				if (mask & IFF_PROMISC)
3666 					bge_setpromisc(sc);
3667 				if (mask & IFF_ALLMULTI)
3668 					bge_setmulti(sc);
3669 			} else {
3670 				bge_init(sc);
3671 			}
3672 		} else if (ifp->if_flags & IFF_RUNNING) {
3673 			bge_stop(sc);
3674 		}
3675 		sc->bge_if_flags = ifp->if_flags;
3676 		break;
3677 	case SIOCADDMULTI:
3678 	case SIOCDELMULTI:
3679 		if (ifp->if_flags & IFF_RUNNING)
3680 			bge_setmulti(sc);
3681 		break;
3682 	case SIOCSIFMEDIA:
3683 	case SIOCGIFMEDIA:
3684 		if (sc->bge_flags & BGE_FLAG_TBI) {
3685 			error = ifmedia_ioctl(ifp, ifr,
3686 			    &sc->bge_ifmedia, command);
3687 		} else {
3688 			struct mii_data *mii;
3689 
3690 			mii = device_get_softc(sc->bge_miibus);
3691 			error = ifmedia_ioctl(ifp, ifr,
3692 					      &mii->mii_media, command);
3693 		}
3694 		break;
3695         case SIOCSIFCAP:
3696 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3697 		if (mask & IFCAP_HWCSUM) {
3698 			ifp->if_capenable ^= (mask & IFCAP_HWCSUM);
3699 			if (IFCAP_HWCSUM & ifp->if_capenable)
3700 				ifp->if_hwassist = BGE_CSUM_FEATURES;
3701 			else
3702 				ifp->if_hwassist = 0;
3703 		}
3704 		break;
3705 	default:
3706 		error = ether_ioctl(ifp, command, data);
3707 		break;
3708 	}
3709 	return error;
3710 }
3711 
3712 static void
3713 bge_watchdog(struct ifnet *ifp)
3714 {
3715 	struct bge_softc *sc = ifp->if_softc;
3716 
3717 	if_printf(ifp, "watchdog timeout -- resetting\n");
3718 
3719 	bge_init(sc);
3720 
3721 	ifp->if_oerrors++;
3722 
3723 	if (!ifq_is_empty(&ifp->if_snd))
3724 		if_devstart(ifp);
3725 }
3726 
3727 /*
3728  * Stop the adapter and free any mbufs allocated to the
3729  * RX and TX lists.
3730  */
3731 static void
3732 bge_stop(struct bge_softc *sc)
3733 {
3734 	struct ifnet *ifp = &sc->arpcom.ac_if;
3735 
3736 	ASSERT_SERIALIZED(ifp->if_serializer);
3737 
3738 	callout_stop(&sc->bge_stat_timer);
3739 
3740 	/*
3741 	 * Disable all of the receiver blocks
3742 	 */
3743 	bge_stop_block(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
3744 	bge_stop_block(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
3745 	bge_stop_block(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
3746 	if (BGE_IS_5700_FAMILY(sc))
3747 		bge_stop_block(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE);
3748 	bge_stop_block(sc, BGE_RDBDI_MODE, BGE_RBDIMODE_ENABLE);
3749 	bge_stop_block(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
3750 	bge_stop_block(sc, BGE_RBDC_MODE, BGE_RBDCMODE_ENABLE);
3751 
3752 	/*
3753 	 * Disable all of the transmit blocks
3754 	 */
3755 	bge_stop_block(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
3756 	bge_stop_block(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
3757 	bge_stop_block(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
3758 	bge_stop_block(sc, BGE_RDMA_MODE, BGE_RDMAMODE_ENABLE);
3759 	bge_stop_block(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE);
3760 	if (BGE_IS_5700_FAMILY(sc))
3761 		bge_stop_block(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE);
3762 	bge_stop_block(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
3763 
3764 	/*
3765 	 * Shut down all of the memory managers and related
3766 	 * state machines.
3767 	 */
3768 	bge_stop_block(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE);
3769 	bge_stop_block(sc, BGE_WDMA_MODE, BGE_WDMAMODE_ENABLE);
3770 	if (BGE_IS_5700_FAMILY(sc))
3771 		bge_stop_block(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE);
3772 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
3773 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
3774 	if (!BGE_IS_5705_PLUS(sc)) {
3775 		BGE_CLRBIT(sc, BGE_BMAN_MODE, BGE_BMANMODE_ENABLE);
3776 		BGE_CLRBIT(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
3777 	}
3778 
3779 	/* Disable host interrupts. */
3780 	bge_disable_intr(sc);
3781 
3782 	/*
3783 	 * Tell firmware we're shutting down.
3784 	 */
3785 	BGE_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3786 
3787 	/* Free the RX lists. */
3788 	bge_free_rx_ring_std(sc);
3789 
3790 	/* Free jumbo RX list. */
3791 	if (BGE_IS_JUMBO_CAPABLE(sc))
3792 		bge_free_rx_ring_jumbo(sc);
3793 
3794 	/* Free TX buffers. */
3795 	bge_free_tx_ring(sc);
3796 
3797 	sc->bge_status_tag = 0;
3798 	sc->bge_link = 0;
3799 	sc->bge_coal_chg = 0;
3800 
3801 	sc->bge_tx_saved_considx = BGE_TXCONS_UNSET;
3802 
3803 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
3804 	ifp->if_timer = 0;
3805 }
3806 
3807 /*
3808  * Stop all chip I/O so that the kernel's probe routines don't
3809  * get confused by errant DMAs when rebooting.
3810  */
3811 static void
3812 bge_shutdown(device_t dev)
3813 {
3814 	struct bge_softc *sc = device_get_softc(dev);
3815 	struct ifnet *ifp = &sc->arpcom.ac_if;
3816 
3817 	lwkt_serialize_enter(ifp->if_serializer);
3818 	bge_stop(sc);
3819 	bge_reset(sc);
3820 	lwkt_serialize_exit(ifp->if_serializer);
3821 }
3822 
3823 static int
3824 bge_suspend(device_t dev)
3825 {
3826 	struct bge_softc *sc = device_get_softc(dev);
3827 	struct ifnet *ifp = &sc->arpcom.ac_if;
3828 
3829 	lwkt_serialize_enter(ifp->if_serializer);
3830 	bge_stop(sc);
3831 	lwkt_serialize_exit(ifp->if_serializer);
3832 
3833 	return 0;
3834 }
3835 
3836 static int
3837 bge_resume(device_t dev)
3838 {
3839 	struct bge_softc *sc = device_get_softc(dev);
3840 	struct ifnet *ifp = &sc->arpcom.ac_if;
3841 
3842 	lwkt_serialize_enter(ifp->if_serializer);
3843 
3844 	if (ifp->if_flags & IFF_UP) {
3845 		bge_init(sc);
3846 
3847 		if (!ifq_is_empty(&ifp->if_snd))
3848 			if_devstart(ifp);
3849 	}
3850 
3851 	lwkt_serialize_exit(ifp->if_serializer);
3852 
3853 	return 0;
3854 }
3855 
3856 static void
3857 bge_setpromisc(struct bge_softc *sc)
3858 {
3859 	struct ifnet *ifp = &sc->arpcom.ac_if;
3860 
3861 	if (ifp->if_flags & IFF_PROMISC)
3862 		BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
3863 	else
3864 		BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
3865 }
3866 
3867 static void
3868 bge_dma_free(struct bge_softc *sc)
3869 {
3870 	int i;
3871 
3872 	/* Destroy RX mbuf DMA stuffs. */
3873 	if (sc->bge_cdata.bge_rx_mtag != NULL) {
3874 		for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
3875 			bus_dmamap_destroy(sc->bge_cdata.bge_rx_mtag,
3876 			    sc->bge_cdata.bge_rx_std_dmamap[i]);
3877 		}
3878 		bus_dmamap_destroy(sc->bge_cdata.bge_rx_mtag,
3879 				   sc->bge_cdata.bge_rx_tmpmap);
3880 		bus_dma_tag_destroy(sc->bge_cdata.bge_rx_mtag);
3881 	}
3882 
3883 	/* Destroy TX mbuf DMA stuffs. */
3884 	if (sc->bge_cdata.bge_tx_mtag != NULL) {
3885 		for (i = 0; i < BGE_TX_RING_CNT; i++) {
3886 			bus_dmamap_destroy(sc->bge_cdata.bge_tx_mtag,
3887 			    sc->bge_cdata.bge_tx_dmamap[i]);
3888 		}
3889 		bus_dma_tag_destroy(sc->bge_cdata.bge_tx_mtag);
3890 	}
3891 
3892 	/* Destroy standard RX ring */
3893 	bge_dma_block_free(sc->bge_cdata.bge_rx_std_ring_tag,
3894 			   sc->bge_cdata.bge_rx_std_ring_map,
3895 			   sc->bge_ldata.bge_rx_std_ring);
3896 
3897 	if (BGE_IS_JUMBO_CAPABLE(sc))
3898 		bge_free_jumbo_mem(sc);
3899 
3900 	/* Destroy RX return ring */
3901 	bge_dma_block_free(sc->bge_cdata.bge_rx_return_ring_tag,
3902 			   sc->bge_cdata.bge_rx_return_ring_map,
3903 			   sc->bge_ldata.bge_rx_return_ring);
3904 
3905 	/* Destroy TX ring */
3906 	bge_dma_block_free(sc->bge_cdata.bge_tx_ring_tag,
3907 			   sc->bge_cdata.bge_tx_ring_map,
3908 			   sc->bge_ldata.bge_tx_ring);
3909 
3910 	/* Destroy status block */
3911 	bge_dma_block_free(sc->bge_cdata.bge_status_tag,
3912 			   sc->bge_cdata.bge_status_map,
3913 			   sc->bge_ldata.bge_status_block);
3914 
3915 	/* Destroy statistics block */
3916 	bge_dma_block_free(sc->bge_cdata.bge_stats_tag,
3917 			   sc->bge_cdata.bge_stats_map,
3918 			   sc->bge_ldata.bge_stats);
3919 
3920 	/* Destroy the parent tag */
3921 	if (sc->bge_cdata.bge_parent_tag != NULL)
3922 		bus_dma_tag_destroy(sc->bge_cdata.bge_parent_tag);
3923 }
3924 
3925 static int
3926 bge_dma_alloc(struct bge_softc *sc)
3927 {
3928 	struct ifnet *ifp = &sc->arpcom.ac_if;
3929 	int i, error;
3930 	bus_addr_t lowaddr;
3931 
3932 	lowaddr = BUS_SPACE_MAXADDR;
3933 	if (sc->bge_flags & BGE_FLAG_MAXADDR_40BIT)
3934 		lowaddr = BGE_DMA_MAXADDR_40BIT;
3935 
3936 	/*
3937 	 * Allocate the parent bus DMA tag appropriate for PCI.
3938 	 *
3939 	 * All of the NetExtreme/NetLink controllers have 4GB boundary
3940 	 * DMA bug.
3941 	 * Whenever an address crosses a multiple of the 4GB boundary
3942 	 * (including 4GB, 8Gb, 12Gb, etc.) and makes the transition
3943 	 * from 0xX_FFFF_FFFF to 0x(X+1)_0000_0000 an internal DMA
3944 	 * state machine will lockup and cause the device to hang.
3945 	 */
3946 	error = bus_dma_tag_create(NULL, 1, BGE_DMA_BOUNDARY_4G,
3947 				   lowaddr, BUS_SPACE_MAXADDR,
3948 				   NULL, NULL,
3949 				   BUS_SPACE_MAXSIZE_32BIT, 0,
3950 				   BUS_SPACE_MAXSIZE_32BIT,
3951 				   0, &sc->bge_cdata.bge_parent_tag);
3952 	if (error) {
3953 		if_printf(ifp, "could not allocate parent dma tag\n");
3954 		return error;
3955 	}
3956 
3957 	/*
3958 	 * Create DMA tag and maps for RX mbufs.
3959 	 */
3960 	error = bus_dma_tag_create(sc->bge_cdata.bge_parent_tag, 1, 0,
3961 				   BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
3962 				   NULL, NULL, MCLBYTES, 1, MCLBYTES,
3963 				   BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK,
3964 				   &sc->bge_cdata.bge_rx_mtag);
3965 	if (error) {
3966 		if_printf(ifp, "could not allocate RX mbuf dma tag\n");
3967 		return error;
3968 	}
3969 
3970 	error = bus_dmamap_create(sc->bge_cdata.bge_rx_mtag,
3971 				  BUS_DMA_WAITOK, &sc->bge_cdata.bge_rx_tmpmap);
3972 	if (error) {
3973 		bus_dma_tag_destroy(sc->bge_cdata.bge_rx_mtag);
3974 		sc->bge_cdata.bge_rx_mtag = NULL;
3975 		return error;
3976 	}
3977 
3978 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
3979 		error = bus_dmamap_create(sc->bge_cdata.bge_rx_mtag,
3980 					  BUS_DMA_WAITOK,
3981 					  &sc->bge_cdata.bge_rx_std_dmamap[i]);
3982 		if (error) {
3983 			int j;
3984 
3985 			for (j = 0; j < i; ++j) {
3986 				bus_dmamap_destroy(sc->bge_cdata.bge_rx_mtag,
3987 					sc->bge_cdata.bge_rx_std_dmamap[j]);
3988 			}
3989 			bus_dma_tag_destroy(sc->bge_cdata.bge_rx_mtag);
3990 			sc->bge_cdata.bge_rx_mtag = NULL;
3991 
3992 			if_printf(ifp, "could not create DMA map for RX\n");
3993 			return error;
3994 		}
3995 	}
3996 
3997 	/*
3998 	 * Create DMA tag and maps for TX mbufs.
3999 	 */
4000 	error = bus_dma_tag_create(sc->bge_cdata.bge_parent_tag, 1, 0,
4001 				   BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
4002 				   NULL, NULL,
4003 				   BGE_JUMBO_FRAMELEN, BGE_NSEG_NEW, MCLBYTES,
4004 				   BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK |
4005 				   BUS_DMA_ONEBPAGE,
4006 				   &sc->bge_cdata.bge_tx_mtag);
4007 	if (error) {
4008 		if_printf(ifp, "could not allocate TX mbuf dma tag\n");
4009 		return error;
4010 	}
4011 
4012 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
4013 		error = bus_dmamap_create(sc->bge_cdata.bge_tx_mtag,
4014 					  BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
4015 					  &sc->bge_cdata.bge_tx_dmamap[i]);
4016 		if (error) {
4017 			int j;
4018 
4019 			for (j = 0; j < i; ++j) {
4020 				bus_dmamap_destroy(sc->bge_cdata.bge_tx_mtag,
4021 					sc->bge_cdata.bge_tx_dmamap[j]);
4022 			}
4023 			bus_dma_tag_destroy(sc->bge_cdata.bge_tx_mtag);
4024 			sc->bge_cdata.bge_tx_mtag = NULL;
4025 
4026 			if_printf(ifp, "could not create DMA map for TX\n");
4027 			return error;
4028 		}
4029 	}
4030 
4031 	/*
4032 	 * Create DMA stuffs for standard RX ring.
4033 	 */
4034 	error = bge_dma_block_alloc(sc, BGE_STD_RX_RING_SZ,
4035 				    &sc->bge_cdata.bge_rx_std_ring_tag,
4036 				    &sc->bge_cdata.bge_rx_std_ring_map,
4037 				    (void *)&sc->bge_ldata.bge_rx_std_ring,
4038 				    &sc->bge_ldata.bge_rx_std_ring_paddr);
4039 	if (error) {
4040 		if_printf(ifp, "could not create std RX ring\n");
4041 		return error;
4042 	}
4043 
4044 	/*
4045 	 * Create jumbo buffer pool.
4046 	 */
4047 	if (BGE_IS_JUMBO_CAPABLE(sc)) {
4048 		error = bge_alloc_jumbo_mem(sc);
4049 		if (error) {
4050 			if_printf(ifp, "could not create jumbo buffer pool\n");
4051 			return error;
4052 		}
4053 	}
4054 
4055 	/*
4056 	 * Create DMA stuffs for RX return ring.
4057 	 */
4058 	error = bge_dma_block_alloc(sc,
4059 	    BGE_RX_RTN_RING_SZ(sc->bge_return_ring_cnt),
4060 	    &sc->bge_cdata.bge_rx_return_ring_tag,
4061 	    &sc->bge_cdata.bge_rx_return_ring_map,
4062 	    (void *)&sc->bge_ldata.bge_rx_return_ring,
4063 	    &sc->bge_ldata.bge_rx_return_ring_paddr);
4064 	if (error) {
4065 		if_printf(ifp, "could not create RX ret ring\n");
4066 		return error;
4067 	}
4068 
4069 	/*
4070 	 * Create DMA stuffs for TX ring.
4071 	 */
4072 	error = bge_dma_block_alloc(sc, BGE_TX_RING_SZ,
4073 				    &sc->bge_cdata.bge_tx_ring_tag,
4074 				    &sc->bge_cdata.bge_tx_ring_map,
4075 				    (void *)&sc->bge_ldata.bge_tx_ring,
4076 				    &sc->bge_ldata.bge_tx_ring_paddr);
4077 	if (error) {
4078 		if_printf(ifp, "could not create TX ring\n");
4079 		return error;
4080 	}
4081 
4082 	/*
4083 	 * Create DMA stuffs for status block.
4084 	 */
4085 	error = bge_dma_block_alloc(sc, BGE_STATUS_BLK_SZ,
4086 				    &sc->bge_cdata.bge_status_tag,
4087 				    &sc->bge_cdata.bge_status_map,
4088 				    (void *)&sc->bge_ldata.bge_status_block,
4089 				    &sc->bge_ldata.bge_status_block_paddr);
4090 	if (error) {
4091 		if_printf(ifp, "could not create status block\n");
4092 		return error;
4093 	}
4094 
4095 	/*
4096 	 * Create DMA stuffs for statistics block.
4097 	 */
4098 	error = bge_dma_block_alloc(sc, BGE_STATS_SZ,
4099 				    &sc->bge_cdata.bge_stats_tag,
4100 				    &sc->bge_cdata.bge_stats_map,
4101 				    (void *)&sc->bge_ldata.bge_stats,
4102 				    &sc->bge_ldata.bge_stats_paddr);
4103 	if (error) {
4104 		if_printf(ifp, "could not create stats block\n");
4105 		return error;
4106 	}
4107 	return 0;
4108 }
4109 
4110 static int
4111 bge_dma_block_alloc(struct bge_softc *sc, bus_size_t size, bus_dma_tag_t *tag,
4112 		    bus_dmamap_t *map, void **addr, bus_addr_t *paddr)
4113 {
4114 	bus_dmamem_t dmem;
4115 	int error;
4116 
4117 	error = bus_dmamem_coherent(sc->bge_cdata.bge_parent_tag, PAGE_SIZE, 0,
4118 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
4119 				    size, BUS_DMA_WAITOK | BUS_DMA_ZERO, &dmem);
4120 	if (error)
4121 		return error;
4122 
4123 	*tag = dmem.dmem_tag;
4124 	*map = dmem.dmem_map;
4125 	*addr = dmem.dmem_addr;
4126 	*paddr = dmem.dmem_busaddr;
4127 
4128 	return 0;
4129 }
4130 
4131 static void
4132 bge_dma_block_free(bus_dma_tag_t tag, bus_dmamap_t map, void *addr)
4133 {
4134 	if (tag != NULL) {
4135 		bus_dmamap_unload(tag, map);
4136 		bus_dmamem_free(tag, addr, map);
4137 		bus_dma_tag_destroy(tag);
4138 	}
4139 }
4140 
4141 /*
4142  * Grrr. The link status word in the status block does
4143  * not work correctly on the BCM5700 rev AX and BX chips,
4144  * according to all available information. Hence, we have
4145  * to enable MII interrupts in order to properly obtain
4146  * async link changes. Unfortunately, this also means that
4147  * we have to read the MAC status register to detect link
4148  * changes, thereby adding an additional register access to
4149  * the interrupt handler.
4150  *
4151  * XXX: perhaps link state detection procedure used for
4152  * BGE_CHIPID_BCM5700_B2 can be used for others BCM5700 revisions.
4153  */
4154 static void
4155 bge_bcm5700_link_upd(struct bge_softc *sc, uint32_t status __unused)
4156 {
4157 	struct ifnet *ifp = &sc->arpcom.ac_if;
4158 	struct mii_data *mii = device_get_softc(sc->bge_miibus);
4159 
4160 	mii_pollstat(mii);
4161 
4162 	if (!sc->bge_link &&
4163 	    (mii->mii_media_status & IFM_ACTIVE) &&
4164 	    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
4165 		sc->bge_link++;
4166 		if (bootverbose)
4167 			if_printf(ifp, "link UP\n");
4168 	} else if (sc->bge_link &&
4169 	    (!(mii->mii_media_status & IFM_ACTIVE) ||
4170 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE)) {
4171 		sc->bge_link = 0;
4172 		if (bootverbose)
4173 			if_printf(ifp, "link DOWN\n");
4174 	}
4175 
4176 	/* Clear the interrupt. */
4177 	CSR_WRITE_4(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_MI_INTERRUPT);
4178 	bge_miibus_readreg(sc->bge_dev, 1, BRGPHY_MII_ISR);
4179 	bge_miibus_writereg(sc->bge_dev, 1, BRGPHY_MII_IMR, BRGPHY_INTRS);
4180 }
4181 
4182 static void
4183 bge_tbi_link_upd(struct bge_softc *sc, uint32_t status)
4184 {
4185 	struct ifnet *ifp = &sc->arpcom.ac_if;
4186 
4187 #define PCS_ENCODE_ERR	(BGE_MACSTAT_PORT_DECODE_ERROR|BGE_MACSTAT_MI_COMPLETE)
4188 
4189 	/*
4190 	 * Sometimes PCS encoding errors are detected in
4191 	 * TBI mode (on fiber NICs), and for some reason
4192 	 * the chip will signal them as link changes.
4193 	 * If we get a link change event, but the 'PCS
4194 	 * encoding error' bit in the MAC status register
4195 	 * is set, don't bother doing a link check.
4196 	 * This avoids spurious "gigabit link up" messages
4197 	 * that sometimes appear on fiber NICs during
4198 	 * periods of heavy traffic.
4199 	 */
4200 	if (status & BGE_MACSTAT_TBI_PCS_SYNCHED) {
4201 		if (!sc->bge_link) {
4202 			sc->bge_link++;
4203 			if (sc->bge_asicrev == BGE_ASICREV_BCM5704) {
4204 				BGE_CLRBIT(sc, BGE_MAC_MODE,
4205 				    BGE_MACMODE_TBI_SEND_CFGS);
4206 			}
4207 			CSR_WRITE_4(sc, BGE_MAC_STS, 0xFFFFFFFF);
4208 
4209 			if (bootverbose)
4210 				if_printf(ifp, "link UP\n");
4211 
4212 			ifp->if_link_state = LINK_STATE_UP;
4213 			if_link_state_change(ifp);
4214 		}
4215 	} else if ((status & PCS_ENCODE_ERR) != PCS_ENCODE_ERR) {
4216 		if (sc->bge_link) {
4217 			sc->bge_link = 0;
4218 
4219 			if (bootverbose)
4220 				if_printf(ifp, "link DOWN\n");
4221 
4222 			ifp->if_link_state = LINK_STATE_DOWN;
4223 			if_link_state_change(ifp);
4224 		}
4225 	}
4226 
4227 #undef PCS_ENCODE_ERR
4228 
4229 	/* Clear the attention. */
4230 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
4231 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
4232 	    BGE_MACSTAT_LINK_CHANGED);
4233 }
4234 
4235 static void
4236 bge_copper_link_upd(struct bge_softc *sc, uint32_t status __unused)
4237 {
4238 	struct ifnet *ifp = &sc->arpcom.ac_if;
4239 	struct mii_data *mii = device_get_softc(sc->bge_miibus);
4240 
4241 	mii_pollstat(mii);
4242 	bge_miibus_statchg(sc->bge_dev);
4243 
4244 	if (bootverbose) {
4245 		if (sc->bge_link)
4246 			if_printf(ifp, "link UP\n");
4247 		else
4248 			if_printf(ifp, "link DOWN\n");
4249 	}
4250 
4251 	/* Clear the attention. */
4252 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
4253 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
4254 	    BGE_MACSTAT_LINK_CHANGED);
4255 }
4256 
4257 static void
4258 bge_autopoll_link_upd(struct bge_softc *sc, uint32_t status __unused)
4259 {
4260 	struct ifnet *ifp = &sc->arpcom.ac_if;
4261 	struct mii_data *mii = device_get_softc(sc->bge_miibus);
4262 
4263 	mii_pollstat(mii);
4264 
4265 	if (!sc->bge_link &&
4266 	    (mii->mii_media_status & IFM_ACTIVE) &&
4267 	    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
4268 		sc->bge_link++;
4269 		if (bootverbose)
4270 			if_printf(ifp, "link UP\n");
4271 	} else if (sc->bge_link &&
4272 	    (!(mii->mii_media_status & IFM_ACTIVE) ||
4273 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE)) {
4274 		sc->bge_link = 0;
4275 		if (bootverbose)
4276 			if_printf(ifp, "link DOWN\n");
4277 	}
4278 
4279 	/* Clear the attention. */
4280 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
4281 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
4282 	    BGE_MACSTAT_LINK_CHANGED);
4283 }
4284 
4285 static int
4286 bge_sysctl_rx_coal_ticks(SYSCTL_HANDLER_ARGS)
4287 {
4288 	struct bge_softc *sc = arg1;
4289 
4290 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4291 	    &sc->bge_rx_coal_ticks,
4292 	    BGE_RX_COAL_TICKS_MIN, BGE_RX_COAL_TICKS_MAX,
4293 	    BGE_RX_COAL_TICKS_CHG);
4294 }
4295 
4296 static int
4297 bge_sysctl_tx_coal_ticks(SYSCTL_HANDLER_ARGS)
4298 {
4299 	struct bge_softc *sc = arg1;
4300 
4301 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4302 	    &sc->bge_tx_coal_ticks,
4303 	    BGE_TX_COAL_TICKS_MIN, BGE_TX_COAL_TICKS_MAX,
4304 	    BGE_TX_COAL_TICKS_CHG);
4305 }
4306 
4307 static int
4308 bge_sysctl_rx_coal_bds(SYSCTL_HANDLER_ARGS)
4309 {
4310 	struct bge_softc *sc = arg1;
4311 
4312 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4313 	    &sc->bge_rx_coal_bds,
4314 	    BGE_RX_COAL_BDS_MIN, BGE_RX_COAL_BDS_MAX,
4315 	    BGE_RX_COAL_BDS_CHG);
4316 }
4317 
4318 static int
4319 bge_sysctl_tx_coal_bds(SYSCTL_HANDLER_ARGS)
4320 {
4321 	struct bge_softc *sc = arg1;
4322 
4323 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4324 	    &sc->bge_tx_coal_bds,
4325 	    BGE_TX_COAL_BDS_MIN, BGE_TX_COAL_BDS_MAX,
4326 	    BGE_TX_COAL_BDS_CHG);
4327 }
4328 
4329 static int
4330 bge_sysctl_rx_coal_ticks_int(SYSCTL_HANDLER_ARGS)
4331 {
4332 	struct bge_softc *sc = arg1;
4333 
4334 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4335 	    &sc->bge_rx_coal_ticks_int,
4336 	    BGE_RX_COAL_TICKS_MIN, BGE_RX_COAL_TICKS_MAX,
4337 	    BGE_RX_COAL_TICKS_INT_CHG);
4338 }
4339 
4340 static int
4341 bge_sysctl_tx_coal_ticks_int(SYSCTL_HANDLER_ARGS)
4342 {
4343 	struct bge_softc *sc = arg1;
4344 
4345 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4346 	    &sc->bge_tx_coal_ticks_int,
4347 	    BGE_TX_COAL_TICKS_MIN, BGE_TX_COAL_TICKS_MAX,
4348 	    BGE_TX_COAL_TICKS_INT_CHG);
4349 }
4350 
4351 static int
4352 bge_sysctl_rx_coal_bds_int(SYSCTL_HANDLER_ARGS)
4353 {
4354 	struct bge_softc *sc = arg1;
4355 
4356 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4357 	    &sc->bge_rx_coal_bds_int,
4358 	    BGE_RX_COAL_BDS_MIN, BGE_RX_COAL_BDS_MAX,
4359 	    BGE_RX_COAL_BDS_INT_CHG);
4360 }
4361 
4362 static int
4363 bge_sysctl_tx_coal_bds_int(SYSCTL_HANDLER_ARGS)
4364 {
4365 	struct bge_softc *sc = arg1;
4366 
4367 	return bge_sysctl_coal_chg(oidp, arg1, arg2, req,
4368 	    &sc->bge_tx_coal_bds_int,
4369 	    BGE_TX_COAL_BDS_MIN, BGE_TX_COAL_BDS_MAX,
4370 	    BGE_TX_COAL_BDS_INT_CHG);
4371 }
4372 
4373 static int
4374 bge_sysctl_coal_chg(SYSCTL_HANDLER_ARGS, uint32_t *coal,
4375     int coal_min, int coal_max, uint32_t coal_chg_mask)
4376 {
4377 	struct bge_softc *sc = arg1;
4378 	struct ifnet *ifp = &sc->arpcom.ac_if;
4379 	int error = 0, v;
4380 
4381 	lwkt_serialize_enter(ifp->if_serializer);
4382 
4383 	v = *coal;
4384 	error = sysctl_handle_int(oidp, &v, 0, req);
4385 	if (!error && req->newptr != NULL) {
4386 		if (v < coal_min || v > coal_max) {
4387 			error = EINVAL;
4388 		} else {
4389 			*coal = v;
4390 			sc->bge_coal_chg |= coal_chg_mask;
4391 		}
4392 	}
4393 
4394 	lwkt_serialize_exit(ifp->if_serializer);
4395 	return error;
4396 }
4397 
4398 static void
4399 bge_coal_change(struct bge_softc *sc)
4400 {
4401 	struct ifnet *ifp = &sc->arpcom.ac_if;
4402 	uint32_t val;
4403 
4404 	ASSERT_SERIALIZED(ifp->if_serializer);
4405 
4406 	if (sc->bge_coal_chg & BGE_RX_COAL_TICKS_CHG) {
4407 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS,
4408 			    sc->bge_rx_coal_ticks);
4409 		DELAY(10);
4410 		val = CSR_READ_4(sc, BGE_HCC_RX_COAL_TICKS);
4411 
4412 		if (bootverbose) {
4413 			if_printf(ifp, "rx_coal_ticks -> %u\n",
4414 				  sc->bge_rx_coal_ticks);
4415 		}
4416 	}
4417 
4418 	if (sc->bge_coal_chg & BGE_TX_COAL_TICKS_CHG) {
4419 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS,
4420 			    sc->bge_tx_coal_ticks);
4421 		DELAY(10);
4422 		val = CSR_READ_4(sc, BGE_HCC_TX_COAL_TICKS);
4423 
4424 		if (bootverbose) {
4425 			if_printf(ifp, "tx_coal_ticks -> %u\n",
4426 				  sc->bge_tx_coal_ticks);
4427 		}
4428 	}
4429 
4430 	if (sc->bge_coal_chg & BGE_RX_COAL_BDS_CHG) {
4431 		CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS,
4432 			    sc->bge_rx_coal_bds);
4433 		DELAY(10);
4434 		val = CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS);
4435 
4436 		if (bootverbose) {
4437 			if_printf(ifp, "rx_coal_bds -> %u\n",
4438 				  sc->bge_rx_coal_bds);
4439 		}
4440 	}
4441 
4442 	if (sc->bge_coal_chg & BGE_TX_COAL_BDS_CHG) {
4443 		CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS,
4444 			    sc->bge_tx_coal_bds);
4445 		DELAY(10);
4446 		val = CSR_READ_4(sc, BGE_HCC_TX_MAX_COAL_BDS);
4447 
4448 		if (bootverbose) {
4449 			if_printf(ifp, "tx_max_coal_bds -> %u\n",
4450 				  sc->bge_tx_coal_bds);
4451 		}
4452 	}
4453 
4454 	if (sc->bge_coal_chg & BGE_RX_COAL_TICKS_INT_CHG) {
4455 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS_INT,
4456 		    sc->bge_rx_coal_ticks_int);
4457 		DELAY(10);
4458 		val = CSR_READ_4(sc, BGE_HCC_RX_COAL_TICKS_INT);
4459 
4460 		if (bootverbose) {
4461 			if_printf(ifp, "rx_coal_ticks_int -> %u\n",
4462 			    sc->bge_rx_coal_ticks_int);
4463 		}
4464 	}
4465 
4466 	if (sc->bge_coal_chg & BGE_TX_COAL_TICKS_INT_CHG) {
4467 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS_INT,
4468 		    sc->bge_tx_coal_ticks_int);
4469 		DELAY(10);
4470 		val = CSR_READ_4(sc, BGE_HCC_TX_COAL_TICKS_INT);
4471 
4472 		if (bootverbose) {
4473 			if_printf(ifp, "tx_coal_ticks_int -> %u\n",
4474 			    sc->bge_tx_coal_ticks_int);
4475 		}
4476 	}
4477 
4478 	if (sc->bge_coal_chg & BGE_RX_COAL_BDS_INT_CHG) {
4479 		CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT,
4480 		    sc->bge_rx_coal_bds_int);
4481 		DELAY(10);
4482 		val = CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT);
4483 
4484 		if (bootverbose) {
4485 			if_printf(ifp, "rx_coal_bds_int -> %u\n",
4486 			    sc->bge_rx_coal_bds_int);
4487 		}
4488 	}
4489 
4490 	if (sc->bge_coal_chg & BGE_TX_COAL_BDS_INT_CHG) {
4491 		CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT,
4492 		    sc->bge_tx_coal_bds_int);
4493 		DELAY(10);
4494 		val = CSR_READ_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT);
4495 
4496 		if (bootverbose) {
4497 			if_printf(ifp, "tx_coal_bds_int -> %u\n",
4498 			    sc->bge_tx_coal_bds_int);
4499 		}
4500 	}
4501 
4502 	sc->bge_coal_chg = 0;
4503 }
4504 
4505 static void
4506 bge_enable_intr(struct bge_softc *sc)
4507 {
4508 	struct ifnet *ifp = &sc->arpcom.ac_if;
4509 
4510 	lwkt_serialize_handler_enable(ifp->if_serializer);
4511 
4512 	/*
4513 	 * Enable interrupt.
4514 	 */
4515 	bge_writembx(sc, BGE_MBX_IRQ0_LO, sc->bge_status_tag << 24);
4516 
4517 	/*
4518 	 * Unmask the interrupt when we stop polling.
4519 	 */
4520 	PCI_CLRBIT(sc->bge_dev, BGE_PCI_MISC_CTL,
4521 	    BGE_PCIMISCCTL_MASK_PCI_INTR, 4);
4522 
4523 	/*
4524 	 * Trigger another interrupt, since above writing
4525 	 * to interrupt mailbox0 may acknowledge pending
4526 	 * interrupt.
4527 	 */
4528 	BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
4529 }
4530 
4531 static void
4532 bge_disable_intr(struct bge_softc *sc)
4533 {
4534 	struct ifnet *ifp = &sc->arpcom.ac_if;
4535 
4536 	/*
4537 	 * Mask the interrupt when we start polling.
4538 	 */
4539 	PCI_SETBIT(sc->bge_dev, BGE_PCI_MISC_CTL,
4540 	    BGE_PCIMISCCTL_MASK_PCI_INTR, 4);
4541 
4542 	/*
4543 	 * Acknowledge possible asserted interrupt.
4544 	 */
4545 	bge_writembx(sc, BGE_MBX_IRQ0_LO, 1);
4546 
4547 	lwkt_serialize_handler_disable(ifp->if_serializer);
4548 }
4549 
4550 static int
4551 bge_get_eaddr_mem(struct bge_softc *sc, uint8_t ether_addr[])
4552 {
4553 	uint32_t mac_addr;
4554 	int ret = 1;
4555 
4556 	mac_addr = bge_readmem_ind(sc, 0x0c14);
4557 	if ((mac_addr >> 16) == 0x484b) {
4558 		ether_addr[0] = (uint8_t)(mac_addr >> 8);
4559 		ether_addr[1] = (uint8_t)mac_addr;
4560 		mac_addr = bge_readmem_ind(sc, 0x0c18);
4561 		ether_addr[2] = (uint8_t)(mac_addr >> 24);
4562 		ether_addr[3] = (uint8_t)(mac_addr >> 16);
4563 		ether_addr[4] = (uint8_t)(mac_addr >> 8);
4564 		ether_addr[5] = (uint8_t)mac_addr;
4565 		ret = 0;
4566 	}
4567 	return ret;
4568 }
4569 
4570 static int
4571 bge_get_eaddr_nvram(struct bge_softc *sc, uint8_t ether_addr[])
4572 {
4573 	int mac_offset = BGE_EE_MAC_OFFSET;
4574 
4575 	if (sc->bge_asicrev == BGE_ASICREV_BCM5906)
4576 		mac_offset = BGE_EE_MAC_OFFSET_5906;
4577 
4578 	return bge_read_nvram(sc, ether_addr, mac_offset + 2, ETHER_ADDR_LEN);
4579 }
4580 
4581 static int
4582 bge_get_eaddr_eeprom(struct bge_softc *sc, uint8_t ether_addr[])
4583 {
4584 	if (sc->bge_flags & BGE_FLAG_NO_EEPROM)
4585 		return 1;
4586 
4587 	return bge_read_eeprom(sc, ether_addr, BGE_EE_MAC_OFFSET + 2,
4588 			       ETHER_ADDR_LEN);
4589 }
4590 
4591 static int
4592 bge_get_eaddr(struct bge_softc *sc, uint8_t eaddr[])
4593 {
4594 	static const bge_eaddr_fcn_t bge_eaddr_funcs[] = {
4595 		/* NOTE: Order is critical */
4596 		bge_get_eaddr_mem,
4597 		bge_get_eaddr_nvram,
4598 		bge_get_eaddr_eeprom,
4599 		NULL
4600 	};
4601 	const bge_eaddr_fcn_t *func;
4602 
4603 	for (func = bge_eaddr_funcs; *func != NULL; ++func) {
4604 		if ((*func)(sc, eaddr) == 0)
4605 			break;
4606 	}
4607 	return (*func == NULL ? ENXIO : 0);
4608 }
4609 
4610 /*
4611  * NOTE: 'm' is not freed upon failure
4612  */
4613 struct mbuf *
4614 bge_defrag_shortdma(struct mbuf *m)
4615 {
4616 	struct mbuf *n;
4617 	int found;
4618 
4619 	/*
4620 	 * If device receive two back-to-back send BDs with less than
4621 	 * or equal to 8 total bytes then the device may hang.  The two
4622 	 * back-to-back send BDs must in the same frame for this failure
4623 	 * to occur.  Scan mbuf chains and see whether two back-to-back
4624 	 * send BDs are there.  If this is the case, allocate new mbuf
4625 	 * and copy the frame to workaround the silicon bug.
4626 	 */
4627 	for (n = m, found = 0; n != NULL; n = n->m_next) {
4628 		if (n->m_len < 8) {
4629 			found++;
4630 			if (found > 1)
4631 				break;
4632 			continue;
4633 		}
4634 		found = 0;
4635 	}
4636 
4637 	if (found > 1)
4638 		n = m_defrag(m, MB_DONTWAIT);
4639 	else
4640 		n = m;
4641 	return n;
4642 }
4643 
4644 static void
4645 bge_stop_block(struct bge_softc *sc, bus_size_t reg, uint32_t bit)
4646 {
4647 	int i;
4648 
4649 	BGE_CLRBIT(sc, reg, bit);
4650 	for (i = 0; i < BGE_TIMEOUT; i++) {
4651 		if ((CSR_READ_4(sc, reg) & bit) == 0)
4652 			return;
4653 		DELAY(100);
4654 	}
4655 }
4656 
4657 static void
4658 bge_link_poll(struct bge_softc *sc)
4659 {
4660 	uint32_t status;
4661 
4662 	status = CSR_READ_4(sc, BGE_MAC_STS);
4663 	if ((status & sc->bge_link_chg) || sc->bge_link_evt) {
4664 		sc->bge_link_evt = 0;
4665 		sc->bge_link_upd(sc, status);
4666 	}
4667 }
4668